diff options
Diffstat (limited to 'arch/sh/lib')
| -rw-r--r-- | arch/sh/lib/Makefile | 13 | ||||
| -rw-r--r-- | arch/sh/lib/__clear_user.S (renamed from arch/sh/lib/clear_page.S) | 48 | ||||
| -rw-r--r-- | arch/sh/lib/copy_page.S | 11 | ||||
| -rw-r--r-- | arch/sh/lib/delay.c | 5 | ||||
| -rw-r--r-- | arch/sh/lib/libgcc.h | 3 | ||||
| -rw-r--r-- | arch/sh/lib/mcount.S | 228 | ||||
| -rw-r--r-- | arch/sh/lib/memset-sh4.S | 107 | ||||
| -rw-r--r-- | arch/sh/lib/strlen.S | 2 |
8 files changed, 348 insertions, 69 deletions
diff --git a/arch/sh/lib/Makefile b/arch/sh/lib/Makefile index aaea580b65bb..7b95f29e3174 100644 --- a/arch/sh/lib/Makefile +++ b/arch/sh/lib/Makefile @@ -2,7 +2,7 @@ # Makefile for SuperH-specific library files.. # -lib-y = delay.o memset.o memmove.o memchr.o \ +lib-y = delay.o memmove.o memchr.o \ checksum.o strlen.o div64.o div64-generic.o # Extracted from libgcc @@ -23,8 +23,11 @@ obj-y += io.o memcpy-y := memcpy.o memcpy-$(CONFIG_CPU_SH4) := memcpy-sh4.o -lib-$(CONFIG_MMU) += copy_page.o clear_page.o -lib-$(CONFIG_FUNCTION_TRACER) += mcount.o -lib-y += $(memcpy-y) $(udivsi3-y) +memset-y := memset.o +memset-$(CONFIG_CPU_SH4) := memset-sh4.o -EXTRA_CFLAGS += -Werror +lib-$(CONFIG_MMU) += copy_page.o __clear_user.o +lib-$(CONFIG_MCOUNT) += mcount.o +lib-y += $(memcpy-y) $(memset-y) $(udivsi3-y) + +ccflags-y := -Werror diff --git a/arch/sh/lib/clear_page.S b/arch/sh/lib/__clear_user.S index 8342bfbde64c..db1dca7aad14 100644 --- a/arch/sh/lib/clear_page.S +++ b/arch/sh/lib/__clear_user.S @@ -8,56 +8,10 @@ #include <linux/linkage.h> #include <asm/page.h> -/* - * clear_page - * @to: P1 address - * - * void clear_page(void *to) - */ - -/* - * r0 --- scratch - * r4 --- to - * r5 --- to + PAGE_SIZE - */ -ENTRY(clear_page) - mov r4,r5 - mov.l .Llimit,r0 - add r0,r5 - mov #0,r0 - ! -1: -#if defined(CONFIG_CPU_SH4) - movca.l r0,@r4 - mov r4,r1 -#else - mov.l r0,@r4 -#endif - add #32,r4 - mov.l r0,@-r4 - mov.l r0,@-r4 - mov.l r0,@-r4 - mov.l r0,@-r4 - mov.l r0,@-r4 - mov.l r0,@-r4 - mov.l r0,@-r4 -#if defined(CONFIG_CPU_SH4) - ocbwb @r1 -#endif - cmp/eq r5,r4 - bf/s 1b - add #28,r4 - ! - rts - nop - - .balign 4 -.Llimit: .long (PAGE_SIZE-28) - ENTRY(__clear_user) ! mov #0, r0 - mov #0xe0, r1 ! 0xffffffe0 + mov #0xffffffe0, r1 ! ! r4..(r4+31)&~32 -------- not aligned [ Area 0 ] ! (r4+31)&~32..(r4+r5)&~32 -------- aligned [ Area 1 ] diff --git a/arch/sh/lib/copy_page.S b/arch/sh/lib/copy_page.S index 43de7e8e4e17..9d7b8bc51866 100644 --- a/arch/sh/lib/copy_page.S +++ b/arch/sh/lib/copy_page.S @@ -30,7 +30,9 @@ ENTRY(copy_page) mov r4,r10 mov r5,r11 mov r5,r8 - mov.l .Lpsz,r0 + mov #(PAGE_SIZE >> 10), r0 + shll8 r0 + shll2 r0 add r0,r8 ! 1: mov.l @r11+,r0 @@ -43,7 +45,6 @@ ENTRY(copy_page) mov.l @r11+,r7 #if defined(CONFIG_CPU_SH4) movca.l r0,@r10 - mov r10,r0 #else mov.l r0,@r10 #endif @@ -55,9 +56,6 @@ ENTRY(copy_page) mov.l r3,@-r10 mov.l r2,@-r10 mov.l r1,@-r10 -#if defined(CONFIG_CPU_SH4) - ocbwb @r0 -#endif cmp/eq r11,r8 bf/s 1b add #28,r10 @@ -68,9 +66,6 @@ ENTRY(copy_page) rts nop - .balign 4 -.Lpsz: .long PAGE_SIZE - /* * __kernel_size_t __copy_user(void *to, const void *from, __kernel_size_t n); * Return the number of bytes NOT copied diff --git a/arch/sh/lib/delay.c b/arch/sh/lib/delay.c index f3ddd2133e6f..faa8f86c0db4 100644 --- a/arch/sh/lib/delay.c +++ b/arch/sh/lib/delay.c @@ -21,13 +21,14 @@ void __delay(unsigned long loops) inline void __const_udelay(unsigned long xloops) { + xloops *= 4; __asm__("dmulu.l %0, %2\n\t" "sts mach, %0" : "=r" (xloops) : "0" (xloops), - "r" (HZ * cpu_data[raw_smp_processor_id()].loops_per_jiffy) + "r" (cpu_data[raw_smp_processor_id()].loops_per_jiffy * (HZ/4)) : "macl", "mach"); - __delay(xloops); + __delay(++xloops); } void __udelay(unsigned long usecs) diff --git a/arch/sh/lib/libgcc.h b/arch/sh/lib/libgcc.h index 3f19d1c5d942..05909d58e2fe 100644 --- a/arch/sh/lib/libgcc.h +++ b/arch/sh/lib/libgcc.h @@ -17,8 +17,7 @@ struct DWstruct { #error I feel sick. #endif -typedef union -{ +typedef union { struct DWstruct s; long long ll; } DWunion; diff --git a/arch/sh/lib/mcount.S b/arch/sh/lib/mcount.S index 110fbfe1831f..84a57761f17e 100644 --- a/arch/sh/lib/mcount.S +++ b/arch/sh/lib/mcount.S @@ -1,14 +1,16 @@ /* * arch/sh/lib/mcount.S * - * Copyright (C) 2008 Paul Mundt - * Copyright (C) 2008 Matt Fleming + * Copyright (C) 2008, 2009 Paul Mundt + * Copyright (C) 2008, 2009 Matt Fleming * * This file is subject to the terms and conditions of the GNU General Public * License. See the file "COPYING" in the main directory of this archive * for more details. */ #include <asm/ftrace.h> +#include <asm/thread_info.h> +#include <asm/asm-offsets.h> #define MCOUNT_ENTER() \ mov.l r4, @-r15; \ @@ -28,6 +30,55 @@ rts; \ mov.l @r15+, r4 +#ifdef CONFIG_STACK_DEBUG +/* + * Perform diagnostic checks on the state of the kernel stack. + * + * Check for stack overflow. If there is less than 1KB free + * then it has overflowed. + * + * Make sure the stack pointer contains a valid address. Valid + * addresses for kernel stacks are anywhere after the bss + * (after _ebss) and anywhere in init_thread_union (init_stack). + */ +#define STACK_CHECK() \ + mov #(THREAD_SIZE >> 10), r0; \ + shll8 r0; \ + shll2 r0; \ + \ + /* r1 = sp & (THREAD_SIZE - 1) */ \ + mov #-1, r1; \ + add r0, r1; \ + and r15, r1; \ + \ + mov #TI_SIZE, r3; \ + mov #(STACK_WARN >> 8), r2; \ + shll8 r2; \ + add r3, r2; \ + \ + /* Is the stack overflowing? */ \ + cmp/hi r2, r1; \ + bf stack_panic; \ + \ + /* If sp > _ebss then we're OK. */ \ + mov.l .L_ebss, r1; \ + cmp/hi r1, r15; \ + bt 1f; \ + \ + /* If sp < init_stack, we're not OK. */ \ + mov.l .L_init_thread_union, r1; \ + cmp/hs r1, r15; \ + bf stack_panic; \ + \ + /* If sp > init_stack && sp < _ebss, not OK. */ \ + add r0, r1; \ + cmp/hs r1, r15; \ + bt stack_panic; \ +1: +#else +#define STACK_CHECK() +#endif /* CONFIG_STACK_DEBUG */ + .align 2 .globl _mcount .type _mcount,@function @@ -35,6 +86,19 @@ .type mcount,@function _mcount: mcount: + STACK_CHECK() + +#ifndef CONFIG_FUNCTION_TRACER + rts + nop +#else +#ifndef CONFIG_DYNAMIC_FTRACE + mov.l .Lfunction_trace_stop, r0 + mov.l @r0, r0 + tst r0, r0 + bf ftrace_stub +#endif + MCOUNT_ENTER() #ifdef CONFIG_DYNAMIC_FTRACE @@ -52,16 +116,69 @@ mcount_call: jsr @r6 nop +#ifdef CONFIG_FUNCTION_GRAPH_TRACER + mov.l .Lftrace_graph_return, r6 + mov.l .Lftrace_stub, r7 + cmp/eq r6, r7 + bt 1f + + mov.l .Lftrace_graph_caller, r0 + jmp @r0 + nop + +1: + mov.l .Lftrace_graph_entry, r6 + mov.l .Lftrace_graph_entry_stub, r7 + cmp/eq r6, r7 + bt skip_trace + + mov.l .Lftrace_graph_caller, r0 + jmp @r0 + nop + + .align 2 +.Lftrace_graph_return: + .long ftrace_graph_return +.Lftrace_graph_entry: + .long ftrace_graph_entry +.Lftrace_graph_entry_stub: + .long ftrace_graph_entry_stub +.Lftrace_graph_caller: + .long ftrace_graph_caller +#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ + + .globl skip_trace skip_trace: MCOUNT_LEAVE() .align 2 .Lftrace_trace_function: - .long ftrace_trace_function + .long ftrace_trace_function #ifdef CONFIG_DYNAMIC_FTRACE +#ifdef CONFIG_FUNCTION_GRAPH_TRACER +/* + * NOTE: Do not move either ftrace_graph_call or ftrace_caller + * as this will affect the calculation of GRAPH_INSN_OFFSET. + */ + .globl ftrace_graph_call +ftrace_graph_call: + mov.l .Lskip_trace, r0 + jmp @r0 + nop + + .align 2 +.Lskip_trace: + .long skip_trace +#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ + .globl ftrace_caller ftrace_caller: + mov.l .Lfunction_trace_stop, r0 + mov.l @r0, r0 + tst r0, r0 + bf ftrace_stub + MCOUNT_ENTER() .globl ftrace_call @@ -70,9 +187,18 @@ ftrace_call: jsr @r6 nop +#ifdef CONFIG_FUNCTION_GRAPH_TRACER + bra ftrace_graph_call + nop +#else MCOUNT_LEAVE() +#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ #endif /* CONFIG_DYNAMIC_FTRACE */ + .align 2 +.Lfunction_trace_stop: + .long function_trace_stop + /* * NOTE: From here on the locations of the .Lftrace_stub label and * ftrace_stub itself are fixed. Adding additional data here will skew @@ -80,7 +206,6 @@ ftrace_call: * Place new labels either after the ftrace_stub body, or before * ftrace_caller. You have been warned. */ - .align 2 .Lftrace_stub: .long ftrace_stub @@ -88,3 +213,98 @@ ftrace_call: ftrace_stub: rts nop + +#ifdef CONFIG_FUNCTION_GRAPH_TRACER + .globl ftrace_graph_caller +ftrace_graph_caller: + mov.l 2f, r0 + mov.l @r0, r0 + tst r0, r0 + bt 1f + + mov.l 3f, r1 + jmp @r1 + nop +1: + /* + * MCOUNT_ENTER() pushed 5 registers onto the stack, so + * the stack address containing our return address is + * r15 + 20. + */ + mov #20, r0 + add r15, r0 + mov r0, r4 + + mov.l .Lprepare_ftrace_return, r0 + jsr @r0 + nop + + MCOUNT_LEAVE() + + .align 2 +2: .long function_trace_stop +3: .long skip_trace +.Lprepare_ftrace_return: + .long prepare_ftrace_return + + .globl return_to_handler +return_to_handler: + /* + * Save the return values. + */ + mov.l r0, @-r15 + mov.l r1, @-r15 + + mov #0, r4 + + mov.l .Lftrace_return_to_handler, r0 + jsr @r0 + nop + + /* + * The return value from ftrace_return_handler has the real + * address that we should return to. + */ + lds r0, pr + mov.l @r15+, r1 + rts + mov.l @r15+, r0 + + + .align 2 +.Lftrace_return_to_handler: + .long ftrace_return_to_handler +#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ +#endif /* CONFIG_FUNCTION_TRACER */ + +#ifdef CONFIG_STACK_DEBUG + .globl stack_panic +stack_panic: + mov.l .Ldump_stack, r0 + jsr @r0 + nop + + mov.l .Lpanic, r0 + jsr @r0 + mov.l .Lpanic_s, r4 + + rts + nop + + .align 2 +.L_ebss: + .long _ebss +.L_init_thread_union: + .long init_thread_union +.Lpanic: + .long panic +.Lpanic_s: + .long .Lpanic_str +.Ldump_stack: + .long dump_stack + + .section .rodata + .align 2 +.Lpanic_str: + .string "Stack error" +#endif /* CONFIG_STACK_DEBUG */ diff --git a/arch/sh/lib/memset-sh4.S b/arch/sh/lib/memset-sh4.S new file mode 100644 index 000000000000..1a6e32cc4e4d --- /dev/null +++ b/arch/sh/lib/memset-sh4.S @@ -0,0 +1,107 @@ +/* + * "memset" implementation for SH4 + * + * Copyright (C) 1999 Niibe Yutaka + * Copyright (c) 2009 STMicroelectronics Limited + * Author: Stuart Menefy <stuart.menefy:st.com> + */ + +/* + * void *memset(void *s, int c, size_t n); + */ + +#include <linux/linkage.h> + +ENTRY(memset) + mov #12,r0 + add r6,r4 + cmp/gt r6,r0 + bt/s 40f ! if it's too small, set a byte at once + mov r4,r0 + and #3,r0 + cmp/eq #0,r0 + bt/s 2f ! It's aligned + sub r0,r6 +1: + dt r0 + bf/s 1b + mov.b r5,@-r4 +2: ! make VVVV + extu.b r5,r5 + swap.b r5,r0 ! V0 + or r0,r5 ! VV + swap.w r5,r0 ! VV00 + or r0,r5 ! VVVV + + ! Check if enough bytes need to be copied to be worth the big loop + mov #0x40, r0 ! (MT) + cmp/gt r6,r0 ! (MT) 64 > len => slow loop + + bt/s 22f + mov r6,r0 + + ! align the dst to the cache block size if necessary + mov r4, r3 + mov #~(0x1f), r1 + + and r3, r1 + cmp/eq r3, r1 + + bt/s 11f ! dst is already aligned + sub r1, r3 ! r3-r1 -> r3 + shlr2 r3 ! number of loops + +10: mov.l r5,@-r4 + dt r3 + bf/s 10b + add #-4, r6 + +11: ! dst is 32byte aligned + mov r6,r2 + mov #-5,r0 + shld r0,r2 ! number of loops + + add #-32, r4 + mov r5, r0 +12: + movca.l r0,@r4 + mov.l r5,@(4, r4) + mov.l r5,@(8, r4) + mov.l r5,@(12,r4) + mov.l r5,@(16,r4) + mov.l r5,@(20,r4) + add #-0x20, r6 + mov.l r5,@(24,r4) + dt r2 + mov.l r5,@(28,r4) + bf/s 12b + add #-32, r4 + + add #32, r4 + mov #8, r0 + cmp/ge r0, r6 + bf 40f + + mov r6,r0 +22: + shlr2 r0 + shlr r0 ! r0 = r6 >> 3 +3: + dt r0 + mov.l r5,@-r4 ! set 8-byte at once + bf/s 3b + mov.l r5,@-r4 + ! + mov #7,r0 + and r0,r6 + + ! fill bytes (length may be zero) +40: tst r6,r6 + bt 5f +4: + dt r6 + bf/s 4b + mov.b r5,@-r4 +5: + rts + mov r4,r0 diff --git a/arch/sh/lib/strlen.S b/arch/sh/lib/strlen.S index f8ab296047b3..1bcc13f05962 100644 --- a/arch/sh/lib/strlen.S +++ b/arch/sh/lib/strlen.S @@ -35,7 +35,7 @@ ENTRY(strlen) mov.b @r4+,r1 tst r1,r1 bt 8f - add #1,r2 + add #1,r2 1: mov #0,r3 |