diff options
author | Dave Airlie <airlied@redhat.com> | 2022-11-09 10:48:44 +1000 |
---|---|---|
committer | Dave Airlie <airlied@redhat.com> | 2022-11-09 11:18:56 +1000 |
commit | a143bc517bf31c4575191efbaac216a11ec016e0 (patch) | |
tree | e88d120c60d148c165551d17493cc3abef651019 /drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c | |
parent | 49e8e6343df688d68b12c2af50791ca37520f0b7 (diff) | |
parent | c4bdac754ca0bb21bad598b47dd8c2c86eee1cc1 (diff) | |
download | linux-a143bc517bf31c4575191efbaac216a11ec016e0.tar.gz |
Merge branch '00.06-gr-ampere' of https://gitlab.freedesktop.org/skeggsb/nouveau into drm-next
This is the pull request for a whole bunch of fixes and prep-work that
was done to support Ampere acceleration prior to GSP-RM being
available. It uses the ACR firmware released by NVIDIA in
linux-firmware, as we do on earlier GPUs. The work to support running
on top of GSP-RM also heavily depends on various pieces of this
series.
In addition to the new HW support, general stability of the driver
should be improved, especially around recovering HW from bugs that can
be generated by userspace driver components.
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Ben Skeggs <bskeggs@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CABDvA==s+nZD0n7CuRWLPE=Pj+02CN13r+ZQJxoHQ_EmR+o=XQ@mail.gmail.com
Diffstat (limited to 'drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c')
-rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c | 116 |
1 files changed, 42 insertions, 74 deletions
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c b/drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c index 44e39f5743d5..c64013d10500 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c +++ b/drivers/gpu/drm/nouveau/nvkm/engine/sec2/gp102.c @@ -74,16 +74,6 @@ gp102_sec2_acr_bootstrap_falcon(struct nvkm_falcon *falcon, msecs_to_jiffies(1000)); } -static int -gp102_sec2_acr_boot(struct nvkm_falcon *falcon) -{ - struct nv_sec2_args args = {}; - nvkm_falcon_load_dmem(falcon, &args, - falcon->func->emem_addr, sizeof(args), 0); - nvkm_falcon_start(falcon); - return 0; -} - static void gp102_sec2_acr_bld_patch(struct nvkm_acr *acr, u32 bld, s64 adjust) { @@ -122,7 +112,6 @@ gp102_sec2_acr_0 = { .bld_size = sizeof(struct loader_config_v1), .bld_write = gp102_sec2_acr_bld_write, .bld_patch = gp102_sec2_acr_bld_patch, - .boot = gp102_sec2_acr_boot, .bootstrap_falcons = BIT_ULL(NVKM_ACR_LSF_FECS) | BIT_ULL(NVKM_ACR_LSF_GPCCS) | BIT_ULL(NVKM_ACR_LSF_SEC2), @@ -160,89 +149,68 @@ gp102_sec2_initmsg(struct nvkm_sec2 *sec2) return 0; } -void -gp102_sec2_intr(struct nvkm_sec2 *sec2) +irqreturn_t +gp102_sec2_intr(struct nvkm_inth *inth) { + struct nvkm_sec2 *sec2 = container_of(inth, typeof(*sec2), engine.subdev.inth); struct nvkm_subdev *subdev = &sec2->engine.subdev; struct nvkm_falcon *falcon = &sec2->falcon; u32 disp = nvkm_falcon_rd32(falcon, 0x01c); u32 intr = nvkm_falcon_rd32(falcon, 0x008) & disp & ~(disp >> 16); if (intr & 0x00000040) { - schedule_work(&sec2->work); + if (unlikely(atomic_read(&sec2->initmsg) == 0)) { + int ret = sec2->func->initmsg(sec2); + + if (ret) + nvkm_error(subdev, "error parsing init message: %d\n", ret); + + atomic_set(&sec2->initmsg, ret ?: 1); + } + + if (atomic_read(&sec2->initmsg) > 0) { + if (!nvkm_falcon_msgq_empty(sec2->msgq)) + nvkm_falcon_msgq_recv(sec2->msgq); + } + nvkm_falcon_wr32(falcon, 0x004, 0x00000040); intr &= ~0x00000040; } + if (intr & 0x00000010) { + if (atomic_read(&sec2->running)) { + FLCN_ERR(falcon, "halted"); + gm200_flcn_tracepc(falcon); + } + + nvkm_falcon_wr32(falcon, 0x004, 0x00000010); + intr &= ~0x00000010; + } + if (intr) { nvkm_error(subdev, "unhandled intr %08x\n", intr); nvkm_falcon_wr32(falcon, 0x004, intr); } -} -int -gp102_sec2_flcn_enable(struct nvkm_falcon *falcon) -{ - nvkm_falcon_mask(falcon, 0x3c0, 0x00000001, 0x00000001); - udelay(10); - nvkm_falcon_mask(falcon, 0x3c0, 0x00000001, 0x00000000); - return nvkm_falcon_v1_enable(falcon); -} - -void -gp102_sec2_flcn_bind_context(struct nvkm_falcon *falcon, - struct nvkm_memory *ctx) -{ - struct nvkm_device *device = falcon->owner->device; - - nvkm_falcon_v1_bind_context(falcon, ctx); - if (!ctx) - return; - - /* Not sure if this is a WAR for a HW issue, or some additional - * programming sequence that's needed to properly complete the - * context switch we trigger above. - * - * Fixes unreliability of booting the SEC2 RTOS on Quadro P620, - * particularly when resuming from suspend. - * - * Also removes the need for an odd workaround where we needed - * to program SEC2's FALCON_CPUCTL_ALIAS_STARTCPU twice before - * the SEC2 RTOS would begin executing. - */ - nvkm_msec(device, 10, - u32 irqstat = nvkm_falcon_rd32(falcon, 0x008); - u32 flcn0dc = nvkm_falcon_rd32(falcon, 0x0dc); - if ((irqstat & 0x00000008) && - (flcn0dc & 0x00007000) == 0x00005000) - break; - ); - - nvkm_falcon_mask(falcon, 0x004, 0x00000008, 0x00000008); - nvkm_falcon_mask(falcon, 0x058, 0x00000002, 0x00000002); - - nvkm_msec(device, 10, - u32 flcn0dc = nvkm_falcon_rd32(falcon, 0x0dc); - if ((flcn0dc & 0x00007000) == 0x00000000) - break; - ); + return IRQ_HANDLED; } static const struct nvkm_falcon_func gp102_sec2_flcn = { + .disable = gm200_flcn_disable, + .enable = gm200_flcn_enable, + .reset_pmc = true, + .reset_eng = gp102_flcn_reset_eng, + .reset_wait_mem_scrubbing = gm200_flcn_reset_wait_mem_scrubbing, .debug = 0x408, - .fbif = 0x600, - .load_imem = nvkm_falcon_v1_load_imem, - .load_dmem = nvkm_falcon_v1_load_dmem, - .read_dmem = nvkm_falcon_v1_read_dmem, + .bind_inst = gm200_flcn_bind_inst, + .bind_stat = gm200_flcn_bind_stat, + .bind_intr = true, + .imem_pio = &gm200_flcn_imem_pio, + .dmem_pio = &gm200_flcn_dmem_pio, .emem_addr = 0x01000000, - .bind_context = gp102_sec2_flcn_bind_context, - .wait_for_halt = nvkm_falcon_v1_wait_for_halt, - .clear_interrupt = nvkm_falcon_v1_clear_interrupt, - .set_start_addr = nvkm_falcon_v1_set_start_addr, + .emem_pio = &gp102_flcn_emem_pio, .start = nvkm_falcon_v1_start, - .enable = gp102_sec2_flcn_enable, - .disable = nvkm_falcon_v1_disable, .cmdq = { 0xa00, 0xa04, 8 }, .msgq = { 0xa30, 0xa34, 8 }, }; @@ -250,6 +218,7 @@ gp102_sec2_flcn = { const struct nvkm_sec2_func gp102_sec2 = { .flcn = &gp102_sec2_flcn, + .unit_unload = NV_SEC2_UNIT_UNLOAD, .unit_acr = NV_SEC2_UNIT_ACR, .intr = gp102_sec2_intr, .initmsg = gp102_sec2_initmsg, @@ -268,7 +237,7 @@ MODULE_FIRMWARE("nvidia/gp107/sec2/desc.bin"); MODULE_FIRMWARE("nvidia/gp107/sec2/image.bin"); MODULE_FIRMWARE("nvidia/gp107/sec2/sig.bin"); -static void +void gp102_sec2_acr_bld_patch_1(struct nvkm_acr *acr, u32 bld, s64 adjust) { struct flcn_bl_dmem_desc_v2 hdr; @@ -279,7 +248,7 @@ gp102_sec2_acr_bld_patch_1(struct nvkm_acr *acr, u32 bld, s64 adjust) flcn_bl_dmem_desc_v2_dump(&acr->subdev, &hdr); } -static void +void gp102_sec2_acr_bld_write_1(struct nvkm_acr *acr, u32 bld, struct nvkm_acr_lsfw *lsfw) { @@ -304,7 +273,6 @@ gp102_sec2_acr_1 = { .bld_size = sizeof(struct flcn_bl_dmem_desc_v2), .bld_write = gp102_sec2_acr_bld_write_1, .bld_patch = gp102_sec2_acr_bld_patch_1, - .boot = gp102_sec2_acr_boot, .bootstrap_falcons = BIT_ULL(NVKM_ACR_LSF_FECS) | BIT_ULL(NVKM_ACR_LSF_GPCCS) | BIT_ULL(NVKM_ACR_LSF_SEC2), |