From b0884e25fe361f2ca228808fb5fd1b74cb04e711 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Thu, 11 Dec 2008 13:45:23 +0100 Subject: [PATCH 1/9] x86, bts: turn BUG_ON into WARN_ON_ONCE Impact: make the ds code more debuggable Turn BUG_ON's into WARN_ON_ONCE. Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/kernel/ds.c | 4 ++-- arch/x86/kernel/ptrace.c | 5 +++-- 2 files changed, 5 insertions(+), 4 deletions(-) diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index 19a8c2c0389f..095306988667 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -452,7 +452,7 @@ struct pebs_tracer *ds_request_pebs(struct task_struct *task, static void ds_release(struct ds_tracer *tracer, enum ds_qualifier qual) { - BUG_ON(tracer->context->owner[qual] != tracer); + WARN_ON_ONCE(tracer->context->owner[qual] != tracer); tracer->context->owner[qual] = NULL; put_tracer(tracer->context->task); @@ -774,7 +774,7 @@ ds_configure(const struct ds_configuration *cfg) printk(KERN_INFO "DS available\n"); - BUG_ON(MAX_SIZEOF_DS < ds_cfg.sizeof_ds); + WARN_ON_ONCE(MAX_SIZEOF_DS < ds_cfg.sizeof_ds); } void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index 2c8ec1ba75e6..b2998fe1166b 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -878,7 +878,8 @@ static int ptrace_bts_write_record(struct task_struct *child, { unsigned char bts_record[BTS_MAX_RECORD_SIZE]; - BUG_ON(BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts); + if (BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts) + return -EOVERFLOW; memset(bts_record, 0, bts_cfg.sizeof_bts); switch (in->qualifier) { @@ -1133,7 +1134,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data) ret = ds_get_bts_index(child->bts, &size); if (ret == 0) { - BUG_ON(size != (int) size); + WARN_ON_ONCE(size != (int) size); ret = (int) size; } break; From c2724775ce57c98b8af9694857b941dc61056516 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Thu, 11 Dec 2008 13:49:59 +0100 Subject: [PATCH 2/9] x86, bts: provide in-kernel branch-trace interface Impact: cleanup Move the BTS bits from ptrace.c into ds.c. Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/include/asm/ds.h | 241 ++++--- arch/x86/include/asm/processor.h | 13 + arch/x86/include/asm/ptrace.h | 36 - arch/x86/include/asm/thread_info.h | 5 +- arch/x86/kernel/cpu/intel.c | 4 - arch/x86/kernel/ds.c | 1035 +++++++++++++++++----------- arch/x86/kernel/process_32.c | 59 +- arch/x86/kernel/process_64.c | 50 +- arch/x86/kernel/ptrace.c | 434 +++--------- include/linux/sched.h | 1 + 10 files changed, 909 insertions(+), 969 deletions(-) diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h index 99b6c39774a4..ee0ea3a96c11 100644 --- a/arch/x86/include/asm/ds.h +++ b/arch/x86/include/asm/ds.h @@ -6,13 +6,13 @@ * precise-event based sampling (PEBS). * * It manages: - * - per-thread and per-cpu allocation of BTS and PEBS + * - DS and BTS hardware configuration * - buffer overflow handling (to be done) * - buffer access * - * It assumes: - * - get_task_struct on all traced tasks - * - current is allowed to trace tasks + * It does not do: + * - security checking (is the caller allowed to trace the task) + * - buffer allocation (memory accounting) * * * Copyright (C) 2007-2008 Intel Corporation. @@ -31,6 +31,7 @@ #ifdef CONFIG_X86_DS struct task_struct; +struct ds_context; struct ds_tracer; struct bts_tracer; struct pebs_tracer; @@ -38,6 +39,38 @@ struct pebs_tracer; typedef void (*bts_ovfl_callback_t)(struct bts_tracer *); typedef void (*pebs_ovfl_callback_t)(struct pebs_tracer *); + +/* + * A list of features plus corresponding macros to talk about them in + * the ds_request function's flags parameter. + * + * We use the enum to index an array of corresponding control bits; + * we use the macro to index a flags bit-vector. + */ +enum ds_feature { + dsf_bts = 0, + dsf_bts_kernel, +#define BTS_KERNEL (1 << dsf_bts_kernel) + /* trace kernel-mode branches */ + + dsf_bts_user, +#define BTS_USER (1 << dsf_bts_user) + /* trace user-mode branches */ + + dsf_bts_overflow, + dsf_bts_max, + dsf_pebs = dsf_bts_max, + + dsf_pebs_max, + dsf_ctl_max = dsf_pebs_max, + dsf_bts_timestamps = dsf_ctl_max, +#define BTS_TIMESTAMPS (1 << dsf_bts_timestamps) + /* add timestamps into BTS trace */ + +#define BTS_USER_FLAGS (BTS_KERNEL | BTS_USER | BTS_TIMESTAMPS) +}; + + /* * Request BTS or PEBS * @@ -58,92 +91,135 @@ typedef void (*pebs_ovfl_callback_t)(struct pebs_tracer *); * NULL if cyclic buffer requested * th: the interrupt threshold in records from the end of the buffer; * -1 if no interrupt threshold is requested. + * flags: a bit-mask of the above flags */ extern struct bts_tracer *ds_request_bts(struct task_struct *task, void *base, size_t size, - bts_ovfl_callback_t ovfl, size_t th); + bts_ovfl_callback_t ovfl, + size_t th, unsigned int flags); extern struct pebs_tracer *ds_request_pebs(struct task_struct *task, void *base, size_t size, pebs_ovfl_callback_t ovfl, - size_t th); + size_t th, unsigned int flags); /* * Release BTS or PEBS resources - * - * Returns 0 on success; -Eerrno otherwise + * Suspend and resume BTS or PEBS tracing * * tracer: the tracer handle returned from ds_request_~() */ -extern int ds_release_bts(struct bts_tracer *tracer); -extern int ds_release_pebs(struct pebs_tracer *tracer); +extern void ds_release_bts(struct bts_tracer *tracer); +extern void ds_suspend_bts(struct bts_tracer *tracer); +extern void ds_resume_bts(struct bts_tracer *tracer); +extern void ds_release_pebs(struct pebs_tracer *tracer); +extern void ds_suspend_pebs(struct pebs_tracer *tracer); +extern void ds_resume_pebs(struct pebs_tracer *tracer); + /* - * Get the (array) index of the write pointer. - * (assuming an array of BTS/PEBS records) + * The raw DS buffer state as it is used for BTS and PEBS recording. * - * Returns 0 on success; -Eerrno on error - * - * tracer: the tracer handle returned from ds_request_~() - * pos (out): will hold the result + * This is the low-level, arch-dependent interface for working + * directly on the raw trace data. */ -extern int ds_get_bts_index(struct bts_tracer *tracer, size_t *pos); -extern int ds_get_pebs_index(struct pebs_tracer *tracer, size_t *pos); +struct ds_trace { + /* the number of bts/pebs records */ + size_t n; + /* the size of a bts/pebs record in bytes */ + size_t size; + /* pointers into the raw buffer: + - to the first entry */ + void *begin; + /* - one beyond the last entry */ + void *end; + /* - one beyond the newest entry */ + void *top; + /* - the interrupt threshold */ + void *ith; + /* flags given on ds_request() */ + unsigned int flags; +}; /* - * Get the (array) index one record beyond the end of the array. - * (assuming an array of BTS/PEBS records) - * - * Returns 0 on success; -Eerrno on error - * - * tracer: the tracer handle returned from ds_request_~() - * pos (out): will hold the result + * An arch-independent view on branch trace data. */ -extern int ds_get_bts_end(struct bts_tracer *tracer, size_t *pos); -extern int ds_get_pebs_end(struct pebs_tracer *tracer, size_t *pos); +enum bts_qualifier { + bts_invalid, +#define BTS_INVALID bts_invalid + + bts_branch, +#define BTS_BRANCH bts_branch + + bts_task_arrives, +#define BTS_TASK_ARRIVES bts_task_arrives + + bts_task_departs, +#define BTS_TASK_DEPARTS bts_task_departs + + bts_qual_bit_size = 4, + bts_qual_max = (1 << bts_qual_bit_size), +}; + +struct bts_struct { + __u64 qualifier; + union { + /* BTS_BRANCH */ + struct { + __u64 from; + __u64 to; + } lbr; + /* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */ + struct { + __u64 jiffies; + pid_t pid; + } timestamp; + } variant; +}; + /* - * Provide a pointer to the BTS/PEBS record at parameter index. - * (assuming an array of BTS/PEBS records) + * The BTS state. * - * The pointer points directly into the buffer. The user is - * responsible for copying the record. - * - * Returns the size of a single record on success; -Eerrno on error - * - * tracer: the tracer handle returned from ds_request_~() - * index: the index of the requested record - * record (out): pointer to the requested record + * This gives access to the raw DS state and adds functions to provide + * an arch-independent view of the BTS data. */ -extern int ds_access_bts(struct bts_tracer *tracer, - size_t index, const void **record); -extern int ds_access_pebs(struct pebs_tracer *tracer, - size_t index, const void **record); +struct bts_trace { + struct ds_trace ds; + + int (*read)(struct bts_tracer *tracer, const void *at, + struct bts_struct *out); + int (*write)(struct bts_tracer *tracer, const struct bts_struct *in); +}; + /* - * Write one or more BTS/PEBS records at the write pointer index and - * advance the write pointer. + * The PEBS state. * - * If size is not a multiple of the record size, trailing bytes are - * zeroed out. + * This gives access to the raw DS state and the PEBS-specific counter + * reset value. + */ +struct pebs_trace { + struct ds_trace ds; + + /* the PEBS reset value */ + unsigned long long reset_value; +}; + + +/* + * Read the BTS or PEBS trace. * - * May result in one or more overflow notifications. + * Returns a view on the trace collected for the parameter tracer. * - * If called during overflow handling, that is, with index >= - * interrupt threshold, the write will wrap around. - * - * An overflow notification is given if and when the interrupt - * threshold is reached during or after the write. - * - * Returns the number of bytes written or -Eerrno. + * The view remains valid as long as the traced task is not running or + * the tracer is suspended. + * Writes into the trace buffer are not reflected. * * tracer: the tracer handle returned from ds_request_~() - * buffer: the buffer to write - * size: the size of the buffer */ -extern int ds_write_bts(struct bts_tracer *tracer, - const void *buffer, size_t size); -extern int ds_write_pebs(struct pebs_tracer *tracer, - const void *buffer, size_t size); +extern const struct bts_trace *ds_read_bts(struct bts_tracer *tracer); +extern const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer); + /* * Reset the write pointer of the BTS/PEBS buffer. @@ -155,27 +231,6 @@ extern int ds_write_pebs(struct pebs_tracer *tracer, extern int ds_reset_bts(struct bts_tracer *tracer); extern int ds_reset_pebs(struct pebs_tracer *tracer); -/* - * Clear the BTS/PEBS buffer and reset the write pointer. - * The entire buffer will be zeroed out. - * - * Returns 0 on success; -Eerrno on error - * - * tracer: the tracer handle returned from ds_request_~() - */ -extern int ds_clear_bts(struct bts_tracer *tracer); -extern int ds_clear_pebs(struct pebs_tracer *tracer); - -/* - * Provide the PEBS counter reset value. - * - * Returns 0 on success; -Eerrno on error - * - * tracer: the tracer handle returned from ds_request_pebs() - * value (out): the counter reset value - */ -extern int ds_get_pebs_reset(struct pebs_tracer *tracer, u64 *value); - /* * Set the PEBS counter reset value. * @@ -192,35 +247,17 @@ extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value); struct cpuinfo_x86; extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *); - - /* - * The DS context - part of struct thread_struct. + * Context switch work */ -#define MAX_SIZEOF_DS (12 * 8) - -struct ds_context { - /* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */ - unsigned char ds[MAX_SIZEOF_DS]; - /* the owner of the BTS and PEBS configuration, respectively */ - struct ds_tracer *owner[2]; - /* use count */ - unsigned long count; - /* a pointer to the context location inside the thread_struct - * or the per_cpu context array */ - struct ds_context **this; - /* a pointer to the task owning this context, or NULL, if the - * context is owned by a cpu */ - struct task_struct *task; -}; - -/* called by exit_thread() to free leftover contexts */ -extern void ds_free(struct ds_context *context); +extern void ds_switch_to(struct task_struct *prev, struct task_struct *next); #else /* CONFIG_X86_DS */ struct cpuinfo_x86; static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {} +static inline void ds_switch_to(struct task_struct *prev, + struct task_struct *next) {} #endif /* CONFIG_X86_DS */ #endif /* _ASM_X86_DS_H */ diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h index 5ca01e383269..aa5914f8e501 100644 --- a/arch/x86/include/asm/processor.h +++ b/arch/x86/include/asm/processor.h @@ -752,6 +752,19 @@ extern void switch_to_new_gdt(void); extern void cpu_init(void); extern void init_gdt(int cpu); +static inline unsigned long get_debugctlmsr(void) +{ + unsigned long debugctlmsr = 0; + +#ifndef CONFIG_X86_DEBUGCTLMSR + if (boot_cpu_data.x86 < 6) + return 0; +#endif + rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); + + return debugctlmsr; +} + static inline void update_debugctlmsr(unsigned long debugctlmsr) { #ifndef CONFIG_X86_DEBUGCTLMSR diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h index eefb0594b058..fbf744215911 100644 --- a/arch/x86/include/asm/ptrace.h +++ b/arch/x86/include/asm/ptrace.h @@ -6,7 +6,6 @@ #include #ifdef __KERNEL__ -#include /* the DS BTS struct is used for ptrace too */ #include #endif @@ -128,34 +127,6 @@ struct pt_regs { #endif /* !__i386__ */ -#ifdef CONFIG_X86_PTRACE_BTS -/* a branch trace record entry - * - * In order to unify the interface between various processor versions, - * we use the below data structure for all processors. - */ -enum bts_qualifier { - BTS_INVALID = 0, - BTS_BRANCH, - BTS_TASK_ARRIVES, - BTS_TASK_DEPARTS -}; - -struct bts_struct { - __u64 qualifier; - union { - /* BTS_BRANCH */ - struct { - __u64 from_ip; - __u64 to_ip; - } lbr; - /* BTS_TASK_ARRIVES or - BTS_TASK_DEPARTS */ - __u64 jiffies; - } variant; -}; -#endif /* CONFIG_X86_PTRACE_BTS */ - #ifdef __KERNEL__ #include @@ -163,13 +134,6 @@ struct bts_struct { struct cpuinfo_x86; struct task_struct; -#ifdef CONFIG_X86_PTRACE_BTS -extern void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *); -extern void ptrace_bts_take_timestamp(struct task_struct *, enum bts_qualifier); -#else -#define ptrace_bts_init_intel(config) do {} while (0) -#endif /* CONFIG_X86_PTRACE_BTS */ - extern unsigned long profile_pc(struct pt_regs *regs); extern unsigned long diff --git a/arch/x86/include/asm/thread_info.h b/arch/x86/include/asm/thread_info.h index 0921b4018c11..bf8113d16a33 100644 --- a/arch/x86/include/asm/thread_info.h +++ b/arch/x86/include/asm/thread_info.h @@ -93,7 +93,6 @@ struct thread_info { #define TIF_FORCED_TF 24 /* true if TF in eflags artificially */ #define TIF_DEBUGCTLMSR 25 /* uses thread_struct.debugctlmsr */ #define TIF_DS_AREA_MSR 26 /* uses thread_struct.ds_area_msr */ -#define TIF_BTS_TRACE_TS 27 /* record scheduling event timestamps */ #define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE) #define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME) @@ -115,7 +114,6 @@ struct thread_info { #define _TIF_FORCED_TF (1 << TIF_FORCED_TF) #define _TIF_DEBUGCTLMSR (1 << TIF_DEBUGCTLMSR) #define _TIF_DS_AREA_MSR (1 << TIF_DS_AREA_MSR) -#define _TIF_BTS_TRACE_TS (1 << TIF_BTS_TRACE_TS) /* work to do in syscall_trace_enter() */ #define _TIF_WORK_SYSCALL_ENTRY \ @@ -141,8 +139,7 @@ struct thread_info { /* flags to check in __switch_to() */ #define _TIF_WORK_CTXSW \ - (_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_BTS_TRACE_TS| \ - _TIF_NOTSC) + (_TIF_IO_BITMAP|_TIF_DEBUGCTLMSR|_TIF_DS_AREA_MSR|_TIF_NOTSC) #define _TIF_WORK_CTXSW_PREV _TIF_WORK_CTXSW #define _TIF_WORK_CTXSW_NEXT (_TIF_WORK_CTXSW|_TIF_DEBUG) diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c index 816f27f289b1..cd413d9a0218 100644 --- a/arch/x86/kernel/cpu/intel.c +++ b/arch/x86/kernel/cpu/intel.c @@ -11,7 +11,6 @@ #include #include #include -#include #include #include @@ -309,9 +308,6 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c) set_cpu_cap(c, X86_FEATURE_P3); #endif - if (cpu_has_bts) - ptrace_bts_init_intel(c); - detect_extended_topology(c); if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) { /* diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index 095306988667..f0583005b75e 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -6,13 +6,13 @@ * precise-event based sampling (PEBS). * * It manages: - * - per-thread and per-cpu allocation of BTS and PEBS + * - DS and BTS hardware configuration * - buffer overflow handling (to be done) * - buffer access * - * It assumes: - * - get_task_struct on all traced tasks - * - current is allowed to trace tasks + * It does not do: + * - security checking (is the caller allowed to trace the task) + * - buffer allocation (memory accounting) * * * Copyright (C) 2007-2008 Intel Corporation. @@ -34,15 +34,30 @@ * The configuration for a particular DS hardware implementation. */ struct ds_configuration { - /* the size of the DS structure in bytes */ - unsigned char sizeof_ds; - /* the size of one pointer-typed field in the DS structure in bytes; - this covers the first 8 fields related to buffer management. */ + /* the name of the configuration */ + const char *name; + /* the size of one pointer-typed field in the DS structure and + in the BTS and PEBS buffers in bytes; + this covers the first 8 DS fields related to buffer management. */ unsigned char sizeof_field; /* the size of a BTS/PEBS record in bytes */ unsigned char sizeof_rec[2]; + /* a series of bit-masks to control various features indexed + * by enum ds_feature */ + unsigned long ctl[dsf_ctl_max]; }; -static struct ds_configuration ds_cfg; +static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array); + +#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id()) + +#define MAX_SIZEOF_DS (12 * 8) /* maximal size of a DS configuration */ +#define MAX_SIZEOF_BTS (3 * 8) /* maximal size of a BTS record */ +#define DS_ALIGNMENT (1 << 3) /* BTS and PEBS buffer alignment */ + +#define BTS_CONTROL \ + (ds_cfg.ctl[dsf_bts] | ds_cfg.ctl[dsf_bts_kernel] | ds_cfg.ctl[dsf_bts_user] |\ + ds_cfg.ctl[dsf_bts_overflow]) + /* * A BTS or PEBS tracer. @@ -61,6 +76,8 @@ struct ds_tracer { struct bts_tracer { /* the common DS part */ struct ds_tracer ds; + /* the trace including the DS configuration */ + struct bts_trace trace; /* buffer overflow notification function */ bts_ovfl_callback_t ovfl; }; @@ -68,6 +85,8 @@ struct bts_tracer { struct pebs_tracer { /* the common DS part */ struct ds_tracer ds; + /* the trace including the DS configuration */ + struct pebs_trace trace; /* buffer overflow notification function */ pebs_ovfl_callback_t ovfl; }; @@ -134,13 +153,11 @@ static inline void ds_set(unsigned char *base, enum ds_qualifier qual, (*(unsigned long *)base) = value; } -#define DS_ALIGNMENT (1 << 3) /* BTS and PEBS buffer alignment */ - /* * Locking is done only for allocating BTS or PEBS resources. */ -static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock); +static DEFINE_SPINLOCK(ds_lock); /* @@ -156,27 +173,32 @@ static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock); * >0 number of per-thread tracers * <0 number of per-cpu tracers * - * The below functions to get and put tracers and to check the - * allocation type require the ds_lock to be held by the caller. - * * Tracers essentially gives the number of ds contexts for a certain * type of allocation. */ -static long tracers; +static atomic_t tracers = ATOMIC_INIT(0); static inline void get_tracer(struct task_struct *task) { - tracers += (task ? 1 : -1); + if (task) + atomic_inc(&tracers); + else + atomic_dec(&tracers); } static inline void put_tracer(struct task_struct *task) { - tracers -= (task ? 1 : -1); + if (task) + atomic_dec(&tracers); + else + atomic_inc(&tracers); } static inline int check_tracer(struct task_struct *task) { - return (task ? (tracers >= 0) : (tracers <= 0)); + return task ? + (atomic_read(&tracers) >= 0) : + (atomic_read(&tracers) <= 0); } @@ -190,14 +212,30 @@ static inline int check_tracer(struct task_struct *task) * Contexts are use-counted. They are allocated on first access and * deallocated when the last user puts the context. */ -static DEFINE_PER_CPU(struct ds_context *, system_context); +struct ds_context { + /* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */ + unsigned char ds[MAX_SIZEOF_DS]; + /* the owner of the BTS and PEBS configuration, respectively */ + struct bts_tracer *bts_master; + struct pebs_tracer *pebs_master; + /* use count */ + unsigned long count; + /* a pointer to the context location inside the thread_struct + * or the per_cpu context array */ + struct ds_context **this; + /* a pointer to the task owning this context, or NULL, if the + * context is owned by a cpu */ + struct task_struct *task; +}; -#define this_system_context per_cpu(system_context, smp_processor_id()) +static DEFINE_PER_CPU(struct ds_context *, system_context_array); + +#define system_context per_cpu(system_context_array, smp_processor_id()) static inline struct ds_context *ds_get_context(struct task_struct *task) { struct ds_context **p_context = - (task ? &task->thread.ds_ctx : &this_system_context); + (task ? &task->thread.ds_ctx : &system_context); struct ds_context *context = *p_context; unsigned long irq; @@ -225,10 +263,22 @@ static inline struct ds_context *ds_get_context(struct task_struct *task) wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds); } - spin_unlock_irqrestore(&ds_lock, irq); - } - context->count++; + context->count++; + + spin_unlock_irqrestore(&ds_lock, irq); + } else { + spin_lock_irqsave(&ds_lock, irq); + + context = *p_context; + if (context) + context->count++; + + spin_unlock_irqrestore(&ds_lock, irq); + + if (!context) + context = ds_get_context(task); + } return context; } @@ -242,8 +292,10 @@ static inline void ds_put_context(struct ds_context *context) spin_lock_irqsave(&ds_lock, irq); - if (--context->count) - goto out; + if (--context->count) { + spin_unlock_irqrestore(&ds_lock, irq); + return; + } *(context->this) = NULL; @@ -253,14 +305,14 @@ static inline void ds_put_context(struct ds_context *context) if (!context->task || (context->task == current)) wrmsrl(MSR_IA32_DS_AREA, 0); - kfree(context); - out: spin_unlock_irqrestore(&ds_lock, irq); + + kfree(context); } /* - * Handle a buffer overflow + * Call the tracer's callback on a buffer overflow. * * context: the ds context * qual: the buffer type @@ -268,328 +320,30 @@ static inline void ds_put_context(struct ds_context *context) static void ds_overflow(struct ds_context *context, enum ds_qualifier qual) { switch (qual) { - case ds_bts: { - struct bts_tracer *tracer = - container_of(context->owner[qual], - struct bts_tracer, ds); - if (tracer->ovfl) - tracer->ovfl(tracer); - } + case ds_bts: + if (context->bts_master && + context->bts_master->ovfl) + context->bts_master->ovfl(context->bts_master); break; - case ds_pebs: { - struct pebs_tracer *tracer = - container_of(context->owner[qual], - struct pebs_tracer, ds); - if (tracer->ovfl) - tracer->ovfl(tracer); - } + case ds_pebs: + if (context->pebs_master && + context->pebs_master->ovfl) + context->pebs_master->ovfl(context->pebs_master); break; } } -static void ds_install_ds_config(struct ds_context *context, - enum ds_qualifier qual, - void *base, size_t size, size_t ith) -{ - unsigned long buffer, adj; - - /* adjust the buffer address and size to meet alignment - * constraints: - * - buffer is double-word aligned - * - size is multiple of record size - * - * We checked the size at the very beginning; we have enough - * space to do the adjustment. - */ - buffer = (unsigned long)base; - - adj = ALIGN(buffer, DS_ALIGNMENT) - buffer; - buffer += adj; - size -= adj; - - size /= ds_cfg.sizeof_rec[qual]; - size *= ds_cfg.sizeof_rec[qual]; - - ds_set(context->ds, qual, ds_buffer_base, buffer); - ds_set(context->ds, qual, ds_index, buffer); - ds_set(context->ds, qual, ds_absolute_maximum, buffer + size); - - /* The value for 'no threshold' is -1, which will set the - * threshold outside of the buffer, just like we want it. - */ - ds_set(context->ds, qual, - ds_interrupt_threshold, buffer + size - ith); -} - -static int ds_request(struct ds_tracer *tracer, enum ds_qualifier qual, - struct task_struct *task, - void *base, size_t size, size_t th) -{ - struct ds_context *context; - unsigned long irq; - int error; - - error = -EOPNOTSUPP; - if (!ds_cfg.sizeof_ds) - goto out; - - error = -EINVAL; - if (!base) - goto out; - - /* we require some space to do alignment adjustments below */ - error = -EINVAL; - if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual])) - goto out; - - if (th != (size_t)-1) { - th *= ds_cfg.sizeof_rec[qual]; - - error = -EINVAL; - if (size <= th) - goto out; - } - - tracer->buffer = base; - tracer->size = size; - - error = -ENOMEM; - context = ds_get_context(task); - if (!context) - goto out; - tracer->context = context; - - - spin_lock_irqsave(&ds_lock, irq); - - error = -EPERM; - if (!check_tracer(task)) - goto out_unlock; - get_tracer(task); - - error = -EPERM; - if (context->owner[qual]) - goto out_put_tracer; - context->owner[qual] = tracer; - - spin_unlock_irqrestore(&ds_lock, irq); - - - ds_install_ds_config(context, qual, base, size, th); - - return 0; - - out_put_tracer: - put_tracer(task); - out_unlock: - spin_unlock_irqrestore(&ds_lock, irq); - ds_put_context(context); - tracer->context = NULL; - out: - return error; -} - -struct bts_tracer *ds_request_bts(struct task_struct *task, - void *base, size_t size, - bts_ovfl_callback_t ovfl, size_t th) -{ - struct bts_tracer *tracer; - int error; - - /* buffer overflow notification is not yet implemented */ - error = -EOPNOTSUPP; - if (ovfl) - goto out; - - error = -ENOMEM; - tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); - if (!tracer) - goto out; - tracer->ovfl = ovfl; - - error = ds_request(&tracer->ds, ds_bts, task, base, size, th); - if (error < 0) - goto out_tracer; - - return tracer; - - out_tracer: - kfree(tracer); - out: - return ERR_PTR(error); -} - -struct pebs_tracer *ds_request_pebs(struct task_struct *task, - void *base, size_t size, - pebs_ovfl_callback_t ovfl, size_t th) -{ - struct pebs_tracer *tracer; - int error; - - /* buffer overflow notification is not yet implemented */ - error = -EOPNOTSUPP; - if (ovfl) - goto out; - - error = -ENOMEM; - tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); - if (!tracer) - goto out; - tracer->ovfl = ovfl; - - error = ds_request(&tracer->ds, ds_pebs, task, base, size, th); - if (error < 0) - goto out_tracer; - - return tracer; - - out_tracer: - kfree(tracer); - out: - return ERR_PTR(error); -} - -static void ds_release(struct ds_tracer *tracer, enum ds_qualifier qual) -{ - WARN_ON_ONCE(tracer->context->owner[qual] != tracer); - tracer->context->owner[qual] = NULL; - - put_tracer(tracer->context->task); - ds_put_context(tracer->context); -} - -int ds_release_bts(struct bts_tracer *tracer) -{ - if (!tracer) - return -EINVAL; - - ds_release(&tracer->ds, ds_bts); - kfree(tracer); - - return 0; -} - -int ds_release_pebs(struct pebs_tracer *tracer) -{ - if (!tracer) - return -EINVAL; - - ds_release(&tracer->ds, ds_pebs); - kfree(tracer); - - return 0; -} - -static size_t ds_get_index(struct ds_context *context, enum ds_qualifier qual) -{ - unsigned long base, index; - - base = ds_get(context->ds, qual, ds_buffer_base); - index = ds_get(context->ds, qual, ds_index); - - return (index - base) / ds_cfg.sizeof_rec[qual]; -} - -int ds_get_bts_index(struct bts_tracer *tracer, size_t *pos) -{ - if (!tracer) - return -EINVAL; - - if (!pos) - return -EINVAL; - - *pos = ds_get_index(tracer->ds.context, ds_bts); - - return 0; -} - -int ds_get_pebs_index(struct pebs_tracer *tracer, size_t *pos) -{ - if (!tracer) - return -EINVAL; - - if (!pos) - return -EINVAL; - - *pos = ds_get_index(tracer->ds.context, ds_pebs); - - return 0; -} - -static size_t ds_get_end(struct ds_context *context, enum ds_qualifier qual) -{ - unsigned long base, max; - - base = ds_get(context->ds, qual, ds_buffer_base); - max = ds_get(context->ds, qual, ds_absolute_maximum); - - return (max - base) / ds_cfg.sizeof_rec[qual]; -} - -int ds_get_bts_end(struct bts_tracer *tracer, size_t *pos) -{ - if (!tracer) - return -EINVAL; - - if (!pos) - return -EINVAL; - - *pos = ds_get_end(tracer->ds.context, ds_bts); - - return 0; -} - -int ds_get_pebs_end(struct pebs_tracer *tracer, size_t *pos) -{ - if (!tracer) - return -EINVAL; - - if (!pos) - return -EINVAL; - - *pos = ds_get_end(tracer->ds.context, ds_pebs); - - return 0; -} - -static int ds_access(struct ds_context *context, enum ds_qualifier qual, - size_t index, const void **record) -{ - unsigned long base, idx; - - if (!record) - return -EINVAL; - - base = ds_get(context->ds, qual, ds_buffer_base); - idx = base + (index * ds_cfg.sizeof_rec[qual]); - - if (idx > ds_get(context->ds, qual, ds_absolute_maximum)) - return -EINVAL; - - *record = (const void *)idx; - - return ds_cfg.sizeof_rec[qual]; -} - -int ds_access_bts(struct bts_tracer *tracer, size_t index, - const void **record) -{ - if (!tracer) - return -EINVAL; - - return ds_access(tracer->ds.context, ds_bts, index, record); -} - -int ds_access_pebs(struct pebs_tracer *tracer, size_t index, - const void **record) -{ - if (!tracer) - return -EINVAL; - - return ds_access(tracer->ds.context, ds_pebs, index, record); -} - +/* + * Write raw data into the BTS or PEBS buffer. + * + * The remainder of any partially written record is zeroed out. + * + * context: the DS context + * qual: the buffer type + * record: the data to write + * size: the size of the data + */ static int ds_write(struct ds_context *context, enum ds_qualifier qual, const void *record, size_t size) { @@ -653,34 +407,455 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual, return bytes_written; } -int ds_write_bts(struct bts_tracer *tracer, const void *record, size_t size) + +/* + * Branch Trace Store (BTS) uses the following format. Different + * architectures vary in the size of those fields. + * - source linear address + * - destination linear address + * - flags + * + * Later architectures use 64bit pointers throughout, whereas earlier + * architectures use 32bit pointers in 32bit mode. + * + * We compute the base address for the first 8 fields based on: + * - the field size stored in the DS configuration + * - the relative field position + * + * In order to store additional information in the BTS buffer, we use + * a special source address to indicate that the record requires + * special interpretation. + * + * Netburst indicated via a bit in the flags field whether the branch + * was predicted; this is ignored. + * + * We use two levels of abstraction: + * - the raw data level defined here + * - an arch-independent level defined in ds.h + */ + +enum bts_field { + bts_from, + bts_to, + bts_flags, + + bts_qual = bts_from, + bts_jiffies = bts_to, + bts_pid = bts_flags, + + bts_qual_mask = (bts_qual_max - 1), + bts_escape = ((unsigned long)-1 & ~bts_qual_mask) +}; + +static inline unsigned long bts_get(const char *base, enum bts_field field) +{ + base += (ds_cfg.sizeof_field * field); + return *(unsigned long *)base; +} + +static inline void bts_set(char *base, enum bts_field field, unsigned long val) +{ + base += (ds_cfg.sizeof_field * field);; + (*(unsigned long *)base) = val; +} + + +/* + * The raw BTS data is architecture dependent. + * + * For higher-level users, we give an arch-independent view. + * - ds.h defines struct bts_struct + * - bts_read translates one raw bts record into a bts_struct + * - bts_write translates one bts_struct into the raw format and + * writes it into the top of the parameter tracer's buffer. + * + * return: bytes read/written on success; -Eerrno, otherwise + */ +static int bts_read(struct bts_tracer *tracer, const void *at, + struct bts_struct *out) { if (!tracer) return -EINVAL; - return ds_write(tracer->ds.context, ds_bts, record, size); + if (at < tracer->trace.ds.begin) + return -EINVAL; + + if (tracer->trace.ds.end < (at + tracer->trace.ds.size)) + return -EINVAL; + + memset(out, 0, sizeof(*out)); + if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) { + out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask); + out->variant.timestamp.jiffies = bts_get(at, bts_jiffies); + out->variant.timestamp.pid = bts_get(at, bts_pid); + } else { + out->qualifier = bts_branch; + out->variant.lbr.from = bts_get(at, bts_from); + out->variant.lbr.to = bts_get(at, bts_to); + } + + return ds_cfg.sizeof_rec[ds_bts]; } -int ds_write_pebs(struct pebs_tracer *tracer, const void *record, size_t size) +static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in) { + unsigned char raw[MAX_SIZEOF_BTS]; + if (!tracer) return -EINVAL; - return ds_write(tracer->ds.context, ds_pebs, record, size); + if (MAX_SIZEOF_BTS < ds_cfg.sizeof_rec[ds_bts]) + return -EOVERFLOW; + + switch (in->qualifier) { + case bts_invalid: + bts_set(raw, bts_from, 0); + bts_set(raw, bts_to, 0); + bts_set(raw, bts_flags, 0); + break; + case bts_branch: + bts_set(raw, bts_from, in->variant.lbr.from); + bts_set(raw, bts_to, in->variant.lbr.to); + bts_set(raw, bts_flags, 0); + break; + case bts_task_arrives: + case bts_task_departs: + bts_set(raw, bts_qual, (bts_escape | in->qualifier)); + bts_set(raw, bts_jiffies, in->variant.timestamp.jiffies); + bts_set(raw, bts_pid, in->variant.timestamp.pid); + break; + default: + return -EINVAL; + } + + return ds_write(tracer->ds.context, ds_bts, raw, + ds_cfg.sizeof_rec[ds_bts]); } -static void ds_reset_or_clear(struct ds_context *context, - enum ds_qualifier qual, int clear) + +static void ds_write_config(struct ds_context *context, + struct ds_trace *cfg, enum ds_qualifier qual) { - unsigned long base, end; + unsigned char *ds = context->ds; - base = ds_get(context->ds, qual, ds_buffer_base); - end = ds_get(context->ds, qual, ds_absolute_maximum); + ds_set(ds, qual, ds_buffer_base, (unsigned long)cfg->begin); + ds_set(ds, qual, ds_index, (unsigned long)cfg->top); + ds_set(ds, qual, ds_absolute_maximum, (unsigned long)cfg->end); + ds_set(ds, qual, ds_interrupt_threshold, (unsigned long)cfg->ith); +} - if (clear) - memset((void *)base, 0, end - base); +static void ds_read_config(struct ds_context *context, + struct ds_trace *cfg, enum ds_qualifier qual) +{ + unsigned char *ds = context->ds; - ds_set(context->ds, qual, ds_index, base); + cfg->begin = (void *)ds_get(ds, qual, ds_buffer_base); + cfg->top = (void *)ds_get(ds, qual, ds_index); + cfg->end = (void *)ds_get(ds, qual, ds_absolute_maximum); + cfg->ith = (void *)ds_get(ds, qual, ds_interrupt_threshold); +} + +static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual, + void *base, size_t size, size_t ith, + unsigned int flags) { + unsigned long buffer, adj; + + /* adjust the buffer address and size to meet alignment + * constraints: + * - buffer is double-word aligned + * - size is multiple of record size + * + * We checked the size at the very beginning; we have enough + * space to do the adjustment. + */ + buffer = (unsigned long)base; + + adj = ALIGN(buffer, DS_ALIGNMENT) - buffer; + buffer += adj; + size -= adj; + + trace->n = size / ds_cfg.sizeof_rec[qual]; + trace->size = ds_cfg.sizeof_rec[qual]; + + size = (trace->n * trace->size); + + trace->begin = (void *)buffer; + trace->top = trace->begin; + trace->end = (void *)(buffer + size); + /* The value for 'no threshold' is -1, which will set the + * threshold outside of the buffer, just like we want it. + */ + trace->ith = (void *)(buffer + size - ith); + + trace->flags = flags; +} + + +static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace, + enum ds_qualifier qual, struct task_struct *task, + void *base, size_t size, size_t th, unsigned int flags) +{ + struct ds_context *context; + int error; + + error = -EINVAL; + if (!base) + goto out; + + /* we require some space to do alignment adjustments below */ + error = -EINVAL; + if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual])) + goto out; + + if (th != (size_t)-1) { + th *= ds_cfg.sizeof_rec[qual]; + + error = -EINVAL; + if (size <= th) + goto out; + } + + tracer->buffer = base; + tracer->size = size; + + error = -ENOMEM; + context = ds_get_context(task); + if (!context) + goto out; + tracer->context = context; + + ds_init_ds_trace(trace, qual, base, size, th, flags); + + error = 0; + out: + return error; +} + +struct bts_tracer *ds_request_bts(struct task_struct *task, + void *base, size_t size, + bts_ovfl_callback_t ovfl, size_t th, + unsigned int flags) +{ + struct bts_tracer *tracer; + unsigned long irq; + int error; + + error = -EOPNOTSUPP; + if (!ds_cfg.ctl[dsf_bts]) + goto out; + + /* buffer overflow notification is not yet implemented */ + error = -EOPNOTSUPP; + if (ovfl) + goto out; + + error = -ENOMEM; + tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); + if (!tracer) + goto out; + tracer->ovfl = ovfl; + + error = ds_request(&tracer->ds, &tracer->trace.ds, + ds_bts, task, base, size, th, flags); + if (error < 0) + goto out_tracer; + + + spin_lock_irqsave(&ds_lock, irq); + + error = -EPERM; + if (!check_tracer(task)) + goto out_unlock; + get_tracer(task); + + error = -EPERM; + if (tracer->ds.context->bts_master) + goto out_put_tracer; + tracer->ds.context->bts_master = tracer; + + spin_unlock_irqrestore(&ds_lock, irq); + + + tracer->trace.read = bts_read; + tracer->trace.write = bts_write; + + ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts); + ds_resume_bts(tracer); + + return tracer; + + out_put_tracer: + put_tracer(task); + out_unlock: + spin_unlock_irqrestore(&ds_lock, irq); + ds_put_context(tracer->ds.context); + out_tracer: + kfree(tracer); + out: + return ERR_PTR(error); +} + +struct pebs_tracer *ds_request_pebs(struct task_struct *task, + void *base, size_t size, + pebs_ovfl_callback_t ovfl, size_t th, + unsigned int flags) +{ + struct pebs_tracer *tracer; + unsigned long irq; + int error; + + /* buffer overflow notification is not yet implemented */ + error = -EOPNOTSUPP; + if (ovfl) + goto out; + + error = -ENOMEM; + tracer = kzalloc(sizeof(*tracer), GFP_KERNEL); + if (!tracer) + goto out; + tracer->ovfl = ovfl; + + error = ds_request(&tracer->ds, &tracer->trace.ds, + ds_pebs, task, base, size, th, flags); + if (error < 0) + goto out_tracer; + + spin_lock_irqsave(&ds_lock, irq); + + error = -EPERM; + if (!check_tracer(task)) + goto out_unlock; + get_tracer(task); + + error = -EPERM; + if (tracer->ds.context->pebs_master) + goto out_put_tracer; + tracer->ds.context->pebs_master = tracer; + + spin_unlock_irqrestore(&ds_lock, irq); + + ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts); + ds_resume_pebs(tracer); + + return tracer; + + out_put_tracer: + put_tracer(task); + out_unlock: + spin_unlock_irqrestore(&ds_lock, irq); + ds_put_context(tracer->ds.context); + out_tracer: + kfree(tracer); + out: + return ERR_PTR(error); +} + +void ds_release_bts(struct bts_tracer *tracer) +{ + if (!tracer) + return; + + ds_suspend_bts(tracer); + + WARN_ON_ONCE(tracer->ds.context->bts_master != tracer); + tracer->ds.context->bts_master = NULL; + + put_tracer(tracer->ds.context->task); + ds_put_context(tracer->ds.context); + + kfree(tracer); +} + +void ds_suspend_bts(struct bts_tracer *tracer) +{ + struct task_struct *task; + + if (!tracer) + return; + + task = tracer->ds.context->task; + + if (!task || (task == current)) + update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL); + + if (task) { + task->thread.debugctlmsr &= ~BTS_CONTROL; + + if (!task->thread.debugctlmsr) + clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR); + } +} + +void ds_resume_bts(struct bts_tracer *tracer) +{ + struct task_struct *task; + unsigned long control; + + if (!tracer) + return; + + task = tracer->ds.context->task; + + control = ds_cfg.ctl[dsf_bts]; + if (!(tracer->trace.ds.flags & BTS_KERNEL)) + control |= ds_cfg.ctl[dsf_bts_kernel]; + if (!(tracer->trace.ds.flags & BTS_USER)) + control |= ds_cfg.ctl[dsf_bts_user]; + + if (task) { + task->thread.debugctlmsr |= control; + set_tsk_thread_flag(task, TIF_DEBUGCTLMSR); + } + + if (!task || (task == current)) + update_debugctlmsr(get_debugctlmsr() | control); +} + +void ds_release_pebs(struct pebs_tracer *tracer) +{ + if (!tracer) + return; + + ds_suspend_pebs(tracer); + + WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer); + tracer->ds.context->pebs_master = NULL; + + put_tracer(tracer->ds.context->task); + ds_put_context(tracer->ds.context); + + kfree(tracer); +} + +void ds_suspend_pebs(struct pebs_tracer *tracer) +{ + +} + +void ds_resume_pebs(struct pebs_tracer *tracer) +{ + +} + +const struct bts_trace *ds_read_bts(struct bts_tracer *tracer) +{ + if (!tracer) + return NULL; + + ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_bts); + return &tracer->trace; +} + +const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer) +{ + if (!tracer) + return NULL; + + ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs); + tracer->trace.reset_value = + *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)); + + return &tracer->trace; } int ds_reset_bts(struct bts_tracer *tracer) @@ -688,7 +863,10 @@ int ds_reset_bts(struct bts_tracer *tracer) if (!tracer) return -EINVAL; - ds_reset_or_clear(tracer->ds.context, ds_bts, /* clear = */ 0); + tracer->trace.ds.top = tracer->trace.ds.begin; + + ds_set(tracer->ds.context->ds, ds_bts, ds_index, + (unsigned long)tracer->trace.ds.top); return 0; } @@ -698,40 +876,10 @@ int ds_reset_pebs(struct pebs_tracer *tracer) if (!tracer) return -EINVAL; - ds_reset_or_clear(tracer->ds.context, ds_pebs, /* clear = */ 0); + tracer->trace.ds.top = tracer->trace.ds.begin; - return 0; -} - -int ds_clear_bts(struct bts_tracer *tracer) -{ - if (!tracer) - return -EINVAL; - - ds_reset_or_clear(tracer->ds.context, ds_bts, /* clear = */ 1); - - return 0; -} - -int ds_clear_pebs(struct pebs_tracer *tracer) -{ - if (!tracer) - return -EINVAL; - - ds_reset_or_clear(tracer->ds.context, ds_pebs, /* clear = */ 1); - - return 0; -} - -int ds_get_pebs_reset(struct pebs_tracer *tracer, u64 *value) -{ - if (!tracer) - return -EINVAL; - - if (!value) - return -EINVAL; - - *value = *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)); + ds_set(tracer->ds.context->ds, ds_bts, ds_index, + (unsigned long)tracer->trace.ds.top); return 0; } @@ -746,35 +894,59 @@ int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value) return 0; } -static const struct ds_configuration ds_cfg_var = { - .sizeof_ds = sizeof(long) * 12, - .sizeof_field = sizeof(long), - .sizeof_rec[ds_bts] = sizeof(long) * 3, +static const struct ds_configuration ds_cfg_netburst = { + .name = "netburst", + .ctl[dsf_bts] = (1 << 2) | (1 << 3), + .ctl[dsf_bts_kernel] = (1 << 5), + .ctl[dsf_bts_user] = (1 << 6), + + .sizeof_field = sizeof(long), + .sizeof_rec[ds_bts] = sizeof(long) * 3, #ifdef __i386__ - .sizeof_rec[ds_pebs] = sizeof(long) * 10 + .sizeof_rec[ds_pebs] = sizeof(long) * 10, #else - .sizeof_rec[ds_pebs] = sizeof(long) * 18 + .sizeof_rec[ds_pebs] = sizeof(long) * 18, #endif }; -static const struct ds_configuration ds_cfg_64 = { - .sizeof_ds = 8 * 12, - .sizeof_field = 8, - .sizeof_rec[ds_bts] = 8 * 3, +static const struct ds_configuration ds_cfg_pentium_m = { + .name = "pentium m", + .ctl[dsf_bts] = (1 << 6) | (1 << 7), + + .sizeof_field = sizeof(long), + .sizeof_rec[ds_bts] = sizeof(long) * 3, #ifdef __i386__ - .sizeof_rec[ds_pebs] = 8 * 10 + .sizeof_rec[ds_pebs] = sizeof(long) * 10, #else - .sizeof_rec[ds_pebs] = 8 * 18 + .sizeof_rec[ds_pebs] = sizeof(long) * 18, #endif }; +static const struct ds_configuration ds_cfg_core2 = { + .name = "core 2", + .ctl[dsf_bts] = (1 << 6) | (1 << 7), + .ctl[dsf_bts_kernel] = (1 << 9), + .ctl[dsf_bts_user] = (1 << 10), + + .sizeof_field = 8, + .sizeof_rec[ds_bts] = 8 * 3, + .sizeof_rec[ds_pebs] = 8 * 18, +}; -static inline void +static void ds_configure(const struct ds_configuration *cfg) { + memset(&ds_cfg, 0, sizeof(ds_cfg)); ds_cfg = *cfg; - printk(KERN_INFO "DS available\n"); + printk(KERN_INFO "[ds] using %s configuration\n", ds_cfg.name); - WARN_ON_ONCE(MAX_SIZEOF_DS < ds_cfg.sizeof_ds); + if (!cpu_has_bts) { + ds_cfg.ctl[dsf_bts] = 0; + printk(KERN_INFO "[ds] bts not available\n"); + } + if (!cpu_has_pebs) + printk(KERN_INFO "[ds] pebs not available\n"); + + WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_field)); } void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) @@ -787,10 +959,10 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) break; case 0xD: case 0xE: /* Pentium M */ - ds_configure(&ds_cfg_var); + ds_configure(&ds_cfg_pentium_m); break; default: /* Core2, Atom, ... */ - ds_configure(&ds_cfg_64); + ds_configure(&ds_cfg_core2); break; } break; @@ -799,7 +971,7 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) case 0x0: case 0x1: case 0x2: /* Netburst */ - ds_configure(&ds_cfg_var); + ds_configure(&ds_cfg_netburst); break; default: /* sorry, don't know about them */ @@ -812,14 +984,41 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) } } -void ds_free(struct ds_context *context) +/* + * Change the DS configuration from tracing prev to tracing next. + */ +void ds_switch_to(struct task_struct *prev, struct task_struct *next) { - /* This is called when the task owning the parameter context - * is dying. There should not be any user of that context left - * to disturb us, anymore. */ - unsigned long leftovers = context->count; - while (leftovers--) { - put_tracer(context->task); - ds_put_context(context); + struct ds_context *prev_ctx = prev->thread.ds_ctx; + struct ds_context *next_ctx = next->thread.ds_ctx; + + if (prev_ctx) { + update_debugctlmsr(0); + + if (prev_ctx->bts_master && + (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) { + struct bts_struct ts = { + .qualifier = bts_task_departs, + .variant.timestamp.jiffies = jiffies_64, + .variant.timestamp.pid = prev->pid + }; + bts_write(prev_ctx->bts_master, &ts); + } } + + if (next_ctx) { + if (next_ctx->bts_master && + (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) { + struct bts_struct ts = { + .qualifier = bts_task_arrives, + .variant.timestamp.jiffies = jiffies_64, + .variant.timestamp.pid = next->pid + }; + bts_write(next_ctx->bts_master, &ts); + } + + wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds); + } + + update_debugctlmsr(next->thread.debugctlmsr); } diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c index 24c2276aa453..605eff9a8ac0 100644 --- a/arch/x86/kernel/process_32.c +++ b/arch/x86/kernel/process_32.c @@ -252,11 +252,14 @@ void exit_thread(void) put_cpu(); } #ifdef CONFIG_X86_DS - /* Free any DS contexts that have not been properly released. */ - if (unlikely(current->thread.ds_ctx)) { - /* we clear debugctl to make sure DS is not used. */ - update_debugctlmsr(0); - ds_free(current->thread.ds_ctx); + /* Free any BTS tracers that have not been properly released. */ + if (unlikely(current->bts)) { + ds_release_bts(current->bts); + current->bts = NULL; + + kfree(current->bts_buffer); + current->bts_buffer = NULL; + current->bts_size = 0; } #endif /* CONFIG_X86_DS */ } @@ -420,48 +423,19 @@ int set_tsc_mode(unsigned int val) return 0; } -#ifdef CONFIG_X86_DS -static int update_debugctl(struct thread_struct *prev, - struct thread_struct *next, unsigned long debugctl) -{ - unsigned long ds_prev = 0; - unsigned long ds_next = 0; - - if (prev->ds_ctx) - ds_prev = (unsigned long)prev->ds_ctx->ds; - if (next->ds_ctx) - ds_next = (unsigned long)next->ds_ctx->ds; - - if (ds_next != ds_prev) { - /* we clear debugctl to make sure DS - * is not in use when we change it */ - debugctl = 0; - update_debugctlmsr(0); - wrmsr(MSR_IA32_DS_AREA, ds_next, 0); - } - return debugctl; -} -#else -static int update_debugctl(struct thread_struct *prev, - struct thread_struct *next, unsigned long debugctl) -{ - return debugctl; -} -#endif /* CONFIG_X86_DS */ - static noinline void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, struct tss_struct *tss) { struct thread_struct *prev, *next; - unsigned long debugctl; prev = &prev_p->thread; next = &next_p->thread; - debugctl = update_debugctl(prev, next, prev->debugctlmsr); - - if (next->debugctlmsr != debugctl) + if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) || + test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR)) + ds_switch_to(prev_p, next_p); + else if (next->debugctlmsr != prev->debugctlmsr) update_debugctlmsr(next->debugctlmsr); if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { @@ -483,15 +457,6 @@ __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, hard_enable_TSC(); } -#ifdef CONFIG_X86_PTRACE_BTS - if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) - ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); - - if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) - ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); -#endif /* CONFIG_X86_PTRACE_BTS */ - - if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { /* * Disable the bitmap via an invalid offset. We still cache diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c index fbb321d53d34..1cfd2a4bf853 100644 --- a/arch/x86/kernel/process_64.c +++ b/arch/x86/kernel/process_64.c @@ -237,11 +237,14 @@ void exit_thread(void) put_cpu(); } #ifdef CONFIG_X86_DS - /* Free any DS contexts that have not been properly released. */ - if (unlikely(t->ds_ctx)) { - /* we clear debugctl to make sure DS is not used. */ - update_debugctlmsr(0); - ds_free(t->ds_ctx); + /* Free any BTS tracers that have not been properly released. */ + if (unlikely(current->bts)) { + ds_release_bts(current->bts); + current->bts = NULL; + + kfree(current->bts_buffer); + current->bts_buffer = NULL; + current->bts_size = 0; } #endif /* CONFIG_X86_DS */ } @@ -471,35 +474,14 @@ static inline void __switch_to_xtra(struct task_struct *prev_p, struct tss_struct *tss) { struct thread_struct *prev, *next; - unsigned long debugctl; prev = &prev_p->thread, next = &next_p->thread; - debugctl = prev->debugctlmsr; - -#ifdef CONFIG_X86_DS - { - unsigned long ds_prev = 0, ds_next = 0; - - if (prev->ds_ctx) - ds_prev = (unsigned long)prev->ds_ctx->ds; - if (next->ds_ctx) - ds_next = (unsigned long)next->ds_ctx->ds; - - if (ds_next != ds_prev) { - /* - * We clear debugctl to make sure DS - * is not in use when we change it: - */ - debugctl = 0; - update_debugctlmsr(0); - wrmsrl(MSR_IA32_DS_AREA, ds_next); - } - } -#endif /* CONFIG_X86_DS */ - - if (next->debugctlmsr != debugctl) + if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) || + test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR)) + ds_switch_to(prev_p, next_p); + else if (next->debugctlmsr != prev->debugctlmsr) update_debugctlmsr(next->debugctlmsr); if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { @@ -534,14 +516,6 @@ static inline void __switch_to_xtra(struct task_struct *prev_p, */ memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); } - -#ifdef CONFIG_X86_PTRACE_BTS - if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS)) - ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS); - - if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS)) - ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES); -#endif /* CONFIG_X86_PTRACE_BTS */ } /* diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index b2998fe1166b..45e9855da2d2 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -581,153 +581,73 @@ static int ioperm_get(struct task_struct *target, } #ifdef CONFIG_X86_PTRACE_BTS -/* - * The configuration for a particular BTS hardware implementation. - */ -struct bts_configuration { - /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */ - unsigned char sizeof_bts; - /* the size of a field in the BTS record in bytes */ - unsigned char sizeof_field; - /* a bitmask to enable/disable BTS in DEBUGCTL MSR */ - unsigned long debugctl_mask; -}; -static struct bts_configuration bts_cfg; - -#define BTS_MAX_RECORD_SIZE (8 * 3) - - -/* - * Branch Trace Store (BTS) uses the following format. Different - * architectures vary in the size of those fields. - * - source linear address - * - destination linear address - * - flags - * - * Later architectures use 64bit pointers throughout, whereas earlier - * architectures use 32bit pointers in 32bit mode. - * - * We compute the base address for the first 8 fields based on: - * - the field size stored in the DS configuration - * - the relative field position - * - * In order to store additional information in the BTS buffer, we use - * a special source address to indicate that the record requires - * special interpretation. - * - * Netburst indicated via a bit in the flags field whether the branch - * was predicted; this is ignored. - */ - -enum bts_field { - bts_from = 0, - bts_to, - bts_flags, - - bts_escape = (unsigned long)-1, - bts_qual = bts_to, - bts_jiffies = bts_flags -}; - -static inline unsigned long bts_get(const char *base, enum bts_field field) -{ - base += (bts_cfg.sizeof_field * field); - return *(unsigned long *)base; -} - -static inline void bts_set(char *base, enum bts_field field, unsigned long val) -{ - base += (bts_cfg.sizeof_field * field);; - (*(unsigned long *)base) = val; -} - -/* - * Translate a BTS record from the raw format into the bts_struct format - * - * out (out): bts_struct interpretation - * raw: raw BTS record - */ -static void ptrace_bts_translate_record(struct bts_struct *out, const void *raw) -{ - memset(out, 0, sizeof(*out)); - if (bts_get(raw, bts_from) == bts_escape) { - out->qualifier = bts_get(raw, bts_qual); - out->variant.jiffies = bts_get(raw, bts_jiffies); - } else { - out->qualifier = BTS_BRANCH; - out->variant.lbr.from_ip = bts_get(raw, bts_from); - out->variant.lbr.to_ip = bts_get(raw, bts_to); - } -} - static int ptrace_bts_read_record(struct task_struct *child, size_t index, struct bts_struct __user *out) { - struct bts_struct ret; - const void *bts_record; - size_t bts_index, bts_end; + const struct bts_trace *trace; + struct bts_struct bts; + const unsigned char *at; int error; - error = ds_get_bts_end(child->bts, &bts_end); + trace = ds_read_bts(child->bts); + if (!trace) + return -EPERM; + + at = trace->ds.top - ((index + 1) * trace->ds.size); + if ((void *)at < trace->ds.begin) + at += (trace->ds.n * trace->ds.size); + + if (!trace->read) + return -EOPNOTSUPP; + + error = trace->read(child->bts, at, &bts); if (error < 0) return error; - if (bts_end <= index) - return -EINVAL; - - error = ds_get_bts_index(child->bts, &bts_index); - if (error < 0) - return error; - - /* translate the ptrace bts index into the ds bts index */ - bts_index += bts_end - (index + 1); - if (bts_end <= bts_index) - bts_index -= bts_end; - - error = ds_access_bts(child->bts, bts_index, &bts_record); - if (error < 0) - return error; - - ptrace_bts_translate_record(&ret, bts_record); - - if (copy_to_user(out, &ret, sizeof(ret))) + if (copy_to_user(out, &bts, sizeof(bts))) return -EFAULT; - return sizeof(ret); + return sizeof(bts); } static int ptrace_bts_drain(struct task_struct *child, long size, struct bts_struct __user *out) { - struct bts_struct ret; - const unsigned char *raw; - size_t end, i; - int error; + const struct bts_trace *trace; + const unsigned char *at; + int error, drained = 0; - error = ds_get_bts_index(child->bts, &end); - if (error < 0) - return error; + trace = ds_read_bts(child->bts); + if (!trace) + return -EPERM; - if (size < (end * sizeof(struct bts_struct))) + if (!trace->read) + return -EOPNOTSUPP; + + if (size < (trace->ds.top - trace->ds.begin)) return -EIO; - error = ds_access_bts(child->bts, 0, (const void **)&raw); - if (error < 0) - return error; + for (at = trace->ds.begin; (void *)at < trace->ds.top; + out++, drained++, at += trace->ds.size) { + struct bts_struct bts; + int error; - for (i = 0; i < end; i++, out++, raw += bts_cfg.sizeof_bts) { - ptrace_bts_translate_record(&ret, raw); + error = trace->read(child->bts, at, &bts); + if (error < 0) + return error; - if (copy_to_user(out, &ret, sizeof(ret))) + if (copy_to_user(out, &bts, sizeof(bts))) return -EFAULT; } - error = ds_clear_bts(child->bts); + memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); + + error = ds_reset_bts(child->bts); if (error < 0) return error; - return end; + return drained; } static int ptrace_bts_config(struct task_struct *child, @@ -735,136 +655,89 @@ static int ptrace_bts_config(struct task_struct *child, const struct ptrace_bts_config __user *ucfg) { struct ptrace_bts_config cfg; - int error = 0; + unsigned int flags = 0; - error = -EOPNOTSUPP; - if (!bts_cfg.sizeof_bts) - goto errout; - - error = -EIO; if (cfg_size < sizeof(cfg)) - goto errout; + return -EIO; - error = -EFAULT; if (copy_from_user(&cfg, ucfg, sizeof(cfg))) - goto errout; + return -EFAULT; - error = -EINVAL; - if ((cfg.flags & PTRACE_BTS_O_SIGNAL) && - !(cfg.flags & PTRACE_BTS_O_ALLOC)) - goto errout; - - if (cfg.flags & PTRACE_BTS_O_ALLOC) { - bts_ovfl_callback_t ovfl = NULL; - unsigned int sig = 0; - - error = -EINVAL; - if (cfg.size < (10 * bts_cfg.sizeof_bts)) - goto errout; - - if (cfg.flags & PTRACE_BTS_O_SIGNAL) { - if (!cfg.signal) - goto errout; - - error = -EOPNOTSUPP; - goto errout; - - sig = cfg.signal; - } - - if (child->bts) { - (void)ds_release_bts(child->bts); - kfree(child->bts_buffer); - - child->bts = NULL; - child->bts_buffer = NULL; - } - - error = -ENOMEM; - child->bts_buffer = kzalloc(cfg.size, GFP_KERNEL); - if (!child->bts_buffer) - goto errout; - - child->bts = ds_request_bts(child, child->bts_buffer, cfg.size, - ovfl, /* th = */ (size_t)-1); - if (IS_ERR(child->bts)) { - error = PTR_ERR(child->bts); - kfree(child->bts_buffer); - child->bts = NULL; - child->bts_buffer = NULL; - goto errout; - } - - child->thread.bts_ovfl_signal = sig; + if (child->bts) { + ds_release_bts(child->bts); + child->bts = NULL; } - error = -EINVAL; - if (!child->thread.ds_ctx && cfg.flags) - goto errout; + if (cfg.flags & PTRACE_BTS_O_SIGNAL) { + if (!cfg.signal) + return -EINVAL; + + return -EOPNOTSUPP; + + child->thread.bts_ovfl_signal = cfg.signal; + } + + if ((cfg.flags & PTRACE_BTS_O_ALLOC) && + (cfg.size != child->bts_size)) { + kfree(child->bts_buffer); + + child->bts_size = cfg.size; + child->bts_buffer = kzalloc(cfg.size, GFP_KERNEL); + if (!child->bts_buffer) { + child->bts_size = 0; + return -ENOMEM; + } + } if (cfg.flags & PTRACE_BTS_O_TRACE) - child->thread.debugctlmsr |= bts_cfg.debugctl_mask; - else - child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask; + flags |= BTS_USER; if (cfg.flags & PTRACE_BTS_O_SCHED) - set_tsk_thread_flag(child, TIF_BTS_TRACE_TS); - else - clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); + flags |= BTS_TIMESTAMPS; - error = sizeof(cfg); + child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size, + /* ovfl = */ NULL, /* th = */ (size_t)-1, + flags); + if (IS_ERR(child->bts)) { + int error = PTR_ERR(child->bts); -out: - if (child->thread.debugctlmsr) - set_tsk_thread_flag(child, TIF_DEBUGCTLMSR); - else - clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR); + kfree(child->bts_buffer); + child->bts = NULL; + child->bts_buffer = NULL; + child->bts_size = 0; - return error; + return error; + } -errout: - child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask; - clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); - goto out; + return sizeof(cfg); } static int ptrace_bts_status(struct task_struct *child, long cfg_size, struct ptrace_bts_config __user *ucfg) { + const struct bts_trace *trace; struct ptrace_bts_config cfg; - size_t end; - const void *base, *max; - int error; if (cfg_size < sizeof(cfg)) return -EIO; - error = ds_get_bts_end(child->bts, &end); - if (error < 0) - return error; - - error = ds_access_bts(child->bts, /* index = */ 0, &base); - if (error < 0) - return error; - - error = ds_access_bts(child->bts, /* index = */ end, &max); - if (error < 0) - return error; + trace = ds_read_bts(child->bts); + if (!trace) + return -EPERM; memset(&cfg, 0, sizeof(cfg)); - cfg.size = (max - base); + cfg.size = trace->ds.end - trace->ds.begin; cfg.signal = child->thread.bts_ovfl_signal; cfg.bts_size = sizeof(struct bts_struct); if (cfg.signal) cfg.flags |= PTRACE_BTS_O_SIGNAL; - if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) && - child->thread.debugctlmsr & bts_cfg.debugctl_mask) + if (trace->ds.flags & BTS_USER) cfg.flags |= PTRACE_BTS_O_TRACE; - if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS)) + if (trace->ds.flags & BTS_TIMESTAMPS) cfg.flags |= PTRACE_BTS_O_SCHED; if (copy_to_user(ucfg, &cfg, sizeof(cfg))) @@ -873,105 +746,28 @@ static int ptrace_bts_status(struct task_struct *child, return sizeof(cfg); } -static int ptrace_bts_write_record(struct task_struct *child, - const struct bts_struct *in) +static int ptrace_bts_clear(struct task_struct *child) { - unsigned char bts_record[BTS_MAX_RECORD_SIZE]; + const struct bts_trace *trace; - if (BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts) - return -EOVERFLOW; + trace = ds_read_bts(child->bts); + if (!trace) + return -EPERM; - memset(bts_record, 0, bts_cfg.sizeof_bts); - switch (in->qualifier) { - case BTS_INVALID: - break; + memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size); - case BTS_BRANCH: - bts_set(bts_record, bts_from, in->variant.lbr.from_ip); - bts_set(bts_record, bts_to, in->variant.lbr.to_ip); - break; - - case BTS_TASK_ARRIVES: - case BTS_TASK_DEPARTS: - bts_set(bts_record, bts_from, bts_escape); - bts_set(bts_record, bts_qual, in->qualifier); - bts_set(bts_record, bts_jiffies, in->variant.jiffies); - break; - - default: - return -EINVAL; - } - - return ds_write_bts(child->bts, bts_record, bts_cfg.sizeof_bts); + return ds_reset_bts(child->bts); } -void ptrace_bts_take_timestamp(struct task_struct *tsk, - enum bts_qualifier qualifier) +static int ptrace_bts_size(struct task_struct *child) { - struct bts_struct rec = { - .qualifier = qualifier, - .variant.jiffies = jiffies_64 - }; + const struct bts_trace *trace; - ptrace_bts_write_record(tsk, &rec); -} + trace = ds_read_bts(child->bts); + if (!trace) + return -EPERM; -static const struct bts_configuration bts_cfg_netburst = { - .sizeof_bts = sizeof(long) * 3, - .sizeof_field = sizeof(long), - .debugctl_mask = (1<<2)|(1<<3)|(1<<5) -}; - -static const struct bts_configuration bts_cfg_pentium_m = { - .sizeof_bts = sizeof(long) * 3, - .sizeof_field = sizeof(long), - .debugctl_mask = (1<<6)|(1<<7) -}; - -static const struct bts_configuration bts_cfg_core2 = { - .sizeof_bts = 8 * 3, - .sizeof_field = 8, - .debugctl_mask = (1<<6)|(1<<7)|(1<<9) -}; - -static inline void bts_configure(const struct bts_configuration *cfg) -{ - bts_cfg = *cfg; -} - -void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c) -{ - switch (c->x86) { - case 0x6: - switch (c->x86_model) { - case 0 ... 0xC: - /* sorry, don't know about them */ - break; - case 0xD: - case 0xE: /* Pentium M */ - bts_configure(&bts_cfg_pentium_m); - break; - default: /* Core2, Atom, ... */ - bts_configure(&bts_cfg_core2); - break; - } - break; - case 0xF: - switch (c->x86_model) { - case 0x0: - case 0x1: - case 0x2: /* Netburst */ - bts_configure(&bts_cfg_netburst); - break; - default: - /* sorry, don't know about them */ - break; - } - break; - default: - /* sorry, don't know about them */ - break; - } + return (trace->ds.top - trace->ds.begin) / trace->ds.size; } #endif /* CONFIG_X86_PTRACE_BTS */ @@ -988,15 +784,12 @@ void ptrace_disable(struct task_struct *child) #endif #ifdef CONFIG_X86_PTRACE_BTS if (child->bts) { - (void)ds_release_bts(child->bts); + ds_release_bts(child->bts); + child->bts = NULL; + kfree(child->bts_buffer); child->bts_buffer = NULL; - - child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask; - if (!child->thread.debugctlmsr) - clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR); - - clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS); + child->bts_size = 0; } #endif /* CONFIG_X86_PTRACE_BTS */ } @@ -1129,16 +922,9 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data) (child, data, (struct ptrace_bts_config __user *)addr); break; - case PTRACE_BTS_SIZE: { - size_t size; - - ret = ds_get_bts_index(child->bts, &size); - if (ret == 0) { - WARN_ON_ONCE(size != (int) size); - ret = (int) size; - } + case PTRACE_BTS_SIZE: + ret = ptrace_bts_size(child); break; - } case PTRACE_BTS_GET: ret = ptrace_bts_read_record @@ -1146,7 +932,7 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data) break; case PTRACE_BTS_CLEAR: - ret = ds_clear_bts(child->bts); + ret = ptrace_bts_clear(child); break; case PTRACE_BTS_DRAIN: @@ -1409,6 +1195,14 @@ long compat_arch_ptrace(struct task_struct *child, compat_long_t request, case PTRACE_GET_THREAD_AREA: case PTRACE_SET_THREAD_AREA: +#ifdef CONFIG_X86_PTRACE_BTS + case PTRACE_BTS_CONFIG: + case PTRACE_BTS_STATUS: + case PTRACE_BTS_SIZE: + case PTRACE_BTS_GET: + case PTRACE_BTS_CLEAR: + case PTRACE_BTS_DRAIN: +#endif /* CONFIG_X86_PTRACE_BTS */ return arch_ptrace(child, request, addr, data); default: diff --git a/include/linux/sched.h b/include/linux/sched.h index 4b81fc5f7731..dc5ea65dc716 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1176,6 +1176,7 @@ struct task_struct { * The buffer to hold the BTS data. */ void *bts_buffer; + size_t bts_size; #endif /* CONFIG_X86_PTRACE_BTS */ /* PID/PID hash table linkage. */ From a93751cab71d63126687551823ed3e70cd85854a Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Thu, 11 Dec 2008 13:53:26 +0100 Subject: [PATCH 3/9] x86, bts, ftrace: adapt the hw-branch-tracer to the ds.c interface Impact: restructure code, cleanup Remove BTS bits from the hw-branch-tracer (renamed from bts-tracer) and use the ds interface. Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- kernel/trace/Kconfig | 4 +- kernel/trace/Makefile | 2 +- kernel/trace/trace.c | 2 +- kernel/trace/trace.h | 14 +- .../{trace_bts.c => trace_hw_branches.c} | 163 +++++------------- 5 files changed, 56 insertions(+), 129 deletions(-) rename kernel/trace/{trace_bts.c => trace_hw_branches.c} (53%) diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index bde6f03512d5..d8bae6f4219e 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -251,9 +251,9 @@ config STACK_TRACER Say N if unsure. -config BTS_TRACER +config HW_BRANCH_TRACER depends on HAVE_HW_BRANCH_TRACER - bool "Trace branches" + bool "Trace hw branches" select TRACING help This tracer records all branches on the system in a circular diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile index 62dc561b6676..349d5a93653f 100644 --- a/kernel/trace/Makefile +++ b/kernel/trace/Makefile @@ -31,7 +31,7 @@ obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o obj-$(CONFIG_BOOT_TRACER) += trace_boot.o obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o -obj-$(CONFIG_BTS_TRACER) += trace_bts.o +obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o obj-$(CONFIG_POWER_TRACER) += trace_power.o libftrace-y := ftrace.o diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 8ebe0070c47a..639344a4d3a2 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2425,7 +2425,7 @@ __tracing_open(struct inode *inode, struct file *file, int *ret) /* Notify the tracer early; before we stop tracing. */ if (iter->trace && iter->trace->open) - iter->trace->open(iter); + iter->trace->open(iter); /* Annotate start of buffers if we had overruns */ if (ring_buffer_overruns(iter->tr->buffer)) diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index 5ac697065a48..f07c246dd73d 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -28,7 +28,7 @@ enum trace_type { TRACE_GRAPH_RET, TRACE_GRAPH_ENT, TRACE_USER_STACK, - TRACE_BTS, + TRACE_HW_BRANCHES, TRACE_POWER, __TRACE_LAST_TYPE @@ -159,10 +159,10 @@ struct trace_branch { char correct; }; -struct bts_entry { +struct hw_branch_entry { struct trace_entry ent; - unsigned long from; - unsigned long to; + u64 from; + u64 to; }; struct trace_power { @@ -278,7 +278,7 @@ extern void __ftrace_bad_type(void); TRACE_GRAPH_ENT); \ IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry, \ TRACE_GRAPH_RET); \ - IF_ASSIGN(var, ent, struct bts_entry, TRACE_BTS);\ + IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\ IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \ __ftrace_bad_type(); \ } while (0) @@ -414,9 +414,7 @@ void trace_function(struct trace_array *tr, void trace_graph_return(struct ftrace_graph_ret *trace); int trace_graph_entry(struct ftrace_graph_ent *trace); -void trace_bts(struct trace_array *tr, - unsigned long from, - unsigned long to); +void trace_hw_branch(struct trace_array *tr, u64 from, u64 to); void tracing_start_cmdline_record(void); void tracing_stop_cmdline_record(void); diff --git a/kernel/trace/trace_bts.c b/kernel/trace/trace_hw_branches.c similarity index 53% rename from kernel/trace/trace_bts.c rename to kernel/trace/trace_hw_branches.c index 23b76e4690ef..ee29e012aa97 100644 --- a/kernel/trace/trace_bts.c +++ b/kernel/trace/trace_hw_branches.c @@ -1,5 +1,5 @@ /* - * BTS tracer + * h/w branch tracer for x86 based on bts * * Copyright (C) 2008 Markus Metzger * @@ -25,68 +25,6 @@ static DEFINE_PER_CPU(unsigned char[SIZEOF_BTS], buffer); #define this_buffer per_cpu(buffer, smp_processor_id()) -/* - * Information to interpret a BTS record. - * This will go into an in-kernel BTS interface. - */ -static unsigned char sizeof_field; -static unsigned long debugctl_mask; - -#define sizeof_bts (3 * sizeof_field) - -static void bts_trace_cpuinit(struct cpuinfo_x86 *c) -{ - switch (c->x86) { - case 0x6: - switch (c->x86_model) { - case 0x0 ... 0xC: - break; - case 0xD: - case 0xE: /* Pentium M */ - sizeof_field = sizeof(long); - debugctl_mask = (1<<6)|(1<<7); - break; - default: - sizeof_field = 8; - debugctl_mask = (1<<6)|(1<<7); - break; - } - break; - case 0xF: - switch (c->x86_model) { - case 0x0: - case 0x1: - case 0x2: /* Netburst */ - sizeof_field = sizeof(long); - debugctl_mask = (1<<2)|(1<<3); - break; - default: - /* sorry, don't know about them */ - break; - } - break; - default: - /* sorry, don't know about them */ - break; - } -} - -static inline void bts_enable(void) -{ - unsigned long debugctl; - - rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); - wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl | debugctl_mask); -} - -static inline void bts_disable(void) -{ - unsigned long debugctl; - - rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); - wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl & ~debugctl_mask); -} - static void bts_trace_reset(struct trace_array *tr) { int cpu; @@ -99,15 +37,17 @@ static void bts_trace_reset(struct trace_array *tr) static void bts_trace_start_cpu(void *arg) { + if (this_tracer) + ds_release_bts(this_tracer); + this_tracer = ds_request_bts(/* task = */ NULL, this_buffer, SIZEOF_BTS, - /* ovfl = */ NULL, /* th = */ (size_t)-1); + /* ovfl = */ NULL, /* th = */ (size_t)-1, + BTS_KERNEL); if (IS_ERR(this_tracer)) { this_tracer = NULL; return; } - - bts_enable(); } static void bts_trace_start(struct trace_array *tr) @@ -123,8 +63,6 @@ static void bts_trace_start(struct trace_array *tr) static void bts_trace_stop_cpu(void *arg) { if (this_tracer) { - bts_disable(); - ds_release_bts(this_tracer); this_tracer = NULL; } @@ -140,7 +78,6 @@ static void bts_trace_stop(struct trace_array *tr) static int bts_trace_init(struct trace_array *tr) { - bts_trace_cpuinit(&boot_cpu_data); bts_trace_reset(tr); bts_trace_start(tr); @@ -149,47 +86,37 @@ static int bts_trace_init(struct trace_array *tr) static void bts_trace_print_header(struct seq_file *m) { -#ifdef __i386__ - seq_puts(m, "# CPU# FROM TO FUNCTION\n"); - seq_puts(m, "# | | | |\n"); -#else seq_puts(m, "# CPU# FROM TO FUNCTION\n"); seq_puts(m, "# | | | |\n"); -#endif } static enum print_line_t bts_trace_print_line(struct trace_iterator *iter) { struct trace_entry *entry = iter->ent; struct trace_seq *seq = &iter->seq; - struct bts_entry *it; + struct hw_branch_entry *it; trace_assign_type(it, entry); - if (entry->type == TRACE_BTS) { - int ret; -#ifdef CONFIG_KALLSYMS - char function[KSYM_SYMBOL_LEN]; - sprint_symbol(function, it->from); -#else - char *function = ""; -#endif - - ret = trace_seq_printf(seq, "%4d 0x%lx -> 0x%lx [%s]\n", - entry->cpu, it->from, it->to, function); - if (!ret) - return TRACE_TYPE_PARTIAL_LINE;; - return TRACE_TYPE_HANDLED; + if (entry->type == TRACE_HW_BRANCHES) { + if (trace_seq_printf(seq, "%4d ", entry->cpu) && + trace_seq_printf(seq, "0x%016llx -> 0x%016llx ", + it->from, it->to) && + (!it->from || + seq_print_ip_sym(seq, it->from, /* sym_flags = */ 0)) && + trace_seq_printf(seq, "\n")) + return TRACE_TYPE_HANDLED; + return TRACE_TYPE_PARTIAL_LINE;; } return TRACE_TYPE_UNHANDLED; } -void trace_bts(struct trace_array *tr, unsigned long from, unsigned long to) +void trace_hw_branch(struct trace_array *tr, u64 from, u64 to) { struct ring_buffer_event *event; - struct bts_entry *entry; + struct hw_branch_entry *entry; unsigned long irq; event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), &irq); @@ -197,56 +124,58 @@ void trace_bts(struct trace_array *tr, unsigned long from, unsigned long to) return; entry = ring_buffer_event_data(event); tracing_generic_entry_update(&entry->ent, 0, from); - entry->ent.type = TRACE_BTS; + entry->ent.type = TRACE_HW_BRANCHES; entry->ent.cpu = smp_processor_id(); entry->from = from; entry->to = to; ring_buffer_unlock_commit(tr->buffer, event, irq); } -static void trace_bts_at(struct trace_array *tr, size_t index) +static void trace_bts_at(struct trace_array *tr, + const struct bts_trace *trace, void *at) { - const void *raw = NULL; - unsigned long from, to; - int err; + struct bts_struct bts; + int err = 0; - err = ds_access_bts(this_tracer, index, &raw); + WARN_ON_ONCE(!trace->read); + if (!trace->read) + return; + + err = trace->read(this_tracer, at, &bts); if (err < 0) return; - from = *(const unsigned long *)raw; - to = *(const unsigned long *)((const char *)raw + sizeof_field); - - trace_bts(tr, from, to); + switch (bts.qualifier) { + case BTS_BRANCH: + trace_hw_branch(tr, bts.variant.lbr.from, bts.variant.lbr.to); + break; + } } static void trace_bts_cpu(void *arg) { struct trace_array *tr = (struct trace_array *) arg; - size_t index = 0, end = 0, i; - int err; + const struct bts_trace *trace; + unsigned char *at; if (!this_tracer) return; - bts_disable(); - - err = ds_get_bts_index(this_tracer, &index); - if (err < 0) + ds_suspend_bts(this_tracer); + trace = ds_read_bts(this_tracer); + if (!trace) goto out; - err = ds_get_bts_end(this_tracer, &end); - if (err < 0) - goto out; + for (at = trace->ds.top; (void *)at < trace->ds.end; + at += trace->ds.size) + trace_bts_at(tr, trace, at); - for (i = index; i < end; i++) - trace_bts_at(tr, i); - - for (i = 0; i < index; i++) - trace_bts_at(tr, i); + for (at = trace->ds.begin; (void *)at < trace->ds.top; + at += trace->ds.size) + trace_bts_at(tr, trace, at); out: - bts_enable(); + ds_resume_bts(this_tracer); } static void trace_bts_prepare(struct trace_iterator *iter) @@ -259,7 +188,7 @@ static void trace_bts_prepare(struct trace_iterator *iter) struct tracer bts_tracer __read_mostly = { - .name = "bts", + .name = "hw-branch-tracer", .init = bts_trace_init, .reset = bts_trace_stop, .print_header = bts_trace_print_header, From ffc2238af8431d930d2c15f16feecf1fd6d75642 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 12 Dec 2008 08:21:19 +0100 Subject: [PATCH 4/9] x86, bts: fix build error Impact: build fix arch/x86/kernel/ds.c: In function 'ds_request': arch/x86/kernel/ds.c:236: sorry, unimplemented: inlining failed in call to 'ds_get_context': recursive inlining but the recursion here is scary ... Signed-off-by: Ingo Molnar --- arch/x86/kernel/ds.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index f0583005b75e..dc1e7123ea4e 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -232,7 +232,7 @@ static DEFINE_PER_CPU(struct ds_context *, system_context_array); #define system_context per_cpu(system_context_array, smp_processor_id()) -static inline struct ds_context *ds_get_context(struct task_struct *task) +static struct ds_context *ds_get_context(struct task_struct *task) { struct ds_context **p_context = (task ? &task->thread.ds_ctx : &system_context); From a0343e823184070f55364d8359f832dcb33c57c7 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 9 Dec 2008 23:53:16 +0100 Subject: [PATCH 5/9] tracing/function-graph-tracer: add a new .irqentry.text section Impact: let the function-graph-tracer be aware of the irq entrypoints Add a new .irqentry.text section to store the irq entrypoints functions inside the same section. This way, the tracer will be able to signal an interrupts triggering on output by recognizing these entrypoints. Also, make this section recordable for dynamic tracing. Signed-off-by: Frederic Weisbecker Signed-off-by: Ingo Molnar --- arch/x86/kernel/vmlinux_64.lds.S | 1 + include/asm-generic/vmlinux.lds.h | 10 ++++++++++ scripts/recordmcount.pl | 1 + 3 files changed, 12 insertions(+) diff --git a/arch/x86/kernel/vmlinux_64.lds.S b/arch/x86/kernel/vmlinux_64.lds.S index 46e05447405b..1a614c0e6bef 100644 --- a/arch/x86/kernel/vmlinux_64.lds.S +++ b/arch/x86/kernel/vmlinux_64.lds.S @@ -35,6 +35,7 @@ SECTIONS SCHED_TEXT LOCK_TEXT KPROBES_TEXT + IRQENTRY_TEXT *(.fixup) *(.gnu.warning) _etext = .; /* End of text section */ diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h index eba835a2c2cd..c61fab1dd2f8 100644 --- a/include/asm-generic/vmlinux.lds.h +++ b/include/asm-generic/vmlinux.lds.h @@ -288,6 +288,16 @@ *(.kprobes.text) \ VMLINUX_SYMBOL(__kprobes_text_end) = .; +#ifdef CONFIG_FUNCTION_GRAPH_TRACER +#define IRQENTRY_TEXT \ + ALIGN_FUNCTION(); \ + VMLINUX_SYMBOL(__irqentry_text_start) = .; \ + *(.irqentry.text) \ + VMLINUX_SYMBOL(__irqentry_text_end) = .; +#else +#define IRQENTRY_TEXT +#endif + /* Section used for early init (in .S files) */ #define HEAD_TEXT *(.head.text) diff --git a/scripts/recordmcount.pl b/scripts/recordmcount.pl index 0b1dc9f9bb06..fe831412bea9 100755 --- a/scripts/recordmcount.pl +++ b/scripts/recordmcount.pl @@ -114,6 +114,7 @@ my %text_sections = ( ".text" => 1, ".sched.text" => 1, ".spinlock.text" => 1, + ".irqentry.text" => 1, ); $objdump = "objdump" if ((length $objdump) == 0); From bcbc4f20b52c2c40c43a4d2337707dcdfe81bc3a Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 9 Dec 2008 23:54:20 +0100 Subject: [PATCH 6/9] tracing/function-graph-tracer: annotate do_IRQ and smp_apic_timer_interrupt Impact: move most important x86 irq entry-points to a separate subsection Annotate do_IRQ and smp_apic_timer_interrupt to put them into the .irqentry.text subsection. These function will so be recognized as hardirq entrypoints for the function-graph-tracer. We could also annotate other irq entries but the others are far less important but they can be added on request. Signed-off-by: Frederic Weisbecker Signed-off-by: Ingo Molnar --- arch/x86/kernel/apic.c | 3 ++- arch/x86/kernel/irq_64.c | 3 ++- include/linux/ftrace.h | 11 +++++++++++ 3 files changed, 15 insertions(+), 2 deletions(-) diff --git a/arch/x86/kernel/apic.c b/arch/x86/kernel/apic.c index 16f94879b525..b946ac19753b 100644 --- a/arch/x86/kernel/apic.c +++ b/arch/x86/kernel/apic.c @@ -30,6 +30,7 @@ #include #include #include +#include #include #include @@ -800,7 +801,7 @@ static void local_apic_timer_interrupt(void) * [ if a single-CPU system runs an SMP kernel then we call the local * interrupt as well. Thus we cannot inline the local irq ... ] */ -void smp_apic_timer_interrupt(struct pt_regs *regs) +void __irq_entry smp_apic_timer_interrupt(struct pt_regs *regs) { struct pt_regs *old_regs = set_irq_regs(regs); diff --git a/arch/x86/kernel/irq_64.c b/arch/x86/kernel/irq_64.c index 60eb84eb77a0..11c65e811ffe 100644 --- a/arch/x86/kernel/irq_64.c +++ b/arch/x86/kernel/irq_64.c @@ -13,6 +13,7 @@ #include #include #include +#include #include #include #include @@ -47,7 +48,7 @@ static inline void stack_overflow_check(struct pt_regs *regs) * SMP cross-CPU interrupts have their own specific * handlers). */ -asmlinkage unsigned int do_IRQ(struct pt_regs *regs) +asmlinkage unsigned int __irq_entry do_IRQ(struct pt_regs *regs) { struct pt_regs *old_regs = set_irq_regs(regs); struct irq_desc *desc; diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h index 11cac81eed08..44020f31bd81 100644 --- a/include/linux/ftrace.h +++ b/include/linux/ftrace.h @@ -377,6 +377,16 @@ struct ftrace_graph_ret { */ #define __notrace_funcgraph notrace +/* + * We want to which function is an entrypoint of a hardirq. + * That will help us to put a signal on output. + */ +#define __irq_entry __attribute__((__section__(".irqentry.text"))) + +/* Limits of hardirq entrypoints */ +extern char __irqentry_text_start[]; +extern char __irqentry_text_end[]; + #define FTRACE_RETFUNC_DEPTH 50 #define FTRACE_RETSTACK_ALLOC_SIZE 32 /* Type of the callback handlers for tracing function graph*/ @@ -414,6 +424,7 @@ static inline void unpause_graph_tracing(void) #else #define __notrace_funcgraph +#define __irq_entry static inline void ftrace_graph_init_task(struct task_struct *t) { } static inline void ftrace_graph_exit_task(struct task_struct *t) { } From f8b755ac8e0cc3f330269e4c4504514f987167a2 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Tue, 9 Dec 2008 23:55:25 +0100 Subject: [PATCH 7/9] tracing/function-graph-tracer: Output arrows signal on hardirq call/return Impact: make more obvious the hardirq calls in the output When a hardirq is triggered inside the codeflow on output, we have now two arrows that indicate the entry and return of the hardirq. 0) | bit_waitqueue() { 0) 0.880 us | __phys_addr(); 0) 2.699 us | } 0) | __wake_up_bit() { 0) ==========> | smp_apic_timer_interrupt() { 0) 0.797 us | native_apic_mem_write(); 0) 0.715 us | exit_idle(); 0) | irq_enter() { 0) 0.722 us | idle_cpu(); 0) 5.519 us | } 0) | hrtimer_interrupt() { 0) | ktime_get() { 0) | ktime_get_ts() { 0) 0.805 us | getnstimeofday(); [...] 0) ! 108.528 us | } 0) | irq_exit() { 0) | do_softirq() { 0) | __do_softirq() { 0) 0.895 us | __local_bh_disable(); 0) | run_timer_softirq() { 0) 0.827 us | hrtimer_run_pending(); 0) 1.226 us | _spin_lock_irq(); 0) | _spin_unlock_irq() { 0) 6.550 us | } 0) 0.924 us | _local_bh_enable(); 0) + 12.129 us | } 0) + 13.911 us | } 0) 0.707 us | idle_cpu(); 0) + 17.009 us | } 0) ! 137.419 us | } 0) <========== | 0) 1.045 us | } 0) ! 148.908 us | } 0) ! 151.022 us | } 0) ! 153.022 us | } 0) 0.963 us | journal_mark_dirty(); 0) 0.925 us | __brelse(); Signed-off-by: Frederic Weisbecker Signed-off-by: Ingo Molnar --- kernel/trace/trace_functions_graph.c | 66 ++++++++++++++++++++++++++-- 1 file changed, 62 insertions(+), 4 deletions(-) diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c index af60eef4cbcc..4bf39fcae97a 100644 --- a/kernel/trace/trace_functions_graph.c +++ b/kernel/trace/trace_functions_graph.c @@ -231,6 +231,49 @@ trace_branch_is_leaf(struct trace_iterator *iter, return true; } +static enum print_line_t +print_graph_irq(struct trace_seq *s, unsigned long addr, + enum trace_type type, int cpu, pid_t pid) +{ + int ret; + + if (addr < (unsigned long)__irqentry_text_start || + addr >= (unsigned long)__irqentry_text_end) + return TRACE_TYPE_UNHANDLED; + + if (type == TRACE_GRAPH_ENT) { + ret = trace_seq_printf(s, "==========> | "); + } else { + /* Cpu */ + if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) { + ret = print_graph_cpu(s, cpu); + if (ret == TRACE_TYPE_PARTIAL_LINE) + return TRACE_TYPE_PARTIAL_LINE; + } + /* Proc */ + if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) { + ret = print_graph_proc(s, pid); + if (ret == TRACE_TYPE_PARTIAL_LINE) + return TRACE_TYPE_PARTIAL_LINE; + + ret = trace_seq_printf(s, " | "); + if (!ret) + return TRACE_TYPE_PARTIAL_LINE; + } + + /* No overhead */ + if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { + ret = trace_seq_printf(s, " "); + if (!ret) + return TRACE_TYPE_PARTIAL_LINE; + } + + ret = trace_seq_printf(s, "<========== |\n"); + } + if (!ret) + return TRACE_TYPE_PARTIAL_LINE; + return TRACE_TYPE_HANDLED; +} static enum print_line_t print_graph_duration(unsigned long long duration, struct trace_seq *s) @@ -344,7 +387,7 @@ print_graph_entry_leaf(struct trace_iterator *iter, static enum print_line_t print_graph_entry_nested(struct ftrace_graph_ent_entry *entry, - struct trace_seq *s) + struct trace_seq *s, pid_t pid, int cpu) { int i; int ret; @@ -357,8 +400,18 @@ print_graph_entry_nested(struct ftrace_graph_ent_entry *entry, return TRACE_TYPE_PARTIAL_LINE; } - /* No time */ - ret = trace_seq_printf(s, " | "); + /* Interrupt */ + ret = print_graph_irq(s, call->func, TRACE_GRAPH_ENT, cpu, pid); + if (ret == TRACE_TYPE_UNHANDLED) { + /* No time */ + ret = trace_seq_printf(s, " | "); + if (!ret) + return TRACE_TYPE_PARTIAL_LINE; + } else { + if (ret == TRACE_TYPE_PARTIAL_LINE) + return TRACE_TYPE_PARTIAL_LINE; + } + /* Function */ for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) { @@ -410,7 +463,7 @@ print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s, if (trace_branch_is_leaf(iter, field)) return print_graph_entry_leaf(iter, field, s); else - return print_graph_entry_nested(field, s); + return print_graph_entry_nested(field, s, iter->ent->pid, cpu); } @@ -474,6 +527,11 @@ print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s, if (!ret) return TRACE_TYPE_PARTIAL_LINE; } + + ret = print_graph_irq(s, trace->func, TRACE_GRAPH_RET, cpu, ent->pid); + if (ret == TRACE_TYPE_PARTIAL_LINE) + return TRACE_TYPE_PARTIAL_LINE; + return TRACE_TYPE_HANDLED; } From cbc34ed1ac36690f75fd272e19e7b4fc29aae5a2 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 10 Dec 2008 08:08:22 +0100 Subject: [PATCH 8/9] sched: fix tracepoints in scheduler The trace point only caught one of many places where a task changes cpu, put it in the right place to we get all of them. Change the signature while we're at it. Signed-off-by: Peter Zijlstra Signed-off-by: Ingo Molnar --- include/trace/sched.h | 4 ++-- kernel/sched.c | 3 ++- 2 files changed, 4 insertions(+), 3 deletions(-) diff --git a/include/trace/sched.h b/include/trace/sched.h index 9b2854abf7e2..f4549d506b16 100644 --- a/include/trace/sched.h +++ b/include/trace/sched.h @@ -30,8 +30,8 @@ DECLARE_TRACE(sched_switch, TPARGS(rq, prev, next)); DECLARE_TRACE(sched_migrate_task, - TPPROTO(struct rq *rq, struct task_struct *p, int dest_cpu), - TPARGS(rq, p, dest_cpu)); + TPPROTO(struct task_struct *p, int orig_cpu, int dest_cpu), + TPARGS(p, orig_cpu, dest_cpu)); DECLARE_TRACE(sched_process_free, TPPROTO(struct task_struct *p), diff --git a/kernel/sched.c b/kernel/sched.c index 7729c4bbc8ba..d377097572f9 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -1851,6 +1851,8 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) clock_offset = old_rq->clock - new_rq->clock; + trace_sched_migrate_task(p, task_cpu(p), new_cpu); + #ifdef CONFIG_SCHEDSTATS if (p->se.wait_start) p->se.wait_start -= clock_offset; @@ -2868,7 +2870,6 @@ static void sched_migrate_task(struct task_struct *p, int dest_cpu) || unlikely(!cpu_active(dest_cpu))) goto out; - trace_sched_migrate_task(rq, p, dest_cpu); /* force the process onto the specified CPU */ if (migrate_task(p, dest_cpu, &req)) { /* Need to wait for migration thread (might exit: take ref). */ From 2bed8446819a7c5033aa1da138d9f230ae212edc Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Fri, 12 Dec 2008 12:13:36 +0100 Subject: [PATCH 9/9] tracing/function-graph-tracer: add a new .irqentry.text section, fix Impact: build fix 32-bit x86 needs this section too. Signed-off-by: Ingo Molnar --- arch/x86/kernel/vmlinux_32.lds.S | 1 + 1 file changed, 1 insertion(+) diff --git a/arch/x86/kernel/vmlinux_32.lds.S b/arch/x86/kernel/vmlinux_32.lds.S index a9b8560adbc2..82c67559dde7 100644 --- a/arch/x86/kernel/vmlinux_32.lds.S +++ b/arch/x86/kernel/vmlinux_32.lds.S @@ -44,6 +44,7 @@ SECTIONS SCHED_TEXT LOCK_TEXT KPROBES_TEXT + IRQENTRY_TEXT *(.fixup) *(.gnu.warning) _etext = .; /* End of text section */