diff options
Diffstat (limited to 'arch/powerpc/oprofile')
| -rw-r--r-- | arch/powerpc/oprofile/Makefile | 2 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/backtrace.c | 126 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/common.c | 9 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/op_model_7450.c | 4 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/op_model_fsl_booke.c | 4 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/op_model_power4.c | 46 | ||||
| -rw-r--r-- | arch/powerpc/oprofile/op_model_rs64.c | 5 |
7 files changed, 147 insertions, 49 deletions
diff --git a/arch/powerpc/oprofile/Makefile b/arch/powerpc/oprofile/Makefile index 554cd7c7532..f5f9859a833 100644 --- a/arch/powerpc/oprofile/Makefile +++ b/arch/powerpc/oprofile/Makefile @@ -6,7 +6,7 @@ DRIVER_OBJS := $(addprefix ../../../drivers/oprofile/, \ oprofilefs.o oprofile_stats.o \ timer_int.o ) -oprofile-y := $(DRIVER_OBJS) common.o +oprofile-y := $(DRIVER_OBJS) common.o backtrace.o oprofile-$(CONFIG_PPC64) += op_model_rs64.o op_model_power4.o oprofile-$(CONFIG_FSL_BOOKE) += op_model_fsl_booke.o oprofile-$(CONFIG_PPC32) += op_model_7450.o diff --git a/arch/powerpc/oprofile/backtrace.c b/arch/powerpc/oprofile/backtrace.c new file mode 100644 index 00000000000..75f57bc96b4 --- /dev/null +++ b/arch/powerpc/oprofile/backtrace.c @@ -0,0 +1,126 @@ +/** + * Copyright (C) 2005 Brian Rogan <bcr6@cornell.edu>, IBM + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. +**/ + +#include <linux/oprofile.h> +#include <linux/sched.h> +#include <asm/processor.h> +#include <asm/uaccess.h> + +#define STACK_SP(STACK) *(STACK) + +#define STACK_LR64(STACK) *((unsigned long *)(STACK) + 2) +#define STACK_LR32(STACK) *((unsigned int *)(STACK) + 1) + +#ifdef CONFIG_PPC64 +#define STACK_LR(STACK) STACK_LR64(STACK) +#else +#define STACK_LR(STACK) STACK_LR32(STACK) +#endif + +static unsigned int user_getsp32(unsigned int sp, int is_first) +{ + unsigned int stack_frame[2]; + + if (!access_ok(VERIFY_READ, sp, sizeof(stack_frame))) + return 0; + + /* + * The most likely reason for this is that we returned -EFAULT, + * which means that we've done all that we can do from + * interrupt context. + */ + if (__copy_from_user_inatomic(stack_frame, (void *)(long)sp, + sizeof(stack_frame))) + return 0; + + if (!is_first) + oprofile_add_trace(STACK_LR32(stack_frame)); + + /* + * We do not enforce increasing stack addresses here because + * we may transition to a different stack, eg a signal handler. + */ + return STACK_SP(stack_frame); +} + +#ifdef CONFIG_PPC64 +static unsigned long user_getsp64(unsigned long sp, int is_first) +{ + unsigned long stack_frame[3]; + + if (!access_ok(VERIFY_READ, sp, sizeof(stack_frame))) + return 0; + + if (__copy_from_user_inatomic(stack_frame, (void *)sp, + sizeof(stack_frame))) + return 0; + + if (!is_first) + oprofile_add_trace(STACK_LR64(stack_frame)); + + return STACK_SP(stack_frame); +} +#endif + +static unsigned long kernel_getsp(unsigned long sp, int is_first) +{ + unsigned long *stack_frame = (unsigned long *)sp; + + if (!validate_sp(sp, current, STACK_FRAME_OVERHEAD)) + return 0; + + if (!is_first) + oprofile_add_trace(STACK_LR(stack_frame)); + + /* + * We do not enforce increasing stack addresses here because + * we might be transitioning from an interrupt stack to a kernel + * stack. validate_sp() is designed to understand this, so just + * use it. + */ + return STACK_SP(stack_frame); +} + +void op_powerpc_backtrace(struct pt_regs * const regs, unsigned int depth) +{ + unsigned long sp = regs->gpr[1]; + int first_frame = 1; + + /* We ditch the top stackframe so need to loop through an extra time */ + depth += 1; + + if (!user_mode(regs)) { + while (depth--) { + sp = kernel_getsp(sp, first_frame); + if (!sp) + break; + first_frame = 0; + } + } else { +#ifdef CONFIG_PPC64 + if (!test_thread_flag(TIF_32BIT)) { + while (depth--) { + sp = user_getsp64(sp, first_frame); + if (!sp) + break; + first_frame = 0; + } + + return; + } +#endif + + while (depth--) { + sp = user_getsp32(sp, first_frame); + if (!sp) + break; + first_frame = 0; + } + } +} diff --git a/arch/powerpc/oprofile/common.c b/arch/powerpc/oprofile/common.c index cc2535be3a7..5b1de7e8041 100644 --- a/arch/powerpc/oprofile/common.c +++ b/arch/powerpc/oprofile/common.c @@ -117,18 +117,10 @@ static int op_powerpc_create_files(struct super_block *sb, struct dentry *root) oprofilefs_create_ulong(sb, root, "enable_kernel", &sys.enable_kernel); oprofilefs_create_ulong(sb, root, "enable_user", &sys.enable_user); -#ifdef CONFIG_PPC64 - oprofilefs_create_ulong(sb, root, "backtrace_spinlocks", - &sys.backtrace_spinlocks); -#endif /* Default to tracing both kernel and user */ sys.enable_kernel = 1; sys.enable_user = 1; -#ifdef CONFIG_PPC64 - /* Turn on backtracing through spinlocks by default */ - sys.backtrace_spinlocks = 1; -#endif return 0; } @@ -168,6 +160,7 @@ int __init oprofile_arch_init(struct oprofile_operations *ops) ops->shutdown = op_powerpc_shutdown; ops->start = op_powerpc_start; ops->stop = op_powerpc_stop; + ops->backtrace = op_powerpc_backtrace; printk(KERN_INFO "oprofile: using %s performance monitoring.\n", ops->cpu_type); diff --git a/arch/powerpc/oprofile/op_model_7450.c b/arch/powerpc/oprofile/op_model_7450.c index 32abfdbb0eb..e0491c3c71f 100644 --- a/arch/powerpc/oprofile/op_model_7450.c +++ b/arch/powerpc/oprofile/op_model_7450.c @@ -176,13 +176,13 @@ static void fsl7450_handle_interrupt(struct pt_regs *regs, mtmsr(mfmsr() | MSR_PMM); pc = mfspr(SPRN_SIAR); - is_kernel = (pc >= KERNELBASE); + is_kernel = is_kernel_addr(pc); for (i = 0; i < NUM_CTRS; ++i) { val = ctr_read(i); if (val < 0) { if (oprofile_running && ctr[i].enabled) { - oprofile_add_pc(pc, is_kernel, i); + oprofile_add_ext_sample(pc, regs, i, is_kernel); ctr_write(i, reset_value[i]); } else { ctr_write(i, 0); diff --git a/arch/powerpc/oprofile/op_model_fsl_booke.c b/arch/powerpc/oprofile/op_model_fsl_booke.c index 26539cda602..93d63e62662 100644 --- a/arch/powerpc/oprofile/op_model_fsl_booke.c +++ b/arch/powerpc/oprofile/op_model_fsl_booke.c @@ -154,13 +154,13 @@ static void fsl_booke_handle_interrupt(struct pt_regs *regs, mtmsr(mfmsr() | MSR_PMM); pc = regs->nip; - is_kernel = (pc >= KERNELBASE); + is_kernel = is_kernel_addr(pc); for (i = 0; i < num_counters; ++i) { val = ctr_read(i); if (val < 0) { if (oprofile_running && ctr[i].enabled) { - oprofile_add_pc(pc, is_kernel, i); + oprofile_add_ext_sample(pc, regs, i, is_kernel); ctr_write(i, reset_value[i]); } else { ctr_write(i, 0); diff --git a/arch/powerpc/oprofile/op_model_power4.c b/arch/powerpc/oprofile/op_model_power4.c index 659a021da0c..4c2beab1fdc 100644 --- a/arch/powerpc/oprofile/op_model_power4.c +++ b/arch/powerpc/oprofile/op_model_power4.c @@ -10,6 +10,7 @@ #include <linux/oprofile.h> #include <linux/init.h> #include <linux/smp.h> +#include <asm/firmware.h> #include <asm/ptrace.h> #include <asm/system.h> #include <asm/processor.h> @@ -24,18 +25,14 @@ static unsigned long reset_value[OP_MAX_COUNTER]; static int oprofile_running; static int mmcra_has_sihv; +/* Unfortunately these bits vary between CPUs */ +static unsigned long mmcra_sihv = MMCRA_SIHV; +static unsigned long mmcra_sipr = MMCRA_SIPR; /* mmcr values are set in power4_reg_setup, used in power4_cpu_setup */ static u32 mmcr0_val; static u64 mmcr1_val; -static u32 mmcra_val; - -/* - * Since we do not have an NMI, backtracing through spinlocks is - * only a best guess. In light of this, allow it to be disabled at - * runtime. - */ -static int backtrace_spinlocks; +static u64 mmcra_val; static void power4_reg_setup(struct op_counter_config *ctr, struct op_system_config *sys, @@ -62,8 +59,6 @@ static void power4_reg_setup(struct op_counter_config *ctr, mmcr1_val = sys->mmcr1; mmcra_val = sys->mmcra; - backtrace_spinlocks = sys->backtrace_spinlocks; - for (i = 0; i < cur_cpu_spec->num_pmcs; ++i) reset_value[i] = 0x80000000UL - ctr[i].count; @@ -196,25 +191,6 @@ static void __attribute_used__ kernel_unknown_bucket(void) { } -static unsigned long check_spinlock_pc(struct pt_regs *regs, - unsigned long profile_pc) -{ - unsigned long pc = instruction_pointer(regs); - - /* - * If both the SIAR (sampled instruction) and the perfmon exception - * occurred in a spinlock region then we account the sample to the - * calling function. This isnt 100% correct, we really need soft - * IRQ disable so we always get the perfmon exception at the - * point at which the SIAR is set. - */ - if (backtrace_spinlocks && in_lock_functions(pc) && - in_lock_functions(profile_pc)) - return regs->link; - else - return profile_pc; -} - /* * On GQ and newer the MMCRA stores the HV and PR bits at the time * the SIAR was sampled. We use that to work out if the SIAR was sampled in @@ -227,17 +203,17 @@ static unsigned long get_pc(struct pt_regs *regs) /* Cant do much about it */ if (!mmcra_has_sihv) - return check_spinlock_pc(regs, pc); + return pc; mmcra = mfspr(SPRN_MMCRA); /* Were we in the hypervisor? */ - if (platform_is_lpar() && (mmcra & MMCRA_SIHV)) + if (firmware_has_feature(FW_FEATURE_LPAR) && (mmcra & mmcra_sihv)) /* function descriptor madness */ return *((unsigned long *)hypervisor_bucket); /* We were in userspace, nothing to do */ - if (mmcra & MMCRA_SIPR) + if (mmcra & mmcra_sipr) return pc; #ifdef CONFIG_PPC_RTAS @@ -256,7 +232,7 @@ static unsigned long get_pc(struct pt_regs *regs) /* function descriptor madness */ return *((unsigned long *)kernel_unknown_bucket); - return check_spinlock_pc(regs, pc); + return pc; } static int get_kernel(unsigned long pc) @@ -267,7 +243,7 @@ static int get_kernel(unsigned long pc) is_kernel = is_kernel_addr(pc); } else { unsigned long mmcra = mfspr(SPRN_MMCRA); - is_kernel = ((mmcra & MMCRA_SIPR) == 0); + is_kernel = ((mmcra & mmcra_sipr) == 0); } return is_kernel; @@ -292,7 +268,7 @@ static void power4_handle_interrupt(struct pt_regs *regs, val = ctr_read(i); if (val < 0) { if (oprofile_running && ctr[i].enabled) { - oprofile_add_pc(pc, is_kernel, i); + oprofile_add_ext_sample(pc, regs, i, is_kernel); ctr_write(i, reset_value[i]); } else { ctr_write(i, 0); diff --git a/arch/powerpc/oprofile/op_model_rs64.c b/arch/powerpc/oprofile/op_model_rs64.c index 5c909ee609f..042f8f4867a 100644 --- a/arch/powerpc/oprofile/op_model_rs64.c +++ b/arch/powerpc/oprofile/op_model_rs64.c @@ -175,10 +175,13 @@ static void rs64_handle_interrupt(struct pt_regs *regs, struct op_counter_config *ctr) { unsigned int mmcr0; + int is_kernel; int val; int i; unsigned long pc = mfspr(SPRN_SIAR); + is_kernel = is_kernel_addr(pc); + /* set the PMM bit (see comment below) */ mtmsrd(mfmsr() | MSR_PMM); @@ -186,7 +189,7 @@ static void rs64_handle_interrupt(struct pt_regs *regs, val = ctr_read(i); if (val < 0) { if (ctr[i].enabled) { - oprofile_add_pc(pc, is_kernel_addr(pc), i); + oprofile_add_ext_sample(pc, regs, i, is_kernel); ctr_write(i, reset_value[i]); } else { ctr_write(i, 0); |