diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gvt')
36 files changed, 16870 insertions, 25 deletions
diff --git a/drivers/gpu/drm/i915/gvt/Makefile b/drivers/gpu/drm/i915/gvt/Makefile index d0f21a6ad60d..8a46a7f31d53 100644 --- a/drivers/gpu/drm/i915/gvt/Makefile +++ b/drivers/gpu/drm/i915/gvt/Makefile @@ -1,5 +1,10 @@ GVT_DIR := gvt -GVT_SOURCE := gvt.o +GVT_SOURCE := gvt.o aperture_gm.o handlers.o vgpu.o trace_points.o firmware.o \ + interrupt.o gtt.o cfg_space.o opregion.o mmio.o display.o edid.o \ + execlist.o scheduler.o sched_policy.o render.o cmd_parser.o -ccflags-y += -I$(src) -I$(src)/$(GVT_DIR) -Wall -i915-y += $(addprefix $(GVT_DIR)/, $(GVT_SOURCE)) +ccflags-y += -I$(src) -I$(src)/$(GVT_DIR) -Wall +i915-y += $(addprefix $(GVT_DIR)/, $(GVT_SOURCE)) + +CFLAGS_kvmgt.o := -Wno-unused-function +obj-$(CONFIG_DRM_I915_GVT_KVMGT) += $(GVT_DIR)/kvmgt.o diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c new file mode 100644 index 000000000000..0d41ebc4aea6 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c @@ -0,0 +1,352 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Dexuan Cui + * + * Contributors: + * Pei Zhang <pei.zhang@intel.com> + * Min He <min.he@intel.com> + * Niu Bing <bing.niu@intel.com> + * Yulei Zhang <yulei.zhang@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +#define MB_TO_BYTES(mb) ((mb) << 20ULL) +#define BYTES_TO_MB(b) ((b) >> 20ULL) + +#define HOST_LOW_GM_SIZE MB_TO_BYTES(128) +#define HOST_HIGH_GM_SIZE MB_TO_BYTES(384) +#define HOST_FENCE 4 + +static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + u32 alloc_flag, search_flag; + u64 start, end, size; + struct drm_mm_node *node; + int retried = 0; + int ret; + + if (high_gm) { + search_flag = DRM_MM_SEARCH_BELOW; + alloc_flag = DRM_MM_CREATE_TOP; + node = &vgpu->gm.high_gm_node; + size = vgpu_hidden_sz(vgpu); + start = gvt_hidden_gmadr_base(gvt); + end = gvt_hidden_gmadr_end(gvt); + } else { + search_flag = DRM_MM_SEARCH_DEFAULT; + alloc_flag = DRM_MM_CREATE_DEFAULT; + node = &vgpu->gm.low_gm_node; + size = vgpu_aperture_sz(vgpu); + start = gvt_aperture_gmadr_base(gvt); + end = gvt_aperture_gmadr_end(gvt); + } + + mutex_lock(&dev_priv->drm.struct_mutex); +search_again: + ret = drm_mm_insert_node_in_range_generic(&dev_priv->ggtt.base.mm, + node, size, 4096, 0, + start, end, search_flag, + alloc_flag); + if (ret) { + ret = i915_gem_evict_something(&dev_priv->ggtt.base, + size, 4096, 0, start, end, 0); + if (ret == 0 && ++retried < 3) + goto search_again; + + gvt_err("fail to alloc %s gm space from host, retried %d\n", + high_gm ? "high" : "low", retried); + } + mutex_unlock(&dev_priv->drm.struct_mutex); + return ret; +} + +static int alloc_vgpu_gm(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ret; + + ret = alloc_gm(vgpu, false); + if (ret) + return ret; + + ret = alloc_gm(vgpu, true); + if (ret) + goto out_free_aperture; + + gvt_dbg_core("vgpu%d: alloc low GM start %llx size %llx\n", vgpu->id, + vgpu_aperture_offset(vgpu), vgpu_aperture_sz(vgpu)); + + gvt_dbg_core("vgpu%d: alloc high GM start %llx size %llx\n", vgpu->id, + vgpu_hidden_offset(vgpu), vgpu_hidden_sz(vgpu)); + + return 0; +out_free_aperture: + mutex_lock(&dev_priv->drm.struct_mutex); + drm_mm_remove_node(&vgpu->gm.low_gm_node); + mutex_unlock(&dev_priv->drm.struct_mutex); + return ret; +} + +static void free_vgpu_gm(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + mutex_lock(&dev_priv->drm.struct_mutex); + drm_mm_remove_node(&vgpu->gm.low_gm_node); + drm_mm_remove_node(&vgpu->gm.high_gm_node); + mutex_unlock(&dev_priv->drm.struct_mutex); +} + +/** + * intel_vgpu_write_fence - write fence registers owned by a vGPU + * @vgpu: vGPU instance + * @fence: vGPU fence register number + * @value: Fence register value to be written + * + * This function is used to write fence registers owned by a vGPU. The vGPU + * fence register number will be translated into HW fence register number. + * + */ +void intel_vgpu_write_fence(struct intel_vgpu *vgpu, + u32 fence, u64 value) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + struct drm_i915_fence_reg *reg; + i915_reg_t fence_reg_lo, fence_reg_hi; + + assert_rpm_wakelock_held(dev_priv); + + if (WARN_ON(fence > vgpu_fence_sz(vgpu))) + return; + + reg = vgpu->fence.regs[fence]; + if (WARN_ON(!reg)) + return; + + fence_reg_lo = FENCE_REG_GEN6_LO(reg->id); + fence_reg_hi = FENCE_REG_GEN6_HI(reg->id); + + I915_WRITE(fence_reg_lo, 0); + POSTING_READ(fence_reg_lo); + + I915_WRITE(fence_reg_hi, upper_32_bits(value)); + I915_WRITE(fence_reg_lo, lower_32_bits(value)); + POSTING_READ(fence_reg_lo); +} + +static void free_vgpu_fence(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + struct drm_i915_fence_reg *reg; + u32 i; + + if (WARN_ON(!vgpu_fence_sz(vgpu))) + return; + + intel_runtime_pm_get(dev_priv); + + mutex_lock(&dev_priv->drm.struct_mutex); + for (i = 0; i < vgpu_fence_sz(vgpu); i++) { + reg = vgpu->fence.regs[i]; + intel_vgpu_write_fence(vgpu, i, 0); + list_add_tail(®->link, + &dev_priv->mm.fence_list); + } + mutex_unlock(&dev_priv->drm.struct_mutex); + + intel_runtime_pm_put(dev_priv); +} + +static int alloc_vgpu_fence(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct drm_i915_private *dev_priv = gvt->dev_priv; + struct drm_i915_fence_reg *reg; + int i; + struct list_head *pos, *q; + + intel_runtime_pm_get(dev_priv); + + /* Request fences from host */ + mutex_lock(&dev_priv->drm.struct_mutex); + i = 0; + list_for_each_safe(pos, q, &dev_priv->mm.fence_list) { + reg = list_entry(pos, struct drm_i915_fence_reg, link); + if (reg->pin_count || reg->vma) + continue; + list_del(pos); + vgpu->fence.regs[i] = reg; + intel_vgpu_write_fence(vgpu, i, 0); + if (++i == vgpu_fence_sz(vgpu)) + break; + } + if (i != vgpu_fence_sz(vgpu)) + goto out_free_fence; + + mutex_unlock(&dev_priv->drm.struct_mutex); + intel_runtime_pm_put(dev_priv); + return 0; +out_free_fence: + /* Return fences to host, if fail */ + for (i = 0; i < vgpu_fence_sz(vgpu); i++) { + reg = vgpu->fence.regs[i]; + if (!reg) + continue; + list_add_tail(®->link, + &dev_priv->mm.fence_list); + } + mutex_unlock(&dev_priv->drm.struct_mutex); + intel_runtime_pm_put(dev_priv); + return -ENOSPC; +} + +static void free_resource(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + + gvt->gm.vgpu_allocated_low_gm_size -= vgpu_aperture_sz(vgpu); + gvt->gm.vgpu_allocated_high_gm_size -= vgpu_hidden_sz(vgpu); + gvt->fence.vgpu_allocated_fence_num -= vgpu_fence_sz(vgpu); +} + +static int alloc_resource(struct intel_vgpu *vgpu, + struct intel_vgpu_creation_params *param) +{ + struct intel_gvt *gvt = vgpu->gvt; + unsigned long request, avail, max, taken; + const char *item; + + if (!param->low_gm_sz || !param->high_gm_sz || !param->fence_sz) { + gvt_err("Invalid vGPU creation params\n"); + return -EINVAL; + } + + item = "low GM space"; + max = gvt_aperture_sz(gvt) - HOST_LOW_GM_SIZE; + taken = gvt->gm.vgpu_allocated_low_gm_size; + avail = max - taken; + request = MB_TO_BYTES(param->low_gm_sz); + + if (request > avail) + goto no_enough_resource; + + vgpu_aperture_sz(vgpu) = request; + + item = "high GM space"; + max = gvt_hidden_sz(gvt) - HOST_HIGH_GM_SIZE; + taken = gvt->gm.vgpu_allocated_high_gm_size; + avail = max - taken; + request = MB_TO_BYTES(param->high_gm_sz); + + if (request > avail) + goto no_enough_resource; + + vgpu_hidden_sz(vgpu) = request; + + item = "fence"; + max = gvt_fence_sz(gvt) - HOST_FENCE; + taken = gvt->fence.vgpu_allocated_fence_num; + avail = max - taken; + request = param->fence_sz; + + if (request > avail) + goto no_enough_resource; + + vgpu_fence_sz(vgpu) = request; + + gvt->gm.vgpu_allocated_low_gm_size += MB_TO_BYTES(param->low_gm_sz); + gvt->gm.vgpu_allocated_high_gm_size += MB_TO_BYTES(param->high_gm_sz); + gvt->fence.vgpu_allocated_fence_num += param->fence_sz; + return 0; + +no_enough_resource: + gvt_err("vgpu%d: fail to allocate resource %s\n", vgpu->id, item); + gvt_err("vgpu%d: request %luMB avail %luMB max %luMB taken %luMB\n", + vgpu->id, BYTES_TO_MB(request), BYTES_TO_MB(avail), + BYTES_TO_MB(max), BYTES_TO_MB(taken)); + return -ENOSPC; +} + +/** + * inte_gvt_free_vgpu_resource - free HW resource owned by a vGPU + * @vgpu: a vGPU + * + * This function is used to free the HW resource owned by a vGPU. + * + */ +void intel_vgpu_free_resource(struct intel_vgpu *vgpu) +{ + free_vgpu_gm(vgpu); + free_vgpu_fence(vgpu); + free_resource(vgpu); +} + +/** + * intel_alloc_vgpu_resource - allocate HW resource for a vGPU + * @vgpu: vGPU + * @param: vGPU creation params + * + * This function is used to allocate HW resource for a vGPU. User specifies + * the resource configuration through the creation params. + * + * Returns: + * zero on success, negative error code if failed. + * + */ +int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu, + struct intel_vgpu_creation_params *param) +{ + int ret; + + ret = alloc_resource(vgpu, param); + if (ret) + return ret; + + ret = alloc_vgpu_gm(vgpu); + if (ret) + goto out_free_resource; + + ret = alloc_vgpu_fence(vgpu); + if (ret) + goto out_free_vgpu_gm; + + return 0; + +out_free_vgpu_gm: + free_vgpu_gm(vgpu); +out_free_resource: + free_resource(vgpu); + return ret; +} diff --git a/drivers/gpu/drm/i915/gvt/cfg_space.c b/drivers/gpu/drm/i915/gvt/cfg_space.c new file mode 100644 index 000000000000..db516382a4d4 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/cfg_space.c @@ -0,0 +1,284 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Jike Song <jike.song@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +enum { + INTEL_GVT_PCI_BAR_GTTMMIO = 0, + INTEL_GVT_PCI_BAR_APERTURE, + INTEL_GVT_PCI_BAR_PIO, + INTEL_GVT_PCI_BAR_MAX, +}; + +/** + * intel_vgpu_emulate_cfg_read - emulate vGPU configuration space read + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_emulate_cfg_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + if (WARN_ON(bytes > 4)) + return -EINVAL; + + if (WARN_ON(offset + bytes > INTEL_GVT_MAX_CFG_SPACE_SZ)) + return -EINVAL; + + memcpy(p_data, vgpu_cfg_space(vgpu) + offset, bytes); + return 0; +} + +static int map_aperture(struct intel_vgpu *vgpu, bool map) +{ + u64 first_gfn, first_mfn; + u64 val; + int ret; + + if (map == vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_APERTURE].tracked) + return 0; + + val = vgpu_cfg_space(vgpu)[PCI_BASE_ADDRESS_2]; + if (val & PCI_BASE_ADDRESS_MEM_TYPE_64) + val = *(u64 *)(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_2); + else + val = *(u32 *)(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_2); + + first_gfn = (val + vgpu_aperture_offset(vgpu)) >> PAGE_SHIFT; + first_mfn = vgpu_aperture_pa_base(vgpu) >> PAGE_SHIFT; + + ret = intel_gvt_hypervisor_map_gfn_to_mfn(vgpu, first_gfn, + first_mfn, + vgpu_aperture_sz(vgpu) >> + PAGE_SHIFT, map); + if (ret) + return ret; + + vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_APERTURE].tracked = map; + return 0; +} + +static int trap_gttmmio(struct intel_vgpu *vgpu, bool trap) +{ + u64 start, end; + u64 val; + int ret; + + if (trap == vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_GTTMMIO].tracked) + return 0; + + val = vgpu_cfg_space(vgpu)[PCI_BASE_ADDRESS_0]; + if (val & PCI_BASE_ADDRESS_MEM_TYPE_64) + start = *(u64 *)(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_0); + else + start = *(u32 *)(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_0); + + start &= ~GENMASK(3, 0); + end = start + vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_GTTMMIO].size - 1; + + ret = intel_gvt_hypervisor_set_trap_area(vgpu, start, end, trap); + if (ret) + return ret; + + vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_GTTMMIO].tracked = trap; + return 0; +} + +static int emulate_pci_command_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u8 old = vgpu_cfg_space(vgpu)[offset]; + u8 new = *(u8 *)p_data; + u8 changed = old ^ new; + int ret; + + if (!(changed & PCI_COMMAND_MEMORY)) + return 0; + + if (old & PCI_COMMAND_MEMORY) { + ret = trap_gttmmio(vgpu, false); + if (ret) + return ret; + ret = map_aperture(vgpu, false); + if (ret) + return ret; + } else { + ret = trap_gttmmio(vgpu, true); + if (ret) + return ret; + ret = map_aperture(vgpu, true); + if (ret) + return ret; + } + + memcpy(vgpu_cfg_space(vgpu) + offset, p_data, bytes); + return 0; +} + +static int emulate_pci_bar_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + unsigned int bar_index = + (rounddown(offset, 8) % PCI_BASE_ADDRESS_0) / 8; + u32 new = *(u32 *)(p_data); + bool lo = IS_ALIGNED(offset, 8); + u64 size; + int ret = 0; + bool mmio_enabled = + vgpu_cfg_space(vgpu)[PCI_COMMAND] & PCI_COMMAND_MEMORY; + + if (WARN_ON(bar_index >= INTEL_GVT_PCI_BAR_MAX)) + return -EINVAL; + + if (new == 0xffffffff) { + /* + * Power-up software can determine how much address + * space the device requires by writing a value of + * all 1's to the register and then reading the value + * back. The device will return 0's in all don't-care + * address bits. + */ + size = vgpu->cfg_space.bar[bar_index].size; + if (lo) { + new = rounddown(new, size); + } else { + u32 val = vgpu_cfg_space(vgpu)[rounddown(offset, 8)]; + /* for 32bit mode bar it returns all-0 in upper 32 + * bit, for 64bit mode bar it will calculate the + * size with lower 32bit and return the corresponding + * value + */ + if (val & PCI_BASE_ADDRESS_MEM_TYPE_64) + new &= (~(size-1)) >> 32; + else + new = 0; + } + /* + * Unmapp & untrap the BAR, since guest hasn't configured a + * valid GPA + */ + switch (bar_index) { + case INTEL_GVT_PCI_BAR_GTTMMIO: + ret = trap_gttmmio(vgpu, false); + break; + case INTEL_GVT_PCI_BAR_APERTURE: + ret = map_aperture(vgpu, false); + break; + } + intel_vgpu_write_pci_bar(vgpu, offset, new, lo); + } else { + /* + * Unmapp & untrap the old BAR first, since guest has + * re-configured the BAR + */ + switch (bar_index) { + case INTEL_GVT_PCI_BAR_GTTMMIO: + ret = trap_gttmmio(vgpu, false); + break; + case INTEL_GVT_PCI_BAR_APERTURE: + ret = map_aperture(vgpu, false); + break; + } + intel_vgpu_write_pci_bar(vgpu, offset, new, lo); + /* Track the new BAR */ + if (mmio_enabled) { + switch (bar_index) { + case INTEL_GVT_PCI_BAR_GTTMMIO: + ret = trap_gttmmio(vgpu, true); + break; + case INTEL_GVT_PCI_BAR_APERTURE: + ret = map_aperture(vgpu, true); + break; + } + } + } + return ret; +} + +/** + * intel_vgpu_emulate_cfg_read - emulate vGPU configuration space write + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + int ret; + + if (WARN_ON(bytes > 4)) + return -EINVAL; + + if (WARN_ON(offset + bytes >= INTEL_GVT_MAX_CFG_SPACE_SZ)) + return -EINVAL; + + /* First check if it's PCI_COMMAND */ + if (IS_ALIGNED(offset, 2) && offset == PCI_COMMAND) { + if (WARN_ON(bytes > 2)) + return -EINVAL; + return emulate_pci_command_write(vgpu, offset, p_data, bytes); + } + + switch (rounddown(offset, 4)) { + case PCI_BASE_ADDRESS_0: + case PCI_BASE_ADDRESS_1: + case PCI_BASE_ADDRESS_2: + case PCI_BASE_ADDRESS_3: + if (WARN_ON(!IS_ALIGNED(offset, 4))) + return -EINVAL; + return emulate_pci_bar_write(vgpu, offset, p_data, bytes); + + case INTEL_GVT_PCI_SWSCI: + if (WARN_ON(!IS_ALIGNED(offset, 4))) + return -EINVAL; + ret = intel_vgpu_emulate_opregion_request(vgpu, *(u32 *)p_data); + if (ret) + return ret; + break; + + case INTEL_GVT_PCI_OPREGION: + if (WARN_ON(!IS_ALIGNED(offset, 4))) + return -EINVAL; + ret = intel_vgpu_init_opregion(vgpu, *(u32 *)p_data); + if (ret) + return ret; + + memcpy(vgpu_cfg_space(vgpu) + offset, p_data, bytes); + break; + default: + memcpy(vgpu_cfg_space(vgpu) + offset, p_data, bytes); + break; + } + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/cmd_parser.c b/drivers/gpu/drm/i915/gvt/cmd_parser.c new file mode 100644 index 000000000000..d26a092c70e8 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/cmd_parser.c @@ -0,0 +1,2831 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Kevin Tian <kevin.tian@intel.com> + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Yulei Zhang <yulei.zhang@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include <linux/slab.h> +#include "i915_drv.h" +#include "gvt.h" +#include "i915_pvinfo.h" +#include "trace.h" + +#define INVALID_OP (~0U) + +#define OP_LEN_MI 9 +#define OP_LEN_2D 10 +#define OP_LEN_3D_MEDIA 16 +#define OP_LEN_MFX_VC 16 +#define OP_LEN_VEBOX 16 + +#define CMD_TYPE(cmd) (((cmd) >> 29) & 7) + +struct sub_op_bits { + int hi; + int low; +}; +struct decode_info { + char *name; + int op_len; + int nr_sub_op; + struct sub_op_bits *sub_op; +}; + +#define MAX_CMD_BUDGET 0x7fffffff +#define MI_WAIT_FOR_PLANE_C_FLIP_PENDING (1<<15) +#define MI_WAIT_FOR_PLANE_B_FLIP_PENDING (1<<9) +#define MI_WAIT_FOR_PLANE_A_FLIP_PENDING (1<<1) + +#define MI_WAIT_FOR_SPRITE_C_FLIP_PENDING (1<<20) +#define MI_WAIT_FOR_SPRITE_B_FLIP_PENDING (1<<10) +#define MI_WAIT_FOR_SPRITE_A_FLIP_PENDING (1<<2) + +/* Render Command Map */ + +/* MI_* command Opcode (28:23) */ +#define OP_MI_NOOP 0x0 +#define OP_MI_SET_PREDICATE 0x1 /* HSW+ */ +#define OP_MI_USER_INTERRUPT 0x2 +#define OP_MI_WAIT_FOR_EVENT 0x3 +#define OP_MI_FLUSH 0x4 +#define OP_MI_ARB_CHECK 0x5 +#define OP_MI_RS_CONTROL 0x6 /* HSW+ */ +#define OP_MI_REPORT_HEAD 0x7 +#define OP_MI_ARB_ON_OFF 0x8 +#define OP_MI_URB_ATOMIC_ALLOC 0x9 /* HSW+ */ +#define OP_MI_BATCH_BUFFER_END 0xA +#define OP_MI_SUSPEND_FLUSH 0xB +#define OP_MI_PREDICATE 0xC /* IVB+ */ +#define OP_MI_TOPOLOGY_FILTER 0xD /* IVB+ */ +#define OP_MI_SET_APPID 0xE /* IVB+ */ +#define OP_MI_RS_CONTEXT 0xF /* HSW+ */ +#define OP_MI_LOAD_SCAN_LINES_INCL 0x12 /* HSW+ */ +#define OP_MI_DISPLAY_FLIP 0x14 +#define OP_MI_SEMAPHORE_MBOX 0x16 +#define OP_MI_SET_CONTEXT 0x18 +#define OP_MI_MATH 0x1A +#define OP_MI_URB_CLEAR 0x19 +#define OP_MI_SEMAPHORE_SIGNAL 0x1B /* BDW+ */ +#define OP_MI_SEMAPHORE_WAIT 0x1C /* BDW+ */ + +#define OP_MI_STORE_DATA_IMM 0x20 +#define OP_MI_STORE_DATA_INDEX 0x21 +#define OP_MI_LOAD_REGISTER_IMM 0x22 +#define OP_MI_UPDATE_GTT 0x23 +#define OP_MI_STORE_REGISTER_MEM 0x24 +#define OP_MI_FLUSH_DW 0x26 +#define OP_MI_CLFLUSH 0x27 +#define OP_MI_REPORT_PERF_COUNT 0x28 +#define OP_MI_LOAD_REGISTER_MEM 0x29 /* HSW+ */ +#define OP_MI_LOAD_REGISTER_REG 0x2A /* HSW+ */ +#define OP_MI_RS_STORE_DATA_IMM 0x2B /* HSW+ */ +#define OP_MI_LOAD_URB_MEM 0x2C /* HSW+ */ +#define OP_MI_STORE_URM_MEM 0x2D /* HSW+ */ +#define OP_MI_2E 0x2E /* BDW+ */ +#define OP_MI_2F 0x2F /* BDW+ */ +#define OP_MI_BATCH_BUFFER_START 0x31 + +/* Bit definition for dword 0 */ +#define _CMDBIT_BB_START_IN_PPGTT (1UL << 8) + +#define OP_MI_CONDITIONAL_BATCH_BUFFER_END 0x36 + +#define BATCH_BUFFER_ADDR_MASK ((1UL << 32) - (1U << 2)) +#define BATCH_BUFFER_ADDR_HIGH_MASK ((1UL << 16) - (1U)) +#define BATCH_BUFFER_ADR_SPACE_BIT(x) (((x) >> 8) & 1U) +#define BATCH_BUFFER_2ND_LEVEL_BIT(x) ((x) >> 22 & 1U) + +/* 2D command: Opcode (28:22) */ +#define OP_2D(x) ((2<<7) | x) + +#define OP_XY_SETUP_BLT OP_2D(0x1) +#define OP_XY_SETUP_CLIP_BLT OP_2D(0x3) +#define OP_XY_SETUP_MONO_PATTERN_SL_BLT OP_2D(0x11) +#define OP_XY_PIXEL_BLT OP_2D(0x24) +#define OP_XY_SCANLINES_BLT OP_2D(0x25) +#define OP_XY_TEXT_BLT OP_2D(0x26) +#define OP_XY_TEXT_IMMEDIATE_BLT OP_2D(0x31) +#define OP_XY_COLOR_BLT OP_2D(0x50) +#define OP_XY_PAT_BLT OP_2D(0x51) +#define OP_XY_MONO_PAT_BLT OP_2D(0x52) +#define OP_XY_SRC_COPY_BLT OP_2D(0x53) +#define OP_XY_MONO_SRC_COPY_BLT OP_2D(0x54) +#define OP_XY_FULL_BLT OP_2D(0x55) +#define OP_XY_FULL_MONO_SRC_BLT OP_2D(0x56) +#define OP_XY_FULL_MONO_PATTERN_BLT OP_2D(0x57) +#define OP_XY_FULL_MONO_PATTERN_MONO_SRC_BLT OP_2D(0x58) +#define OP_XY_MONO_PAT_FIXED_BLT OP_2D(0x59) +#define OP_XY_MONO_SRC_COPY_IMMEDIATE_BLT OP_2D(0x71) +#define OP_XY_PAT_BLT_IMMEDIATE OP_2D(0x72) +#define OP_XY_SRC_COPY_CHROMA_BLT OP_2D(0x73) +#define OP_XY_FULL_IMMEDIATE_PATTERN_BLT OP_2D(0x74) +#define OP_XY_FULL_MONO_SRC_IMMEDIATE_PATTERN_BLT OP_2D(0x75) +#define OP_XY_PAT_CHROMA_BLT OP_2D(0x76) +#define OP_XY_PAT_CHROMA_BLT_IMMEDIATE OP_2D(0x77) + +/* 3D/Media Command: Pipeline Type(28:27) Opcode(26:24) Sub Opcode(23:16) */ +#define OP_3D_MEDIA(sub_type, opcode, sub_opcode) \ + ((3 << 13) | ((sub_type) << 11) | ((opcode) << 8) | (sub_opcode)) + +#define OP_STATE_PREFETCH OP_3D_MEDIA(0x0, 0x0, 0x03) + +#define OP_STATE_BASE_ADDRESS OP_3D_MEDIA(0x0, 0x1, 0x01) +#define OP_STATE_SIP OP_3D_MEDIA(0x0, 0x1, 0x02) +#define OP_3D_MEDIA_0_1_4 OP_3D_MEDIA(0x0, 0x1, 0x04) + +#define OP_3DSTATE_VF_STATISTICS_GM45 OP_3D_MEDIA(0x1, 0x0, 0x0B) + +#define OP_PIPELINE_SELECT OP_3D_MEDIA(0x1, 0x1, 0x04) + +#define OP_MEDIA_VFE_STATE OP_3D_MEDIA(0x2, 0x0, 0x0) +#define OP_MEDIA_CURBE_LOAD OP_3D_MEDIA(0x2, 0x0, 0x1) +#define OP_MEDIA_INTERFACE_DESCRIPTOR_LOAD OP_3D_MEDIA(0x2, 0x0, 0x2) +#define OP_MEDIA_GATEWAY_STATE OP_3D_MEDIA(0x2, 0x0, 0x3) +#define OP_MEDIA_STATE_FLUSH OP_3D_MEDIA(0x2, 0x0, 0x4) + +#define OP_MEDIA_OBJECT OP_3D_MEDIA(0x2, 0x1, 0x0) +#define OP_MEDIA_OBJECT_PRT OP_3D_MEDIA(0x2, 0x1, 0x2) +#define OP_MEDIA_OBJECT_WALKER OP_3D_MEDIA(0x2, 0x1, 0x3) +#define OP_GPGPU_WALKER OP_3D_MEDIA(0x2, 0x1, 0x5) + +#define OP_3DSTATE_CLEAR_PARAMS OP_3D_MEDIA(0x3, 0x0, 0x04) /* IVB+ */ +#define OP_3DSTATE_DEPTH_BUFFER OP_3D_MEDIA(0x3, 0x0, 0x05) /* IVB+ */ +#define OP_3DSTATE_STENCIL_BUFFER OP_3D_MEDIA(0x3, 0x0, 0x06) /* IVB+ */ +#define OP_3DSTATE_HIER_DEPTH_BUFFER OP_3D_MEDIA(0x3, 0x0, 0x07) /* IVB+ */ +#define OP_3DSTATE_VERTEX_BUFFERS OP_3D_MEDIA(0x3, 0x0, 0x08) +#define OP_3DSTATE_VERTEX_ELEMENTS OP_3D_MEDIA(0x3, 0x0, 0x09) +#define OP_3DSTATE_INDEX_BUFFER OP_3D_MEDIA(0x3, 0x0, 0x0A) +#define OP_3DSTATE_VF_STATISTICS OP_3D_MEDIA(0x3, 0x0, 0x0B) +#define OP_3DSTATE_VF OP_3D_MEDIA(0x3, 0x0, 0x0C) /* HSW+ */ +#define OP_3DSTATE_CC_STATE_POINTERS OP_3D_MEDIA(0x3, 0x0, 0x0E) +#define OP_3DSTATE_SCISSOR_STATE_POINTERS OP_3D_MEDIA(0x3, 0x0, 0x0F) +#define OP_3DSTATE_VS OP_3D_MEDIA(0x3, 0x0, 0x10) +#define OP_3DSTATE_GS OP_3D_MEDIA(0x3, 0x0, 0x11) +#define OP_3DSTATE_CLIP OP_3D_MEDIA(0x3, 0x0, 0x12) +#define OP_3DSTATE_SF OP_3D_MEDIA(0x3, 0x0, 0x13) +#define OP_3DSTATE_WM OP_3D_MEDIA(0x3, 0x0, 0x14) +#define OP_3DSTATE_CONSTANT_VS OP_3D_MEDIA(0x3, 0x0, 0x15) +#define OP_3DSTATE_CONSTANT_GS OP_3D_MEDIA(0x3, 0x0, 0x16) +#define OP_3DSTATE_CONSTANT_PS OP_3D_MEDIA(0x3, 0x0, 0x17) +#define OP_3DSTATE_SAMPLE_MASK OP_3D_MEDIA(0x3, 0x0, 0x18) +#define OP_3DSTATE_CONSTANT_HS OP_3D_MEDIA(0x3, 0x0, 0x19) /* IVB+ */ +#define OP_3DSTATE_CONSTANT_DS OP_3D_MEDIA(0x3, 0x0, 0x1A) /* IVB+ */ +#define OP_3DSTATE_HS OP_3D_MEDIA(0x3, 0x0, 0x1B) /* IVB+ */ +#define OP_3DSTATE_TE OP_3D_MEDIA(0x3, 0x0, 0x1C) /* IVB+ */ +#define OP_3DSTATE_DS OP_3D_MEDIA(0x3, 0x0, 0x1D) /* IVB+ */ +#define OP_3DSTATE_STREAMOUT OP_3D_MEDIA(0x3, 0x0, 0x1E) /* IVB+ */ +#define OP_3DSTATE_SBE OP_3D_MEDIA(0x3, 0x0, 0x1F) /* IVB+ */ +#define OP_3DSTATE_PS OP_3D_MEDIA(0x3, 0x0, 0x20) /* IVB+ */ +#define OP_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP OP_3D_MEDIA(0x3, 0x0, 0x21) /* IVB+ */ +#define OP_3DSTATE_VIEWPORT_STATE_POINTERS_CC OP_3D_MEDIA(0x3, 0x0, 0x23) /* IVB+ */ +#define OP_3DSTATE_BLEND_STATE_POINTERS OP_3D_MEDIA(0x3, 0x0, 0x24) /* IVB+ */ +#define OP_3DSTATE_DEPTH_STENCIL_STATE_POINTERS OP_3D_MEDIA(0x3, 0x0, 0x25) /* IVB+ */ +#define OP_3DSTATE_BINDING_TABLE_POINTERS_VS OP_3D_MEDIA(0x3, 0x0, 0x26) /* IVB+ */ +#define OP_3DSTATE_BINDING_TABLE_POINTERS_HS OP_3D_MEDIA(0x3, 0x0, 0x27) /* IVB+ */ +#define OP_3DSTATE_BINDING_TABLE_POINTERS_DS OP_3D_MEDIA(0x3, 0x0, 0x28) /* IVB+ */ +#define OP_3DSTATE_BINDING_TABLE_POINTERS_GS OP_3D_MEDIA(0x3, 0x0, 0x29) /* IVB+ */ +#define OP_3DSTATE_BINDING_TABLE_POINTERS_PS OP_3D_MEDIA(0x3, 0x0, 0x2A) /* IVB+ */ +#define OP_3DSTATE_SAMPLER_STATE_POINTERS_VS OP_3D_MEDIA(0x3, 0x0, 0x2B) /* IVB+ */ +#define OP_3DSTATE_SAMPLER_STATE_POINTERS_HS OP_3D_MEDIA(0x3, 0x0, 0x2C) /* IVB+ */ +#define OP_3DSTATE_SAMPLER_STATE_POINTERS_DS OP_3D_MEDIA(0x3, 0x0, 0x2D) /* IVB+ */ +#define OP_3DSTATE_SAMPLER_STATE_POINTERS_GS OP_3D_MEDIA(0x3, 0x0, 0x2E) /* IVB+ */ +#define OP_3DSTATE_SAMPLER_STATE_POINTERS_PS OP_3D_MEDIA(0x3, 0x0, 0x2F) /* IVB+ */ +#define OP_3DSTATE_URB_VS OP_3D_MEDIA(0x3, 0x0, 0x30) /* IVB+ */ +#define OP_3DSTATE_URB_HS OP_3D_MEDIA(0x3, 0x0, 0x31) /* IVB+ */ +#define OP_3DSTATE_URB_DS OP_3D_MEDIA(0x3, 0x0, 0x32) /* IVB+ */ +#define OP_3DSTATE_URB_GS OP_3D_MEDIA(0x3, 0x0, 0x33) /* IVB+ */ +#define OP_3DSTATE_GATHER_CONSTANT_VS OP_3D_MEDIA(0x3, 0x0, 0x34) /* HSW+ */ +#define OP_3DSTATE_GATHER_CONSTANT_GS OP_3D_MEDIA(0x3, 0x0, 0x35) /* HSW+ */ +#define OP_3DSTATE_GATHER_CONSTANT_HS OP_3D_MEDIA(0x3, 0x0, 0x36) /* HSW+ */ +#define OP_3DSTATE_GATHER_CONSTANT_DS OP_3D_MEDIA(0x3, 0x0, 0x37) /* HSW+ */ +#define OP_3DSTATE_GATHER_CONSTANT_PS OP_3D_MEDIA(0x3, 0x0, 0x38) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTF_VS OP_3D_MEDIA(0x3, 0x0, 0x39) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTF_PS OP_3D_MEDIA(0x3, 0x0, 0x3A) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTI_VS OP_3D_MEDIA(0x3, 0x0, 0x3B) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTI_PS OP_3D_MEDIA(0x3, 0x0, 0x3C) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTB_VS OP_3D_MEDIA(0x3, 0x0, 0x3D) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANTB_PS OP_3D_MEDIA(0x3, 0x0, 0x3E) /* HSW+ */ +#define OP_3DSTATE_DX9_LOCAL_VALID_VS OP_3D_MEDIA(0x3, 0x0, 0x3F) /* HSW+ */ +#define OP_3DSTATE_DX9_LOCAL_VALID_PS OP_3D_MEDIA(0x3, 0x0, 0x40) /* HSW+ */ +#define OP_3DSTATE_DX9_GENERATE_ACTIVE_VS OP_3D_MEDIA(0x3, 0x0, 0x41) /* HSW+ */ +#define OP_3DSTATE_DX9_GENERATE_ACTIVE_PS OP_3D_MEDIA(0x3, 0x0, 0x42) /* HSW+ */ +#define OP_3DSTATE_BINDING_TABLE_EDIT_VS OP_3D_MEDIA(0x3, 0x0, 0x43) /* HSW+ */ +#define OP_3DSTATE_BINDING_TABLE_EDIT_GS OP_3D_MEDIA(0x3, 0x0, 0x44) /* HSW+ */ +#define OP_3DSTATE_BINDING_TABLE_EDIT_HS OP_3D_MEDIA(0x3, 0x0, 0x45) /* HSW+ */ +#define OP_3DSTATE_BINDING_TABLE_EDIT_DS OP_3D_MEDIA(0x3, 0x0, 0x46) /* HSW+ */ +#define OP_3DSTATE_BINDING_TABLE_EDIT_PS OP_3D_MEDIA(0x3, 0x0, 0x47) /* HSW+ */ + +#define OP_3DSTATE_VF_INSTANCING OP_3D_MEDIA(0x3, 0x0, 0x49) /* BDW+ */ +#define OP_3DSTATE_VF_SGVS OP_3D_MEDIA(0x3, 0x0, 0x4A) /* BDW+ */ +#define OP_3DSTATE_VF_TOPOLOGY OP_3D_MEDIA(0x3, 0x0, 0x4B) /* BDW+ */ +#define OP_3DSTATE_WM_CHROMAKEY OP_3D_MEDIA(0x3, 0x0, 0x4C) /* BDW+ */ +#define OP_3DSTATE_PS_BLEND OP_3D_MEDIA(0x3, 0x0, 0x4D) /* BDW+ */ +#define OP_3DSTATE_WM_DEPTH_STENCIL OP_3D_MEDIA(0x3, 0x0, 0x4E) /* BDW+ */ +#define OP_3DSTATE_PS_EXTRA OP_3D_MEDIA(0x3, 0x0, 0x4F) /* BDW+ */ +#define OP_3DSTATE_RASTER OP_3D_MEDIA(0x3, 0x0, 0x50) /* BDW+ */ +#define OP_3DSTATE_SBE_SWIZ OP_3D_MEDIA(0x3, 0x0, 0x51) /* BDW+ */ +#define OP_3DSTATE_WM_HZ_OP OP_3D_MEDIA(0x3, 0x0, 0x52) /* BDW+ */ +#define OP_3DSTATE_COMPONENT_PACKING OP_3D_MEDIA(0x3, 0x0, 0x55) /* SKL+ */ + +#define OP_3DSTATE_DRAWING_RECTANGLE OP_3D_MEDIA(0x3, 0x1, 0x00) +#define OP_3DSTATE_SAMPLER_PALETTE_LOAD0 OP_3D_MEDIA(0x3, 0x1, 0x02) +#define OP_3DSTATE_CHROMA_KEY OP_3D_MEDIA(0x3, 0x1, 0x04) +#define OP_SNB_3DSTATE_DEPTH_BUFFER OP_3D_MEDIA(0x3, 0x1, 0x05) +#define OP_3DSTATE_POLY_STIPPLE_OFFSET OP_3D_MEDIA(0x3, 0x1, 0x06) +#define OP_3DSTATE_POLY_STIPPLE_PATTERN OP_3D_MEDIA(0x3, 0x1, 0x07) +#define OP_3DSTATE_LINE_STIPPLE OP_3D_MEDIA(0x3, 0x1, 0x08) +#define OP_3DSTATE_AA_LINE_PARAMS OP_3D_MEDIA(0x3, 0x1, 0x0A) +#define OP_3DSTATE_GS_SVB_INDEX OP_3D_MEDIA(0x3, 0x1, 0x0B) +#define OP_3DSTATE_SAMPLER_PALETTE_LOAD1 OP_3D_MEDIA(0x3, 0x1, 0x0C) +#define OP_3DSTATE_MULTISAMPLE_BDW OP_3D_MEDIA(0x3, 0x0, 0x0D) +#define OP_SNB_3DSTATE_STENCIL_BUFFER OP_3D_MEDIA(0x3, 0x1, 0x0E) +#define OP_SNB_3DSTATE_HIER_DEPTH_BUFFER OP_3D_MEDIA(0x3, 0x1, 0x0F) +#define OP_SNB_3DSTATE_CLEAR_PARAMS OP_3D_MEDIA(0x3, 0x1, 0x10) +#define OP_3DSTATE_MONOFILTER_SIZE OP_3D_MEDIA(0x3, 0x1, 0x11) +#define OP_3DSTATE_PUSH_CONSTANT_ALLOC_VS OP_3D_MEDIA(0x3, 0x1, 0x12) /* IVB+ */ +#define OP_3DSTATE_PUSH_CONSTANT_ALLOC_HS OP_3D_MEDIA(0x3, 0x1, 0x13) /* IVB+ */ +#define OP_3DSTATE_PUSH_CONSTANT_ALLOC_DS OP_3D_MEDIA(0x3, 0x1, 0x14) /* IVB+ */ +#define OP_3DSTATE_PUSH_CONSTANT_ALLOC_GS OP_3D_MEDIA(0x3, 0x1, 0x15) /* IVB+ */ +#define OP_3DSTATE_PUSH_CONSTANT_ALLOC_PS OP_3D_MEDIA(0x3, 0x1, 0x16) /* IVB+ */ +#define OP_3DSTATE_SO_DECL_LIST OP_3D_MEDIA(0x3, 0x1, 0x17) +#define OP_3DSTATE_SO_BUFFER OP_3D_MEDIA(0x3, 0x1, 0x18) +#define OP_3DSTATE_BINDING_TABLE_POOL_ALLOC OP_3D_MEDIA(0x3, 0x1, 0x19) /* HSW+ */ +#define OP_3DSTATE_GATHER_POOL_ALLOC OP_3D_MEDIA(0x3, 0x1, 0x1A) /* HSW+ */ +#define OP_3DSTATE_DX9_CONSTANT_BUFFER_POOL_ALLOC OP_3D_MEDIA(0x3, 0x1, 0x1B) /* HSW+ */ +#define OP_3DSTATE_SAMPLE_PATTERN OP_3D_MEDIA(0x3, 0x1, 0x1C) +#define OP_PIPE_CONTROL OP_3D_MEDIA(0x3, 0x2, 0x00) +#define OP_3DPRIMITIVE OP_3D_MEDIA(0x3, 0x3, 0x00) + +/* VCCP Command Parser */ + +/* + * Below MFX and VBE cmd definition is from vaapi intel driver project (BSD License) + * git://anongit.freedesktop.org/vaapi/intel-driver + * src/i965_defines.h + * + */ + +#define OP_MFX(pipeline, op, sub_opa, sub_opb) \ + (3 << 13 | \ + (pipeline) << 11 | \ + (op) << 8 | \ + (sub_opa) << 5 | \ + (sub_opb)) + +#define OP_MFX_PIPE_MODE_SELECT OP_MFX(2, 0, 0, 0) /* ALL */ +#define OP_MFX_SURFACE_STATE OP_MFX(2, 0, 0, 1) /* ALL */ +#define OP_MFX_PIPE_BUF_ADDR_STATE OP_MFX(2, 0, 0, 2) /* ALL */ +#define OP_MFX_IND_OBJ_BASE_ADDR_STATE OP_MFX(2, 0, 0, 3) /* ALL */ +#define OP_MFX_BSP_BUF_BASE_ADDR_STATE OP_MFX(2, 0, 0, 4) /* ALL */ +#define OP_2_0_0_5 OP_MFX(2, 0, 0, 5) /* ALL */ +#define OP_MFX_STATE_POINTER OP_MFX(2, 0, 0, 6) /* ALL */ +#define OP_MFX_QM_STATE OP_MFX(2, 0, 0, 7) /* IVB+ */ +#define OP_MFX_FQM_STATE OP_MFX(2, 0, 0, 8) /* IVB+ */ +#define OP_MFX_PAK_INSERT_OBJECT OP_MFX(2, 0, 2, 8) /* IVB+ */ +#define OP_MFX_STITCH_OBJECT OP_MFX(2, 0, 2, 0xA) /* IVB+ */ + +#define OP_MFD_IT_OBJECT OP_MFX(2, 0, 1, 9) /* ALL */ + +#define OP_MFX_WAIT OP_MFX(1, 0, 0, 0) /* IVB+ */ +#define OP_MFX_AVC_IMG_STATE OP_MFX(2, 1, 0, 0) /* ALL */ +#define OP_MFX_AVC_QM_STATE OP_MFX(2, 1, 0, 1) /* ALL */ +#define OP_MFX_AVC_DIRECTMODE_STATE OP_MFX(2, 1, 0, 2) /* ALL */ +#define OP_MFX_AVC_SLICE_STATE OP_MFX(2, 1, 0, 3) /* ALL */ +#define OP_MFX_AVC_REF_IDX_STATE OP_MFX(2, 1, 0, 4) /* ALL */ +#define OP_MFX_AVC_WEIGHTOFFSET_STATE OP_MFX(2, 1, 0, 5) /* ALL */ +#define OP_MFD_AVC_PICID_STATE OP_MFX(2, 1, 1, 5) /* HSW+ */ +#define OP_MFD_AVC_DPB_STATE OP_MFX(2, 1, 1, 6) /* IVB+ */ +#define OP_MFD_AVC_SLICEADDR OP_MFX(2, 1, 1, 7) /* IVB+ */ +#define OP_MFD_AVC_BSD_OBJECT OP_MFX(2, 1, 1, 8) /* ALL */ +#define OP_MFC_AVC_PAK_OBJECT OP_MFX(2, 1, 2, 9) /* ALL */ + +#define OP_MFX_VC1_PRED_PIPE_STATE OP_MFX(2, 2, 0, 1) /* ALL */ +#define OP_MFX_VC1_DIRECTMODE_STATE OP_MFX(2, 2, 0, 2) /* ALL */ +#define OP_MFD_VC1_SHORT_PIC_STATE OP_MFX(2, 2, 1, 0) /* IVB+ */ +#define OP_MFD_VC1_LONG_PIC_STATE OP_MFX(2, 2, 1, 1) /* IVB+ */ +#define OP_MFD_VC1_BSD_OBJECT OP_MFX(2, 2, 1, 8) /* ALL */ + +#define OP_MFX_MPEG2_PIC_STATE OP_MFX(2, 3, 0, 0) /* ALL */ +#define OP_MFX_MPEG2_QM_STATE OP_MFX(2, 3, 0, 1) /* ALL */ +#define OP_MFD_MPEG2_BSD_OBJECT OP_MFX(2, 3, 1, 8) /* ALL */ +#define OP_MFC_MPEG2_SLICEGROUP_STATE OP_MFX(2, 3, 2, 3) /* ALL */ +#define OP_MFC_MPEG2_PAK_OBJECT OP_MFX(2, 3, 2, 9) /* ALL */ + +#define OP_MFX_2_6_0_0 OP_MFX(2, 6, 0, 0) /* IVB+ */ +#define OP_MFX_2_6_0_8 OP_MFX(2, 6, 0, 8) /* IVB+ */ +#define OP_MFX_2_6_0_9 OP_MFX(2, 6, 0, 9) /* IVB+ */ + +#define OP_MFX_JPEG_PIC_STATE OP_MFX(2, 7, 0, 0) +#define OP_MFX_JPEG_HUFF_TABLE_STATE OP_MFX(2, 7, 0, 2) +#define OP_MFD_JPEG_BSD_OBJECT OP_MFX(2, 7, 1, 8) + +#define OP_VEB(pipeline, op, sub_opa, sub_opb) \ + (3 << 13 | \ + (pipeline) << 11 | \ + (op) << 8 | \ + (sub_opa) << 5 | \ + (sub_opb)) + +#define OP_VEB_SURFACE_STATE OP_VEB(2, 4, 0, 0) +#define OP_VEB_STATE OP_VEB(2, 4, 0, 2) +#define OP_VEB_DNDI_IECP_STATE OP_VEB(2, 4, 0, 3) + +struct parser_exec_state; + +typedef int (*parser_cmd_handler)(struct parser_exec_state *s); + +#define GVT_CMD_HASH_BITS 7 + +/* which DWords need address fix */ +#define ADDR_FIX_1(x1) (1 << (x1)) +#define ADDR_FIX_2(x1, x2) (ADDR_FIX_1(x1) | ADDR_FIX_1(x2)) +#define ADDR_FIX_3(x1, x2, x3) (ADDR_FIX_1(x1) | ADDR_FIX_2(x2, x3)) +#define ADDR_FIX_4(x1, x2, x3, x4) (ADDR_FIX_1(x1) | ADDR_FIX_3(x2, x3, x4)) +#define ADDR_FIX_5(x1, x2, x3, x4, x5) (ADDR_FIX_1(x1) | ADDR_FIX_4(x2, x3, x4, x5)) + +struct cmd_info { + char *name; + u32 opcode; + +#define F_LEN_MASK (1U<<0) +#define F_LEN_CONST 1U +#define F_LEN_VAR 0U + +/* + * command has its own ip advance logic + * e.g. MI_BATCH_START, MI_BATCH_END + */ +#define F_IP_ADVANCE_CUSTOM (1<<1) + +#define F_POST_HANDLE (1<<2) + u32 flag; + +#define R_RCS (1 << RCS) +#define R_VCS1 (1 << VCS) +#define R_VCS2 (1 << VCS2) +#define R_VCS (R_VCS1 | R_VCS2) +#define R_BCS (1 << BCS) +#define R_VECS (1 << VECS) +#define R_ALL (R_RCS | R_VCS | R_BCS | R_VECS) + /* rings that support this cmd: BLT/RCS/VCS/VECS */ + uint16_t rings; + + /* devices that support this cmd: SNB/IVB/HSW/... */ + uint16_t devices; + + /* which DWords are address that need fix up. + * bit 0 means a 32-bit non address operand in command + * bit 1 means address operand, which could be 32-bit + * or 64-bit depending on different architectures.( + * defined by "gmadr_bytes_in_cmd" in intel_gvt. + * No matter the address length, each address only takes + * one bit in the bitmap. + */ + uint16_t addr_bitmap; + + /* flag == F_LEN_CONST : command length + * flag == F_LEN_VAR : length bias bits + * Note: length is in DWord + */ + uint8_t len; + + parser_cmd_handler handler; +}; + +struct cmd_entry { + struct hlist_node hlist; + struct cmd_info *info; +}; + +enum { + RING_BUFFER_INSTRUCTION, + BATCH_BUFFER_INSTRUCTION, + BATCH_BUFFER_2ND_LEVEL, +}; + +enum { + GTT_BUFFER, + PPGTT_BUFFER +}; + +struct parser_exec_state { + struct intel_vgpu *vgpu; + int ring_id; + + int buf_type; + + /* batch buffer address type */ + int buf_addr_type; + + /* graphics memory address of ring buffer start */ + unsigned long ring_start; + unsigned long ring_size; + unsigned long ring_head; + unsigned long ring_tail; + + /* instruction graphics memory address */ + unsigned long ip_gma; + + /* mapped va of the instr_gma */ + void *ip_va; + void *rb_va; + + void *ret_bb_va; + /* next instruction when return from batch buffer to ring buffer */ + unsigned long ret_ip_gma_ring; + + /* next instruction when return from 2nd batch buffer to batch buffer */ + unsigned long ret_ip_gma_bb; + + /* batch buffer address type (GTT or PPGTT) + * used when ret from 2nd level batch buffer + */ + int saved_buf_addr_type; + + struct cmd_info *info; + + struct intel_vgpu_workload *workload; +}; + +#define gmadr_dw_number(s) \ + (s->vgpu->gvt->device_info.gmadr_bytes_in_cmd >> 2) + +static unsigned long bypass_scan_mask = 0; +static bool bypass_batch_buffer_scan = true; + +/* ring ALL, type = 0 */ +static struct sub_op_bits sub_op_mi[] = { + {31, 29}, + {28, 23}, +}; + +static struct decode_info decode_info_mi = { + "MI", + OP_LEN_MI, + ARRAY_SIZE(sub_op_mi), + sub_op_mi, +}; + +/* ring RCS, command type 2 */ +static struct sub_op_bits sub_op_2d[] = { + {31, 29}, + {28, 22}, +}; + +static struct decode_info decode_info_2d = { + "2D", + OP_LEN_2D, + ARRAY_SIZE(sub_op_2d), + sub_op_2d, +}; + +/* ring RCS, command type 3 */ +static struct sub_op_bits sub_op_3d_media[] = { + {31, 29}, + {28, 27}, + {26, 24}, + {23, 16}, +}; + +static struct decode_info decode_info_3d_media = { + "3D_Media", + OP_LEN_3D_MEDIA, + ARRAY_SIZE(sub_op_3d_media), + sub_op_3d_media, +}; + +/* ring VCS, command type 3 */ +static struct sub_op_bits sub_op_mfx_vc[] = { + {31, 29}, + {28, 27}, + {26, 24}, + {23, 21}, + {20, 16}, +}; + +static struct decode_info decode_info_mfx_vc = { + "MFX_VC", + OP_LEN_MFX_VC, + ARRAY_SIZE(sub_op_mfx_vc), + sub_op_mfx_vc, +}; + +/* ring VECS, command type 3 */ +static struct sub_op_bits sub_op_vebox[] = { + {31, 29}, + {28, 27}, + {26, 24}, + {23, 21}, + {20, 16}, +}; + +static struct decode_info decode_info_vebox = { + "VEBOX", + OP_LEN_VEBOX, + ARRAY_SIZE(sub_op_vebox), + sub_op_vebox, +}; + +static struct decode_info *ring_decode_info[I915_NUM_ENGINES][8] = { + [RCS] = { + &decode_info_mi, + NULL, + NULL, + &decode_info_3d_media, + NULL, + NULL, + NULL, + NULL, + }, + + [VCS] = { + &decode_info_mi, + NULL, + NULL, + &decode_info_mfx_vc, + NULL, + NULL, + NULL, + NULL, + }, + + [BCS] = { + &decode_info_mi, + NULL, + &decode_info_2d, + NULL, + NULL, + NULL, + NULL, + NULL, + }, + + [VECS] = { + &decode_info_mi, + NULL, + NULL, + &decode_info_vebox, + NULL, + NULL, + NULL, + NULL, + }, + + [VCS2] = { + &decode_info_mi, + NULL, + NULL, + &decode_info_mfx_vc, + NULL, + NULL, + NULL, + NULL, + }, +}; + +static inline u32 get_opcode(u32 cmd, int ring_id) +{ + struct decode_info *d_info; + + if (ring_id >= I915_NUM_ENGINES) + return INVALID_OP; + + d_info = ring_decode_info[ring_id][CMD_TYPE(cmd)]; + if (d_info == NULL) + return INVALID_OP; + + return cmd >> (32 - d_info->op_len); +} + +static inline struct cmd_info *find_cmd_entry(struct intel_gvt *gvt, + unsigned int opcode, int ring_id) +{ + struct cmd_entry *e; + + hash_for_each_possible(gvt->cmd_table, e, hlist, opcode) { + if ((opcode == e->info->opcode) && + (e->info->rings & (1 << ring_id))) + return e->info; + } + return NULL; +} + +static inline struct cmd_info *get_cmd_info(struct intel_gvt *gvt, + u32 cmd, int ring_id) +{ + u32 opcode; + + opcode = get_opcode(cmd, ring_id); + if (opcode == INVALID_OP) + return NULL; + + return find_cmd_entry(gvt, opcode, ring_id); +} + +static inline u32 sub_op_val(u32 cmd, u32 hi, u32 low) +{ + return (cmd >> low) & ((1U << (hi - low + 1)) - 1); +} + +static inline void print_opcode(u32 cmd, int ring_id) +{ + struct decode_info *d_info; + int i; + + if (ring_id >= I915_NUM_ENGINES) + return; + + d_info = ring_decode_info[ring_id][CMD_TYPE(cmd)]; + if (d_info == NULL) + return; + + gvt_err("opcode=0x%x %s sub_ops:", + cmd >> (32 - d_info->op_len), d_info->name); + + for (i = 0; i < d_info->nr_sub_op; i++) + pr_err("0x%x ", sub_op_val(cmd, d_info->sub_op[i].hi, + d_info->sub_op[i].low)); + + pr_err("\n"); +} + +static inline u32 *cmd_ptr(struct parser_exec_state *s, int index) +{ + return s->ip_va + (index << 2); +} + +static inline u32 cmd_val(struct parser_exec_state *s, int index) +{ + return *cmd_ptr(s, index); +} + +static void parser_exec_state_dump(struct parser_exec_state *s) +{ + int cnt = 0; + int i; + + gvt_err(" vgpu%d RING%d: ring_start(%08lx) ring_end(%08lx)" + " ring_head(%08lx) ring_tail(%08lx)\n", s->vgpu->id, + s->ring_id, s->ring_start, s->ring_start + s->ring_size, + s->ring_head, s->ring_tail); + + gvt_err(" %s %s ip_gma(%08lx) ", + s->buf_type == RING_BUFFER_INSTRUCTION ? + "RING_BUFFER" : "BATCH_BUFFER", + s->buf_addr_type == GTT_BUFFER ? + "GTT" : "PPGTT", s->ip_gma); + + if (s->ip_va == NULL) { + gvt_err(" ip_va(NULL)"); + return; + } + + gvt_err(" ip_va=%p: %08x %08x %08x %08x\n", + s->ip_va, cmd_val(s, 0), cmd_val(s, 1), + cmd_val(s, 2), cmd_val(s, 3)); + + print_opcode(cmd_val(s, 0), s->ring_id); + + /* print the whole page to trace */ + pr_err(" ip_va=%p: %08x %08x %08x %08x\n", + s->ip_va, cmd_val(s, 0), cmd_val(s, 1), + cmd_val(s, 2), cmd_val(s, 3)); + + s->ip_va = (u32 *)((((u64)s->ip_va) >> 12) << 12); + + while (cnt < 1024) { + pr_err("ip_va=%p: ", s->ip_va); + for (i = 0; i < 8; i++) + pr_err("%08x ", cmd_val(s, i)); + pr_err("\n"); + + s->ip_va += 8 * sizeof(u32); + cnt += 8; + } +} + +static inline void update_ip_va(struct parser_exec_state *s) +{ + unsigned long len = 0; + + if (WARN_ON(s->ring_head == s->ring_tail)) + return; + + if (s->buf_type == RING_BUFFER_INSTRUCTION) { + unsigned long ring_top = s->ring_start + s->ring_size; + + if (s->ring_head > s->ring_tail) { + if (s->ip_gma >= s->ring_head && s->ip_gma < ring_top) + len = (s->ip_gma - s->ring_head); + else if (s->ip_gma >= s->ring_start && + s->ip_gma <= s->ring_tail) + len = (ring_top - s->ring_head) + + (s->ip_gma - s->ring_start); + } else + len = (s->ip_gma - s->ring_head); + + s->ip_va = s->rb_va + len; + } else {/* shadow batch buffer */ + s->ip_va = s->ret_bb_va; + } +} + +static inline int ip_gma_set(struct parser_exec_state *s, + unsigned long ip_gma) +{ + WARN_ON(!IS_ALIGNED(ip_gma, 4)); + + s->ip_gma = ip_gma; + update_ip_va(s); + return 0; +} + +static inline int ip_gma_advance(struct parser_exec_state *s, + unsigned int dw_len) +{ + s->ip_gma += (dw_len << 2); + + if (s->buf_type == RING_BUFFER_INSTRUCTION) { + if (s->ip_gma >= s->ring_start + s->ring_size) + s->ip_gma -= s->ring_size; + update_ip_va(s); + } else { + s->ip_va += (dw_len << 2); + } + + return 0; +} + +static inline int get_cmd_length(struct cmd_info *info, u32 cmd) +{ + if ((info->flag & F_LEN_MASK) == F_LEN_CONST) + return info->len; + else + return (cmd & ((1U << info->len) - 1)) + 2; + return 0; +} + +static inline int cmd_length(struct parser_exec_state *s) +{ + return get_cmd_length(s->info, cmd_val(s, 0)); +} + +/* do not remove this, some platform may need clflush here */ +#define patch_value(s, addr, val) do { \ + *addr = val; \ +} while (0) + +static bool is_shadowed_mmio(unsigned int offset) +{ + bool ret = false; + + if ((offset == 0x2168) || /*BB current head register UDW */ + (offset == 0x2140) || /*BB current header register */ + (offset == 0x211c) || /*second BB header register UDW */ + (offset == 0x2114)) { /*second BB header register UDW */ + ret = true; + } + return ret; +} + +static int cmd_reg_handler(struct parser_exec_state *s, + unsigned int offset, unsigned int index, char *cmd) +{ + struct intel_vgpu *vgpu = s->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + + if (offset + 4 > gvt->device_info.mmio_size) { + gvt_err("%s access to (%x) outside of MMIO range\n", + cmd, offset); + return -EINVAL; + } + + if (!intel_gvt_mmio_is_cmd_access(gvt, offset)) { + gvt_err("vgpu%d: %s access to non-render register (%x)\n", + s->vgpu->id, cmd, offset); + return 0; + } + + if (is_shadowed_mmio(offset)) { + gvt_err("vgpu%d: found access of shadowed MMIO %x\n", + s->vgpu->id, offset); + return 0; + } + + if (offset == i915_mmio_reg_offset(DERRMR) || + offset == i915_mmio_reg_offset(FORCEWAKE_MT)) { + /* Writing to HW VGT_PVINFO_PAGE offset will be discarded */ + patch_value(s, cmd_ptr(s, index), VGT_PVINFO_PAGE); + } + + /* TODO: Update the global mask if this MMIO is a masked-MMIO */ + intel_gvt_mmio_set_cmd_accessed(gvt, offset); + return 0; +} + +#define cmd_reg(s, i) \ + (cmd_val(s, i) & GENMASK(22, 2)) + +#define cmd_reg_inhibit(s, i) \ + (cmd_val(s, i) & GENMASK(22, 18)) + +#define cmd_gma(s, i) \ + (cmd_val(s, i) & GENMASK(31, 2)) + +#define cmd_gma_hi(s, i) \ + (cmd_val(s, i) & GENMASK(15, 0)) + +static int cmd_handler_lri(struct parser_exec_state *s) +{ + int i, ret = 0; + int cmd_len = cmd_length(s); + struct intel_gvt *gvt = s->vgpu->gvt; + + for (i = 1; i < cmd_len; i += 2) { + if (IS_BROADWELL(gvt->dev_priv) && + (s->ring_id != RCS)) { + if (s->ring_id == BCS && + cmd_reg(s, i) == + i915_mmio_reg_offset(DERRMR)) + ret |= 0; + else + ret |= (cmd_reg_inhibit(s, i)) ? -EINVAL : 0; + } + if (ret) + break; + ret |= cmd_reg_handler(s, cmd_reg(s, i), i, "lri"); + } + return ret; +} + +static int cmd_handler_lrr(struct parser_exec_state *s) +{ + int i, ret = 0; + int cmd_len = cmd_length(s); + + for (i = 1; i < cmd_len; i += 2) { + if (IS_BROADWELL(s->vgpu->gvt->dev_priv)) + ret |= ((cmd_reg_inhibit(s, i) || + (cmd_reg_inhibit(s, i + 1)))) ? + -EINVAL : 0; + if (ret) + break; + ret |= cmd_reg_handler(s, cmd_reg(s, i), i, "lrr-src"); + ret |= cmd_reg_handler(s, cmd_reg(s, i + 1), i, "lrr-dst"); + } + return ret; +} + +static inline int cmd_address_audit(struct parser_exec_state *s, + unsigned long guest_gma, int op_size, bool index_mode); + +static int cmd_handler_lrm(struct parser_exec_state *s) +{ + struct intel_gvt *gvt = s->vgpu->gvt; + int gmadr_bytes = gvt->device_info.gmadr_bytes_in_cmd; + unsigned long gma; + int i, ret = 0; + int cmd_len = cmd_length(s); + + for (i = 1; i < cmd_len;) { + if (IS_BROADWELL(gvt->dev_priv)) + ret |= (cmd_reg_inhibit(s, i)) ? -EINVAL : 0; + if (ret) + break; + ret |= cmd_reg_handler(s, cmd_reg(s, i), i, "lrm"); + if (cmd_val(s, 0) & (1 << 22)) { + gma = cmd_gma(s, i + 1); + if (gmadr_bytes == 8) + gma |= (cmd_gma_hi(s, i + 2)) << 32; + ret |= cmd_address_audit(s, gma, sizeof(u32), false); + } + i += gmadr_dw_number(s) + 1; + } + return ret; +} + +static int cmd_handler_srm(struct parser_exec_state *s) +{ + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + unsigned long gma; + int i, ret = 0; + int cmd_len = cmd_length(s); + + for (i = 1; i < cmd_len;) { + ret |= cmd_reg_handler(s, cmd_reg(s, i), i, "srm"); + if (cmd_val(s, 0) & (1 << 22)) { + gma = cmd_gma(s, i + 1); + if (gmadr_bytes == 8) + gma |= (cmd_gma_hi(s, i + 2)) << 32; + ret |= cmd_address_audit(s, gma, sizeof(u32), false); + } + i += gmadr_dw_number(s) + 1; + } + return ret; +} + +struct cmd_interrupt_event { + int pipe_control_notify; + int mi_flush_dw; + int mi_user_interrupt; +}; + +static struct cmd_interrupt_event cmd_interrupt_events[] = { + [RCS] = { + .pipe_control_notify = RCS_PIPE_CONTROL, + .mi_flush_dw = INTEL_GVT_EVENT_RESERVED, + .mi_user_interrupt = RCS_MI_USER_INTERRUPT, + }, + [BCS] = { + .pipe_control_notify = INTEL_GVT_EVENT_RESERVED, + .mi_flush_dw = BCS_MI_FLUSH_DW, + .mi_user_interrupt = BCS_MI_USER_INTERRUPT, + }, + [VCS] = { + .pipe_control_notify = INTEL_GVT_EVENT_RESERVED, + .mi_flush_dw = VCS_MI_FLUSH_DW, + .mi_user_interrupt = VCS_MI_USER_INTERRUPT, + }, + [VCS2] = { + .pipe_control_notify = INTEL_GVT_EVENT_RESERVED, + .mi_flush_dw = VCS2_MI_FLUSH_DW, + .mi_user_interrupt = VCS2_MI_USER_INTERRUPT, + }, + [VECS] = { + .pipe_control_notify = INTEL_GVT_EVENT_RESERVED, + .mi_flush_dw = VECS_MI_FLUSH_DW, + .mi_user_interrupt = VECS_MI_USER_INTERRUPT, + }, +}; + +static int cmd_handler_pipe_control(struct parser_exec_state *s) +{ + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + unsigned long gma; + bool index_mode = false; + unsigned int post_sync; + int ret = 0; + + post_sync = (cmd_val(s, 1) & PIPE_CONTROL_POST_SYNC_OP_MASK) >> 14; + + /* LRI post sync */ + if (cmd_val(s, 1) & PIPE_CONTROL_MMIO_WRITE) + ret = cmd_reg_handler(s, cmd_reg(s, 2), 1, "pipe_ctrl"); + /* post sync */ + else if (post_sync) { + if (post_sync == 2) + ret = cmd_reg_handler(s, 0x2350, 1, "pipe_ctrl"); + else if (post_sync == 3) + ret = cmd_reg_handler(s, 0x2358, 1, "pipe_ctrl"); + else if (post_sync == 1) { + /* check ggtt*/ + if ((cmd_val(s, 2) & (1 << 2))) { + gma = cmd_val(s, 2) & GENMASK(31, 3); + if (gmadr_bytes == 8) + gma |= (cmd_gma_hi(s, 3)) << 32; + /* Store Data Index */ + if (cmd_val(s, 1) & (1 << 21)) + index_mode = true; + ret |= cmd_address_audit(s, gma, sizeof(u64), + index_mode); + } + } + } + + if (ret) + return ret; + + if (cmd_val(s, 1) & PIPE_CONTROL_NOTIFY) + set_bit(cmd_interrupt_events[s->ring_id].pipe_control_notify, + s->workload->pending_events); + return 0; +} + +static int cmd_handler_mi_user_interrupt(struct parser_exec_state *s) +{ + set_bit(cmd_interrupt_events[s->ring_id].mi_user_interrupt, + s->workload->pending_events); + return 0; +} + +static int cmd_advance_default(struct parser_exec_state *s) +{ + return ip_gma_advance(s, cmd_length(s)); +} + +static int cmd_handler_mi_batch_buffer_end(struct parser_exec_state *s) +{ + int ret; + + if (s->buf_type == BATCH_BUFFER_2ND_LEVEL) { + s->buf_type = BATCH_BUFFER_INSTRUCTION; + ret = ip_gma_set(s, s->ret_ip_gma_bb); + s->buf_addr_type = s->saved_buf_addr_type; + } else { + s->buf_type = RING_BUFFER_INSTRUCTION; + s->buf_addr_type = GTT_BUFFER; + if (s->ret_ip_gma_ring >= s->ring_start + s->ring_size) + s->ret_ip_gma_ring -= s->ring_size; + ret = ip_gma_set(s, s->ret_ip_gma_ring); + } + return ret; +} + +struct mi_display_flip_command_info { + int pipe; + int plane; + int event; + i915_reg_t stride_reg; + i915_reg_t ctrl_reg; + i915_reg_t surf_reg; + u64 stride_val; + u64 tile_val; + u64 surf_val; + bool async_flip; +}; + +struct plane_code_mapping { + int pipe; + int plane; + int event; +}; + +static int gen8_decode_mi_display_flip(struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + struct plane_code_mapping gen8_plane_code[] = { + [0] = {PIPE_A, PLANE_A, PRIMARY_A_FLIP_DONE}, + [1] = {PIPE_B, PLANE_A, PRIMARY_B_FLIP_DONE}, + [2] = {PIPE_A, PLANE_B, SPRITE_A_FLIP_DONE}, + [3] = {PIPE_B, PLANE_B, SPRITE_B_FLIP_DONE}, + [4] = {PIPE_C, PLANE_A, PRIMARY_C_FLIP_DONE}, + [5] = {PIPE_C, PLANE_B, SPRITE_C_FLIP_DONE}, + }; + u32 dword0, dword1, dword2; + u32 v; + + dword0 = cmd_val(s, 0); + dword1 = cmd_val(s, 1); + dword2 = cmd_val(s, 2); + + v = (dword0 & GENMASK(21, 19)) >> 19; + if (WARN_ON(v >= ARRAY_SIZE(gen8_plane_code))) + return -EINVAL; + + info->pipe = gen8_plane_code[v].pipe; + info->plane = gen8_plane_code[v].plane; + info->event = gen8_plane_code[v].event; + info->stride_val = (dword1 & GENMASK(15, 6)) >> 6; + info->tile_val = (dword1 & 0x1); + info->surf_val = (dword2 & GENMASK(31, 12)) >> 12; + info->async_flip = ((dword2 & GENMASK(1, 0)) == 0x1); + + if (info->plane == PLANE_A) { + info->ctrl_reg = DSPCNTR(info->pipe); + info->stride_reg = DSPSTRIDE(info->pipe); + info->surf_reg = DSPSURF(info->pipe); + } else if (info->plane == PLANE_B) { + info->ctrl_reg = SPRCTL(info->pipe); + info->stride_reg = SPRSTRIDE(info->pipe); + info->surf_reg = SPRSURF(info->pipe); + } else { + WARN_ON(1); + return -EINVAL; + } + return 0; +} + +static int skl_decode_mi_display_flip(struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + u32 dword0 = cmd_val(s, 0); + u32 dword1 = cmd_val(s, 1); + u32 dword2 = cmd_val(s, 2); + u32 plane = (dword0 & GENMASK(12, 8)) >> 8; + + switch (plane) { + case MI_DISPLAY_FLIP_SKL_PLANE_1_A: + info->pipe = PIPE_A; + info->event = PRIMARY_A_FLIP_DONE; + break; + case MI_DISPLAY_FLIP_SKL_PLANE_1_B: + info->pipe = PIPE_B; + info->event = PRIMARY_B_FLIP_DONE; + break; + case MI_DISPLAY_FLIP_SKL_PLANE_1_C: + info->pipe = PIPE_C; + info->event = PRIMARY_C_FLIP_DONE; + break; + default: + gvt_err("unknown plane code %d\n", plane); + return -EINVAL; + } + + info->pipe = PRIMARY_PLANE; + info->stride_val = (dword1 & GENMASK(15, 6)) >> 6; + info->tile_val = (dword1 & GENMASK(2, 0)); + info->surf_val = (dword2 & GENMASK(31, 12)) >> 12; + info->async_flip = ((dword2 & GENMASK(1, 0)) == 0x1); + + info->ctrl_reg = DSPCNTR(info->pipe); + info->stride_reg = DSPSTRIDE(info->pipe); + info->surf_reg = DSPSURF(info->pipe); + + return 0; +} + +static int gen8_check_mi_display_flip(struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + u32 stride, tile; + + if (!info->async_flip) + return 0; + + if (IS_SKYLAKE(dev_priv)) { + stride = vgpu_vreg(s->vgpu, info->stride_reg) & GENMASK(9, 0); + tile = (vgpu_vreg(s->vgpu, info->ctrl_reg) & + GENMASK(12, 10)) >> 10; + } else { + stride = (vgpu_vreg(s->vgpu, info->stride_reg) & + GENMASK(15, 6)) >> 6; + tile = (vgpu_vreg(s->vgpu, info->ctrl_reg) & (1 << 10)) >> 10; + } + + if (stride != info->stride_val) + gvt_dbg_cmd("cannot change stride during async flip\n"); + + if (tile != info->tile_val) + gvt_dbg_cmd("cannot change tile during async flip\n"); + + return 0; +} + +static int gen8_update_plane_mmio_from_mi_display_flip( + struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + struct intel_vgpu *vgpu = s->vgpu; + + set_mask_bits(&vgpu_vreg(vgpu, info->surf_reg), GENMASK(31, 12), + info->surf_val << 12); + if (IS_SKYLAKE(dev_priv)) { + set_mask_bits(&vgpu_vreg(vgpu, info->stride_reg), GENMASK(9, 0), + info->stride_val); + set_mask_bits(&vgpu_vreg(vgpu, info->ctrl_reg), GENMASK(12, 10), + info->tile_val << 10); + } else { + set_mask_bits(&vgpu_vreg(vgpu, info->stride_reg), GENMASK(15, 6), + info->stride_val << 6); + set_mask_bits(&vgpu_vreg(vgpu, info->ctrl_reg), GENMASK(10, 10), + info->tile_val << 10); + } + + vgpu_vreg(vgpu, PIPE_FRMCOUNT_G4X(info->pipe))++; + intel_vgpu_trigger_virtual_event(vgpu, info->event); + return 0; +} + +static int decode_mi_display_flip(struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + + if (IS_BROADWELL(dev_priv)) + return gen8_decode_mi_display_flip(s, info); + if (IS_SKYLAKE(dev_priv)) + return skl_decode_mi_display_flip(s, info); + + return -ENODEV; +} + +static int check_mi_display_flip(struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + + if (IS_BROADWELL(dev_priv) || IS_SKYLAKE(dev_priv)) + return gen8_check_mi_display_flip(s, info); + return -ENODEV; +} + +static int update_plane_mmio_from_mi_display_flip( + struct parser_exec_state *s, + struct mi_display_flip_command_info *info) +{ + struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv; + + if (IS_BROADWELL(dev_priv) || IS_SKYLAKE(dev_priv)) + return gen8_update_plane_mmio_from_mi_display_flip(s, info); + return -ENODEV; +} + +static int cmd_handler_mi_display_flip(struct parser_exec_state *s) +{ + struct mi_display_flip_command_info info; + int ret; + int i; + int len = cmd_length(s); + + ret = decode_mi_display_flip(s, &info); + if (ret) { + gvt_err("fail to decode MI display flip command\n"); + return ret; + } + + ret = check_mi_display_flip(s, &info); + if (ret) { + gvt_err("invalid MI display flip command\n"); + return ret; + } + + ret = update_plane_mmio_from_mi_display_flip(s, &info); + if (ret) { + gvt_err("fail to update plane mmio\n"); + return ret; + } + + for (i = 0; i < len; i++) + patch_value(s, cmd_ptr(s, i), MI_NOOP); + return 0; +} + +static bool is_wait_for_flip_pending(u32 cmd) +{ + return cmd & (MI_WAIT_FOR_PLANE_A_FLIP_PENDING | + MI_WAIT_FOR_PLANE_B_FLIP_PENDING | + MI_WAIT_FOR_PLANE_C_FLIP_PENDING | + MI_WAIT_FOR_SPRITE_A_FLIP_PENDING | + MI_WAIT_FOR_SPRITE_B_FLIP_PENDING | + MI_WAIT_FOR_SPRITE_C_FLIP_PENDING); +} + +static int cmd_handler_mi_wait_for_event(struct parser_exec_state *s) +{ + u32 cmd = cmd_val(s, 0); + + if (!is_wait_for_flip_pending(cmd)) + return 0; + + patch_value(s, cmd_ptr(s, 0), MI_NOOP); + return 0; +} + +static unsigned long get_gma_bb_from_cmd(struct parser_exec_state *s, int index) +{ + unsigned long addr; + unsigned long gma_high, gma_low; + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + + if (WARN_ON(gmadr_bytes != 4 && gmadr_bytes != 8)) + return INTEL_GVT_INVALID_ADDR; + + gma_low = cmd_val(s, index) & BATCH_BUFFER_ADDR_MASK; + if (gmadr_bytes == 4) { + addr = gma_low; + } else { + gma_high = cmd_val(s, index + 1) & BATCH_BUFFER_ADDR_HIGH_MASK; + addr = (((unsigned long)gma_high) << 32) | gma_low; + } + return addr; +} + +static inline int cmd_address_audit(struct parser_exec_state *s, + unsigned long guest_gma, int op_size, bool index_mode) +{ + struct intel_vgpu *vgpu = s->vgpu; + u32 max_surface_size = vgpu->gvt->device_info.max_surface_size; + int i; + int ret; + + if (op_size > max_surface_size) { + gvt_err("command address audit fail name %s\n", s->info->name); + return -EINVAL; + } + + if (index_mode) { + if (guest_gma >= GTT_PAGE_SIZE / sizeof(u64)) { + ret = -EINVAL; + goto err; + } + } else if ((!vgpu_gmadr_is_valid(s->vgpu, guest_gma)) || + (!vgpu_gmadr_is_valid(s->vgpu, + guest_gma + op_size - 1))) { + ret = -EINVAL; + goto err; + } + return 0; +err: + gvt_err("cmd_parser: Malicious %s detected, addr=0x%lx, len=%d!\n", + s->info->name, guest_gma, op_size); + + pr_err("cmd dump: "); + for (i = 0; i < cmd_length(s); i++) { + if (!(i % 4)) + pr_err("\n%08x ", cmd_val(s, i)); + else + pr_err("%08x ", cmd_val(s, i)); + } + pr_err("\nvgpu%d: aperture 0x%llx - 0x%llx, hidden 0x%llx - 0x%llx\n", + vgpu->id, + vgpu_aperture_gmadr_base(vgpu), + vgpu_aperture_gmadr_end(vgpu), + vgpu_hidden_gmadr_base(vgpu), + vgpu_hidden_gmadr_end(vgpu)); + return ret; +} + +static int cmd_handler_mi_store_data_imm(struct parser_exec_state *s) +{ + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + int op_size = (cmd_length(s) - 3) * sizeof(u32); + int core_id = (cmd_val(s, 2) & (1 << 0)) ? 1 : 0; + unsigned long gma, gma_low, gma_high; + int ret = 0; + + /* check ppggt */ + if (!(cmd_val(s, 0) & (1 << 22))) + return 0; + + gma = cmd_val(s, 2) & GENMASK(31, 2); + + if (gmadr_bytes == 8) { + gma_low = cmd_val(s, 1) & GENMASK(31, 2); + gma_high = cmd_val(s, 2) & GENMASK(15, 0); + gma = (gma_high << 32) | gma_low; + core_id = (cmd_val(s, 1) & (1 << 0)) ? 1 : 0; + } + ret = cmd_address_audit(s, gma + op_size * core_id, op_size, false); + return ret; +} + +static inline int unexpected_cmd(struct parser_exec_state *s) +{ + gvt_err("vgpu%d: Unexpected %s in command buffer!\n", + s->vgpu->id, s->info->name); + return -EINVAL; +} + +static int cmd_handler_mi_semaphore_wait(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_report_perf_count(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_op_2e(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_op_2f(struct parser_exec_state *s) +{ + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + int op_size = (1 << ((cmd_val(s, 0) & GENMASK(20, 19)) >> 19)) * + sizeof(u32); + unsigned long gma, gma_high; + int ret = 0; + + if (!(cmd_val(s, 0) & (1 << 22))) + return ret; + + gma = cmd_val(s, 1) & GENMASK(31, 2); + if (gmadr_bytes == 8) { + gma_high = cmd_val(s, 2) & GENMASK(15, 0); + gma = (gma_high << 32) | gma; + } + ret = cmd_address_audit(s, gma, op_size, false); + return ret; +} + +static int cmd_handler_mi_store_data_index(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_clflush(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_conditional_batch_buffer_end( + struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_update_gtt(struct parser_exec_state *s) +{ + return unexpected_cmd(s); +} + +static int cmd_handler_mi_flush_dw(struct parser_exec_state *s) +{ + int gmadr_bytes = s->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + unsigned long gma; + bool index_mode = false; + int ret = 0; + + /* Check post-sync and ppgtt bit */ + if (((cmd_val(s, 0) >> 14) & 0x3) && (cmd_val(s, 1) & (1 << 2))) { + gma = cmd_val(s, 1) & GENMASK(31, 3); + if (gmadr_bytes == 8) + gma |= (cmd_val(s, 2) & GENMASK(15, 0)) << 32; + /* Store Data Index */ + if (cmd_val(s, 0) & (1 << 21)) + index_mode = true; + ret = cmd_address_audit(s, gma, sizeof(u64), index_mode); + } + /* Check notify bit */ + if ((cmd_val(s, 0) & (1 << 8))) + set_bit(cmd_interrupt_events[s->ring_id].mi_flush_dw, + s->workload->pending_events); + return ret; +} + +static void addr_type_update_snb(struct parser_exec_state *s) +{ + if ((s->buf_type == RING_BUFFER_INSTRUCTION) && + (BATCH_BUFFER_ADR_SPACE_BIT(cmd_val(s, 0)) == 1)) { + s->buf_addr_type = PPGTT_BUFFER; + } +} + + +static int copy_gma_to_hva(struct intel_vgpu *vgpu, struct intel_vgpu_mm *mm, + unsigned long gma, unsigned long end_gma, void *va) +{ + unsigned long copy_len, offset; + unsigned long len = 0; + unsigned long gpa; + + while (gma != end_gma) { + gpa = intel_vgpu_gma_to_gpa(mm, gma); + if (gpa == INTEL_GVT_INVALID_ADDR) { + gvt_err("invalid gma address: %lx\n", gma); + return -EFAULT; + } + + offset = gma & (GTT_PAGE_SIZE - 1); + + copy_len = (end_gma - gma) >= (GTT_PAGE_SIZE - offset) ? + GTT_PAGE_SIZE - offset : end_gma - gma; + + intel_gvt_hypervisor_read_gpa(vgpu, gpa, va + len, copy_len); + + len += copy_len; + gma += copy_len; + } + return 0; +} + + +/* + * Check whether a batch buffer needs to be scanned. Currently + * the only criteria is based on privilege. + */ +static int batch_buffer_needs_scan(struct parser_exec_state *s) +{ + struct intel_gvt *gvt = s->vgpu->gvt; + + if (bypass_batch_buffer_scan) + return 0; + + if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) { + /* BDW decides privilege based on address space */ + if (cmd_val(s, 0) & (1 << 8)) + return 0; + } + return 1; +} + +static uint32_t find_bb_size(struct parser_exec_state *s) +{ + unsigned long gma = 0; + struct cmd_info *info; + uint32_t bb_size = 0; + uint32_t cmd_len = 0; + bool met_bb_end = false; + u32 cmd; + + /* get the start gm address of the batch buffer */ + gma = get_gma_bb_from_cmd(s, 1); + cmd = cmd_val(s, 0); + + info = get_cmd_info(s->vgpu->gvt, cmd, s->ring_id); + if (info == NULL) { + gvt_err("unknown cmd 0x%x, opcode=0x%x\n", + cmd, get_opcode(cmd, s->ring_id)); + return -EINVAL; + } + do { + copy_gma_to_hva(s->vgpu, s->vgpu->gtt.ggtt_mm, + gma, gma + 4, &cmd); + info = get_cmd_info(s->vgpu->gvt, cmd, s->ring_id); + if (info == NULL) { + gvt_err("unknown cmd 0x%x, opcode=0x%x\n", + cmd, get_opcode(cmd, s->ring_id)); + return -EINVAL; + } + + if (info->opcode == OP_MI_BATCH_BUFFER_END) { + met_bb_end = true; + } else if (info->opcode == OP_MI_BATCH_BUFFER_START) { + if (BATCH_BUFFER_2ND_LEVEL_BIT(cmd) == 0) { + /* chained batch buffer */ + met_bb_end = true; + } + } + cmd_len = get_cmd_length(info, cmd) << 2; + bb_size += cmd_len; + gma += cmd_len; + + } while (!met_bb_end); + + return bb_size; +} + +static int perform_bb_shadow(struct parser_exec_state *s) +{ + struct intel_shadow_bb_entry *entry_obj; + unsigned long gma = 0; + uint32_t bb_size; + void *dst = NULL; + int ret = 0; + + /* get the start gm address of the batch buffer */ + gma = get_gma_bb_from_cmd(s, 1); + + /* get the size of the batch buffer */ + bb_size = find_bb_size(s); + + /* allocate shadow batch buffer */ + entry_obj = kmalloc(sizeof(*entry_obj), GFP_KERNEL); + if (entry_obj == NULL) + return -ENOMEM; + + entry_obj->obj = + i915_gem_object_create(&(s->vgpu->gvt->dev_priv->drm), + roundup(bb_size, PAGE_SIZE)); + if (IS_ERR(entry_obj->obj)) { + ret = PTR_ERR(entry_obj->obj); + goto free_entry; + } + entry_obj->len = bb_size; + INIT_LIST_HEAD(&entry_obj->list); + + dst = i915_gem_object_pin_map(entry_obj->obj, I915_MAP_WB); + if (IS_ERR(dst)) { + ret = PTR_ERR(dst); + goto put_obj; + } + + ret = i915_gem_object_set_to_cpu_domain(entry_obj->obj, false); + if (ret) { + gvt_err("failed to set shadow batch to CPU\n"); + goto unmap_src; + } + + entry_obj->va = dst; + entry_obj->bb_start_cmd_va = s->ip_va; + + /* copy batch buffer to shadow batch buffer*/ + ret = copy_gma_to_hva(s->vgpu, s->vgpu->gtt.ggtt_mm, + gma, gma + bb_size, + dst); + if (ret) { + gvt_err("fail to copy guest ring buffer\n"); + goto unmap_src; + } + + list_add(&entry_obj->list, &s->workload->shadow_bb); + /* + * ip_va saves the virtual address of the shadow batch buffer, while + * ip_gma saves the graphics address of the original batch buffer. + * As the shadow batch buffer is just a copy from the originial one, + * it should be right to use shadow batch buffer'va and original batch + * buffer's gma in pair. After all, we don't want to pin the shadow + * buffer here (too early). + */ + s->ip_va = dst; + s->ip_gma = gma; + + return 0; + +unmap_src: + i915_gem_object_unpin_map(entry_obj->obj); +put_obj: + i915_gem_object_put(entry_obj->obj); +free_entry: + kfree(entry_obj); + return ret; +} + +static int cmd_handler_mi_batch_buffer_start(struct parser_exec_state *s) +{ + bool second_level; + int ret = 0; + + if (s->buf_type == BATCH_BUFFER_2ND_LEVEL) { + gvt_err("Found MI_BATCH_BUFFER_START in 2nd level BB\n"); + return -EINVAL; + } + + second_level = BATCH_BUFFER_2ND_LEVEL_BIT(cmd_val(s, 0)) == 1; + if (second_level && (s->buf_type != BATCH_BUFFER_INSTRUCTION)) { + gvt_err("Jumping to 2nd level BB from RB is not allowed\n"); + return -EINVAL; + } + + s->saved_buf_addr_type = s->buf_addr_type; + addr_type_update_snb(s); + if (s->buf_type == RING_BUFFER_INSTRUCTION) { + s->ret_ip_gma_ring = s->ip_gma + cmd_length(s) * sizeof(u32); + s->buf_type = BATCH_BUFFER_INSTRUCTION; + } else if (second_level) { + s->buf_type = BATCH_BUFFER_2ND_LEVEL; + s->ret_ip_gma_bb = s->ip_gma + cmd_length(s) * sizeof(u32); + s->ret_bb_va = s->ip_va + cmd_length(s) * sizeof(u32); + } + + if (batch_buffer_needs_scan(s)) { + ret = perform_bb_shadow(s); + if (ret < 0) + gvt_err("invalid shadow batch buffer\n"); + } else { + /* emulate a batch buffer end to do return right */ + ret = cmd_handler_mi_batch_buffer_end(s); + if (ret < 0) + return ret; + } + + return ret; +} + +static struct cmd_info cmd_info[] = { + {"MI_NOOP", OP_MI_NOOP, F_LEN_CONST, R_ALL, D_ALL, 0, 1, NULL}, + + {"MI_SET_PREDICATE", OP_MI_SET_PREDICATE, F_LEN_CONST, R_ALL, D_ALL, + 0, 1, NULL}, + + {"MI_USER_INTERRUPT", OP_MI_USER_INTERRUPT, F_LEN_CONST, R_ALL, D_ALL, + 0, 1, cmd_handler_mi_user_interrupt}, + + {"MI_WAIT_FOR_EVENT", OP_MI_WAIT_FOR_EVENT, F_LEN_CONST, R_RCS | R_BCS, + D_ALL, 0, 1, cmd_handler_mi_wait_for_event}, + + {"MI_FLUSH", OP_MI_FLUSH, F_LEN_CONST, R_ALL, D_ALL, 0, 1, NULL}, + + {"MI_ARB_CHECK", OP_MI_ARB_CHECK, F_LEN_CONST, R_ALL, D_ALL, 0, 1, + NULL}, + + {"MI_RS_CONTROL", OP_MI_RS_CONTROL, F_LEN_CONST, R_RCS, D_ALL, 0, 1, + NULL}, + + {"MI_REPORT_HEAD", OP_MI_REPORT_HEAD, F_LEN_CONST, R_ALL, D_ALL, 0, 1, + NULL}, + + {"MI_ARB_ON_OFF", OP_MI_ARB_ON_OFF, F_LEN_CONST, R_ALL, D_ALL, 0, 1, + NULL}, + + {"MI_URB_ATOMIC_ALLOC", OP_MI_URB_ATOMIC_ALLOC, F_LEN_CONST, R_RCS, + D_ALL, 0, 1, NULL}, + + {"MI_BATCH_BUFFER_END", OP_MI_BATCH_BUFFER_END, + F_IP_ADVANCE_CUSTOM | F_LEN_CONST, R_ALL, D_ALL, 0, 1, + cmd_handler_mi_batch_buffer_end}, + + {"MI_SUSPEND_FLUSH", OP_MI_SUSPEND_FLUSH, F_LEN_CONST, R_ALL, D_ALL, + 0, 1, NULL}, + + {"MI_PREDICATE", OP_MI_PREDICATE, F_LEN_CONST, R_RCS, D_ALL, 0, 1, + NULL}, + + {"MI_TOPOLOGY_FILTER", OP_MI_TOPOLOGY_FILTER, F_LEN_CONST, R_ALL, + D_ALL, 0, 1, NULL}, + + {"MI_SET_APPID", OP_MI_SET_APPID, F_LEN_CONST, R_ALL, D_ALL, 0, 1, + NULL}, + + {"MI_RS_CONTEXT", OP_MI_RS_CONTEXT, F_LEN_CONST, R_RCS, D_ALL, 0, 1, + NULL}, + + {"MI_DISPLAY_FLIP", OP_MI_DISPLAY_FLIP, F_LEN_VAR | F_POST_HANDLE, + R_RCS | R_BCS, D_ALL, 0, 8, cmd_handler_mi_display_flip}, + + {"MI_SEMAPHORE_MBOX", OP_MI_SEMAPHORE_MBOX, F_LEN_VAR, R_ALL, D_ALL, + 0, 8, NULL}, + + {"MI_MATH", OP_MI_MATH, F_LEN_VAR, R_ALL, D_ALL, 0, 8, NULL}, + + {"MI_URB_CLEAR", OP_MI_URB_CLEAR, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"ME_SEMAPHORE_SIGNAL", OP_MI_SEMAPHORE_SIGNAL, F_LEN_VAR, R_ALL, + D_BDW_PLUS, 0, 8, NULL}, + + {"ME_SEMAPHORE_WAIT", OP_MI_SEMAPHORE_WAIT, F_LEN_VAR, R_ALL, D_BDW_PLUS, + ADDR_FIX_1(2), 8, cmd_handler_mi_semaphore_wait}, + + {"MI_STORE_DATA_IMM", OP_MI_STORE_DATA_IMM, F_LEN_VAR, R_ALL, D_BDW_PLUS, + ADDR_FIX_1(1), 10, cmd_handler_mi_store_data_imm}, + + {"MI_STORE_DATA_INDEX", OP_MI_STORE_DATA_INDEX, F_LEN_VAR, R_ALL, D_ALL, + 0, 8, cmd_handler_mi_store_data_index}, + + {"MI_LOAD_REGISTER_IMM", OP_MI_LOAD_REGISTER_IMM, F_LEN_VAR, R_ALL, + D_ALL, 0, 8, cmd_handler_lri}, + + {"MI_UPDATE_GTT", OP_MI_UPDATE_GTT, F_LEN_VAR, R_ALL, D_BDW_PLUS, 0, 10, + cmd_handler_mi_update_gtt}, + + {"MI_STORE_REGISTER_MEM", OP_MI_STORE_REGISTER_MEM, F_LEN_VAR, R_ALL, + D_ALL, ADDR_FIX_1(2), 8, cmd_handler_srm}, + + {"MI_FLUSH_DW", OP_MI_FLUSH_DW, F_LEN_VAR, R_ALL, D_ALL, 0, 6, + cmd_handler_mi_flush_dw}, + + {"MI_CLFLUSH", OP_MI_CLFLUSH, F_LEN_VAR, R_ALL, D_ALL, ADDR_FIX_1(1), + 10, cmd_handler_mi_clflush}, + + {"MI_REPORT_PERF_COUNT", OP_MI_REPORT_PERF_COUNT, F_LEN_VAR, R_ALL, + D_ALL, ADDR_FIX_1(1), 6, cmd_handler_mi_report_perf_count}, + + {"MI_LOAD_REGISTER_MEM", OP_MI_LOAD_REGISTER_MEM, F_LEN_VAR, R_ALL, + D_ALL, ADDR_FIX_1(2), 8, cmd_handler_lrm}, + + {"MI_LOAD_REGISTER_REG", OP_MI_LOAD_REGISTER_REG, F_LEN_VAR, R_ALL, + D_ALL, 0, 8, cmd_handler_lrr}, + + {"MI_RS_STORE_DATA_IMM", OP_MI_RS_STORE_DATA_IMM, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"MI_LOAD_URB_MEM", OP_MI_LOAD_URB_MEM, F_LEN_VAR, R_RCS, D_ALL, + ADDR_FIX_1(2), 8, NULL}, + + {"MI_STORE_URM_MEM", OP_MI_STORE_URM_MEM, F_LEN_VAR, R_RCS, D_ALL, + ADDR_FIX_1(2), 8, NULL}, + + {"MI_OP_2E", OP_MI_2E, F_LEN_VAR, R_ALL, D_BDW_PLUS, ADDR_FIX_2(1, 2), + 8, cmd_handler_mi_op_2e}, + + {"MI_OP_2F", OP_MI_2F, F_LEN_VAR, R_ALL, D_BDW_PLUS, ADDR_FIX_1(1), + 8, cmd_handler_mi_op_2f}, + + {"MI_BATCH_BUFFER_START", OP_MI_BATCH_BUFFER_START, + F_IP_ADVANCE_CUSTOM, R_ALL, D_ALL, 0, 8, + cmd_handler_mi_batch_buffer_start}, + + {"MI_CONDITIONAL_BATCH_BUFFER_END", OP_MI_CONDITIONAL_BATCH_BUFFER_END, + F_LEN_VAR, R_ALL, D_ALL, ADDR_FIX_1(2), 8, + cmd_handler_mi_conditional_batch_buffer_end}, + + {"MI_LOAD_SCAN_LINES_INCL", OP_MI_LOAD_SCAN_LINES_INCL, F_LEN_CONST, + R_RCS | R_BCS, D_ALL, 0, 2, NULL}, + + {"XY_SETUP_BLT", OP_XY_SETUP_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_2(4, 7), 8, NULL}, + + {"XY_SETUP_CLIP_BLT", OP_XY_SETUP_CLIP_BLT, F_LEN_VAR, R_BCS, D_ALL, + 0, 8, NULL}, + + {"XY_SETUP_MONO_PATTERN_SL_BLT", OP_XY_SETUP_MONO_PATTERN_SL_BLT, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_1(4), 8, NULL}, + + {"XY_PIXEL_BLT", OP_XY_PIXEL_BLT, F_LEN_VAR, R_BCS, D_ALL, 0, 8, NULL}, + + {"XY_SCANLINES_BLT", OP_XY_SCANLINES_BLT, F_LEN_VAR, R_BCS, D_ALL, + 0, 8, NULL}, + + {"XY_TEXT_BLT", OP_XY_TEXT_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_1(3), 8, NULL}, + + {"XY_TEXT_IMMEDIATE_BLT", OP_XY_TEXT_IMMEDIATE_BLT, F_LEN_VAR, R_BCS, + D_ALL, 0, 8, NULL}, + + {"XY_COLOR_BLT", OP_XY_COLOR_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_1(4), 8, NULL}, + + {"XY_PAT_BLT", OP_XY_PAT_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_2(4, 5), 8, NULL}, + + {"XY_MONO_PAT_BLT", OP_XY_MONO_PAT_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_1(4), 8, NULL}, + + {"XY_SRC_COPY_BLT", OP_XY_SRC_COPY_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_2(4, 7), 8, NULL}, + + {"XY_MONO_SRC_COPY_BLT", OP_XY_MONO_SRC_COPY_BLT, F_LEN_VAR, R_BCS, + D_ALL, ADDR_FIX_2(4, 5), 8, NULL}, + + {"XY_FULL_BLT", OP_XY_FULL_BLT, F_LEN_VAR, R_BCS, D_ALL, 0, 8, NULL}, + + {"XY_FULL_MONO_SRC_BLT", OP_XY_FULL_MONO_SRC_BLT, F_LEN_VAR, R_BCS, + D_ALL, ADDR_FIX_3(4, 5, 8), 8, NULL}, + + {"XY_FULL_MONO_PATTERN_BLT", OP_XY_FULL_MONO_PATTERN_BLT, F_LEN_VAR, + R_BCS, D_ALL, ADDR_FIX_2(4, 7), 8, NULL}, + + {"XY_FULL_MONO_PATTERN_MONO_SRC_BLT", + OP_XY_FULL_MONO_PATTERN_MONO_SRC_BLT, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_2(4, 5), 8, NULL}, + + {"XY_MONO_PAT_FIXED_BLT", OP_XY_MONO_PAT_FIXED_BLT, F_LEN_VAR, R_BCS, + D_ALL, ADDR_FIX_1(4), 8, NULL}, + + {"XY_MONO_SRC_COPY_IMMEDIATE_BLT", OP_XY_MONO_SRC_COPY_IMMEDIATE_BLT, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_1(4), 8, NULL}, + + {"XY_PAT_BLT_IMMEDIATE", OP_XY_PAT_BLT_IMMEDIATE, F_LEN_VAR, R_BCS, + D_ALL, ADDR_FIX_1(4), 8, NULL}, + + {"XY_SRC_COPY_CHROMA_BLT", OP_XY_SRC_COPY_CHROMA_BLT, F_LEN_VAR, R_BCS, + D_ALL, ADDR_FIX_2(4, 7), 8, NULL}, + + {"XY_FULL_IMMEDIATE_PATTERN_BLT", OP_XY_FULL_IMMEDIATE_PATTERN_BLT, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_2(4, 7), 8, NULL}, + + {"XY_FULL_MONO_SRC_IMMEDIATE_PATTERN_BLT", + OP_XY_FULL_MONO_SRC_IMMEDIATE_PATTERN_BLT, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_2(4, 5), 8, NULL}, + + {"XY_PAT_CHROMA_BLT", OP_XY_PAT_CHROMA_BLT, F_LEN_VAR, R_BCS, D_ALL, + ADDR_FIX_2(4, 5), 8, NULL}, + + {"XY_PAT_CHROMA_BLT_IMMEDIATE", OP_XY_PAT_CHROMA_BLT_IMMEDIATE, + F_LEN_VAR, R_BCS, D_ALL, ADDR_FIX_1(4), 8, NULL}, + + {"3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", + OP_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_VIEWPORT_STATE_POINTERS_CC", + OP_3DSTATE_VIEWPORT_STATE_POINTERS_CC, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BLEND_STATE_POINTERS", + OP_3DSTATE_BLEND_STATE_POINTERS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DEPTH_STENCIL_STATE_POINTERS", + OP_3DSTATE_DEPTH_STENCIL_STATE_POINTERS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POINTERS_VS", + OP_3DSTATE_BINDING_TABLE_POINTERS_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POINTERS_HS", + OP_3DSTATE_BINDING_TABLE_POINTERS_HS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POINTERS_DS", + OP_3DSTATE_BINDING_TABLE_POINTERS_DS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POINTERS_GS", + OP_3DSTATE_BINDING_TABLE_POINTERS_GS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POINTERS_PS", + OP_3DSTATE_BINDING_TABLE_POINTERS_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_STATE_POINTERS_VS", + OP_3DSTATE_SAMPLER_STATE_POINTERS_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_STATE_POINTERS_HS", + OP_3DSTATE_SAMPLER_STATE_POINTERS_HS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_STATE_POINTERS_DS", + OP_3DSTATE_SAMPLER_STATE_POINTERS_DS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_STATE_POINTERS_GS", + OP_3DSTATE_SAMPLER_STATE_POINTERS_GS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_STATE_POINTERS_PS", + OP_3DSTATE_SAMPLER_STATE_POINTERS_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_URB_VS", OP_3DSTATE_URB_VS, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"3DSTATE_URB_HS", OP_3DSTATE_URB_HS, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"3DSTATE_URB_DS", OP_3DSTATE_URB_DS, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"3DSTATE_URB_GS", OP_3DSTATE_URB_GS, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"3DSTATE_GATHER_CONSTANT_VS", OP_3DSTATE_GATHER_CONSTANT_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_GATHER_CONSTANT_GS", OP_3DSTATE_GATHER_CONSTANT_GS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_GATHER_CONSTANT_HS", OP_3DSTATE_GATHER_CONSTANT_HS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_GATHER_CONSTANT_DS", OP_3DSTATE_GATHER_CONSTANT_DS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_GATHER_CONSTANT_PS", OP_3DSTATE_GATHER_CONSTANT_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_CONSTANTF_VS", OP_3DSTATE_DX9_CONSTANTF_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 11, NULL}, + + {"3DSTATE_DX9_CONSTANTF_PS", OP_3DSTATE_DX9_CONSTANTF_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 11, NULL}, + + {"3DSTATE_DX9_CONSTANTI_VS", OP_3DSTATE_DX9_CONSTANTI_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_CONSTANTI_PS", OP_3DSTATE_DX9_CONSTANTI_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_CONSTANTB_VS", OP_3DSTATE_DX9_CONSTANTB_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_CONSTANTB_PS", OP_3DSTATE_DX9_CONSTANTB_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_LOCAL_VALID_VS", OP_3DSTATE_DX9_LOCAL_VALID_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_LOCAL_VALID_PS", OP_3DSTATE_DX9_LOCAL_VALID_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_GENERATE_ACTIVE_VS", OP_3DSTATE_DX9_GENERATE_ACTIVE_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DX9_GENERATE_ACTIVE_PS", OP_3DSTATE_DX9_GENERATE_ACTIVE_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_BINDING_TABLE_EDIT_VS", OP_3DSTATE_BINDING_TABLE_EDIT_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 9, NULL}, + + {"3DSTATE_BINDING_TABLE_EDIT_GS", OP_3DSTATE_BINDING_TABLE_EDIT_GS, + F_LEN_VAR, R_RCS, D_ALL, 0, 9, NULL}, + + {"3DSTATE_BINDING_TABLE_EDIT_HS", OP_3DSTATE_BINDING_TABLE_EDIT_HS, + F_LEN_VAR, R_RCS, D_ALL, 0, 9, NULL}, + + {"3DSTATE_BINDING_TABLE_EDIT_DS", OP_3DSTATE_BINDING_TABLE_EDIT_DS, + F_LEN_VAR, R_RCS, D_ALL, 0, 9, NULL}, + + {"3DSTATE_BINDING_TABLE_EDIT_PS", OP_3DSTATE_BINDING_TABLE_EDIT_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 9, NULL}, + + {"3DSTATE_VF_INSTANCING", OP_3DSTATE_VF_INSTANCING, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_VF_SGVS", OP_3DSTATE_VF_SGVS, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, 8, + NULL}, + + {"3DSTATE_VF_TOPOLOGY", OP_3DSTATE_VF_TOPOLOGY, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_WM_CHROMAKEY", OP_3DSTATE_WM_CHROMAKEY, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_PS_BLEND", OP_3DSTATE_PS_BLEND, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, + 8, NULL}, + + {"3DSTATE_WM_DEPTH_STENCIL", OP_3DSTATE_WM_DEPTH_STENCIL, F_LEN_VAR, + R_RCS, D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_PS_EXTRA", OP_3DSTATE_PS_EXTRA, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, + 8, NULL}, + + {"3DSTATE_RASTER", OP_3DSTATE_RASTER, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, 8, + NULL}, + + {"3DSTATE_SBE_SWIZ", OP_3DSTATE_SBE_SWIZ, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, 8, + NULL}, + + {"3DSTATE_WM_HZ_OP", OP_3DSTATE_WM_HZ_OP, F_LEN_VAR, R_RCS, D_BDW_PLUS, 0, 8, + NULL}, + + {"3DSTATE_VERTEX_BUFFERS", OP_3DSTATE_VERTEX_BUFFERS, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_VERTEX_ELEMENTS", OP_3DSTATE_VERTEX_ELEMENTS, F_LEN_VAR, + R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_INDEX_BUFFER", OP_3DSTATE_INDEX_BUFFER, F_LEN_VAR, R_RCS, + D_BDW_PLUS, ADDR_FIX_1(2), 8, NULL}, + + {"3DSTATE_VF_STATISTICS", OP_3DSTATE_VF_STATISTICS, F_LEN_CONST, + R_RCS, D_ALL, 0, 1, NULL}, + + {"3DSTATE_VF", OP_3DSTATE_VF, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_CC_STATE_POINTERS", OP_3DSTATE_CC_STATE_POINTERS, F_LEN_VAR, + R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SCISSOR_STATE_POINTERS", OP_3DSTATE_SCISSOR_STATE_POINTERS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_GS", OP_3DSTATE_GS, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_CLIP", OP_3DSTATE_CLIP, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_WM", OP_3DSTATE_WM, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_CONSTANT_GS", OP_3DSTATE_CONSTANT_GS, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_CONSTANT_PS", OP_3DSTATE_CONSTANT_PS, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_SAMPLE_MASK", OP_3DSTATE_SAMPLE_MASK, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"3DSTATE_CONSTANT_HS", OP_3DSTATE_CONSTANT_HS, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_CONSTANT_DS", OP_3DSTATE_CONSTANT_DS, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_HS", OP_3DSTATE_HS, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_TE", OP_3DSTATE_TE, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DS", OP_3DSTATE_DS, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_STREAMOUT", OP_3DSTATE_STREAMOUT, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"3DSTATE_SBE", OP_3DSTATE_SBE, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PS", OP_3DSTATE_PS, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_DRAWING_RECTANGLE", OP_3DSTATE_DRAWING_RECTANGLE, F_LEN_VAR, + R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_PALETTE_LOAD0", OP_3DSTATE_SAMPLER_PALETTE_LOAD0, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_CHROMA_KEY", OP_3DSTATE_CHROMA_KEY, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"3DSTATE_DEPTH_BUFFER", OP_3DSTATE_DEPTH_BUFFER, F_LEN_VAR, R_RCS, + D_ALL, ADDR_FIX_1(2), 8, NULL}, + + {"3DSTATE_POLY_STIPPLE_OFFSET", OP_3DSTATE_POLY_STIPPLE_OFFSET, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_POLY_STIPPLE_PATTERN", OP_3DSTATE_POLY_STIPPLE_PATTERN, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_LINE_STIPPLE", OP_3DSTATE_LINE_STIPPLE, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"3DSTATE_AA_LINE_PARAMS", OP_3DSTATE_AA_LINE_PARAMS, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"3DSTATE_GS_SVB_INDEX", OP_3DSTATE_GS_SVB_INDEX, F_LEN_VAR, R_RCS, + D_ALL, 0, 8, NULL}, + + {"3DSTATE_SAMPLER_PALETTE_LOAD1", OP_3DSTATE_SAMPLER_PALETTE_LOAD1, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_MULTISAMPLE", OP_3DSTATE_MULTISAMPLE_BDW, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"3DSTATE_STENCIL_BUFFER", OP_3DSTATE_STENCIL_BUFFER, F_LEN_VAR, R_RCS, + D_ALL, ADDR_FIX_1(2), 8, NULL}, + + {"3DSTATE_HIER_DEPTH_BUFFER", OP_3DSTATE_HIER_DEPTH_BUFFER, F_LEN_VAR, + R_RCS, D_ALL, ADDR_FIX_1(2), 8, NULL}, + + {"3DSTATE_CLEAR_PARAMS", OP_3DSTATE_CLEAR_PARAMS, F_LEN_VAR, + R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PUSH_CONSTANT_ALLOC_VS", OP_3DSTATE_PUSH_CONSTANT_ALLOC_VS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PUSH_CONSTANT_ALLOC_HS", OP_3DSTATE_PUSH_CONSTANT_ALLOC_HS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PUSH_CONSTANT_ALLOC_DS", OP_3DSTATE_PUSH_CONSTANT_ALLOC_DS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PUSH_CONSTANT_ALLOC_GS", OP_3DSTATE_PUSH_CONSTANT_ALLOC_GS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_PUSH_CONSTANT_ALLOC_PS", OP_3DSTATE_PUSH_CONSTANT_ALLOC_PS, + F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_MONOFILTER_SIZE", OP_3DSTATE_MONOFILTER_SIZE, F_LEN_VAR, + R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SO_DECL_LIST", OP_3DSTATE_SO_DECL_LIST, F_LEN_VAR, R_RCS, + D_ALL, 0, 9, NULL}, + + {"3DSTATE_SO_BUFFER", OP_3DSTATE_SO_BUFFER, F_LEN_VAR, R_RCS, D_BDW_PLUS, + ADDR_FIX_2(2, 4), 8, NULL}, + + {"3DSTATE_BINDING_TABLE_POOL_ALLOC", + OP_3DSTATE_BINDING_TABLE_POOL_ALLOC, + F_LEN_VAR, R_RCS, D_BDW_PLUS, ADDR_FIX_1(1), 8, NULL}, + + {"3DSTATE_GATHER_POOL_ALLOC", OP_3DSTATE_GATHER_POOL_ALLOC, + F_LEN_VAR, R_RCS, D_BDW_PLUS, ADDR_FIX_1(1), 8, NULL}, + + {"3DSTATE_DX9_CONSTANT_BUFFER_POOL_ALLOC", + OP_3DSTATE_DX9_CONSTANT_BUFFER_POOL_ALLOC, + F_LEN_VAR, R_RCS, D_BDW_PLUS, ADDR_FIX_1(1), 8, NULL}, + + {"3DSTATE_SAMPLE_PATTERN", OP_3DSTATE_SAMPLE_PATTERN, F_LEN_VAR, R_RCS, + D_BDW_PLUS, 0, 8, NULL}, + + {"PIPE_CONTROL", OP_PIPE_CONTROL, F_LEN_VAR, R_RCS, D_ALL, + ADDR_FIX_1(2), 8, cmd_handler_pipe_control}, + + {"3DPRIMITIVE", OP_3DPRIMITIVE, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"PIPELINE_SELECT", OP_PIPELINE_SELECT, F_LEN_CONST, R_RCS, D_ALL, 0, + 1, NULL}, + + {"STATE_PREFETCH", OP_STATE_PREFETCH, F_LEN_VAR, R_RCS, D_ALL, + ADDR_FIX_1(1), 8, NULL}, + + {"STATE_SIP", OP_STATE_SIP, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"STATE_BASE_ADDRESS", OP_STATE_BASE_ADDRESS, F_LEN_VAR, R_RCS, D_BDW_PLUS, + ADDR_FIX_5(1, 3, 4, 5, 6), 8, NULL}, + + {"OP_3D_MEDIA_0_1_4", OP_3D_MEDIA_0_1_4, F_LEN_VAR, R_RCS, D_ALL, + ADDR_FIX_1(1), 8, NULL}, + + {"3DSTATE_VS", OP_3DSTATE_VS, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_SF", OP_3DSTATE_SF, F_LEN_VAR, R_RCS, D_ALL, 0, 8, NULL}, + + {"3DSTATE_CONSTANT_VS", OP_3DSTATE_CONSTANT_VS, F_LEN_VAR, R_RCS, D_BDW_PLUS, + 0, 8, NULL}, + + {"3DSTATE_COMPONENT_PACKING", OP_3DSTATE_COMPONENT_PACKING, F_LEN_VAR, R_RCS, + D_SKL_PLUS, 0, 8, NULL}, + + {"MEDIA_INTERFACE_DESCRIPTOR_LOAD", OP_MEDIA_INTERFACE_DESCRIPTOR_LOAD, + F_LEN_VAR, R_RCS, D_ALL, 0, 16, NULL}, + + {"MEDIA_GATEWAY_STATE", OP_MEDIA_GATEWAY_STATE, F_LEN_VAR, R_RCS, D_ALL, + 0, 16, NULL}, + + {"MEDIA_STATE_FLUSH", OP_MEDIA_STATE_FLUSH, F_LEN_VAR, R_RCS, D_ALL, + 0, 16, NULL}, + + {"MEDIA_OBJECT", OP_MEDIA_OBJECT, F_LEN_VAR, R_RCS, D_ALL, 0, 16, NULL}, + + {"MEDIA_CURBE_LOAD", OP_MEDIA_CURBE_LOAD, F_LEN_VAR, R_RCS, D_ALL, + 0, 16, NULL}, + + {"MEDIA_OBJECT_PRT", OP_MEDIA_OBJECT_PRT, F_LEN_VAR, R_RCS, D_ALL, + 0, 16, NULL}, + + {"MEDIA_OBJECT_WALKER", OP_MEDIA_OBJECT_WALKER, F_LEN_VAR, R_RCS, D_ALL, + 0, 16, NULL}, + + {"GPGPU_WALKER", OP_GPGPU_WALKER, F_LEN_VAR, R_RCS, D_ALL, + 0, 8, NULL}, + + {"MEDIA_VFE_STATE", OP_MEDIA_VFE_STATE, F_LEN_VAR, R_RCS, D_ALL, 0, 16, + NULL}, + + {"3DSTATE_VF_STATISTICS_GM45", OP_3DSTATE_VF_STATISTICS_GM45, + F_LEN_CONST, R_ALL, D_ALL, 0, 1, NULL}, + + {"MFX_PIPE_MODE_SELECT", OP_MFX_PIPE_MODE_SELECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_SURFACE_STATE", OP_MFX_SURFACE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_PIPE_BUF_ADDR_STATE", OP_MFX_PIPE_BUF_ADDR_STATE, F_LEN_VAR, + R_VCS, D_BDW_PLUS, 0, 12, NULL}, + + {"MFX_IND_OBJ_BASE_ADDR_STATE", OP_MFX_IND_OBJ_BASE_ADDR_STATE, + F_LEN_VAR, R_VCS, D_BDW_PLUS, 0, 12, NULL}, + + {"MFX_BSP_BUF_BASE_ADDR_STATE", OP_MFX_BSP_BUF_BASE_ADDR_STATE, + F_LEN_VAR, R_VCS, D_BDW_PLUS, ADDR_FIX_3(1, 3, 5), 12, NULL}, + + {"OP_2_0_0_5", OP_2_0_0_5, F_LEN_VAR, R_VCS, D_BDW_PLUS, 0, 12, NULL}, + + {"MFX_STATE_POINTER", OP_MFX_STATE_POINTER, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_QM_STATE", OP_MFX_QM_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_FQM_STATE", OP_MFX_FQM_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_PAK_INSERT_OBJECT", OP_MFX_PAK_INSERT_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_STITCH_OBJECT", OP_MFX_STITCH_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_IT_OBJECT", OP_MFD_IT_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_WAIT", OP_MFX_WAIT, F_LEN_VAR, + R_VCS, D_ALL, 0, 6, NULL}, + + {"MFX_AVC_IMG_STATE", OP_MFX_AVC_IMG_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_AVC_QM_STATE", OP_MFX_AVC_QM_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_AVC_DIRECTMODE_STATE", OP_MFX_AVC_DIRECTMODE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_AVC_SLICE_STATE", OP_MFX_AVC_SLICE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_AVC_REF_IDX_STATE", OP_MFX_AVC_REF_IDX_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_AVC_WEIGHTOFFSET_STATE", OP_MFX_AVC_WEIGHTOFFSET_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_AVC_PICID_STATE", OP_MFD_AVC_PICID_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + {"MFD_AVC_DPB_STATE", OP_MFD_AVC_DPB_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_AVC_BSD_OBJECT", OP_MFD_AVC_BSD_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_AVC_SLICEADDR", OP_MFD_AVC_SLICEADDR, F_LEN_VAR, + R_VCS, D_ALL, ADDR_FIX_1(2), 12, NULL}, + + {"MFC_AVC_PAK_OBJECT", OP_MFC_AVC_PAK_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_VC1_PRED_PIPE_STATE", OP_MFX_VC1_PRED_PIPE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_VC1_DIRECTMODE_STATE", OP_MFX_VC1_DIRECTMODE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_VC1_SHORT_PIC_STATE", OP_MFD_VC1_SHORT_PIC_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_VC1_LONG_PIC_STATE", OP_MFD_VC1_LONG_PIC_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_VC1_BSD_OBJECT", OP_MFD_VC1_BSD_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFC_MPEG2_SLICEGROUP_STATE", OP_MFC_MPEG2_SLICEGROUP_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFC_MPEG2_PAK_OBJECT", OP_MFC_MPEG2_PAK_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_MPEG2_PIC_STATE", OP_MFX_MPEG2_PIC_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_MPEG2_QM_STATE", OP_MFX_MPEG2_QM_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_MPEG2_BSD_OBJECT", OP_MFD_MPEG2_BSD_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_2_6_0_0", OP_MFX_2_6_0_0, F_LEN_VAR, R_VCS, D_ALL, + 0, 16, NULL}, + + {"MFX_2_6_0_9", OP_MFX_2_6_0_9, F_LEN_VAR, R_VCS, D_ALL, 0, 16, NULL}, + + {"MFX_2_6_0_8", OP_MFX_2_6_0_8, F_LEN_VAR, R_VCS, D_ALL, 0, 16, NULL}, + + {"MFX_JPEG_PIC_STATE", OP_MFX_JPEG_PIC_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFX_JPEG_HUFF_TABLE_STATE", OP_MFX_JPEG_HUFF_TABLE_STATE, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"MFD_JPEG_BSD_OBJECT", OP_MFD_JPEG_BSD_OBJECT, F_LEN_VAR, + R_VCS, D_ALL, 0, 12, NULL}, + + {"VEBOX_STATE", OP_VEB_STATE, F_LEN_VAR, R_VECS, D_ALL, 0, 12, NULL}, + + {"VEBOX_SURFACE_STATE", OP_VEB_SURFACE_STATE, F_LEN_VAR, R_VECS, D_ALL, + 0, 12, NULL}, + + {"VEB_DI_IECP", OP_VEB_DNDI_IECP_STATE, F_LEN_VAR, R_VECS, D_BDW_PLUS, + 0, 20, NULL}, +}; + +static void add_cmd_entry(struct intel_gvt *gvt, struct cmd_entry *e) +{ + hash_add(gvt->cmd_table, &e->hlist, e->info->opcode); +} + +#define GVT_MAX_CMD_LENGTH 20 /* In Dword */ + +static void trace_cs_command(struct parser_exec_state *s, + cycles_t cost_pre_cmd_handler, cycles_t cost_cmd_handler) +{ + /* This buffer is used by ftrace to store all commands copied from + * guest gma space. Sometimes commands can cross pages, this should + * not be handled in ftrace logic. So this is just used as a + * 'bounce buffer' + */ + u32 cmd_trace_buf[GVT_MAX_CMD_LENGTH]; + int i; + u32 cmd_len = cmd_length(s); + /* The chosen value of GVT_MAX_CMD_LENGTH are just based on + * following two considerations: + * 1) From observation, most common ring commands is not that long. + * But there are execeptions. So it indeed makes sence to observe + * longer commands. + * 2) From the performance and debugging point of view, dumping all + * contents of very commands is not necessary. + * We mgith shrink GVT_MAX_CMD_LENGTH or remove this trace event in + * future for performance considerations. + */ + if (unlikely(cmd_len > GVT_MAX_CMD_LENGTH)) { + gvt_dbg_cmd("cmd length exceed tracing limitation!\n"); + cmd_len = GVT_MAX_CMD_LENGTH; + } + + for (i = 0; i < cmd_len; i++) + cmd_trace_buf[i] = cmd_val(s, i); + + trace_gvt_command(s->vgpu->id, s->ring_id, s->ip_gma, cmd_trace_buf, + cmd_len, s->buf_type == RING_BUFFER_INSTRUCTION, + cost_pre_cmd_handler, cost_cmd_handler); +} + +/* call the cmd handler, and advance ip */ +static int cmd_parser_exec(struct parser_exec_state *s) +{ + struct cmd_info *info; + u32 cmd; + int ret = 0; + cycles_t t0, t1, t2; + struct parser_exec_state s_before_advance_custom; + + t0 = get_cycles(); + + cmd = cmd_val(s, 0); + + info = get_cmd_info(s->vgpu->gvt, cmd, s->ring_id); + if (info == NULL) { + gvt_err("unknown cmd 0x%x, opcode=0x%x\n", + cmd, get_opcode(cmd, s->ring_id)); + return -EINVAL; + } + + gvt_dbg_cmd("%s\n", info->name); + + s->info = info; + + t1 = get_cycles(); + + memcpy(&s_before_advance_custom, s, sizeof(struct parser_exec_state)); + + if (info->handler) { + ret = info->handler(s); + if (ret < 0) { + gvt_err("%s handler error\n", info->name); + return ret; + } + } + t2 = get_cycles(); + + trace_cs_command(&s_before_advance_custom, t1 - t0, t2 - t1); + + if (!(info->flag & F_IP_ADVANCE_CUSTOM)) { + ret = cmd_advance_default(s); + if (ret) { + gvt_err("%s IP advance error\n", info->name); + return ret; + } + } + return 0; +} + +static inline bool gma_out_of_range(unsigned long gma, + unsigned long gma_head, unsigned int gma_tail) +{ + if (gma_tail >= gma_head) + return (gma < gma_head) || (gma > gma_tail); + else + return (gma > gma_tail) && (gma < gma_head); +} + +static int command_scan(struct parser_exec_state *s, + unsigned long rb_head, unsigned long rb_tail, + unsigned long rb_start, unsigned long rb_len) +{ + + unsigned long gma_head, gma_tail, gma_bottom; + int ret = 0; + + gma_head = rb_start + rb_head; + gma_tail = rb_start + rb_tail; + gma_bottom = rb_start + rb_len; + + gvt_dbg_cmd("scan_start: start=%lx end=%lx\n", gma_head, gma_tail); + + while (s->ip_gma != gma_tail) { + if (s->buf_type == RING_BUFFER_INSTRUCTION) { + if (!(s->ip_gma >= rb_start) || + !(s->ip_gma < gma_bottom)) { + gvt_err("ip_gma %lx out of ring scope." + "(base:0x%lx, bottom: 0x%lx)\n", + s->ip_gma, rb_start, + gma_bottom); + parser_exec_state_dump(s); + return -EINVAL; + } + if (gma_out_of_range(s->ip_gma, gma_head, gma_tail)) { + gvt_err("ip_gma %lx out of range." + "base 0x%lx head 0x%lx tail 0x%lx\n", + s->ip_gma, rb_start, + rb_head, rb_tail); + parser_exec_state_dump(s); + break; + } + } + ret = cmd_parser_exec(s); + if (ret) { + gvt_err("cmd parser error\n"); + parser_exec_state_dump(s); + break; + } + } + + gvt_dbg_cmd("scan_end\n"); + + return ret; +} + +static int scan_workload(struct intel_vgpu_workload *workload) +{ + unsigned long gma_head, gma_tail, gma_bottom; + struct parser_exec_state s; + int ret = 0; + + /* ring base is page aligned */ + if (WARN_ON(!IS_ALIGNED(workload->rb_start, GTT_PAGE_SIZE))) + return -EINVAL; + + gma_head = workload->rb_start + workload->rb_head; + gma_tail = workload->rb_start + workload->rb_tail; + gma_bottom = workload->rb_start + _RING_CTL_BUF_SIZE(workload->rb_ctl); + + s.buf_type = RING_BUFFER_INSTRUCTION; + s.buf_addr_type = GTT_BUFFER; + s.vgpu = workload->vgpu; + s.ring_id = workload->ring_id; + s.ring_start = workload->rb_start; + s.ring_size = _RING_CTL_BUF_SIZE(workload->rb_ctl); + s.ring_head = gma_head; + s.ring_tail = gma_tail; + s.rb_va = workload->shadow_ring_buffer_va; + s.workload = workload; + + if ((bypass_scan_mask & (1 << workload->ring_id)) || + gma_head == gma_tail) + return 0; + + ret = ip_gma_set(&s, gma_head); + if (ret) + goto out; + + ret = command_scan(&s, workload->rb_head, workload->rb_tail, + workload->rb_start, _RING_CTL_BUF_SIZE(workload->rb_ctl)); + +out: + return ret; +} + +static int scan_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + + unsigned long gma_head, gma_tail, gma_bottom, ring_size, ring_tail; + struct parser_exec_state s; + int ret = 0; + + /* ring base is page aligned */ + if (WARN_ON(!IS_ALIGNED(wa_ctx->indirect_ctx.guest_gma, GTT_PAGE_SIZE))) + return -EINVAL; + + ring_tail = wa_ctx->indirect_ctx.size + 3 * sizeof(uint32_t); + ring_size = round_up(wa_ctx->indirect_ctx.size + CACHELINE_BYTES, + PAGE_SIZE); + gma_head = wa_ctx->indirect_ctx.guest_gma; + gma_tail = wa_ctx->indirect_ctx.guest_gma + ring_tail; + gma_bottom = wa_ctx->indirect_ctx.guest_gma + ring_size; + + s.buf_type = RING_BUFFER_INSTRUCTION; + s.buf_addr_type = GTT_BUFFER; + s.vgpu = wa_ctx->workload->vgpu; + s.ring_id = wa_ctx->workload->ring_id; + s.ring_start = wa_ctx->indirect_ctx.guest_gma; + s.ring_size = ring_size; + s.ring_head = gma_head; + s.ring_tail = gma_tail; + s.rb_va = wa_ctx->indirect_ctx.shadow_va; + s.workload = wa_ctx->workload; + + ret = ip_gma_set(&s, gma_head); + if (ret) + goto out; + + ret = command_scan(&s, 0, ring_tail, + wa_ctx->indirect_ctx.guest_gma, ring_size); +out: + return ret; +} + +static int shadow_workload_ring_buffer(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + int ring_id = workload->ring_id; + struct i915_gem_context *shadow_ctx = vgpu->shadow_ctx; + struct intel_ring *ring = shadow_ctx->engine[ring_id].ring; + unsigned long gma_head, gma_tail, gma_top, guest_rb_size; + unsigned int copy_len = 0; + int ret; + + guest_rb_size = _RING_CTL_BUF_SIZE(workload->rb_ctl); + + /* calculate workload ring buffer size */ + workload->rb_len = (workload->rb_tail + guest_rb_size - + workload->rb_head) % guest_rb_size; + + gma_head = workload->rb_start + workload->rb_head; + gma_tail = workload->rb_start + workload->rb_tail; + gma_top = workload->rb_start + guest_rb_size; + + /* allocate shadow ring buffer */ + ret = intel_ring_begin(workload->req, workload->rb_len / 4); + if (ret) + return ret; + + /* get shadow ring buffer va */ + workload->shadow_ring_buffer_va = ring->vaddr + ring->tail; + + /* head > tail --> copy head <-> top */ + if (gma_head > gma_tail) { + ret = copy_gma_to_hva(vgpu, vgpu->gtt.ggtt_mm, + gma_head, gma_top, + workload->shadow_ring_buffer_va); + if (ret) { + gvt_err("fail to copy guest ring buffer\n"); + return ret; + } + copy_len = gma_top - gma_head; + gma_head = workload->rb_start; + } + + /* copy head or start <-> tail */ + ret = copy_gma_to_hva(vgpu, vgpu->gtt.ggtt_mm, + gma_head, gma_tail, + workload->shadow_ring_buffer_va + copy_len); + if (ret) { + gvt_err("fail to copy guest ring buffer\n"); + return ret; + } + ring->tail += workload->rb_len; + intel_ring_advance(ring); + return 0; +} + +int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload) +{ + int ret; + + ret = shadow_workload_ring_buffer(workload); + if (ret) { + gvt_err("fail to shadow workload ring_buffer\n"); + return ret; + } + + ret = scan_workload(workload); + if (ret) { + gvt_err("scan workload error\n"); + return ret; + } + return 0; +} + +static int shadow_indirect_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + struct drm_device *dev = &wa_ctx->workload->vgpu->gvt->dev_priv->drm; + int ctx_size = wa_ctx->indirect_ctx.size; + unsigned long guest_gma = wa_ctx->indirect_ctx.guest_gma; + struct drm_i915_gem_object *obj; + int ret = 0; + void *map; + + obj = i915_gem_object_create(dev, + roundup(ctx_size + CACHELINE_BYTES, + PAGE_SIZE)); + if (IS_ERR(obj)) + return PTR_ERR(obj); + + /* get the va of the shadow batch buffer */ + map = i915_gem_object_pin_map(obj, I915_MAP_WB); + if (IS_ERR(map)) { + gvt_err("failed to vmap shadow indirect ctx\n"); + ret = PTR_ERR(map); + goto put_obj; + } + + ret = i915_gem_object_set_to_cpu_domain(obj, false); + if (ret) { + gvt_err("failed to set shadow indirect ctx to CPU\n"); + goto unmap_src; + } + + ret = copy_gma_to_hva(wa_ctx->workload->vgpu, + wa_ctx->workload->vgpu->gtt.ggtt_mm, + guest_gma, guest_gma + ctx_size, + map); + if (ret) { + gvt_err("fail to copy guest indirect ctx\n"); + goto unmap_src; + } + + wa_ctx->indirect_ctx.obj = obj; + wa_ctx->indirect_ctx.shadow_va = map; + return 0; + +unmap_src: + i915_gem_object_unpin_map(obj); +put_obj: + i915_gem_object_put(wa_ctx->indirect_ctx.obj); + return ret; +} + +static int combine_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + uint32_t per_ctx_start[CACHELINE_DWORDS] = {0}; + unsigned char *bb_start_sva; + + per_ctx_start[0] = 0x18800001; + per_ctx_start[1] = wa_ctx->per_ctx.guest_gma; + + bb_start_sva = (unsigned char *)wa_ctx->indirect_ctx.shadow_va + + wa_ctx->indirect_ctx.size; + + memcpy(bb_start_sva, per_ctx_start, CACHELINE_BYTES); + + return 0; +} + +int intel_gvt_scan_and_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + int ret; + + if (wa_ctx->indirect_ctx.size == 0) + return 0; + + ret = shadow_indirect_ctx(wa_ctx); + if (ret) { + gvt_err("fail to shadow indirect ctx\n"); + return ret; + } + + combine_wa_ctx(wa_ctx); + + ret = scan_wa_ctx(wa_ctx); + if (ret) { + gvt_err("scan wa ctx error\n"); + return ret; + } + + return 0; +} + +static struct cmd_info *find_cmd_entry_any_ring(struct intel_gvt *gvt, + unsigned int opcode, int rings) +{ + struct cmd_info *info = NULL; + unsigned int ring; + + for_each_set_bit(ring, (unsigned long *)&rings, I915_NUM_ENGINES) { + info = find_cmd_entry(gvt, opcode, ring); + if (info) + break; + } + return info; +} + +static int init_cmd_table(struct intel_gvt *gvt) +{ + int i; + struct cmd_entry *e; + struct cmd_info *info; + unsigned int gen_type; + + gen_type = intel_gvt_get_device_type(gvt); + + for (i = 0; i < ARRAY_SIZE(cmd_info); i++) { + if (!(cmd_info[i].devices & gen_type)) + continue; + + e = kzalloc(sizeof(*e), GFP_KERNEL); + if (!e) + return -ENOMEM; + + e->info = &cmd_info[i]; + info = find_cmd_entry_any_ring(gvt, + e->info->opcode, e->info->rings); + if (info) { + gvt_err("%s %s duplicated\n", e->info->name, + info->name); + return -EEXIST; + } + + INIT_HLIST_NODE(&e->hlist); + add_cmd_entry(gvt, e); + gvt_dbg_cmd("add %-30s op %04x flag %x devs %02x rings %02x\n", + e->info->name, e->info->opcode, e->info->flag, + e->info->devices, e->info->rings); + } + return 0; +} + +static void clean_cmd_table(struct intel_gvt *gvt) +{ + struct hlist_node *tmp; + struct cmd_entry *e; + int i; + + hash_for_each_safe(gvt->cmd_table, i, tmp, e, hlist) + kfree(e); + + hash_init(gvt->cmd_table); +} + +void intel_gvt_clean_cmd_parser(struct intel_gvt *gvt) +{ + clean_cmd_table(gvt); +} + +int intel_gvt_init_cmd_parser(struct intel_gvt *gvt) +{ + int ret; + + ret = init_cmd_table(gvt); + if (ret) { + intel_gvt_clean_cmd_parser(gvt); + return ret; + } + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/cmd_parser.h b/drivers/gpu/drm/i915/gvt/cmd_parser.h new file mode 100644 index 000000000000..bed33514103c --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/cmd_parser.h @@ -0,0 +1,49 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Kevin Tian <kevin.tian@intel.com> + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Yulei Zhang <yulei.zhang@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ +#ifndef _GVT_CMD_PARSER_H_ +#define _GVT_CMD_PARSER_H_ + +#define GVT_CMD_HASH_BITS 7 + +void intel_gvt_clean_cmd_parser(struct intel_gvt *gvt); + +int intel_gvt_init_cmd_parser(struct intel_gvt *gvt); + +int intel_gvt_scan_and_shadow_workload(struct intel_vgpu_workload *workload); + +int intel_gvt_scan_and_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx); + +#endif diff --git a/drivers/gpu/drm/i915/gvt/debug.h b/drivers/gpu/drm/i915/gvt/debug.h index 7ef412be665f..68cba7bd980a 100644 --- a/drivers/gpu/drm/i915/gvt/debug.h +++ b/drivers/gpu/drm/i915/gvt/debug.h @@ -24,11 +24,34 @@ #ifndef __GVT_DEBUG_H__ #define __GVT_DEBUG_H__ +#define gvt_err(fmt, args...) \ + DRM_ERROR("gvt: "fmt, ##args) + #define gvt_dbg_core(fmt, args...) \ DRM_DEBUG_DRIVER("gvt: core: "fmt, ##args) -/* - * Other GVT debug stuff will be introduced in the GVT device model patches. - */ +#define gvt_dbg_irq(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: irq: "fmt, ##args) + +#define gvt_dbg_mm(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: mm: "fmt, ##args) + +#define gvt_dbg_mmio(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: mmio: "fmt, ##args) + +#define gvt_dbg_dpy(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: dpy: "fmt, ##args) + +#define gvt_dbg_el(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: el: "fmt, ##args) + +#define gvt_dbg_sched(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: sched: "fmt, ##args) + +#define gvt_dbg_render(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: render: "fmt, ##args) + +#define gvt_dbg_cmd(fmt, args...) \ + DRM_DEBUG_DRIVER("gvt: cmd: "fmt, ##args) #endif diff --git a/drivers/gpu/drm/i915/gvt/display.c b/drivers/gpu/drm/i915/gvt/display.c new file mode 100644 index 000000000000..c0c884aeb30e --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/display.c @@ -0,0 +1,330 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Terrence Xu <terrence.xu@intel.com> + * Changbin Du <changbin.du@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +static int get_edp_pipe(struct intel_vgpu *vgpu) +{ + u32 data = vgpu_vreg(vgpu, _TRANS_DDI_FUNC_CTL_EDP); + int pipe = -1; + + switch (data & TRANS_DDI_EDP_INPUT_MASK) { + case TRANS_DDI_EDP_INPUT_A_ON: + case TRANS_DDI_EDP_INPUT_A_ONOFF: + pipe = PIPE_A; + break; + case TRANS_DDI_EDP_INPUT_B_ONOFF: + pipe = PIPE_B; + break; + case TRANS_DDI_EDP_INPUT_C_ONOFF: + pipe = PIPE_C; + break; + } + return pipe; +} + +static int edp_pipe_is_enabled(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + if (!(vgpu_vreg(vgpu, PIPECONF(_PIPE_EDP)) & PIPECONF_ENABLE)) + return 0; + + if (!(vgpu_vreg(vgpu, _TRANS_DDI_FUNC_CTL_EDP) & TRANS_DDI_FUNC_ENABLE)) + return 0; + return 1; +} + +static int pipe_is_enabled(struct intel_vgpu *vgpu, int pipe) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + if (WARN_ON(pipe < PIPE_A || pipe >= I915_MAX_PIPES)) + return -EINVAL; + + if (vgpu_vreg(vgpu, PIPECONF(pipe)) & PIPECONF_ENABLE) + return 1; + + if (edp_pipe_is_enabled(vgpu) && + get_edp_pipe(vgpu) == pipe) + return 1; + return 0; +} + +/* EDID with 1024x768 as its resolution */ +static unsigned char virtual_dp_monitor_edid[] = { + /*Header*/ + 0x00, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x00, + /* Vendor & Product Identification */ + 0x22, 0xf0, 0x54, 0x29, 0x00, 0x00, 0x00, 0x00, 0x04, 0x17, + /* Version & Revision */ + 0x01, 0x04, + /* Basic Display Parameters & Features */ + 0xa5, 0x34, 0x20, 0x78, 0x23, + /* Color Characteristics */ + 0xfc, 0x81, 0xa4, 0x55, 0x4d, 0x9d, 0x25, 0x12, 0x50, 0x54, + /* Established Timings: maximum resolution is 1024x768 */ + 0x21, 0x08, 0x00, + /* Standard Timings. All invalid */ + 0x00, 0xc0, 0x00, 0xc0, 0x00, 0x40, 0x00, 0x80, 0x00, 0x00, + 0x00, 0x40, 0x00, 0x00, 0x00, 0x01, + /* 18 Byte Data Blocks 1: invalid */ + 0x00, 0x00, 0x80, 0xa0, 0x70, 0xb0, + 0x23, 0x40, 0x30, 0x20, 0x36, 0x00, 0x06, 0x44, 0x21, 0x00, 0x00, 0x1a, + /* 18 Byte Data Blocks 2: invalid */ + 0x00, 0x00, 0x00, 0xfd, 0x00, 0x18, 0x3c, 0x18, 0x50, 0x11, 0x00, 0x0a, + 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, + /* 18 Byte Data Blocks 3: invalid */ + 0x00, 0x00, 0x00, 0xfc, 0x00, 0x48, + 0x50, 0x20, 0x5a, 0x52, 0x32, 0x34, 0x34, 0x30, 0x77, 0x0a, 0x20, 0x20, + /* 18 Byte Data Blocks 4: invalid */ + 0x00, 0x00, 0x00, 0xff, 0x00, 0x43, 0x4e, 0x34, 0x33, 0x30, 0x34, 0x30, + 0x44, 0x58, 0x51, 0x0a, 0x20, 0x20, + /* Extension Block Count */ + 0x00, + /* Checksum */ + 0xef, +}; + +#define DPCD_HEADER_SIZE 0xb + +static u8 dpcd_fix_data[DPCD_HEADER_SIZE] = { + 0x11, 0x0a, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 +}; + +static void emulate_monitor_status_change(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + vgpu_vreg(vgpu, SDEISR) &= ~(SDE_PORTB_HOTPLUG_CPT | + SDE_PORTC_HOTPLUG_CPT | + SDE_PORTD_HOTPLUG_CPT); + + if (IS_SKYLAKE(dev_priv)) + vgpu_vreg(vgpu, SDEISR) &= ~(SDE_PORTA_HOTPLUG_SPT | + SDE_PORTE_HOTPLUG_SPT); + + if (intel_vgpu_has_monitor_on_port(vgpu, PORT_B)) + vgpu_vreg(vgpu, SDEISR) |= SDE_PORTB_HOTPLUG_CPT; + + if (intel_vgpu_has_monitor_on_port(vgpu, PORT_C)) + vgpu_vreg(vgpu, SDEISR) |= SDE_PORTC_HOTPLUG_CPT; + + if (intel_vgpu_has_monitor_on_port(vgpu, PORT_D)) + vgpu_vreg(vgpu, SDEISR) |= SDE_PORTD_HOTPLUG_CPT; + + if (IS_SKYLAKE(dev_priv) && + intel_vgpu_has_monitor_on_port(vgpu, PORT_E)) { + vgpu_vreg(vgpu, SDEISR) |= SDE_PORTE_HOTPLUG_SPT; + } + + if (intel_vgpu_has_monitor_on_port(vgpu, PORT_A)) { + if (IS_BROADWELL(dev_priv)) + vgpu_vreg(vgpu, GEN8_DE_PORT_ISR) |= + GEN8_PORT_DP_A_HOTPLUG; + else + vgpu_vreg(vgpu, SDEISR) |= SDE_PORTA_HOTPLUG_SPT; + } +} + +static void clean_virtual_dp_monitor(struct intel_vgpu *vgpu, int port_num) +{ + struct intel_vgpu_port *port = intel_vgpu_port(vgpu, port_num); + + kfree(port->edid); + port->edid = NULL; + + kfree(port->dpcd); + port->dpcd = NULL; +} + +static int setup_virtual_dp_monitor(struct intel_vgpu *vgpu, int port_num, + int type) +{ + struct intel_vgpu_port *port = intel_vgpu_port(vgpu, port_num); + + port->edid = kzalloc(sizeof(*(port->edid)), GFP_KERNEL); + if (!port->edid) + return -ENOMEM; + + port->dpcd = kzalloc(sizeof(*(port->dpcd)), GFP_KERNEL); + if (!port->dpcd) { + kfree(port->edid); + return -ENOMEM; + } + + memcpy(port->edid->edid_block, virtual_dp_monitor_edid, + EDID_SIZE); + port->edid->data_valid = true; + + memcpy(port->dpcd->data, dpcd_fix_data, DPCD_HEADER_SIZE); + port->dpcd->data_valid = true; + port->dpcd->data[DPCD_SINK_COUNT] = 0x1; + port->type = type; + + emulate_monitor_status_change(vgpu); + return 0; +} + +/** + * intel_gvt_check_vblank_emulation - check if vblank emulation timer should + * be turned on/off when a virtual pipe is enabled/disabled. + * @gvt: a GVT device + * + * This function is used to turn on/off vblank timer according to currently + * enabled/disabled virtual pipes. + * + */ +void intel_gvt_check_vblank_emulation(struct intel_gvt *gvt) +{ + struct intel_gvt_irq *irq = &gvt->irq; + struct intel_vgpu *vgpu; + bool have_enabled_pipe = false; + int pipe, id; + + if (WARN_ON(!mutex_is_locked(&gvt->lock))) + return; + + hrtimer_cancel(&irq->vblank_timer.timer); + + for_each_active_vgpu(gvt, vgpu, id) { + for (pipe = 0; pipe < I915_MAX_PIPES; pipe++) { + have_enabled_pipe = + pipe_is_enabled(vgpu, pipe); + if (have_enabled_pipe) + break; + } + } + + if (have_enabled_pipe) + hrtimer_start(&irq->vblank_timer.timer, + ktime_add_ns(ktime_get(), irq->vblank_timer.period), + HRTIMER_MODE_ABS); +} + +static void emulate_vblank_on_pipe(struct intel_vgpu *vgpu, int pipe) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct intel_vgpu_irq *irq = &vgpu->irq; + int vblank_event[] = { + [PIPE_A] = PIPE_A_VBLANK, + [PIPE_B] = PIPE_B_VBLANK, + [PIPE_C] = PIPE_C_VBLANK, + }; + int event; + + if (pipe < PIPE_A || pipe > PIPE_C) + return; + + for_each_set_bit(event, irq->flip_done_event[pipe], + INTEL_GVT_EVENT_MAX) { + clear_bit(event, irq->flip_done_event[pipe]); + if (!pipe_is_enabled(vgpu, pipe)) + continue; + + vgpu_vreg(vgpu, PIPE_FLIPCOUNT_G4X(pipe))++; + intel_vgpu_trigger_virtual_event(vgpu, event); + } + + if (pipe_is_enabled(vgpu, pipe)) { + vgpu_vreg(vgpu, PIPE_FRMCOUNT_G4X(pipe))++; + intel_vgpu_trigger_virtual_event(vgpu, vblank_event[pipe]); + } +} + +static void emulate_vblank(struct intel_vgpu *vgpu) +{ + int pipe; + + for_each_pipe(vgpu->gvt->dev_priv, pipe) + emulate_vblank_on_pipe(vgpu, pipe); +} + +/** + * intel_gvt_emulate_vblank - trigger vblank events for vGPUs on GVT device + * @gvt: a GVT device + * + * This function is used to trigger vblank interrupts for vGPUs on GVT device + * + */ +void intel_gvt_emulate_vblank(struct intel_gvt *gvt) +{ + struct intel_vgpu *vgpu; + int id; + + if (WARN_ON(!mutex_is_locked(&gvt->lock))) + return; + + for_each_active_vgpu(gvt, vgpu, id) + emulate_vblank(vgpu); +} + +/** + * intel_vgpu_clean_display - clean vGPU virtual display emulation + * @vgpu: a vGPU + * + * This function is used to clean vGPU virtual display emulation stuffs + * + */ +void intel_vgpu_clean_display(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + if (IS_SKYLAKE(dev_priv)) + clean_virtual_dp_monitor(vgpu, PORT_D); + else + clean_virtual_dp_monitor(vgpu, PORT_B); +} + +/** + * intel_vgpu_init_display- initialize vGPU virtual display emulation + * @vgpu: a vGPU + * + * This function is used to initialize vGPU virtual display emulation stuffs + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_vgpu_init_display(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + intel_vgpu_init_i2c_edid(vgpu); + + if (IS_SKYLAKE(dev_priv)) + return setup_virtual_dp_monitor(vgpu, PORT_D, GVT_DP_D); + else + return setup_virtual_dp_monitor(vgpu, PORT_B, GVT_DP_B); +} diff --git a/drivers/gpu/drm/i915/gvt/display.h b/drivers/gpu/drm/i915/gvt/display.h new file mode 100644 index 000000000000..7a60cb848268 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/display.h @@ -0,0 +1,163 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Terrence Xu <terrence.xu@intel.com> + * Changbin Du <changbin.du@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#ifndef _GVT_DISPLAY_H_ +#define _GVT_DISPLAY_H_ + +#define SBI_REG_MAX 20 +#define DPCD_SIZE 0x700 + +#define intel_vgpu_port(vgpu, port) \ + (&(vgpu->display.ports[port])) + +#define intel_vgpu_has_monitor_on_port(vgpu, port) \ + (intel_vgpu_port(vgpu, port)->edid && \ + intel_vgpu_port(vgpu, port)->edid->data_valid) + +#define intel_vgpu_port_is_dp(vgpu, port) \ + ((intel_vgpu_port(vgpu, port)->type == GVT_DP_A) || \ + (intel_vgpu_port(vgpu, port)->type == GVT_DP_B) || \ + (intel_vgpu_port(vgpu, port)->type == GVT_DP_C) || \ + (intel_vgpu_port(vgpu, port)->type == GVT_DP_D)) + +#define INTEL_GVT_MAX_UEVENT_VARS 3 + +/* DPCD start */ +#define DPCD_SIZE 0x700 + +/* DPCD */ +#define DP_SET_POWER 0x600 +#define DP_SET_POWER_D0 0x1 +#define AUX_NATIVE_WRITE 0x8 +#define AUX_NATIVE_READ 0x9 + +#define AUX_NATIVE_REPLY_MASK (0x3 << 4) +#define AUX_NATIVE_REPLY_ACK (0x0 << 4) +#define AUX_NATIVE_REPLY_NAK (0x1 << 4) +#define AUX_NATIVE_REPLY_DEFER (0x2 << 4) + +#define AUX_BURST_SIZE 16 + +/* DPCD addresses */ +#define DPCD_REV 0x000 +#define DPCD_MAX_LINK_RATE 0x001 +#define DPCD_MAX_LANE_COUNT 0x002 + +#define DPCD_TRAINING_PATTERN_SET 0x102 +#define DPCD_SINK_COUNT 0x200 +#define DPCD_LANE0_1_STATUS 0x202 +#define DPCD_LANE2_3_STATUS 0x203 +#define DPCD_LANE_ALIGN_STATUS_UPDATED 0x204 +#define DPCD_SINK_STATUS 0x205 + +/* link training */ +#define DPCD_TRAINING_PATTERN_SET_MASK 0x03 +#define DPCD_LINK_TRAINING_DISABLED 0x00 +#define DPCD_TRAINING_PATTERN_1 0x01 +#define DPCD_TRAINING_PATTERN_2 0x02 + +#define DPCD_CP_READY_MASK (1 << 6) + +/* lane status */ +#define DPCD_LANES_CR_DONE 0x11 +#define DPCD_LANES_EQ_DONE 0x22 +#define DPCD_SYMBOL_LOCKED 0x44 + +#define DPCD_INTERLANE_ALIGN_DONE 0x01 + +#define DPCD_SINK_IN_SYNC 0x03 +/* DPCD end */ + +#define SBI_RESPONSE_MASK 0x3 +#define SBI_RESPONSE_SHIFT 0x1 +#define SBI_STAT_MASK 0x1 +#define SBI_STAT_SHIFT 0x0 +#define SBI_OPCODE_SHIFT 8 +#define SBI_OPCODE_MASK (0xff << SBI_OPCODE_SHIFT) +#define SBI_CMD_IORD 2 +#define SBI_CMD_IOWR 3 +#define SBI_CMD_CRRD 6 +#define SBI_CMD_CRWR 7 +#define SBI_ADDR_OFFSET_SHIFT 16 +#define SBI_ADDR_OFFSET_MASK (0xffff << SBI_ADDR_OFFSET_SHIFT) + +struct intel_vgpu_sbi_register { + unsigned int offset; + u32 value; +}; + +struct intel_vgpu_sbi { + int number; + struct intel_vgpu_sbi_register registers[SBI_REG_MAX]; +}; + +enum intel_gvt_plane_type { + PRIMARY_PLANE = 0, + CURSOR_PLANE, + SPRITE_PLANE, + MAX_PLANE +}; + +struct intel_vgpu_dpcd_data { + bool data_valid; + u8 data[DPCD_SIZE]; +}; + +enum intel_vgpu_port_type { + GVT_CRT = 0, + GVT_DP_A, + GVT_DP_B, + GVT_DP_C, + GVT_DP_D, + GVT_HDMI_B, + GVT_HDMI_C, + GVT_HDMI_D, + GVT_PORT_MAX +}; + +struct intel_vgpu_port { + /* per display EDID information */ + struct intel_vgpu_edid_data *edid; + /* per display DPCD information */ + struct intel_vgpu_dpcd_data *dpcd; + int type; +}; + +void intel_gvt_emulate_vblank(struct intel_gvt *gvt); +void intel_gvt_check_vblank_emulation(struct intel_gvt *gvt); + +int intel_vgpu_init_display(struct intel_vgpu *vgpu); +void intel_vgpu_clean_display(struct intel_vgpu *vgpu); + +#endif diff --git a/drivers/gpu/drm/i915/gvt/edid.c b/drivers/gpu/drm/i915/gvt/edid.c new file mode 100644 index 000000000000..bda85dff7b2a --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/edid.c @@ -0,0 +1,531 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Terrence Xu <terrence.xu@intel.com> + * Changbin Du <changbin.du@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +#define GMBUS1_TOTAL_BYTES_SHIFT 16 +#define GMBUS1_TOTAL_BYTES_MASK 0x1ff +#define gmbus1_total_byte_count(v) (((v) >> \ + GMBUS1_TOTAL_BYTES_SHIFT) & GMBUS1_TOTAL_BYTES_MASK) +#define gmbus1_slave_addr(v) (((v) & 0xff) >> 1) +#define gmbus1_slave_index(v) (((v) >> 8) & 0xff) +#define gmbus1_bus_cycle(v) (((v) >> 25) & 0x7) + +/* GMBUS0 bits definitions */ +#define _GMBUS_PIN_SEL_MASK (0x7) + +static unsigned char edid_get_byte(struct intel_vgpu *vgpu) +{ + struct intel_vgpu_i2c_edid *edid = &vgpu->display.i2c_edid; + unsigned char chr = 0; + + if (edid->state == I2C_NOT_SPECIFIED || !edid->slave_selected) { + gvt_err("Driver tries to read EDID without proper sequence!\n"); + return 0; + } + if (edid->current_edid_read >= EDID_SIZE) { + gvt_err("edid_get_byte() exceeds the size of EDID!\n"); + return 0; + } + + if (!edid->edid_available) { + gvt_err("Reading EDID but EDID is not available!\n"); + return 0; + } + + if (intel_vgpu_has_monitor_on_port(vgpu, edid->port)) { + struct intel_vgpu_edid_data *edid_data = + intel_vgpu_port(vgpu, edid->port)->edid; + + chr = edid_data->edid_block[edid->current_edid_read]; + edid->current_edid_read++; + } else { + gvt_err("No EDID available during the reading?\n"); + } + return chr; +} + +static inline int get_port_from_gmbus0(u32 gmbus0) +{ + int port_select = gmbus0 & _GMBUS_PIN_SEL_MASK; + int port = -EINVAL; + + if (port_select == 2) + port = PORT_E; + else if (port_select == 4) + port = PORT_C; + else if (port_select == 5) + port = PORT_B; + else if (port_select == 6) + port = PORT_D; + return port; +} + +static void reset_gmbus_controller(struct intel_vgpu *vgpu) +{ + vgpu_vreg(vgpu, PCH_GMBUS2) = GMBUS_HW_RDY; + if (!vgpu->display.i2c_edid.edid_available) + vgpu_vreg(vgpu, PCH_GMBUS2) |= GMBUS_SATOER; + vgpu->display.i2c_edid.gmbus.phase = GMBUS_IDLE_PHASE; +} + +/* GMBUS0 */ +static int gmbus0_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + int port, pin_select; + + memcpy(&vgpu_vreg(vgpu, offset), p_data, bytes); + + pin_select = vgpu_vreg(vgpu, offset) & _GMBUS_PIN_SEL_MASK; + + intel_vgpu_init_i2c_edid(vgpu); + + if (pin_select == 0) + return 0; + + port = get_port_from_gmbus0(pin_select); + if (WARN_ON(port < 0)) + return 0; + + vgpu->display.i2c_edid.state = I2C_GMBUS; + vgpu->display.i2c_edid.gmbus.phase = GMBUS_IDLE_PHASE; + + vgpu_vreg(vgpu, PCH_GMBUS2) &= ~GMBUS_ACTIVE; + vgpu_vreg(vgpu, PCH_GMBUS2) |= GMBUS_HW_RDY | GMBUS_HW_WAIT_PHASE; + + if (intel_vgpu_has_monitor_on_port(vgpu, port) && + !intel_vgpu_port_is_dp(vgpu, port)) { + vgpu->display.i2c_edid.port = port; + vgpu->display.i2c_edid.edid_available = true; + vgpu_vreg(vgpu, PCH_GMBUS2) &= ~GMBUS_SATOER; + } else + vgpu_vreg(vgpu, PCH_GMBUS2) |= GMBUS_SATOER; + return 0; +} + +static int gmbus1_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + struct intel_vgpu_i2c_edid *i2c_edid = &vgpu->display.i2c_edid; + u32 slave_addr; + u32 wvalue = *(u32 *)p_data; + + if (vgpu_vreg(vgpu, offset) & GMBUS_SW_CLR_INT) { + if (!(wvalue & GMBUS_SW_CLR_INT)) { + vgpu_vreg(vgpu, offset) &= ~GMBUS_SW_CLR_INT; + reset_gmbus_controller(vgpu); + } + /* + * TODO: "This bit is cleared to zero when an event + * causes the HW_RDY bit transition to occur " + */ + } else { + /* + * per bspec setting this bit can cause: + * 1) INT status bit cleared + * 2) HW_RDY bit asserted + */ + if (wvalue & GMBUS_SW_CLR_INT) { + vgpu_vreg(vgpu, PCH_GMBUS2) &= ~GMBUS_INT; + vgpu_vreg(vgpu, PCH_GMBUS2) |= GMBUS_HW_RDY; + } + + /* For virtualization, we suppose that HW is always ready, + * so GMBUS_SW_RDY should always be cleared + */ + if (wvalue & GMBUS_SW_RDY) + wvalue &= ~GMBUS_SW_RDY; + + i2c_edid->gmbus.total_byte_count = + gmbus1_total_byte_count(wvalue); + slave_addr = gmbus1_slave_addr(wvalue); + + /* vgpu gmbus only support EDID */ + if (slave_addr == EDID_ADDR) { + i2c_edid->slave_selected = true; + } else if (slave_addr != 0) { + gvt_dbg_dpy( + "vgpu%d: unsupported gmbus slave addr(0x%x)\n" + " gmbus operations will be ignored.\n", + vgpu->id, slave_addr); + } + + if (wvalue & GMBUS_CYCLE_INDEX) + i2c_edid->current_edid_read = + gmbus1_slave_index(wvalue); + + i2c_edid->gmbus.cycle_type = gmbus1_bus_cycle(wvalue); + switch (gmbus1_bus_cycle(wvalue)) { + case GMBUS_NOCYCLE: + break; + case GMBUS_STOP: + /* From spec: + * This can only cause a STOP to be generated + * if a GMBUS cycle is generated, the GMBUS is + * currently in a data/wait/idle phase, or it is in a + * WAIT phase + */ + if (gmbus1_bus_cycle(vgpu_vreg(vgpu, offset)) + != GMBUS_NOCYCLE) { + intel_vgpu_init_i2c_edid(vgpu); + /* After the 'stop' cycle, hw state would become + * 'stop phase' and then 'idle phase' after a + * few milliseconds. In emulation, we just set + * it as 'idle phase' ('stop phase' is not + * visible in gmbus interface) + */ + i2c_edid->gmbus.phase = GMBUS_IDLE_PHASE; + vgpu_vreg(vgpu, PCH_GMBUS2) &= ~GMBUS_ACTIVE; + } + break; + case NIDX_NS_W: + case IDX_NS_W: + case NIDX_STOP: + case IDX_STOP: + /* From hw spec the GMBUS phase + * transition like this: + * START (-->INDEX) -->DATA + */ + i2c_edid->gmbus.phase = GMBUS_DATA_PHASE; + vgpu_vreg(vgpu, PCH_GMBUS2) |= GMBUS_ACTIVE; + break; + default: + gvt_err("Unknown/reserved GMBUS cycle detected!\n"); + break; + } + /* + * From hw spec the WAIT state will be + * cleared: + * (1) in a new GMBUS cycle + * (2) by generating a stop + */ + vgpu_vreg(vgpu, offset) = wvalue; + } + return 0; +} + +static int gmbus3_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + WARN_ON(1); + return 0; +} + +static int gmbus3_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + int i; + unsigned char byte_data; + struct intel_vgpu_i2c_edid *i2c_edid = &vgpu->display.i2c_edid; + int byte_left = i2c_edid->gmbus.total_byte_count - + i2c_edid->current_edid_read; + int byte_count = byte_left; + u32 reg_data = 0; + + /* Data can only be recevied if previous settings correct */ + if (vgpu_vreg(vgpu, PCH_GMBUS1) & GMBUS_SLAVE_READ) { + if (byte_left <= 0) { + memcpy(p_data, &vgpu_vreg(vgpu, offset), bytes); + return 0; + } + + if (byte_count > 4) + byte_count = 4; + for (i = 0; i < byte_count; i++) { + byte_data = edid_get_byte(vgpu); + reg_data |= (byte_data << (i << 3)); + } + + memcpy(&vgpu_vreg(vgpu, offset), ®_data, byte_count); + memcpy(p_data, &vgpu_vreg(vgpu, offset), bytes); + + if (byte_left <= 4) { + switch (i2c_edid->gmbus.cycle_type) { + case NIDX_STOP: + case IDX_STOP: + i2c_edid->gmbus.phase = GMBUS_IDLE_PHASE; + break; + case NIDX_NS_W: + case IDX_NS_W: + default: + i2c_edid->gmbus.phase = GMBUS_WAIT_PHASE; + break; + } + intel_vgpu_init_i2c_edid(vgpu); + } + /* + * Read GMBUS3 during send operation, + * return the latest written value + */ + } else { + memcpy(p_data, &vgpu_vreg(vgpu, offset), bytes); + gvt_err("vgpu%d: warning: gmbus3 read with nothing returned\n", + vgpu->id); + } + return 0; +} + +static int gmbus2_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 value = vgpu_vreg(vgpu, offset); + + if (!(vgpu_vreg(vgpu, offset) & GMBUS_INUSE)) + vgpu_vreg(vgpu, offset) |= GMBUS_INUSE; + memcpy(p_data, (void *)&value, bytes); + return 0; +} + +static int gmbus2_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 wvalue = *(u32 *)p_data; + + if (wvalue & GMBUS_INUSE) + vgpu_vreg(vgpu, offset) &= ~GMBUS_INUSE; + /* All other bits are read-only */ + return 0; +} + +/** + * intel_gvt_i2c_handle_gmbus_read - emulate gmbus register mmio read + * @vgpu: a vGPU + * + * This function is used to emulate gmbus register mmio read + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_gvt_i2c_handle_gmbus_read(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + if (WARN_ON(bytes > 8 && (offset & (bytes - 1)))) + return -EINVAL; + + if (offset == i915_mmio_reg_offset(PCH_GMBUS2)) + return gmbus2_mmio_read(vgpu, offset, p_data, bytes); + else if (offset == i915_mmio_reg_offset(PCH_GMBUS3)) + return gmbus3_mmio_read(vgpu, offset, p_data, bytes); + + memcpy(p_data, &vgpu_vreg(vgpu, offset), bytes); + return 0; +} + +/** + * intel_gvt_i2c_handle_gmbus_write - emulate gmbus register mmio write + * @vgpu: a vGPU + * + * This function is used to emulate gmbus register mmio write + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_gvt_i2c_handle_gmbus_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + if (WARN_ON(bytes > 8 && (offset & (bytes - 1)))) + return -EINVAL; + + if (offset == i915_mmio_reg_offset(PCH_GMBUS0)) + return gmbus0_mmio_write(vgpu, offset, p_data, bytes); + else if (offset == i915_mmio_reg_offset(PCH_GMBUS1)) + return gmbus1_mmio_write(vgpu, offset, p_data, bytes); + else if (offset == i915_mmio_reg_offset(PCH_GMBUS2)) + return gmbus2_mmio_write(vgpu, offset, p_data, bytes); + else if (offset == i915_mmio_reg_offset(PCH_GMBUS3)) + return gmbus3_mmio_write(vgpu, offset, p_data, bytes); + + memcpy(&vgpu_vreg(vgpu, offset), p_data, bytes); + return 0; +} + +enum { + AUX_CH_CTL = 0, + AUX_CH_DATA1, + AUX_CH_DATA2, + AUX_CH_DATA3, + AUX_CH_DATA4, + AUX_CH_DATA5 +}; + +static inline int get_aux_ch_reg(unsigned int offset) +{ + int reg; + + switch (offset & 0xff) { + case 0x10: + reg = AUX_CH_CTL; + break; + case 0x14: + reg = AUX_CH_DATA1; + break; + case 0x18: + reg = AUX_CH_DATA2; + break; + case 0x1c: + reg = AUX_CH_DATA3; + break; + case 0x20: + reg = AUX_CH_DATA4; + break; + case 0x24: + reg = AUX_CH_DATA5; + break; + default: + reg = -1; + break; + } + return reg; +} + +#define AUX_CTL_MSG_LENGTH(reg) \ + ((reg & DP_AUX_CH_CTL_MESSAGE_SIZE_MASK) >> \ + DP_AUX_CH_CTL_MESSAGE_SIZE_SHIFT) + +/** + * intel_gvt_i2c_handle_aux_ch_write - emulate AUX channel register write + * @vgpu: a vGPU + * + * This function is used to emulate AUX channel register write + * + */ +void intel_gvt_i2c_handle_aux_ch_write(struct intel_vgpu *vgpu, + int port_idx, + unsigned int offset, + void *p_data) +{ + struct intel_vgpu_i2c_edid *i2c_edid = &vgpu->display.i2c_edid; + int msg_length, ret_msg_size; + int msg, addr, ctrl, op; + u32 value = *(u32 *)p_data; + int aux_data_for_write = 0; + int reg = get_aux_ch_reg(offset); + + if (reg != AUX_CH_CTL) { + vgpu_vreg(vgpu, offset) = value; + return; + } + + msg_length = AUX_CTL_MSG_LENGTH(value); + // check the msg in DATA register. + msg = vgpu_vreg(vgpu, offset + 4); + addr = (msg >> 8) & 0xffff; + ctrl = (msg >> 24) & 0xff; + op = ctrl >> 4; + if (!(value & DP_AUX_CH_CTL_SEND_BUSY)) { + /* The ctl write to clear some states */ + return; + } + + /* Always set the wanted value for vms. */ + ret_msg_size = (((op & 0x1) == GVT_AUX_I2C_READ) ? 2 : 1); + vgpu_vreg(vgpu, offset) = + DP_AUX_CH_CTL_DONE | + ((ret_msg_size << DP_AUX_CH_CTL_MESSAGE_SIZE_SHIFT) & + DP_AUX_CH_CTL_MESSAGE_SIZE_MASK); + + if (msg_length == 3) { + if (!(op & GVT_AUX_I2C_MOT)) { + /* stop */ + intel_vgpu_init_i2c_edid(vgpu); + } else { + /* start or restart */ + i2c_edid->aux_ch.i2c_over_aux_ch = true; + i2c_edid->aux_ch.aux_ch_mot = true; + if (addr == 0) { + /* reset the address */ + intel_vgpu_init_i2c_edid(vgpu); + } else if (addr == EDID_ADDR) { + i2c_edid->state = I2C_AUX_CH; + i2c_edid->port = port_idx; + i2c_edid->slave_selected = true; + if (intel_vgpu_has_monitor_on_port(vgpu, + port_idx) && + intel_vgpu_port_is_dp(vgpu, port_idx)) + i2c_edid->edid_available = true; + } + } + } else if ((op & 0x1) == GVT_AUX_I2C_WRITE) { + /* TODO + * We only support EDID reading from I2C_over_AUX. And + * we do not expect the index mode to be used. Right now + * the WRITE operation is ignored. It is good enough to + * support the gfx driver to do EDID access. + */ + } else { + if (WARN_ON((op & 0x1) != GVT_AUX_I2C_READ)) + return; + if (WARN_ON(msg_length != 4)) + return; + if (i2c_edid->edid_available && i2c_edid->slave_selected) { + unsigned char val = edid_get_byte(vgpu); + + aux_data_for_write = (val << 16); + } + } + /* write the return value in AUX_CH_DATA reg which includes: + * ACK of I2C_WRITE + * returned byte if it is READ + */ + aux_data_for_write |= GVT_AUX_I2C_REPLY_ACK << 24; + vgpu_vreg(vgpu, offset + 4) = aux_data_for_write; +} + +/** + * intel_vgpu_init_i2c_edid - initialize vGPU i2c edid emulation + * @vgpu: a vGPU + * + * This function is used to initialize vGPU i2c edid emulation stuffs + * + */ +void intel_vgpu_init_i2c_edid(struct intel_vgpu *vgpu) +{ + struct intel_vgpu_i2c_edid *edid = &vgpu->display.i2c_edid; + + edid->state = I2C_NOT_SPECIFIED; + + edid->port = -1; + edid->slave_selected = false; + edid->edid_available = false; + edid->current_edid_read = 0; + + memset(&edid->gmbus, 0, sizeof(struct intel_vgpu_i2c_gmbus)); + + edid->aux_ch.i2c_over_aux_ch = false; + edid->aux_ch.aux_ch_mot = false; +} diff --git a/drivers/gpu/drm/i915/gvt/edid.h b/drivers/gpu/drm/i915/gvt/edid.h new file mode 100644 index 000000000000..f6dfc8b795ec --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/edid.h @@ -0,0 +1,150 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Terrence Xu <terrence.xu@intel.com> + * Changbin Du <changbin.du@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#ifndef _GVT_EDID_H_ +#define _GVT_EDID_H_ + +#define EDID_SIZE 128 +#define EDID_ADDR 0x50 /* Linux hvm EDID addr */ + +#define GVT_AUX_NATIVE_WRITE 0x8 +#define GVT_AUX_NATIVE_READ 0x9 +#define GVT_AUX_I2C_WRITE 0x0 +#define GVT_AUX_I2C_READ 0x1 +#define GVT_AUX_I2C_STATUS 0x2 +#define GVT_AUX_I2C_MOT 0x4 +#define GVT_AUX_I2C_REPLY_ACK 0x0 + +struct intel_vgpu_edid_data { + bool data_valid; + unsigned char edid_block[EDID_SIZE]; +}; + +enum gmbus_cycle_type { + GMBUS_NOCYCLE = 0x0, + NIDX_NS_W = 0x1, + IDX_NS_W = 0x3, + GMBUS_STOP = 0x4, + NIDX_STOP = 0x5, + IDX_STOP = 0x7 +}; + +/* + * States of GMBUS + * + * GMBUS0-3 could be related to the EDID virtualization. Another two GMBUS + * registers, GMBUS4 (interrupt mask) and GMBUS5 (2 byte indes register), are + * not considered here. Below describes the usage of GMBUS registers that are + * cared by the EDID virtualization + * + * GMBUS0: + * R/W + * port selection. value of bit0 - bit2 corresponds to the GPIO registers. + * + * GMBUS1: + * R/W Protect + * Command and Status. + * bit0 is the direction bit: 1 is read; 0 is write. + * bit1 - bit7 is slave 7-bit address. + * bit16 - bit24 total byte count (ignore?) + * + * GMBUS2: + * Most of bits are read only except bit 15 (IN_USE) + * Status register + * bit0 - bit8 current byte count + * bit 11: hardware ready; + * + * GMBUS3: + * Read/Write + * Data for transfer + */ + +/* From hw specs, Other phases like START, ADDRESS, INDEX + * are invisible to GMBUS MMIO interface. So no definitions + * in below enum types + */ +enum gvt_gmbus_phase { + GMBUS_IDLE_PHASE = 0, + GMBUS_DATA_PHASE, + GMBUS_WAIT_PHASE, + //GMBUS_STOP_PHASE, + GMBUS_MAX_PHASE +}; + +struct intel_vgpu_i2c_gmbus { + unsigned int total_byte_count; /* from GMBUS1 */ + enum gmbus_cycle_type cycle_type; + enum gvt_gmbus_phase phase; +}; + +struct intel_vgpu_i2c_aux_ch { + bool i2c_over_aux_ch; + bool aux_ch_mot; +}; + +enum i2c_state { + I2C_NOT_SPECIFIED = 0, + I2C_GMBUS = 1, + I2C_AUX_CH = 2 +}; + +/* I2C sequences cannot interleave. + * GMBUS and AUX_CH sequences cannot interleave. + */ +struct intel_vgpu_i2c_edid { + enum i2c_state state; + + unsigned int port; + bool slave_selected; + bool edid_available; + unsigned int current_edid_read; + + struct intel_vgpu_i2c_gmbus gmbus; + struct intel_vgpu_i2c_aux_ch aux_ch; +}; + +void intel_vgpu_init_i2c_edid(struct intel_vgpu *vgpu); + +int intel_gvt_i2c_handle_gmbus_read(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes); + +int intel_gvt_i2c_handle_gmbus_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes); + +void intel_gvt_i2c_handle_aux_ch_write(struct intel_vgpu *vgpu, + int port_idx, + unsigned int offset, + void *p_data); + +#endif /*_GVT_EDID_H_*/ diff --git a/drivers/gpu/drm/i915/gvt/execlist.c b/drivers/gpu/drm/i915/gvt/execlist.c new file mode 100644 index 000000000000..f32bb6f6495c --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/execlist.c @@ -0,0 +1,858 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +#define _EL_OFFSET_STATUS 0x234 +#define _EL_OFFSET_STATUS_BUF 0x370 +#define _EL_OFFSET_STATUS_PTR 0x3A0 + +#define execlist_ring_mmio(gvt, ring_id, offset) \ + (gvt->dev_priv->engine[ring_id]->mmio_base + (offset)) + +#define valid_context(ctx) ((ctx)->valid) +#define same_context(a, b) (((a)->context_id == (b)->context_id) && \ + ((a)->lrca == (b)->lrca)) + +static int context_switch_events[] = { + [RCS] = RCS_AS_CONTEXT_SWITCH, + [BCS] = BCS_AS_CONTEXT_SWITCH, + [VCS] = VCS_AS_CONTEXT_SWITCH, + [VCS2] = VCS2_AS_CONTEXT_SWITCH, + [VECS] = VECS_AS_CONTEXT_SWITCH, +}; + +static int ring_id_to_context_switch_event(int ring_id) +{ + if (WARN_ON(ring_id < RCS && ring_id > + ARRAY_SIZE(context_switch_events))) + return -EINVAL; + + return context_switch_events[ring_id]; +} + +static void switch_virtual_execlist_slot(struct intel_vgpu_execlist *execlist) +{ + gvt_dbg_el("[before] running slot %d/context %x pending slot %d\n", + execlist->running_slot ? + execlist->running_slot->index : -1, + execlist->running_context ? + execlist->running_context->context_id : 0, + execlist->pending_slot ? + execlist->pending_slot->index : -1); + + execlist->running_slot = execlist->pending_slot; + execlist->pending_slot = NULL; + execlist->running_context = execlist->running_context ? + &execlist->running_slot->ctx[0] : NULL; + + gvt_dbg_el("[after] running slot %d/context %x pending slot %d\n", + execlist->running_slot ? + execlist->running_slot->index : -1, + execlist->running_context ? + execlist->running_context->context_id : 0, + execlist->pending_slot ? + execlist->pending_slot->index : -1); +} + +static void emulate_execlist_status(struct intel_vgpu_execlist *execlist) +{ + struct intel_vgpu_execlist_slot *running = execlist->running_slot; + struct intel_vgpu_execlist_slot *pending = execlist->pending_slot; + struct execlist_ctx_descriptor_format *desc = execlist->running_context; + struct intel_vgpu *vgpu = execlist->vgpu; + struct execlist_status_format status; + int ring_id = execlist->ring_id; + u32 status_reg = execlist_ring_mmio(vgpu->gvt, + ring_id, _EL_OFFSET_STATUS); + + status.ldw = vgpu_vreg(vgpu, status_reg); + status.udw = vgpu_vreg(vgpu, status_reg + 4); + + if (running) { + status.current_execlist_pointer = !!running->index; + status.execlist_write_pointer = !!!running->index; + status.execlist_0_active = status.execlist_0_valid = + !!!(running->index); + status.execlist_1_active = status.execlist_1_valid = + !!(running->index); + } else { + status.context_id = 0; + status.execlist_0_active = status.execlist_0_valid = 0; + status.execlist_1_active = status.execlist_1_valid = 0; + } + + status.context_id = desc ? desc->context_id : 0; + status.execlist_queue_full = !!(pending); + + vgpu_vreg(vgpu, status_reg) = status.ldw; + vgpu_vreg(vgpu, status_reg + 4) = status.udw; + + gvt_dbg_el("vgpu%d: status reg offset %x ldw %x udw %x\n", + vgpu->id, status_reg, status.ldw, status.udw); +} + +static void emulate_csb_update(struct intel_vgpu_execlist *execlist, + struct execlist_context_status_format *status, + bool trigger_interrupt_later) +{ + struct intel_vgpu *vgpu = execlist->vgpu; + int ring_id = execlist->ring_id; + struct execlist_context_status_pointer_format ctx_status_ptr; + u32 write_pointer; + u32 ctx_status_ptr_reg, ctx_status_buf_reg, offset; + + ctx_status_ptr_reg = execlist_ring_mmio(vgpu->gvt, ring_id, + _EL_OFFSET_STATUS_PTR); + ctx_status_buf_reg = execlist_ring_mmio(vgpu->gvt, ring_id, + _EL_OFFSET_STATUS_BUF); + + ctx_status_ptr.dw = vgpu_vreg(vgpu, ctx_status_ptr_reg); + + write_pointer = ctx_status_ptr.write_ptr; + + if (write_pointer == 0x7) + write_pointer = 0; + else { + ++write_pointer; + write_pointer %= 0x6; + } + + offset = ctx_status_buf_reg + write_pointer * 8; + + vgpu_vreg(vgpu, offset) = status->ldw; + vgpu_vreg(vgpu, offset + 4) = status->udw; + + ctx_status_ptr.write_ptr = write_pointer; + vgpu_vreg(vgpu, ctx_status_ptr_reg) = ctx_status_ptr.dw; + + gvt_dbg_el("vgpu%d: w pointer %u reg %x csb l %x csb h %x\n", + vgpu->id, write_pointer, offset, status->ldw, status->udw); + + if (trigger_interrupt_later) + return; + + intel_vgpu_trigger_virtual_event(vgpu, + ring_id_to_context_switch_event(execlist->ring_id)); +} + +static int emulate_execlist_ctx_schedule_out( + struct intel_vgpu_execlist *execlist, + struct execlist_ctx_descriptor_format *ctx) +{ + struct intel_vgpu_execlist_slot *running = execlist->running_slot; + struct intel_vgpu_execlist_slot *pending = execlist->pending_slot; + struct execlist_ctx_descriptor_format *ctx0 = &running->ctx[0]; + struct execlist_ctx_descriptor_format *ctx1 = &running->ctx[1]; + struct execlist_context_status_format status; + + memset(&status, 0, sizeof(status)); + + gvt_dbg_el("schedule out context id %x\n", ctx->context_id); + + if (WARN_ON(!same_context(ctx, execlist->running_context))) { + gvt_err("schedule out context is not running context," + "ctx id %x running ctx id %x\n", + ctx->context_id, + execlist->running_context->context_id); + return -EINVAL; + } + + /* ctx1 is valid, ctx0/ctx is scheduled-out -> element switch */ + if (valid_context(ctx1) && same_context(ctx0, ctx)) { + gvt_dbg_el("ctx 1 valid, ctx/ctx 0 is scheduled-out\n"); + + execlist->running_context = ctx1; + + emulate_execlist_status(execlist); + + status.context_complete = status.element_switch = 1; + status.context_id = ctx->context_id; + + emulate_csb_update(execlist, &status, false); + /* + * ctx1 is not valid, ctx == ctx0 + * ctx1 is valid, ctx1 == ctx + * --> last element is finished + * emulate: + * active-to-idle if there is *no* pending execlist + * context-complete if there *is* pending execlist + */ + } else if ((!valid_context(ctx1) && same_context(ctx0, ctx)) + || (valid_context(ctx1) && same_context(ctx1, ctx))) { + gvt_dbg_el("need to switch virtual execlist slot\n"); + + switch_virtual_execlist_slot(execlist); + + emulate_execlist_status(execlist); + + status.context_complete = status.active_to_idle = 1; + status.context_id = ctx->context_id; + + if (!pending) { + emulate_csb_update(execlist, &status, false); + } else { + emulate_csb_update(execlist, &status, true); + + memset(&status, 0, sizeof(status)); + + status.idle_to_active = 1; + status.context_id = 0; + + emulate_csb_update(execlist, &status, false); + } + } else { + WARN_ON(1); + return -EINVAL; + } + + return 0; +} + +static struct intel_vgpu_execlist_slot *get_next_execlist_slot( + struct intel_vgpu_execlist *execlist) +{ + struct intel_vgpu *vgpu = execlist->vgpu; + int ring_id = execlist->ring_id; + u32 status_reg = execlist_ring_mmio(vgpu->gvt, ring_id, + _EL_OFFSET_STATUS); + struct execlist_status_format status; + + status.ldw = vgpu_vreg(vgpu, status_reg); + status.udw = vgpu_vreg(vgpu, status_reg + 4); + + if (status.execlist_queue_full) { + gvt_err("virtual execlist slots are full\n"); + return NULL; + } + + return &execlist->slot[status.execlist_write_pointer]; +} + +static int emulate_execlist_schedule_in(struct intel_vgpu_execlist *execlist, + struct execlist_ctx_descriptor_format ctx[2]) +{ + struct intel_vgpu_execlist_slot *running = execlist->running_slot; + struct intel_vgpu_execlist_slot *slot = + get_next_execlist_slot(execlist); + + struct execlist_ctx_descriptor_format *ctx0, *ctx1; + struct execlist_context_status_format status; + + gvt_dbg_el("emulate schedule-in\n"); + + if (!slot) { + gvt_err("no available execlist slot\n"); + return -EINVAL; + } + + memset(&status, 0, sizeof(status)); + memset(slot->ctx, 0, sizeof(slot->ctx)); + + slot->ctx[0] = ctx[0]; + slot->ctx[1] = ctx[1]; + + gvt_dbg_el("alloc slot index %d ctx 0 %x ctx 1 %x\n", + slot->index, ctx[0].context_id, + ctx[1].context_id); + + /* + * no running execlist, make this write bundle as running execlist + * -> idle-to-active + */ + if (!running) { + gvt_dbg_el("no current running execlist\n"); + + execlist->running_slot = slot; + execlist->pending_slot = NULL; + execlist->running_context = &slot->ctx[0]; + + gvt_dbg_el("running slot index %d running context %x\n", + execlist->running_slot->index, + execlist->running_context->context_id); + + emulate_execlist_status(execlist); + + status.idle_to_active = 1; + status.context_id = 0; + + emulate_csb_update(execlist, &status, false); + return 0; + } + + ctx0 = &running->ctx[0]; + ctx1 = &running->ctx[1]; + + gvt_dbg_el("current running slot index %d ctx 0 %x ctx 1 %x\n", + running->index, ctx0->context_id, ctx1->context_id); + + /* + * already has an running execlist + * a. running ctx1 is valid, + * ctx0 is finished, and running ctx1 == new execlist ctx[0] + * b. running ctx1 is not valid, + * ctx0 == new execlist ctx[0] + * ----> lite-restore + preempted + */ + if ((valid_context(ctx1) && same_context(ctx1, &slot->ctx[0]) && + /* condition a */ + (!same_context(ctx0, execlist->running_context))) || + (!valid_context(ctx1) && + same_context(ctx0, &slot->ctx[0]))) { /* condition b */ + gvt_dbg_el("need to switch virtual execlist slot\n"); + + execlist->pending_slot = slot; + switch_virtual_execlist_slot(execlist); + + emulate_execlist_status(execlist); + + status.lite_restore = status.preempted = 1; + status.context_id = ctx[0].context_id; + + emulate_csb_update(execlist, &status, false); + } else { + gvt_dbg_el("emulate as pending slot\n"); + /* + * otherwise + * --> emulate pending execlist exist + but no preemption case + */ + execlist->pending_slot = slot; + emulate_execlist_status(execlist); + } + return 0; +} + +static void free_workload(struct intel_vgpu_workload *workload) +{ + intel_vgpu_unpin_mm(workload->shadow_mm); + intel_gvt_mm_unreference(workload->shadow_mm); + kmem_cache_free(workload->vgpu->workloads, workload); +} + +#define get_desc_from_elsp_dwords(ed, i) \ + ((struct execlist_ctx_descriptor_format *)&((ed)->data[i * 2])) + + +#define BATCH_BUFFER_ADDR_MASK ((1UL << 32) - (1U << 2)) +#define BATCH_BUFFER_ADDR_HIGH_MASK ((1UL << 16) - (1U)) +static int set_gma_to_bb_cmd(struct intel_shadow_bb_entry *entry_obj, + unsigned long add, int gmadr_bytes) +{ + if (WARN_ON(gmadr_bytes != 4 && gmadr_bytes != 8)) + return -1; + + *((u32 *)(entry_obj->bb_start_cmd_va + (1 << 2))) = add & + BATCH_BUFFER_ADDR_MASK; + if (gmadr_bytes == 8) { + *((u32 *)(entry_obj->bb_start_cmd_va + (2 << 2))) = + add & BATCH_BUFFER_ADDR_HIGH_MASK; + } + + return 0; +} + +static void prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload) +{ + int gmadr_bytes = workload->vgpu->gvt->device_info.gmadr_bytes_in_cmd; + + /* pin the gem object to ggtt */ + if (!list_empty(&workload->shadow_bb)) { + struct intel_shadow_bb_entry *entry_obj = + list_first_entry(&workload->shadow_bb, + struct intel_shadow_bb_entry, + list); + struct intel_shadow_bb_entry *temp; + + list_for_each_entry_safe(entry_obj, temp, &workload->shadow_bb, + list) { + struct i915_vma *vma; + + vma = i915_gem_object_ggtt_pin(entry_obj->obj, NULL, 0, + 4, 0); + if (IS_ERR(vma)) { + gvt_err("Cannot pin\n"); + return; + } + + /* FIXME: we are not tracking our pinned VMA leaving it + * up to the core to fix up the stray pin_count upon + * free. + */ + + /* update the relocate gma with shadow batch buffer*/ + set_gma_to_bb_cmd(entry_obj, + i915_ggtt_offset(vma), + gmadr_bytes); + } + } +} + +static int update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + int ring_id = wa_ctx->workload->ring_id; + struct i915_gem_context *shadow_ctx = + wa_ctx->workload->vgpu->shadow_ctx; + struct drm_i915_gem_object *ctx_obj = + shadow_ctx->engine[ring_id].state->obj; + struct execlist_ring_context *shadow_ring_context; + struct page *page; + + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); + shadow_ring_context = kmap_atomic(page); + + shadow_ring_context->bb_per_ctx_ptr.val = + (shadow_ring_context->bb_per_ctx_ptr.val & + (~PER_CTX_ADDR_MASK)) | wa_ctx->per_ctx.shadow_gma; + shadow_ring_context->rcs_indirect_ctx.val = + (shadow_ring_context->rcs_indirect_ctx.val & + (~INDIRECT_CTX_ADDR_MASK)) | wa_ctx->indirect_ctx.shadow_gma; + + kunmap_atomic(shadow_ring_context); + return 0; +} + +static void prepare_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + struct i915_vma *vma; + unsigned char *per_ctx_va = + (unsigned char *)wa_ctx->indirect_ctx.shadow_va + + wa_ctx->indirect_ctx.size; + + if (wa_ctx->indirect_ctx.size == 0) + return; + + vma = i915_gem_object_ggtt_pin(wa_ctx->indirect_ctx.obj, NULL, + 0, CACHELINE_BYTES, 0); + if (IS_ERR(vma)) { + gvt_err("Cannot pin indirect ctx obj\n"); + return; + } + + /* FIXME: we are not tracking our pinned VMA leaving it + * up to the core to fix up the stray pin_count upon + * free. + */ + + wa_ctx->indirect_ctx.shadow_gma = i915_ggtt_offset(vma); + + wa_ctx->per_ctx.shadow_gma = *((unsigned int *)per_ctx_va + 1); + memset(per_ctx_va, 0, CACHELINE_BYTES); + + update_wa_ctx_2_shadow_ctx(wa_ctx); +} + +static int prepare_execlist_workload(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct execlist_ctx_descriptor_format ctx[2]; + int ring_id = workload->ring_id; + + intel_vgpu_pin_mm(workload->shadow_mm); + intel_vgpu_sync_oos_pages(workload->vgpu); + intel_vgpu_flush_post_shadow(workload->vgpu); + prepare_shadow_batch_buffer(workload); + prepare_shadow_wa_ctx(&workload->wa_ctx); + if (!workload->emulate_schedule_in) + return 0; + + ctx[0] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 1); + ctx[1] = *get_desc_from_elsp_dwords(&workload->elsp_dwords, 0); + + return emulate_execlist_schedule_in(&vgpu->execlist[ring_id], ctx); +} + +static void release_shadow_batch_buffer(struct intel_vgpu_workload *workload) +{ + /* release all the shadow batch buffer */ + if (!list_empty(&workload->shadow_bb)) { + struct intel_shadow_bb_entry *entry_obj = + list_first_entry(&workload->shadow_bb, + struct intel_shadow_bb_entry, + list); + struct intel_shadow_bb_entry *temp; + + list_for_each_entry_safe(entry_obj, temp, &workload->shadow_bb, + list) { + i915_gem_object_unpin_map(entry_obj->obj); + i915_gem_object_put(entry_obj->obj); + list_del(&entry_obj->list); + kfree(entry_obj); + } + } +} + +static void release_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx) +{ + if (wa_ctx->indirect_ctx.size == 0) + return; + + i915_gem_object_unpin_map(wa_ctx->indirect_ctx.obj); + i915_gem_object_put(wa_ctx->indirect_ctx.obj); +} + +static int complete_execlist_workload(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct intel_vgpu_execlist *execlist = + &vgpu->execlist[workload->ring_id]; + struct intel_vgpu_workload *next_workload; + struct list_head *next = workload_q_head(vgpu, workload->ring_id)->next; + bool lite_restore = false; + int ret; + + gvt_dbg_el("complete workload %p status %d\n", workload, + workload->status); + + release_shadow_batch_buffer(workload); + release_shadow_wa_ctx(&workload->wa_ctx); + + if (workload->status || vgpu->resetting) + goto out; + + if (!list_empty(workload_q_head(vgpu, workload->ring_id))) { + struct execlist_ctx_descriptor_format *this_desc, *next_desc; + + next_workload = container_of(next, + struct intel_vgpu_workload, list); + this_desc = &workload->ctx_desc; + next_desc = &next_workload->ctx_desc; + + lite_restore = same_context(this_desc, next_desc); + } + + if (lite_restore) { + gvt_dbg_el("next context == current - no schedule-out\n"); + free_workload(workload); + return 0; + } + + ret = emulate_execlist_ctx_schedule_out(execlist, &workload->ctx_desc); + if (ret) + goto err; +out: + free_workload(workload); + return 0; +err: + free_workload(workload); + return ret; +} + +#define RING_CTX_OFF(x) \ + offsetof(struct execlist_ring_context, x) + +static void read_guest_pdps(struct intel_vgpu *vgpu, + u64 ring_context_gpa, u32 pdp[8]) +{ + u64 gpa; + int i; + + gpa = ring_context_gpa + RING_CTX_OFF(pdp3_UDW.val); + + for (i = 0; i < 8; i++) + intel_gvt_hypervisor_read_gpa(vgpu, + gpa + i * 8, &pdp[7 - i], 4); +} + +static int prepare_mm(struct intel_vgpu_workload *workload) +{ + struct execlist_ctx_descriptor_format *desc = &workload->ctx_desc; + struct intel_vgpu_mm *mm; + int page_table_level; + u32 pdp[8]; + + if (desc->addressing_mode == 1) { /* legacy 32-bit */ + page_table_level = 3; + } else if (desc->addressing_mode == 3) { /* legacy 64 bit */ + page_table_level = 4; + } else { + gvt_err("Advanced Context mode(SVM) is not supported!\n"); + return -EINVAL; + } + + read_guest_pdps(workload->vgpu, workload->ring_context_gpa, pdp); + + mm = intel_vgpu_find_ppgtt_mm(workload->vgpu, page_table_level, pdp); + if (mm) { + intel_gvt_mm_reference(mm); + } else { + + mm = intel_vgpu_create_mm(workload->vgpu, INTEL_GVT_MM_PPGTT, + pdp, page_table_level, 0); + if (IS_ERR(mm)) { + gvt_err("fail to create mm object.\n"); + return PTR_ERR(mm); + } + } + workload->shadow_mm = mm; + return 0; +} + +#define get_last_workload(q) \ + (list_empty(q) ? NULL : container_of(q->prev, \ + struct intel_vgpu_workload, list)) + +static int submit_context(struct intel_vgpu *vgpu, int ring_id, + struct execlist_ctx_descriptor_format *desc, + bool emulate_schedule_in) +{ + struct list_head *q = workload_q_head(vgpu, ring_id); + struct intel_vgpu_workload *last_workload = get_last_workload(q); + struct intel_vgpu_workload *workload = NULL; + u64 ring_context_gpa; + u32 head, tail, start, ctl, ctx_ctl, per_ctx, indirect_ctx; + int ret; + + ring_context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm, + (u32)((desc->lrca + 1) << GTT_PAGE_SHIFT)); + if (ring_context_gpa == INTEL_GVT_INVALID_ADDR) { + gvt_err("invalid guest context LRCA: %x\n", desc->lrca); + return -EINVAL; + } + + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ring_header.val), &head, 4); + + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ring_tail.val), &tail, 4); + + head &= RB_HEAD_OFF_MASK; + tail &= RB_TAIL_OFF_MASK; + + if (last_workload && same_context(&last_workload->ctx_desc, desc)) { + gvt_dbg_el("ring id %d cur workload == last\n", ring_id); + gvt_dbg_el("ctx head %x real head %lx\n", head, + last_workload->rb_tail); + /* + * cannot use guest context head pointer here, + * as it might not be updated at this time + */ + head = last_workload->rb_tail; + } + + gvt_dbg_el("ring id %d begin a new workload\n", ring_id); + + workload = kmem_cache_zalloc(vgpu->workloads, GFP_KERNEL); + if (!workload) + return -ENOMEM; + + /* record some ring buffer register values for scan and shadow */ + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(rb_start.val), &start, 4); + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(rb_ctrl.val), &ctl, 4); + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(ctx_ctrl.val), &ctx_ctl, 4); + + INIT_LIST_HEAD(&workload->list); + INIT_LIST_HEAD(&workload->shadow_bb); + + init_waitqueue_head(&workload->shadow_ctx_status_wq); + atomic_set(&workload->shadow_ctx_active, 0); + + workload->vgpu = vgpu; + workload->ring_id = ring_id; + workload->ctx_desc = *desc; + workload->ring_context_gpa = ring_context_gpa; + workload->rb_head = head; + workload->rb_tail = tail; + workload->rb_start = start; + workload->rb_ctl = ctl; + workload->prepare = prepare_execlist_workload; + workload->complete = complete_execlist_workload; + workload->status = -EINPROGRESS; + workload->emulate_schedule_in = emulate_schedule_in; + + if (ring_id == RCS) { + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(bb_per_ctx_ptr.val), &per_ctx, 4); + intel_gvt_hypervisor_read_gpa(vgpu, ring_context_gpa + + RING_CTX_OFF(rcs_indirect_ctx.val), &indirect_ctx, 4); + + workload->wa_ctx.indirect_ctx.guest_gma = + indirect_ctx & INDIRECT_CTX_ADDR_MASK; + workload->wa_ctx.indirect_ctx.size = + (indirect_ctx & INDIRECT_CTX_SIZE_MASK) * + CACHELINE_BYTES; + workload->wa_ctx.per_ctx.guest_gma = + per_ctx & PER_CTX_ADDR_MASK; + workload->wa_ctx.workload = workload; + + WARN_ON(workload->wa_ctx.indirect_ctx.size && !(per_ctx & 0x1)); + } + + if (emulate_schedule_in) + memcpy(&workload->elsp_dwords, + &vgpu->execlist[ring_id].elsp_dwords, + sizeof(workload->elsp_dwords)); + + gvt_dbg_el("workload %p ring id %d head %x tail %x start %x ctl %x\n", + workload, ring_id, head, tail, start, ctl); + + gvt_dbg_el("workload %p emulate schedule_in %d\n", workload, + emulate_schedule_in); + + ret = prepare_mm(workload); + if (ret) { + kmem_cache_free(vgpu->workloads, workload); + return ret; + } + + queue_workload(workload); + return 0; +} + +int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id) +{ + struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id]; + struct execlist_ctx_descriptor_format *desc[2], valid_desc[2]; + unsigned long valid_desc_bitmap = 0; + bool emulate_schedule_in = true; + int ret; + int i; + + memset(valid_desc, 0, sizeof(valid_desc)); + + desc[0] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 1); + desc[1] = get_desc_from_elsp_dwords(&execlist->elsp_dwords, 0); + + for (i = 0; i < 2; i++) { + if (!desc[i]->valid) + continue; + + if (!desc[i]->privilege_access) { + gvt_err("vgpu%d: unexpected GGTT elsp submission\n", + vgpu->id); + return -EINVAL; + } + + /* TODO: add another guest context checks here. */ + set_bit(i, &valid_desc_bitmap); + valid_desc[i] = *desc[i]; + } + + if (!valid_desc_bitmap) { + gvt_err("vgpu%d: no valid desc in a elsp submission\n", + vgpu->id); + return -EINVAL; + } + + if (!test_bit(0, (void *)&valid_desc_bitmap) && + test_bit(1, (void *)&valid_desc_bitmap)) { + gvt_err("vgpu%d: weird elsp submission, desc 0 is not valid\n", + vgpu->id); + return -EINVAL; + } + + /* submit workload */ + for_each_set_bit(i, (void *)&valid_desc_bitmap, 2) { + ret = submit_context(vgpu, ring_id, &valid_desc[i], + emulate_schedule_in); + if (ret) { + gvt_err("vgpu%d: fail to schedule workload\n", + vgpu->id); + return ret; + } + emulate_schedule_in = false; + } + return 0; +} + +static void init_vgpu_execlist(struct intel_vgpu *vgpu, int ring_id) +{ + struct intel_vgpu_execlist *execlist = &vgpu->execlist[ring_id]; + struct execlist_context_status_pointer_format ctx_status_ptr; + u32 ctx_status_ptr_reg; + + memset(execlist, 0, sizeof(*execlist)); + + execlist->vgpu = vgpu; + execlist->ring_id = ring_id; + execlist->slot[0].index = 0; + execlist->slot[1].index = 1; + + ctx_status_ptr_reg = execlist_ring_mmio(vgpu->gvt, ring_id, + _EL_OFFSET_STATUS_PTR); + + ctx_status_ptr.dw = vgpu_vreg(vgpu, ctx_status_ptr_reg); + ctx_status_ptr.read_ptr = ctx_status_ptr.write_ptr = 0x7; + vgpu_vreg(vgpu, ctx_status_ptr_reg) = ctx_status_ptr.dw; +} + +void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu) +{ + kmem_cache_destroy(vgpu->workloads); +} + +int intel_vgpu_init_execlist(struct intel_vgpu *vgpu) +{ + enum intel_engine_id i; + struct intel_engine_cs *engine; + + /* each ring has a virtual execlist engine */ + for_each_engine(engine, vgpu->gvt->dev_priv, i) { + init_vgpu_execlist(vgpu, i); + INIT_LIST_HEAD(&vgpu->workload_q_head[i]); + } + + vgpu->workloads = kmem_cache_create("gvt-g vgpu workload", + sizeof(struct intel_vgpu_workload), 0, + SLAB_HWCACHE_ALIGN, + NULL); + + if (!vgpu->workloads) + return -ENOMEM; + + return 0; +} + +void intel_vgpu_reset_execlist(struct intel_vgpu *vgpu, + unsigned long engine_mask) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct intel_engine_cs *engine; + struct intel_vgpu_workload *pos, *n; + unsigned int tmp; + + for_each_engine_masked(engine, dev_priv, engine_mask, tmp) { + /* free the unsubmited workload in the queue */ + list_for_each_entry_safe(pos, n, + &vgpu->workload_q_head[engine->id], list) { + list_del_init(&pos->list); + free_workload(pos); + } + + init_vgpu_execlist(vgpu, engine->id); + } +} diff --git a/drivers/gpu/drm/i915/gvt/execlist.h b/drivers/gpu/drm/i915/gvt/execlist.h new file mode 100644 index 000000000000..7eced40a1e30 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/execlist.h @@ -0,0 +1,188 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * + */ + +#ifndef _GVT_EXECLIST_H_ +#define _GVT_EXECLIST_H_ + +struct execlist_ctx_descriptor_format { + union { + u32 udw; + u32 context_id; + }; + union { + u32 ldw; + struct { + u32 valid : 1; + u32 force_pd_restore : 1; + u32 force_restore : 1; + u32 addressing_mode : 2; + u32 llc_coherency : 1; + u32 fault_handling : 2; + u32 privilege_access : 1; + u32 reserved : 3; + u32 lrca : 20; + }; + }; +}; + +struct execlist_status_format { + union { + u32 ldw; + struct { + u32 current_execlist_pointer :1; + u32 execlist_write_pointer :1; + u32 execlist_queue_full :1; + u32 execlist_1_valid :1; + u32 execlist_0_valid :1; + u32 last_ctx_switch_reason :9; + u32 current_active_elm_status :2; + u32 arbitration_enable :1; + u32 execlist_1_active :1; + u32 execlist_0_active :1; + u32 reserved :13; + }; + }; + union { + u32 udw; + u32 context_id; + }; +}; + +struct execlist_context_status_pointer_format { + union { + u32 dw; + struct { + u32 write_ptr :3; + u32 reserved :5; + u32 read_ptr :3; + u32 reserved2 :5; + u32 mask :16; + }; + }; +}; + +struct execlist_context_status_format { + union { + u32 ldw; + struct { + u32 idle_to_active :1; + u32 preempted :1; + u32 element_switch :1; + u32 active_to_idle :1; + u32 context_complete :1; + u32 wait_on_sync_flip :1; + u32 wait_on_vblank :1; + u32 wait_on_semaphore :1; + u32 wait_on_scanline :1; + u32 reserved :2; + u32 semaphore_wait_mode :1; + u32 display_plane :3; + u32 lite_restore :1; + u32 reserved_2 :16; + }; + }; + union { + u32 udw; + u32 context_id; + }; +}; + +struct execlist_mmio_pair { + u32 addr; + u32 val; +}; + +/* The first 52 dwords in register state context */ +struct execlist_ring_context { + u32 nop1; + u32 lri_cmd_1; + struct execlist_mmio_pair ctx_ctrl; + struct execlist_mmio_pair ring_header; + struct execlist_mmio_pair ring_tail; + struct execlist_mmio_pair rb_start; + struct execlist_mmio_pair rb_ctrl; + struct execlist_mmio_pair bb_cur_head_UDW; + struct execlist_mmio_pair bb_cur_head_LDW; + struct execlist_mmio_pair bb_state; + struct execlist_mmio_pair second_bb_addr_UDW; + struct execlist_mmio_pair second_bb_addr_LDW; + struct execlist_mmio_pair second_bb_state; + struct execlist_mmio_pair bb_per_ctx_ptr; + struct execlist_mmio_pair rcs_indirect_ctx; + struct execlist_mmio_pair rcs_indirect_ctx_offset; + u32 nop2; + u32 nop3; + u32 nop4; + u32 lri_cmd_2; + struct execlist_mmio_pair ctx_timestamp; + struct execlist_mmio_pair pdp3_UDW; + struct execlist_mmio_pair pdp3_LDW; + struct execlist_mmio_pair pdp2_UDW; + struct execlist_mmio_pair pdp2_LDW; + struct execlist_mmio_pair pdp1_UDW; + struct execlist_mmio_pair pdp1_LDW; + struct execlist_mmio_pair pdp0_UDW; + struct execlist_mmio_pair pdp0_LDW; +}; + +struct intel_vgpu_elsp_dwords { + u32 data[4]; + u32 index; +}; + +struct intel_vgpu_execlist_slot { + struct execlist_ctx_descriptor_format ctx[2]; + u32 index; +}; + +struct intel_vgpu_execlist { + struct intel_vgpu_execlist_slot slot[2]; + struct intel_vgpu_execlist_slot *running_slot; + struct intel_vgpu_execlist_slot *pending_slot; + struct execlist_ctx_descriptor_format *running_context; + int ring_id; + struct intel_vgpu *vgpu; + struct intel_vgpu_elsp_dwords elsp_dwords; +}; + +void intel_vgpu_clean_execlist(struct intel_vgpu *vgpu); + +int intel_vgpu_init_execlist(struct intel_vgpu *vgpu); + +int intel_vgpu_submit_execlist(struct intel_vgpu *vgpu, int ring_id); + +void intel_vgpu_reset_execlist(struct intel_vgpu *vgpu, + unsigned long engine_mask); + +#endif /*_GVT_EXECLIST_H_*/ diff --git a/drivers/gpu/drm/i915/gvt/firmware.c b/drivers/gpu/drm/i915/gvt/firmware.c new file mode 100644 index 000000000000..2fae2a2ca96f --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/firmware.c @@ -0,0 +1,312 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Changbin Du <changbin.du@intel.com> + * + */ + +#include <linux/firmware.h> +#include <linux/crc32.h> + +#include "i915_drv.h" +#include "gvt.h" +#include "i915_pvinfo.h" + +#define FIRMWARE_VERSION (0x0) + +struct gvt_firmware_header { + u64 magic; + u32 crc32; /* protect the data after this field */ + u32 version; + u64 cfg_space_size; + u64 cfg_space_offset; /* offset in the file */ + u64 mmio_size; + u64 mmio_offset; /* offset in the file */ + unsigned char data[1]; +}; + +#define RD(offset) (readl(mmio + offset.reg)) +#define WR(v, offset) (writel(v, mmio + offset.reg)) + +static void bdw_forcewake_get(void __iomem *mmio) +{ + WR(_MASKED_BIT_DISABLE(0xffff), FORCEWAKE_MT); + + RD(ECOBUS); + + if (wait_for((RD(FORCEWAKE_ACK_HSW) & FORCEWAKE_KERNEL) == 0, 50)) + gvt_err("fail to wait forcewake idle\n"); + + WR(_MASKED_BIT_ENABLE(FORCEWAKE_KERNEL), FORCEWAKE_MT); + + if (wait_for((RD(FORCEWAKE_ACK_HSW) & FORCEWAKE_KERNEL), 50)) + gvt_err("fail to wait forcewake ack\n"); + + if (wait_for((RD(GEN6_GT_THREAD_STATUS_REG) & + GEN6_GT_THREAD_STATUS_CORE_MASK) == 0, 50)) + gvt_err("fail to wait c0 wake up\n"); +} + +#undef RD +#undef WR + +#define dev_to_drm_minor(d) dev_get_drvdata((d)) + +static ssize_t +gvt_firmware_read(struct file *filp, struct kobject *kobj, + struct bin_attribute *attr, char *buf, + loff_t offset, size_t count) +{ + memcpy(buf, attr->private + offset, count); + return count; +} + +static struct bin_attribute firmware_attr = { + .attr = {.name = "gvt_firmware", .mode = (S_IRUSR)}, + .read = gvt_firmware_read, + .write = NULL, + .mmap = NULL, +}; + +static int expose_firmware_sysfs(struct intel_gvt *gvt, + void __iomem *mmio) +{ + struct intel_gvt_device_info *info = &gvt->device_info; + struct pci_dev *pdev = gvt->dev_priv->drm.pdev; + struct intel_gvt_mmio_info *e; + struct gvt_firmware_header *h; + void *firmware; + void *p; + unsigned long size; + int i; + int ret; + + size = sizeof(*h) + info->mmio_size + info->cfg_space_size - 1; + firmware = vmalloc(size); + if (!firmware) + return -ENOMEM; + + h = firmware; + + h->magic = VGT_MAGIC; + h->version = FIRMWARE_VERSION; + h->cfg_space_size = info->cfg_space_size; + h->cfg_space_offset = offsetof(struct gvt_firmware_header, data); + h->mmio_size = info->mmio_size; + h->mmio_offset = h->cfg_space_offset + h->cfg_space_size; + + p = firmware + h->cfg_space_offset; + + for (i = 0; i < h->cfg_space_size; i += 4) + pci_read_config_dword(pdev, i, p + i); + + memcpy(gvt->firmware.cfg_space, p, info->cfg_space_size); + + p = firmware + h->mmio_offset; + + hash_for_each(gvt->mmio.mmio_info_table, i, e, node) { + int j; + + for (j = 0; j < e->length; j += 4) + *(u32 *)(p + e->offset + j) = + readl(mmio + e->offset + j); + } + + memcpy(gvt->firmware.mmio, p, info->mmio_size); + + firmware_attr.size = size; + firmware_attr.private = firmware; + + ret = device_create_bin_file(&pdev->dev, &firmware_attr); + if (ret) { + vfree(firmware); + return ret; + } + return 0; +} + +static void clean_firmware_sysfs(struct intel_gvt *gvt) +{ + struct pci_dev *pdev = gvt->dev_priv->drm.pdev; + + device_remove_bin_file(&pdev->dev, &firmware_attr); + vfree(firmware_attr.private); +} + +/** + * intel_gvt_free_firmware - free GVT firmware + * @gvt: intel gvt device + * + */ +void intel_gvt_free_firmware(struct intel_gvt *gvt) +{ + if (!gvt->firmware.firmware_loaded) + clean_firmware_sysfs(gvt); + + kfree(gvt->firmware.cfg_space); + kfree(gvt->firmware.mmio); +} + +static int verify_firmware(struct intel_gvt *gvt, + const struct firmware *fw) +{ + struct intel_gvt_device_info *info = &gvt->device_info; + struct drm_i915_private *dev_priv = gvt->dev_priv; + struct pci_dev *pdev = dev_priv->drm.pdev; + struct gvt_firmware_header *h; + unsigned long id, crc32_start; + const void *mem; + const char *item; + u64 file, request; + + h = (struct gvt_firmware_header *)fw->data; + + crc32_start = offsetof(struct gvt_firmware_header, crc32) + 4; + mem = fw->data + crc32_start; + +#define VERIFY(s, a, b) do { \ + item = (s); file = (u64)(a); request = (u64)(b); \ + if ((a) != (b)) \ + goto invalid_firmware; \ +} while (0) + + VERIFY("magic number", h->magic, VGT_MAGIC); + VERIFY("version", h->version, FIRMWARE_VERSION); + VERIFY("crc32", h->crc32, crc32_le(0, mem, fw->size - crc32_start)); + VERIFY("cfg space size", h->cfg_space_size, info->cfg_space_size); + VERIFY("mmio size", h->mmio_size, info->mmio_size); + + mem = (fw->data + h->cfg_space_offset); + + id = *(u16 *)(mem + PCI_VENDOR_ID); + VERIFY("vender id", id, pdev->vendor); + + id = *(u16 *)(mem + PCI_DEVICE_ID); + VERIFY("device id", id, pdev->device); + + id = *(u8 *)(mem + PCI_REVISION_ID); + VERIFY("revision id", id, pdev->revision); + +#undef VERIFY + return 0; + +invalid_firmware: + gvt_dbg_core("Invalid firmware: %s [file] 0x%llx [request] 0x%llx\n", + item, file, request); + return -EINVAL; +} + +#define GVT_FIRMWARE_PATH "i915/gvt" + +/** + * intel_gvt_load_firmware - load GVT firmware + * @gvt: intel gvt device + * + */ +int intel_gvt_load_firmware(struct intel_gvt *gvt) +{ + struct intel_gvt_device_info *info = &gvt->device_info; + struct drm_i915_private *dev_priv = gvt->dev_priv; + struct pci_dev *pdev = dev_priv->drm.pdev; + struct intel_gvt_firmware *firmware = &gvt->firmware; + struct gvt_firmware_header *h; + const struct firmware *fw; + char *path; + void __iomem *mmio; + void *mem; + int ret; + + path = kmalloc(PATH_MAX, GFP_KERNEL); + if (!path) + return -ENOMEM; + + mem = kmalloc(info->cfg_space_size, GFP_KERNEL); + if (!mem) { + kfree(path); + return -ENOMEM; + } + + firmware->cfg_space = mem; + + mem = kmalloc(info->mmio_size, GFP_KERNEL); + if (!mem) { + kfree(path); + kfree(firmware->cfg_space); + return -ENOMEM; + } + + firmware->mmio = mem; + + mmio = pci_iomap(pdev, info->mmio_bar, info->mmio_size); + if (!mmio) { + kfree(path); + kfree(firmware->cfg_space); + kfree(firmware->mmio); + return -EINVAL; + } + + if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) + bdw_forcewake_get(mmio); + + sprintf(path, "%s/vid_0x%04x_did_0x%04x_rid_0x%04x.golden_hw_state", + GVT_FIRMWARE_PATH, pdev->vendor, pdev->device, + pdev->revision); + + gvt_dbg_core("request hw state firmware %s...\n", path); + + ret = request_firmware(&fw, path, &dev_priv->drm.pdev->dev); + kfree(path); + + if (ret) + goto expose_firmware; + + gvt_dbg_core("success.\n"); + + ret = verify_firmware(gvt, fw); + if (ret) + goto out_free_fw; + + gvt_dbg_core("verified.\n"); + + h = (struct gvt_firmware_header *)fw->data; + + memcpy(firmware->cfg_space, fw->data + h->cfg_space_offset, + h->cfg_space_size); + memcpy(firmware->mmio, fw->data + h->mmio_offset, + h->mmio_size); + + release_firmware(fw); + firmware->firmware_loaded = true; + pci_iounmap(pdev, mmio); + return 0; + +out_free_fw: + release_firmware(fw); +expose_firmware: + expose_firmware_sysfs(gvt, mmio); + pci_iounmap(pdev, mmio); + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c new file mode 100644 index 000000000000..7eaaf1c9ed2b --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/gtt.c @@ -0,0 +1,2244 @@ +/* + * GTT virtualization + * + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhi Wang <zhi.a.wang@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * Xiao Zheng <xiao.zheng@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" +#include "i915_pvinfo.h" +#include "trace.h" + +static bool enable_out_of_sync = false; +static int preallocated_oos_pages = 8192; + +/* + * validate a gm address and related range size, + * translate it to host gm address + */ +bool intel_gvt_ggtt_validate_range(struct intel_vgpu *vgpu, u64 addr, u32 size) +{ + if ((!vgpu_gmadr_is_valid(vgpu, addr)) || (size + && !vgpu_gmadr_is_valid(vgpu, addr + size - 1))) { + gvt_err("vgpu%d: invalid range gmadr 0x%llx size 0x%x\n", + vgpu->id, addr, size); + return false; + } + return true; +} + +/* translate a guest gmadr to host gmadr */ +int intel_gvt_ggtt_gmadr_g2h(struct intel_vgpu *vgpu, u64 g_addr, u64 *h_addr) +{ + if (WARN(!vgpu_gmadr_is_valid(vgpu, g_addr), + "invalid guest gmadr %llx\n", g_addr)) + return -EACCES; + + if (vgpu_gmadr_is_aperture(vgpu, g_addr)) + *h_addr = vgpu_aperture_gmadr_base(vgpu) + + (g_addr - vgpu_aperture_offset(vgpu)); + else + *h_addr = vgpu_hidden_gmadr_base(vgpu) + + (g_addr - vgpu_hidden_offset(vgpu)); + return 0; +} + +/* translate a host gmadr to guest gmadr */ +int intel_gvt_ggtt_gmadr_h2g(struct intel_vgpu *vgpu, u64 h_addr, u64 *g_addr) +{ + if (WARN(!gvt_gmadr_is_valid(vgpu->gvt, h_addr), + "invalid host gmadr %llx\n", h_addr)) + return -EACCES; + + if (gvt_gmadr_is_aperture(vgpu->gvt, h_addr)) + *g_addr = vgpu_aperture_gmadr_base(vgpu) + + (h_addr - gvt_aperture_gmadr_base(vgpu->gvt)); + else + *g_addr = vgpu_hidden_gmadr_base(vgpu) + + (h_addr - gvt_hidden_gmadr_base(vgpu->gvt)); + return 0; +} + +int intel_gvt_ggtt_index_g2h(struct intel_vgpu *vgpu, unsigned long g_index, + unsigned long *h_index) +{ + u64 h_addr; + int ret; + + ret = intel_gvt_ggtt_gmadr_g2h(vgpu, g_index << GTT_PAGE_SHIFT, + &h_addr); + if (ret) + return ret; + + *h_index = h_addr >> GTT_PAGE_SHIFT; + return 0; +} + +int intel_gvt_ggtt_h2g_index(struct intel_vgpu *vgpu, unsigned long h_index, + unsigned long *g_index) +{ + u64 g_addr; + int ret; + + ret = intel_gvt_ggtt_gmadr_h2g(vgpu, h_index << GTT_PAGE_SHIFT, + &g_addr); + if (ret) + return ret; + + *g_index = g_addr >> GTT_PAGE_SHIFT; + return 0; +} + +#define gtt_type_is_entry(type) \ + (type > GTT_TYPE_INVALID && type < GTT_TYPE_PPGTT_ENTRY \ + && type != GTT_TYPE_PPGTT_PTE_ENTRY \ + && type != GTT_TYPE_PPGTT_ROOT_ENTRY) + +#define gtt_type_is_pt(type) \ + (type >= GTT_TYPE_PPGTT_PTE_PT && type < GTT_TYPE_MAX) + +#define gtt_type_is_pte_pt(type) \ + (type == GTT_TYPE_PPGTT_PTE_PT) + +#define gtt_type_is_root_pointer(type) \ + (gtt_type_is_entry(type) && type > GTT_TYPE_PPGTT_ROOT_ENTRY) + +#define gtt_init_entry(e, t, p, v) do { \ + (e)->type = t; \ + (e)->pdev = p; \ + memcpy(&(e)->val64, &v, sizeof(v)); \ +} while (0) + +/* + * Mappings between GTT_TYPE* enumerations. + * Following information can be found according to the given type: + * - type of next level page table + * - type of entry inside this level page table + * - type of entry with PSE set + * + * If the given type doesn't have such a kind of information, + * e.g. give a l4 root entry type, then request to get its PSE type, + * give a PTE page table type, then request to get its next level page + * table type, as we know l4 root entry doesn't have a PSE bit, + * and a PTE page table doesn't have a next level page table type, + * GTT_TYPE_INVALID will be returned. This is useful when traversing a + * page table. + */ + +struct gtt_type_table_entry { + int entry_type; + int next_pt_type; + int pse_entry_type; +}; + +#define GTT_TYPE_TABLE_ENTRY(type, e_type, npt_type, pse_type) \ + [type] = { \ + .entry_type = e_type, \ + .next_pt_type = npt_type, \ + .pse_entry_type = pse_type, \ + } + +static struct gtt_type_table_entry gtt_type_table[] = { + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_ROOT_L4_ENTRY, + GTT_TYPE_PPGTT_ROOT_L4_ENTRY, + GTT_TYPE_PPGTT_PML4_PT, + GTT_TYPE_INVALID), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PML4_PT, + GTT_TYPE_PPGTT_PML4_ENTRY, + GTT_TYPE_PPGTT_PDP_PT, + GTT_TYPE_INVALID), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PML4_ENTRY, + GTT_TYPE_PPGTT_PML4_ENTRY, + GTT_TYPE_PPGTT_PDP_PT, + GTT_TYPE_INVALID), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PDP_PT, + GTT_TYPE_PPGTT_PDP_ENTRY, + GTT_TYPE_PPGTT_PDE_PT, + GTT_TYPE_PPGTT_PTE_1G_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_ROOT_L3_ENTRY, + GTT_TYPE_PPGTT_ROOT_L3_ENTRY, + GTT_TYPE_PPGTT_PDE_PT, + GTT_TYPE_PPGTT_PTE_1G_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PDP_ENTRY, + GTT_TYPE_PPGTT_PDP_ENTRY, + GTT_TYPE_PPGTT_PDE_PT, + GTT_TYPE_PPGTT_PTE_1G_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PDE_PT, + GTT_TYPE_PPGTT_PDE_ENTRY, + GTT_TYPE_PPGTT_PTE_PT, + GTT_TYPE_PPGTT_PTE_2M_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PDE_ENTRY, + GTT_TYPE_PPGTT_PDE_ENTRY, + GTT_TYPE_PPGTT_PTE_PT, + GTT_TYPE_PPGTT_PTE_2M_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PTE_PT, + GTT_TYPE_PPGTT_PTE_4K_ENTRY, + GTT_TYPE_INVALID, + GTT_TYPE_INVALID), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PTE_4K_ENTRY, + GTT_TYPE_PPGTT_PTE_4K_ENTRY, + GTT_TYPE_INVALID, + GTT_TYPE_INVALID), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PTE_2M_ENTRY, + GTT_TYPE_PPGTT_PDE_ENTRY, + GTT_TYPE_INVALID, + GTT_TYPE_PPGTT_PTE_2M_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_PPGTT_PTE_1G_ENTRY, + GTT_TYPE_PPGTT_PDP_ENTRY, + GTT_TYPE_INVALID, + GTT_TYPE_PPGTT_PTE_1G_ENTRY), + GTT_TYPE_TABLE_ENTRY(GTT_TYPE_GGTT_PTE, + GTT_TYPE_GGTT_PTE, + GTT_TYPE_INVALID, + GTT_TYPE_INVALID), +}; + +static inline int get_next_pt_type(int type) +{ + return gtt_type_table[type].next_pt_type; +} + +static inline int get_entry_type(int type) +{ + return gtt_type_table[type].entry_type; +} + +static inline int get_pse_type(int type) +{ + return gtt_type_table[type].pse_entry_type; +} + +static u64 read_pte64(struct drm_i915_private *dev_priv, unsigned long index) +{ + void __iomem *addr = (gen8_pte_t __iomem *)dev_priv->ggtt.gsm + index; + u64 pte; + +#ifdef readq + pte = readq(addr); +#else + pte = ioread32(addr); + pte |= (u64)ioread32(addr + 4) << 32; +#endif + return pte; +} + +static void write_pte64(struct drm_i915_private *dev_priv, + unsigned long index, u64 pte) +{ + void __iomem *addr = (gen8_pte_t __iomem *)dev_priv->ggtt.gsm + index; + +#ifdef writeq + writeq(pte, addr); +#else + iowrite32((u32)pte, addr); + iowrite32(pte >> 32, addr + 4); +#endif + I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN); + POSTING_READ(GFX_FLSH_CNTL_GEN6); +} + +static inline struct intel_gvt_gtt_entry *gtt_get_entry64(void *pt, + struct intel_gvt_gtt_entry *e, + unsigned long index, bool hypervisor_access, unsigned long gpa, + struct intel_vgpu *vgpu) +{ + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + int ret; + + if (WARN_ON(info->gtt_entry_size != 8)) + return e; + + if (hypervisor_access) { + ret = intel_gvt_hypervisor_read_gpa(vgpu, gpa + + (index << info->gtt_entry_size_shift), + &e->val64, 8); + WARN_ON(ret); + } else if (!pt) { + e->val64 = read_pte64(vgpu->gvt->dev_priv, index); + } else { + e->val64 = *((u64 *)pt + index); + } + return e; +} + +static inline struct intel_gvt_gtt_entry *gtt_set_entry64(void *pt, + struct intel_gvt_gtt_entry *e, + unsigned long index, bool hypervisor_access, unsigned long gpa, + struct intel_vgpu *vgpu) +{ + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + int ret; + + if (WARN_ON(info->gtt_entry_size != 8)) + return e; + + if (hypervisor_access) { + ret = intel_gvt_hypervisor_write_gpa(vgpu, gpa + + (index << info->gtt_entry_size_shift), + &e->val64, 8); + WARN_ON(ret); + } else if (!pt) { + write_pte64(vgpu->gvt->dev_priv, index, e->val64); + } else { + *((u64 *)pt + index) = e->val64; + } + return e; +} + +#define GTT_HAW 46 + +#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30 + 1)) - 1) << 30) +#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21 + 1)) - 1) << 21) +#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12 + 1)) - 1) << 12) + +static unsigned long gen8_gtt_get_pfn(struct intel_gvt_gtt_entry *e) +{ + unsigned long pfn; + + if (e->type == GTT_TYPE_PPGTT_PTE_1G_ENTRY) + pfn = (e->val64 & ADDR_1G_MASK) >> 12; + else if (e->type == GTT_TYPE_PPGTT_PTE_2M_ENTRY) + pfn = (e->val64 & ADDR_2M_MASK) >> 12; + else + pfn = (e->val64 & ADDR_4K_MASK) >> 12; + return pfn; +} + +static void gen8_gtt_set_pfn(struct intel_gvt_gtt_entry *e, unsigned long pfn) +{ + if (e->type == GTT_TYPE_PPGTT_PTE_1G_ENTRY) { + e->val64 &= ~ADDR_1G_MASK; + pfn &= (ADDR_1G_MASK >> 12); + } else if (e->type == GTT_TYPE_PPGTT_PTE_2M_ENTRY) { + e->val64 &= ~ADDR_2M_MASK; + pfn &= (ADDR_2M_MASK >> 12); + } else { + e->val64 &= ~ADDR_4K_MASK; + pfn &= (ADDR_4K_MASK >> 12); + } + + e->val64 |= (pfn << 12); +} + +static bool gen8_gtt_test_pse(struct intel_gvt_gtt_entry *e) +{ + /* Entry doesn't have PSE bit. */ + if (get_pse_type(e->type) == GTT_TYPE_INVALID) + return false; + + e->type = get_entry_type(e->type); + if (!(e->val64 & (1 << 7))) + return false; + + e->type = get_pse_type(e->type); + return true; +} + +static bool gen8_gtt_test_present(struct intel_gvt_gtt_entry *e) +{ + /* + * i915 writes PDP root pointer registers without present bit, + * it also works, so we need to treat root pointer entry + * specifically. + */ + if (e->type == GTT_TYPE_PPGTT_ROOT_L3_ENTRY + || e->type == GTT_TYPE_PPGTT_ROOT_L4_ENTRY) + return (e->val64 != 0); + else + return (e->val64 & (1 << 0)); +} + +static void gtt_entry_clear_present(struct intel_gvt_gtt_entry *e) +{ + e->val64 &= ~(1 << 0); +} + +/* + * Per-platform GMA routines. + */ +static unsigned long gma_to_ggtt_pte_index(unsigned long gma) +{ + unsigned long x = (gma >> GTT_PAGE_SHIFT); + + trace_gma_index(__func__, gma, x); + return x; +} + +#define DEFINE_PPGTT_GMA_TO_INDEX(prefix, ename, exp) \ +static unsigned long prefix##_gma_to_##ename##_index(unsigned long gma) \ +{ \ + unsigned long x = (exp); \ + trace_gma_index(__func__, gma, x); \ + return x; \ +} + +DEFINE_PPGTT_GMA_TO_INDEX(gen8, pte, (gma >> 12 & 0x1ff)); +DEFINE_PPGTT_GMA_TO_INDEX(gen8, pde, (gma >> 21 & 0x1ff)); +DEFINE_PPGTT_GMA_TO_INDEX(gen8, l3_pdp, (gma >> 30 & 0x3)); +DEFINE_PPGTT_GMA_TO_INDEX(gen8, l4_pdp, (gma >> 30 & 0x1ff)); +DEFINE_PPGTT_GMA_TO_INDEX(gen8, pml4, (gma >> 39 & 0x1ff)); + +static struct intel_gvt_gtt_pte_ops gen8_gtt_pte_ops = { + .get_entry = gtt_get_entry64, + .set_entry = gtt_set_entry64, + .clear_present = gtt_entry_clear_present, + .test_present = gen8_gtt_test_present, + .test_pse = gen8_gtt_test_pse, + .get_pfn = gen8_gtt_get_pfn, + .set_pfn = gen8_gtt_set_pfn, +}; + +static struct intel_gvt_gtt_gma_ops gen8_gtt_gma_ops = { + .gma_to_ggtt_pte_index = gma_to_ggtt_pte_index, + .gma_to_pte_index = gen8_gma_to_pte_index, + .gma_to_pde_index = gen8_gma_to_pde_index, + .gma_to_l3_pdp_index = gen8_gma_to_l3_pdp_index, + .gma_to_l4_pdp_index = gen8_gma_to_l4_pdp_index, + .gma_to_pml4_index = gen8_gma_to_pml4_index, +}; + +static int gtt_entry_p2m(struct intel_vgpu *vgpu, struct intel_gvt_gtt_entry *p, + struct intel_gvt_gtt_entry *m) +{ + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + unsigned long gfn, mfn; + + *m = *p; + + if (!ops->test_present(p)) + return 0; + + gfn = ops->get_pfn(p); + + mfn = intel_gvt_hypervisor_gfn_to_mfn(vgpu, gfn); + if (mfn == INTEL_GVT_INVALID_ADDR) { + gvt_err("fail to translate gfn: 0x%lx\n", gfn); + return -ENXIO; + } + + ops->set_pfn(m, mfn); + return 0; +} + +/* + * MM helpers. + */ +struct intel_gvt_gtt_entry *intel_vgpu_mm_get_entry(struct intel_vgpu_mm *mm, + void *page_table, struct intel_gvt_gtt_entry *e, + unsigned long index) +{ + struct intel_gvt *gvt = mm->vgpu->gvt; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + + e->type = mm->page_table_entry_type; + + ops->get_entry(page_table, e, index, false, 0, mm->vgpu); + ops->test_pse(e); + return e; +} + +struct intel_gvt_gtt_entry *intel_vgpu_mm_set_entry(struct intel_vgpu_mm *mm, + void *page_table, struct intel_gvt_gtt_entry *e, + unsigned long index) +{ + struct intel_gvt *gvt = mm->vgpu->gvt; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + + return ops->set_entry(page_table, e, index, false, 0, mm->vgpu); +} + +/* + * PPGTT shadow page table helpers. + */ +static inline struct intel_gvt_gtt_entry *ppgtt_spt_get_entry( + struct intel_vgpu_ppgtt_spt *spt, + void *page_table, int type, + struct intel_gvt_gtt_entry *e, unsigned long index, + bool guest) +{ + struct intel_gvt *gvt = spt->vgpu->gvt; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + + e->type = get_entry_type(type); + + if (WARN(!gtt_type_is_entry(e->type), "invalid entry type\n")) + return e; + + ops->get_entry(page_table, e, index, guest, + spt->guest_page.gfn << GTT_PAGE_SHIFT, + spt->vgpu); + ops->test_pse(e); + return e; +} + +static inline struct intel_gvt_gtt_entry *ppgtt_spt_set_entry( + struct intel_vgpu_ppgtt_spt *spt, + void *page_table, int type, + struct intel_gvt_gtt_entry *e, unsigned long index, + bool guest) +{ + struct intel_gvt *gvt = spt->vgpu->gvt; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + + if (WARN(!gtt_type_is_entry(e->type), "invalid entry type\n")) + return e; + + return ops->set_entry(page_table, e, index, guest, + spt->guest_page.gfn << GTT_PAGE_SHIFT, + spt->vgpu); +} + +#define ppgtt_get_guest_entry(spt, e, index) \ + ppgtt_spt_get_entry(spt, NULL, \ + spt->guest_page_type, e, index, true) + +#define ppgtt_set_guest_entry(spt, e, index) \ + ppgtt_spt_set_entry(spt, NULL, \ + spt->guest_page_type, e, index, true) + +#define ppgtt_get_shadow_entry(spt, e, index) \ + ppgtt_spt_get_entry(spt, spt->shadow_page.vaddr, \ + spt->shadow_page.type, e, index, false) + +#define ppgtt_set_shadow_entry(spt, e, index) \ + ppgtt_spt_set_entry(spt, spt->shadow_page.vaddr, \ + spt->shadow_page.type, e, index, false) + +/** + * intel_vgpu_init_guest_page - init a guest page data structure + * @vgpu: a vGPU + * @p: a guest page data structure + * @gfn: guest memory page frame number + * @handler: function will be called when target guest memory page has + * been modified. + * + * This function is called when user wants to track a guest memory page. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_init_guest_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *p, + unsigned long gfn, + int (*handler)(void *, u64, void *, int), + void *data) +{ + INIT_HLIST_NODE(&p->node); + + p->writeprotection = false; + p->gfn = gfn; + p->handler = handler; + p->data = data; + p->oos_page = NULL; + p->write_cnt = 0; + + hash_add(vgpu->gtt.guest_page_hash_table, &p->node, p->gfn); + return 0; +} + +static int detach_oos_page(struct intel_vgpu *vgpu, + struct intel_vgpu_oos_page *oos_page); + +/** + * intel_vgpu_clean_guest_page - release the resource owned by guest page data + * structure + * @vgpu: a vGPU + * @p: a tracked guest page + * + * This function is called when user tries to stop tracking a guest memory + * page. + */ +void intel_vgpu_clean_guest_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *p) +{ + if (!hlist_unhashed(&p->node)) + hash_del(&p->node); + + if (p->oos_page) + detach_oos_page(vgpu, p->oos_page); + + if (p->writeprotection) + intel_gvt_hypervisor_unset_wp_page(vgpu, p); +} + +/** + * intel_vgpu_find_guest_page - find a guest page data structure by GFN. + * @vgpu: a vGPU + * @gfn: guest memory page frame number + * + * This function is called when emulation logic wants to know if a trapped GFN + * is a tracked guest page. + * + * Returns: + * Pointer to guest page data structure, NULL if failed. + */ +struct intel_vgpu_guest_page *intel_vgpu_find_guest_page( + struct intel_vgpu *vgpu, unsigned long gfn) +{ + struct intel_vgpu_guest_page *p; + + hash_for_each_possible(vgpu->gtt.guest_page_hash_table, + p, node, gfn) { + if (p->gfn == gfn) + return p; + } + return NULL; +} + +static inline int init_shadow_page(struct intel_vgpu *vgpu, + struct intel_vgpu_shadow_page *p, int type) +{ + p->vaddr = page_address(p->page); + p->type = type; + + INIT_HLIST_NODE(&p->node); + + p->mfn = intel_gvt_hypervisor_virt_to_mfn(p->vaddr); + if (p->mfn == INTEL_GVT_INVALID_ADDR) + return -EFAULT; + + hash_add(vgpu->gtt.shadow_page_hash_table, &p->node, p->mfn); + return 0; +} + +static inline void clean_shadow_page(struct intel_vgpu_shadow_page *p) +{ + if (!hlist_unhashed(&p->node)) + hash_del(&p->node); +} + +static inline struct intel_vgpu_shadow_page *find_shadow_page( + struct intel_vgpu *vgpu, unsigned long mfn) +{ + struct intel_vgpu_shadow_page *p; + + hash_for_each_possible(vgpu->gtt.shadow_page_hash_table, + p, node, mfn) { + if (p->mfn == mfn) + return p; + } + return NULL; +} + +#define guest_page_to_ppgtt_spt(ptr) \ + container_of(ptr, struct intel_vgpu_ppgtt_spt, guest_page) + +#define shadow_page_to_ppgtt_spt(ptr) \ + container_of(ptr, struct intel_vgpu_ppgtt_spt, shadow_page) + +static void *alloc_spt(gfp_t gfp_mask) +{ + struct intel_vgpu_ppgtt_spt *spt; + + spt = kzalloc(sizeof(*spt), gfp_mask); + if (!spt) + return NULL; + + spt->shadow_page.page = alloc_page(gfp_mask); + if (!spt->shadow_page.page) { + kfree(spt); + return NULL; + } + return spt; +} + +static void free_spt(struct intel_vgpu_ppgtt_spt *spt) +{ + __free_page(spt->shadow_page.page); + kfree(spt); +} + +static void ppgtt_free_shadow_page(struct intel_vgpu_ppgtt_spt *spt) +{ + trace_spt_free(spt->vgpu->id, spt, spt->shadow_page.type); + + clean_shadow_page(&spt->shadow_page); + intel_vgpu_clean_guest_page(spt->vgpu, &spt->guest_page); + list_del_init(&spt->post_shadow_list); + + free_spt(spt); +} + +static void ppgtt_free_all_shadow_page(struct intel_vgpu *vgpu) +{ + struct hlist_node *n; + struct intel_vgpu_shadow_page *sp; + int i; + + hash_for_each_safe(vgpu->gtt.shadow_page_hash_table, i, n, sp, node) + ppgtt_free_shadow_page(shadow_page_to_ppgtt_spt(sp)); +} + +static int ppgtt_handle_guest_write_page_table_bytes(void *gp, + u64 pa, void *p_data, int bytes); + +static int ppgtt_write_protection_handler(void *gp, u64 pa, + void *p_data, int bytes) +{ + struct intel_vgpu_guest_page *gpt = (struct intel_vgpu_guest_page *)gp; + int ret; + + if (bytes != 4 && bytes != 8) + return -EINVAL; + + if (!gpt->writeprotection) + return -EINVAL; + + ret = ppgtt_handle_guest_write_page_table_bytes(gp, + pa, p_data, bytes); + if (ret) + return ret; + return ret; +} + +static int reclaim_one_mm(struct intel_gvt *gvt); + +static struct intel_vgpu_ppgtt_spt *ppgtt_alloc_shadow_page( + struct intel_vgpu *vgpu, int type, unsigned long gfn) +{ + struct intel_vgpu_ppgtt_spt *spt = NULL; + int ret; + +retry: + spt = alloc_spt(GFP_KERNEL | __GFP_ZERO); + if (!spt) { + if (reclaim_one_mm(vgpu->gvt)) + goto retry; + + gvt_err("fail to allocate ppgtt shadow page\n"); + return ERR_PTR(-ENOMEM); + } + + spt->vgpu = vgpu; + spt->guest_page_type = type; + atomic_set(&spt->refcount, 1); + INIT_LIST_HEAD(&spt->post_shadow_list); + + /* + * TODO: guest page type may be different with shadow page type, + * when we support PSE page in future. + */ + ret = init_shadow_page(vgpu, &spt->shadow_page, type); + if (ret) { + gvt_err("fail to initialize shadow page for spt\n"); + goto err; + } + + ret = intel_vgpu_init_guest_page(vgpu, &spt->guest_page, + gfn, ppgtt_write_protection_handler, NULL); + if (ret) { + gvt_err("fail to initialize guest page for spt\n"); + goto err; + } + + trace_spt_alloc(vgpu->id, spt, type, spt->shadow_page.mfn, gfn); + return spt; +err: + ppgtt_free_shadow_page(spt); + return ERR_PTR(ret); +} + +static struct intel_vgpu_ppgtt_spt *ppgtt_find_shadow_page( + struct intel_vgpu *vgpu, unsigned long mfn) +{ + struct intel_vgpu_shadow_page *p = find_shadow_page(vgpu, mfn); + + if (p) + return shadow_page_to_ppgtt_spt(p); + + gvt_err("vgpu%d: fail to find ppgtt shadow page: 0x%lx\n", + vgpu->id, mfn); + return NULL; +} + +#define pt_entry_size_shift(spt) \ + ((spt)->vgpu->gvt->device_info.gtt_entry_size_shift) + +#define pt_entries(spt) \ + (GTT_PAGE_SIZE >> pt_entry_size_shift(spt)) + +#define for_each_present_guest_entry(spt, e, i) \ + for (i = 0; i < pt_entries(spt); i++) \ + if (spt->vgpu->gvt->gtt.pte_ops->test_present( \ + ppgtt_get_guest_entry(spt, e, i))) + +#define for_each_present_shadow_entry(spt, e, i) \ + for (i = 0; i < pt_entries(spt); i++) \ + if (spt->vgpu->gvt->gtt.pte_ops->test_present( \ + ppgtt_get_shadow_entry(spt, e, i))) + +static void ppgtt_get_shadow_page(struct intel_vgpu_ppgtt_spt *spt) +{ + int v = atomic_read(&spt->refcount); + + trace_spt_refcount(spt->vgpu->id, "inc", spt, v, (v + 1)); + + atomic_inc(&spt->refcount); +} + +static int ppgtt_invalidate_shadow_page(struct intel_vgpu_ppgtt_spt *spt); + +static int ppgtt_invalidate_shadow_page_by_shadow_entry(struct intel_vgpu *vgpu, + struct intel_gvt_gtt_entry *e) +{ + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + struct intel_vgpu_ppgtt_spt *s; + intel_gvt_gtt_type_t cur_pt_type; + + if (WARN_ON(!gtt_type_is_pt(get_next_pt_type(e->type)))) + return -EINVAL; + + if (e->type != GTT_TYPE_PPGTT_ROOT_L3_ENTRY + && e->type != GTT_TYPE_PPGTT_ROOT_L4_ENTRY) { + cur_pt_type = get_next_pt_type(e->type) + 1; + if (ops->get_pfn(e) == + vgpu->gtt.scratch_pt[cur_pt_type].page_mfn) + return 0; + } + s = ppgtt_find_shadow_page(vgpu, ops->get_pfn(e)); + if (!s) { + gvt_err("vgpu%d: fail to find shadow page: mfn: 0x%lx\n", + vgpu->id, ops->get_pfn(e)); + return -ENXIO; + } + return ppgtt_invalidate_shadow_page(s); +} + +static int ppgtt_invalidate_shadow_page(struct intel_vgpu_ppgtt_spt *spt) +{ + struct intel_gvt_gtt_entry e; + unsigned long index; + int ret; + int v = atomic_read(&spt->refcount); + + trace_spt_change(spt->vgpu->id, "die", spt, + spt->guest_page.gfn, spt->shadow_page.type); + + trace_spt_refcount(spt->vgpu->id, "dec", spt, v, (v - 1)); + + if (atomic_dec_return(&spt->refcount) > 0) + return 0; + + if (gtt_type_is_pte_pt(spt->shadow_page.type)) + goto release; + + for_each_present_shadow_entry(spt, &e, index) { + if (!gtt_type_is_pt(get_next_pt_type(e.type))) { + gvt_err("GVT doesn't support pse bit for now\n"); + return -EINVAL; + } + ret = ppgtt_invalidate_shadow_page_by_shadow_entry( + spt->vgpu, &e); + if (ret) + goto fail; + } +release: + trace_spt_change(spt->vgpu->id, "release", spt, + spt->guest_page.gfn, spt->shadow_page.type); + ppgtt_free_shadow_page(spt); + return 0; +fail: + gvt_err("vgpu%d: fail: shadow page %p shadow entry 0x%llx type %d\n", + spt->vgpu->id, spt, e.val64, e.type); + return ret; +} + +static int ppgtt_populate_shadow_page(struct intel_vgpu_ppgtt_spt *spt); + +static struct intel_vgpu_ppgtt_spt *ppgtt_populate_shadow_page_by_guest_entry( + struct intel_vgpu *vgpu, struct intel_gvt_gtt_entry *we) +{ + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + struct intel_vgpu_ppgtt_spt *s = NULL; + struct intel_vgpu_guest_page *g; + int ret; + + if (WARN_ON(!gtt_type_is_pt(get_next_pt_type(we->type)))) { + ret = -EINVAL; + goto fail; + } + + g = intel_vgpu_find_guest_page(vgpu, ops->get_pfn(we)); + if (g) { + s = guest_page_to_ppgtt_spt(g); + ppgtt_get_shadow_page(s); + } else { + int type = get_next_pt_type(we->type); + + s = ppgtt_alloc_shadow_page(vgpu, type, ops->get_pfn(we)); + if (IS_ERR(s)) { + ret = PTR_ERR(s); + goto fail; + } + + ret = intel_gvt_hypervisor_set_wp_page(vgpu, &s->guest_page); + if (ret) + goto fail; + + ret = ppgtt_populate_shadow_page(s); + if (ret) + goto fail; + + trace_spt_change(vgpu->id, "new", s, s->guest_page.gfn, + s->shadow_page.type); + } + return s; +fail: + gvt_err("vgpu%d: fail: shadow page %p guest entry 0x%llx type %d\n", + vgpu->id, s, we->val64, we->type); + return ERR_PTR(ret); +} + +static inline void ppgtt_generate_shadow_entry(struct intel_gvt_gtt_entry *se, + struct intel_vgpu_ppgtt_spt *s, struct intel_gvt_gtt_entry *ge) +{ + struct intel_gvt_gtt_pte_ops *ops = s->vgpu->gvt->gtt.pte_ops; + + se->type = ge->type; + se->val64 = ge->val64; + + ops->set_pfn(se, s->shadow_page.mfn); +} + +static int ppgtt_populate_shadow_page(struct intel_vgpu_ppgtt_spt *spt) +{ + struct intel_vgpu *vgpu = spt->vgpu; + struct intel_vgpu_ppgtt_spt *s; + struct intel_gvt_gtt_entry se, ge; + unsigned long i; + int ret; + + trace_spt_change(spt->vgpu->id, "born", spt, + spt->guest_page.gfn, spt->shadow_page.type); + + if (gtt_type_is_pte_pt(spt->shadow_page.type)) { + for_each_present_guest_entry(spt, &ge, i) { + ret = gtt_entry_p2m(vgpu, &ge, &se); + if (ret) + goto fail; + ppgtt_set_shadow_entry(spt, &se, i); + } + return 0; + } + + for_each_present_guest_entry(spt, &ge, i) { + if (!gtt_type_is_pt(get_next_pt_type(ge.type))) { + gvt_err("GVT doesn't support pse bit now\n"); + ret = -EINVAL; + goto fail; + } + + s = ppgtt_populate_shadow_page_by_guest_entry(vgpu, &ge); + if (IS_ERR(s)) { + ret = PTR_ERR(s); + goto fail; + } + ppgtt_get_shadow_entry(spt, &se, i); + ppgtt_generate_shadow_entry(&se, s, &ge); + ppgtt_set_shadow_entry(spt, &se, i); + } + return 0; +fail: + gvt_err("vgpu%d: fail: shadow page %p guest entry 0x%llx type %d\n", + vgpu->id, spt, ge.val64, ge.type); + return ret; +} + +static int ppgtt_handle_guest_entry_removal(struct intel_vgpu_guest_page *gpt, + unsigned long index) +{ + struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt); + struct intel_vgpu_shadow_page *sp = &spt->shadow_page; + struct intel_vgpu *vgpu = spt->vgpu; + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + struct intel_gvt_gtt_entry e; + int ret; + + ppgtt_get_shadow_entry(spt, &e, index); + + trace_gpt_change(spt->vgpu->id, "remove", spt, sp->type, e.val64, + index); + + if (!ops->test_present(&e)) + return 0; + + if (ops->get_pfn(&e) == vgpu->gtt.scratch_pt[sp->type].page_mfn) + return 0; + + if (gtt_type_is_pt(get_next_pt_type(e.type))) { + struct intel_vgpu_ppgtt_spt *s = + ppgtt_find_shadow_page(vgpu, ops->get_pfn(&e)); + if (!s) { + gvt_err("fail to find guest page\n"); + ret = -ENXIO; + goto fail; + } + ret = ppgtt_invalidate_shadow_page(s); + if (ret) + goto fail; + } + ops->set_pfn(&e, vgpu->gtt.scratch_pt[sp->type].page_mfn); + ppgtt_set_shadow_entry(spt, &e, index); + return 0; +fail: + gvt_err("vgpu%d: fail: shadow page %p guest entry 0x%llx type %d\n", + vgpu->id, spt, e.val64, e.type); + return ret; +} + +static int ppgtt_handle_guest_entry_add(struct intel_vgpu_guest_page *gpt, + struct intel_gvt_gtt_entry *we, unsigned long index) +{ + struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt); + struct intel_vgpu_shadow_page *sp = &spt->shadow_page; + struct intel_vgpu *vgpu = spt->vgpu; + struct intel_gvt_gtt_entry m; + struct intel_vgpu_ppgtt_spt *s; + int ret; + + trace_gpt_change(spt->vgpu->id, "add", spt, sp->type, + we->val64, index); + + if (gtt_type_is_pt(get_next_pt_type(we->type))) { + s = ppgtt_populate_shadow_page_by_guest_entry(vgpu, we); + if (IS_ERR(s)) { + ret = PTR_ERR(s); + goto fail; + } + ppgtt_get_shadow_entry(spt, &m, index); + ppgtt_generate_shadow_entry(&m, s, we); + ppgtt_set_shadow_entry(spt, &m, index); + } else { + ret = gtt_entry_p2m(vgpu, we, &m); + if (ret) + goto fail; + ppgtt_set_shadow_entry(spt, &m, index); + } + return 0; +fail: + gvt_err("vgpu%d: fail: spt %p guest entry 0x%llx type %d\n", vgpu->id, + spt, we->val64, we->type); + return ret; +} + +static int sync_oos_page(struct intel_vgpu *vgpu, + struct intel_vgpu_oos_page *oos_page) +{ + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + struct intel_vgpu_ppgtt_spt *spt = + guest_page_to_ppgtt_spt(oos_page->guest_page); + struct intel_gvt_gtt_entry old, new, m; + int index; + int ret; + + trace_oos_change(vgpu->id, "sync", oos_page->id, + oos_page->guest_page, spt->guest_page_type); + + old.type = new.type = get_entry_type(spt->guest_page_type); + old.val64 = new.val64 = 0; + + for (index = 0; index < (GTT_PAGE_SIZE >> info->gtt_entry_size_shift); + index++) { + ops->get_entry(oos_page->mem, &old, index, false, 0, vgpu); + ops->get_entry(NULL, &new, index, true, + oos_page->guest_page->gfn << PAGE_SHIFT, vgpu); + + if (old.val64 == new.val64 + && !test_and_clear_bit(index, spt->post_shadow_bitmap)) + continue; + + trace_oos_sync(vgpu->id, oos_page->id, + oos_page->guest_page, spt->guest_page_type, + new.val64, index); + + ret = gtt_entry_p2m(vgpu, &new, &m); + if (ret) + return ret; + + ops->set_entry(oos_page->mem, &new, index, false, 0, vgpu); + ppgtt_set_shadow_entry(spt, &m, index); + } + + oos_page->guest_page->write_cnt = 0; + list_del_init(&spt->post_shadow_list); + return 0; +} + +static int detach_oos_page(struct intel_vgpu *vgpu, + struct intel_vgpu_oos_page *oos_page) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_vgpu_ppgtt_spt *spt = + guest_page_to_ppgtt_spt(oos_page->guest_page); + + trace_oos_change(vgpu->id, "detach", oos_page->id, + oos_page->guest_page, spt->guest_page_type); + + oos_page->guest_page->write_cnt = 0; + oos_page->guest_page->oos_page = NULL; + oos_page->guest_page = NULL; + + list_del_init(&oos_page->vm_list); + list_move_tail(&oos_page->list, &gvt->gtt.oos_page_free_list_head); + + return 0; +} + +static int attach_oos_page(struct intel_vgpu *vgpu, + struct intel_vgpu_oos_page *oos_page, + struct intel_vgpu_guest_page *gpt) +{ + struct intel_gvt *gvt = vgpu->gvt; + int ret; + + ret = intel_gvt_hypervisor_read_gpa(vgpu, gpt->gfn << GTT_PAGE_SHIFT, + oos_page->mem, GTT_PAGE_SIZE); + if (ret) + return ret; + + oos_page->guest_page = gpt; + gpt->oos_page = oos_page; + + list_move_tail(&oos_page->list, &gvt->gtt.oos_page_use_list_head); + + trace_oos_change(vgpu->id, "attach", gpt->oos_page->id, + gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type); + return 0; +} + +static int ppgtt_set_guest_page_sync(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *gpt) +{ + int ret; + + ret = intel_gvt_hypervisor_set_wp_page(vgpu, gpt); + if (ret) + return ret; + + trace_oos_change(vgpu->id, "set page sync", gpt->oos_page->id, + gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type); + + list_del_init(&gpt->oos_page->vm_list); + return sync_oos_page(vgpu, gpt->oos_page); +} + +static int ppgtt_allocate_oos_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *gpt) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct intel_vgpu_oos_page *oos_page = gpt->oos_page; + int ret; + + WARN(oos_page, "shadow PPGTT page has already has a oos page\n"); + + if (list_empty(>t->oos_page_free_list_head)) { + oos_page = container_of(gtt->oos_page_use_list_head.next, + struct intel_vgpu_oos_page, list); + ret = ppgtt_set_guest_page_sync(vgpu, oos_page->guest_page); + if (ret) + return ret; + ret = detach_oos_page(vgpu, oos_page); + if (ret) + return ret; + } else + oos_page = container_of(gtt->oos_page_free_list_head.next, + struct intel_vgpu_oos_page, list); + return attach_oos_page(vgpu, oos_page, gpt); +} + +static int ppgtt_set_guest_page_oos(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *gpt) +{ + struct intel_vgpu_oos_page *oos_page = gpt->oos_page; + + if (WARN(!oos_page, "shadow PPGTT page should have a oos page\n")) + return -EINVAL; + + trace_oos_change(vgpu->id, "set page out of sync", gpt->oos_page->id, + gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type); + + list_add_tail(&oos_page->vm_list, &vgpu->gtt.oos_page_list_head); + return intel_gvt_hypervisor_unset_wp_page(vgpu, gpt); +} + +/** + * intel_vgpu_sync_oos_pages - sync all the out-of-synced shadow for vGPU + * @vgpu: a vGPU + * + * This function is called before submitting a guest workload to host, + * to sync all the out-of-synced shadow for vGPU + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_sync_oos_pages(struct intel_vgpu *vgpu) +{ + struct list_head *pos, *n; + struct intel_vgpu_oos_page *oos_page; + int ret; + + if (!enable_out_of_sync) + return 0; + + list_for_each_safe(pos, n, &vgpu->gtt.oos_page_list_head) { + oos_page = container_of(pos, + struct intel_vgpu_oos_page, vm_list); + ret = ppgtt_set_guest_page_sync(vgpu, oos_page->guest_page); + if (ret) + return ret; + } + return 0; +} + +/* + * The heart of PPGTT shadow page table. + */ +static int ppgtt_handle_guest_write_page_table( + struct intel_vgpu_guest_page *gpt, + struct intel_gvt_gtt_entry *we, unsigned long index) +{ + struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt); + struct intel_vgpu *vgpu = spt->vgpu; + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + + int ret; + int new_present; + + new_present = ops->test_present(we); + + ret = ppgtt_handle_guest_entry_removal(gpt, index); + if (ret) + goto fail; + + if (new_present) { + ret = ppgtt_handle_guest_entry_add(gpt, we, index); + if (ret) + goto fail; + } + return 0; +fail: + gvt_err("vgpu%d: fail: shadow page %p guest entry 0x%llx type %d.\n", + vgpu->id, spt, we->val64, we->type); + return ret; +} + +static inline bool can_do_out_of_sync(struct intel_vgpu_guest_page *gpt) +{ + return enable_out_of_sync + && gtt_type_is_pte_pt( + guest_page_to_ppgtt_spt(gpt)->guest_page_type) + && gpt->write_cnt >= 2; +} + +static void ppgtt_set_post_shadow(struct intel_vgpu_ppgtt_spt *spt, + unsigned long index) +{ + set_bit(index, spt->post_shadow_bitmap); + if (!list_empty(&spt->post_shadow_list)) + return; + + list_add_tail(&spt->post_shadow_list, + &spt->vgpu->gtt.post_shadow_list_head); +} + +/** + * intel_vgpu_flush_post_shadow - flush the post shadow transactions + * @vgpu: a vGPU + * + * This function is called before submitting a guest workload to host, + * to flush all the post shadows for a vGPU. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_flush_post_shadow(struct intel_vgpu *vgpu) +{ + struct list_head *pos, *n; + struct intel_vgpu_ppgtt_spt *spt; + struct intel_gvt_gtt_entry ge; + unsigned long index; + int ret; + + list_for_each_safe(pos, n, &vgpu->gtt.post_shadow_list_head) { + spt = container_of(pos, struct intel_vgpu_ppgtt_spt, + post_shadow_list); + + for_each_set_bit(index, spt->post_shadow_bitmap, + GTT_ENTRY_NUM_IN_ONE_PAGE) { + ppgtt_get_guest_entry(spt, &ge, index); + + ret = ppgtt_handle_guest_write_page_table( + &spt->guest_page, &ge, index); + if (ret) + return ret; + clear_bit(index, spt->post_shadow_bitmap); + } + list_del_init(&spt->post_shadow_list); + } + return 0; +} + +static int ppgtt_handle_guest_write_page_table_bytes(void *gp, + u64 pa, void *p_data, int bytes) +{ + struct intel_vgpu_guest_page *gpt = (struct intel_vgpu_guest_page *)gp; + struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt); + struct intel_vgpu *vgpu = spt->vgpu; + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + struct intel_gvt_gtt_entry we; + unsigned long index; + int ret; + + index = (pa & (PAGE_SIZE - 1)) >> info->gtt_entry_size_shift; + + ppgtt_get_guest_entry(spt, &we, index); + + ops->test_pse(&we); + + if (bytes == info->gtt_entry_size) { + ret = ppgtt_handle_guest_write_page_table(gpt, &we, index); + if (ret) + return ret; + } else { + if (!test_bit(index, spt->post_shadow_bitmap)) { + ret = ppgtt_handle_guest_entry_removal(gpt, index); + if (ret) + return ret; + } + + ppgtt_set_post_shadow(spt, index); + } + + if (!enable_out_of_sync) + return 0; + + gpt->write_cnt++; + + if (gpt->oos_page) + ops->set_entry(gpt->oos_page->mem, &we, index, + false, 0, vgpu); + + if (can_do_out_of_sync(gpt)) { + if (!gpt->oos_page) + ppgtt_allocate_oos_page(vgpu, gpt); + + ret = ppgtt_set_guest_page_oos(vgpu, gpt); + if (ret < 0) + return ret; + } + return 0; +} + +/* + * mm page table allocation policy for bdw+ + * - for ggtt, only virtual page table will be allocated. + * - for ppgtt, dedicated virtual/shadow page table will be allocated. + */ +static int gen8_mm_alloc_page_table(struct intel_vgpu_mm *mm) +{ + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + const struct intel_gvt_device_info *info = &gvt->device_info; + void *mem; + + if (mm->type == INTEL_GVT_MM_PPGTT) { + mm->page_table_entry_cnt = 4; + mm->page_table_entry_size = mm->page_table_entry_cnt * + info->gtt_entry_size; + mem = kzalloc(mm->has_shadow_page_table ? + mm->page_table_entry_size * 2 + : mm->page_table_entry_size, + GFP_ATOMIC); + if (!mem) + return -ENOMEM; + mm->virtual_page_table = mem; + if (!mm->has_shadow_page_table) + return 0; + mm->shadow_page_table = mem + mm->page_table_entry_size; + } else if (mm->type == INTEL_GVT_MM_GGTT) { + mm->page_table_entry_cnt = + (gvt_ggtt_gm_sz(gvt) >> GTT_PAGE_SHIFT); + mm->page_table_entry_size = mm->page_table_entry_cnt * + info->gtt_entry_size; + mem = vzalloc(mm->page_table_entry_size); + if (!mem) + return -ENOMEM; + mm->virtual_page_table = mem; + } + return 0; +} + +static void gen8_mm_free_page_table(struct intel_vgpu_mm *mm) +{ + if (mm->type == INTEL_GVT_MM_PPGTT) { + kfree(mm->virtual_page_table); + } else if (mm->type == INTEL_GVT_MM_GGTT) { + if (mm->virtual_page_table) + vfree(mm->virtual_page_table); + } + mm->virtual_page_table = mm->shadow_page_table = NULL; +} + +static void invalidate_mm(struct intel_vgpu_mm *mm) +{ + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct intel_gvt_gtt_pte_ops *ops = gtt->pte_ops; + struct intel_gvt_gtt_entry se; + int i; + + if (WARN_ON(!mm->has_shadow_page_table || !mm->shadowed)) + return; + + for (i = 0; i < mm->page_table_entry_cnt; i++) { + ppgtt_get_shadow_root_entry(mm, &se, i); + if (!ops->test_present(&se)) + continue; + ppgtt_invalidate_shadow_page_by_shadow_entry( + vgpu, &se); + se.val64 = 0; + ppgtt_set_shadow_root_entry(mm, &se, i); + + trace_gpt_change(vgpu->id, "destroy root pointer", + NULL, se.type, se.val64, i); + } + mm->shadowed = false; +} + +/** + * intel_vgpu_destroy_mm - destroy a mm object + * @mm: a kref object + * + * This function is used to destroy a mm object for vGPU + * + */ +void intel_vgpu_destroy_mm(struct kref *mm_ref) +{ + struct intel_vgpu_mm *mm = container_of(mm_ref, typeof(*mm), ref); + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt *gtt = &gvt->gtt; + + if (!mm->initialized) + goto out; + + list_del(&mm->list); + list_del(&mm->lru_list); + + if (mm->has_shadow_page_table) + invalidate_mm(mm); + + gtt->mm_free_page_table(mm); +out: + kfree(mm); +} + +static int shadow_mm(struct intel_vgpu_mm *mm) +{ + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct intel_gvt_gtt_pte_ops *ops = gtt->pte_ops; + struct intel_vgpu_ppgtt_spt *spt; + struct intel_gvt_gtt_entry ge, se; + int i; + int ret; + + if (WARN_ON(!mm->has_shadow_page_table || mm->shadowed)) + return 0; + + mm->shadowed = true; + + for (i = 0; i < mm->page_table_entry_cnt; i++) { + ppgtt_get_guest_root_entry(mm, &ge, i); + if (!ops->test_present(&ge)) + continue; + + trace_gpt_change(vgpu->id, __func__, NULL, + ge.type, ge.val64, i); + + spt = ppgtt_populate_shadow_page_by_guest_entry(vgpu, &ge); + if (IS_ERR(spt)) { + gvt_err("fail to populate guest root pointer\n"); + ret = PTR_ERR(spt); + goto fail; + } + ppgtt_generate_shadow_entry(&se, spt, &ge); + ppgtt_set_shadow_root_entry(mm, &se, i); + + trace_gpt_change(vgpu->id, "populate root pointer", + NULL, se.type, se.val64, i); + } + return 0; +fail: + invalidate_mm(mm); + return ret; +} + +/** + * intel_vgpu_create_mm - create a mm object for a vGPU + * @vgpu: a vGPU + * @mm_type: mm object type, should be PPGTT or GGTT + * @virtual_page_table: page table root pointers. Could be NULL if user wants + * to populate shadow later. + * @page_table_level: describe the page table level of the mm object + * @pde_base_index: pde root pointer base in GGTT MMIO. + * + * This function is used to create a mm object for a vGPU. + * + * Returns: + * Zero on success, negative error code in pointer if failed. + */ +struct intel_vgpu_mm *intel_vgpu_create_mm(struct intel_vgpu *vgpu, + int mm_type, void *virtual_page_table, int page_table_level, + u32 pde_base_index) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct intel_vgpu_mm *mm; + int ret; + + mm = kzalloc(sizeof(*mm), GFP_ATOMIC); + if (!mm) { + ret = -ENOMEM; + goto fail; + } + + mm->type = mm_type; + + if (page_table_level == 1) + mm->page_table_entry_type = GTT_TYPE_GGTT_PTE; + else if (page_table_level == 3) + mm->page_table_entry_type = GTT_TYPE_PPGTT_ROOT_L3_ENTRY; + else if (page_table_level == 4) + mm->page_table_entry_type = GTT_TYPE_PPGTT_ROOT_L4_ENTRY; + else { + WARN_ON(1); + ret = -EINVAL; + goto fail; + } + + mm->page_table_level = page_table_level; + mm->pde_base_index = pde_base_index; + + mm->vgpu = vgpu; + mm->has_shadow_page_table = !!(mm_type == INTEL_GVT_MM_PPGTT); + + kref_init(&mm->ref); + atomic_set(&mm->pincount, 0); + INIT_LIST_HEAD(&mm->list); + INIT_LIST_HEAD(&mm->lru_list); + list_add_tail(&mm->list, &vgpu->gtt.mm_list_head); + + ret = gtt->mm_alloc_page_table(mm); + if (ret) { + gvt_err("fail to allocate page table for mm\n"); + goto fail; + } + + mm->initialized = true; + + if (virtual_page_table) + memcpy(mm->virtual_page_table, virtual_page_table, + mm->page_table_entry_size); + + if (mm->has_shadow_page_table) { + ret = shadow_mm(mm); + if (ret) + goto fail; + list_add_tail(&mm->lru_list, &gvt->gtt.mm_lru_list_head); + } + return mm; +fail: + gvt_err("fail to create mm\n"); + if (mm) + intel_gvt_mm_unreference(mm); + return ERR_PTR(ret); +} + +/** + * intel_vgpu_unpin_mm - decrease the pin count of a vGPU mm object + * @mm: a vGPU mm object + * + * This function is called when user doesn't want to use a vGPU mm object + */ +void intel_vgpu_unpin_mm(struct intel_vgpu_mm *mm) +{ + if (WARN_ON(mm->type != INTEL_GVT_MM_PPGTT)) + return; + + atomic_dec(&mm->pincount); +} + +/** + * intel_vgpu_pin_mm - increase the pin count of a vGPU mm object + * @vgpu: a vGPU + * + * This function is called when user wants to use a vGPU mm object. If this + * mm object hasn't been shadowed yet, the shadow will be populated at this + * time. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_pin_mm(struct intel_vgpu_mm *mm) +{ + int ret; + + if (WARN_ON(mm->type != INTEL_GVT_MM_PPGTT)) + return 0; + + atomic_inc(&mm->pincount); + + if (!mm->shadowed) { + ret = shadow_mm(mm); + if (ret) + return ret; + } + + list_del_init(&mm->lru_list); + list_add_tail(&mm->lru_list, &mm->vgpu->gvt->gtt.mm_lru_list_head); + return 0; +} + +static int reclaim_one_mm(struct intel_gvt *gvt) +{ + struct intel_vgpu_mm *mm; + struct list_head *pos, *n; + + list_for_each_safe(pos, n, &gvt->gtt.mm_lru_list_head) { + mm = container_of(pos, struct intel_vgpu_mm, lru_list); + + if (mm->type != INTEL_GVT_MM_PPGTT) + continue; + if (atomic_read(&mm->pincount)) + continue; + + list_del_init(&mm->lru_list); + invalidate_mm(mm); + return 1; + } + return 0; +} + +/* + * GMA translation APIs. + */ +static inline int ppgtt_get_next_level_entry(struct intel_vgpu_mm *mm, + struct intel_gvt_gtt_entry *e, unsigned long index, bool guest) +{ + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + struct intel_vgpu_ppgtt_spt *s; + + if (WARN_ON(!mm->has_shadow_page_table)) + return -EINVAL; + + s = ppgtt_find_shadow_page(vgpu, ops->get_pfn(e)); + if (!s) + return -ENXIO; + + if (!guest) + ppgtt_get_shadow_entry(s, e, index); + else + ppgtt_get_guest_entry(s, e, index); + return 0; +} + +/** + * intel_vgpu_gma_to_gpa - translate a gma to GPA + * @mm: mm object. could be a PPGTT or GGTT mm object + * @gma: graphics memory address in this mm object + * + * This function is used to translate a graphics memory address in specific + * graphics memory space to guest physical address. + * + * Returns: + * Guest physical address on success, INTEL_GVT_INVALID_ADDR if failed. + */ +unsigned long intel_vgpu_gma_to_gpa(struct intel_vgpu_mm *mm, unsigned long gma) +{ + struct intel_vgpu *vgpu = mm->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_gtt_pte_ops *pte_ops = gvt->gtt.pte_ops; + struct intel_gvt_gtt_gma_ops *gma_ops = gvt->gtt.gma_ops; + unsigned long gpa = INTEL_GVT_INVALID_ADDR; + unsigned long gma_index[4]; + struct intel_gvt_gtt_entry e; + int i, index; + int ret; + + if (mm->type != INTEL_GVT_MM_GGTT && mm->type != INTEL_GVT_MM_PPGTT) + return INTEL_GVT_INVALID_ADDR; + + if (mm->type == INTEL_GVT_MM_GGTT) { + if (!vgpu_gmadr_is_valid(vgpu, gma)) + goto err; + + ggtt_get_guest_entry(mm, &e, + gma_ops->gma_to_ggtt_pte_index(gma)); + gpa = (pte_ops->get_pfn(&e) << GTT_PAGE_SHIFT) + + (gma & ~GTT_PAGE_MASK); + + trace_gma_translate(vgpu->id, "ggtt", 0, 0, gma, gpa); + return gpa; + } + + switch (mm->page_table_level) { + case 4: + ppgtt_get_shadow_root_entry(mm, &e, 0); + gma_index[0] = gma_ops->gma_to_pml4_index(gma); + gma_index[1] = gma_ops->gma_to_l4_pdp_index(gma); + gma_index[2] = gma_ops->gma_to_pde_index(gma); + gma_index[3] = gma_ops->gma_to_pte_index(gma); + index = 4; + break; + case 3: + ppgtt_get_shadow_root_entry(mm, &e, + gma_ops->gma_to_l3_pdp_index(gma)); + gma_index[0] = gma_ops->gma_to_pde_index(gma); + gma_index[1] = gma_ops->gma_to_pte_index(gma); + index = 2; + break; + case 2: + ppgtt_get_shadow_root_entry(mm, &e, + gma_ops->gma_to_pde_index(gma)); + gma_index[0] = gma_ops->gma_to_pte_index(gma); + index = 1; + break; + default: + WARN_ON(1); + goto err; + } + + /* walk into the shadow page table and get gpa from guest entry */ + for (i = 0; i < index; i++) { + ret = ppgtt_get_next_level_entry(mm, &e, gma_index[i], + (i == index - 1)); + if (ret) + goto err; + } + + gpa = (pte_ops->get_pfn(&e) << GTT_PAGE_SHIFT) + + (gma & ~GTT_PAGE_MASK); + + trace_gma_translate(vgpu->id, "ppgtt", 0, + mm->page_table_level, gma, gpa); + return gpa; +err: + gvt_err("invalid mm type: %d gma %lx\n", mm->type, gma); + return INTEL_GVT_INVALID_ADDR; +} + +static int emulate_gtt_mmio_read(struct intel_vgpu *vgpu, + unsigned int off, void *p_data, unsigned int bytes) +{ + struct intel_vgpu_mm *ggtt_mm = vgpu->gtt.ggtt_mm; + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + unsigned long index = off >> info->gtt_entry_size_shift; + struct intel_gvt_gtt_entry e; + + if (bytes != 4 && bytes != 8) + return -EINVAL; + + ggtt_get_guest_entry(ggtt_mm, &e, index); + memcpy(p_data, (void *)&e.val64 + (off & (info->gtt_entry_size - 1)), + bytes); + return 0; +} + +/** + * intel_vgpu_emulate_gtt_mmio_read - emulate GTT MMIO register read + * @vgpu: a vGPU + * @off: register offset + * @p_data: data will be returned to guest + * @bytes: data length + * + * This function is used to emulate the GTT MMIO register read + * + * Returns: + * Zero on success, error code if failed. + */ +int intel_vgpu_emulate_gtt_mmio_read(struct intel_vgpu *vgpu, unsigned int off, + void *p_data, unsigned int bytes) +{ + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + int ret; + + if (bytes != 4 && bytes != 8) + return -EINVAL; + + off -= info->gtt_start_offset; + ret = emulate_gtt_mmio_read(vgpu, off, p_data, bytes); + return ret; +} + +static int emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off, + void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + const struct intel_gvt_device_info *info = &gvt->device_info; + struct intel_vgpu_mm *ggtt_mm = vgpu->gtt.ggtt_mm; + struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops; + unsigned long g_gtt_index = off >> info->gtt_entry_size_shift; + unsigned long gma; + struct intel_gvt_gtt_entry e, m; + int ret; + + if (bytes != 4 && bytes != 8) + return -EINVAL; + + gma = g_gtt_index << GTT_PAGE_SHIFT; + + /* the VM may configure the whole GM space when ballooning is used */ + if (WARN_ONCE(!vgpu_gmadr_is_valid(vgpu, gma), + "vgpu%d: found oob ggtt write, offset %x\n", + vgpu->id, off)) { + return 0; + } + + ggtt_get_guest_entry(ggtt_mm, &e, g_gtt_index); + + memcpy((void *)&e.val64 + (off & (info->gtt_entry_size - 1)), p_data, + bytes); + + if (ops->test_present(&e)) { + ret = gtt_entry_p2m(vgpu, &e, &m); + if (ret) { + gvt_err("vgpu%d: fail to translate guest gtt entry\n", + vgpu->id); + return ret; + } + } else { + m = e; + m.val64 = 0; + } + + ggtt_set_shadow_entry(ggtt_mm, &m, g_gtt_index); + ggtt_set_guest_entry(ggtt_mm, &e, g_gtt_index); + return 0; +} + +/* + * intel_vgpu_emulate_gtt_mmio_write - emulate GTT MMIO register write + * @vgpu: a vGPU + * @off: register offset + * @p_data: data from guest write + * @bytes: data length + * + * This function is used to emulate the GTT MMIO register write + * + * Returns: + * Zero on success, error code if failed. + */ +int intel_vgpu_emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off, + void *p_data, unsigned int bytes) +{ + const struct intel_gvt_device_info *info = &vgpu->gvt->device_info; + int ret; + + if (bytes != 4 && bytes != 8) + return -EINVAL; + + off -= info->gtt_start_offset; + ret = emulate_gtt_mmio_write(vgpu, off, p_data, bytes); + return ret; +} + +static int alloc_scratch_pages(struct intel_vgpu *vgpu, + intel_gvt_gtt_type_t type) +{ + struct intel_vgpu_gtt *gtt = &vgpu->gtt; + struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops; + int page_entry_num = GTT_PAGE_SIZE >> + vgpu->gvt->device_info.gtt_entry_size_shift; + struct page *scratch_pt; + unsigned long mfn; + int i; + void *p; + + if (WARN_ON(type < GTT_TYPE_PPGTT_PTE_PT || type >= GTT_TYPE_MAX)) + return -EINVAL; + + scratch_pt = alloc_page(GFP_KERNEL | GFP_ATOMIC | __GFP_ZERO); + if (!scratch_pt) { + gvt_err("fail to allocate scratch page\n"); + return -ENOMEM; + } + + p = kmap_atomic(scratch_pt); + mfn = intel_gvt_hypervisor_virt_to_mfn(p); + if (mfn == INTEL_GVT_INVALID_ADDR) { + gvt_err("fail to translate vaddr:0x%llx\n", (u64)p); + kunmap_atomic(p); + __free_page(scratch_pt); + return -EFAULT; + } + gtt->scratch_pt[type].page_mfn = mfn; + gtt->scratch_pt[type].page = scratch_pt; + gvt_dbg_mm("vgpu%d create scratch_pt: type %d mfn=0x%lx\n", + vgpu->id, type, mfn); + + /* Build the tree by full filled the scratch pt with the entries which + * point to the next level scratch pt or scratch page. The + * scratch_pt[type] indicate the scratch pt/scratch page used by the + * 'type' pt. + * e.g. scratch_pt[GTT_TYPE_PPGTT_PDE_PT] is used by + * GTT_TYPE_PPGTT_PDE_PT level pt, that means this scatch_pt it self + * is GTT_TYPE_PPGTT_PTE_PT, and full filled by scratch page mfn. + */ + if (type > GTT_TYPE_PPGTT_PTE_PT && type < GTT_TYPE_MAX) { + struct intel_gvt_gtt_entry se; + + memset(&se, 0, sizeof(struct intel_gvt_gtt_entry)); + se.type = get_entry_type(type - 1); + ops->set_pfn(&se, gtt->scratch_pt[type - 1].page_mfn); + + /* The entry parameters like present/writeable/cache type + * set to the same as i915's scratch page tree. + */ + se.val64 |= _PAGE_PRESENT | _PAGE_RW; + if (type == GTT_TYPE_PPGTT_PDE_PT) + se.val64 |= PPAT_CACHED_INDEX; + + for (i = 0; i < page_entry_num; i++) + ops->set_entry(p, &se, i, false, 0, vgpu); + } + + kunmap_atomic(p); + + return 0; +} + +static int release_scratch_page_tree(struct intel_vgpu *vgpu) +{ + int i; + + for (i = GTT_TYPE_PPGTT_PTE_PT; i < GTT_TYPE_MAX; i++) { + if (vgpu->gtt.scratch_pt[i].page != NULL) { + __free_page(vgpu->gtt.scratch_pt[i].page); + vgpu->gtt.scratch_pt[i].page = NULL; + vgpu->gtt.scratch_pt[i].page_mfn = 0; + } + } + + return 0; +} + +static int create_scratch_page_tree(struct intel_vgpu *vgpu) +{ + int i, ret; + + for (i = GTT_TYPE_PPGTT_PTE_PT; i < GTT_TYPE_MAX; i++) { + ret = alloc_scratch_pages(vgpu, i); + if (ret) + goto err; + } + + return 0; + +err: + release_scratch_page_tree(vgpu); + return ret; +} + +/** + * intel_vgpu_init_gtt - initialize per-vGPU graphics memory virulization + * @vgpu: a vGPU + * + * This function is used to initialize per-vGPU graphics memory virtualization + * components. + * + * Returns: + * Zero on success, error code if failed. + */ +int intel_vgpu_init_gtt(struct intel_vgpu *vgpu) +{ + struct intel_vgpu_gtt *gtt = &vgpu->gtt; + struct intel_vgpu_mm *ggtt_mm; + + hash_init(gtt->guest_page_hash_table); + hash_init(gtt->shadow_page_hash_table); + + INIT_LIST_HEAD(>t->mm_list_head); + INIT_LIST_HEAD(>t->oos_page_list_head); + INIT_LIST_HEAD(>t->post_shadow_list_head); + + ggtt_mm = intel_vgpu_create_mm(vgpu, INTEL_GVT_MM_GGTT, + NULL, 1, 0); + if (IS_ERR(ggtt_mm)) { + gvt_err("fail to create mm for ggtt.\n"); + return PTR_ERR(ggtt_mm); + } + + gtt->ggtt_mm = ggtt_mm; + + return create_scratch_page_tree(vgpu); +} + +/** + * intel_vgpu_clean_gtt - clean up per-vGPU graphics memory virulization + * @vgpu: a vGPU + * + * This function is used to clean up per-vGPU graphics memory virtualization + * components. + * + * Returns: + * Zero on success, error code if failed. + */ +void intel_vgpu_clean_gtt(struct intel_vgpu *vgpu) +{ + struct list_head *pos, *n; + struct intel_vgpu_mm *mm; + + ppgtt_free_all_shadow_page(vgpu); + release_scratch_page_tree(vgpu); + + list_for_each_safe(pos, n, &vgpu->gtt.mm_list_head) { + mm = container_of(pos, struct intel_vgpu_mm, list); + vgpu->gvt->gtt.mm_free_page_table(mm); + list_del(&mm->list); + list_del(&mm->lru_list); + kfree(mm); + } +} + +static void clean_spt_oos(struct intel_gvt *gvt) +{ + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct list_head *pos, *n; + struct intel_vgpu_oos_page *oos_page; + + WARN(!list_empty(>t->oos_page_use_list_head), + "someone is still using oos page\n"); + + list_for_each_safe(pos, n, >t->oos_page_free_list_head) { + oos_page = container_of(pos, struct intel_vgpu_oos_page, list); + list_del(&oos_page->list); + kfree(oos_page); + } +} + +static int setup_spt_oos(struct intel_gvt *gvt) +{ + struct intel_gvt_gtt *gtt = &gvt->gtt; + struct intel_vgpu_oos_page *oos_page; + int i; + int ret; + + INIT_LIST_HEAD(>t->oos_page_free_list_head); + INIT_LIST_HEAD(>t->oos_page_use_list_head); + + for (i = 0; i < preallocated_oos_pages; i++) { + oos_page = kzalloc(sizeof(*oos_page), GFP_KERNEL); + if (!oos_page) { + gvt_err("fail to pre-allocate oos page\n"); + ret = -ENOMEM; + goto fail; + } + + INIT_LIST_HEAD(&oos_page->list); + INIT_LIST_HEAD(&oos_page->vm_list); + oos_page->id = i; + list_add_tail(&oos_page->list, >t->oos_page_free_list_head); + } + + gvt_dbg_mm("%d oos pages preallocated\n", i); + + return 0; +fail: + clean_spt_oos(gvt); + return ret; +} + +/** + * intel_vgpu_find_ppgtt_mm - find a PPGTT mm object + * @vgpu: a vGPU + * @page_table_level: PPGTT page table level + * @root_entry: PPGTT page table root pointers + * + * This function is used to find a PPGTT mm object from mm object pool + * + * Returns: + * pointer to mm object on success, NULL if failed. + */ +struct intel_vgpu_mm *intel_vgpu_find_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level, void *root_entry) +{ + struct list_head *pos; + struct intel_vgpu_mm *mm; + u64 *src, *dst; + + list_for_each(pos, &vgpu->gtt.mm_list_head) { + mm = container_of(pos, struct intel_vgpu_mm, list); + if (mm->type != INTEL_GVT_MM_PPGTT) + continue; + + if (mm->page_table_level != page_table_level) + continue; + + src = root_entry; + dst = mm->virtual_page_table; + + if (page_table_level == 3) { + if (src[0] == dst[0] + && src[1] == dst[1] + && src[2] == dst[2] + && src[3] == dst[3]) + return mm; + } else { + if (src[0] == dst[0]) + return mm; + } + } + return NULL; +} + +/** + * intel_vgpu_g2v_create_ppgtt_mm - create a PPGTT mm object from + * g2v notification + * @vgpu: a vGPU + * @page_table_level: PPGTT page table level + * + * This function is used to create a PPGTT mm object from a guest to GVT-g + * notification. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_g2v_create_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level) +{ + u64 *pdp = (u64 *)&vgpu_vreg64(vgpu, vgtif_reg(pdp[0])); + struct intel_vgpu_mm *mm; + + if (WARN_ON((page_table_level != 4) && (page_table_level != 3))) + return -EINVAL; + + mm = intel_vgpu_find_ppgtt_mm(vgpu, page_table_level, pdp); + if (mm) { + intel_gvt_mm_reference(mm); + } else { + mm = intel_vgpu_create_mm(vgpu, INTEL_GVT_MM_PPGTT, + pdp, page_table_level, 0); + if (IS_ERR(mm)) { + gvt_err("fail to create mm\n"); + return PTR_ERR(mm); + } + } + return 0; +} + +/** + * intel_vgpu_g2v_destroy_ppgtt_mm - destroy a PPGTT mm object from + * g2v notification + * @vgpu: a vGPU + * @page_table_level: PPGTT page table level + * + * This function is used to create a PPGTT mm object from a guest to GVT-g + * notification. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_g2v_destroy_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level) +{ + u64 *pdp = (u64 *)&vgpu_vreg64(vgpu, vgtif_reg(pdp[0])); + struct intel_vgpu_mm *mm; + + if (WARN_ON((page_table_level != 4) && (page_table_level != 3))) + return -EINVAL; + + mm = intel_vgpu_find_ppgtt_mm(vgpu, page_table_level, pdp); + if (!mm) { + gvt_err("fail to find ppgtt instance.\n"); + return -EINVAL; + } + intel_gvt_mm_unreference(mm); + return 0; +} + +/** + * intel_gvt_init_gtt - initialize mm components of a GVT device + * @gvt: GVT device + * + * This function is called at the initialization stage, to initialize + * the mm components of a GVT device. + * + * Returns: + * zero on success, negative error code if failed. + */ +int intel_gvt_init_gtt(struct intel_gvt *gvt) +{ + int ret; + + gvt_dbg_core("init gtt\n"); + + if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) { + gvt->gtt.pte_ops = &gen8_gtt_pte_ops; + gvt->gtt.gma_ops = &gen8_gtt_gma_ops; + gvt->gtt.mm_alloc_page_table = gen8_mm_alloc_page_table; + gvt->gtt.mm_free_page_table = gen8_mm_free_page_table; + } else { + return -ENODEV; + } + + if (enable_out_of_sync) { + ret = setup_spt_oos(gvt); + if (ret) { + gvt_err("fail to initialize SPT oos\n"); + return ret; + } + } + INIT_LIST_HEAD(&gvt->gtt.mm_lru_list_head); + return 0; +} + +/** + * intel_gvt_clean_gtt - clean up mm components of a GVT device + * @gvt: GVT device + * + * This function is called at the driver unloading stage, to clean up the + * the mm components of a GVT device. + * + */ +void intel_gvt_clean_gtt(struct intel_gvt *gvt) +{ + if (enable_out_of_sync) + clean_spt_oos(gvt); +} diff --git a/drivers/gpu/drm/i915/gvt/gtt.h b/drivers/gpu/drm/i915/gvt/gtt.h new file mode 100644 index 000000000000..d250013bc37b --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/gtt.h @@ -0,0 +1,306 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhi Wang <zhi.a.wang@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * Xiao Zheng <xiao.zheng@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#ifndef _GVT_GTT_H_ +#define _GVT_GTT_H_ + +#define GTT_PAGE_SHIFT 12 +#define GTT_PAGE_SIZE (1UL << GTT_PAGE_SHIFT) +#define GTT_PAGE_MASK (~(GTT_PAGE_SIZE-1)) + +struct intel_vgpu_mm; + +#define INTEL_GVT_GTT_HASH_BITS 8 +#define INTEL_GVT_INVALID_ADDR (~0UL) + +struct intel_gvt_gtt_entry { + u64 val64; + int type; +}; + +struct intel_gvt_gtt_pte_ops { + struct intel_gvt_gtt_entry *(*get_entry)(void *pt, + struct intel_gvt_gtt_entry *e, + unsigned long index, bool hypervisor_access, unsigned long gpa, + struct intel_vgpu *vgpu); + struct intel_gvt_gtt_entry *(*set_entry)(void *pt, + struct intel_gvt_gtt_entry *e, + unsigned long index, bool hypervisor_access, unsigned long gpa, + struct intel_vgpu *vgpu); + bool (*test_present)(struct intel_gvt_gtt_entry *e); + void (*clear_present)(struct intel_gvt_gtt_entry *e); + bool (*test_pse)(struct intel_gvt_gtt_entry *e); + void (*set_pfn)(struct intel_gvt_gtt_entry *e, unsigned long pfn); + unsigned long (*get_pfn)(struct intel_gvt_gtt_entry *e); +}; + +struct intel_gvt_gtt_gma_ops { + unsigned long (*gma_to_ggtt_pte_index)(unsigned long gma); + unsigned long (*gma_to_pte_index)(unsigned long gma); + unsigned long (*gma_to_pde_index)(unsigned long gma); + unsigned long (*gma_to_l3_pdp_index)(unsigned long gma); + unsigned long (*gma_to_l4_pdp_index)(unsigned long gma); + unsigned long (*gma_to_pml4_index)(unsigned long gma); +}; + +struct intel_gvt_gtt { + struct intel_gvt_gtt_pte_ops *pte_ops; + struct intel_gvt_gtt_gma_ops *gma_ops; + int (*mm_alloc_page_table)(struct intel_vgpu_mm *mm); + void (*mm_free_page_table)(struct intel_vgpu_mm *mm); + struct list_head oos_page_use_list_head; + struct list_head oos_page_free_list_head; + struct list_head mm_lru_list_head; +}; + +enum { + INTEL_GVT_MM_GGTT = 0, + INTEL_GVT_MM_PPGTT, +}; + +typedef enum { + GTT_TYPE_INVALID = -1, + + GTT_TYPE_GGTT_PTE, + + GTT_TYPE_PPGTT_PTE_4K_ENTRY, + GTT_TYPE_PPGTT_PTE_2M_ENTRY, + GTT_TYPE_PPGTT_PTE_1G_ENTRY, + + GTT_TYPE_PPGTT_PTE_ENTRY, + + GTT_TYPE_PPGTT_PDE_ENTRY, + GTT_TYPE_PPGTT_PDP_ENTRY, + GTT_TYPE_PPGTT_PML4_ENTRY, + + GTT_TYPE_PPGTT_ROOT_ENTRY, + + GTT_TYPE_PPGTT_ROOT_L3_ENTRY, + GTT_TYPE_PPGTT_ROOT_L4_ENTRY, + + GTT_TYPE_PPGTT_ENTRY, + + GTT_TYPE_PPGTT_PTE_PT, + GTT_TYPE_PPGTT_PDE_PT, + GTT_TYPE_PPGTT_PDP_PT, + GTT_TYPE_PPGTT_PML4_PT, + + GTT_TYPE_MAX, +} intel_gvt_gtt_type_t; + +struct intel_vgpu_mm { + int type; + bool initialized; + bool shadowed; + + int page_table_entry_type; + u32 page_table_entry_size; + u32 page_table_entry_cnt; + void *virtual_page_table; + void *shadow_page_table; + + int page_table_level; + bool has_shadow_page_table; + u32 pde_base_index; + + struct list_head list; + struct kref ref; + atomic_t pincount; + struct list_head lru_list; + struct intel_vgpu *vgpu; +}; + +extern struct intel_gvt_gtt_entry *intel_vgpu_mm_get_entry( + struct intel_vgpu_mm *mm, + void *page_table, struct intel_gvt_gtt_entry *e, + unsigned long index); + +extern struct intel_gvt_gtt_entry *intel_vgpu_mm_set_entry( + struct intel_vgpu_mm *mm, + void *page_table, struct intel_gvt_gtt_entry *e, + unsigned long index); + +#define ggtt_get_guest_entry(mm, e, index) \ + intel_vgpu_mm_get_entry(mm, mm->virtual_page_table, e, index) + +#define ggtt_set_guest_entry(mm, e, index) \ + intel_vgpu_mm_set_entry(mm, mm->virtual_page_table, e, index) + +#define ggtt_get_shadow_entry(mm, e, index) \ + intel_vgpu_mm_get_entry(mm, mm->shadow_page_table, e, index) + +#define ggtt_set_shadow_entry(mm, e, index) \ + intel_vgpu_mm_set_entry(mm, mm->shadow_page_table, e, index) + +#define ppgtt_get_guest_root_entry(mm, e, index) \ + intel_vgpu_mm_get_entry(mm, mm->virtual_page_table, e, index) + +#define ppgtt_set_guest_root_entry(mm, e, index) \ + intel_vgpu_mm_set_entry(mm, mm->virtual_page_table, e, index) + +#define ppgtt_get_shadow_root_entry(mm, e, index) \ + intel_vgpu_mm_get_entry(mm, mm->shadow_page_table, e, index) + +#define ppgtt_set_shadow_root_entry(mm, e, index) \ + intel_vgpu_mm_set_entry(mm, mm->shadow_page_table, e, index) + +extern struct intel_vgpu_mm *intel_vgpu_create_mm(struct intel_vgpu *vgpu, + int mm_type, void *virtual_page_table, int page_table_level, + u32 pde_base_index); +extern void intel_vgpu_destroy_mm(struct kref *mm_ref); + +struct intel_vgpu_guest_page; + +struct intel_vgpu_scratch_pt { + struct page *page; + unsigned long page_mfn; +}; + + +struct intel_vgpu_gtt { + struct intel_vgpu_mm *ggtt_mm; + unsigned long active_ppgtt_mm_bitmap; + struct list_head mm_list_head; + DECLARE_HASHTABLE(shadow_page_hash_table, INTEL_GVT_GTT_HASH_BITS); + DECLARE_HASHTABLE(guest_page_hash_table, INTEL_GVT_GTT_HASH_BITS); + atomic_t n_write_protected_guest_page; + struct list_head oos_page_list_head; + struct list_head post_shadow_list_head; + struct intel_vgpu_scratch_pt scratch_pt[GTT_TYPE_MAX]; + +}; + +extern int intel_vgpu_init_gtt(struct intel_vgpu *vgpu); +extern void intel_vgpu_clean_gtt(struct intel_vgpu *vgpu); + +extern int intel_gvt_init_gtt(struct intel_gvt *gvt); +extern void intel_gvt_clean_gtt(struct intel_gvt *gvt); + +extern struct intel_vgpu_mm *intel_gvt_find_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level, void *root_entry); + +struct intel_vgpu_oos_page; + +struct intel_vgpu_shadow_page { + void *vaddr; + struct page *page; + int type; + struct hlist_node node; + unsigned long mfn; +}; + +struct intel_vgpu_guest_page { + struct hlist_node node; + bool writeprotection; + unsigned long gfn; + int (*handler)(void *, u64, void *, int); + void *data; + unsigned long write_cnt; + struct intel_vgpu_oos_page *oos_page; +}; + +struct intel_vgpu_oos_page { + struct intel_vgpu_guest_page *guest_page; + struct list_head list; + struct list_head vm_list; + int id; + unsigned char mem[GTT_PAGE_SIZE]; +}; + +#define GTT_ENTRY_NUM_IN_ONE_PAGE 512 + +struct intel_vgpu_ppgtt_spt { + struct intel_vgpu_shadow_page shadow_page; + struct intel_vgpu_guest_page guest_page; + int guest_page_type; + atomic_t refcount; + struct intel_vgpu *vgpu; + DECLARE_BITMAP(post_shadow_bitmap, GTT_ENTRY_NUM_IN_ONE_PAGE); + struct list_head post_shadow_list; +}; + +int intel_vgpu_init_guest_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *guest_page, + unsigned long gfn, + int (*handler)(void *gp, u64, void *, int), + void *data); + +void intel_vgpu_clean_guest_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *guest_page); + +int intel_vgpu_set_guest_page_writeprotection(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *guest_page); + +void intel_vgpu_clear_guest_page_writeprotection(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *guest_page); + +struct intel_vgpu_guest_page *intel_vgpu_find_guest_page( + struct intel_vgpu *vgpu, unsigned long gfn); + +int intel_vgpu_sync_oos_pages(struct intel_vgpu *vgpu); + +int intel_vgpu_flush_post_shadow(struct intel_vgpu *vgpu); + +static inline void intel_gvt_mm_reference(struct intel_vgpu_mm *mm) +{ + kref_get(&mm->ref); +} + +static inline void intel_gvt_mm_unreference(struct intel_vgpu_mm *mm) +{ + kref_put(&mm->ref, intel_vgpu_destroy_mm); +} + +int intel_vgpu_pin_mm(struct intel_vgpu_mm *mm); + +void intel_vgpu_unpin_mm(struct intel_vgpu_mm *mm); + +unsigned long intel_vgpu_gma_to_gpa(struct intel_vgpu_mm *mm, + unsigned long gma); + +struct intel_vgpu_mm *intel_vgpu_find_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level, void *root_entry); + +int intel_vgpu_g2v_create_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level); + +int intel_vgpu_g2v_destroy_ppgtt_mm(struct intel_vgpu *vgpu, + int page_table_level); + +int intel_vgpu_emulate_gtt_mmio_read(struct intel_vgpu *vgpu, + unsigned int off, void *p_data, unsigned int bytes); + +int intel_vgpu_emulate_gtt_mmio_write(struct intel_vgpu *vgpu, + unsigned int off, void *p_data, unsigned int bytes); + +#endif /* _GVT_GTT_H_ */ diff --git a/drivers/gpu/drm/i915/gvt/gvt.c b/drivers/gpu/drm/i915/gvt/gvt.c index 927f4579f5b6..398877c3d2fd 100644 --- a/drivers/gpu/drm/i915/gvt/gvt.c +++ b/drivers/gpu/drm/i915/gvt/gvt.c @@ -19,12 +19,23 @@ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Eddie Dong <eddie.dong@intel.com> + * + * Contributors: + * Niu Bing <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * */ #include <linux/types.h> #include <xen/xen.h> +#include <linux/kthread.h> #include "i915_drv.h" +#include "gvt.h" struct intel_gvt_host intel_gvt_host; @@ -33,6 +44,16 @@ static const char * const supported_hypervisors[] = { [INTEL_GVT_HYPERVISOR_KVM] = "KVM", }; +static const struct intel_gvt_ops intel_gvt_ops = { + .emulate_cfg_read = intel_vgpu_emulate_cfg_read, + .emulate_cfg_write = intel_vgpu_emulate_cfg_write, + .emulate_mmio_read = intel_vgpu_emulate_mmio_read, + .emulate_mmio_write = intel_vgpu_emulate_mmio_write, + .vgpu_create = intel_gvt_create_vgpu, + .vgpu_destroy = intel_gvt_destroy_vgpu, + .vgpu_reset = intel_gvt_reset_vgpu, +}; + /** * intel_gvt_init_host - Load MPT modules and detect if we're running in host * @gvt: intel gvt device @@ -47,6 +68,8 @@ static const char * const supported_hypervisors[] = { */ int intel_gvt_init_host(void) { + int ret; + if (intel_gvt_host.initialized) return 0; @@ -61,10 +84,12 @@ int intel_gvt_init_host(void) symbol_get(xengt_mpt), "xengt"); intel_gvt_host.hypervisor_type = INTEL_GVT_HYPERVISOR_XEN; } else { +#if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT) /* not in Xen. Try KVMGT */ intel_gvt_host.mpt = try_then_request_module( - symbol_get(kvmgt_mpt), "kvm"); + symbol_get(kvmgt_mpt), "kvmgt"); intel_gvt_host.hypervisor_type = INTEL_GVT_HYPERVISOR_KVM; +#endif } /* Fail to load MPT modules - bail out */ @@ -72,7 +97,8 @@ int intel_gvt_init_host(void) return -EINVAL; /* Try to detect if we're running in host instead of VM. */ - if (!intel_gvt_hypervisor_detect_host()) + ret = intel_gvt_hypervisor_detect_host(); + if (ret) return -ENODEV; gvt_dbg_core("Running with hypervisor %s in host mode\n", @@ -84,9 +110,67 @@ int intel_gvt_init_host(void) static void init_device_info(struct intel_gvt *gvt) { - if (IS_BROADWELL(gvt->dev_priv)) - gvt->device_info.max_support_vgpus = 8; - /* This function will grow large in GVT device model patches. */ + struct intel_gvt_device_info *info = &gvt->device_info; + struct pci_dev *pdev = gvt->dev_priv->drm.pdev; + + if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) { + info->max_support_vgpus = 8; + info->cfg_space_size = 256; + info->mmio_size = 2 * 1024 * 1024; + info->mmio_bar = 0; + info->gtt_start_offset = 8 * 1024 * 1024; + info->gtt_entry_size = 8; + info->gtt_entry_size_shift = 3; + info->gmadr_bytes_in_cmd = 8; + info->max_surface_size = 36 * 1024 * 1024; + } + info->msi_cap_offset = pdev->msi_cap; +} + +static int gvt_service_thread(void *data) +{ + struct intel_gvt *gvt = (struct intel_gvt *)data; + int ret; + + gvt_dbg_core("service thread start\n"); + + while (!kthread_should_stop()) { + ret = wait_event_interruptible(gvt->service_thread_wq, + kthread_should_stop() || gvt->service_request); + + if (kthread_should_stop()) + break; + + if (WARN_ONCE(ret, "service thread is waken up by signal.\n")) + continue; + + if (test_and_clear_bit(INTEL_GVT_REQUEST_EMULATE_VBLANK, + (void *)&gvt->service_request)) { + mutex_lock(&gvt->lock); + intel_gvt_emulate_vblank(gvt); + mutex_unlock(&gvt->lock); + } + } + + return 0; +} + +static void clean_service_thread(struct intel_gvt *gvt) +{ + kthread_stop(gvt->service_thread); +} + +static int init_service_thread(struct intel_gvt *gvt) +{ + init_waitqueue_head(&gvt->service_thread_wq); + + gvt->service_thread = kthread_run(gvt_service_thread, + gvt, "gvt_service_thread"); + if (IS_ERR(gvt->service_thread)) { + gvt_err("fail to start service thread.\n"); + return PTR_ERR(gvt->service_thread); + } + return 0; } /** @@ -99,14 +183,26 @@ static void init_device_info(struct intel_gvt *gvt) */ void intel_gvt_clean_device(struct drm_i915_private *dev_priv) { - struct intel_gvt *gvt = &dev_priv->gvt; + struct intel_gvt *gvt = to_gvt(dev_priv); - if (WARN_ON(!gvt->initialized)) + if (WARN_ON(!gvt)) return; - /* Other de-initialization of GVT components will be introduced. */ + clean_service_thread(gvt); + intel_gvt_clean_cmd_parser(gvt); + intel_gvt_clean_sched_policy(gvt); + intel_gvt_clean_workload_scheduler(gvt); + intel_gvt_clean_opregion(gvt); + intel_gvt_clean_gtt(gvt); + intel_gvt_clean_irq(gvt); + intel_gvt_clean_mmio_info(gvt); + intel_gvt_free_firmware(gvt); + + intel_gvt_hypervisor_host_exit(&dev_priv->drm.pdev->dev, gvt); + intel_gvt_clean_vgpu_types(gvt); - gvt->initialized = false; + kfree(dev_priv->gvt); + dev_priv->gvt = NULL; } /** @@ -122,7 +218,9 @@ void intel_gvt_clean_device(struct drm_i915_private *dev_priv) */ int intel_gvt_init_device(struct drm_i915_private *dev_priv) { - struct intel_gvt *gvt = &dev_priv->gvt; + struct intel_gvt *gvt; + int ret; + /* * Cannot initialize GVT device without intel_gvt_host gets * initialized first. @@ -130,16 +228,91 @@ int intel_gvt_init_device(struct drm_i915_private *dev_priv) if (WARN_ON(!intel_gvt_host.initialized)) return -EINVAL; - if (WARN_ON(gvt->initialized)) + if (WARN_ON(dev_priv->gvt)) return -EEXIST; + gvt = kzalloc(sizeof(struct intel_gvt), GFP_KERNEL); + if (!gvt) + return -ENOMEM; + gvt_dbg_core("init gvt device\n"); + mutex_init(&gvt->lock); + gvt->dev_priv = dev_priv; + init_device_info(gvt); - /* - * Other initialization of GVT components will be introduce here. - */ - gvt_dbg_core("gvt device creation is done\n"); - gvt->initialized = true; + + ret = intel_gvt_setup_mmio_info(gvt); + if (ret) + return ret; + + ret = intel_gvt_load_firmware(gvt); + if (ret) + goto out_clean_mmio_info; + + ret = intel_gvt_init_irq(gvt); + if (ret) + goto out_free_firmware; + + ret = intel_gvt_init_gtt(gvt); + if (ret) + goto out_clean_irq; + + ret = intel_gvt_init_opregion(gvt); + if (ret) + goto out_clean_gtt; + + ret = intel_gvt_init_workload_scheduler(gvt); + if (ret) + goto out_clean_opregion; + + ret = intel_gvt_init_sched_policy(gvt); + if (ret) + goto out_clean_workload_scheduler; + + ret = intel_gvt_init_cmd_parser(gvt); + if (ret) + goto out_clean_sched_policy; + + ret = init_service_thread(gvt); + if (ret) + goto out_clean_cmd_parser; + + ret = intel_gvt_init_vgpu_types(gvt); + if (ret) + goto out_clean_thread; + + ret = intel_gvt_hypervisor_host_init(&dev_priv->drm.pdev->dev, gvt, + &intel_gvt_ops); + if (ret) { + gvt_err("failed to register gvt-g host device: %d\n", ret); + goto out_clean_types; + } + + gvt_dbg_core("gvt device initialization is done\n"); + dev_priv->gvt = gvt; return 0; + +out_clean_types: + intel_gvt_clean_vgpu_types(gvt); +out_clean_thread: + clean_service_thread(gvt); +out_clean_cmd_parser: + intel_gvt_clean_cmd_parser(gvt); +out_clean_sched_policy: + intel_gvt_clean_sched_policy(gvt); +out_clean_workload_scheduler: + intel_gvt_clean_workload_scheduler(gvt); +out_clean_opregion: + intel_gvt_clean_opregion(gvt); +out_clean_gtt: + intel_gvt_clean_gtt(gvt); +out_clean_irq: + intel_gvt_clean_irq(gvt); +out_free_firmware: + intel_gvt_free_firmware(gvt); +out_clean_mmio_info: + intel_gvt_clean_mmio_info(gvt); + kfree(gvt); + return ret; } diff --git a/drivers/gpu/drm/i915/gvt/gvt.h b/drivers/gpu/drm/i915/gvt/gvt.h index fb619a6e519d..b1a7c8dd4b5f 100644 --- a/drivers/gpu/drm/i915/gvt/gvt.h +++ b/drivers/gpu/drm/i915/gvt/gvt.h @@ -19,6 +19,15 @@ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Eddie Dong <eddie.dong@intel.com> + * + * Contributors: + * Niu Bing <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * */ #ifndef _GVT_H_ @@ -26,6 +35,17 @@ #include "debug.h" #include "hypercall.h" +#include "mmio.h" +#include "reg.h" +#include "interrupt.h" +#include "gtt.h" +#include "display.h" +#include "edid.h" +#include "execlist.h" +#include "scheduler.h" +#include "sched_policy.h" +#include "render.h" +#include "cmd_parser.h" #define GVT_MAX_VGPU 8 @@ -45,25 +65,381 @@ extern struct intel_gvt_host intel_gvt_host; /* Describe per-platform limitations. */ struct intel_gvt_device_info { u32 max_support_vgpus; - /* This data structure will grow bigger in GVT device model patches */ + u32 cfg_space_size; + u32 mmio_size; + u32 mmio_bar; + unsigned long msi_cap_offset; + u32 gtt_start_offset; + u32 gtt_entry_size; + u32 gtt_entry_size_shift; + int gmadr_bytes_in_cmd; + u32 max_surface_size; +}; + +/* GM resources owned by a vGPU */ +struct intel_vgpu_gm { + u64 aperture_sz; + u64 hidden_sz; + struct drm_mm_node low_gm_node; + struct drm_mm_node high_gm_node; +}; + +#define INTEL_GVT_MAX_NUM_FENCES 32 + +/* Fences owned by a vGPU */ +struct intel_vgpu_fence { + struct drm_i915_fence_reg *regs[INTEL_GVT_MAX_NUM_FENCES]; + u32 base; + u32 size; +}; + +struct intel_vgpu_mmio { + void *vreg; + void *sreg; + bool disable_warn_untrack; +}; + +#define INTEL_GVT_MAX_CFG_SPACE_SZ 256 +#define INTEL_GVT_MAX_BAR_NUM 4 + +struct intel_vgpu_pci_bar { + u64 size; + bool tracked; +}; + +struct intel_vgpu_cfg_space { + unsigned char virtual_cfg_space[INTEL_GVT_MAX_CFG_SPACE_SZ]; + struct intel_vgpu_pci_bar bar[INTEL_GVT_MAX_BAR_NUM]; +}; + +#define vgpu_cfg_space(vgpu) ((vgpu)->cfg_space.virtual_cfg_space) + +#define INTEL_GVT_MAX_PIPE 4 + +struct intel_vgpu_irq { + bool irq_warn_once[INTEL_GVT_EVENT_MAX]; + DECLARE_BITMAP(flip_done_event[INTEL_GVT_MAX_PIPE], + INTEL_GVT_EVENT_MAX); +}; + +struct intel_vgpu_opregion { + void *va; + u32 gfn[INTEL_GVT_OPREGION_PAGES]; + struct page *pages[INTEL_GVT_OPREGION_PAGES]; +}; + +#define vgpu_opregion(vgpu) (&(vgpu->opregion)) + +#define INTEL_GVT_MAX_PORT 5 + +struct intel_vgpu_display { + struct intel_vgpu_i2c_edid i2c_edid; + struct intel_vgpu_port ports[INTEL_GVT_MAX_PORT]; + struct intel_vgpu_sbi sbi; }; struct intel_vgpu { struct intel_gvt *gvt; int id; unsigned long handle; /* vGPU handle used by hypervisor MPT modules */ + bool active; + bool resetting; + void *sched_data; + + struct intel_vgpu_fence fence; + struct intel_vgpu_gm gm; + struct intel_vgpu_cfg_space cfg_space; + struct intel_vgpu_mmio mmio; + struct intel_vgpu_irq irq; + struct intel_vgpu_gtt gtt; + struct intel_vgpu_opregion opregion; + struct intel_vgpu_display display; + struct intel_vgpu_execlist execlist[I915_NUM_ENGINES]; + struct list_head workload_q_head[I915_NUM_ENGINES]; + struct kmem_cache *workloads; + atomic_t running_workload_num; + DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES); + struct i915_gem_context *shadow_ctx; + struct notifier_block shadow_ctx_notifier_block; + +#if IS_ENABLED(CONFIG_DRM_I915_GVT_KVMGT) + struct { + struct device *mdev; + struct vfio_region *region; + int num_regions; + struct eventfd_ctx *intx_trigger; + struct eventfd_ctx *msi_trigger; + struct rb_root cache; + struct mutex cache_lock; + void *vfio_group; + struct notifier_block iommu_notifier; + } vdev; +#endif +}; + +struct intel_gvt_gm { + unsigned long vgpu_allocated_low_gm_size; + unsigned long vgpu_allocated_high_gm_size; +}; + +struct intel_gvt_fence { + unsigned long vgpu_allocated_fence_num; +}; + +#define INTEL_GVT_MMIO_HASH_BITS 9 + +struct intel_gvt_mmio { + u32 *mmio_attribute; + DECLARE_HASHTABLE(mmio_info_table, INTEL_GVT_MMIO_HASH_BITS); +}; + +struct intel_gvt_firmware { + void *cfg_space; + void *mmio; + bool firmware_loaded; +}; + +struct intel_gvt_opregion { + void __iomem *opregion_va; + u32 opregion_pa; +}; + +#define NR_MAX_INTEL_VGPU_TYPES 20 +struct intel_vgpu_type { + char name[16]; + unsigned int max_instance; + unsigned int avail_instance; + unsigned int low_gm_size; + unsigned int high_gm_size; + unsigned int fence; }; struct intel_gvt { struct mutex lock; - bool initialized; - struct drm_i915_private *dev_priv; struct idr vgpu_idr; /* vGPU IDR pool */ struct intel_gvt_device_info device_info; + struct intel_gvt_gm gm; + struct intel_gvt_fence fence; + struct intel_gvt_mmio mmio; + struct intel_gvt_firmware firmware; + struct intel_gvt_irq irq; + struct intel_gvt_gtt gtt; + struct intel_gvt_opregion opregion; + struct intel_gvt_workload_scheduler scheduler; + DECLARE_HASHTABLE(cmd_table, GVT_CMD_HASH_BITS); + struct intel_vgpu_type *types; + unsigned int num_types; + + struct task_struct *service_thread; + wait_queue_head_t service_thread_wq; + unsigned long service_request; +}; + +static inline struct intel_gvt *to_gvt(struct drm_i915_private *i915) +{ + return i915->gvt; +} + +enum { + INTEL_GVT_REQUEST_EMULATE_VBLANK = 0, }; +static inline void intel_gvt_request_service(struct intel_gvt *gvt, + int service) +{ + set_bit(service, (void *)&gvt->service_request); + wake_up(&gvt->service_thread_wq); +} + +void intel_gvt_free_firmware(struct intel_gvt *gvt); +int intel_gvt_load_firmware(struct intel_gvt *gvt); + +/* Aperture/GM space definitions for GVT device */ +#define MB_TO_BYTES(mb) ((mb) << 20ULL) +#define BYTES_TO_MB(b) ((b) >> 20ULL) + +#define HOST_LOW_GM_SIZE MB_TO_BYTES(128) +#define HOST_HIGH_GM_SIZE MB_TO_BYTES(384) +#define HOST_FENCE 4 + +/* Aperture/GM space definitions for GVT device */ +#define gvt_aperture_sz(gvt) (gvt->dev_priv->ggtt.mappable_end) +#define gvt_aperture_pa_base(gvt) (gvt->dev_priv->ggtt.mappable_base) + +#define gvt_ggtt_gm_sz(gvt) (gvt->dev_priv->ggtt.base.total) +#define gvt_ggtt_sz(gvt) \ + ((gvt->dev_priv->ggtt.base.total >> PAGE_SHIFT) << 3) +#define gvt_hidden_sz(gvt) (gvt_ggtt_gm_sz(gvt) - gvt_aperture_sz(gvt)) + +#define gvt_aperture_gmadr_base(gvt) (0) +#define gvt_aperture_gmadr_end(gvt) (gvt_aperture_gmadr_base(gvt) \ + + gvt_aperture_sz(gvt) - 1) + +#define gvt_hidden_gmadr_base(gvt) (gvt_aperture_gmadr_base(gvt) \ + + gvt_aperture_sz(gvt)) +#define gvt_hidden_gmadr_end(gvt) (gvt_hidden_gmadr_base(gvt) \ + + gvt_hidden_sz(gvt) - 1) + +#define gvt_fence_sz(gvt) (gvt->dev_priv->num_fence_regs) + +/* Aperture/GM space definitions for vGPU */ +#define vgpu_aperture_offset(vgpu) ((vgpu)->gm.low_gm_node.start) +#define vgpu_hidden_offset(vgpu) ((vgpu)->gm.high_gm_node.start) +#define vgpu_aperture_sz(vgpu) ((vgpu)->gm.aperture_sz) +#define vgpu_hidden_sz(vgpu) ((vgpu)->gm.hidden_sz) + +#define vgpu_aperture_pa_base(vgpu) \ + (gvt_aperture_pa_base(vgpu->gvt) + vgpu_aperture_offset(vgpu)) + +#define vgpu_ggtt_gm_sz(vgpu) ((vgpu)->gm.aperture_sz + (vgpu)->gm.hidden_sz) + +#define vgpu_aperture_pa_end(vgpu) \ + (vgpu_aperture_pa_base(vgpu) + vgpu_aperture_sz(vgpu) - 1) + +#define vgpu_aperture_gmadr_base(vgpu) (vgpu_aperture_offset(vgpu)) +#define vgpu_aperture_gmadr_end(vgpu) \ + (vgpu_aperture_gmadr_base(vgpu) + vgpu_aperture_sz(vgpu) - 1) + +#define vgpu_hidden_gmadr_base(vgpu) (vgpu_hidden_offset(vgpu)) +#define vgpu_hidden_gmadr_end(vgpu) \ + (vgpu_hidden_gmadr_base(vgpu) + vgpu_hidden_sz(vgpu) - 1) + +#define vgpu_fence_base(vgpu) (vgpu->fence.base) +#define vgpu_fence_sz(vgpu) (vgpu->fence.size) + +struct intel_vgpu_creation_params { + __u64 handle; + __u64 low_gm_sz; /* in MB */ + __u64 high_gm_sz; /* in MB */ + __u64 fence_sz; + __s32 primary; + __u64 vgpu_id; +}; + +int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu, + struct intel_vgpu_creation_params *param); +void intel_vgpu_free_resource(struct intel_vgpu *vgpu); +void intel_vgpu_write_fence(struct intel_vgpu *vgpu, + u32 fence, u64 value); + +/* Macros for easily accessing vGPU virtual/shadow register */ +#define vgpu_vreg(vgpu, reg) \ + (*(u32 *)(vgpu->mmio.vreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_vreg8(vgpu, reg) \ + (*(u8 *)(vgpu->mmio.vreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_vreg16(vgpu, reg) \ + (*(u16 *)(vgpu->mmio.vreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_vreg64(vgpu, reg) \ + (*(u64 *)(vgpu->mmio.vreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_sreg(vgpu, reg) \ + (*(u32 *)(vgpu->mmio.sreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_sreg8(vgpu, reg) \ + (*(u8 *)(vgpu->mmio.sreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_sreg16(vgpu, reg) \ + (*(u16 *)(vgpu->mmio.sreg + INTEL_GVT_MMIO_OFFSET(reg))) +#define vgpu_sreg64(vgpu, reg) \ + (*(u64 *)(vgpu->mmio.sreg + INTEL_GVT_MMIO_OFFSET(reg))) + +#define for_each_active_vgpu(gvt, vgpu, id) \ + idr_for_each_entry((&(gvt)->vgpu_idr), (vgpu), (id)) \ + for_each_if(vgpu->active) + +static inline void intel_vgpu_write_pci_bar(struct intel_vgpu *vgpu, + u32 offset, u32 val, bool low) +{ + u32 *pval; + + /* BAR offset should be 32 bits algiend */ + offset = rounddown(offset, 4); + pval = (u32 *)(vgpu_cfg_space(vgpu) + offset); + + if (low) { + /* + * only update bit 31 - bit 4, + * leave the bit 3 - bit 0 unchanged. + */ + *pval = (val & GENMASK(31, 4)) | (*pval & GENMASK(3, 0)); + } else { + *pval = val; + } +} + +int intel_gvt_init_vgpu_types(struct intel_gvt *gvt); +void intel_gvt_clean_vgpu_types(struct intel_gvt *gvt); + +struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt, + struct intel_vgpu_type *type); +void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu); +void intel_gvt_reset_vgpu(struct intel_vgpu *vgpu); + + +/* validating GM functions */ +#define vgpu_gmadr_is_aperture(vgpu, gmadr) \ + ((gmadr >= vgpu_aperture_gmadr_base(vgpu)) && \ + (gmadr <= vgpu_aperture_gmadr_end(vgpu))) + +#define vgpu_gmadr_is_hidden(vgpu, gmadr) \ + ((gmadr >= vgpu_hidden_gmadr_base(vgpu)) && \ + (gmadr <= vgpu_hidden_gmadr_end(vgpu))) + +#define vgpu_gmadr_is_valid(vgpu, gmadr) \ + ((vgpu_gmadr_is_aperture(vgpu, gmadr) || \ + (vgpu_gmadr_is_hidden(vgpu, gmadr)))) + +#define gvt_gmadr_is_aperture(gvt, gmadr) \ + ((gmadr >= gvt_aperture_gmadr_base(gvt)) && \ + (gmadr <= gvt_aperture_gmadr_end(gvt))) + +#define gvt_gmadr_is_hidden(gvt, gmadr) \ + ((gmadr >= gvt_hidden_gmadr_base(gvt)) && \ + (gmadr <= gvt_hidden_gmadr_end(gvt))) + +#define gvt_gmadr_is_valid(gvt, gmadr) \ + (gvt_gmadr_is_aperture(gvt, gmadr) || \ + gvt_gmadr_is_hidden(gvt, gmadr)) + +bool intel_gvt_ggtt_validate_range(struct intel_vgpu *vgpu, u64 addr, u32 size); +int intel_gvt_ggtt_gmadr_g2h(struct intel_vgpu *vgpu, u64 g_addr, u64 *h_addr); +int intel_gvt_ggtt_gmadr_h2g(struct intel_vgpu *vgpu, u64 h_addr, u64 *g_addr); +int intel_gvt_ggtt_index_g2h(struct intel_vgpu *vgpu, unsigned long g_index, + unsigned long *h_index); +int intel_gvt_ggtt_h2g_index(struct intel_vgpu *vgpu, unsigned long h_index, + unsigned long *g_index); + +int intel_vgpu_emulate_cfg_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes); + +int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes); + +void intel_gvt_clean_opregion(struct intel_gvt *gvt); +int intel_gvt_init_opregion(struct intel_gvt *gvt); + +void intel_vgpu_clean_opregion(struct intel_vgpu *vgpu); +int intel_vgpu_init_opregion(struct intel_vgpu *vgpu, u32 gpa); + +int intel_vgpu_emulate_opregion_request(struct intel_vgpu *vgpu, u32 swsci); +int setup_vgpu_mmio(struct intel_vgpu *vgpu); +void populate_pvinfo_page(struct intel_vgpu *vgpu); + +struct intel_gvt_ops { + int (*emulate_cfg_read)(struct intel_vgpu *, unsigned int, void *, + unsigned int); + int (*emulate_cfg_write)(struct intel_vgpu *, unsigned int, void *, + unsigned int); + int (*emulate_mmio_read)(struct intel_vgpu *, u64, void *, + unsigned int); + int (*emulate_mmio_write)(struct intel_vgpu *, u64, void *, + unsigned int); + struct intel_vgpu *(*vgpu_create)(struct intel_gvt *, + struct intel_vgpu_type *); + void (*vgpu_destroy)(struct intel_vgpu *); + void (*vgpu_reset)(struct intel_vgpu *); +}; + + #include "mpt.h" #endif diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c new file mode 100644 index 000000000000..522809710312 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/handlers.c @@ -0,0 +1,2848 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Eddie Dong <eddie.dong@intel.com> + * Zhiyuan Lv <zhiyuan.lv@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Pei Zhang <pei.zhang@intel.com> + * Niu Bing <bing.niu@intel.com> + * Ping Gao <ping.a.gao@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + + */ + +#include "i915_drv.h" +#include "gvt.h" +#include "i915_pvinfo.h" + +/* XXX FIXME i915 has changed PP_XXX definition */ +#define PCH_PP_STATUS _MMIO(0xc7200) +#define PCH_PP_CONTROL _MMIO(0xc7204) +#define PCH_PP_ON_DELAYS _MMIO(0xc7208) +#define PCH_PP_OFF_DELAYS _MMIO(0xc720c) +#define PCH_PP_DIVISOR _MMIO(0xc7210) + +/* Register contains RO bits */ +#define F_RO (1 << 0) +/* Register contains graphics address */ +#define F_GMADR (1 << 1) +/* Mode mask registers with high 16 bits as the mask bits */ +#define F_MODE_MASK (1 << 2) +/* This reg can be accessed by GPU commands */ +#define F_CMD_ACCESS (1 << 3) +/* This reg has been accessed by a VM */ +#define F_ACCESSED (1 << 4) +/* This reg has been accessed through GPU commands */ +#define F_CMD_ACCESSED (1 << 5) +/* This reg could be accessed by unaligned address */ +#define F_UNALIGN (1 << 6) + +unsigned long intel_gvt_get_device_type(struct intel_gvt *gvt) +{ + if (IS_BROADWELL(gvt->dev_priv)) + return D_BDW; + else if (IS_SKYLAKE(gvt->dev_priv)) + return D_SKL; + + return 0; +} + +bool intel_gvt_match_device(struct intel_gvt *gvt, + unsigned long device) +{ + return intel_gvt_get_device_type(gvt) & device; +} + +static void read_vreg(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + memcpy(p_data, &vgpu_vreg(vgpu, offset), bytes); +} + +static void write_vreg(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + memcpy(&vgpu_vreg(vgpu, offset), p_data, bytes); +} + +static int new_mmio_info(struct intel_gvt *gvt, + u32 offset, u32 flags, u32 size, + u32 addr_mask, u32 ro_mask, u32 device, + void *read, void *write) +{ + struct intel_gvt_mmio_info *info, *p; + u32 start, end, i; + + if (!intel_gvt_match_device(gvt, device)) + return 0; + + if (WARN_ON(!IS_ALIGNED(offset, 4))) + return -EINVAL; + + start = offset; + end = offset + size; + + for (i = start; i < end; i += 4) { + info = kzalloc(sizeof(*info), GFP_KERNEL); + if (!info) + return -ENOMEM; + + info->offset = i; + p = intel_gvt_find_mmio_info(gvt, info->offset); + if (p) + gvt_err("dup mmio definition offset %x\n", + info->offset); + info->size = size; + info->length = (i + 4) < end ? 4 : (end - i); + info->addr_mask = addr_mask; + info->device = device; + info->read = read ? read : intel_vgpu_default_mmio_read; + info->write = write ? write : intel_vgpu_default_mmio_write; + gvt->mmio.mmio_attribute[info->offset / 4] = flags; + INIT_HLIST_NODE(&info->node); + hash_add(gvt->mmio.mmio_info_table, &info->node, info->offset); + } + return 0; +} + +static int render_mmio_to_ring_id(struct intel_gvt *gvt, unsigned int reg) +{ + enum intel_engine_id id; + struct intel_engine_cs *engine; + + reg &= ~GENMASK(11, 0); + for_each_engine(engine, gvt->dev_priv, id) { + if (engine->mmio_base == reg) + return id; + } + return -1; +} + +#define offset_to_fence_num(offset) \ + ((offset - i915_mmio_reg_offset(FENCE_REG_GEN6_LO(0))) >> 3) + +#define fence_num_to_offset(num) \ + (num * 8 + i915_mmio_reg_offset(FENCE_REG_GEN6_LO(0))) + +static int sanitize_fence_mmio_access(struct intel_vgpu *vgpu, + unsigned int fence_num, void *p_data, unsigned int bytes) +{ + if (fence_num >= vgpu_fence_sz(vgpu)) { + gvt_err("vgpu%d: found oob fence register access\n", + vgpu->id); + gvt_err("vgpu%d: total fence num %d access fence num %d\n", + vgpu->id, vgpu_fence_sz(vgpu), fence_num); + memset(p_data, 0, bytes); + } + return 0; +} + +static int fence_mmio_read(struct intel_vgpu *vgpu, unsigned int off, + void *p_data, unsigned int bytes) +{ + int ret; + + ret = sanitize_fence_mmio_access(vgpu, offset_to_fence_num(off), + p_data, bytes); + if (ret) + return ret; + read_vreg(vgpu, off, p_data, bytes); + return 0; +} + +static int fence_mmio_write(struct intel_vgpu *vgpu, unsigned int off, + void *p_data, unsigned int bytes) +{ + unsigned int fence_num = offset_to_fence_num(off); + int ret; + + ret = sanitize_fence_mmio_access(vgpu, fence_num, p_data, bytes); + if (ret) + return ret; + write_vreg(vgpu, off, p_data, bytes); + + intel_vgpu_write_fence(vgpu, fence_num, + vgpu_vreg64(vgpu, fence_num_to_offset(fence_num))); + return 0; +} + +#define CALC_MODE_MASK_REG(old, new) \ + (((new) & GENMASK(31, 16)) \ + | ((((old) & GENMASK(15, 0)) & ~((new) >> 16)) \ + | ((new) & ((new) >> 16)))) + +static int mul_force_wake_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 old, new; + uint32_t ack_reg_offset; + + old = vgpu_vreg(vgpu, offset); + new = CALC_MODE_MASK_REG(old, *(u32 *)p_data); + + if (IS_SKYLAKE(vgpu->gvt->dev_priv)) { + switch (offset) { + case FORCEWAKE_RENDER_GEN9_REG: + ack_reg_offset = FORCEWAKE_ACK_RENDER_GEN9_REG; + break; + case FORCEWAKE_BLITTER_GEN9_REG: + ack_reg_offset = FORCEWAKE_ACK_BLITTER_GEN9_REG; + break; + case FORCEWAKE_MEDIA_GEN9_REG: + ack_reg_offset = FORCEWAKE_ACK_MEDIA_GEN9_REG; + break; + default: + /*should not hit here*/ + gvt_err("invalid forcewake offset 0x%x\n", offset); + return 1; + } + } else { + ack_reg_offset = FORCEWAKE_ACK_HSW_REG; + } + + vgpu_vreg(vgpu, offset) = new; + vgpu_vreg(vgpu, ack_reg_offset) = (new & GENMASK(15, 0)); + return 0; +} + +static int handle_device_reset(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes, unsigned long bitmap) +{ + struct intel_gvt_workload_scheduler *scheduler = + &vgpu->gvt->scheduler; + + vgpu->resetting = true; + + intel_vgpu_stop_schedule(vgpu); + /* + * The current_vgpu will set to NULL after stopping the + * scheduler when the reset is triggered by current vgpu. + */ + if (scheduler->current_vgpu == NULL) { + mutex_unlock(&vgpu->gvt->lock); + intel_gvt_wait_vgpu_idle(vgpu); + mutex_lock(&vgpu->gvt->lock); + } + + intel_vgpu_reset_execlist(vgpu, bitmap); + + /* full GPU reset */ + if (bitmap == 0xff) { + mutex_unlock(&vgpu->gvt->lock); + intel_vgpu_clean_gtt(vgpu); + mutex_lock(&vgpu->gvt->lock); + setup_vgpu_mmio(vgpu); + populate_pvinfo_page(vgpu); + intel_vgpu_init_gtt(vgpu); + } + + vgpu->resetting = false; + + return 0; +} + +static int gdrst_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 data; + u64 bitmap = 0; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (data & GEN6_GRDOM_FULL) { + gvt_dbg_mmio("vgpu%d: request full GPU reset\n", vgpu->id); + bitmap = 0xff; + } + if (data & GEN6_GRDOM_RENDER) { + gvt_dbg_mmio("vgpu%d: request RCS reset\n", vgpu->id); + bitmap |= (1 << RCS); + } + if (data & GEN6_GRDOM_MEDIA) { + gvt_dbg_mmio("vgpu%d: request VCS reset\n", vgpu->id); + bitmap |= (1 << VCS); + } + if (data & GEN6_GRDOM_BLT) { + gvt_dbg_mmio("vgpu%d: request BCS Reset\n", vgpu->id); + bitmap |= (1 << BCS); + } + if (data & GEN6_GRDOM_VECS) { + gvt_dbg_mmio("vgpu%d: request VECS Reset\n", vgpu->id); + bitmap |= (1 << VECS); + } + if (data & GEN8_GRDOM_MEDIA2) { + gvt_dbg_mmio("vgpu%d: request VCS2 Reset\n", vgpu->id); + if (HAS_BSD2(vgpu->gvt->dev_priv)) + bitmap |= (1 << VCS2); + } + return handle_device_reset(vgpu, offset, p_data, bytes, bitmap); +} + +static int gmbus_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + return intel_gvt_i2c_handle_gmbus_read(vgpu, offset, p_data, bytes); +} + +static int gmbus_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + return intel_gvt_i2c_handle_gmbus_write(vgpu, offset, p_data, bytes); +} + +static int pch_pp_control_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & PANEL_POWER_ON) { + vgpu_vreg(vgpu, PCH_PP_STATUS) |= PP_ON; + vgpu_vreg(vgpu, PCH_PP_STATUS) |= PP_SEQUENCE_STATE_ON_IDLE; + vgpu_vreg(vgpu, PCH_PP_STATUS) &= ~PP_SEQUENCE_POWER_DOWN; + vgpu_vreg(vgpu, PCH_PP_STATUS) &= ~PP_CYCLE_DELAY_ACTIVE; + + } else + vgpu_vreg(vgpu, PCH_PP_STATUS) &= + ~(PP_ON | PP_SEQUENCE_POWER_DOWN + | PP_CYCLE_DELAY_ACTIVE); + return 0; +} + +static int transconf_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & TRANS_ENABLE) + vgpu_vreg(vgpu, offset) |= TRANS_STATE_ENABLE; + else + vgpu_vreg(vgpu, offset) &= ~TRANS_STATE_ENABLE; + return 0; +} + +static int lcpll_ctl_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & LCPLL_PLL_DISABLE) + vgpu_vreg(vgpu, offset) &= ~LCPLL_PLL_LOCK; + else + vgpu_vreg(vgpu, offset) |= LCPLL_PLL_LOCK; + + if (vgpu_vreg(vgpu, offset) & LCPLL_CD_SOURCE_FCLK) + vgpu_vreg(vgpu, offset) |= LCPLL_CD_SOURCE_FCLK_DONE; + else + vgpu_vreg(vgpu, offset) &= ~LCPLL_CD_SOURCE_FCLK_DONE; + + return 0; +} + +static int dpy_reg_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + *(u32 *)p_data = (1 << 17); + return 0; +} + +static int dpy_reg_mmio_read_2(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + *(u32 *)p_data = 3; + return 0; +} + +static int dpy_reg_mmio_read_3(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + *(u32 *)p_data = (0x2f << 16); + return 0; +} + +static int pipeconf_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (data & PIPECONF_ENABLE) + vgpu_vreg(vgpu, offset) |= I965_PIPECONF_ACTIVE; + else + vgpu_vreg(vgpu, offset) &= ~I965_PIPECONF_ACTIVE; + intel_gvt_check_vblank_emulation(vgpu->gvt); + return 0; +} + +static int ddi_buf_ctl_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & DDI_BUF_CTL_ENABLE) { + vgpu_vreg(vgpu, offset) &= ~DDI_BUF_IS_IDLE; + } else { + vgpu_vreg(vgpu, offset) |= DDI_BUF_IS_IDLE; + if (offset == i915_mmio_reg_offset(DDI_BUF_CTL(PORT_E))) + vgpu_vreg(vgpu, DP_TP_STATUS(PORT_E)) + &= ~DP_TP_STATUS_AUTOTRAIN_DONE; + } + return 0; +} + +static int fdi_rx_iir_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + vgpu_vreg(vgpu, offset) &= ~*(u32 *)p_data; + return 0; +} + +#define FDI_LINK_TRAIN_PATTERN1 0 +#define FDI_LINK_TRAIN_PATTERN2 1 + +static int fdi_auto_training_started(struct intel_vgpu *vgpu) +{ + u32 ddi_buf_ctl = vgpu_vreg(vgpu, DDI_BUF_CTL(PORT_E)); + u32 rx_ctl = vgpu_vreg(vgpu, _FDI_RXA_CTL); + u32 tx_ctl = vgpu_vreg(vgpu, DP_TP_CTL(PORT_E)); + + if ((ddi_buf_ctl & DDI_BUF_CTL_ENABLE) && + (rx_ctl & FDI_RX_ENABLE) && + (rx_ctl & FDI_AUTO_TRAINING) && + (tx_ctl & DP_TP_CTL_ENABLE) && + (tx_ctl & DP_TP_CTL_FDI_AUTOTRAIN)) + return 1; + else + return 0; +} + +static int check_fdi_rx_train_status(struct intel_vgpu *vgpu, + enum pipe pipe, unsigned int train_pattern) +{ + i915_reg_t fdi_rx_imr, fdi_tx_ctl, fdi_rx_ctl; + unsigned int fdi_rx_check_bits, fdi_tx_check_bits; + unsigned int fdi_rx_train_bits, fdi_tx_train_bits; + unsigned int fdi_iir_check_bits; + + fdi_rx_imr = FDI_RX_IMR(pipe); + fdi_tx_ctl = FDI_TX_CTL(pipe); + fdi_rx_ctl = FDI_RX_CTL(pipe); + + if (train_pattern == FDI_LINK_TRAIN_PATTERN1) { + fdi_rx_train_bits = FDI_LINK_TRAIN_PATTERN_1_CPT; + fdi_tx_train_bits = FDI_LINK_TRAIN_PATTERN_1; + fdi_iir_check_bits = FDI_RX_BIT_LOCK; + } else if (train_pattern == FDI_LINK_TRAIN_PATTERN2) { + fdi_rx_train_bits = FDI_LINK_TRAIN_PATTERN_2_CPT; + fdi_tx_train_bits = FDI_LINK_TRAIN_PATTERN_2; + fdi_iir_check_bits = FDI_RX_SYMBOL_LOCK; + } else { + gvt_err("Invalid train pattern %d\n", train_pattern); + return -EINVAL; + } + + fdi_rx_check_bits = FDI_RX_ENABLE | fdi_rx_train_bits; + fdi_tx_check_bits = FDI_TX_ENABLE | fdi_tx_train_bits; + + /* If imr bit has been masked */ + if (vgpu_vreg(vgpu, fdi_rx_imr) & fdi_iir_check_bits) + return 0; + + if (((vgpu_vreg(vgpu, fdi_tx_ctl) & fdi_tx_check_bits) + == fdi_tx_check_bits) + && ((vgpu_vreg(vgpu, fdi_rx_ctl) & fdi_rx_check_bits) + == fdi_rx_check_bits)) + return 1; + else + return 0; +} + +#define INVALID_INDEX (~0U) + +static unsigned int calc_index(unsigned int offset, unsigned int start, + unsigned int next, unsigned int end, i915_reg_t i915_end) +{ + unsigned int range = next - start; + + if (!end) + end = i915_mmio_reg_offset(i915_end); + if (offset < start || offset > end) + return INVALID_INDEX; + offset -= start; + return offset / range; +} + +#define FDI_RX_CTL_TO_PIPE(offset) \ + calc_index(offset, _FDI_RXA_CTL, _FDI_RXB_CTL, 0, FDI_RX_CTL(PIPE_C)) + +#define FDI_TX_CTL_TO_PIPE(offset) \ + calc_index(offset, _FDI_TXA_CTL, _FDI_TXB_CTL, 0, FDI_TX_CTL(PIPE_C)) + +#define FDI_RX_IMR_TO_PIPE(offset) \ + calc_index(offset, _FDI_RXA_IMR, _FDI_RXB_IMR, 0, FDI_RX_IMR(PIPE_C)) + +static int update_fdi_rx_iir_status(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + i915_reg_t fdi_rx_iir; + unsigned int index; + int ret; + + if (FDI_RX_CTL_TO_PIPE(offset) != INVALID_INDEX) + index = FDI_RX_CTL_TO_PIPE(offset); + else if (FDI_TX_CTL_TO_PIPE(offset) != INVALID_INDEX) + index = FDI_TX_CTL_TO_PIPE(offset); + else if (FDI_RX_IMR_TO_PIPE(offset) != INVALID_INDEX) + index = FDI_RX_IMR_TO_PIPE(offset); + else { + gvt_err("Unsupport registers %x\n", offset); + return -EINVAL; + } + + write_vreg(vgpu, offset, p_data, bytes); + + fdi_rx_iir = FDI_RX_IIR(index); + + ret = check_fdi_rx_train_status(vgpu, index, FDI_LINK_TRAIN_PATTERN1); + if (ret < 0) + return ret; + if (ret) + vgpu_vreg(vgpu, fdi_rx_iir) |= FDI_RX_BIT_LOCK; + + ret = check_fdi_rx_train_status(vgpu, index, FDI_LINK_TRAIN_PATTERN2); + if (ret < 0) + return ret; + if (ret) + vgpu_vreg(vgpu, fdi_rx_iir) |= FDI_RX_SYMBOL_LOCK; + + if (offset == _FDI_RXA_CTL) + if (fdi_auto_training_started(vgpu)) + vgpu_vreg(vgpu, DP_TP_STATUS(PORT_E)) |= + DP_TP_STATUS_AUTOTRAIN_DONE; + return 0; +} + +#define DP_TP_CTL_TO_PORT(offset) \ + calc_index(offset, _DP_TP_CTL_A, _DP_TP_CTL_B, 0, DP_TP_CTL(PORT_E)) + +static int dp_tp_ctl_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + i915_reg_t status_reg; + unsigned int index; + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + + index = DP_TP_CTL_TO_PORT(offset); + data = (vgpu_vreg(vgpu, offset) & GENMASK(10, 8)) >> 8; + if (data == 0x2) { + status_reg = DP_TP_STATUS(index); + vgpu_vreg(vgpu, status_reg) |= (1 << 25); + } + return 0; +} + +static int dp_tp_status_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 reg_val; + u32 sticky_mask; + + reg_val = *((u32 *)p_data); + sticky_mask = GENMASK(27, 26) | (1 << 24); + + vgpu_vreg(vgpu, offset) = (reg_val & ~sticky_mask) | + (vgpu_vreg(vgpu, offset) & sticky_mask); + vgpu_vreg(vgpu, offset) &= ~(reg_val & sticky_mask); + return 0; +} + +static int pch_adpa_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (data & ADPA_CRT_HOTPLUG_FORCE_TRIGGER) + vgpu_vreg(vgpu, offset) &= ~ADPA_CRT_HOTPLUG_FORCE_TRIGGER; + return 0; +} + +static int south_chicken2_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (data & FDI_MPHY_IOSFSB_RESET_CTL) + vgpu_vreg(vgpu, offset) |= FDI_MPHY_IOSFSB_RESET_STATUS; + else + vgpu_vreg(vgpu, offset) &= ~FDI_MPHY_IOSFSB_RESET_STATUS; + return 0; +} + +#define DSPSURF_TO_PIPE(offset) \ + calc_index(offset, _DSPASURF, _DSPBSURF, 0, DSPSURF(PIPE_C)) + +static int pri_surf_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + unsigned int index = DSPSURF_TO_PIPE(offset); + i915_reg_t surflive_reg = DSPSURFLIVE(index); + int flip_event[] = { + [PIPE_A] = PRIMARY_A_FLIP_DONE, + [PIPE_B] = PRIMARY_B_FLIP_DONE, + [PIPE_C] = PRIMARY_C_FLIP_DONE, + }; + + write_vreg(vgpu, offset, p_data, bytes); + vgpu_vreg(vgpu, surflive_reg) = vgpu_vreg(vgpu, offset); + + set_bit(flip_event[index], vgpu->irq.flip_done_event[index]); + return 0; +} + +#define SPRSURF_TO_PIPE(offset) \ + calc_index(offset, _SPRA_SURF, _SPRB_SURF, 0, SPRSURF(PIPE_C)) + +static int spr_surf_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + unsigned int index = SPRSURF_TO_PIPE(offset); + i915_reg_t surflive_reg = SPRSURFLIVE(index); + int flip_event[] = { + [PIPE_A] = SPRITE_A_FLIP_DONE, + [PIPE_B] = SPRITE_B_FLIP_DONE, + [PIPE_C] = SPRITE_C_FLIP_DONE, + }; + + write_vreg(vgpu, offset, p_data, bytes); + vgpu_vreg(vgpu, surflive_reg) = vgpu_vreg(vgpu, offset); + + set_bit(flip_event[index], vgpu->irq.flip_done_event[index]); + return 0; +} + +static int trigger_aux_channel_interrupt(struct intel_vgpu *vgpu, + unsigned int reg) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + enum intel_gvt_event_type event; + + if (reg == _DPA_AUX_CH_CTL) + event = AUX_CHANNEL_A; + else if (reg == _PCH_DPB_AUX_CH_CTL || reg == _DPB_AUX_CH_CTL) + event = AUX_CHANNEL_B; + else if (reg == _PCH_DPC_AUX_CH_CTL || reg == _DPC_AUX_CH_CTL) + event = AUX_CHANNEL_C; + else if (reg == _PCH_DPD_AUX_CH_CTL || reg == _DPD_AUX_CH_CTL) + event = AUX_CHANNEL_D; + else { + WARN_ON(true); + return -EINVAL; + } + + intel_vgpu_trigger_virtual_event(vgpu, event); + return 0; +} + +static int dp_aux_ch_ctl_trans_done(struct intel_vgpu *vgpu, u32 value, + unsigned int reg, int len, bool data_valid) +{ + /* mark transaction done */ + value |= DP_AUX_CH_CTL_DONE; + value &= ~DP_AUX_CH_CTL_SEND_BUSY; + value &= ~DP_AUX_CH_CTL_RECEIVE_ERROR; + + if (data_valid) + value &= ~DP_AUX_CH_CTL_TIME_OUT_ERROR; + else + value |= DP_AUX_CH_CTL_TIME_OUT_ERROR; + + /* message size */ + value &= ~(0xf << 20); + value |= (len << 20); + vgpu_vreg(vgpu, reg) = value; + + if (value & DP_AUX_CH_CTL_INTERRUPT) + return trigger_aux_channel_interrupt(vgpu, reg); + return 0; +} + +static void dp_aux_ch_ctl_link_training(struct intel_vgpu_dpcd_data *dpcd, + uint8_t t) +{ + if ((t & DPCD_TRAINING_PATTERN_SET_MASK) == DPCD_TRAINING_PATTERN_1) { + /* training pattern 1 for CR */ + /* set LANE0_CR_DONE, LANE1_CR_DONE */ + dpcd->data[DPCD_LANE0_1_STATUS] |= DPCD_LANES_CR_DONE; + /* set LANE2_CR_DONE, LANE3_CR_DONE */ + dpcd->data[DPCD_LANE2_3_STATUS] |= DPCD_LANES_CR_DONE; + } else if ((t & DPCD_TRAINING_PATTERN_SET_MASK) == + DPCD_TRAINING_PATTERN_2) { + /* training pattern 2 for EQ */ + /* Set CHANNEL_EQ_DONE and SYMBOL_LOCKED for Lane0_1 */ + dpcd->data[DPCD_LANE0_1_STATUS] |= DPCD_LANES_EQ_DONE; + dpcd->data[DPCD_LANE0_1_STATUS] |= DPCD_SYMBOL_LOCKED; + /* Set CHANNEL_EQ_DONE and SYMBOL_LOCKED for Lane2_3 */ + dpcd->data[DPCD_LANE2_3_STATUS] |= DPCD_LANES_EQ_DONE; + dpcd->data[DPCD_LANE2_3_STATUS] |= DPCD_SYMBOL_LOCKED; + /* set INTERLANE_ALIGN_DONE */ + dpcd->data[DPCD_LANE_ALIGN_STATUS_UPDATED] |= + DPCD_INTERLANE_ALIGN_DONE; + } else if ((t & DPCD_TRAINING_PATTERN_SET_MASK) == + DPCD_LINK_TRAINING_DISABLED) { + /* finish link training */ + /* set sink status as synchronized */ + dpcd->data[DPCD_SINK_STATUS] = DPCD_SINK_IN_SYNC; + } +} + +#define _REG_HSW_DP_AUX_CH_CTL(dp) \ + ((dp) ? (_PCH_DPB_AUX_CH_CTL + ((dp)-1)*0x100) : 0x64010) + +#define _REG_SKL_DP_AUX_CH_CTL(dp) (0x64010 + (dp) * 0x100) + +#define OFFSET_TO_DP_AUX_PORT(offset) (((offset) & 0xF00) >> 8) + +#define dpy_is_valid_port(port) \ + (((port) >= PORT_A) && ((port) < I915_MAX_PORTS)) + +static int dp_aux_ch_ctl_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + struct intel_vgpu_display *display = &vgpu->display; + int msg, addr, ctrl, op, len; + int port_index = OFFSET_TO_DP_AUX_PORT(offset); + struct intel_vgpu_dpcd_data *dpcd = NULL; + struct intel_vgpu_port *port = NULL; + u32 data; + + if (!dpy_is_valid_port(port_index)) { + gvt_err("GVT(%d): Unsupported DP port access!\n", vgpu->id); + return 0; + } + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (IS_SKYLAKE(vgpu->gvt->dev_priv) && + offset != _REG_SKL_DP_AUX_CH_CTL(port_index)) { + /* SKL DPB/C/D aux ctl register changed */ + return 0; + } else if (IS_BROADWELL(vgpu->gvt->dev_priv) && + offset != _REG_HSW_DP_AUX_CH_CTL(port_index)) { + /* write to the data registers */ + return 0; + } + + if (!(data & DP_AUX_CH_CTL_SEND_BUSY)) { + /* just want to clear the sticky bits */ + vgpu_vreg(vgpu, offset) = 0; + return 0; + } + + port = &display->ports[port_index]; + dpcd = port->dpcd; + + /* read out message from DATA1 register */ + msg = vgpu_vreg(vgpu, offset + 4); + addr = (msg >> 8) & 0xffff; + ctrl = (msg >> 24) & 0xff; + len = msg & 0xff; + op = ctrl >> 4; + + if (op == GVT_AUX_NATIVE_WRITE) { + int t; + uint8_t buf[16]; + + if ((addr + len + 1) >= DPCD_SIZE) { + /* + * Write request exceeds what we supported, + * DCPD spec: When a Source Device is writing a DPCD + * address not supported by the Sink Device, the Sink + * Device shall reply with AUX NACK and “M” equal to + * zero. + */ + + /* NAK the write */ + vgpu_vreg(vgpu, offset + 4) = AUX_NATIVE_REPLY_NAK; + dp_aux_ch_ctl_trans_done(vgpu, data, offset, 2, true); + return 0; + } + + /* + * Write request format: (command + address) occupies + * 3 bytes, followed by (len + 1) bytes of data. + */ + if (WARN_ON((len + 4) > AUX_BURST_SIZE)) + return -EINVAL; + + /* unpack data from vreg to buf */ + for (t = 0; t < 4; t++) { + u32 r = vgpu_vreg(vgpu, offset + 8 + t * 4); + + buf[t * 4] = (r >> 24) & 0xff; + buf[t * 4 + 1] = (r >> 16) & 0xff; + buf[t * 4 + 2] = (r >> 8) & 0xff; + buf[t * 4 + 3] = r & 0xff; + } + + /* write to virtual DPCD */ + if (dpcd && dpcd->data_valid) { + for (t = 0; t <= len; t++) { + int p = addr + t; + + dpcd->data[p] = buf[t]; + /* check for link training */ + if (p == DPCD_TRAINING_PATTERN_SET) + dp_aux_ch_ctl_link_training(dpcd, + buf[t]); + } + } + + /* ACK the write */ + vgpu_vreg(vgpu, offset + 4) = 0; + dp_aux_ch_ctl_trans_done(vgpu, data, offset, 1, + dpcd && dpcd->data_valid); + return 0; + } + + if (op == GVT_AUX_NATIVE_READ) { + int idx, i, ret = 0; + + if ((addr + len + 1) >= DPCD_SIZE) { + /* + * read request exceeds what we supported + * DPCD spec: A Sink Device receiving a Native AUX CH + * read request for an unsupported DPCD address must + * reply with an AUX ACK and read data set equal to + * zero instead of replying with AUX NACK. + */ + + /* ACK the READ*/ + vgpu_vreg(vgpu, offset + 4) = 0; + vgpu_vreg(vgpu, offset + 8) = 0; + vgpu_vreg(vgpu, offset + 12) = 0; + vgpu_vreg(vgpu, offset + 16) = 0; + vgpu_vreg(vgpu, offset + 20) = 0; + + dp_aux_ch_ctl_trans_done(vgpu, data, offset, len + 2, + true); + return 0; + } + + for (idx = 1; idx <= 5; idx++) { + /* clear the data registers */ + vgpu_vreg(vgpu, offset + 4 * idx) = 0; + } + + /* + * Read reply format: ACK (1 byte) plus (len + 1) bytes of data. + */ + if (WARN_ON((len + 2) > AUX_BURST_SIZE)) + return -EINVAL; + + /* read from virtual DPCD to vreg */ + /* first 4 bytes: [ACK][addr][addr+1][addr+2] */ + if (dpcd && dpcd->data_valid) { + for (i = 1; i <= (len + 1); i++) { + int t; + + t = dpcd->data[addr + i - 1]; + t <<= (24 - 8 * (i % 4)); + ret |= t; + + if ((i % 4 == 3) || (i == (len + 1))) { + vgpu_vreg(vgpu, offset + + (i / 4 + 1) * 4) = ret; + ret = 0; + } + } + } + dp_aux_ch_ctl_trans_done(vgpu, data, offset, len + 2, + dpcd && dpcd->data_valid); + return 0; + } + + /* i2c transaction starts */ + intel_gvt_i2c_handle_aux_ch_write(vgpu, port_index, offset, p_data); + + if (data & DP_AUX_CH_CTL_INTERRUPT) + trigger_aux_channel_interrupt(vgpu, offset); + return 0; +} + +static int vga_control_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + bool vga_disable; + + write_vreg(vgpu, offset, p_data, bytes); + vga_disable = vgpu_vreg(vgpu, offset) & VGA_DISP_DISABLE; + + gvt_dbg_core("vgpu%d: %s VGA mode\n", vgpu->id, + vga_disable ? "Disable" : "Enable"); + return 0; +} + +static u32 read_virtual_sbi_register(struct intel_vgpu *vgpu, + unsigned int sbi_offset) +{ + struct intel_vgpu_display *display = &vgpu->display; + int num = display->sbi.number; + int i; + + for (i = 0; i < num; ++i) + if (display->sbi.registers[i].offset == sbi_offset) + break; + + if (i == num) + return 0; + + return display->sbi.registers[i].value; +} + +static void write_virtual_sbi_register(struct intel_vgpu *vgpu, + unsigned int offset, u32 value) +{ + struct intel_vgpu_display *display = &vgpu->display; + int num = display->sbi.number; + int i; + + for (i = 0; i < num; ++i) { + if (display->sbi.registers[i].offset == offset) + break; + } + + if (i == num) { + if (num == SBI_REG_MAX) { + gvt_err("vgpu%d: SBI caching meets maximum limits\n", + vgpu->id); + return; + } + display->sbi.number++; + } + + display->sbi.registers[i].offset = offset; + display->sbi.registers[i].value = value; +} + +static int sbi_data_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + if (((vgpu_vreg(vgpu, SBI_CTL_STAT) & SBI_OPCODE_MASK) >> + SBI_OPCODE_SHIFT) == SBI_CMD_CRRD) { + unsigned int sbi_offset = (vgpu_vreg(vgpu, SBI_ADDR) & + SBI_ADDR_OFFSET_MASK) >> SBI_ADDR_OFFSET_SHIFT; + vgpu_vreg(vgpu, offset) = read_virtual_sbi_register(vgpu, + sbi_offset); + } + read_vreg(vgpu, offset, p_data, bytes); + return 0; +} + +static bool sbi_ctl_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + data &= ~(SBI_STAT_MASK << SBI_STAT_SHIFT); + data |= SBI_READY; + + data &= ~(SBI_RESPONSE_MASK << SBI_RESPONSE_SHIFT); + data |= SBI_RESPONSE_SUCCESS; + + vgpu_vreg(vgpu, offset) = data; + + if (((vgpu_vreg(vgpu, SBI_CTL_STAT) & SBI_OPCODE_MASK) >> + SBI_OPCODE_SHIFT) == SBI_CMD_CRWR) { + unsigned int sbi_offset = (vgpu_vreg(vgpu, SBI_ADDR) & + SBI_ADDR_OFFSET_MASK) >> SBI_ADDR_OFFSET_SHIFT; + + write_virtual_sbi_register(vgpu, sbi_offset, + vgpu_vreg(vgpu, SBI_DATA)); + } + return 0; +} + +#define _vgtif_reg(x) \ + (VGT_PVINFO_PAGE + offsetof(struct vgt_if, x)) + +static int pvinfo_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + bool invalid_read = false; + + read_vreg(vgpu, offset, p_data, bytes); + + switch (offset) { + case _vgtif_reg(magic) ... _vgtif_reg(vgt_id): + if (offset + bytes > _vgtif_reg(vgt_id) + 4) + invalid_read = true; + break; + case _vgtif_reg(avail_rs.mappable_gmadr.base) ... + _vgtif_reg(avail_rs.fence_num): + if (offset + bytes > + _vgtif_reg(avail_rs.fence_num) + 4) + invalid_read = true; + break; + case 0x78010: /* vgt_caps */ + case 0x7881c: + break; + default: + invalid_read = true; + break; + } + if (invalid_read) + gvt_err("invalid pvinfo read: [%x:%x] = %x\n", + offset, bytes, *(u32 *)p_data); + return 0; +} + +static int handle_g2v_notification(struct intel_vgpu *vgpu, int notification) +{ + int ret = 0; + + switch (notification) { + case VGT_G2V_PPGTT_L3_PAGE_TABLE_CREATE: + ret = intel_vgpu_g2v_create_ppgtt_mm(vgpu, 3); + break; + case VGT_G2V_PPGTT_L3_PAGE_TABLE_DESTROY: + ret = intel_vgpu_g2v_destroy_ppgtt_mm(vgpu, 3); + break; + case VGT_G2V_PPGTT_L4_PAGE_TABLE_CREATE: + ret = intel_vgpu_g2v_create_ppgtt_mm(vgpu, 4); + break; + case VGT_G2V_PPGTT_L4_PAGE_TABLE_DESTROY: + ret = intel_vgpu_g2v_destroy_ppgtt_mm(vgpu, 4); + break; + case VGT_G2V_EXECLIST_CONTEXT_CREATE: + case VGT_G2V_EXECLIST_CONTEXT_DESTROY: + case 1: /* Remove this in guest driver. */ + break; + default: + gvt_err("Invalid PV notification %d\n", notification); + } + return ret; +} + +static int send_display_ready_uevent(struct intel_vgpu *vgpu, int ready) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct kobject *kobj = &dev_priv->drm.primary->kdev->kobj; + char *env[3] = {NULL, NULL, NULL}; + char vmid_str[20]; + char display_ready_str[20]; + + snprintf(display_ready_str, 20, "GVT_DISPLAY_READY=%d\n", ready); + env[0] = display_ready_str; + + snprintf(vmid_str, 20, "VMID=%d", vgpu->id); + env[1] = vmid_str; + + return kobject_uevent_env(kobj, KOBJ_ADD, env); +} + +static int pvinfo_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 data; + int ret; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + switch (offset) { + case _vgtif_reg(display_ready): + send_display_ready_uevent(vgpu, data ? 1 : 0); + break; + case _vgtif_reg(g2v_notify): + ret = handle_g2v_notification(vgpu, data); + break; + /* add xhot and yhot to handled list to avoid error log */ + case 0x78830: + case 0x78834: + case _vgtif_reg(pdp[0].lo): + case _vgtif_reg(pdp[0].hi): + case _vgtif_reg(pdp[1].lo): + case _vgtif_reg(pdp[1].hi): + case _vgtif_reg(pdp[2].lo): + case _vgtif_reg(pdp[2].hi): + case _vgtif_reg(pdp[3].lo): + case _vgtif_reg(pdp[3].hi): + case _vgtif_reg(execlist_context_descriptor_lo): + case _vgtif_reg(execlist_context_descriptor_hi): + break; + default: + gvt_err("invalid pvinfo write offset %x bytes %x data %x\n", + offset, bytes, data); + break; + } + return 0; +} + +static int pf_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 val = *(u32 *)p_data; + + if ((offset == _PS_1A_CTRL || offset == _PS_2A_CTRL || + offset == _PS_1B_CTRL || offset == _PS_2B_CTRL || + offset == _PS_1C_CTRL) && (val & PS_PLANE_SEL_MASK) != 0) { + WARN_ONCE(true, "VM(%d): guest is trying to scaling a plane\n", + vgpu->id); + return 0; + } + + return intel_vgpu_default_mmio_write(vgpu, offset, p_data, bytes); +} + +static int power_well_ctl_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & HSW_PWR_WELL_ENABLE_REQUEST) + vgpu_vreg(vgpu, offset) |= HSW_PWR_WELL_STATE_ENABLED; + else + vgpu_vreg(vgpu, offset) &= ~HSW_PWR_WELL_STATE_ENABLED; + return 0; +} + +static int fpga_dbg_mmio_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + + if (vgpu_vreg(vgpu, offset) & FPGA_DBG_RM_NOCLAIM) + vgpu_vreg(vgpu, offset) &= ~FPGA_DBG_RM_NOCLAIM; + return 0; +} + +static int dma_ctrl_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 mode; + + write_vreg(vgpu, offset, p_data, bytes); + mode = vgpu_vreg(vgpu, offset); + + if (GFX_MODE_BIT_SET_IN_MASK(mode, START_DMA)) { + WARN_ONCE(1, "VM(%d): iGVT-g doesn't supporte GuC\n", + vgpu->id); + return 0; + } + + return 0; +} + +static int gen9_trtte_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + u32 trtte = *(u32 *)p_data; + + if ((trtte & 1) && (trtte & (1 << 1)) == 0) { + WARN(1, "VM(%d): Use physical address for TRTT!\n", + vgpu->id); + return -EINVAL; + } + write_vreg(vgpu, offset, p_data, bytes); + /* TRTTE is not per-context */ + I915_WRITE(_MMIO(offset), vgpu_vreg(vgpu, offset)); + + return 0; +} + +static int gen9_trtt_chicken_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + u32 val = *(u32 *)p_data; + + if (val & 1) { + /* unblock hw logic */ + I915_WRITE(_MMIO(offset), val); + } + write_vreg(vgpu, offset, p_data, bytes); + return 0; +} + +static int dpll_status_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 v = 0; + + if (vgpu_vreg(vgpu, 0x46010) & (1 << 31)) + v |= (1 << 0); + + if (vgpu_vreg(vgpu, 0x46014) & (1 << 31)) + v |= (1 << 8); + + if (vgpu_vreg(vgpu, 0x46040) & (1 << 31)) + v |= (1 << 16); + + if (vgpu_vreg(vgpu, 0x46060) & (1 << 31)) + v |= (1 << 24); + + vgpu_vreg(vgpu, offset) = v; + + return intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes); +} + +static int mailbox_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 value = *(u32 *)p_data; + u32 cmd = value & 0xff; + u32 *data0 = &vgpu_vreg(vgpu, GEN6_PCODE_DATA); + + switch (cmd) { + case 0x6: + /** + * "Read memory latency" command on gen9. + * Below memory latency values are read + * from skylake platform. + */ + if (!*data0) + *data0 = 0x1e1a1100; + else + *data0 = 0x61514b3d; + break; + case 0x5: + *data0 |= 0x1; + break; + } + + gvt_dbg_core("VM(%d) write %x to mailbox, return data0 %x\n", + vgpu->id, value, *data0); + + value &= ~(1 << 31); + return intel_vgpu_default_mmio_write(vgpu, offset, &value, bytes); +} + +static int skl_power_well_ctl_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 v = *(u32 *)p_data; + + v &= (1 << 31) | (1 << 29) | (1 << 9) | + (1 << 7) | (1 << 5) | (1 << 3) | (1 << 1); + v |= (v >> 1); + + return intel_vgpu_default_mmio_write(vgpu, offset, &v, bytes); +} + +static int skl_misc_ctl_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + i915_reg_t reg = {.reg = offset}; + + switch (offset) { + case 0x4ddc: + vgpu_vreg(vgpu, offset) = 0x8000003c; + /* WaCompressedResourceSamplerPbeMediaNewHashMode:skl */ + I915_WRITE(reg, vgpu_vreg(vgpu, offset)); + break; + case 0x42080: + vgpu_vreg(vgpu, offset) = 0x8000; + /* WaCompressedResourceDisplayNewHashMode:skl */ + I915_WRITE(reg, vgpu_vreg(vgpu, offset)); + break; + default: + return -EINVAL; + } + + return 0; +} + +static int skl_lcpll_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 v = *(u32 *)p_data; + + /* other bits are MBZ. */ + v &= (1 << 31) | (1 << 30); + v & (1 << 31) ? (v |= (1 << 30)) : (v &= ~(1 << 30)); + + vgpu_vreg(vgpu, offset) = v; + + return 0; +} + +static int ring_timestamp_mmio_read(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + vgpu_vreg(vgpu, offset) = I915_READ(_MMIO(offset)); + return intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes); +} + +static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset); + struct intel_vgpu_execlist *execlist; + u32 data = *(u32 *)p_data; + int ret = 0; + + if (WARN_ON(ring_id < 0 || ring_id > I915_NUM_ENGINES - 1)) + return -EINVAL; + + execlist = &vgpu->execlist[ring_id]; + + execlist->elsp_dwords.data[execlist->elsp_dwords.index] = data; + if (execlist->elsp_dwords.index == 3) { + ret = intel_vgpu_submit_execlist(vgpu, ring_id); + if(ret) + gvt_err("fail submit workload on ring %d\n", ring_id); + } + + ++execlist->elsp_dwords.index; + execlist->elsp_dwords.index &= 0x3; + return ret; +} + +static int ring_mode_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + u32 data = *(u32 *)p_data; + int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset); + bool enable_execlist; + + write_vreg(vgpu, offset, p_data, bytes); + if ((data & _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)) + || (data & _MASKED_BIT_DISABLE(GFX_RUN_LIST_ENABLE))) { + enable_execlist = !!(data & GFX_RUN_LIST_ENABLE); + + gvt_dbg_core("EXECLIST %s on ring %d\n", + (enable_execlist ? "enabling" : "disabling"), + ring_id); + + if (enable_execlist) + intel_vgpu_start_schedule(vgpu); + } + return 0; +} + +static int gvt_reg_tlb_control_handler(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + int rc = 0; + unsigned int id = 0; + + write_vreg(vgpu, offset, p_data, bytes); + vgpu_vreg(vgpu, offset) = 0; + + switch (offset) { + case 0x4260: + id = RCS; + break; + case 0x4264: + id = VCS; + break; + case 0x4268: + id = VCS2; + break; + case 0x426c: + id = BCS; + break; + case 0x4270: + id = VECS; + break; + default: + rc = -EINVAL; + break; + } + set_bit(id, (void *)vgpu->tlb_handle_pending); + + return rc; +} + +static int ring_reset_ctl_write(struct intel_vgpu *vgpu, + unsigned int offset, void *p_data, unsigned int bytes) +{ + u32 data; + + write_vreg(vgpu, offset, p_data, bytes); + data = vgpu_vreg(vgpu, offset); + + if (data & _MASKED_BIT_ENABLE(RESET_CTL_REQUEST_RESET)) + data |= RESET_CTL_READY_TO_RESET; + else if (data & _MASKED_BIT_DISABLE(RESET_CTL_REQUEST_RESET)) + data &= ~RESET_CTL_READY_TO_RESET; + + vgpu_vreg(vgpu, offset) = data; + return 0; +} + +#define MMIO_F(reg, s, f, am, rm, d, r, w) do { \ + ret = new_mmio_info(gvt, INTEL_GVT_MMIO_OFFSET(reg), \ + f, s, am, rm, d, r, w); \ + if (ret) \ + return ret; \ +} while (0) + +#define MMIO_D(reg, d) \ + MMIO_F(reg, 4, 0, 0, 0, d, NULL, NULL) + +#define MMIO_DH(reg, d, r, w) \ + MMIO_F(reg, 4, 0, 0, 0, d, r, w) + +#define MMIO_DFH(reg, d, f, r, w) \ + MMIO_F(reg, 4, f, 0, 0, d, r, w) + +#define MMIO_GM(reg, d, r, w) \ + MMIO_F(reg, 4, F_GMADR, 0xFFFFF000, 0, d, r, w) + +#define MMIO_RO(reg, d, f, rm, r, w) \ + MMIO_F(reg, 4, F_RO | f, 0, rm, d, r, w) + +#define MMIO_RING_F(prefix, s, f, am, rm, d, r, w) do { \ + MMIO_F(prefix(RENDER_RING_BASE), s, f, am, rm, d, r, w); \ + MMIO_F(prefix(BLT_RING_BASE), s, f, am, rm, d, r, w); \ + MMIO_F(prefix(GEN6_BSD_RING_BASE), s, f, am, rm, d, r, w); \ + MMIO_F(prefix(VEBOX_RING_BASE), s, f, am, rm, d, r, w); \ +} while (0) + +#define MMIO_RING_D(prefix, d) \ + MMIO_RING_F(prefix, 4, 0, 0, 0, d, NULL, NULL) + +#define MMIO_RING_DFH(prefix, d, f, r, w) \ + MMIO_RING_F(prefix, 4, f, 0, 0, d, r, w) + +#define MMIO_RING_GM(prefix, d, r, w) \ + MMIO_RING_F(prefix, 4, F_GMADR, 0xFFFF0000, 0, d, r, w) + +#define MMIO_RING_RO(prefix, d, f, rm, r, w) \ + MMIO_RING_F(prefix, 4, F_RO | f, 0, rm, d, r, w) + +static int init_generic_mmio_info(struct intel_gvt *gvt) +{ + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ret; + + MMIO_RING_DFH(RING_IMR, D_ALL, 0, NULL, intel_vgpu_reg_imr_handler); + + MMIO_DFH(SDEIMR, D_ALL, 0, NULL, intel_vgpu_reg_imr_handler); + MMIO_DFH(SDEIER, D_ALL, 0, NULL, intel_vgpu_reg_ier_handler); + MMIO_DFH(SDEIIR, D_ALL, 0, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(SDEISR, D_ALL); + + MMIO_RING_D(RING_HWSTAM, D_ALL); + + MMIO_GM(RENDER_HWS_PGA_GEN7, D_ALL, NULL, NULL); + MMIO_GM(BSD_HWS_PGA_GEN7, D_ALL, NULL, NULL); + MMIO_GM(BLT_HWS_PGA_GEN7, D_ALL, NULL, NULL); + MMIO_GM(VEBOX_HWS_PGA_GEN7, D_ALL, NULL, NULL); + +#define RING_REG(base) (base + 0x28) + MMIO_RING_D(RING_REG, D_ALL); +#undef RING_REG + +#define RING_REG(base) (base + 0x134) + MMIO_RING_D(RING_REG, D_ALL); +#undef RING_REG + + MMIO_GM(0x2148, D_ALL, NULL, NULL); + MMIO_GM(CCID, D_ALL, NULL, NULL); + MMIO_GM(0x12198, D_ALL, NULL, NULL); + MMIO_D(GEN7_CXT_SIZE, D_ALL); + + MMIO_RING_D(RING_TAIL, D_ALL); + MMIO_RING_D(RING_HEAD, D_ALL); + MMIO_RING_D(RING_CTL, D_ALL); + MMIO_RING_D(RING_ACTHD, D_ALL); + MMIO_RING_GM(RING_START, D_ALL, NULL, NULL); + + /* RING MODE */ +#define RING_REG(base) (base + 0x29c) + MMIO_RING_DFH(RING_REG, D_ALL, F_MODE_MASK, NULL, ring_mode_mmio_write); +#undef RING_REG + + MMIO_RING_DFH(RING_MI_MODE, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_RING_DFH(RING_INSTPM, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_RING_DFH(RING_TIMESTAMP, D_ALL, F_CMD_ACCESS, + ring_timestamp_mmio_read, NULL); + MMIO_RING_DFH(RING_TIMESTAMP_UDW, D_ALL, F_CMD_ACCESS, + ring_timestamp_mmio_read, NULL); + + MMIO_DFH(GEN7_GT_MODE, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(CACHE_MODE_0_GEN7, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(CACHE_MODE_1, D_ALL, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + + MMIO_DFH(0x20dc, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(_3D_CHICKEN3, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(0x2088, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(0x20e4, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(0x2470, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_D(GAM_ECOCHK, D_ALL); + MMIO_DFH(GEN7_COMMON_SLICE_CHICKEN1, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(COMMON_SLICE_CHICKEN2, D_ALL, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + MMIO_D(0x9030, D_ALL); + MMIO_D(0x20a0, D_ALL); + MMIO_D(0x2420, D_ALL); + MMIO_D(0x2430, D_ALL); + MMIO_D(0x2434, D_ALL); + MMIO_D(0x2438, D_ALL); + MMIO_D(0x243c, D_ALL); + MMIO_DFH(0x7018, D_ALL, F_MODE_MASK, NULL, NULL); + MMIO_DFH(HALF_SLICE_CHICKEN3, D_ALL, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0xe100, D_ALL, F_MODE_MASK, NULL, NULL); + + /* display */ + MMIO_F(0x60220, 0x20, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_D(0x602a0, D_ALL); + + MMIO_D(0x65050, D_ALL); + MMIO_D(0x650b4, D_ALL); + + MMIO_D(0xc4040, D_ALL); + MMIO_D(DERRMR, D_ALL); + + MMIO_D(PIPEDSL(PIPE_A), D_ALL); + MMIO_D(PIPEDSL(PIPE_B), D_ALL); + MMIO_D(PIPEDSL(PIPE_C), D_ALL); + MMIO_D(PIPEDSL(_PIPE_EDP), D_ALL); + + MMIO_DH(PIPECONF(PIPE_A), D_ALL, NULL, pipeconf_mmio_write); + MMIO_DH(PIPECONF(PIPE_B), D_ALL, NULL, pipeconf_mmio_write); + MMIO_DH(PIPECONF(PIPE_C), D_ALL, NULL, pipeconf_mmio_write); + MMIO_DH(PIPECONF(_PIPE_EDP), D_ALL, NULL, pipeconf_mmio_write); + + MMIO_D(PIPESTAT(PIPE_A), D_ALL); + MMIO_D(PIPESTAT(PIPE_B), D_ALL); + MMIO_D(PIPESTAT(PIPE_C), D_ALL); + MMIO_D(PIPESTAT(_PIPE_EDP), D_ALL); + + MMIO_D(PIPE_FLIPCOUNT_G4X(PIPE_A), D_ALL); + MMIO_D(PIPE_FLIPCOUNT_G4X(PIPE_B), D_ALL); + MMIO_D(PIPE_FLIPCOUNT_G4X(PIPE_C), D_ALL); + MMIO_D(PIPE_FLIPCOUNT_G4X(_PIPE_EDP), D_ALL); + + MMIO_D(PIPE_FRMCOUNT_G4X(PIPE_A), D_ALL); + MMIO_D(PIPE_FRMCOUNT_G4X(PIPE_B), D_ALL); + MMIO_D(PIPE_FRMCOUNT_G4X(PIPE_C), D_ALL); + MMIO_D(PIPE_FRMCOUNT_G4X(_PIPE_EDP), D_ALL); + + MMIO_D(CURCNTR(PIPE_A), D_ALL); + MMIO_D(CURCNTR(PIPE_B), D_ALL); + MMIO_D(CURCNTR(PIPE_C), D_ALL); + + MMIO_D(CURPOS(PIPE_A), D_ALL); + MMIO_D(CURPOS(PIPE_B), D_ALL); + MMIO_D(CURPOS(PIPE_C), D_ALL); + + MMIO_D(CURBASE(PIPE_A), D_ALL); + MMIO_D(CURBASE(PIPE_B), D_ALL); + MMIO_D(CURBASE(PIPE_C), D_ALL); + + MMIO_D(0x700ac, D_ALL); + MMIO_D(0x710ac, D_ALL); + MMIO_D(0x720ac, D_ALL); + + MMIO_D(0x70090, D_ALL); + MMIO_D(0x70094, D_ALL); + MMIO_D(0x70098, D_ALL); + MMIO_D(0x7009c, D_ALL); + + MMIO_D(DSPCNTR(PIPE_A), D_ALL); + MMIO_D(DSPADDR(PIPE_A), D_ALL); + MMIO_D(DSPSTRIDE(PIPE_A), D_ALL); + MMIO_D(DSPPOS(PIPE_A), D_ALL); + MMIO_D(DSPSIZE(PIPE_A), D_ALL); + MMIO_DH(DSPSURF(PIPE_A), D_ALL, NULL, pri_surf_mmio_write); + MMIO_D(DSPOFFSET(PIPE_A), D_ALL); + MMIO_D(DSPSURFLIVE(PIPE_A), D_ALL); + + MMIO_D(DSPCNTR(PIPE_B), D_ALL); + MMIO_D(DSPADDR(PIPE_B), D_ALL); + MMIO_D(DSPSTRIDE(PIPE_B), D_ALL); + MMIO_D(DSPPOS(PIPE_B), D_ALL); + MMIO_D(DSPSIZE(PIPE_B), D_ALL); + MMIO_DH(DSPSURF(PIPE_B), D_ALL, NULL, pri_surf_mmio_write); + MMIO_D(DSPOFFSET(PIPE_B), D_ALL); + MMIO_D(DSPSURFLIVE(PIPE_B), D_ALL); + + MMIO_D(DSPCNTR(PIPE_C), D_ALL); + MMIO_D(DSPADDR(PIPE_C), D_ALL); + MMIO_D(DSPSTRIDE(PIPE_C), D_ALL); + MMIO_D(DSPPOS(PIPE_C), D_ALL); + MMIO_D(DSPSIZE(PIPE_C), D_ALL); + MMIO_DH(DSPSURF(PIPE_C), D_ALL, NULL, pri_surf_mmio_write); + MMIO_D(DSPOFFSET(PIPE_C), D_ALL); + MMIO_D(DSPSURFLIVE(PIPE_C), D_ALL); + + MMIO_D(SPRCTL(PIPE_A), D_ALL); + MMIO_D(SPRLINOFF(PIPE_A), D_ALL); + MMIO_D(SPRSTRIDE(PIPE_A), D_ALL); + MMIO_D(SPRPOS(PIPE_A), D_ALL); + MMIO_D(SPRSIZE(PIPE_A), D_ALL); + MMIO_D(SPRKEYVAL(PIPE_A), D_ALL); + MMIO_D(SPRKEYMSK(PIPE_A), D_ALL); + MMIO_DH(SPRSURF(PIPE_A), D_ALL, NULL, spr_surf_mmio_write); + MMIO_D(SPRKEYMAX(PIPE_A), D_ALL); + MMIO_D(SPROFFSET(PIPE_A), D_ALL); + MMIO_D(SPRSCALE(PIPE_A), D_ALL); + MMIO_D(SPRSURFLIVE(PIPE_A), D_ALL); + + MMIO_D(SPRCTL(PIPE_B), D_ALL); + MMIO_D(SPRLINOFF(PIPE_B), D_ALL); + MMIO_D(SPRSTRIDE(PIPE_B), D_ALL); + MMIO_D(SPRPOS(PIPE_B), D_ALL); + MMIO_D(SPRSIZE(PIPE_B), D_ALL); + MMIO_D(SPRKEYVAL(PIPE_B), D_ALL); + MMIO_D(SPRKEYMSK(PIPE_B), D_ALL); + MMIO_DH(SPRSURF(PIPE_B), D_ALL, NULL, spr_surf_mmio_write); + MMIO_D(SPRKEYMAX(PIPE_B), D_ALL); + MMIO_D(SPROFFSET(PIPE_B), D_ALL); + MMIO_D(SPRSCALE(PIPE_B), D_ALL); + MMIO_D(SPRSURFLIVE(PIPE_B), D_ALL); + + MMIO_D(SPRCTL(PIPE_C), D_ALL); + MMIO_D(SPRLINOFF(PIPE_C), D_ALL); + MMIO_D(SPRSTRIDE(PIPE_C), D_ALL); + MMIO_D(SPRPOS(PIPE_C), D_ALL); + MMIO_D(SPRSIZE(PIPE_C), D_ALL); + MMIO_D(SPRKEYVAL(PIPE_C), D_ALL); + MMIO_D(SPRKEYMSK(PIPE_C), D_ALL); + MMIO_DH(SPRSURF(PIPE_C), D_ALL, NULL, spr_surf_mmio_write); + MMIO_D(SPRKEYMAX(PIPE_C), D_ALL); + MMIO_D(SPROFFSET(PIPE_C), D_ALL); + MMIO_D(SPRSCALE(PIPE_C), D_ALL); + MMIO_D(SPRSURFLIVE(PIPE_C), D_ALL); + + MMIO_F(LGC_PALETTE(PIPE_A, 0), 4 * 256, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(LGC_PALETTE(PIPE_B, 0), 4 * 256, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(LGC_PALETTE(PIPE_C, 0), 4 * 256, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(HTOTAL(TRANSCODER_A), D_ALL); + MMIO_D(HBLANK(TRANSCODER_A), D_ALL); + MMIO_D(HSYNC(TRANSCODER_A), D_ALL); + MMIO_D(VTOTAL(TRANSCODER_A), D_ALL); + MMIO_D(VBLANK(TRANSCODER_A), D_ALL); + MMIO_D(VSYNC(TRANSCODER_A), D_ALL); + MMIO_D(BCLRPAT(TRANSCODER_A), D_ALL); + MMIO_D(VSYNCSHIFT(TRANSCODER_A), D_ALL); + MMIO_D(PIPESRC(TRANSCODER_A), D_ALL); + + MMIO_D(HTOTAL(TRANSCODER_B), D_ALL); + MMIO_D(HBLANK(TRANSCODER_B), D_ALL); + MMIO_D(HSYNC(TRANSCODER_B), D_ALL); + MMIO_D(VTOTAL(TRANSCODER_B), D_ALL); + MMIO_D(VBLANK(TRANSCODER_B), D_ALL); + MMIO_D(VSYNC(TRANSCODER_B), D_ALL); + MMIO_D(BCLRPAT(TRANSCODER_B), D_ALL); + MMIO_D(VSYNCSHIFT(TRANSCODER_B), D_ALL); + MMIO_D(PIPESRC(TRANSCODER_B), D_ALL); + + MMIO_D(HTOTAL(TRANSCODER_C), D_ALL); + MMIO_D(HBLANK(TRANSCODER_C), D_ALL); + MMIO_D(HSYNC(TRANSCODER_C), D_ALL); + MMIO_D(VTOTAL(TRANSCODER_C), D_ALL); + MMIO_D(VBLANK(TRANSCODER_C), D_ALL); + MMIO_D(VSYNC(TRANSCODER_C), D_ALL); + MMIO_D(BCLRPAT(TRANSCODER_C), D_ALL); + MMIO_D(VSYNCSHIFT(TRANSCODER_C), D_ALL); + MMIO_D(PIPESRC(TRANSCODER_C), D_ALL); + + MMIO_D(HTOTAL(TRANSCODER_EDP), D_ALL); + MMIO_D(HBLANK(TRANSCODER_EDP), D_ALL); + MMIO_D(HSYNC(TRANSCODER_EDP), D_ALL); + MMIO_D(VTOTAL(TRANSCODER_EDP), D_ALL); + MMIO_D(VBLANK(TRANSCODER_EDP), D_ALL); + MMIO_D(VSYNC(TRANSCODER_EDP), D_ALL); + MMIO_D(BCLRPAT(TRANSCODER_EDP), D_ALL); + MMIO_D(VSYNCSHIFT(TRANSCODER_EDP), D_ALL); + + MMIO_D(PIPE_DATA_M1(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_DATA_N1(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_DATA_M2(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_DATA_N2(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_LINK_M1(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_LINK_N1(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_LINK_M2(TRANSCODER_A), D_ALL); + MMIO_D(PIPE_LINK_N2(TRANSCODER_A), D_ALL); + + MMIO_D(PIPE_DATA_M1(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_DATA_N1(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_DATA_M2(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_DATA_N2(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_LINK_M1(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_LINK_N1(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_LINK_M2(TRANSCODER_B), D_ALL); + MMIO_D(PIPE_LINK_N2(TRANSCODER_B), D_ALL); + + MMIO_D(PIPE_DATA_M1(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_DATA_N1(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_DATA_M2(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_DATA_N2(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_LINK_M1(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_LINK_N1(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_LINK_M2(TRANSCODER_C), D_ALL); + MMIO_D(PIPE_LINK_N2(TRANSCODER_C), D_ALL); + + MMIO_D(PIPE_DATA_M1(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_DATA_N1(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_DATA_M2(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_DATA_N2(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_LINK_M1(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_LINK_N1(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_LINK_M2(TRANSCODER_EDP), D_ALL); + MMIO_D(PIPE_LINK_N2(TRANSCODER_EDP), D_ALL); + + MMIO_D(PF_CTL(PIPE_A), D_ALL); + MMIO_D(PF_WIN_SZ(PIPE_A), D_ALL); + MMIO_D(PF_WIN_POS(PIPE_A), D_ALL); + MMIO_D(PF_VSCALE(PIPE_A), D_ALL); + MMIO_D(PF_HSCALE(PIPE_A), D_ALL); + + MMIO_D(PF_CTL(PIPE_B), D_ALL); + MMIO_D(PF_WIN_SZ(PIPE_B), D_ALL); + MMIO_D(PF_WIN_POS(PIPE_B), D_ALL); + MMIO_D(PF_VSCALE(PIPE_B), D_ALL); + MMIO_D(PF_HSCALE(PIPE_B), D_ALL); + + MMIO_D(PF_CTL(PIPE_C), D_ALL); + MMIO_D(PF_WIN_SZ(PIPE_C), D_ALL); + MMIO_D(PF_WIN_POS(PIPE_C), D_ALL); + MMIO_D(PF_VSCALE(PIPE_C), D_ALL); + MMIO_D(PF_HSCALE(PIPE_C), D_ALL); + + MMIO_D(WM0_PIPEA_ILK, D_ALL); + MMIO_D(WM0_PIPEB_ILK, D_ALL); + MMIO_D(WM0_PIPEC_IVB, D_ALL); + MMIO_D(WM1_LP_ILK, D_ALL); + MMIO_D(WM2_LP_ILK, D_ALL); + MMIO_D(WM3_LP_ILK, D_ALL); + MMIO_D(WM1S_LP_ILK, D_ALL); + MMIO_D(WM2S_LP_IVB, D_ALL); + MMIO_D(WM3S_LP_IVB, D_ALL); + + MMIO_D(BLC_PWM_CPU_CTL2, D_ALL); + MMIO_D(BLC_PWM_CPU_CTL, D_ALL); + MMIO_D(BLC_PWM_PCH_CTL1, D_ALL); + MMIO_D(BLC_PWM_PCH_CTL2, D_ALL); + + MMIO_D(0x48268, D_ALL); + + MMIO_F(PCH_GMBUS0, 4 * 4, 0, 0, 0, D_ALL, gmbus_mmio_read, + gmbus_mmio_write); + MMIO_F(PCH_GPIOA, 6 * 4, F_UNALIGN, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0xe4f00, 0x28, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_F(_PCH_DPB_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_PRE_SKL, NULL, + dp_aux_ch_ctl_mmio_write); + MMIO_F(_PCH_DPC_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_PRE_SKL, NULL, + dp_aux_ch_ctl_mmio_write); + MMIO_F(_PCH_DPD_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_PRE_SKL, NULL, + dp_aux_ch_ctl_mmio_write); + + MMIO_RO(PCH_ADPA, D_ALL, 0, ADPA_CRT_HOTPLUG_MONITOR_MASK, NULL, pch_adpa_mmio_write); + + MMIO_DH(_PCH_TRANSACONF, D_ALL, NULL, transconf_mmio_write); + MMIO_DH(_PCH_TRANSBCONF, D_ALL, NULL, transconf_mmio_write); + + MMIO_DH(FDI_RX_IIR(PIPE_A), D_ALL, NULL, fdi_rx_iir_mmio_write); + MMIO_DH(FDI_RX_IIR(PIPE_B), D_ALL, NULL, fdi_rx_iir_mmio_write); + MMIO_DH(FDI_RX_IIR(PIPE_C), D_ALL, NULL, fdi_rx_iir_mmio_write); + MMIO_DH(FDI_RX_IMR(PIPE_A), D_ALL, NULL, update_fdi_rx_iir_status); + MMIO_DH(FDI_RX_IMR(PIPE_B), D_ALL, NULL, update_fdi_rx_iir_status); + MMIO_DH(FDI_RX_IMR(PIPE_C), D_ALL, NULL, update_fdi_rx_iir_status); + MMIO_DH(FDI_RX_CTL(PIPE_A), D_ALL, NULL, update_fdi_rx_iir_status); + MMIO_DH(FDI_RX_CTL(PIPE_B), D_ALL, NULL, update_fdi_rx_iir_status); + MMIO_DH(FDI_RX_CTL(PIPE_C), D_ALL, NULL, update_fdi_rx_iir_status); + + MMIO_D(_PCH_TRANS_HTOTAL_A, D_ALL); + MMIO_D(_PCH_TRANS_HBLANK_A, D_ALL); + MMIO_D(_PCH_TRANS_HSYNC_A, D_ALL); + MMIO_D(_PCH_TRANS_VTOTAL_A, D_ALL); + MMIO_D(_PCH_TRANS_VBLANK_A, D_ALL); + MMIO_D(_PCH_TRANS_VSYNC_A, D_ALL); + MMIO_D(_PCH_TRANS_VSYNCSHIFT_A, D_ALL); + + MMIO_D(_PCH_TRANS_HTOTAL_B, D_ALL); + MMIO_D(_PCH_TRANS_HBLANK_B, D_ALL); + MMIO_D(_PCH_TRANS_HSYNC_B, D_ALL); + MMIO_D(_PCH_TRANS_VTOTAL_B, D_ALL); + MMIO_D(_PCH_TRANS_VBLANK_B, D_ALL); + MMIO_D(_PCH_TRANS_VSYNC_B, D_ALL); + MMIO_D(_PCH_TRANS_VSYNCSHIFT_B, D_ALL); + + MMIO_D(_PCH_TRANSA_DATA_M1, D_ALL); + MMIO_D(_PCH_TRANSA_DATA_N1, D_ALL); + MMIO_D(_PCH_TRANSA_DATA_M2, D_ALL); + MMIO_D(_PCH_TRANSA_DATA_N2, D_ALL); + MMIO_D(_PCH_TRANSA_LINK_M1, D_ALL); + MMIO_D(_PCH_TRANSA_LINK_N1, D_ALL); + MMIO_D(_PCH_TRANSA_LINK_M2, D_ALL); + MMIO_D(_PCH_TRANSA_LINK_N2, D_ALL); + + MMIO_D(TRANS_DP_CTL(PIPE_A), D_ALL); + MMIO_D(TRANS_DP_CTL(PIPE_B), D_ALL); + MMIO_D(TRANS_DP_CTL(PIPE_C), D_ALL); + + MMIO_D(TVIDEO_DIP_CTL(PIPE_A), D_ALL); + MMIO_D(TVIDEO_DIP_DATA(PIPE_A), D_ALL); + MMIO_D(TVIDEO_DIP_GCP(PIPE_A), D_ALL); + + MMIO_D(TVIDEO_DIP_CTL(PIPE_B), D_ALL); + MMIO_D(TVIDEO_DIP_DATA(PIPE_B), D_ALL); + MMIO_D(TVIDEO_DIP_GCP(PIPE_B), D_ALL); + + MMIO_D(TVIDEO_DIP_CTL(PIPE_C), D_ALL); + MMIO_D(TVIDEO_DIP_DATA(PIPE_C), D_ALL); + MMIO_D(TVIDEO_DIP_GCP(PIPE_C), D_ALL); + + MMIO_D(_FDI_RXA_MISC, D_ALL); + MMIO_D(_FDI_RXB_MISC, D_ALL); + MMIO_D(_FDI_RXA_TUSIZE1, D_ALL); + MMIO_D(_FDI_RXA_TUSIZE2, D_ALL); + MMIO_D(_FDI_RXB_TUSIZE1, D_ALL); + MMIO_D(_FDI_RXB_TUSIZE2, D_ALL); + + MMIO_DH(PCH_PP_CONTROL, D_ALL, NULL, pch_pp_control_mmio_write); + MMIO_D(PCH_PP_DIVISOR, D_ALL); + MMIO_D(PCH_PP_STATUS, D_ALL); + MMIO_D(PCH_LVDS, D_ALL); + MMIO_D(_PCH_DPLL_A, D_ALL); + MMIO_D(_PCH_DPLL_B, D_ALL); + MMIO_D(_PCH_FPA0, D_ALL); + MMIO_D(_PCH_FPA1, D_ALL); + MMIO_D(_PCH_FPB0, D_ALL); + MMIO_D(_PCH_FPB1, D_ALL); + MMIO_D(PCH_DREF_CONTROL, D_ALL); + MMIO_D(PCH_RAWCLK_FREQ, D_ALL); + MMIO_D(PCH_DPLL_SEL, D_ALL); + + MMIO_D(0x61208, D_ALL); + MMIO_D(0x6120c, D_ALL); + MMIO_D(PCH_PP_ON_DELAYS, D_ALL); + MMIO_D(PCH_PP_OFF_DELAYS, D_ALL); + + MMIO_DH(0xe651c, D_ALL, dpy_reg_mmio_read, NULL); + MMIO_DH(0xe661c, D_ALL, dpy_reg_mmio_read, NULL); + MMIO_DH(0xe671c, D_ALL, dpy_reg_mmio_read, NULL); + MMIO_DH(0xe681c, D_ALL, dpy_reg_mmio_read, NULL); + MMIO_DH(0xe6c04, D_ALL, dpy_reg_mmio_read_2, NULL); + MMIO_DH(0xe6e1c, D_ALL, dpy_reg_mmio_read_3, NULL); + + MMIO_RO(PCH_PORT_HOTPLUG, D_ALL, 0, + PORTA_HOTPLUG_STATUS_MASK + | PORTB_HOTPLUG_STATUS_MASK + | PORTC_HOTPLUG_STATUS_MASK + | PORTD_HOTPLUG_STATUS_MASK, + NULL, NULL); + + MMIO_DH(LCPLL_CTL, D_ALL, NULL, lcpll_ctl_mmio_write); + MMIO_D(FUSE_STRAP, D_ALL); + MMIO_D(DIGITAL_PORT_HOTPLUG_CNTRL, D_ALL); + + MMIO_D(DISP_ARB_CTL, D_ALL); + MMIO_D(DISP_ARB_CTL2, D_ALL); + + MMIO_D(ILK_DISPLAY_CHICKEN1, D_ALL); + MMIO_D(ILK_DISPLAY_CHICKEN2, D_ALL); + MMIO_D(ILK_DSPCLK_GATE_D, D_ALL); + + MMIO_D(SOUTH_CHICKEN1, D_ALL); + MMIO_DH(SOUTH_CHICKEN2, D_ALL, NULL, south_chicken2_mmio_write); + MMIO_D(_TRANSA_CHICKEN1, D_ALL); + MMIO_D(_TRANSB_CHICKEN1, D_ALL); + MMIO_D(SOUTH_DSPCLK_GATE_D, D_ALL); + MMIO_D(_TRANSA_CHICKEN2, D_ALL); + MMIO_D(_TRANSB_CHICKEN2, D_ALL); + + MMIO_D(ILK_DPFC_CB_BASE, D_ALL); + MMIO_D(ILK_DPFC_CONTROL, D_ALL); + MMIO_D(ILK_DPFC_RECOMP_CTL, D_ALL); + MMIO_D(ILK_DPFC_STATUS, D_ALL); + MMIO_D(ILK_DPFC_FENCE_YOFF, D_ALL); + MMIO_D(ILK_DPFC_CHICKEN, D_ALL); + MMIO_D(ILK_FBC_RT_BASE, D_ALL); + + MMIO_D(IPS_CTL, D_ALL); + + MMIO_D(PIPE_CSC_COEFF_RY_GY(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BY(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RU_GU(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BU(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RV_GV(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BV(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_MODE(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_HI(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_ME(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_LO(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_HI(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_ME(PIPE_A), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_LO(PIPE_A), D_ALL); + + MMIO_D(PIPE_CSC_COEFF_RY_GY(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BY(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RU_GU(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BU(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RV_GV(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BV(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_MODE(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_HI(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_ME(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_LO(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_HI(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_ME(PIPE_B), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_LO(PIPE_B), D_ALL); + + MMIO_D(PIPE_CSC_COEFF_RY_GY(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BY(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RU_GU(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BU(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_COEFF_RV_GV(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_COEFF_BV(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_MODE(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_HI(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_ME(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_PREOFF_LO(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_HI(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_ME(PIPE_C), D_ALL); + MMIO_D(PIPE_CSC_POSTOFF_LO(PIPE_C), D_ALL); + + MMIO_D(PREC_PAL_INDEX(PIPE_A), D_ALL); + MMIO_D(PREC_PAL_DATA(PIPE_A), D_ALL); + MMIO_F(PREC_PAL_GC_MAX(PIPE_A, 0), 4 * 3, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(PREC_PAL_INDEX(PIPE_B), D_ALL); + MMIO_D(PREC_PAL_DATA(PIPE_B), D_ALL); + MMIO_F(PREC_PAL_GC_MAX(PIPE_B, 0), 4 * 3, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(PREC_PAL_INDEX(PIPE_C), D_ALL); + MMIO_D(PREC_PAL_DATA(PIPE_C), D_ALL); + MMIO_F(PREC_PAL_GC_MAX(PIPE_C, 0), 4 * 3, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(0x60110, D_ALL); + MMIO_D(0x61110, D_ALL); + MMIO_F(0x70400, 0x40, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x71400, 0x40, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x72400, 0x40, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x70440, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + MMIO_F(0x71440, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + MMIO_F(0x72440, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + MMIO_F(0x7044c, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + MMIO_F(0x7144c, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + MMIO_F(0x7244c, 0xc, 0, 0, 0, D_PRE_SKL, NULL, NULL); + + MMIO_D(PIPE_WM_LINETIME(PIPE_A), D_ALL); + MMIO_D(PIPE_WM_LINETIME(PIPE_B), D_ALL); + MMIO_D(PIPE_WM_LINETIME(PIPE_C), D_ALL); + MMIO_D(SPLL_CTL, D_ALL); + MMIO_D(_WRPLL_CTL1, D_ALL); + MMIO_D(_WRPLL_CTL2, D_ALL); + MMIO_D(PORT_CLK_SEL(PORT_A), D_ALL); + MMIO_D(PORT_CLK_SEL(PORT_B), D_ALL); + MMIO_D(PORT_CLK_SEL(PORT_C), D_ALL); + MMIO_D(PORT_CLK_SEL(PORT_D), D_ALL); + MMIO_D(PORT_CLK_SEL(PORT_E), D_ALL); + MMIO_D(TRANS_CLK_SEL(TRANSCODER_A), D_ALL); + MMIO_D(TRANS_CLK_SEL(TRANSCODER_B), D_ALL); + MMIO_D(TRANS_CLK_SEL(TRANSCODER_C), D_ALL); + + MMIO_D(HSW_NDE_RSTWRN_OPT, D_ALL); + MMIO_D(0x46508, D_ALL); + + MMIO_D(0x49080, D_ALL); + MMIO_D(0x49180, D_ALL); + MMIO_D(0x49280, D_ALL); + + MMIO_F(0x49090, 0x14, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x49190, 0x14, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x49290, 0x14, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(GAMMA_MODE(PIPE_A), D_ALL); + MMIO_D(GAMMA_MODE(PIPE_B), D_ALL); + MMIO_D(GAMMA_MODE(PIPE_C), D_ALL); + + MMIO_D(PIPE_MULT(PIPE_A), D_ALL); + MMIO_D(PIPE_MULT(PIPE_B), D_ALL); + MMIO_D(PIPE_MULT(PIPE_C), D_ALL); + + MMIO_D(HSW_TVIDEO_DIP_CTL(TRANSCODER_A), D_ALL); + MMIO_D(HSW_TVIDEO_DIP_CTL(TRANSCODER_B), D_ALL); + MMIO_D(HSW_TVIDEO_DIP_CTL(TRANSCODER_C), D_ALL); + + MMIO_DH(SFUSE_STRAP, D_ALL, NULL, NULL); + MMIO_D(SBI_ADDR, D_ALL); + MMIO_DH(SBI_DATA, D_ALL, sbi_data_mmio_read, NULL); + MMIO_DH(SBI_CTL_STAT, D_ALL, NULL, sbi_ctl_mmio_write); + MMIO_D(PIXCLK_GATE, D_ALL); + + MMIO_F(_DPA_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_ALL, NULL, + dp_aux_ch_ctl_mmio_write); + + MMIO_DH(DDI_BUF_CTL(PORT_A), D_ALL, NULL, ddi_buf_ctl_mmio_write); + MMIO_DH(DDI_BUF_CTL(PORT_B), D_ALL, NULL, ddi_buf_ctl_mmio_write); + MMIO_DH(DDI_BUF_CTL(PORT_C), D_ALL, NULL, ddi_buf_ctl_mmio_write); + MMIO_DH(DDI_BUF_CTL(PORT_D), D_ALL, NULL, ddi_buf_ctl_mmio_write); + MMIO_DH(DDI_BUF_CTL(PORT_E), D_ALL, NULL, ddi_buf_ctl_mmio_write); + + MMIO_DH(DP_TP_CTL(PORT_A), D_ALL, NULL, dp_tp_ctl_mmio_write); + MMIO_DH(DP_TP_CTL(PORT_B), D_ALL, NULL, dp_tp_ctl_mmio_write); + MMIO_DH(DP_TP_CTL(PORT_C), D_ALL, NULL, dp_tp_ctl_mmio_write); + MMIO_DH(DP_TP_CTL(PORT_D), D_ALL, NULL, dp_tp_ctl_mmio_write); + MMIO_DH(DP_TP_CTL(PORT_E), D_ALL, NULL, dp_tp_ctl_mmio_write); + + MMIO_DH(DP_TP_STATUS(PORT_A), D_ALL, NULL, dp_tp_status_mmio_write); + MMIO_DH(DP_TP_STATUS(PORT_B), D_ALL, NULL, dp_tp_status_mmio_write); + MMIO_DH(DP_TP_STATUS(PORT_C), D_ALL, NULL, dp_tp_status_mmio_write); + MMIO_DH(DP_TP_STATUS(PORT_D), D_ALL, NULL, dp_tp_status_mmio_write); + MMIO_DH(DP_TP_STATUS(PORT_E), D_ALL, NULL, NULL); + + MMIO_F(_DDI_BUF_TRANS_A, 0x50, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x64e60, 0x50, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x64eC0, 0x50, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x64f20, 0x50, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x64f80, 0x50, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(HSW_AUD_CFG(PIPE_A), D_ALL); + MMIO_D(HSW_AUD_PIN_ELD_CP_VLD, D_ALL); + + MMIO_DH(_TRANS_DDI_FUNC_CTL_A, D_ALL, NULL, NULL); + MMIO_DH(_TRANS_DDI_FUNC_CTL_B, D_ALL, NULL, NULL); + MMIO_DH(_TRANS_DDI_FUNC_CTL_C, D_ALL, NULL, NULL); + MMIO_DH(_TRANS_DDI_FUNC_CTL_EDP, D_ALL, NULL, NULL); + + MMIO_D(_TRANSA_MSA_MISC, D_ALL); + MMIO_D(_TRANSB_MSA_MISC, D_ALL); + MMIO_D(_TRANSC_MSA_MISC, D_ALL); + MMIO_D(_TRANS_EDP_MSA_MISC, D_ALL); + + MMIO_DH(FORCEWAKE, D_ALL, NULL, NULL); + MMIO_D(FORCEWAKE_ACK, D_ALL); + MMIO_D(GEN6_GT_CORE_STATUS, D_ALL); + MMIO_D(GEN6_GT_THREAD_STATUS_REG, D_ALL); + MMIO_D(GTFIFODBG, D_ALL); + MMIO_D(GTFIFOCTL, D_ALL); + MMIO_DH(FORCEWAKE_MT, D_PRE_SKL, NULL, mul_force_wake_write); + MMIO_DH(FORCEWAKE_ACK_HSW, D_HSW | D_BDW, NULL, NULL); + MMIO_D(ECOBUS, D_ALL); + MMIO_DH(GEN6_RC_CONTROL, D_ALL, NULL, NULL); + MMIO_DH(GEN6_RC_STATE, D_ALL, NULL, NULL); + MMIO_D(GEN6_RPNSWREQ, D_ALL); + MMIO_D(GEN6_RC_VIDEO_FREQ, D_ALL); + MMIO_D(GEN6_RP_DOWN_TIMEOUT, D_ALL); + MMIO_D(GEN6_RP_INTERRUPT_LIMITS, D_ALL); + MMIO_D(GEN6_RPSTAT1, D_ALL); + MMIO_D(GEN6_RP_CONTROL, D_ALL); + MMIO_D(GEN6_RP_UP_THRESHOLD, D_ALL); + MMIO_D(GEN6_RP_DOWN_THRESHOLD, D_ALL); + MMIO_D(GEN6_RP_CUR_UP_EI, D_ALL); + MMIO_D(GEN6_RP_CUR_UP, D_ALL); + MMIO_D(GEN6_RP_PREV_UP, D_ALL); + MMIO_D(GEN6_RP_CUR_DOWN_EI, D_ALL); + MMIO_D(GEN6_RP_CUR_DOWN, D_ALL); + MMIO_D(GEN6_RP_PREV_DOWN, D_ALL); + MMIO_D(GEN6_RP_UP_EI, D_ALL); + MMIO_D(GEN6_RP_DOWN_EI, D_ALL); + MMIO_D(GEN6_RP_IDLE_HYSTERSIS, D_ALL); + MMIO_D(GEN6_RC1_WAKE_RATE_LIMIT, D_ALL); + MMIO_D(GEN6_RC6_WAKE_RATE_LIMIT, D_ALL); + MMIO_D(GEN6_RC6pp_WAKE_RATE_LIMIT, D_ALL); + MMIO_D(GEN6_RC_EVALUATION_INTERVAL, D_ALL); + MMIO_D(GEN6_RC_IDLE_HYSTERSIS, D_ALL); + MMIO_D(GEN6_RC_SLEEP, D_ALL); + MMIO_D(GEN6_RC1e_THRESHOLD, D_ALL); + MMIO_D(GEN6_RC6_THRESHOLD, D_ALL); + MMIO_D(GEN6_RC6p_THRESHOLD, D_ALL); + MMIO_D(GEN6_RC6pp_THRESHOLD, D_ALL); + MMIO_D(GEN6_PMINTRMSK, D_ALL); + MMIO_DH(HSW_PWR_WELL_BIOS, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + MMIO_DH(HSW_PWR_WELL_DRIVER, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + MMIO_DH(HSW_PWR_WELL_KVMR, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + MMIO_DH(HSW_PWR_WELL_DEBUG, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + MMIO_DH(HSW_PWR_WELL_CTL5, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + MMIO_DH(HSW_PWR_WELL_CTL6, D_HSW | D_BDW, NULL, power_well_ctl_mmio_write); + + MMIO_D(RSTDBYCTL, D_ALL); + + MMIO_DH(GEN6_GDRST, D_ALL, NULL, gdrst_mmio_write); + MMIO_F(FENCE_REG_GEN6_LO(0), 0x80, 0, 0, 0, D_ALL, fence_mmio_read, fence_mmio_write); + MMIO_F(VGT_PVINFO_PAGE, VGT_PVINFO_SIZE, F_UNALIGN, 0, 0, D_ALL, pvinfo_mmio_read, pvinfo_mmio_write); + MMIO_DH(CPU_VGACNTRL, D_ALL, NULL, vga_control_mmio_write); + + MMIO_F(MCHBAR_MIRROR_BASE_SNB, 0x40000, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(TILECTL, D_ALL); + + MMIO_D(GEN6_UCGCTL1, D_ALL); + MMIO_D(GEN6_UCGCTL2, D_ALL); + + MMIO_F(0x4f000, 0x90, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_D(GEN6_PCODE_MAILBOX, D_PRE_SKL); + MMIO_D(GEN6_PCODE_DATA, D_ALL); + MMIO_D(0x13812c, D_ALL); + MMIO_DH(GEN7_ERR_INT, D_ALL, NULL, NULL); + MMIO_D(HSW_EDRAM_CAP, D_ALL); + MMIO_D(HSW_IDICR, D_ALL); + MMIO_DH(GFX_FLSH_CNTL_GEN6, D_ALL, NULL, NULL); + + MMIO_D(0x3c, D_ALL); + MMIO_D(0x860, D_ALL); + MMIO_D(ECOSKPD, D_ALL); + MMIO_D(0x121d0, D_ALL); + MMIO_D(GEN6_BLITTER_ECOSKPD, D_ALL); + MMIO_D(0x41d0, D_ALL); + MMIO_D(GAC_ECO_BITS, D_ALL); + MMIO_D(0x6200, D_ALL); + MMIO_D(0x6204, D_ALL); + MMIO_D(0x6208, D_ALL); + MMIO_D(0x7118, D_ALL); + MMIO_D(0x7180, D_ALL); + MMIO_D(0x7408, D_ALL); + MMIO_D(0x7c00, D_ALL); + MMIO_D(GEN6_MBCTL, D_ALL); + MMIO_D(0x911c, D_ALL); + MMIO_D(0x9120, D_ALL); + MMIO_DFH(GEN7_UCGCTL4, D_ALL, F_CMD_ACCESS, NULL, NULL); + + MMIO_D(GAB_CTL, D_ALL); + MMIO_D(0x48800, D_ALL); + MMIO_D(0xce044, D_ALL); + MMIO_D(0xe6500, D_ALL); + MMIO_D(0xe6504, D_ALL); + MMIO_D(0xe6600, D_ALL); + MMIO_D(0xe6604, D_ALL); + MMIO_D(0xe6700, D_ALL); + MMIO_D(0xe6704, D_ALL); + MMIO_D(0xe6800, D_ALL); + MMIO_D(0xe6804, D_ALL); + MMIO_D(PCH_GMBUS4, D_ALL); + MMIO_D(PCH_GMBUS5, D_ALL); + + MMIO_D(0x902c, D_ALL); + MMIO_D(0xec008, D_ALL); + MMIO_D(0xec00c, D_ALL); + MMIO_D(0xec008 + 0x18, D_ALL); + MMIO_D(0xec00c + 0x18, D_ALL); + MMIO_D(0xec008 + 0x18 * 2, D_ALL); + MMIO_D(0xec00c + 0x18 * 2, D_ALL); + MMIO_D(0xec008 + 0x18 * 3, D_ALL); + MMIO_D(0xec00c + 0x18 * 3, D_ALL); + MMIO_D(0xec408, D_ALL); + MMIO_D(0xec40c, D_ALL); + MMIO_D(0xec408 + 0x18, D_ALL); + MMIO_D(0xec40c + 0x18, D_ALL); + MMIO_D(0xec408 + 0x18 * 2, D_ALL); + MMIO_D(0xec40c + 0x18 * 2, D_ALL); + MMIO_D(0xec408 + 0x18 * 3, D_ALL); + MMIO_D(0xec40c + 0x18 * 3, D_ALL); + MMIO_D(0xfc810, D_ALL); + MMIO_D(0xfc81c, D_ALL); + MMIO_D(0xfc828, D_ALL); + MMIO_D(0xfc834, D_ALL); + MMIO_D(0xfcc00, D_ALL); + MMIO_D(0xfcc0c, D_ALL); + MMIO_D(0xfcc18, D_ALL); + MMIO_D(0xfcc24, D_ALL); + MMIO_D(0xfd000, D_ALL); + MMIO_D(0xfd00c, D_ALL); + MMIO_D(0xfd018, D_ALL); + MMIO_D(0xfd024, D_ALL); + MMIO_D(0xfd034, D_ALL); + + MMIO_DH(FPGA_DBG, D_ALL, NULL, fpga_dbg_mmio_write); + MMIO_D(0x2054, D_ALL); + MMIO_D(0x12054, D_ALL); + MMIO_D(0x22054, D_ALL); + MMIO_D(0x1a054, D_ALL); + + MMIO_D(0x44070, D_ALL); + + MMIO_D(0x215c, D_HSW_PLUS); + MMIO_DFH(0x2178, D_ALL, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x217c, D_ALL, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x12178, D_ALL, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x1217c, D_ALL, F_CMD_ACCESS, NULL, NULL); + + MMIO_F(0x2290, 8, 0, 0, 0, D_HSW_PLUS, NULL, NULL); + MMIO_D(OACONTROL, D_HSW); + MMIO_D(0x2b00, D_BDW_PLUS); + MMIO_D(0x2360, D_BDW_PLUS); + MMIO_F(0x5200, 32, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x5240, 32, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(0x5280, 16, 0, 0, 0, D_ALL, NULL, NULL); + + MMIO_DFH(0x1c17c, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x1c178, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_D(BCS_SWCTRL, D_ALL); + + MMIO_F(HS_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(DS_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(IA_VERTICES_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(IA_PRIMITIVES_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(VS_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(GS_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(GS_PRIMITIVES_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(CL_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(CL_PRIMITIVES_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(PS_INVOCATION_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_F(PS_DEPTH_COUNT, 8, 0, 0, 0, D_ALL, NULL, NULL); + MMIO_DH(0x4260, D_BDW_PLUS, NULL, gvt_reg_tlb_control_handler); + MMIO_DH(0x4264, D_BDW_PLUS, NULL, gvt_reg_tlb_control_handler); + MMIO_DH(0x4268, D_BDW_PLUS, NULL, gvt_reg_tlb_control_handler); + MMIO_DH(0x426c, D_BDW_PLUS, NULL, gvt_reg_tlb_control_handler); + MMIO_DH(0x4270, D_BDW_PLUS, NULL, gvt_reg_tlb_control_handler); + MMIO_DFH(0x4094, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + + return 0; +} + +static int init_broadwell_mmio_info(struct intel_gvt *gvt) +{ + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ret; + + MMIO_DH(RING_IMR(GEN8_BSD2_RING_BASE), D_BDW_PLUS, NULL, + intel_vgpu_reg_imr_handler); + + MMIO_DH(GEN8_GT_IMR(0), D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_GT_IER(0), D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_GT_IIR(0), D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_GT_ISR(0), D_BDW_PLUS); + + MMIO_DH(GEN8_GT_IMR(1), D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_GT_IER(1), D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_GT_IIR(1), D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_GT_ISR(1), D_BDW_PLUS); + + MMIO_DH(GEN8_GT_IMR(2), D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_GT_IER(2), D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_GT_IIR(2), D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_GT_ISR(2), D_BDW_PLUS); + + MMIO_DH(GEN8_GT_IMR(3), D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_GT_IER(3), D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_GT_IIR(3), D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_GT_ISR(3), D_BDW_PLUS); + + MMIO_DH(GEN8_DE_PIPE_IMR(PIPE_A), D_BDW_PLUS, NULL, + intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_DE_PIPE_IER(PIPE_A), D_BDW_PLUS, NULL, + intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_DE_PIPE_IIR(PIPE_A), D_BDW_PLUS, NULL, + intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_DE_PIPE_ISR(PIPE_A), D_BDW_PLUS); + + MMIO_DH(GEN8_DE_PIPE_IMR(PIPE_B), D_BDW_PLUS, NULL, + intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_DE_PIPE_IER(PIPE_B), D_BDW_PLUS, NULL, + intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_DE_PIPE_IIR(PIPE_B), D_BDW_PLUS, NULL, + intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_DE_PIPE_ISR(PIPE_B), D_BDW_PLUS); + + MMIO_DH(GEN8_DE_PIPE_IMR(PIPE_C), D_BDW_PLUS, NULL, + intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_DE_PIPE_IER(PIPE_C), D_BDW_PLUS, NULL, + intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_DE_PIPE_IIR(PIPE_C), D_BDW_PLUS, NULL, + intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_DE_PIPE_ISR(PIPE_C), D_BDW_PLUS); + + MMIO_DH(GEN8_DE_PORT_IMR, D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_DE_PORT_IER, D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_DE_PORT_IIR, D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_DE_PORT_ISR, D_BDW_PLUS); + + MMIO_DH(GEN8_DE_MISC_IMR, D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_DE_MISC_IER, D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_DE_MISC_IIR, D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_DE_MISC_ISR, D_BDW_PLUS); + + MMIO_DH(GEN8_PCU_IMR, D_BDW_PLUS, NULL, intel_vgpu_reg_imr_handler); + MMIO_DH(GEN8_PCU_IER, D_BDW_PLUS, NULL, intel_vgpu_reg_ier_handler); + MMIO_DH(GEN8_PCU_IIR, D_BDW_PLUS, NULL, intel_vgpu_reg_iir_handler); + MMIO_D(GEN8_PCU_ISR, D_BDW_PLUS); + + MMIO_DH(GEN8_MASTER_IRQ, D_BDW_PLUS, NULL, + intel_vgpu_reg_master_irq_handler); + + MMIO_D(RING_HWSTAM(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_D(0x1c134, D_BDW_PLUS); + + MMIO_D(RING_TAIL(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_D(RING_HEAD(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_GM(RING_START(GEN8_BSD2_RING_BASE), D_BDW_PLUS, NULL, NULL); + MMIO_D(RING_CTL(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_D(RING_ACTHD(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_D(RING_ACTHD_UDW(GEN8_BSD2_RING_BASE), D_BDW_PLUS); + MMIO_DFH(0x1c29c, D_BDW_PLUS, F_MODE_MASK, NULL, ring_mode_mmio_write); + MMIO_DFH(RING_MI_MODE(GEN8_BSD2_RING_BASE), D_BDW_PLUS, F_MODE_MASK, + NULL, NULL); + MMIO_DFH(RING_INSTPM(GEN8_BSD2_RING_BASE), D_BDW_PLUS, F_MODE_MASK, + NULL, NULL); + MMIO_DFH(RING_TIMESTAMP(GEN8_BSD2_RING_BASE), D_BDW_PLUS, F_CMD_ACCESS, + ring_timestamp_mmio_read, NULL); + + MMIO_RING_D(RING_ACTHD_UDW, D_BDW_PLUS); + +#define RING_REG(base) (base + 0xd0) + MMIO_RING_F(RING_REG, 4, F_RO, 0, + ~_MASKED_BIT_ENABLE(RESET_CTL_REQUEST_RESET), D_BDW_PLUS, NULL, + ring_reset_ctl_write); + MMIO_F(RING_REG(GEN8_BSD2_RING_BASE), 4, F_RO, 0, + ~_MASKED_BIT_ENABLE(RESET_CTL_REQUEST_RESET), D_BDW_PLUS, NULL, + ring_reset_ctl_write); +#undef RING_REG + +#define RING_REG(base) (base + 0x230) + MMIO_RING_DFH(RING_REG, D_BDW_PLUS, 0, NULL, elsp_mmio_write); + MMIO_DH(RING_REG(GEN8_BSD2_RING_BASE), D_BDW_PLUS, NULL, elsp_mmio_write); +#undef RING_REG + +#define RING_REG(base) (base + 0x234) + MMIO_RING_F(RING_REG, 8, F_RO, 0, ~0, D_BDW_PLUS, NULL, NULL); + MMIO_F(RING_REG(GEN8_BSD2_RING_BASE), 4, F_RO, 0, ~0LL, D_BDW_PLUS, NULL, NULL); +#undef RING_REG + +#define RING_REG(base) (base + 0x244) + MMIO_RING_D(RING_REG, D_BDW_PLUS); + MMIO_D(RING_REG(GEN8_BSD2_RING_BASE), D_BDW_PLUS); +#undef RING_REG + +#define RING_REG(base) (base + 0x370) + MMIO_RING_F(RING_REG, 48, F_RO, 0, ~0, D_BDW_PLUS, NULL, NULL); + MMIO_F(RING_REG(GEN8_BSD2_RING_BASE), 48, F_RO, 0, ~0, D_BDW_PLUS, + NULL, NULL); +#undef RING_REG + +#define RING_REG(base) (base + 0x3a0) + MMIO_RING_DFH(RING_REG, D_BDW_PLUS, F_MODE_MASK, NULL, NULL); + MMIO_DFH(RING_REG(GEN8_BSD2_RING_BASE), D_BDW_PLUS, F_MODE_MASK, NULL, NULL); +#undef RING_REG + + MMIO_D(PIPEMISC(PIPE_A), D_BDW_PLUS); + MMIO_D(PIPEMISC(PIPE_B), D_BDW_PLUS); + MMIO_D(PIPEMISC(PIPE_C), D_BDW_PLUS); + MMIO_D(0x1c1d0, D_BDW_PLUS); + MMIO_D(GEN6_MBCUNIT_SNPCR, D_BDW_PLUS); + MMIO_D(GEN7_MISCCPCTL, D_BDW_PLUS); + MMIO_D(0x1c054, D_BDW_PLUS); + + MMIO_D(GEN8_PRIVATE_PAT_LO, D_BDW_PLUS); + MMIO_D(GEN8_PRIVATE_PAT_HI, D_BDW_PLUS); + + MMIO_D(GAMTARBMODE, D_BDW_PLUS); + +#define RING_REG(base) (base + 0x270) + MMIO_RING_F(RING_REG, 32, 0, 0, 0, D_BDW_PLUS, NULL, NULL); + MMIO_F(RING_REG(GEN8_BSD2_RING_BASE), 32, 0, 0, 0, D_BDW_PLUS, NULL, NULL); +#undef RING_REG + + MMIO_RING_GM(RING_HWS_PGA, D_BDW_PLUS, NULL, NULL); + MMIO_GM(0x1c080, D_BDW_PLUS, NULL, NULL); + + MMIO_DFH(HDC_CHICKEN0, D_BDW_PLUS, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + + MMIO_D(CHICKEN_PIPESL_1(PIPE_A), D_BDW); + MMIO_D(CHICKEN_PIPESL_1(PIPE_B), D_BDW); + MMIO_D(CHICKEN_PIPESL_1(PIPE_C), D_BDW); + + MMIO_D(WM_MISC, D_BDW); + MMIO_D(BDW_EDP_PSR_BASE, D_BDW); + + MMIO_D(0x66c00, D_BDW_PLUS); + MMIO_D(0x66c04, D_BDW_PLUS); + + MMIO_D(HSW_GTT_CACHE_EN, D_BDW_PLUS); + + MMIO_D(GEN8_EU_DISABLE0, D_BDW_PLUS); + MMIO_D(GEN8_EU_DISABLE1, D_BDW_PLUS); + MMIO_D(GEN8_EU_DISABLE2, D_BDW_PLUS); + + MMIO_D(0xfdc, D_BDW); + MMIO_DFH(GEN8_ROW_CHICKEN, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_D(GEN7_ROW_CHICKEN2, D_BDW_PLUS); + MMIO_D(GEN8_UCGCTL6, D_BDW_PLUS); + + MMIO_D(0xb1f0, D_BDW); + MMIO_D(0xb1c0, D_BDW); + MMIO_DFH(GEN8_L3SQCREG4, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_D(0xb100, D_BDW); + MMIO_D(0xb10c, D_BDW); + MMIO_D(0xb110, D_BDW); + + MMIO_DFH(0x24d0, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x24d4, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x24d8, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x24dc, D_BDW_PLUS, F_CMD_ACCESS, NULL, NULL); + + MMIO_D(0x83a4, D_BDW); + MMIO_D(GEN8_L3_LRA_1_GPGPU, D_BDW_PLUS); + + MMIO_D(0x8430, D_BDW); + + MMIO_D(0x110000, D_BDW_PLUS); + + MMIO_D(0x48400, D_BDW_PLUS); + + MMIO_D(0x6e570, D_BDW_PLUS); + MMIO_D(0x65f10, D_BDW_PLUS); + + MMIO_DFH(0xe194, D_BDW_PLUS, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0xe188, D_BDW_PLUS, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(HALF_SLICE_CHICKEN2, D_BDW_PLUS, F_MODE_MASK | F_CMD_ACCESS, NULL, NULL); + MMIO_DFH(0x2580, D_BDW_PLUS, F_MODE_MASK, NULL, NULL); + + MMIO_D(0x2248, D_BDW); + + return 0; +} + +static int init_skl_mmio_info(struct intel_gvt *gvt) +{ + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ret; + + MMIO_DH(FORCEWAKE_RENDER_GEN9, D_SKL_PLUS, NULL, mul_force_wake_write); + MMIO_DH(FORCEWAKE_ACK_RENDER_GEN9, D_SKL_PLUS, NULL, NULL); + MMIO_DH(FORCEWAKE_BLITTER_GEN9, D_SKL_PLUS, NULL, mul_force_wake_write); + MMIO_DH(FORCEWAKE_ACK_BLITTER_GEN9, D_SKL_PLUS, NULL, NULL); + MMIO_DH(FORCEWAKE_MEDIA_GEN9, D_SKL_PLUS, NULL, mul_force_wake_write); + MMIO_DH(FORCEWAKE_ACK_MEDIA_GEN9, D_SKL_PLUS, NULL, NULL); + + MMIO_F(_DPB_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_SKL, NULL, dp_aux_ch_ctl_mmio_write); + MMIO_F(_DPC_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_SKL, NULL, dp_aux_ch_ctl_mmio_write); + MMIO_F(_DPD_AUX_CH_CTL, 6 * 4, 0, 0, 0, D_SKL, NULL, dp_aux_ch_ctl_mmio_write); + + MMIO_D(HSW_PWR_WELL_BIOS, D_SKL); + MMIO_DH(HSW_PWR_WELL_DRIVER, D_SKL, NULL, skl_power_well_ctl_write); + + MMIO_DH(GEN6_PCODE_MAILBOX, D_SKL, NULL, mailbox_write); + MMIO_D(0xa210, D_SKL_PLUS); + MMIO_D(GEN9_MEDIA_PG_IDLE_HYSTERESIS, D_SKL_PLUS); + MMIO_D(GEN9_RENDER_PG_IDLE_HYSTERESIS, D_SKL_PLUS); + MMIO_DFH(GEN9_GAMT_ECO_REG_RW_IA, D_SKL_PLUS, F_CMD_ACCESS, NULL, NULL); + MMIO_DH(0x4ddc, D_SKL, NULL, skl_misc_ctl_write); + MMIO_DH(0x42080, D_SKL, NULL, skl_misc_ctl_write); + MMIO_D(0x45504, D_SKL); + MMIO_D(0x45520, D_SKL); + MMIO_D(0x46000, D_SKL); + MMIO_DH(0x46010, D_SKL, NULL, skl_lcpll_write); + MMIO_DH(0x46014, D_SKL, NULL, skl_lcpll_write); + MMIO_D(0x6C040, D_SKL); + MMIO_D(0x6C048, D_SKL); + MMIO_D(0x6C050, D_SKL); + MMIO_D(0x6C044, D_SKL); + MMIO_D(0x6C04C, D_SKL); + MMIO_D(0x6C054, D_SKL); + MMIO_D(0x6c058, D_SKL); + MMIO_D(0x6c05c, D_SKL); + MMIO_DH(0X6c060, D_SKL, dpll_status_read, NULL); + + MMIO_DH(SKL_PS_WIN_POS(PIPE_A, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_POS(PIPE_A, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_POS(PIPE_B, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_POS(PIPE_B, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_POS(PIPE_C, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_POS(PIPE_C, 1), D_SKL, NULL, pf_write); + + MMIO_DH(SKL_PS_WIN_SZ(PIPE_A, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_SZ(PIPE_A, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_SZ(PIPE_B, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_SZ(PIPE_B, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_SZ(PIPE_C, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_WIN_SZ(PIPE_C, 1), D_SKL, NULL, pf_write); + + MMIO_DH(SKL_PS_CTRL(PIPE_A, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_CTRL(PIPE_A, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_CTRL(PIPE_B, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_CTRL(PIPE_B, 1), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_CTRL(PIPE_C, 0), D_SKL, NULL, pf_write); + MMIO_DH(SKL_PS_CTRL(PIPE_C, 1), D_SKL, NULL, pf_write); + + MMIO_DH(PLANE_BUF_CFG(PIPE_A, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_A, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_A, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_A, 3), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_BUF_CFG(PIPE_B, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_B, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_B, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_B, 3), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_BUF_CFG(PIPE_C, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_C, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_C, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_BUF_CFG(PIPE_C, 3), D_SKL, NULL, NULL); + + MMIO_DH(CUR_BUF_CFG(PIPE_A), D_SKL, NULL, NULL); + MMIO_DH(CUR_BUF_CFG(PIPE_B), D_SKL, NULL, NULL); + MMIO_DH(CUR_BUF_CFG(PIPE_C), D_SKL, NULL, NULL); + + MMIO_F(PLANE_WM(PIPE_A, 0, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_A, 1, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_A, 2, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + + MMIO_F(PLANE_WM(PIPE_B, 0, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_B, 1, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_B, 2, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + + MMIO_F(PLANE_WM(PIPE_C, 0, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_C, 1, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(PLANE_WM(PIPE_C, 2, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + + MMIO_F(CUR_WM(PIPE_A, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(CUR_WM(PIPE_B, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_F(CUR_WM(PIPE_C, 0), 4 * 8, 0, 0, 0, D_SKL, NULL, NULL); + + MMIO_DH(PLANE_WM_TRANS(PIPE_A, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_A, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_A, 2), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_WM_TRANS(PIPE_B, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_B, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_B, 2), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_WM_TRANS(PIPE_C, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_C, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_WM_TRANS(PIPE_C, 2), D_SKL, NULL, NULL); + + MMIO_DH(CUR_WM_TRANS(PIPE_A), D_SKL, NULL, NULL); + MMIO_DH(CUR_WM_TRANS(PIPE_B), D_SKL, NULL, NULL); + MMIO_DH(CUR_WM_TRANS(PIPE_C), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_A, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_A, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_A, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_A, 3), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_B, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_B, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_B, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_B, 3), D_SKL, NULL, NULL); + + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_C, 0), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_C, 1), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_C, 2), D_SKL, NULL, NULL); + MMIO_DH(PLANE_NV12_BUF_CFG(PIPE_C, 3), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C0(PIPE_A, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_A, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_A, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_A, 4), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C0(PIPE_B, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_B, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_B, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_B, 4), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C0(PIPE_C, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_C, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_C, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C0(PIPE_C, 4), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C4(PIPE_A, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_A, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_A, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_A, 4), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C4(PIPE_B, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_B, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_B, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_B, 4), D_SKL, NULL, NULL); + + MMIO_DH(_REG_701C4(PIPE_C, 1), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_C, 2), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_C, 3), D_SKL, NULL, NULL); + MMIO_DH(_REG_701C4(PIPE_C, 4), D_SKL, NULL, NULL); + + MMIO_D(0x70380, D_SKL); + MMIO_D(0x71380, D_SKL); + MMIO_D(0x72380, D_SKL); + MMIO_D(0x7039c, D_SKL); + + MMIO_F(0x80000, 0x3000, 0, 0, 0, D_SKL, NULL, NULL); + MMIO_D(0x8f074, D_SKL); + MMIO_D(0x8f004, D_SKL); + MMIO_D(0x8f034, D_SKL); + + MMIO_D(0xb11c, D_SKL); + + MMIO_D(0x51000, D_SKL); + MMIO_D(0x6c00c, D_SKL); + + MMIO_F(0xc800, 0x7f8, F_CMD_ACCESS, 0, 0, D_SKL, NULL, NULL); + MMIO_F(0xb020, 0x80, F_CMD_ACCESS, 0, 0, D_SKL, NULL, NULL); + + MMIO_D(0xd08, D_SKL); + MMIO_D(0x20e0, D_SKL); + MMIO_D(0x20ec, D_SKL); + + /* TRTT */ + MMIO_D(0x4de0, D_SKL); + MMIO_D(0x4de4, D_SKL); + MMIO_D(0x4de8, D_SKL); + MMIO_D(0x4dec, D_SKL); + MMIO_D(0x4df0, D_SKL); + MMIO_DH(0x4df4, D_SKL, NULL, gen9_trtte_write); + MMIO_DH(0x4dfc, D_SKL, NULL, gen9_trtt_chicken_write); + + MMIO_D(0x45008, D_SKL); + + MMIO_D(0x46430, D_SKL); + + MMIO_D(0x46520, D_SKL); + + MMIO_D(0xc403c, D_SKL); + MMIO_D(0xb004, D_SKL); + MMIO_DH(DMA_CTRL, D_SKL_PLUS, NULL, dma_ctrl_write); + + MMIO_D(0x65900, D_SKL); + MMIO_D(0x1082c0, D_SKL); + MMIO_D(0x4068, D_SKL); + MMIO_D(0x67054, D_SKL); + MMIO_D(0x6e560, D_SKL); + MMIO_D(0x6e554, D_SKL); + MMIO_D(0x2b20, D_SKL); + MMIO_D(0x65f00, D_SKL); + MMIO_D(0x65f08, D_SKL); + MMIO_D(0x320f0, D_SKL); + + MMIO_D(_REG_VCS2_EXCC, D_SKL); + MMIO_D(0x70034, D_SKL); + MMIO_D(0x71034, D_SKL); + MMIO_D(0x72034, D_SKL); + + MMIO_D(_PLANE_KEYVAL_1(PIPE_A), D_SKL); + MMIO_D(_PLANE_KEYVAL_1(PIPE_B), D_SKL); + MMIO_D(_PLANE_KEYVAL_1(PIPE_C), D_SKL); + MMIO_D(_PLANE_KEYMSK_1(PIPE_A), D_SKL); + MMIO_D(_PLANE_KEYMSK_1(PIPE_B), D_SKL); + MMIO_D(_PLANE_KEYMSK_1(PIPE_C), D_SKL); + + MMIO_D(0x44500, D_SKL); + return 0; +} + +/** + * intel_gvt_find_mmio_info - find MMIO information entry by aligned offset + * @gvt: GVT device + * @offset: register offset + * + * This function is used to find the MMIO information entry from hash table + * + * Returns: + * pointer to MMIO information entry, NULL if not exists + */ +struct intel_gvt_mmio_info *intel_gvt_find_mmio_info(struct intel_gvt *gvt, + unsigned int offset) +{ + struct intel_gvt_mmio_info *e; + + WARN_ON(!IS_ALIGNED(offset, 4)); + + hash_for_each_possible(gvt->mmio.mmio_info_table, e, node, offset) { + if (e->offset == offset) + return e; + } + return NULL; +} + +/** + * intel_gvt_clean_mmio_info - clean up MMIO information table for GVT device + * @gvt: GVT device + * + * This function is called at the driver unloading stage, to clean up the MMIO + * information table of GVT device + * + */ +void intel_gvt_clean_mmio_info(struct intel_gvt *gvt) +{ + struct hlist_node *tmp; + struct intel_gvt_mmio_info *e; + int i; + + hash_for_each_safe(gvt->mmio.mmio_info_table, i, tmp, e, node) + kfree(e); + + vfree(gvt->mmio.mmio_attribute); + gvt->mmio.mmio_attribute = NULL; +} + +/** + * intel_gvt_setup_mmio_info - setup MMIO information table for GVT device + * @gvt: GVT device + * + * This function is called at the initialization stage, to setup the MMIO + * information table for GVT device + * + * Returns: + * zero on success, negative if failed. + */ +int intel_gvt_setup_mmio_info(struct intel_gvt *gvt) +{ + struct intel_gvt_device_info *info = &gvt->device_info; + struct drm_i915_private *dev_priv = gvt->dev_priv; + int ret; + + gvt->mmio.mmio_attribute = vzalloc(info->mmio_size); + if (!gvt->mmio.mmio_attribute) + return -ENOMEM; + + ret = init_generic_mmio_info(gvt); + if (ret) + goto err; + + if (IS_BROADWELL(dev_priv)) { + ret = init_broadwell_mmio_info(gvt); + if (ret) + goto err; + } else if (IS_SKYLAKE(dev_priv)) { + ret = init_broadwell_mmio_info(gvt); + if (ret) + goto err; + ret = init_skl_mmio_info(gvt); + if (ret) + goto err; + } + return 0; +err: + intel_gvt_clean_mmio_info(gvt); + return ret; +} + +/** + * intel_gvt_mmio_set_accessed - mark a MMIO has been accessed + * @gvt: a GVT device + * @offset: register offset + * + */ +void intel_gvt_mmio_set_accessed(struct intel_gvt *gvt, unsigned int offset) +{ + gvt->mmio.mmio_attribute[offset >> 2] |= + F_ACCESSED; +} + +/** + * intel_gvt_mmio_is_cmd_accessed - mark a MMIO could be accessed by command + * @gvt: a GVT device + * @offset: register offset + * + */ +bool intel_gvt_mmio_is_cmd_access(struct intel_gvt *gvt, + unsigned int offset) +{ + return gvt->mmio.mmio_attribute[offset >> 2] & + F_CMD_ACCESS; +} + +/** + * intel_gvt_mmio_is_unalign - mark a MMIO could be accessed unaligned + * @gvt: a GVT device + * @offset: register offset + * + */ +bool intel_gvt_mmio_is_unalign(struct intel_gvt *gvt, + unsigned int offset) +{ + return gvt->mmio.mmio_attribute[offset >> 2] & + F_UNALIGN; +} + +/** + * intel_gvt_mmio_set_cmd_accessed - mark a MMIO has been accessed by command + * @gvt: a GVT device + * @offset: register offset + * + */ +void intel_gvt_mmio_set_cmd_accessed(struct intel_gvt *gvt, + unsigned int offset) +{ + gvt->mmio.mmio_attribute[offset >> 2] |= + F_CMD_ACCESSED; +} + +/** + * intel_gvt_mmio_has_mode_mask - if a MMIO has a mode mask + * @gvt: a GVT device + * @offset: register offset + * + * Returns: + * True if a MMIO has a mode mask in its higher 16 bits, false if it isn't. + * + */ +bool intel_gvt_mmio_has_mode_mask(struct intel_gvt *gvt, unsigned int offset) +{ + return gvt->mmio.mmio_attribute[offset >> 2] & + F_MODE_MASK; +} + +/** + * intel_vgpu_default_mmio_read - default MMIO read handler + * @vgpu: a vGPU + * @offset: access offset + * @p_data: data return buffer + * @bytes: access data length + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_default_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + read_vreg(vgpu, offset, p_data, bytes); + return 0; +} + +/** + * intel_t_default_mmio_write - default MMIO write handler + * @vgpu: a vGPU + * @offset: access offset + * @p_data: write data buffer + * @bytes: access data length + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_default_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes) +{ + write_vreg(vgpu, offset, p_data, bytes); + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/hypercall.h b/drivers/gpu/drm/i915/gvt/hypercall.h index 254df8bf1f35..30e543f5a703 100644 --- a/drivers/gpu/drm/i915/gvt/hypercall.h +++ b/drivers/gpu/drm/i915/gvt/hypercall.h @@ -19,6 +19,15 @@ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Dexuan Cui + * Jike Song <jike.song@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * */ #ifndef _GVT_HYPERCALL_H_ @@ -30,6 +39,23 @@ */ struct intel_gvt_mpt { int (*detect_host)(void); + int (*host_init)(struct device *dev, void *gvt, const void *ops); + void (*host_exit)(struct device *dev, void *gvt); + int (*attach_vgpu)(void *vgpu, unsigned long *handle); + void (*detach_vgpu)(unsigned long handle); + int (*inject_msi)(unsigned long handle, u32 addr, u16 data); + unsigned long (*from_virt_to_mfn)(void *p); + int (*set_wp_page)(unsigned long handle, u64 gfn); + int (*unset_wp_page)(unsigned long handle, u64 gfn); + int (*read_gpa)(unsigned long handle, unsigned long gpa, void *buf, + unsigned long len); + int (*write_gpa)(unsigned long handle, unsigned long gpa, void *buf, + unsigned long len); + unsigned long (*gfn_to_mfn)(unsigned long handle, unsigned long gfn); + int (*map_gfn_to_mfn)(unsigned long handle, unsigned long gfn, + unsigned long mfn, unsigned int nr, bool map); + int (*set_trap_area)(unsigned long handle, u64 start, u64 end, + bool map); }; extern struct intel_gvt_mpt xengt_mpt; diff --git a/drivers/gpu/drm/i915/gvt/interrupt.c b/drivers/gpu/drm/i915/gvt/interrupt.c new file mode 100644 index 000000000000..f7be02ac4be1 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/interrupt.c @@ -0,0 +1,741 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Min he <min.he@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +/* common offset among interrupt control registers */ +#define regbase_to_isr(base) (base) +#define regbase_to_imr(base) (base + 0x4) +#define regbase_to_iir(base) (base + 0x8) +#define regbase_to_ier(base) (base + 0xC) + +#define iir_to_regbase(iir) (iir - 0x8) +#define ier_to_regbase(ier) (ier - 0xC) + +#define get_event_virt_handler(irq, e) (irq->events[e].v_handler) +#define get_irq_info(irq, e) (irq->events[e].info) + +#define irq_to_gvt(irq) \ + container_of(irq, struct intel_gvt, irq) + +static void update_upstream_irq(struct intel_vgpu *vgpu, + struct intel_gvt_irq_info *info); + +static const char * const irq_name[INTEL_GVT_EVENT_MAX] = { + [RCS_MI_USER_INTERRUPT] = "Render CS MI USER INTERRUPT", + [RCS_DEBUG] = "Render EU debug from SVG", + [RCS_MMIO_SYNC_FLUSH] = "Render MMIO sync flush status", + [RCS_CMD_STREAMER_ERR] = "Render CS error interrupt", + [RCS_PIPE_CONTROL] = "Render PIPE CONTROL notify", + [RCS_WATCHDOG_EXCEEDED] = "Render CS Watchdog counter exceeded", + [RCS_PAGE_DIRECTORY_FAULT] = "Render page directory faults", + [RCS_AS_CONTEXT_SWITCH] = "Render AS Context Switch Interrupt", + + [VCS_MI_USER_INTERRUPT] = "Video CS MI USER INTERRUPT", + [VCS_MMIO_SYNC_FLUSH] = "Video MMIO sync flush status", + [VCS_CMD_STREAMER_ERR] = "Video CS error interrupt", + [VCS_MI_FLUSH_DW] = "Video MI FLUSH DW notify", + [VCS_WATCHDOG_EXCEEDED] = "Video CS Watchdog counter exceeded", + [VCS_PAGE_DIRECTORY_FAULT] = "Video page directory faults", + [VCS_AS_CONTEXT_SWITCH] = "Video AS Context Switch Interrupt", + [VCS2_MI_USER_INTERRUPT] = "VCS2 Video CS MI USER INTERRUPT", + [VCS2_MI_FLUSH_DW] = "VCS2 Video MI FLUSH DW notify", + [VCS2_AS_CONTEXT_SWITCH] = "VCS2 Context Switch Interrupt", + + [BCS_MI_USER_INTERRUPT] = "Blitter CS MI USER INTERRUPT", + [BCS_MMIO_SYNC_FLUSH] = "Billter MMIO sync flush status", + [BCS_CMD_STREAMER_ERR] = "Blitter CS error interrupt", + [BCS_MI_FLUSH_DW] = "Blitter MI FLUSH DW notify", + [BCS_PAGE_DIRECTORY_FAULT] = "Blitter page directory faults", + [BCS_AS_CONTEXT_SWITCH] = "Blitter AS Context Switch Interrupt", + + [VECS_MI_FLUSH_DW] = "Video Enhanced Streamer MI FLUSH DW notify", + [VECS_AS_CONTEXT_SWITCH] = "VECS Context Switch Interrupt", + + [PIPE_A_FIFO_UNDERRUN] = "Pipe A FIFO underrun", + [PIPE_A_CRC_ERR] = "Pipe A CRC error", + [PIPE_A_CRC_DONE] = "Pipe A CRC done", + [PIPE_A_VSYNC] = "Pipe A vsync", + [PIPE_A_LINE_COMPARE] = "Pipe A line compare", + [PIPE_A_ODD_FIELD] = "Pipe A odd field", + [PIPE_A_EVEN_FIELD] = "Pipe A even field", + [PIPE_A_VBLANK] = "Pipe A vblank", + [PIPE_B_FIFO_UNDERRUN] = "Pipe B FIFO underrun", + [PIPE_B_CRC_ERR] = "Pipe B CRC error", + [PIPE_B_CRC_DONE] = "Pipe B CRC done", + [PIPE_B_VSYNC] = "Pipe B vsync", + [PIPE_B_LINE_COMPARE] = "Pipe B line compare", + [PIPE_B_ODD_FIELD] = "Pipe B odd field", + [PIPE_B_EVEN_FIELD] = "Pipe B even field", + [PIPE_B_VBLANK] = "Pipe B vblank", + [PIPE_C_VBLANK] = "Pipe C vblank", + [DPST_PHASE_IN] = "DPST phase in event", + [DPST_HISTOGRAM] = "DPST histogram event", + [GSE] = "GSE", + [DP_A_HOTPLUG] = "DP A Hotplug", + [AUX_CHANNEL_A] = "AUX Channel A", + [PERF_COUNTER] = "Performance counter", + [POISON] = "Poison", + [GTT_FAULT] = "GTT fault", + [PRIMARY_A_FLIP_DONE] = "Primary Plane A flip done", + [PRIMARY_B_FLIP_DONE] = "Primary Plane B flip done", + [PRIMARY_C_FLIP_DONE] = "Primary Plane C flip done", + [SPRITE_A_FLIP_DONE] = "Sprite Plane A flip done", + [SPRITE_B_FLIP_DONE] = "Sprite Plane B flip done", + [SPRITE_C_FLIP_DONE] = "Sprite Plane C flip done", + + [PCU_THERMAL] = "PCU Thermal Event", + [PCU_PCODE2DRIVER_MAILBOX] = "PCU pcode2driver mailbox event", + + [FDI_RX_INTERRUPTS_TRANSCODER_A] = "FDI RX Interrupts Combined A", + [AUDIO_CP_CHANGE_TRANSCODER_A] = "Audio CP Change Transcoder A", + [AUDIO_CP_REQUEST_TRANSCODER_A] = "Audio CP Request Transcoder A", + [FDI_RX_INTERRUPTS_TRANSCODER_B] = "FDI RX Interrupts Combined B", + [AUDIO_CP_CHANGE_TRANSCODER_B] = "Audio CP Change Transcoder B", + [AUDIO_CP_REQUEST_TRANSCODER_B] = "Audio CP Request Transcoder B", + [FDI_RX_INTERRUPTS_TRANSCODER_C] = "FDI RX Interrupts Combined C", + [AUDIO_CP_CHANGE_TRANSCODER_C] = "Audio CP Change Transcoder C", + [AUDIO_CP_REQUEST_TRANSCODER_C] = "Audio CP Request Transcoder C", + [ERR_AND_DBG] = "South Error and Debug Interupts Combined", + [GMBUS] = "Gmbus", + [SDVO_B_HOTPLUG] = "SDVO B hotplug", + [CRT_HOTPLUG] = "CRT Hotplug", + [DP_B_HOTPLUG] = "DisplayPort/HDMI/DVI B Hotplug", + [DP_C_HOTPLUG] = "DisplayPort/HDMI/DVI C Hotplug", + [DP_D_HOTPLUG] = "DisplayPort/HDMI/DVI D Hotplug", + [AUX_CHANNEL_B] = "AUX Channel B", + [AUX_CHANNEL_C] = "AUX Channel C", + [AUX_CHANNEL_D] = "AUX Channel D", + [AUDIO_POWER_STATE_CHANGE_B] = "Audio Power State change Port B", + [AUDIO_POWER_STATE_CHANGE_C] = "Audio Power State change Port C", + [AUDIO_POWER_STATE_CHANGE_D] = "Audio Power State change Port D", + + [INTEL_GVT_EVENT_RESERVED] = "RESERVED EVENTS!!!", +}; + +static inline struct intel_gvt_irq_info *regbase_to_irq_info( + struct intel_gvt *gvt, + unsigned int reg) +{ + struct intel_gvt_irq *irq = &gvt->irq; + int i; + + for_each_set_bit(i, irq->irq_info_bitmap, INTEL_GVT_IRQ_INFO_MAX) { + if (i915_mmio_reg_offset(irq->info[i]->reg_base) == reg) + return irq->info[i]; + } + + return NULL; +} + +/** + * intel_vgpu_reg_imr_handler - Generic IMR register emulation write handler + * @vgpu: a vGPU + * @reg: register offset written by guest + * @p_data: register data written by guest + * @bytes: register data length + * + * This function is used to emulate the generic IMR register bit change + * behavior. + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_vgpu_reg_imr_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_irq_ops *ops = gvt->irq.ops; + u32 changed, masked, unmasked; + u32 imr = *(u32 *)p_data; + + gvt_dbg_irq("write IMR %x with val %x\n", + reg, imr); + + gvt_dbg_irq("old vIMR %x\n", vgpu_vreg(vgpu, reg)); + + /* figure out newly masked/unmasked bits */ + changed = vgpu_vreg(vgpu, reg) ^ imr; + masked = (vgpu_vreg(vgpu, reg) & changed) ^ changed; + unmasked = masked ^ changed; + + gvt_dbg_irq("changed %x, masked %x, unmasked %x\n", + changed, masked, unmasked); + + vgpu_vreg(vgpu, reg) = imr; + + ops->check_pending_irq(vgpu); + gvt_dbg_irq("IRQ: new vIMR %x\n", vgpu_vreg(vgpu, reg)); + return 0; +} + +/** + * intel_vgpu_reg_master_irq_handler - master IRQ write emulation handler + * @vgpu: a vGPU + * @reg: register offset written by guest + * @p_data: register data written by guest + * @bytes: register data length + * + * This function is used to emulate the master IRQ register on gen8+. + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_vgpu_reg_master_irq_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_irq_ops *ops = gvt->irq.ops; + u32 changed, enabled, disabled; + u32 ier = *(u32 *)p_data; + u32 virtual_ier = vgpu_vreg(vgpu, reg); + + gvt_dbg_irq("write master irq reg %x with val %x\n", + reg, ier); + + gvt_dbg_irq("old vreg %x\n", vgpu_vreg(vgpu, reg)); + + /* + * GEN8_MASTER_IRQ is a special irq register, + * only bit 31 is allowed to be modified + * and treated as an IER bit. + */ + ier &= GEN8_MASTER_IRQ_CONTROL; + virtual_ier &= GEN8_MASTER_IRQ_CONTROL; + vgpu_vreg(vgpu, reg) &= ~GEN8_MASTER_IRQ_CONTROL; + vgpu_vreg(vgpu, reg) |= ier; + + /* figure out newly enabled/disable bits */ + changed = virtual_ier ^ ier; + enabled = (virtual_ier & changed) ^ changed; + disabled = enabled ^ changed; + + gvt_dbg_irq("changed %x, enabled %x, disabled %x\n", + changed, enabled, disabled); + + ops->check_pending_irq(vgpu); + gvt_dbg_irq("new vreg %x\n", vgpu_vreg(vgpu, reg)); + return 0; +} + +/** + * intel_vgpu_reg_ier_handler - Generic IER write emulation handler + * @vgpu: a vGPU + * @reg: register offset written by guest + * @p_data: register data written by guest + * @bytes: register data length + * + * This function is used to emulate the generic IER register behavior. + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_vgpu_reg_ier_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_irq_ops *ops = gvt->irq.ops; + struct intel_gvt_irq_info *info; + u32 changed, enabled, disabled; + u32 ier = *(u32 *)p_data; + + gvt_dbg_irq("write IER %x with val %x\n", + reg, ier); + + gvt_dbg_irq("old vIER %x\n", vgpu_vreg(vgpu, reg)); + + /* figure out newly enabled/disable bits */ + changed = vgpu_vreg(vgpu, reg) ^ ier; + enabled = (vgpu_vreg(vgpu, reg) & changed) ^ changed; + disabled = enabled ^ changed; + + gvt_dbg_irq("changed %x, enabled %x, disabled %x\n", + changed, enabled, disabled); + vgpu_vreg(vgpu, reg) = ier; + + info = regbase_to_irq_info(gvt, ier_to_regbase(reg)); + if (WARN_ON(!info)) + return -EINVAL; + + if (info->has_upstream_irq) + update_upstream_irq(vgpu, info); + + ops->check_pending_irq(vgpu); + gvt_dbg_irq("new vIER %x\n", vgpu_vreg(vgpu, reg)); + return 0; +} + +/** + * intel_vgpu_reg_iir_handler - Generic IIR write emulation handler + * @vgpu: a vGPU + * @reg: register offset written by guest + * @p_data: register data written by guest + * @bytes: register data length + * + * This function is used to emulate the generic IIR register behavior. + * + * Returns: + * Zero on success, negative error code if failed. + * + */ +int intel_vgpu_reg_iir_handler(struct intel_vgpu *vgpu, unsigned int reg, + void *p_data, unsigned int bytes) +{ + struct intel_gvt_irq_info *info = regbase_to_irq_info(vgpu->gvt, + iir_to_regbase(reg)); + u32 iir = *(u32 *)p_data; + + gvt_dbg_irq("write IIR %x with val %x\n", reg, iir); + + if (WARN_ON(!info)) + return -EINVAL; + + vgpu_vreg(vgpu, reg) &= ~iir; + + if (info->has_upstream_irq) + update_upstream_irq(vgpu, info); + return 0; +} + +static struct intel_gvt_irq_map gen8_irq_map[] = { + { INTEL_GVT_IRQ_INFO_MASTER, 0, INTEL_GVT_IRQ_INFO_GT0, 0xffff }, + { INTEL_GVT_IRQ_INFO_MASTER, 1, INTEL_GVT_IRQ_INFO_GT0, 0xffff0000 }, + { INTEL_GVT_IRQ_INFO_MASTER, 2, INTEL_GVT_IRQ_INFO_GT1, 0xffff }, + { INTEL_GVT_IRQ_INFO_MASTER, 3, INTEL_GVT_IRQ_INFO_GT1, 0xffff0000 }, + { INTEL_GVT_IRQ_INFO_MASTER, 4, INTEL_GVT_IRQ_INFO_GT2, 0xffff }, + { INTEL_GVT_IRQ_INFO_MASTER, 6, INTEL_GVT_IRQ_INFO_GT3, 0xffff }, + { INTEL_GVT_IRQ_INFO_MASTER, 16, INTEL_GVT_IRQ_INFO_DE_PIPE_A, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 17, INTEL_GVT_IRQ_INFO_DE_PIPE_B, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 18, INTEL_GVT_IRQ_INFO_DE_PIPE_C, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 20, INTEL_GVT_IRQ_INFO_DE_PORT, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 22, INTEL_GVT_IRQ_INFO_DE_MISC, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 23, INTEL_GVT_IRQ_INFO_PCH, ~0 }, + { INTEL_GVT_IRQ_INFO_MASTER, 30, INTEL_GVT_IRQ_INFO_PCU, ~0 }, + { -1, -1, ~0 }, +}; + +static void update_upstream_irq(struct intel_vgpu *vgpu, + struct intel_gvt_irq_info *info) +{ + struct intel_gvt_irq *irq = &vgpu->gvt->irq; + struct intel_gvt_irq_map *map = irq->irq_map; + struct intel_gvt_irq_info *up_irq_info = NULL; + u32 set_bits = 0; + u32 clear_bits = 0; + int bit; + u32 val = vgpu_vreg(vgpu, + regbase_to_iir(i915_mmio_reg_offset(info->reg_base))) + & vgpu_vreg(vgpu, + regbase_to_ier(i915_mmio_reg_offset(info->reg_base))); + + if (!info->has_upstream_irq) + return; + + for (map = irq->irq_map; map->up_irq_bit != -1; map++) { + if (info->group != map->down_irq_group) + continue; + + if (!up_irq_info) + up_irq_info = irq->info[map->up_irq_group]; + else + WARN_ON(up_irq_info != irq->info[map->up_irq_group]); + + bit = map->up_irq_bit; + + if (val & map->down_irq_bitmask) + set_bits |= (1 << bit); + else + clear_bits |= (1 << bit); + } + + WARN_ON(!up_irq_info); + + if (up_irq_info->group == INTEL_GVT_IRQ_INFO_MASTER) { + u32 isr = i915_mmio_reg_offset(up_irq_info->reg_base); + + vgpu_vreg(vgpu, isr) &= ~clear_bits; + vgpu_vreg(vgpu, isr) |= set_bits; + } else { + u32 iir = regbase_to_iir( + i915_mmio_reg_offset(up_irq_info->reg_base)); + u32 imr = regbase_to_imr( + i915_mmio_reg_offset(up_irq_info->reg_base)); + + vgpu_vreg(vgpu, iir) |= (set_bits & ~vgpu_vreg(vgpu, imr)); + } + + if (up_irq_info->has_upstream_irq) + update_upstream_irq(vgpu, up_irq_info); +} + +static void init_irq_map(struct intel_gvt_irq *irq) +{ + struct intel_gvt_irq_map *map; + struct intel_gvt_irq_info *up_info, *down_info; + int up_bit; + + for (map = irq->irq_map; map->up_irq_bit != -1; map++) { + up_info = irq->info[map->up_irq_group]; + up_bit = map->up_irq_bit; + down_info = irq->info[map->down_irq_group]; + + set_bit(up_bit, up_info->downstream_irq_bitmap); + down_info->has_upstream_irq = true; + + gvt_dbg_irq("[up] grp %d bit %d -> [down] grp %d bitmask %x\n", + up_info->group, up_bit, + down_info->group, map->down_irq_bitmask); + } +} + +/* =======================vEvent injection===================== */ +static int inject_virtual_interrupt(struct intel_vgpu *vgpu) +{ + return intel_gvt_hypervisor_inject_msi(vgpu); +} + +static void propagate_event(struct intel_gvt_irq *irq, + enum intel_gvt_event_type event, struct intel_vgpu *vgpu) +{ + struct intel_gvt_irq_info *info; + unsigned int reg_base; + int bit; + + info = get_irq_info(irq, event); + if (WARN_ON(!info)) + return; + + reg_base = i915_mmio_reg_offset(info->reg_base); + bit = irq->events[event].bit; + + if (!test_bit(bit, (void *)&vgpu_vreg(vgpu, + regbase_to_imr(reg_base)))) { + gvt_dbg_irq("set bit (%d) for (%s) for vgpu (%d)\n", + bit, irq_name[event], vgpu->id); + set_bit(bit, (void *)&vgpu_vreg(vgpu, + regbase_to_iir(reg_base))); + } +} + +/* =======================vEvent Handlers===================== */ +static void handle_default_event_virt(struct intel_gvt_irq *irq, + enum intel_gvt_event_type event, struct intel_vgpu *vgpu) +{ + if (!vgpu->irq.irq_warn_once[event]) { + gvt_dbg_core("vgpu%d: IRQ receive event %d (%s)\n", + vgpu->id, event, irq_name[event]); + vgpu->irq.irq_warn_once[event] = true; + } + propagate_event(irq, event, vgpu); +} + +/* =====================GEN specific logic======================= */ +/* GEN8 interrupt routines. */ + +#define DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(regname, regbase) \ +static struct intel_gvt_irq_info gen8_##regname##_info = { \ + .name = #regname"-IRQ", \ + .reg_base = (regbase), \ + .bit_to_event = {[0 ... INTEL_GVT_IRQ_BITWIDTH-1] = \ + INTEL_GVT_EVENT_RESERVED}, \ +} + +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(gt0, GEN8_GT_ISR(0)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(gt1, GEN8_GT_ISR(1)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(gt2, GEN8_GT_ISR(2)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(gt3, GEN8_GT_ISR(3)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(de_pipe_a, GEN8_DE_PIPE_ISR(PIPE_A)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(de_pipe_b, GEN8_DE_PIPE_ISR(PIPE_B)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(de_pipe_c, GEN8_DE_PIPE_ISR(PIPE_C)); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(de_port, GEN8_DE_PORT_ISR); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(de_misc, GEN8_DE_MISC_ISR); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(pcu, GEN8_PCU_ISR); +DEFINE_GVT_GEN8_INTEL_GVT_IRQ_INFO(master, GEN8_MASTER_IRQ); + +static struct intel_gvt_irq_info gvt_base_pch_info = { + .name = "PCH-IRQ", + .reg_base = SDEISR, + .bit_to_event = {[0 ... INTEL_GVT_IRQ_BITWIDTH-1] = + INTEL_GVT_EVENT_RESERVED}, +}; + +static void gen8_check_pending_irq(struct intel_vgpu *vgpu) +{ + struct intel_gvt_irq *irq = &vgpu->gvt->irq; + int i; + + if (!(vgpu_vreg(vgpu, i915_mmio_reg_offset(GEN8_MASTER_IRQ)) & + GEN8_MASTER_IRQ_CONTROL)) + return; + + for_each_set_bit(i, irq->irq_info_bitmap, INTEL_GVT_IRQ_INFO_MAX) { + struct intel_gvt_irq_info *info = irq->info[i]; + u32 reg_base; + + if (!info->has_upstream_irq) + continue; + + reg_base = i915_mmio_reg_offset(info->reg_base); + if ((vgpu_vreg(vgpu, regbase_to_iir(reg_base)) + & vgpu_vreg(vgpu, regbase_to_ier(reg_base)))) + update_upstream_irq(vgpu, info); + } + + if (vgpu_vreg(vgpu, i915_mmio_reg_offset(GEN8_MASTER_IRQ)) + & ~GEN8_MASTER_IRQ_CONTROL) + inject_virtual_interrupt(vgpu); +} + +static void gen8_init_irq( + struct intel_gvt_irq *irq) +{ + struct intel_gvt *gvt = irq_to_gvt(irq); + +#define SET_BIT_INFO(s, b, e, i) \ + do { \ + s->events[e].bit = b; \ + s->events[e].info = s->info[i]; \ + s->info[i]->bit_to_event[b] = e;\ + } while (0) + +#define SET_IRQ_GROUP(s, g, i) \ + do { \ + s->info[g] = i; \ + (i)->group = g; \ + set_bit(g, s->irq_info_bitmap); \ + } while (0) + + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_MASTER, &gen8_master_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_GT0, &gen8_gt0_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_GT1, &gen8_gt1_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_GT2, &gen8_gt2_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_GT3, &gen8_gt3_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_DE_PIPE_A, &gen8_de_pipe_a_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_DE_PIPE_B, &gen8_de_pipe_b_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_DE_PIPE_C, &gen8_de_pipe_c_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_DE_PORT, &gen8_de_port_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_DE_MISC, &gen8_de_misc_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_PCU, &gen8_pcu_info); + SET_IRQ_GROUP(irq, INTEL_GVT_IRQ_INFO_PCH, &gvt_base_pch_info); + + /* GEN8 level 2 interrupts. */ + + /* GEN8 interrupt GT0 events */ + SET_BIT_INFO(irq, 0, RCS_MI_USER_INTERRUPT, INTEL_GVT_IRQ_INFO_GT0); + SET_BIT_INFO(irq, 4, RCS_PIPE_CONTROL, INTEL_GVT_IRQ_INFO_GT0); + SET_BIT_INFO(irq, 8, RCS_AS_CONTEXT_SWITCH, INTEL_GVT_IRQ_INFO_GT0); + + SET_BIT_INFO(irq, 16, BCS_MI_USER_INTERRUPT, INTEL_GVT_IRQ_INFO_GT0); + SET_BIT_INFO(irq, 20, BCS_MI_FLUSH_DW, INTEL_GVT_IRQ_INFO_GT0); + SET_BIT_INFO(irq, 24, BCS_AS_CONTEXT_SWITCH, INTEL_GVT_IRQ_INFO_GT0); + + /* GEN8 interrupt GT1 events */ + SET_BIT_INFO(irq, 0, VCS_MI_USER_INTERRUPT, INTEL_GVT_IRQ_INFO_GT1); + SET_BIT_INFO(irq, 4, VCS_MI_FLUSH_DW, INTEL_GVT_IRQ_INFO_GT1); + SET_BIT_INFO(irq, 8, VCS_AS_CONTEXT_SWITCH, INTEL_GVT_IRQ_INFO_GT1); + + if (HAS_BSD2(gvt->dev_priv)) { + SET_BIT_INFO(irq, 16, VCS2_MI_USER_INTERRUPT, + INTEL_GVT_IRQ_INFO_GT1); + SET_BIT_INFO(irq, 20, VCS2_MI_FLUSH_DW, + INTEL_GVT_IRQ_INFO_GT1); + SET_BIT_INFO(irq, 24, VCS2_AS_CONTEXT_SWITCH, + INTEL_GVT_IRQ_INFO_GT1); + } + + /* GEN8 interrupt GT3 events */ + SET_BIT_INFO(irq, 0, VECS_MI_USER_INTERRUPT, INTEL_GVT_IRQ_INFO_GT3); + SET_BIT_INFO(irq, 4, VECS_MI_FLUSH_DW, INTEL_GVT_IRQ_INFO_GT3); + SET_BIT_INFO(irq, 8, VECS_AS_CONTEXT_SWITCH, INTEL_GVT_IRQ_INFO_GT3); + + SET_BIT_INFO(irq, 0, PIPE_A_VBLANK, INTEL_GVT_IRQ_INFO_DE_PIPE_A); + SET_BIT_INFO(irq, 0, PIPE_B_VBLANK, INTEL_GVT_IRQ_INFO_DE_PIPE_B); + SET_BIT_INFO(irq, 0, PIPE_C_VBLANK, INTEL_GVT_IRQ_INFO_DE_PIPE_C); + + /* GEN8 interrupt DE PORT events */ + SET_BIT_INFO(irq, 0, AUX_CHANNEL_A, INTEL_GVT_IRQ_INFO_DE_PORT); + SET_BIT_INFO(irq, 3, DP_A_HOTPLUG, INTEL_GVT_IRQ_INFO_DE_PORT); + + /* GEN8 interrupt DE MISC events */ + SET_BIT_INFO(irq, 0, GSE, INTEL_GVT_IRQ_INFO_DE_MISC); + + /* PCH events */ + SET_BIT_INFO(irq, 17, GMBUS, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 19, CRT_HOTPLUG, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 21, DP_B_HOTPLUG, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 22, DP_C_HOTPLUG, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 23, DP_D_HOTPLUG, INTEL_GVT_IRQ_INFO_PCH); + + if (IS_BROADWELL(gvt->dev_priv)) { + SET_BIT_INFO(irq, 25, AUX_CHANNEL_B, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 26, AUX_CHANNEL_C, INTEL_GVT_IRQ_INFO_PCH); + SET_BIT_INFO(irq, 27, AUX_CHANNEL_D, INTEL_GVT_IRQ_INFO_PCH); + + SET_BIT_INFO(irq, 4, PRIMARY_A_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_A); + SET_BIT_INFO(irq, 5, SPRITE_A_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_A); + + SET_BIT_INFO(irq, 4, PRIMARY_B_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_B); + SET_BIT_INFO(irq, 5, SPRITE_B_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_B); + + SET_BIT_INFO(irq, 4, PRIMARY_C_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_C); + SET_BIT_INFO(irq, 5, SPRITE_C_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_C); + } else if (IS_SKYLAKE(gvt->dev_priv)) { + SET_BIT_INFO(irq, 25, AUX_CHANNEL_B, INTEL_GVT_IRQ_INFO_DE_PORT); + SET_BIT_INFO(irq, 26, AUX_CHANNEL_C, INTEL_GVT_IRQ_INFO_DE_PORT); + SET_BIT_INFO(irq, 27, AUX_CHANNEL_D, INTEL_GVT_IRQ_INFO_DE_PORT); + + SET_BIT_INFO(irq, 3, PRIMARY_A_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_A); + SET_BIT_INFO(irq, 3, PRIMARY_B_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_B); + SET_BIT_INFO(irq, 3, PRIMARY_C_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_C); + } + + /* GEN8 interrupt PCU events */ + SET_BIT_INFO(irq, 24, PCU_THERMAL, INTEL_GVT_IRQ_INFO_PCU); + SET_BIT_INFO(irq, 25, PCU_PCODE2DRIVER_MAILBOX, INTEL_GVT_IRQ_INFO_PCU); +} + +static struct intel_gvt_irq_ops gen8_irq_ops = { + .init_irq = gen8_init_irq, + .check_pending_irq = gen8_check_pending_irq, +}; + +/** + * intel_vgpu_trigger_virtual_event - Trigger a virtual event for a vGPU + * @vgpu: a vGPU + * @event: interrupt event + * + * This function is used to trigger a virtual interrupt event for vGPU. + * The caller provides the event to be triggered, the framework itself + * will emulate the IRQ register bit change. + * + */ +void intel_vgpu_trigger_virtual_event(struct intel_vgpu *vgpu, + enum intel_gvt_event_type event) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_irq *irq = &gvt->irq; + gvt_event_virt_handler_t handler; + struct intel_gvt_irq_ops *ops = gvt->irq.ops; + + handler = get_event_virt_handler(irq, event); + WARN_ON(!handler); + + handler(irq, event, vgpu); + + ops->check_pending_irq(vgpu); +} + +static void init_events( + struct intel_gvt_irq *irq) +{ + int i; + + for (i = 0; i < INTEL_GVT_EVENT_MAX; i++) { + irq->events[i].info = NULL; + irq->events[i].v_handler = handle_default_event_virt; + } +} + +static enum hrtimer_restart vblank_timer_fn(struct hrtimer *data) +{ + struct intel_gvt_vblank_timer *vblank_timer; + struct intel_gvt_irq *irq; + struct intel_gvt *gvt; + + vblank_timer = container_of(data, struct intel_gvt_vblank_timer, timer); + irq = container_of(vblank_timer, struct intel_gvt_irq, vblank_timer); + gvt = container_of(irq, struct intel_gvt, irq); + + intel_gvt_request_service(gvt, INTEL_GVT_REQUEST_EMULATE_VBLANK); + hrtimer_add_expires_ns(&vblank_timer->timer, vblank_timer->period); + return HRTIMER_RESTART; +} + +/** + * intel_gvt_clean_irq - clean up GVT-g IRQ emulation subsystem + * @gvt: a GVT device + * + * This function is called at driver unloading stage, to clean up GVT-g IRQ + * emulation subsystem. + * + */ +void intel_gvt_clean_irq(struct intel_gvt *gvt) +{ + struct intel_gvt_irq *irq = &gvt->irq; + + hrtimer_cancel(&irq->vblank_timer.timer); +} + +#define VBLNAK_TIMER_PERIOD 16000000 + +/** + * intel_gvt_init_irq - initialize GVT-g IRQ emulation subsystem + * @gvt: a GVT device + * + * This function is called at driver loading stage, to initialize the GVT-g IRQ + * emulation subsystem. + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_gvt_init_irq(struct intel_gvt *gvt) +{ + struct intel_gvt_irq *irq = &gvt->irq; + struct intel_gvt_vblank_timer *vblank_timer = &irq->vblank_timer; + + gvt_dbg_core("init irq framework\n"); + + if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) { + irq->ops = &gen8_irq_ops; + irq->irq_map = gen8_irq_map; + } else { + WARN_ON(1); + return -ENODEV; + } + + /* common event initialization */ + init_events(irq); + + /* gen specific initialization */ + irq->ops->init_irq(irq); + + init_irq_map(irq); + + hrtimer_init(&vblank_timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); + vblank_timer->timer.function = vblank_timer_fn; + vblank_timer->period = VBLNAK_TIMER_PERIOD; + + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/interrupt.h b/drivers/gpu/drm/i915/gvt/interrupt.h new file mode 100644 index 000000000000..5313fb1b33e1 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/interrupt.h @@ -0,0 +1,233 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Min he <min.he@intel.com> + * + */ + +#ifndef _GVT_INTERRUPT_H_ +#define _GVT_INTERRUPT_H_ + +enum intel_gvt_event_type { + RCS_MI_USER_INTERRUPT = 0, + RCS_DEBUG, + RCS_MMIO_SYNC_FLUSH, + RCS_CMD_STREAMER_ERR, + RCS_PIPE_CONTROL, + RCS_L3_PARITY_ERR, + RCS_WATCHDOG_EXCEEDED, + RCS_PAGE_DIRECTORY_FAULT, + RCS_AS_CONTEXT_SWITCH, + RCS_MONITOR_BUFF_HALF_FULL, + + VCS_MI_USER_INTERRUPT, + VCS_MMIO_SYNC_FLUSH, + VCS_CMD_STREAMER_ERR, + VCS_MI_FLUSH_DW, + VCS_WATCHDOG_EXCEEDED, + VCS_PAGE_DIRECTORY_FAULT, + VCS_AS_CONTEXT_SWITCH, + + VCS2_MI_USER_INTERRUPT, + VCS2_MI_FLUSH_DW, + VCS2_AS_CONTEXT_SWITCH, + + BCS_MI_USER_INTERRUPT, + BCS_MMIO_SYNC_FLUSH, + BCS_CMD_STREAMER_ERR, + BCS_MI_FLUSH_DW, + BCS_PAGE_DIRECTORY_FAULT, + BCS_AS_CONTEXT_SWITCH, + + VECS_MI_USER_INTERRUPT, + VECS_MI_FLUSH_DW, + VECS_AS_CONTEXT_SWITCH, + + PIPE_A_FIFO_UNDERRUN, + PIPE_B_FIFO_UNDERRUN, + PIPE_A_CRC_ERR, + PIPE_B_CRC_ERR, + PIPE_A_CRC_DONE, + PIPE_B_CRC_DONE, + PIPE_A_ODD_FIELD, + PIPE_B_ODD_FIELD, + PIPE_A_EVEN_FIELD, + PIPE_B_EVEN_FIELD, + PIPE_A_LINE_COMPARE, + PIPE_B_LINE_COMPARE, + PIPE_C_LINE_COMPARE, + PIPE_A_VBLANK, + PIPE_B_VBLANK, + PIPE_C_VBLANK, + PIPE_A_VSYNC, + PIPE_B_VSYNC, + PIPE_C_VSYNC, + PRIMARY_A_FLIP_DONE, + PRIMARY_B_FLIP_DONE, + PRIMARY_C_FLIP_DONE, + SPRITE_A_FLIP_DONE, + SPRITE_B_FLIP_DONE, + SPRITE_C_FLIP_DONE, + + PCU_THERMAL, + PCU_PCODE2DRIVER_MAILBOX, + + DPST_PHASE_IN, + DPST_HISTOGRAM, + GSE, + DP_A_HOTPLUG, + AUX_CHANNEL_A, + PERF_COUNTER, + POISON, + GTT_FAULT, + ERROR_INTERRUPT_COMBINED, + + FDI_RX_INTERRUPTS_TRANSCODER_A, + AUDIO_CP_CHANGE_TRANSCODER_A, + AUDIO_CP_REQUEST_TRANSCODER_A, + FDI_RX_INTERRUPTS_TRANSCODER_B, + AUDIO_CP_CHANGE_TRANSCODER_B, + AUDIO_CP_REQUEST_TRANSCODER_B, + FDI_RX_INTERRUPTS_TRANSCODER_C, + AUDIO_CP_CHANGE_TRANSCODER_C, + AUDIO_CP_REQUEST_TRANSCODER_C, + ERR_AND_DBG, + GMBUS, + SDVO_B_HOTPLUG, + CRT_HOTPLUG, + DP_B_HOTPLUG, + DP_C_HOTPLUG, + DP_D_HOTPLUG, + AUX_CHANNEL_B, + AUX_CHANNEL_C, + AUX_CHANNEL_D, + AUDIO_POWER_STATE_CHANGE_B, + AUDIO_POWER_STATE_CHANGE_C, + AUDIO_POWER_STATE_CHANGE_D, + + INTEL_GVT_EVENT_RESERVED, + INTEL_GVT_EVENT_MAX, +}; + +struct intel_gvt_irq; +struct intel_gvt; + +typedef void (*gvt_event_virt_handler_t)(struct intel_gvt_irq *irq, + enum intel_gvt_event_type event, struct intel_vgpu *vgpu); + +struct intel_gvt_irq_ops { + void (*init_irq)(struct intel_gvt_irq *irq); + void (*check_pending_irq)(struct intel_vgpu *vgpu); +}; + +/* the list of physical interrupt control register groups */ +enum intel_gvt_irq_type { + INTEL_GVT_IRQ_INFO_GT, + INTEL_GVT_IRQ_INFO_DPY, + INTEL_GVT_IRQ_INFO_PCH, + INTEL_GVT_IRQ_INFO_PM, + + INTEL_GVT_IRQ_INFO_MASTER, + INTEL_GVT_IRQ_INFO_GT0, + INTEL_GVT_IRQ_INFO_GT1, + INTEL_GVT_IRQ_INFO_GT2, + INTEL_GVT_IRQ_INFO_GT3, + INTEL_GVT_IRQ_INFO_DE_PIPE_A, + INTEL_GVT_IRQ_INFO_DE_PIPE_B, + INTEL_GVT_IRQ_INFO_DE_PIPE_C, + INTEL_GVT_IRQ_INFO_DE_PORT, + INTEL_GVT_IRQ_INFO_DE_MISC, + INTEL_GVT_IRQ_INFO_AUD, + INTEL_GVT_IRQ_INFO_PCU, + + INTEL_GVT_IRQ_INFO_MAX, +}; + +#define INTEL_GVT_IRQ_BITWIDTH 32 + +/* device specific interrupt bit definitions */ +struct intel_gvt_irq_info { + char *name; + i915_reg_t reg_base; + enum intel_gvt_event_type bit_to_event[INTEL_GVT_IRQ_BITWIDTH]; + unsigned long warned; + int group; + DECLARE_BITMAP(downstream_irq_bitmap, INTEL_GVT_IRQ_BITWIDTH); + bool has_upstream_irq; +}; + +/* per-event information */ +struct intel_gvt_event_info { + int bit; /* map to register bit */ + int policy; /* forwarding policy */ + struct intel_gvt_irq_info *info; /* register info */ + gvt_event_virt_handler_t v_handler; /* for v_event */ +}; + +struct intel_gvt_irq_map { + int up_irq_group; + int up_irq_bit; + int down_irq_group; + u32 down_irq_bitmask; +}; + +struct intel_gvt_vblank_timer { + struct hrtimer timer; + u64 period; +}; + +/* structure containing device specific IRQ state */ +struct intel_gvt_irq { + struct intel_gvt_irq_ops *ops; + struct intel_gvt_irq_info *info[INTEL_GVT_IRQ_INFO_MAX]; + DECLARE_BITMAP(irq_info_bitmap, INTEL_GVT_IRQ_INFO_MAX); + struct intel_gvt_event_info events[INTEL_GVT_EVENT_MAX]; + DECLARE_BITMAP(pending_events, INTEL_GVT_EVENT_MAX); + struct intel_gvt_irq_map *irq_map; + struct intel_gvt_vblank_timer vblank_timer; +}; + +int intel_gvt_init_irq(struct intel_gvt *gvt); +void intel_gvt_clean_irq(struct intel_gvt *gvt); + +void intel_vgpu_trigger_virtual_event(struct intel_vgpu *vgpu, + enum intel_gvt_event_type event); + +int intel_vgpu_reg_iir_handler(struct intel_vgpu *vgpu, unsigned int reg, + void *p_data, unsigned int bytes); +int intel_vgpu_reg_ier_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes); +int intel_vgpu_reg_master_irq_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes); +int intel_vgpu_reg_imr_handler(struct intel_vgpu *vgpu, + unsigned int reg, void *p_data, unsigned int bytes); + +int gvt_ring_id_to_pipe_control_notify_event(int ring_id); +int gvt_ring_id_to_mi_flush_dw_event(int ring_id); +int gvt_ring_id_to_mi_user_interrupt_event(int ring_id); + +#endif /* _GVT_INTERRUPT_H_ */ diff --git a/drivers/gpu/drm/i915/gvt/kvmgt.c b/drivers/gpu/drm/i915/gvt/kvmgt.c new file mode 100644 index 000000000000..dc0365033157 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/kvmgt.c @@ -0,0 +1,597 @@ +/* + * KVMGT - the implementation of Intel mediated pass-through framework for KVM + * + * Copyright(c) 2014-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Kevin Tian <kevin.tian@intel.com> + * Jike Song <jike.song@intel.com> + * Xiaoguang Chen <xiaoguang.chen@intel.com> + */ + +#include <linux/init.h> +#include <linux/device.h> +#include <linux/mm.h> +#include <linux/types.h> +#include <linux/list.h> +#include <linux/rbtree.h> +#include <linux/spinlock.h> +#include <linux/eventfd.h> +#include <linux/uuid.h> +#include <linux/kvm_host.h> +#include <linux/vfio.h> + +#include "i915_drv.h" +#include "gvt.h" + +static inline long kvmgt_pin_pages(struct device *dev, unsigned long *user_pfn, + long npage, int prot, unsigned long *phys_pfn) +{ + return 0; +} +static inline long kvmgt_unpin_pages(struct device *dev, unsigned long *pfn, + long npage) +{ + return 0; +} + +static const struct intel_gvt_ops *intel_gvt_ops; + + +/* helper macros copied from vfio-pci */ +#define VFIO_PCI_OFFSET_SHIFT 40 +#define VFIO_PCI_OFFSET_TO_INDEX(off) (off >> VFIO_PCI_OFFSET_SHIFT) +#define VFIO_PCI_INDEX_TO_OFFSET(index) ((u64)(index) << VFIO_PCI_OFFSET_SHIFT) +#define VFIO_PCI_OFFSET_MASK (((u64)(1) << VFIO_PCI_OFFSET_SHIFT) - 1) + +struct vfio_region { + u32 type; + u32 subtype; + size_t size; + u32 flags; +}; + +struct kvmgt_pgfn { + gfn_t gfn; + struct hlist_node hnode; +}; + +struct kvmgt_guest_info { + struct kvm *kvm; + struct intel_vgpu *vgpu; + struct kvm_page_track_notifier_node track_node; +#define NR_BKT (1 << 18) + struct hlist_head ptable[NR_BKT]; +#undef NR_BKT +}; + +struct gvt_dma { + struct rb_node node; + gfn_t gfn; + kvm_pfn_t pfn; +}; + +static struct gvt_dma *__gvt_cache_find(struct intel_vgpu *vgpu, gfn_t gfn) +{ + struct rb_node *node = vgpu->vdev.cache.rb_node; + struct gvt_dma *ret = NULL; + + while (node) { + struct gvt_dma *itr = rb_entry(node, struct gvt_dma, node); + + if (gfn < itr->gfn) + node = node->rb_left; + else if (gfn > itr->gfn) + node = node->rb_right; + else { + ret = itr; + goto out; + } + } + +out: + return ret; +} + +static kvm_pfn_t gvt_cache_find(struct intel_vgpu *vgpu, gfn_t gfn) +{ + struct gvt_dma *entry; + + mutex_lock(&vgpu->vdev.cache_lock); + entry = __gvt_cache_find(vgpu, gfn); + mutex_unlock(&vgpu->vdev.cache_lock); + + return entry == NULL ? 0 : entry->pfn; +} + +static void gvt_cache_add(struct intel_vgpu *vgpu, gfn_t gfn, kvm_pfn_t pfn) +{ + struct gvt_dma *new, *itr; + struct rb_node **link = &vgpu->vdev.cache.rb_node, *parent = NULL; + + new = kzalloc(sizeof(struct gvt_dma), GFP_KERNEL); + if (!new) + return; + + new->gfn = gfn; + new->pfn = pfn; + + mutex_lock(&vgpu->vdev.cache_lock); + while (*link) { + parent = *link; + itr = rb_entry(parent, struct gvt_dma, node); + + if (gfn == itr->gfn) + goto out; + else if (gfn < itr->gfn) + link = &parent->rb_left; + else + link = &parent->rb_right; + } + + rb_link_node(&new->node, parent, link); + rb_insert_color(&new->node, &vgpu->vdev.cache); + mutex_unlock(&vgpu->vdev.cache_lock); + return; + +out: + mutex_unlock(&vgpu->vdev.cache_lock); + kfree(new); +} + +static void __gvt_cache_remove_entry(struct intel_vgpu *vgpu, + struct gvt_dma *entry) +{ + rb_erase(&entry->node, &vgpu->vdev.cache); + kfree(entry); +} + +static void gvt_cache_remove(struct intel_vgpu *vgpu, gfn_t gfn) +{ + struct device *dev = vgpu->vdev.mdev; + struct gvt_dma *this; + unsigned long pfn; + + mutex_lock(&vgpu->vdev.cache_lock); + this = __gvt_cache_find(vgpu, gfn); + if (!this) { + mutex_unlock(&vgpu->vdev.cache_lock); + return; + } + + pfn = this->pfn; + WARN_ON((kvmgt_unpin_pages(dev, &pfn, 1) != 1)); + __gvt_cache_remove_entry(vgpu, this); + mutex_unlock(&vgpu->vdev.cache_lock); +} + +static void gvt_cache_init(struct intel_vgpu *vgpu) +{ + vgpu->vdev.cache = RB_ROOT; + mutex_init(&vgpu->vdev.cache_lock); +} + +static void gvt_cache_destroy(struct intel_vgpu *vgpu) +{ + struct gvt_dma *dma; + struct rb_node *node = NULL; + struct device *dev = vgpu->vdev.mdev; + unsigned long pfn; + + mutex_lock(&vgpu->vdev.cache_lock); + while ((node = rb_first(&vgpu->vdev.cache))) { + dma = rb_entry(node, struct gvt_dma, node); + pfn = dma->pfn; + + kvmgt_unpin_pages(dev, &pfn, 1); + __gvt_cache_remove_entry(vgpu, dma); + } + mutex_unlock(&vgpu->vdev.cache_lock); +} + +static struct intel_vgpu_type *intel_gvt_find_vgpu_type(struct intel_gvt *gvt, + const char *name) +{ + int i; + struct intel_vgpu_type *t; + const char *driver_name = dev_driver_string( + &gvt->dev_priv->drm.pdev->dev); + + for (i = 0; i < gvt->num_types; i++) { + t = &gvt->types[i]; + if (!strncmp(t->name, name + strlen(driver_name) + 1, + sizeof(t->name))) + return t; + } + + return NULL; +} + +static struct attribute *type_attrs[] = { + NULL, +}; + +static struct attribute_group *intel_vgpu_type_groups[] = { + [0 ... NR_MAX_INTEL_VGPU_TYPES - 1] = NULL, +}; + +static bool intel_gvt_init_vgpu_type_groups(struct intel_gvt *gvt) +{ + int i, j; + struct intel_vgpu_type *type; + struct attribute_group *group; + + for (i = 0; i < gvt->num_types; i++) { + type = &gvt->types[i]; + + group = kzalloc(sizeof(struct attribute_group), GFP_KERNEL); + if (WARN_ON(!group)) + goto unwind; + + group->name = type->name; + group->attrs = type_attrs; + intel_vgpu_type_groups[i] = group; + } + + return true; + +unwind: + for (j = 0; j < i; j++) { + group = intel_vgpu_type_groups[j]; + kfree(group); + } + + return false; +} + +static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt) +{ + int i; + struct attribute_group *group; + + for (i = 0; i < gvt->num_types; i++) { + group = intel_vgpu_type_groups[i]; + kfree(group); + } +} + +static void kvmgt_protect_table_init(struct kvmgt_guest_info *info) +{ + hash_init(info->ptable); +} + +static void kvmgt_protect_table_destroy(struct kvmgt_guest_info *info) +{ + struct kvmgt_pgfn *p; + struct hlist_node *tmp; + int i; + + hash_for_each_safe(info->ptable, i, tmp, p, hnode) { + hash_del(&p->hnode); + kfree(p); + } +} + +static struct kvmgt_pgfn * +__kvmgt_protect_table_find(struct kvmgt_guest_info *info, gfn_t gfn) +{ + struct kvmgt_pgfn *p, *res = NULL; + + hash_for_each_possible(info->ptable, p, hnode, gfn) { + if (gfn == p->gfn) { + res = p; + break; + } + } + + return res; +} + +static bool kvmgt_gfn_is_write_protected(struct kvmgt_guest_info *info, + gfn_t gfn) +{ + struct kvmgt_pgfn *p; + + p = __kvmgt_protect_table_find(info, gfn); + return !!p; +} + +static void kvmgt_protect_table_add(struct kvmgt_guest_info *info, gfn_t gfn) +{ + struct kvmgt_pgfn *p; + + if (kvmgt_gfn_is_write_protected(info, gfn)) + return; + + p = kmalloc(sizeof(struct kvmgt_pgfn), GFP_ATOMIC); + if (WARN(!p, "gfn: 0x%llx\n", gfn)) + return; + + p->gfn = gfn; + hash_add(info->ptable, &p->hnode, gfn); +} + +static void kvmgt_protect_table_del(struct kvmgt_guest_info *info, + gfn_t gfn) +{ + struct kvmgt_pgfn *p; + + p = __kvmgt_protect_table_find(info, gfn); + if (p) { + hash_del(&p->hnode); + kfree(p); + } +} + +static int kvmgt_host_init(struct device *dev, void *gvt, const void *ops) +{ + if (!intel_gvt_init_vgpu_type_groups(gvt)) + return -EFAULT; + + intel_gvt_ops = ops; + + /* MDEV is not yet available */ + return -ENODEV; +} + +static void kvmgt_host_exit(struct device *dev, void *gvt) +{ + intel_gvt_cleanup_vgpu_type_groups(gvt); +} + +static int kvmgt_write_protect_add(unsigned long handle, u64 gfn) +{ + struct kvmgt_guest_info *info = (struct kvmgt_guest_info *)handle; + struct kvm *kvm = info->kvm; + struct kvm_memory_slot *slot; + int idx; + + idx = srcu_read_lock(&kvm->srcu); + slot = gfn_to_memslot(kvm, gfn); + + spin_lock(&kvm->mmu_lock); + + if (kvmgt_gfn_is_write_protected(info, gfn)) + goto out; + + kvm_slot_page_track_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE); + kvmgt_protect_table_add(info, gfn); + +out: + spin_unlock(&kvm->mmu_lock); + srcu_read_unlock(&kvm->srcu, idx); + return 0; +} + +static int kvmgt_write_protect_remove(unsigned long handle, u64 gfn) +{ + struct kvmgt_guest_info *info = (struct kvmgt_guest_info *)handle; + struct kvm *kvm = info->kvm; + struct kvm_memory_slot *slot; + int idx; + + idx = srcu_read_lock(&kvm->srcu); + slot = gfn_to_memslot(kvm, gfn); + + spin_lock(&kvm->mmu_lock); + + if (!kvmgt_gfn_is_write_protected(info, gfn)) + goto out; + + kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE); + kvmgt_protect_table_del(info, gfn); + +out: + spin_unlock(&kvm->mmu_lock); + srcu_read_unlock(&kvm->srcu, idx); + return 0; +} + +static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, + const u8 *val, int len, + struct kvm_page_track_notifier_node *node) +{ + struct kvmgt_guest_info *info = container_of(node, + struct kvmgt_guest_info, track_node); + + if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa))) + intel_gvt_ops->emulate_mmio_write(info->vgpu, gpa, + (void *)val, len); +} + +static void kvmgt_page_track_flush_slot(struct kvm *kvm, + struct kvm_memory_slot *slot, + struct kvm_page_track_notifier_node *node) +{ + int i; + gfn_t gfn; + struct kvmgt_guest_info *info = container_of(node, + struct kvmgt_guest_info, track_node); + + spin_lock(&kvm->mmu_lock); + for (i = 0; i < slot->npages; i++) { + gfn = slot->base_gfn + i; + if (kvmgt_gfn_is_write_protected(info, gfn)) { + kvm_slot_page_track_remove_page(kvm, slot, gfn, + KVM_PAGE_TRACK_WRITE); + kvmgt_protect_table_del(info, gfn); + } + } + spin_unlock(&kvm->mmu_lock); +} + +static bool kvmgt_check_guest(void) +{ + unsigned int eax, ebx, ecx, edx; + char s[12]; + unsigned int *i; + + eax = KVM_CPUID_SIGNATURE; + ebx = ecx = edx = 0; + + asm volatile ("cpuid" + : "+a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx) + : + : "cc", "memory"); + i = (unsigned int *)s; + i[0] = ebx; + i[1] = ecx; + i[2] = edx; + + return !strncmp(s, "KVMKVMKVM", strlen("KVMKVMKVM")); +} + +/** + * NOTE: + * It's actually impossible to check if we are running in KVM host, + * since the "KVM host" is simply native. So we only dectect guest here. + */ +static int kvmgt_detect_host(void) +{ +#ifdef CONFIG_INTEL_IOMMU + if (intel_iommu_gfx_mapped) { + gvt_err("Hardware IOMMU compatibility not yet supported, try to boot with intel_iommu=igfx_off\n"); + return -ENODEV; + } +#endif + return kvmgt_check_guest() ? -ENODEV : 0; +} + +static int kvmgt_attach_vgpu(void *vgpu, unsigned long *handle) +{ + /* nothing to do here */ + return 0; +} + +static void kvmgt_detach_vgpu(unsigned long handle) +{ + /* nothing to do here */ +} + +static int kvmgt_inject_msi(unsigned long handle, u32 addr, u16 data) +{ + struct kvmgt_guest_info *info = (struct kvmgt_guest_info *)handle; + struct intel_vgpu *vgpu = info->vgpu; + + if (vgpu->vdev.msi_trigger) + return eventfd_signal(vgpu->vdev.msi_trigger, 1) == 1; + + return false; +} + +static unsigned long kvmgt_gfn_to_pfn(unsigned long handle, unsigned long gfn) +{ + unsigned long pfn; + struct kvmgt_guest_info *info = (struct kvmgt_guest_info *)handle; + int rc; + + pfn = gvt_cache_find(info->vgpu, gfn); + if (pfn != 0) + return pfn; + + rc = kvmgt_pin_pages(info->vgpu->vdev.mdev, &gfn, 1, + IOMMU_READ | IOMMU_WRITE, &pfn); + if (rc != 1) { + gvt_err("vfio_pin_pages failed for gfn: 0x%lx\n", gfn); + return 0; + } + + gvt_cache_add(info->vgpu, gfn, pfn); + return pfn; +} + +static void *kvmgt_gpa_to_hva(unsigned long handle, unsigned long gpa) +{ + unsigned long pfn; + gfn_t gfn = gpa_to_gfn(gpa); + + pfn = kvmgt_gfn_to_pfn(handle, gfn); + if (!pfn) + return NULL; + + return (char *)pfn_to_kaddr(pfn) + offset_in_page(gpa); +} + +static int kvmgt_rw_gpa(unsigned long handle, unsigned long gpa, + void *buf, unsigned long len, bool write) +{ + void *hva = NULL; + + hva = kvmgt_gpa_to_hva(handle, gpa); + if (!hva) + return -EFAULT; + + if (write) + memcpy(hva, buf, len); + else + memcpy(buf, hva, len); + + return 0; +} + +static int kvmgt_read_gpa(unsigned long handle, unsigned long gpa, + void *buf, unsigned long len) +{ + return kvmgt_rw_gpa(handle, gpa, buf, len, false); +} + +static int kvmgt_write_gpa(unsigned long handle, unsigned long gpa, + void *buf, unsigned long len) +{ + return kvmgt_rw_gpa(handle, gpa, buf, len, true); +} + +static unsigned long kvmgt_virt_to_pfn(void *addr) +{ + return PFN_DOWN(__pa(addr)); +} + +struct intel_gvt_mpt kvmgt_mpt = { + .detect_host = kvmgt_detect_host, + .host_init = kvmgt_host_init, + .host_exit = kvmgt_host_exit, + .attach_vgpu = kvmgt_attach_vgpu, + .detach_vgpu = kvmgt_detach_vgpu, + .inject_msi = kvmgt_inject_msi, + .from_virt_to_mfn = kvmgt_virt_to_pfn, + .set_wp_page = kvmgt_write_protect_add, + .unset_wp_page = kvmgt_write_protect_remove, + .read_gpa = kvmgt_read_gpa, + .write_gpa = kvmgt_write_gpa, + .gfn_to_mfn = kvmgt_gfn_to_pfn, +}; +EXPORT_SYMBOL_GPL(kvmgt_mpt); + +static int __init kvmgt_init(void) +{ + return 0; +} + +static void __exit kvmgt_exit(void) +{ +} + +module_init(kvmgt_init); +module_exit(kvmgt_exit); + +MODULE_LICENSE("GPL and additional rights"); +MODULE_AUTHOR("Intel Corporation"); diff --git a/drivers/gpu/drm/i915/gvt/mmio.c b/drivers/gpu/drm/i915/gvt/mmio.c new file mode 100644 index 000000000000..09c9450a1946 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/mmio.c @@ -0,0 +1,304 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Kevin Tian <kevin.tian@intel.com> + * Dexuan Cui + * + * Contributors: + * Tina Zhang <tina.zhang@intel.com> + * Min He <min.he@intel.com> + * Niu Bing <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +/** + * intel_vgpu_gpa_to_mmio_offset - translate a GPA to MMIO offset + * @vgpu: a vGPU + * + * Returns: + * Zero on success, negative error code if failed + */ +int intel_vgpu_gpa_to_mmio_offset(struct intel_vgpu *vgpu, u64 gpa) +{ + u64 gttmmio_gpa = *(u64 *)(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_0) & + ~GENMASK(3, 0); + return gpa - gttmmio_gpa; +} + +#define reg_is_mmio(gvt, reg) \ + (reg >= 0 && reg < gvt->device_info.mmio_size) + +#define reg_is_gtt(gvt, reg) \ + (reg >= gvt->device_info.gtt_start_offset \ + && reg < gvt->device_info.gtt_start_offset + gvt_ggtt_sz(gvt)) + +/** + * intel_vgpu_emulate_mmio_read - emulate MMIO read + * @vgpu: a vGPU + * @pa: guest physical address + * @p_data: data return buffer + * @bytes: access data length + * + * Returns: + * Zero on success, negative error code if failed + */ +int intel_vgpu_emulate_mmio_read(struct intel_vgpu *vgpu, uint64_t pa, + void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_mmio_info *mmio; + unsigned int offset = 0; + int ret = -EINVAL; + + mutex_lock(&gvt->lock); + + if (atomic_read(&vgpu->gtt.n_write_protected_guest_page)) { + struct intel_vgpu_guest_page *gp; + + gp = intel_vgpu_find_guest_page(vgpu, pa >> PAGE_SHIFT); + if (gp) { + ret = intel_gvt_hypervisor_read_gpa(vgpu, pa, + p_data, bytes); + if (ret) { + gvt_err("vgpu%d: guest page read error %d, " + "gfn 0x%lx, pa 0x%llx, var 0x%x, len %d\n", + vgpu->id, ret, + gp->gfn, pa, *(u32 *)p_data, bytes); + } + mutex_unlock(&gvt->lock); + return ret; + } + } + + offset = intel_vgpu_gpa_to_mmio_offset(vgpu, pa); + + if (WARN_ON(bytes > 8)) + goto err; + + if (reg_is_gtt(gvt, offset)) { + if (WARN_ON(!IS_ALIGNED(offset, 4) && !IS_ALIGNED(offset, 8))) + goto err; + if (WARN_ON(bytes != 4 && bytes != 8)) + goto err; + if (WARN_ON(!reg_is_gtt(gvt, offset + bytes - 1))) + goto err; + + ret = intel_vgpu_emulate_gtt_mmio_read(vgpu, offset, + p_data, bytes); + if (ret) + goto err; + mutex_unlock(&gvt->lock); + return ret; + } + + if (WARN_ON_ONCE(!reg_is_mmio(gvt, offset))) { + ret = intel_gvt_hypervisor_read_gpa(vgpu, pa, p_data, bytes); + mutex_unlock(&gvt->lock); + return ret; + } + + if (WARN_ON(!reg_is_mmio(gvt, offset + bytes - 1))) + goto err; + + mmio = intel_gvt_find_mmio_info(gvt, rounddown(offset, 4)); + if (!mmio && !vgpu->mmio.disable_warn_untrack) { + gvt_err("vgpu%d: read untracked MMIO %x len %d val %x\n", + vgpu->id, offset, bytes, *(u32 *)p_data); + + if (offset == 0x206c) { + gvt_err("------------------------------------------\n"); + gvt_err("vgpu%d: likely triggers a gfx reset\n", + vgpu->id); + gvt_err("------------------------------------------\n"); + vgpu->mmio.disable_warn_untrack = true; + } + } + + if (!intel_gvt_mmio_is_unalign(gvt, offset)) { + if (WARN_ON(!IS_ALIGNED(offset, bytes))) + goto err; + } + + if (mmio) { + if (!intel_gvt_mmio_is_unalign(gvt, mmio->offset)) { + if (WARN_ON(offset + bytes > mmio->offset + mmio->size)) + goto err; + if (WARN_ON(mmio->offset != offset)) + goto err; + } + ret = mmio->read(vgpu, offset, p_data, bytes); + } else + ret = intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes); + + if (ret) + goto err; + + intel_gvt_mmio_set_accessed(gvt, offset); + mutex_unlock(&gvt->lock); + return 0; +err: + gvt_err("vgpu%d: fail to emulate MMIO read %08x len %d\n", + vgpu->id, offset, bytes); + mutex_unlock(&gvt->lock); + return ret; +} + +/** + * intel_vgpu_emulate_mmio_write - emulate MMIO write + * @vgpu: a vGPU + * @pa: guest physical address + * @p_data: write data buffer + * @bytes: access data length + * + * Returns: + * Zero on success, negative error code if failed + */ +int intel_vgpu_emulate_mmio_write(struct intel_vgpu *vgpu, uint64_t pa, + void *p_data, unsigned int bytes) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_mmio_info *mmio; + unsigned int offset = 0; + u32 old_vreg = 0, old_sreg = 0; + int ret = -EINVAL; + + mutex_lock(&gvt->lock); + + if (atomic_read(&vgpu->gtt.n_write_protected_guest_page)) { + struct intel_vgpu_guest_page *gp; + + gp = intel_vgpu_find_guest_page(vgpu, pa >> PAGE_SHIFT); + if (gp) { + ret = gp->handler(gp, pa, p_data, bytes); + if (ret) { + gvt_err("vgpu%d: guest page write error %d, " + "gfn 0x%lx, pa 0x%llx, var 0x%x, len %d\n", + vgpu->id, ret, + gp->gfn, pa, *(u32 *)p_data, bytes); + } + mutex_unlock(&gvt->lock); + return ret; + } + } + + offset = intel_vgpu_gpa_to_mmio_offset(vgpu, pa); + + if (WARN_ON(bytes > 8)) + goto err; + + if (reg_is_gtt(gvt, offset)) { + if (WARN_ON(!IS_ALIGNED(offset, 4) && !IS_ALIGNED(offset, 8))) + goto err; + if (WARN_ON(bytes != 4 && bytes != 8)) + goto err; + if (WARN_ON(!reg_is_gtt(gvt, offset + bytes - 1))) + goto err; + + ret = intel_vgpu_emulate_gtt_mmio_write(vgpu, offset, + p_data, bytes); + if (ret) + goto err; + mutex_unlock(&gvt->lock); + return ret; + } + + if (WARN_ON_ONCE(!reg_is_mmio(gvt, offset))) { + ret = intel_gvt_hypervisor_write_gpa(vgpu, pa, p_data, bytes); + mutex_unlock(&gvt->lock); + return ret; + } + + mmio = intel_gvt_find_mmio_info(gvt, rounddown(offset, 4)); + if (!mmio && !vgpu->mmio.disable_warn_untrack) + gvt_err("vgpu%d: write untracked MMIO %x len %d val %x\n", + vgpu->id, offset, bytes, *(u32 *)p_data); + + if (!intel_gvt_mmio_is_unalign(gvt, offset)) { + if (WARN_ON(!IS_ALIGNED(offset, bytes))) + goto err; + } + + if (mmio) { + u64 ro_mask = mmio->ro_mask; + + if (!intel_gvt_mmio_is_unalign(gvt, mmio->offset)) { + if (WARN_ON(offset + bytes > mmio->offset + mmio->size)) + goto err; + if (WARN_ON(mmio->offset != offset)) + goto err; + } + + if (intel_gvt_mmio_has_mode_mask(gvt, mmio->offset)) { + old_vreg = vgpu_vreg(vgpu, offset); + old_sreg = vgpu_sreg(vgpu, offset); + } + + if (!ro_mask) { + ret = mmio->write(vgpu, offset, p_data, bytes); + } else { + /* Protect RO bits like HW */ + u64 data = 0; + + /* all register bits are RO. */ + if (ro_mask == ~(u64)0) { + gvt_err("vgpu%d: try to write RO reg %x\n", + vgpu->id, offset); + ret = 0; + goto out; + } + /* keep the RO bits in the virtual register */ + memcpy(&data, p_data, bytes); + data &= ~mmio->ro_mask; + data |= vgpu_vreg(vgpu, offset) & mmio->ro_mask; + ret = mmio->write(vgpu, offset, &data, bytes); + } + + /* higher 16bits of mode ctl regs are mask bits for change */ + if (intel_gvt_mmio_has_mode_mask(gvt, mmio->offset)) { + u32 mask = vgpu_vreg(vgpu, offset) >> 16; + + vgpu_vreg(vgpu, offset) = (old_vreg & ~mask) + | (vgpu_vreg(vgpu, offset) & mask); + vgpu_sreg(vgpu, offset) = (old_sreg & ~mask) + | (vgpu_sreg(vgpu, offset) & mask); + } + } else + ret = intel_vgpu_default_mmio_write(vgpu, offset, p_data, + bytes); + if (ret) + goto err; +out: + intel_gvt_mmio_set_accessed(gvt, offset); + mutex_unlock(&gvt->lock); + return 0; +err: + gvt_err("vgpu%d: fail to emulate MMIO write %08x len %d\n", + vgpu->id, offset, bytes); + mutex_unlock(&gvt->lock); + return ret; +} diff --git a/drivers/gpu/drm/i915/gvt/mmio.h b/drivers/gpu/drm/i915/gvt/mmio.h new file mode 100644 index 000000000000..87d5b5e366a3 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/mmio.h @@ -0,0 +1,106 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Ke Yu + * Kevin Tian <kevin.tian@intel.com> + * Dexuan Cui + * + * Contributors: + * Tina Zhang <tina.zhang@intel.com> + * Min He <min.he@intel.com> + * Niu Bing <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#ifndef _GVT_MMIO_H_ +#define _GVT_MMIO_H_ + +struct intel_gvt; +struct intel_vgpu; + +#define D_SNB (1 << 0) +#define D_IVB (1 << 1) +#define D_HSW (1 << 2) +#define D_BDW (1 << 3) +#define D_SKL (1 << 4) + +#define D_GEN9PLUS (D_SKL) +#define D_GEN8PLUS (D_BDW | D_SKL) +#define D_GEN75PLUS (D_HSW | D_BDW | D_SKL) +#define D_GEN7PLUS (D_IVB | D_HSW | D_BDW | D_SKL) + +#define D_SKL_PLUS (D_SKL) +#define D_BDW_PLUS (D_BDW | D_SKL) +#define D_HSW_PLUS (D_HSW | D_BDW | D_SKL) +#define D_IVB_PLUS (D_IVB | D_HSW | D_BDW | D_SKL) + +#define D_PRE_BDW (D_SNB | D_IVB | D_HSW) +#define D_PRE_SKL (D_SNB | D_IVB | D_HSW | D_BDW) +#define D_ALL (D_SNB | D_IVB | D_HSW | D_BDW | D_SKL) + +struct intel_gvt_mmio_info { + u32 offset; + u32 size; + u32 length; + u32 addr_mask; + u64 ro_mask; + u32 device; + int (*read)(struct intel_vgpu *, unsigned int, void *, unsigned int); + int (*write)(struct intel_vgpu *, unsigned int, void *, unsigned int); + u32 addr_range; + struct hlist_node node; +}; + +unsigned long intel_gvt_get_device_type(struct intel_gvt *gvt); +bool intel_gvt_match_device(struct intel_gvt *gvt, unsigned long device); + +int intel_gvt_setup_mmio_info(struct intel_gvt *gvt); +void intel_gvt_clean_mmio_info(struct intel_gvt *gvt); + +struct intel_gvt_mmio_info *intel_gvt_find_mmio_info(struct intel_gvt *gvt, + unsigned int offset); +#define INTEL_GVT_MMIO_OFFSET(reg) ({ \ + typeof(reg) __reg = reg; \ + u32 *offset = (u32 *)&__reg; \ + *offset; \ +}) + +int intel_vgpu_gpa_to_mmio_offset(struct intel_vgpu *vgpu, u64 gpa); + +int intel_vgpu_emulate_mmio_read(struct intel_vgpu *vgpu, u64 pa, + void *p_data, unsigned int bytes); +int intel_vgpu_emulate_mmio_write(struct intel_vgpu *vgpu, u64 pa, + void *p_data, unsigned int bytes); +bool intel_gvt_mmio_is_cmd_access(struct intel_gvt *gvt, + unsigned int offset); +bool intel_gvt_mmio_is_unalign(struct intel_gvt *gvt, unsigned int offset); +void intel_gvt_mmio_set_accessed(struct intel_gvt *gvt, unsigned int offset); +void intel_gvt_mmio_set_cmd_accessed(struct intel_gvt *gvt, + unsigned int offset); +bool intel_gvt_mmio_has_mode_mask(struct intel_gvt *gvt, unsigned int offset); +int intel_vgpu_default_mmio_read(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes); +int intel_vgpu_default_mmio_write(struct intel_vgpu *vgpu, unsigned int offset, + void *p_data, unsigned int bytes); +#endif diff --git a/drivers/gpu/drm/i915/gvt/mpt.h b/drivers/gpu/drm/i915/gvt/mpt.h index 03601e3ffa7c..1af5830c0a56 100644 --- a/drivers/gpu/drm/i915/gvt/mpt.h +++ b/drivers/gpu/drm/i915/gvt/mpt.h @@ -19,6 +19,15 @@ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Dexuan Cui + * Jike Song <jike.song@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * */ #ifndef _GVT_MPT_H_ @@ -46,4 +55,254 @@ static inline int intel_gvt_hypervisor_detect_host(void) return intel_gvt_host.mpt->detect_host(); } +/** + * intel_gvt_hypervisor_host_init - init GVT-g host side + * + * Returns: + * Zero on success, negative error code if failed + */ +static inline int intel_gvt_hypervisor_host_init(struct device *dev, + void *gvt, const void *ops) +{ + /* optional to provide */ + if (!intel_gvt_host.mpt->host_init) + return 0; + + return intel_gvt_host.mpt->host_init(dev, gvt, ops); +} + +/** + * intel_gvt_hypervisor_host_exit - exit GVT-g host side + */ +static inline void intel_gvt_hypervisor_host_exit(struct device *dev, + void *gvt) +{ + /* optional to provide */ + if (!intel_gvt_host.mpt->host_exit) + return; + + intel_gvt_host.mpt->host_exit(dev, gvt); +} + +/** + * intel_gvt_hypervisor_attach_vgpu - call hypervisor to initialize vGPU + * related stuffs inside hypervisor. + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_attach_vgpu(struct intel_vgpu *vgpu) +{ + /* optional to provide */ + if (!intel_gvt_host.mpt->attach_vgpu) + return 0; + + return intel_gvt_host.mpt->attach_vgpu(vgpu, &vgpu->handle); +} + +/** + * intel_gvt_hypervisor_detach_vgpu - call hypervisor to release vGPU + * related stuffs inside hypervisor. + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline void intel_gvt_hypervisor_detach_vgpu(struct intel_vgpu *vgpu) +{ + /* optional to provide */ + if (!intel_gvt_host.mpt->detach_vgpu) + return; + + intel_gvt_host.mpt->detach_vgpu(vgpu->handle); +} + +#define MSI_CAP_CONTROL(offset) (offset + 2) +#define MSI_CAP_ADDRESS(offset) (offset + 4) +#define MSI_CAP_DATA(offset) (offset + 8) +#define MSI_CAP_EN 0x1 + +/** + * intel_gvt_hypervisor_inject_msi - inject a MSI interrupt into vGPU + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_inject_msi(struct intel_vgpu *vgpu) +{ + unsigned long offset = vgpu->gvt->device_info.msi_cap_offset; + u16 control, data; + u32 addr; + int ret; + + control = *(u16 *)(vgpu_cfg_space(vgpu) + MSI_CAP_CONTROL(offset)); + addr = *(u32 *)(vgpu_cfg_space(vgpu) + MSI_CAP_ADDRESS(offset)); + data = *(u16 *)(vgpu_cfg_space(vgpu) + MSI_CAP_DATA(offset)); + + /* Do not generate MSI if MSIEN is disable */ + if (!(control & MSI_CAP_EN)) + return 0; + + if (WARN(control & GENMASK(15, 1), "only support one MSI format\n")) + return -EINVAL; + + gvt_dbg_irq("vgpu%d: inject msi address %x data%x\n", vgpu->id, addr, + data); + + ret = intel_gvt_host.mpt->inject_msi(vgpu->handle, addr, data); + if (ret) + return ret; + return 0; +} + +/** + * intel_gvt_hypervisor_set_wp_page - translate a host VA into MFN + * @p: host kernel virtual address + * + * Returns: + * MFN on success, INTEL_GVT_INVALID_ADDR if failed. + */ +static inline unsigned long intel_gvt_hypervisor_virt_to_mfn(void *p) +{ + return intel_gvt_host.mpt->from_virt_to_mfn(p); +} + +/** + * intel_gvt_hypervisor_set_wp_page - set a guest page to write-protected + * @vgpu: a vGPU + * @p: intel_vgpu_guest_page + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_set_wp_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *p) +{ + int ret; + + if (p->writeprotection) + return 0; + + ret = intel_gvt_host.mpt->set_wp_page(vgpu->handle, p->gfn); + if (ret) + return ret; + p->writeprotection = true; + atomic_inc(&vgpu->gtt.n_write_protected_guest_page); + return 0; +} + +/** + * intel_gvt_hypervisor_unset_wp_page - remove the write-protection of a + * guest page + * @vgpu: a vGPU + * @p: intel_vgpu_guest_page + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_unset_wp_page(struct intel_vgpu *vgpu, + struct intel_vgpu_guest_page *p) +{ + int ret; + + if (!p->writeprotection) + return 0; + + ret = intel_gvt_host.mpt->unset_wp_page(vgpu->handle, p->gfn); + if (ret) + return ret; + p->writeprotection = false; + atomic_dec(&vgpu->gtt.n_write_protected_guest_page); + return 0; +} + +/** + * intel_gvt_hypervisor_read_gpa - copy data from GPA to host data buffer + * @vgpu: a vGPU + * @gpa: guest physical address + * @buf: host data buffer + * @len: data length + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_read_gpa(struct intel_vgpu *vgpu, + unsigned long gpa, void *buf, unsigned long len) +{ + return intel_gvt_host.mpt->read_gpa(vgpu->handle, gpa, buf, len); +} + +/** + * intel_gvt_hypervisor_write_gpa - copy data from host data buffer to GPA + * @vgpu: a vGPU + * @gpa: guest physical address + * @buf: host data buffer + * @len: data length + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_write_gpa(struct intel_vgpu *vgpu, + unsigned long gpa, void *buf, unsigned long len) +{ + return intel_gvt_host.mpt->write_gpa(vgpu->handle, gpa, buf, len); +} + +/** + * intel_gvt_hypervisor_gfn_to_mfn - translate a GFN to MFN + * @vgpu: a vGPU + * @gpfn: guest pfn + * + * Returns: + * MFN on success, INTEL_GVT_INVALID_ADDR if failed. + */ +static inline unsigned long intel_gvt_hypervisor_gfn_to_mfn( + struct intel_vgpu *vgpu, unsigned long gfn) +{ + return intel_gvt_host.mpt->gfn_to_mfn(vgpu->handle, gfn); +} + +/** + * intel_gvt_hypervisor_map_gfn_to_mfn - map a GFN region to MFN + * @vgpu: a vGPU + * @gfn: guest PFN + * @mfn: host PFN + * @nr: amount of PFNs + * @map: map or unmap + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_map_gfn_to_mfn( + struct intel_vgpu *vgpu, unsigned long gfn, + unsigned long mfn, unsigned int nr, + bool map) +{ + /* a MPT implementation could have MMIO mapped elsewhere */ + if (!intel_gvt_host.mpt->map_gfn_to_mfn) + return 0; + + return intel_gvt_host.mpt->map_gfn_to_mfn(vgpu->handle, gfn, mfn, nr, + map); +} + +/** + * intel_gvt_hypervisor_set_trap_area - Trap a guest PA region + * @vgpu: a vGPU + * @start: the beginning of the guest physical address region + * @end: the end of the guest physical address region + * @map: map or unmap + * + * Returns: + * Zero on success, negative error code if failed. + */ +static inline int intel_gvt_hypervisor_set_trap_area( + struct intel_vgpu *vgpu, u64 start, u64 end, bool map) +{ + /* a MPT implementation could have MMIO trapped elsewhere */ + if (!intel_gvt_host.mpt->set_trap_area) + return 0; + + return intel_gvt_host.mpt->set_trap_area(vgpu->handle, start, end, map); +} + #endif /* _GVT_MPT_H_ */ diff --git a/drivers/gpu/drm/i915/gvt/opregion.c b/drivers/gpu/drm/i915/gvt/opregion.c new file mode 100644 index 000000000000..d2a0fbc896c3 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/opregion.c @@ -0,0 +1,320 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/acpi.h> +#include "i915_drv.h" +#include "gvt.h" + +static int init_vgpu_opregion(struct intel_vgpu *vgpu, u32 gpa) +{ + void __iomem *host_va = vgpu->gvt->opregion.opregion_va; + u8 *buf; + int i; + + if (WARN((vgpu_opregion(vgpu)->va), + "vgpu%d: opregion has been initialized already.\n", + vgpu->id)) + return -EINVAL; + + vgpu_opregion(vgpu)->va = (void *)__get_free_pages(GFP_ATOMIC | + GFP_DMA32 | __GFP_ZERO, + INTEL_GVT_OPREGION_PORDER); + + if (!vgpu_opregion(vgpu)->va) + return -ENOMEM; + + memcpy_fromio(vgpu_opregion(vgpu)->va, host_va, + INTEL_GVT_OPREGION_SIZE); + + for (i = 0; i < INTEL_GVT_OPREGION_PAGES; i++) + vgpu_opregion(vgpu)->gfn[i] = (gpa >> PAGE_SHIFT) + i; + + /* for unknown reason, the value in LID field is incorrect + * which block the windows guest, so workaround it by force + * setting it to "OPEN" + */ + buf = (u8 *)vgpu_opregion(vgpu)->va; + buf[INTEL_GVT_OPREGION_CLID] = 0x3; + + return 0; +} + +static int map_vgpu_opregion(struct intel_vgpu *vgpu, bool map) +{ + u64 mfn; + int i, ret; + + for (i = 0; i < INTEL_GVT_OPREGION_PAGES; i++) { + mfn = intel_gvt_hypervisor_virt_to_mfn(vgpu_opregion(vgpu) + + i * PAGE_SIZE); + if (mfn == INTEL_GVT_INVALID_ADDR) { + gvt_err("fail to get MFN from VA\n"); + return -EINVAL; + } + ret = intel_gvt_hypervisor_map_gfn_to_mfn(vgpu, + vgpu_opregion(vgpu)->gfn[i], + mfn, 1, map); + if (ret) { + gvt_err("fail to map GFN to MFN, errno: %d\n", ret); + return ret; + } + } + return 0; +} + +/** + * intel_vgpu_clean_opregion - clean the stuff used to emulate opregion + * @vgpu: a vGPU + * + */ +void intel_vgpu_clean_opregion(struct intel_vgpu *vgpu) +{ + gvt_dbg_core("vgpu%d: clean vgpu opregion\n", vgpu->id); + + if (!vgpu_opregion(vgpu)->va) + return; + + if (intel_gvt_host.hypervisor_type == INTEL_GVT_HYPERVISOR_XEN) { + map_vgpu_opregion(vgpu, false); + free_pages((unsigned long)vgpu_opregion(vgpu)->va, + INTEL_GVT_OPREGION_PORDER); + + vgpu_opregion(vgpu)->va = NULL; + } +} + +/** + * intel_vgpu_init_opregion - initialize the stuff used to emulate opregion + * @vgpu: a vGPU + * @gpa: guest physical address of opregion + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_vgpu_init_opregion(struct intel_vgpu *vgpu, u32 gpa) +{ + int ret; + + gvt_dbg_core("vgpu%d: init vgpu opregion\n", vgpu->id); + + if (intel_gvt_host.hypervisor_type == INTEL_GVT_HYPERVISOR_XEN) { + gvt_dbg_core("emulate opregion from kernel\n"); + + ret = init_vgpu_opregion(vgpu, gpa); + if (ret) + return ret; + + ret = map_vgpu_opregion(vgpu, true); + if (ret) + return ret; + } + + return 0; +} + +/** + * intel_gvt_clean_opregion - clean host opergion related stuffs + * @gvt: a GVT device + * + */ +void intel_gvt_clean_opregion(struct intel_gvt *gvt) +{ + memunmap(gvt->opregion.opregion_va); + gvt->opregion.opregion_va = NULL; +} + +/** + * intel_gvt_init_opregion - initialize host opergion related stuffs + * @gvt: a GVT device + * + * Returns: + * Zero on success, negative error code if failed. + */ +int intel_gvt_init_opregion(struct intel_gvt *gvt) +{ + gvt_dbg_core("init host opregion\n"); + + pci_read_config_dword(gvt->dev_priv->drm.pdev, INTEL_GVT_PCI_OPREGION, + &gvt->opregion.opregion_pa); + + gvt->opregion.opregion_va = memremap(gvt->opregion.opregion_pa, + INTEL_GVT_OPREGION_SIZE, MEMREMAP_WB); + if (!gvt->opregion.opregion_va) { + gvt_err("fail to map host opregion\n"); + return -EFAULT; + } + return 0; +} + +#define GVT_OPREGION_FUNC(scic) \ + ({ \ + u32 __ret; \ + __ret = (scic & OPREGION_SCIC_FUNC_MASK) >> \ + OPREGION_SCIC_FUNC_SHIFT; \ + __ret; \ + }) + +#define GVT_OPREGION_SUBFUNC(scic) \ + ({ \ + u32 __ret; \ + __ret = (scic & OPREGION_SCIC_SUBFUNC_MASK) >> \ + OPREGION_SCIC_SUBFUNC_SHIFT; \ + __ret; \ + }) + +static const char *opregion_func_name(u32 func) +{ + const char *name = NULL; + + switch (func) { + case 0 ... 3: + case 5: + case 7 ... 15: + name = "Reserved"; + break; + + case 4: + name = "Get BIOS Data"; + break; + + case 6: + name = "System BIOS Callbacks"; + break; + + default: + name = "Unknown"; + break; + } + return name; +} + +static const char *opregion_subfunc_name(u32 subfunc) +{ + const char *name = NULL; + + switch (subfunc) { + case 0: + name = "Supported Calls"; + break; + + case 1: + name = "Requested Callbacks"; + break; + + case 2 ... 3: + case 8 ... 9: + name = "Reserved"; + break; + + case 5: + name = "Boot Display"; + break; + + case 6: + name = "TV-Standard/Video-Connector"; + break; + + case 7: + name = "Internal Graphics"; + break; + + case 10: + name = "Spread Spectrum Clocks"; + break; + + case 11: + name = "Get AKSV"; + break; + + default: + name = "Unknown"; + break; + } + return name; +}; + +static bool querying_capabilities(u32 scic) +{ + u32 func, subfunc; + + func = GVT_OPREGION_FUNC(scic); + subfunc = GVT_OPREGION_SUBFUNC(scic); + + if ((func == INTEL_GVT_OPREGION_SCIC_F_GETBIOSDATA && + subfunc == INTEL_GVT_OPREGION_SCIC_SF_SUPPRTEDCALLS) + || (func == INTEL_GVT_OPREGION_SCIC_F_GETBIOSDATA && + subfunc == INTEL_GVT_OPREGION_SCIC_SF_REQEUSTEDCALLBACKS) + || (func == INTEL_GVT_OPREGION_SCIC_F_GETBIOSCALLBACKS && + subfunc == INTEL_GVT_OPREGION_SCIC_SF_SUPPRTEDCALLS)) { + return true; + } + return false; +} + +/** + * intel_vgpu_emulate_opregion_request - emulating OpRegion request + * @vgpu: a vGPU + * @swsci: SWSCI request + * + * Returns: + * Zero on success, negative error code if failed + */ +int intel_vgpu_emulate_opregion_request(struct intel_vgpu *vgpu, u32 swsci) +{ + u32 *scic, *parm; + u32 func, subfunc; + + scic = vgpu_opregion(vgpu)->va + INTEL_GVT_OPREGION_SCIC; + parm = vgpu_opregion(vgpu)->va + INTEL_GVT_OPREGION_PARM; + + if (!(swsci & SWSCI_SCI_SELECT)) { + gvt_err("vgpu%d: requesting SMI service\n", vgpu->id); + return 0; + } + /* ignore non 0->1 trasitions */ + if ((vgpu_cfg_space(vgpu)[INTEL_GVT_PCI_SWSCI] + & SWSCI_SCI_TRIGGER) || + !(swsci & SWSCI_SCI_TRIGGER)) { + return 0; + } + + func = GVT_OPREGION_FUNC(*scic); + subfunc = GVT_OPREGION_SUBFUNC(*scic); + if (!querying_capabilities(*scic)) { + gvt_err("vgpu%d: requesting runtime service: func \"%s\"," + " subfunc \"%s\"\n", + vgpu->id, + opregion_func_name(func), + opregion_subfunc_name(subfunc)); + /* + * emulate exit status of function call, '0' means + * "failure, generic, unsupported or unknown cause" + */ + *scic &= ~OPREGION_SCIC_EXIT_MASK; + return 0; + } + + *scic = 0; + *parm = 0; + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/reg.h b/drivers/gpu/drm/i915/gvt/reg.h new file mode 100644 index 000000000000..0dfe789d8f02 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/reg.h @@ -0,0 +1,80 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#ifndef _GVT_REG_H +#define _GVT_REG_H + +#define INTEL_GVT_PCI_CLASS_VGA_OTHER 0x80 + +#define INTEL_GVT_PCI_GMCH_CONTROL 0x50 +#define BDW_GMCH_GMS_SHIFT 8 +#define BDW_GMCH_GMS_MASK 0xff + +#define INTEL_GVT_PCI_SWSCI 0xe8 +#define SWSCI_SCI_SELECT (1 << 15) +#define SWSCI_SCI_TRIGGER 1 + +#define INTEL_GVT_PCI_OPREGION 0xfc + +#define INTEL_GVT_OPREGION_CLID 0x1AC +#define INTEL_GVT_OPREGION_SCIC 0x200 +#define OPREGION_SCIC_FUNC_MASK 0x1E +#define OPREGION_SCIC_FUNC_SHIFT 1 +#define OPREGION_SCIC_SUBFUNC_MASK 0xFF00 +#define OPREGION_SCIC_SUBFUNC_SHIFT 8 +#define OPREGION_SCIC_EXIT_MASK 0xE0 +#define INTEL_GVT_OPREGION_SCIC_F_GETBIOSDATA 4 +#define INTEL_GVT_OPREGION_SCIC_F_GETBIOSCALLBACKS 6 +#define INTEL_GVT_OPREGION_SCIC_SF_SUPPRTEDCALLS 0 +#define INTEL_GVT_OPREGION_SCIC_SF_REQEUSTEDCALLBACKS 1 +#define INTEL_GVT_OPREGION_PARM 0x204 + +#define INTEL_GVT_OPREGION_PAGES 2 +#define INTEL_GVT_OPREGION_PORDER 1 +#define INTEL_GVT_OPREGION_SIZE (2 * 4096) + +#define VGT_SPRSTRIDE(pipe) _PIPE(pipe, _SPRA_STRIDE, _PLANE_STRIDE_2_B) + +#define _REG_VECS_EXCC 0x1A028 +#define _REG_VCS2_EXCC 0x1c028 + +#define _REG_701C0(pipe, plane) (0x701c0 + pipe * 0x1000 + (plane - 1) * 0x100) +#define _REG_701C4(pipe, plane) (0x701c4 + pipe * 0x1000 + (plane - 1) * 0x100) + +#define GFX_MODE_BIT_SET_IN_MASK(val, bit) \ + ((((bit) & 0xffff0000) == 0) && !!((val) & (((bit) << 16)))) + +#define FORCEWAKE_RENDER_GEN9_REG 0xa278 +#define FORCEWAKE_ACK_RENDER_GEN9_REG 0x0D84 +#define FORCEWAKE_BLITTER_GEN9_REG 0xa188 +#define FORCEWAKE_ACK_BLITTER_GEN9_REG 0x130044 +#define FORCEWAKE_MEDIA_GEN9_REG 0xa270 +#define FORCEWAKE_ACK_MEDIA_GEN9_REG 0x0D88 +#define FORCEWAKE_ACK_HSW_REG 0x130044 + +#define RB_HEAD_OFF_MASK ((1U << 21) - (1U << 2)) +#define RB_TAIL_OFF_MASK ((1U << 21) - (1U << 3)) +#define RB_TAIL_SIZE_MASK ((1U << 21) - (1U << 12)) +#define _RING_CTL_BUF_SIZE(ctl) (((ctl) & RB_TAIL_SIZE_MASK) + GTT_PAGE_SIZE) + +#endif diff --git a/drivers/gpu/drm/i915/gvt/render.c b/drivers/gpu/drm/i915/gvt/render.c new file mode 100644 index 000000000000..44136b1f3aab --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/render.c @@ -0,0 +1,310 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Kevin Tian <kevin.tian@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * Changbin Du <changbin.du@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +struct render_mmio { + int ring_id; + i915_reg_t reg; + u32 mask; + bool in_context; + u32 value; +}; + +static struct render_mmio gen8_render_mmio_list[] = { + {RCS, _MMIO(0x229c), 0xffff, false}, + {RCS, _MMIO(0x2248), 0x0, false}, + {RCS, _MMIO(0x2098), 0x0, false}, + {RCS, _MMIO(0x20c0), 0xffff, true}, + {RCS, _MMIO(0x24d0), 0, false}, + {RCS, _MMIO(0x24d4), 0, false}, + {RCS, _MMIO(0x24d8), 0, false}, + {RCS, _MMIO(0x24dc), 0, false}, + {RCS, _MMIO(0x7004), 0xffff, true}, + {RCS, _MMIO(0x7008), 0xffff, true}, + {RCS, _MMIO(0x7000), 0xffff, true}, + {RCS, _MMIO(0x7010), 0xffff, true}, + {RCS, _MMIO(0x7300), 0xffff, true}, + {RCS, _MMIO(0x83a4), 0xffff, true}, + + {BCS, _MMIO(0x2229c), 0xffff, false}, + {BCS, _MMIO(0x2209c), 0xffff, false}, + {BCS, _MMIO(0x220c0), 0xffff, false}, + {BCS, _MMIO(0x22098), 0x0, false}, + {BCS, _MMIO(0x22028), 0x0, false}, +}; + +static struct render_mmio gen9_render_mmio_list[] = { + {RCS, _MMIO(0x229c), 0xffff, false}, + {RCS, _MMIO(0x2248), 0x0, false}, + {RCS, _MMIO(0x2098), 0x0, false}, + {RCS, _MMIO(0x20c0), 0xffff, true}, + {RCS, _MMIO(0x24d0), 0, false}, + {RCS, _MMIO(0x24d4), 0, false}, + {RCS, _MMIO(0x24d8), 0, false}, + {RCS, _MMIO(0x24dc), 0, false}, + {RCS, _MMIO(0x7004), 0xffff, true}, + {RCS, _MMIO(0x7008), 0xffff, true}, + {RCS, _MMIO(0x7000), 0xffff, true}, + {RCS, _MMIO(0x7010), 0xffff, true}, + {RCS, _MMIO(0x7300), 0xffff, true}, + {RCS, _MMIO(0x83a4), 0xffff, true}, + + {RCS, _MMIO(0x40e0), 0, false}, + {RCS, _MMIO(0x40e4), 0, false}, + {RCS, _MMIO(0x2580), 0xffff, true}, + {RCS, _MMIO(0x7014), 0xffff, true}, + {RCS, _MMIO(0x20ec), 0xffff, false}, + {RCS, _MMIO(0xb118), 0, false}, + {RCS, _MMIO(0xe100), 0xffff, true}, + {RCS, _MMIO(0xe180), 0xffff, true}, + {RCS, _MMIO(0xe184), 0xffff, true}, + {RCS, _MMIO(0xe188), 0xffff, true}, + {RCS, _MMIO(0xe194), 0xffff, true}, + {RCS, _MMIO(0x4de0), 0, false}, + {RCS, _MMIO(0x4de4), 0, false}, + {RCS, _MMIO(0x4de8), 0, false}, + {RCS, _MMIO(0x4dec), 0, false}, + {RCS, _MMIO(0x4df0), 0, false}, + {RCS, _MMIO(0x4df4), 0, false}, + + {BCS, _MMIO(0x2229c), 0xffff, false}, + {BCS, _MMIO(0x2209c), 0xffff, false}, + {BCS, _MMIO(0x220c0), 0xffff, false}, + {BCS, _MMIO(0x22098), 0x0, false}, + {BCS, _MMIO(0x22028), 0x0, false}, + + {VCS2, _MMIO(0x1c028), 0xffff, false}, + + {VECS, _MMIO(0x1a028), 0xffff, false}, +}; + +static u32 gen9_render_mocs[I915_NUM_ENGINES][64]; +static u32 gen9_render_mocs_L3[32]; + +static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + enum forcewake_domains fw; + i915_reg_t reg; + u32 regs[] = { + [RCS] = 0x4260, + [VCS] = 0x4264, + [VCS2] = 0x4268, + [BCS] = 0x426c, + [VECS] = 0x4270, + }; + + if (WARN_ON(ring_id >= ARRAY_SIZE(regs))) + return; + + if (!test_and_clear_bit(ring_id, (void *)vgpu->tlb_handle_pending)) + return; + + reg = _MMIO(regs[ring_id]); + + /* WaForceWakeRenderDuringMmioTLBInvalidate:skl + * we need to put a forcewake when invalidating RCS TLB caches, + * otherwise device can go to RC6 state and interrupt invalidation + * process + */ + fw = intel_uncore_forcewake_for_reg(dev_priv, reg, + FW_REG_READ | FW_REG_WRITE); + if (ring_id == RCS && IS_SKYLAKE(dev_priv)) + fw |= FORCEWAKE_RENDER; + + intel_uncore_forcewake_get(dev_priv, fw); + + I915_WRITE_FW(reg, 0x1); + + if (wait_for_atomic((I915_READ_FW(reg) == 0), 50)) + gvt_err("timeout in invalidate ring (%d) tlb\n", ring_id); + else + vgpu_vreg(vgpu, regs[ring_id]) = 0; + + intel_uncore_forcewake_put(dev_priv, fw); + + gvt_dbg_core("invalidate TLB for ring %d\n", ring_id); +} + +static void load_mocs(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + i915_reg_t offset, l3_offset; + u32 regs[] = { + [RCS] = 0xc800, + [VCS] = 0xc900, + [VCS2] = 0xca00, + [BCS] = 0xcc00, + [VECS] = 0xcb00, + }; + int i; + + if (WARN_ON(ring_id >= ARRAY_SIZE(regs))) + return; + + if (!IS_SKYLAKE(dev_priv)) + return; + + offset.reg = regs[ring_id]; + for (i = 0; i < 64; i++) { + gen9_render_mocs[ring_id][i] = I915_READ(offset); + I915_WRITE(offset, vgpu_vreg(vgpu, offset)); + POSTING_READ(offset); + offset.reg += 4; + } + + if (ring_id == RCS) { + l3_offset.reg = 0xb020; + for (i = 0; i < 32; i++) { + gen9_render_mocs_L3[i] = I915_READ(l3_offset); + I915_WRITE(l3_offset, vgpu_vreg(vgpu, offset)); + POSTING_READ(l3_offset); + l3_offset.reg += 4; + } + } +} + +static void restore_mocs(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + i915_reg_t offset, l3_offset; + u32 regs[] = { + [RCS] = 0xc800, + [VCS] = 0xc900, + [VCS2] = 0xca00, + [BCS] = 0xcc00, + [VECS] = 0xcb00, + }; + int i; + + if (WARN_ON(ring_id >= ARRAY_SIZE(regs))) + return; + + if (!IS_SKYLAKE(dev_priv)) + return; + + offset.reg = regs[ring_id]; + for (i = 0; i < 64; i++) { + vgpu_vreg(vgpu, offset) = I915_READ(offset); + I915_WRITE(offset, gen9_render_mocs[ring_id][i]); + POSTING_READ(offset); + offset.reg += 4; + } + + if (ring_id == RCS) { + l3_offset.reg = 0xb020; + for (i = 0; i < 32; i++) { + vgpu_vreg(vgpu, l3_offset) = I915_READ(l3_offset); + I915_WRITE(l3_offset, gen9_render_mocs_L3[i]); + POSTING_READ(l3_offset); + l3_offset.reg += 4; + } + } +} + +void intel_gvt_load_render_mmio(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct render_mmio *mmio; + u32 v; + int i, array_size; + + if (IS_SKYLAKE(vgpu->gvt->dev_priv)) { + mmio = gen9_render_mmio_list; + array_size = ARRAY_SIZE(gen9_render_mmio_list); + load_mocs(vgpu, ring_id); + } else { + mmio = gen8_render_mmio_list; + array_size = ARRAY_SIZE(gen8_render_mmio_list); + } + + for (i = 0; i < array_size; i++, mmio++) { + if (mmio->ring_id != ring_id) + continue; + + mmio->value = I915_READ(mmio->reg); + if (mmio->mask) + v = vgpu_vreg(vgpu, mmio->reg) | (mmio->mask << 16); + else + v = vgpu_vreg(vgpu, mmio->reg); + + I915_WRITE(mmio->reg, v); + POSTING_READ(mmio->reg); + + gvt_dbg_render("load reg %x old %x new %x\n", + i915_mmio_reg_offset(mmio->reg), + mmio->value, v); + } + handle_tlb_pending_event(vgpu, ring_id); +} + +void intel_gvt_restore_render_mmio(struct intel_vgpu *vgpu, int ring_id) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + struct render_mmio *mmio; + u32 v; + int i, array_size; + + if (IS_SKYLAKE(dev_priv)) { + mmio = gen9_render_mmio_list; + array_size = ARRAY_SIZE(gen9_render_mmio_list); + restore_mocs(vgpu, ring_id); + } else { + mmio = gen8_render_mmio_list; + array_size = ARRAY_SIZE(gen8_render_mmio_list); + } + + for (i = 0; i < array_size; i++, mmio++) { + if (mmio->ring_id != ring_id) + continue; + + vgpu_vreg(vgpu, mmio->reg) = I915_READ(mmio->reg); + + if (mmio->mask) { + vgpu_vreg(vgpu, mmio->reg) &= ~(mmio->mask << 16); + v = mmio->value | (mmio->mask << 16); + } else + v = mmio->value; + + I915_WRITE(mmio->reg, v); + POSTING_READ(mmio->reg); + + gvt_dbg_render("restore reg %x old %x new %x\n", + i915_mmio_reg_offset(mmio->reg), + mmio->value, v); + } +} diff --git a/drivers/gpu/drm/i915/gvt/render.h b/drivers/gpu/drm/i915/gvt/render.h new file mode 100644 index 000000000000..dac1a3cc458b --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/render.h @@ -0,0 +1,43 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Kevin Tian <kevin.tian@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * Changbin Du <changbin.du@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#ifndef __GVT_RENDER_H__ +#define __GVT_RENDER_H__ + +void intel_gvt_load_render_mmio(struct intel_vgpu *vgpu, int ring_id); + +void intel_gvt_restore_render_mmio(struct intel_vgpu *vgpu, int ring_id); + +#endif diff --git a/drivers/gpu/drm/i915/gvt/sched_policy.c b/drivers/gpu/drm/i915/gvt/sched_policy.c new file mode 100644 index 000000000000..678b0be85376 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/sched_policy.c @@ -0,0 +1,292 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Anhua Xu + * Kevin Tian <kevin.tian@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" + +static bool vgpu_has_pending_workload(struct intel_vgpu *vgpu) +{ + enum intel_engine_id i; + struct intel_engine_cs *engine; + + for_each_engine(engine, vgpu->gvt->dev_priv, i) { + if (!list_empty(workload_q_head(vgpu, i))) + return true; + } + + return false; +} + +static void try_to_schedule_next_vgpu(struct intel_gvt *gvt) +{ + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + enum intel_engine_id i; + struct intel_engine_cs *engine; + + /* no target to schedule */ + if (!scheduler->next_vgpu) + return; + + gvt_dbg_sched("try to schedule next vgpu %d\n", + scheduler->next_vgpu->id); + + /* + * after the flag is set, workload dispatch thread will + * stop dispatching workload for current vgpu + */ + scheduler->need_reschedule = true; + + /* still have uncompleted workload? */ + for_each_engine(engine, gvt->dev_priv, i) { + if (scheduler->current_workload[i]) { + gvt_dbg_sched("still have running workload\n"); + return; + } + } + + gvt_dbg_sched("switch to next vgpu %d\n", + scheduler->next_vgpu->id); + + /* switch current vgpu */ + scheduler->current_vgpu = scheduler->next_vgpu; + scheduler->next_vgpu = NULL; + + scheduler->need_reschedule = false; + + /* wake up workload dispatch thread */ + for_each_engine(engine, gvt->dev_priv, i) + wake_up(&scheduler->waitq[i]); +} + +struct tbs_vgpu_data { + struct list_head list; + struct intel_vgpu *vgpu; + /* put some per-vgpu sched stats here */ +}; + +struct tbs_sched_data { + struct intel_gvt *gvt; + struct delayed_work work; + unsigned long period; + struct list_head runq_head; +}; + +#define GVT_DEFAULT_TIME_SLICE (1 * HZ / 1000) + +static void tbs_sched_func(struct work_struct *work) +{ + struct tbs_sched_data *sched_data = container_of(work, + struct tbs_sched_data, work.work); + struct tbs_vgpu_data *vgpu_data; + + struct intel_gvt *gvt = sched_data->gvt; + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + + struct intel_vgpu *vgpu = NULL; + struct list_head *pos, *head; + + mutex_lock(&gvt->lock); + + /* no vgpu or has already had a target */ + if (list_empty(&sched_data->runq_head) || scheduler->next_vgpu) + goto out; + + if (scheduler->current_vgpu) { + vgpu_data = scheduler->current_vgpu->sched_data; + head = &vgpu_data->list; + } else { + gvt_dbg_sched("no current vgpu search from q head\n"); + head = &sched_data->runq_head; + } + + /* search a vgpu with pending workload */ + list_for_each(pos, head) { + if (pos == &sched_data->runq_head) + continue; + + vgpu_data = container_of(pos, struct tbs_vgpu_data, list); + if (!vgpu_has_pending_workload(vgpu_data->vgpu)) + continue; + + vgpu = vgpu_data->vgpu; + break; + } + + if (vgpu) { + scheduler->next_vgpu = vgpu; + gvt_dbg_sched("pick next vgpu %d\n", vgpu->id); + } +out: + if (scheduler->next_vgpu) { + gvt_dbg_sched("try to schedule next vgpu %d\n", + scheduler->next_vgpu->id); + try_to_schedule_next_vgpu(gvt); + } + + /* + * still have vgpu on runq + * or last schedule haven't finished due to running workload + */ + if (!list_empty(&sched_data->runq_head) || scheduler->next_vgpu) + schedule_delayed_work(&sched_data->work, sched_data->period); + + mutex_unlock(&gvt->lock); +} + +static int tbs_sched_init(struct intel_gvt *gvt) +{ + struct intel_gvt_workload_scheduler *scheduler = + &gvt->scheduler; + + struct tbs_sched_data *data; + + data = kzalloc(sizeof(*data), GFP_KERNEL); + if (!data) + return -ENOMEM; + + INIT_LIST_HEAD(&data->runq_head); + INIT_DELAYED_WORK(&data->work, tbs_sched_func); + data->period = GVT_DEFAULT_TIME_SLICE; + data->gvt = gvt; + + scheduler->sched_data = data; + return 0; +} + +static void tbs_sched_clean(struct intel_gvt *gvt) +{ + struct intel_gvt_workload_scheduler *scheduler = + &gvt->scheduler; + struct tbs_sched_data *data = scheduler->sched_data; + + cancel_delayed_work(&data->work); + kfree(data); + scheduler->sched_data = NULL; +} + +static int tbs_sched_init_vgpu(struct intel_vgpu *vgpu) +{ + struct tbs_vgpu_data *data; + + data = kzalloc(sizeof(*data), GFP_KERNEL); + if (!data) + return -ENOMEM; + + data->vgpu = vgpu; + INIT_LIST_HEAD(&data->list); + + vgpu->sched_data = data; + return 0; +} + +static void tbs_sched_clean_vgpu(struct intel_vgpu *vgpu) +{ + kfree(vgpu->sched_data); + vgpu->sched_data = NULL; +} + +static void tbs_sched_start_schedule(struct intel_vgpu *vgpu) +{ + struct tbs_sched_data *sched_data = vgpu->gvt->scheduler.sched_data; + struct tbs_vgpu_data *vgpu_data = vgpu->sched_data; + + if (!list_empty(&vgpu_data->list)) + return; + + list_add_tail(&vgpu_data->list, &sched_data->runq_head); + schedule_delayed_work(&sched_data->work, sched_data->period); +} + +static void tbs_sched_stop_schedule(struct intel_vgpu *vgpu) +{ + struct tbs_vgpu_data *vgpu_data = vgpu->sched_data; + + list_del_init(&vgpu_data->list); +} + +static struct intel_gvt_sched_policy_ops tbs_schedule_ops = { + .init = tbs_sched_init, + .clean = tbs_sched_clean, + .init_vgpu = tbs_sched_init_vgpu, + .clean_vgpu = tbs_sched_clean_vgpu, + .start_schedule = tbs_sched_start_schedule, + .stop_schedule = tbs_sched_stop_schedule, +}; + +int intel_gvt_init_sched_policy(struct intel_gvt *gvt) +{ + gvt->scheduler.sched_ops = &tbs_schedule_ops; + + return gvt->scheduler.sched_ops->init(gvt); +} + +void intel_gvt_clean_sched_policy(struct intel_gvt *gvt) +{ + gvt->scheduler.sched_ops->clean(gvt); +} + +int intel_vgpu_init_sched_policy(struct intel_vgpu *vgpu) +{ + return vgpu->gvt->scheduler.sched_ops->init_vgpu(vgpu); +} + +void intel_vgpu_clean_sched_policy(struct intel_vgpu *vgpu) +{ + vgpu->gvt->scheduler.sched_ops->clean_vgpu(vgpu); +} + +void intel_vgpu_start_schedule(struct intel_vgpu *vgpu) +{ + gvt_dbg_core("vgpu%d: start schedule\n", vgpu->id); + + vgpu->gvt->scheduler.sched_ops->start_schedule(vgpu); +} + +void intel_vgpu_stop_schedule(struct intel_vgpu *vgpu) +{ + struct intel_gvt_workload_scheduler *scheduler = + &vgpu->gvt->scheduler; + + gvt_dbg_core("vgpu%d: stop schedule\n", vgpu->id); + + scheduler->sched_ops->stop_schedule(vgpu); + + if (scheduler->next_vgpu == vgpu) + scheduler->next_vgpu = NULL; + + if (scheduler->current_vgpu == vgpu) { + /* stop workload dispatching */ + scheduler->need_reschedule = true; + scheduler->current_vgpu = NULL; + } +} diff --git a/drivers/gpu/drm/i915/gvt/sched_policy.h b/drivers/gpu/drm/i915/gvt/sched_policy.h new file mode 100644 index 000000000000..bb8b9097e41a --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/sched_policy.h @@ -0,0 +1,58 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Anhua Xu + * Kevin Tian <kevin.tian@intel.com> + * + * Contributors: + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#ifndef __GVT_SCHED_POLICY__ +#define __GVT_SCHED_POLICY__ + +struct intel_gvt_sched_policy_ops { + int (*init)(struct intel_gvt *gvt); + void (*clean)(struct intel_gvt *gvt); + int (*init_vgpu)(struct intel_vgpu *vgpu); + void (*clean_vgpu)(struct intel_vgpu *vgpu); + void (*start_schedule)(struct intel_vgpu *vgpu); + void (*stop_schedule)(struct intel_vgpu *vgpu); +}; + +int intel_gvt_init_sched_policy(struct intel_gvt *gvt); + +void intel_gvt_clean_sched_policy(struct intel_gvt *gvt); + +int intel_vgpu_init_sched_policy(struct intel_vgpu *vgpu); + +void intel_vgpu_clean_sched_policy(struct intel_vgpu *vgpu); + +void intel_vgpu_start_schedule(struct intel_vgpu *vgpu); + +void intel_vgpu_stop_schedule(struct intel_vgpu *vgpu); + +#endif diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c new file mode 100644 index 000000000000..4db242250235 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/scheduler.c @@ -0,0 +1,583 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Chanbin Du <changbin.du@intel.com> + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * + */ + +#include <linux/kthread.h> + +#include "i915_drv.h" +#include "gvt.h" + +#define RING_CTX_OFF(x) \ + offsetof(struct execlist_ring_context, x) + +static void set_context_pdp_root_pointer( + struct execlist_ring_context *ring_context, + u32 pdp[8]) +{ + struct execlist_mmio_pair *pdp_pair = &ring_context->pdp3_UDW; + int i; + + for (i = 0; i < 8; i++) + pdp_pair[i].val = pdp[7 - i]; +} + +static int populate_shadow_context(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + int ring_id = workload->ring_id; + struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx; + struct drm_i915_gem_object *ctx_obj = + shadow_ctx->engine[ring_id].state->obj; + struct execlist_ring_context *shadow_ring_context; + struct page *page; + void *dst; + unsigned long context_gpa, context_page_num; + int i; + + gvt_dbg_sched("ring id %d workload lrca %x", ring_id, + workload->ctx_desc.lrca); + + context_page_num = intel_lr_context_size( + gvt->dev_priv->engine[ring_id]); + + context_page_num = context_page_num >> PAGE_SHIFT; + + if (IS_BROADWELL(gvt->dev_priv) && ring_id == RCS) + context_page_num = 19; + + i = 2; + + while (i < context_page_num) { + context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm, + (u32)((workload->ctx_desc.lrca + i) << + GTT_PAGE_SHIFT)); + if (context_gpa == INTEL_GVT_INVALID_ADDR) { + gvt_err("Invalid guest context descriptor\n"); + return -EINVAL; + } + + page = i915_gem_object_get_page(ctx_obj, LRC_PPHWSP_PN + i); + dst = kmap(page); + intel_gvt_hypervisor_read_gpa(vgpu, context_gpa, dst, + GTT_PAGE_SIZE); + kunmap(page); + i++; + } + + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); + shadow_ring_context = kmap(page); + +#define COPY_REG(name) \ + intel_gvt_hypervisor_read_gpa(vgpu, workload->ring_context_gpa \ + + RING_CTX_OFF(name.val), &shadow_ring_context->name.val, 4) + + COPY_REG(ctx_ctrl); + COPY_REG(ctx_timestamp); + + if (ring_id == RCS) { + COPY_REG(bb_per_ctx_ptr); + COPY_REG(rcs_indirect_ctx); + COPY_REG(rcs_indirect_ctx_offset); + } +#undef COPY_REG + + set_context_pdp_root_pointer(shadow_ring_context, + workload->shadow_mm->shadow_page_table); + + intel_gvt_hypervisor_read_gpa(vgpu, + workload->ring_context_gpa + + sizeof(*shadow_ring_context), + (void *)shadow_ring_context + + sizeof(*shadow_ring_context), + GTT_PAGE_SIZE - sizeof(*shadow_ring_context)); + + kunmap(page); + return 0; +} + +static int shadow_context_status_change(struct notifier_block *nb, + unsigned long action, void *data) +{ + struct intel_vgpu *vgpu = container_of(nb, + struct intel_vgpu, shadow_ctx_notifier_block); + struct drm_i915_gem_request *req = + (struct drm_i915_gem_request *)data; + struct intel_gvt_workload_scheduler *scheduler = + &vgpu->gvt->scheduler; + struct intel_vgpu_workload *workload = + scheduler->current_workload[req->engine->id]; + + switch (action) { + case INTEL_CONTEXT_SCHEDULE_IN: + intel_gvt_load_render_mmio(workload->vgpu, + workload->ring_id); + atomic_set(&workload->shadow_ctx_active, 1); + break; + case INTEL_CONTEXT_SCHEDULE_OUT: + intel_gvt_restore_render_mmio(workload->vgpu, + workload->ring_id); + atomic_set(&workload->shadow_ctx_active, 0); + break; + default: + WARN_ON(1); + return NOTIFY_OK; + } + wake_up(&workload->shadow_ctx_status_wq); + return NOTIFY_OK; +} + +static int dispatch_workload(struct intel_vgpu_workload *workload) +{ + int ring_id = workload->ring_id; + struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx; + struct drm_i915_private *dev_priv = workload->vgpu->gvt->dev_priv; + struct drm_i915_gem_request *rq; + int ret; + + gvt_dbg_sched("ring id %d prepare to dispatch workload %p\n", + ring_id, workload); + + shadow_ctx->desc_template = workload->ctx_desc.addressing_mode << + GEN8_CTX_ADDRESSING_MODE_SHIFT; + + mutex_lock(&dev_priv->drm.struct_mutex); + + rq = i915_gem_request_alloc(dev_priv->engine[ring_id], shadow_ctx); + if (IS_ERR(rq)) { + gvt_err("fail to allocate gem request\n"); + ret = PTR_ERR(rq); + goto out; + } + + gvt_dbg_sched("ring id %d get i915 gem request %p\n", ring_id, rq); + + workload->req = i915_gem_request_get(rq); + + ret = intel_gvt_scan_and_shadow_workload(workload); + if (ret) + goto out; + + ret = intel_gvt_scan_and_shadow_wa_ctx(&workload->wa_ctx); + if (ret) + goto out; + + ret = populate_shadow_context(workload); + if (ret) + goto out; + + if (workload->prepare) { + ret = workload->prepare(workload); + if (ret) + goto out; + } + + gvt_dbg_sched("ring id %d submit workload to i915 %p\n", + ring_id, workload->req); + + ret = 0; + workload->dispatched = true; +out: + if (ret) + workload->status = ret; + + if (!IS_ERR_OR_NULL(rq)) + i915_add_request_no_flush(rq); + mutex_unlock(&dev_priv->drm.struct_mutex); + return ret; +} + +static struct intel_vgpu_workload *pick_next_workload( + struct intel_gvt *gvt, int ring_id) +{ + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + struct intel_vgpu_workload *workload = NULL; + + mutex_lock(&gvt->lock); + + /* + * no current vgpu / will be scheduled out / no workload + * bail out + */ + if (!scheduler->current_vgpu) { + gvt_dbg_sched("ring id %d stop - no current vgpu\n", ring_id); + goto out; + } + + if (scheduler->need_reschedule) { + gvt_dbg_sched("ring id %d stop - will reschedule\n", ring_id); + goto out; + } + + if (list_empty(workload_q_head(scheduler->current_vgpu, ring_id))) { + gvt_dbg_sched("ring id %d stop - no available workload\n", + ring_id); + goto out; + } + + /* + * still have current workload, maybe the workload disptacher + * fail to submit it for some reason, resubmit it. + */ + if (scheduler->current_workload[ring_id]) { + workload = scheduler->current_workload[ring_id]; + gvt_dbg_sched("ring id %d still have current workload %p\n", + ring_id, workload); + goto out; + } + + /* + * pick a workload as current workload + * once current workload is set, schedule policy routines + * will wait the current workload is finished when trying to + * schedule out a vgpu. + */ + scheduler->current_workload[ring_id] = container_of( + workload_q_head(scheduler->current_vgpu, ring_id)->next, + struct intel_vgpu_workload, list); + + workload = scheduler->current_workload[ring_id]; + + gvt_dbg_sched("ring id %d pick new workload %p\n", ring_id, workload); + + atomic_inc(&workload->vgpu->running_workload_num); +out: + mutex_unlock(&gvt->lock); + return workload; +} + +static void update_guest_context(struct intel_vgpu_workload *workload) +{ + struct intel_vgpu *vgpu = workload->vgpu; + struct intel_gvt *gvt = vgpu->gvt; + int ring_id = workload->ring_id; + struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx; + struct drm_i915_gem_object *ctx_obj = + shadow_ctx->engine[ring_id].state->obj; + struct execlist_ring_context *shadow_ring_context; + struct page *page; + void *src; + unsigned long context_gpa, context_page_num; + int i; + + gvt_dbg_sched("ring id %d workload lrca %x\n", ring_id, + workload->ctx_desc.lrca); + + context_page_num = intel_lr_context_size( + gvt->dev_priv->engine[ring_id]); + + context_page_num = context_page_num >> PAGE_SHIFT; + + if (IS_BROADWELL(gvt->dev_priv) && ring_id == RCS) + context_page_num = 19; + + i = 2; + + while (i < context_page_num) { + context_gpa = intel_vgpu_gma_to_gpa(vgpu->gtt.ggtt_mm, + (u32)((workload->ctx_desc.lrca + i) << + GTT_PAGE_SHIFT)); + if (context_gpa == INTEL_GVT_INVALID_ADDR) { + gvt_err("invalid guest context descriptor\n"); + return; + } + + page = i915_gem_object_get_page(ctx_obj, LRC_PPHWSP_PN + i); + src = kmap(page); + intel_gvt_hypervisor_write_gpa(vgpu, context_gpa, src, + GTT_PAGE_SIZE); + kunmap(page); + i++; + } + + intel_gvt_hypervisor_write_gpa(vgpu, workload->ring_context_gpa + + RING_CTX_OFF(ring_header.val), &workload->rb_tail, 4); + + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); + shadow_ring_context = kmap(page); + +#define COPY_REG(name) \ + intel_gvt_hypervisor_write_gpa(vgpu, workload->ring_context_gpa + \ + RING_CTX_OFF(name.val), &shadow_ring_context->name.val, 4) + + COPY_REG(ctx_ctrl); + COPY_REG(ctx_timestamp); + +#undef COPY_REG + + intel_gvt_hypervisor_write_gpa(vgpu, + workload->ring_context_gpa + + sizeof(*shadow_ring_context), + (void *)shadow_ring_context + + sizeof(*shadow_ring_context), + GTT_PAGE_SIZE - sizeof(*shadow_ring_context)); + + kunmap(page); +} + +static void complete_current_workload(struct intel_gvt *gvt, int ring_id) +{ + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + struct intel_vgpu_workload *workload; + int event; + + mutex_lock(&gvt->lock); + + workload = scheduler->current_workload[ring_id]; + + if (!workload->status && !workload->vgpu->resetting) { + wait_event(workload->shadow_ctx_status_wq, + !atomic_read(&workload->shadow_ctx_active)); + + update_guest_context(workload); + + for_each_set_bit(event, workload->pending_events, + INTEL_GVT_EVENT_MAX) + intel_vgpu_trigger_virtual_event(workload->vgpu, + event); + } + + gvt_dbg_sched("ring id %d complete workload %p status %d\n", + ring_id, workload, workload->status); + + scheduler->current_workload[ring_id] = NULL; + + atomic_dec(&workload->vgpu->running_workload_num); + + list_del_init(&workload->list); + workload->complete(workload); + + wake_up(&scheduler->workload_complete_wq); + mutex_unlock(&gvt->lock); +} + +struct workload_thread_param { + struct intel_gvt *gvt; + int ring_id; +}; + +static DEFINE_MUTEX(scheduler_mutex); + +static int workload_thread(void *priv) +{ + struct workload_thread_param *p = (struct workload_thread_param *)priv; + struct intel_gvt *gvt = p->gvt; + int ring_id = p->ring_id; + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + struct intel_vgpu_workload *workload = NULL; + long lret; + int ret; + bool need_force_wake = IS_SKYLAKE(gvt->dev_priv); + DEFINE_WAIT_FUNC(wait, woken_wake_function); + + kfree(p); + + gvt_dbg_core("workload thread for ring %d started\n", ring_id); + + while (!kthread_should_stop()) { + add_wait_queue(&scheduler->waitq[ring_id], &wait); + do { + workload = pick_next_workload(gvt, ring_id); + if (workload) + break; + wait_woken(&wait, TASK_INTERRUPTIBLE, + MAX_SCHEDULE_TIMEOUT); + } while (!kthread_should_stop()); + remove_wait_queue(&scheduler->waitq[ring_id], &wait); + + if (!workload) + break; + + mutex_lock(&scheduler_mutex); + + gvt_dbg_sched("ring id %d next workload %p vgpu %d\n", + workload->ring_id, workload, + workload->vgpu->id); + + intel_runtime_pm_get(gvt->dev_priv); + + gvt_dbg_sched("ring id %d will dispatch workload %p\n", + workload->ring_id, workload); + + if (need_force_wake) + intel_uncore_forcewake_get(gvt->dev_priv, + FORCEWAKE_ALL); + + mutex_lock(&gvt->lock); + ret = dispatch_workload(workload); + mutex_unlock(&gvt->lock); + + if (ret) { + gvt_err("fail to dispatch workload, skip\n"); + goto complete; + } + + gvt_dbg_sched("ring id %d wait workload %p\n", + workload->ring_id, workload); + + lret = i915_wait_request(workload->req, + 0, MAX_SCHEDULE_TIMEOUT); + if (lret < 0) { + workload->status = lret; + gvt_err("fail to wait workload, skip\n"); + } else { + workload->status = 0; + } + +complete: + gvt_dbg_sched("will complete workload %p\n, status: %d\n", + workload, workload->status); + + complete_current_workload(gvt, ring_id); + + if (workload->req) + i915_gem_request_put(fetch_and_zero(&workload->req)); + + if (need_force_wake) + intel_uncore_forcewake_put(gvt->dev_priv, + FORCEWAKE_ALL); + + intel_runtime_pm_put(gvt->dev_priv); + + mutex_unlock(&scheduler_mutex); + + } + return 0; +} + +void intel_gvt_wait_vgpu_idle(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + + if (atomic_read(&vgpu->running_workload_num)) { + gvt_dbg_sched("wait vgpu idle\n"); + + wait_event(scheduler->workload_complete_wq, + !atomic_read(&vgpu->running_workload_num)); + } +} + +void intel_gvt_clean_workload_scheduler(struct intel_gvt *gvt) +{ + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + int i; + + gvt_dbg_core("clean workload scheduler\n"); + + for (i = 0; i < I915_NUM_ENGINES; i++) { + if (scheduler->thread[i]) { + kthread_stop(scheduler->thread[i]); + scheduler->thread[i] = NULL; + } + } +} + +int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt) +{ + struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler; + struct workload_thread_param *param = NULL; + int ret; + int i; + + gvt_dbg_core("init workload scheduler\n"); + + init_waitqueue_head(&scheduler->workload_complete_wq); + + for (i = 0; i < I915_NUM_ENGINES; i++) { + /* check ring mask at init time */ + if (!HAS_ENGINE(gvt->dev_priv, i)) + continue; + + init_waitqueue_head(&scheduler->waitq[i]); + + param = kzalloc(sizeof(*param), GFP_KERNEL); + if (!param) { + ret = -ENOMEM; + goto err; + } + + param->gvt = gvt; + param->ring_id = i; + + scheduler->thread[i] = kthread_run(workload_thread, param, + "gvt workload %d", i); + if (IS_ERR(scheduler->thread[i])) { + gvt_err("fail to create workload thread\n"); + ret = PTR_ERR(scheduler->thread[i]); + goto err; + } + } + return 0; +err: + intel_gvt_clean_workload_scheduler(gvt); + kfree(param); + param = NULL; + return ret; +} + +void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu) +{ + struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; + + atomic_notifier_chain_unregister(&vgpu->shadow_ctx->status_notifier, + &vgpu->shadow_ctx_notifier_block); + + mutex_lock(&dev_priv->drm.struct_mutex); + + /* a little hacky to mark as ctx closed */ + vgpu->shadow_ctx->closed = true; + i915_gem_context_put(vgpu->shadow_ctx); + + mutex_unlock(&dev_priv->drm.struct_mutex); +} + +int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu) +{ + atomic_set(&vgpu->running_workload_num, 0); + + vgpu->shadow_ctx = i915_gem_context_create_gvt( + &vgpu->gvt->dev_priv->drm); + if (IS_ERR(vgpu->shadow_ctx)) + return PTR_ERR(vgpu->shadow_ctx); + + vgpu->shadow_ctx->engine[RCS].initialised = true; + + vgpu->shadow_ctx_notifier_block.notifier_call = + shadow_context_status_change; + + atomic_notifier_chain_register(&vgpu->shadow_ctx->status_notifier, + &vgpu->shadow_ctx_notifier_block); + return 0; +} diff --git a/drivers/gpu/drm/i915/gvt/scheduler.h b/drivers/gpu/drm/i915/gvt/scheduler.h new file mode 100644 index 000000000000..3b30c28bff51 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/scheduler.h @@ -0,0 +1,139 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Zhi Wang <zhi.a.wang@intel.com> + * + * Contributors: + * Ping Gao <ping.a.gao@intel.com> + * Tina Zhang <tina.zhang@intel.com> + * Chanbin Du <changbin.du@intel.com> + * Min He <min.he@intel.com> + * Bing Niu <bing.niu@intel.com> + * Zhenyu Wang <zhenyuw@linux.intel.com> + * + */ + +#ifndef _GVT_SCHEDULER_H_ +#define _GVT_SCHEDULER_H_ + +struct intel_gvt_workload_scheduler { + struct intel_vgpu *current_vgpu; + struct intel_vgpu *next_vgpu; + struct intel_vgpu_workload *current_workload[I915_NUM_ENGINES]; + bool need_reschedule; + + wait_queue_head_t workload_complete_wq; + struct task_struct *thread[I915_NUM_ENGINES]; + wait_queue_head_t waitq[I915_NUM_ENGINES]; + + void *sched_data; + struct intel_gvt_sched_policy_ops *sched_ops; +}; + +#define INDIRECT_CTX_ADDR_MASK 0xffffffc0 +#define INDIRECT_CTX_SIZE_MASK 0x3f +struct shadow_indirect_ctx { + struct drm_i915_gem_object *obj; + unsigned long guest_gma; + unsigned long shadow_gma; + void *shadow_va; + uint32_t size; +}; + +#define PER_CTX_ADDR_MASK 0xfffff000 +struct shadow_per_ctx { + unsigned long guest_gma; + unsigned long shadow_gma; +}; + +struct intel_shadow_wa_ctx { + struct intel_vgpu_workload *workload; + struct shadow_indirect_ctx indirect_ctx; + struct shadow_per_ctx per_ctx; + +}; + +struct intel_vgpu_workload { + struct intel_vgpu *vgpu; + int ring_id; + struct drm_i915_gem_request *req; + /* if this workload has been dispatched to i915? */ + bool dispatched; + int status; + + struct intel_vgpu_mm *shadow_mm; + + /* different submission model may need different handler */ + int (*prepare)(struct intel_vgpu_workload *); + int (*complete)(struct intel_vgpu_workload *); + struct list_head list; + + DECLARE_BITMAP(pending_events, INTEL_GVT_EVENT_MAX); + void *shadow_ring_buffer_va; + + /* execlist context information */ + struct execlist_ctx_descriptor_format ctx_desc; + struct execlist_ring_context *ring_context; + unsigned long rb_head, rb_tail, rb_ctl, rb_start, rb_len; + bool restore_inhibit; + struct intel_vgpu_elsp_dwords elsp_dwords; + bool emulate_schedule_in; + atomic_t shadow_ctx_active; + wait_queue_head_t shadow_ctx_status_wq; + u64 ring_context_gpa; + + /* shadow batch buffer */ + struct list_head shadow_bb; + struct intel_shadow_wa_ctx wa_ctx; +}; + +/* Intel shadow batch buffer is a i915 gem object */ +struct intel_shadow_bb_entry { + struct list_head list; + struct drm_i915_gem_object *obj; + void *va; + unsigned long len; + void *bb_start_cmd_va; +}; + +#define workload_q_head(vgpu, ring_id) \ + (&(vgpu->workload_q_head[ring_id])) + +#define queue_workload(workload) do { \ + list_add_tail(&workload->list, \ + workload_q_head(workload->vgpu, workload->ring_id)); \ + wake_up(&workload->vgpu->gvt-> \ + scheduler.waitq[workload->ring_id]); \ +} while (0) + +int intel_gvt_init_workload_scheduler(struct intel_gvt *gvt); + +void intel_gvt_clean_workload_scheduler(struct intel_gvt *gvt); + +void intel_gvt_wait_vgpu_idle(struct intel_vgpu *vgpu); + +int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu); + +void intel_vgpu_clean_gvt_context(struct intel_vgpu *vgpu); + +#endif diff --git a/drivers/gpu/drm/i915/gvt/trace.h b/drivers/gpu/drm/i915/gvt/trace.h new file mode 100644 index 000000000000..53a2d10cf3f1 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/trace.h @@ -0,0 +1,286 @@ +/* + * Copyright © 2011-2016 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Jike Song <jike.song@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#if !defined(_GVT_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ) +#define _GVT_TRACE_H_ + +#include <linux/types.h> +#include <linux/stringify.h> +#include <linux/tracepoint.h> +#include <asm/tsc.h> + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM gvt + +TRACE_EVENT(spt_alloc, + TP_PROTO(int id, void *spt, int type, unsigned long mfn, + unsigned long gpt_gfn), + + TP_ARGS(id, spt, type, mfn, gpt_gfn), + + TP_STRUCT__entry( + __field(int, id) + __field(void *, spt) + __field(int, type) + __field(unsigned long, mfn) + __field(unsigned long, gpt_gfn) + ), + + TP_fast_assign( + __entry->id = id; + __entry->spt = spt; + __entry->type = type; + __entry->mfn = mfn; + __entry->gpt_gfn = gpt_gfn; + ), + + TP_printk("VM%d [alloc] spt %p type %d mfn 0x%lx gfn 0x%lx\n", + __entry->id, + __entry->spt, + __entry->type, + __entry->mfn, + __entry->gpt_gfn) +); + +TRACE_EVENT(spt_free, + TP_PROTO(int id, void *spt, int type), + + TP_ARGS(id, spt, type), + + TP_STRUCT__entry( + __field(int, id) + __field(void *, spt) + __field(int, type) + ), + + TP_fast_assign( + __entry->id = id; + __entry->spt = spt; + __entry->type = type; + ), + + TP_printk("VM%u [free] spt %p type %d\n", + __entry->id, + __entry->spt, + __entry->type) +); + +#define MAX_BUF_LEN 256 + +TRACE_EVENT(gma_index, + TP_PROTO(const char *prefix, unsigned long gma, + unsigned long index), + + TP_ARGS(prefix, gma, index), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "%s gma 0x%lx index 0x%lx\n", prefix, gma, index); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(gma_translate, + TP_PROTO(int id, char *type, int ring_id, int pt_level, + unsigned long gma, unsigned long gpa), + + TP_ARGS(id, type, ring_id, pt_level, gma, gpa), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d %s ring %d pt_level %d gma 0x%lx -> gpa 0x%lx\n", + id, type, ring_id, pt_level, gma, gpa); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(spt_refcount, + TP_PROTO(int id, char *action, void *spt, int before, int after), + + TP_ARGS(id, action, spt, before, after), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d [%s] spt %p before %d -> after %d\n", + id, action, spt, before, after); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(spt_change, + TP_PROTO(int id, char *action, void *spt, unsigned long gfn, + int type), + + TP_ARGS(id, action, spt, gfn, type), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d [%s] spt %p gfn 0x%lx type %d\n", + id, action, spt, gfn, type); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(gpt_change, + TP_PROTO(int id, const char *tag, void *spt, int type, u64 v, + unsigned long index), + + TP_ARGS(id, tag, spt, type, v, index), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d [%s] spt %p type %d entry 0x%llx index 0x%lx\n", + id, tag, spt, type, v, index); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(oos_change, + TP_PROTO(int id, const char *tag, int page_id, void *gpt, int type), + + TP_ARGS(id, tag, page_id, gpt, type), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d [oos %s] page id %d gpt %p type %d\n", + id, tag, page_id, gpt, type); + ), + + TP_printk("%s", __entry->buf) +); + +TRACE_EVENT(oos_sync, + TP_PROTO(int id, int page_id, void *gpt, int type, u64 v, + unsigned long index), + + TP_ARGS(id, page_id, gpt, type, v, index), + + TP_STRUCT__entry( + __array(char, buf, MAX_BUF_LEN) + ), + + TP_fast_assign( + snprintf(__entry->buf, MAX_BUF_LEN, + "VM%d [oos sync] page id %d gpt %p type %d entry 0x%llx index 0x%lx\n", + id, page_id, gpt, type, v, index); + ), + + TP_printk("%s", __entry->buf) +); + +#define MAX_CMD_STR_LEN 256 +TRACE_EVENT(gvt_command, + TP_PROTO(u8 vm_id, u8 ring_id, u32 ip_gma, u32 *cmd_va, u32 cmd_len, bool ring_buffer_cmd, cycles_t cost_pre_cmd_handler, cycles_t cost_cmd_handler), + + TP_ARGS(vm_id, ring_id, ip_gma, cmd_va, cmd_len, ring_buffer_cmd, cost_pre_cmd_handler, cost_cmd_handler), + + TP_STRUCT__entry( + __field(u8, vm_id) + __field(u8, ring_id) + __field(int, i) + __array(char, tmp_buf, MAX_CMD_STR_LEN) + __array(char, cmd_str, MAX_CMD_STR_LEN) + ), + + TP_fast_assign( + __entry->vm_id = vm_id; + __entry->ring_id = ring_id; + __entry->cmd_str[0] = '\0'; + snprintf(__entry->tmp_buf, MAX_CMD_STR_LEN, "VM(%d) Ring(%d): %s ip(%08x) pre handler cost (%llu), handler cost (%llu) ", vm_id, ring_id, ring_buffer_cmd ? "RB":"BB", ip_gma, cost_pre_cmd_handler, cost_cmd_handler); + strcat(__entry->cmd_str, __entry->tmp_buf); + entry->i = 0; + while (cmd_len > 0) { + if (cmd_len >= 8) { + snprintf(__entry->tmp_buf, MAX_CMD_STR_LEN, "%08x %08x %08x %08x %08x %08x %08x %08x ", + cmd_va[__entry->i], cmd_va[__entry->i+1], cmd_va[__entry->i+2], cmd_va[__entry->i+3], + cmd_va[__entry->i+4], cmd_va[__entry->i+5], cmd_va[__entry->i+6], cmd_va[__entry->i+7]); + __entry->i += 8; + cmd_len -= 8; + strcat(__entry->cmd_str, __entry->tmp_buf); + } else if (cmd_len >= 4) { + snprintf(__entry->tmp_buf, MAX_CMD_STR_LEN, "%08x %08x %08x %08x ", + cmd_va[__entry->i], cmd_va[__entry->i+1], cmd_va[__entry->i+2], cmd_va[__entry->i+3]); + __entry->i += 4; + cmd_len -= 4; + strcat(__entry->cmd_str, __entry->tmp_buf); + } else if (cmd_len >= 2) { + snprintf(__entry->tmp_buf, MAX_CMD_STR_LEN, "%08x %08x ", cmd_va[__entry->i], cmd_va[__entry->i+1]); + __entry->i += 2; + cmd_len -= 2; + strcat(__entry->cmd_str, __entry->tmp_buf); + } else if (cmd_len == 1) { + snprintf(__entry->tmp_buf, MAX_CMD_STR_LEN, "%08x ", cmd_va[__entry->i]); + __entry->i += 1; + cmd_len -= 1; + strcat(__entry->cmd_str, __entry->tmp_buf); + } + } + strcat(__entry->cmd_str, "\n"); + ), + + TP_printk("%s", __entry->cmd_str) +); +#endif /* _GVT_TRACE_H_ */ + +/* This part must be out of protection */ +#undef TRACE_INCLUDE_PATH +#define TRACE_INCLUDE_PATH . +#undef TRACE_INCLUDE_FILE +#define TRACE_INCLUDE_FILE trace +#include <trace/define_trace.h> diff --git a/drivers/gpu/drm/i915/gvt/trace_points.c b/drivers/gpu/drm/i915/gvt/trace_points.c new file mode 100644 index 000000000000..a3deed692b9c --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/trace_points.c @@ -0,0 +1,36 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Jike Song <jike.song@intel.com> + * + * Contributors: + * Zhi Wang <zhi.a.wang@intel.com> + * + */ + +#include "trace.h" + +#ifndef __CHECKER__ +#define CREATE_TRACE_POINTS +#include "trace.h" +#endif diff --git a/drivers/gpu/drm/i915/gvt/vgpu.c b/drivers/gpu/drm/i915/gvt/vgpu.c new file mode 100644 index 000000000000..536d2b9d5777 --- /dev/null +++ b/drivers/gpu/drm/i915/gvt/vgpu.c @@ -0,0 +1,409 @@ +/* + * Copyright(c) 2011-2016 Intel Corporation. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + * Authors: + * Eddie Dong <eddie.dong@intel.com> + * Kevin Tian <kevin.tian@intel.com> + * + * Contributors: + * Ping Gao <ping.a.gao@intel.com> + * Zhi Wang <zhi.a.wang@intel.com> + * Bing Niu <bing.niu@intel.com> + * + */ + +#include "i915_drv.h" +#include "gvt.h" +#include "i915_pvinfo.h" + +static void clean_vgpu_mmio(struct intel_vgpu *vgpu) +{ + vfree(vgpu->mmio.vreg); + vgpu->mmio.vreg = vgpu->mmio.sreg = NULL; +} + +int setup_vgpu_mmio(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + const struct intel_gvt_device_info *info = &gvt->device_info; + + if (vgpu->mmio.vreg) + memset(vgpu->mmio.vreg, 0, info->mmio_size * 2); + else { + vgpu->mmio.vreg = vzalloc(info->mmio_size * 2); + if (!vgpu->mmio.vreg) + return -ENOMEM; + } + + vgpu->mmio.sreg = vgpu->mmio.vreg + info->mmio_size; + + memcpy(vgpu->mmio.vreg, gvt->firmware.mmio, info->mmio_size); + memcpy(vgpu->mmio.sreg, gvt->firmware.mmio, info->mmio_size); + + vgpu_vreg(vgpu, GEN6_GT_THREAD_STATUS_REG) = 0; + + /* set the bit 0:2(Core C-State ) to C0 */ + vgpu_vreg(vgpu, GEN6_GT_CORE_STATUS) = 0; + return 0; +} + +static void setup_vgpu_cfg_space(struct intel_vgpu *vgpu, + struct intel_vgpu_creation_params *param) +{ + struct intel_gvt *gvt = vgpu->gvt; + const struct intel_gvt_device_info *info = &gvt->device_info; + u16 *gmch_ctl; + int i; + + memcpy(vgpu_cfg_space(vgpu), gvt->firmware.cfg_space, + info->cfg_space_size); + + if (!param->primary) { + vgpu_cfg_space(vgpu)[PCI_CLASS_DEVICE] = + INTEL_GVT_PCI_CLASS_VGA_OTHER; + vgpu_cfg_space(vgpu)[PCI_CLASS_PROG] = + INTEL_GVT_PCI_CLASS_VGA_OTHER; + } + + /* Show guest that there isn't any stolen memory.*/ + gmch_ctl = (u16 *)(vgpu_cfg_space(vgpu) + INTEL_GVT_PCI_GMCH_CONTROL); + *gmch_ctl &= ~(BDW_GMCH_GMS_MASK << BDW_GMCH_GMS_SHIFT); + + intel_vgpu_write_pci_bar(vgpu, PCI_BASE_ADDRESS_2, + gvt_aperture_pa_base(gvt), true); + + vgpu_cfg_space(vgpu)[PCI_COMMAND] &= ~(PCI_COMMAND_IO + | PCI_COMMAND_MEMORY + | PCI_COMMAND_MASTER); + /* + * Clear the bar upper 32bit and let guest to assign the new value + */ + memset(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_1, 0, 4); + memset(vgpu_cfg_space(vgpu) + PCI_BASE_ADDRESS_3, 0, 4); + memset(vgpu_cfg_space(vgpu) + INTEL_GVT_PCI_OPREGION, 0, 4); + + for (i = 0; i < INTEL_GVT_MAX_BAR_NUM; i++) { + vgpu->cfg_space.bar[i].size = pci_resource_len( + gvt->dev_priv->drm.pdev, i * 2); + vgpu->cfg_space.bar[i].tracked = false; + } +} + +void populate_pvinfo_page(struct intel_vgpu *vgpu) +{ + /* setup the ballooning information */ + vgpu_vreg64(vgpu, vgtif_reg(magic)) = VGT_MAGIC; + vgpu_vreg(vgpu, vgtif_reg(version_major)) = 1; + vgpu_vreg(vgpu, vgtif_reg(version_minor)) = 0; + vgpu_vreg(vgpu, vgtif_reg(display_ready)) = 0; + vgpu_vreg(vgpu, vgtif_reg(vgt_id)) = vgpu->id; + vgpu_vreg(vgpu, vgtif_reg(avail_rs.mappable_gmadr.base)) = + vgpu_aperture_gmadr_base(vgpu); + vgpu_vreg(vgpu, vgtif_reg(avail_rs.mappable_gmadr.size)) = + vgpu_aperture_sz(vgpu); + vgpu_vreg(vgpu, vgtif_reg(avail_rs.nonmappable_gmadr.base)) = + vgpu_hidden_gmadr_base(vgpu); + vgpu_vreg(vgpu, vgtif_reg(avail_rs.nonmappable_gmadr.size)) = + vgpu_hidden_sz(vgpu); + + vgpu_vreg(vgpu, vgtif_reg(avail_rs.fence_num)) = vgpu_fence_sz(vgpu); + + gvt_dbg_core("Populate PVINFO PAGE for vGPU %d\n", vgpu->id); + gvt_dbg_core("aperture base [GMADR] 0x%llx size 0x%llx\n", + vgpu_aperture_gmadr_base(vgpu), vgpu_aperture_sz(vgpu)); + gvt_dbg_core("hidden base [GMADR] 0x%llx size=0x%llx\n", + vgpu_hidden_gmadr_base(vgpu), vgpu_hidden_sz(vgpu)); + gvt_dbg_core("fence size %d\n", vgpu_fence_sz(vgpu)); + + WARN_ON(sizeof(struct vgt_if) != VGT_PVINFO_SIZE); +} + +/** + * intel_gvt_init_vgpu_types - initialize vGPU type list + * @gvt : GVT device + * + * Initialize vGPU type list based on available resource. + * + */ +int intel_gvt_init_vgpu_types(struct intel_gvt *gvt) +{ + unsigned int num_types; + unsigned int i, low_avail; + unsigned int min_low; + + /* vGPU type name is defined as GVTg_Vx_y which contains + * physical GPU generation type and 'y' means maximum vGPU + * instances user can create on one physical GPU for this + * type. + * + * Depend on physical SKU resource, might see vGPU types like + * GVTg_V4_8, GVTg_V4_4, GVTg_V4_2, etc. We can create + * different types of vGPU on same physical GPU depending on + * available resource. Each vGPU type will have "avail_instance" + * to indicate how many vGPU instance can be created for this + * type. + * + * Currently use static size here as we init type earlier.. + */ + low_avail = MB_TO_BYTES(256) - HOST_LOW_GM_SIZE; + num_types = 4; + + gvt->types = kzalloc(num_types * sizeof(struct intel_vgpu_type), + GFP_KERNEL); + if (!gvt->types) + return -ENOMEM; + + min_low = MB_TO_BYTES(32); + for (i = 0; i < num_types; ++i) { + if (low_avail / min_low == 0) + break; + gvt->types[i].low_gm_size = min_low; + gvt->types[i].high_gm_size = 3 * gvt->types[i].low_gm_size; + gvt->types[i].fence = 4; + gvt->types[i].max_instance = low_avail / min_low; + gvt->types[i].avail_instance = gvt->types[i].max_instance; + + if (IS_GEN8(gvt->dev_priv)) + sprintf(gvt->types[i].name, "GVTg_V4_%u", + gvt->types[i].max_instance); + else if (IS_GEN9(gvt->dev_priv)) + sprintf(gvt->types[i].name, "GVTg_V5_%u", + gvt->types[i].max_instance); + + min_low <<= 1; + gvt_dbg_core("type[%d]: %s max %u avail %u low %u high %u fence %u\n", + i, gvt->types[i].name, gvt->types[i].max_instance, + gvt->types[i].avail_instance, + gvt->types[i].low_gm_size, + gvt->types[i].high_gm_size, gvt->types[i].fence); + } + + gvt->num_types = i; + return 0; +} + +void intel_gvt_clean_vgpu_types(struct intel_gvt *gvt) +{ + kfree(gvt->types); +} + +static void intel_gvt_update_vgpu_types(struct intel_gvt *gvt) +{ + int i; + unsigned int low_gm_avail, high_gm_avail, fence_avail; + unsigned int low_gm_min, high_gm_min, fence_min, total_min; + + /* Need to depend on maxium hw resource size but keep on + * static config for now. + */ + low_gm_avail = MB_TO_BYTES(256) - HOST_LOW_GM_SIZE - + gvt->gm.vgpu_allocated_low_gm_size; + high_gm_avail = MB_TO_BYTES(256) * 3 - HOST_HIGH_GM_SIZE - + gvt->gm.vgpu_allocated_high_gm_size; + fence_avail = gvt_fence_sz(gvt) - HOST_FENCE - + gvt->fence.vgpu_allocated_fence_num; + + for (i = 0; i < gvt->num_types; i++) { + low_gm_min = low_gm_avail / gvt->types[i].low_gm_size; + high_gm_min = high_gm_avail / gvt->types[i].high_gm_size; + fence_min = fence_avail / gvt->types[i].fence; + total_min = min(min(low_gm_min, high_gm_min), fence_min); + gvt->types[i].avail_instance = min(gvt->types[i].max_instance, + total_min); + + gvt_dbg_core("update type[%d]: %s max %u avail %u low %u high %u fence %u\n", + i, gvt->types[i].name, gvt->types[i].max_instance, + gvt->types[i].avail_instance, gvt->types[i].low_gm_size, + gvt->types[i].high_gm_size, gvt->types[i].fence); + } +} + +/** + * intel_gvt_destroy_vgpu - destroy a virtual GPU + * @vgpu: virtual GPU + * + * This function is called when user wants to destroy a virtual GPU. + * + */ +void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu) +{ + struct intel_gvt *gvt = vgpu->gvt; + + mutex_lock(&gvt->lock); + + vgpu->active = false; + idr_remove(&gvt->vgpu_idr, vgpu->id); + + if (atomic_read(&vgpu->running_workload_num)) { + mutex_unlock(&gvt->lock); + intel_gvt_wait_vgpu_idle(vgpu); + mutex_lock(&gvt->lock); + } + + intel_vgpu_stop_schedule(vgpu); + intel_vgpu_clean_sched_policy(vgpu); + intel_vgpu_clean_gvt_context(vgpu); + intel_vgpu_clean_execlist(vgpu); + intel_vgpu_clean_display(vgpu); + intel_vgpu_clean_opregion(vgpu); + intel_vgpu_clean_gtt(vgpu); + intel_gvt_hypervisor_detach_vgpu(vgpu); + intel_vgpu_free_resource(vgpu); + clean_vgpu_mmio(vgpu); + vfree(vgpu); + + intel_gvt_update_vgpu_types(gvt); + mutex_unlock(&gvt->lock); +} + +static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt, + struct intel_vgpu_creation_params *param) +{ + struct intel_vgpu *vgpu; + int ret; + + gvt_dbg_core("handle %llu low %llu MB high %llu MB fence %llu\n", + param->handle, param->low_gm_sz, param->high_gm_sz, + param->fence_sz); + + vgpu = vzalloc(sizeof(*vgpu)); + if (!vgpu) + return ERR_PTR(-ENOMEM); + + mutex_lock(&gvt->lock); + + ret = idr_alloc(&gvt->vgpu_idr, vgpu, 1, GVT_MAX_VGPU, GFP_KERNEL); + if (ret < 0) + goto out_free_vgpu; + + vgpu->id = ret; + vgpu->handle = param->handle; + vgpu->gvt = gvt; + bitmap_zero(vgpu->tlb_handle_pending, I915_NUM_ENGINES); + + setup_vgpu_cfg_space(vgpu, param); + + ret = setup_vgpu_mmio(vgpu); + if (ret) + goto out_free_vgpu; + + ret = intel_vgpu_alloc_resource(vgpu, param); + if (ret) + goto out_clean_vgpu_mmio; + + populate_pvinfo_page(vgpu); + + ret = intel_gvt_hypervisor_attach_vgpu(vgpu); + if (ret) + goto out_clean_vgpu_resource; + + ret = intel_vgpu_init_gtt(vgpu); + if (ret) + goto out_detach_hypervisor_vgpu; + + ret = intel_vgpu_init_display(vgpu); + if (ret) + goto out_clean_gtt; + + ret = intel_vgpu_init_execlist(vgpu); + if (ret) + goto out_clean_display; + + ret = intel_vgpu_init_gvt_context(vgpu); + if (ret) + goto out_clean_execlist; + + ret = intel_vgpu_init_sched_policy(vgpu); + if (ret) + goto out_clean_shadow_ctx; + + vgpu->active = true; + mutex_unlock(&gvt->lock); + + return vgpu; + +out_clean_shadow_ctx: + intel_vgpu_clean_gvt_context(vgpu); +out_clean_execlist: + intel_vgpu_clean_execlist(vgpu); +out_clean_display: + intel_vgpu_clean_display(vgpu); +out_clean_gtt: + intel_vgpu_clean_gtt(vgpu); +out_detach_hypervisor_vgpu: + intel_gvt_hypervisor_detach_vgpu(vgpu); +out_clean_vgpu_resource: + intel_vgpu_free_resource(vgpu); +out_clean_vgpu_mmio: + clean_vgpu_mmio(vgpu); +out_free_vgpu: + vfree(vgpu); + mutex_unlock(&gvt->lock); + return ERR_PTR(ret); +} + +/** + * intel_gvt_create_vgpu - create a virtual GPU + * @gvt: GVT device + * @type: type of the vGPU to create + * + * This function is called when user wants to create a virtual GPU. + * + * Returns: + * pointer to intel_vgpu, error pointer if failed. + */ +struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt, + struct intel_vgpu_type *type) +{ + struct intel_vgpu_creation_params param; + struct intel_vgpu *vgpu; + + param.handle = 0; + param.primary = 1; + param.low_gm_sz = type->low_gm_size; + param.high_gm_sz = type->high_gm_size; + param.fence_sz = type->fence; + + /* XXX current param based on MB */ + param.low_gm_sz = BYTES_TO_MB(param.low_gm_sz); + param.high_gm_sz = BYTES_TO_MB(param.high_gm_sz); + + vgpu = __intel_gvt_create_vgpu(gvt, ¶m); + if (IS_ERR(vgpu)) + return vgpu; + + /* calculate left instance change for types */ + intel_gvt_update_vgpu_types(gvt); + + return vgpu; +} + +/** + * intel_gvt_reset_vgpu - reset a virtual GPU + * @vgpu: virtual GPU + * + * This function is called when user wants to reset a virtual GPU. + * + */ +void intel_gvt_reset_vgpu(struct intel_vgpu *vgpu) +{ +} |