From 3f6da3905398826d85731247e7fbcf53400c18bd Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 5 Mar 2010 13:01:18 +0100 Subject: perf: Rework and fix the arch CPU-hotplug hooks Remove the hw_perf_event_*() hotplug hooks in favour of per PMU hotplug notifiers. This has the advantage of reducing the static weak interface as well as exposing all hotplug actions to the PMU. Use this to fix x86 hotplug usage where we did things in ONLINE which should have been done in UP_PREPARE or STARTING. Signed-off-by: Peter Zijlstra Cc: Paul Mundt Cc: paulus@samba.org Cc: eranian@google.com Cc: robert.richter@amd.com Cc: fweisbec@gmail.com Cc: Arnaldo Carvalho de Melo LKML-Reference: <20100305154128.736225361@chello.nl> Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 6f8cd7da1a01..80acbf3d5de1 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -936,5 +936,21 @@ static inline void perf_event_disable(struct perf_event *event) { } #define perf_output_put(handle, x) \ perf_output_copy((handle), &(x), sizeof(x)) +/* + * This has to have a higher priority than migration_notifier in sched.c. + */ +#define perf_cpu_notifier(fn) \ +do { \ + static struct notifier_block fn##_nb __cpuinitdata = \ + { .notifier_call = fn, .priority = 20 }; \ + fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ + (void *)(unsigned long)smp_processor_id()); \ + fn(&fn##_nb, (unsigned long)CPU_STARTING, \ + (void *)(unsigned long)smp_processor_id()); \ + fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ + (void *)(unsigned long)smp_processor_id()); \ + register_cpu_notifier(&fn##_nb); \ +} while (0) + #endif /* __KERNEL__ */ #endif /* _LINUX_PERF_EVENT_H */ -- cgit v1.2.2 From 5331d7b84613b8325362dde53dc2bff2fb87d351 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Thu, 4 Mar 2010 21:15:56 +0100 Subject: perf: Introduce new perf_fetch_caller_regs() for hot regs snapshot Events that trigger overflows by interrupting a context can use get_irq_regs() or task_pt_regs() to retrieve the state when the event triggered. But this is not the case for some other class of events like trace events as tracepoints are executed in the same context than the code that triggered the event. It means we need a different api to capture the regs there, namely we need a hot snapshot to get the most important informations for perf: the instruction pointer to get the event origin, the frame pointer for the callchain, the code segment for user_mode() tests (we always use __KERNEL_CS as trace events always occur from the kernel) and the eflags for further purposes. v2: rename perf_save_regs to perf_fetch_caller_regs as per Masami's suggestion. Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Thomas Gleixner Cc: H. Peter Anvin Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Cc: Arnaldo Carvalho de Melo Cc: Masami Hiramatsu Cc: Jason Baron Cc: Archs --- include/linux/perf_event.h | 42 +++++++++++++++++++++++++++++++++++++++++- 1 file changed, 41 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 80acbf3d5de1..70cffd052c04 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -452,6 +452,7 @@ enum perf_callchain_context { #include #include #include +#include #include #define PERF_MAX_STACK_DEPTH 255 @@ -847,6 +848,44 @@ perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) __perf_sw_event(event_id, nr, nmi, regs, addr); } +extern void +perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip); + +/* + * Take a snapshot of the regs. Skip ip and frame pointer to + * the nth caller. We only need a few of the regs: + * - ip for PERF_SAMPLE_IP + * - cs for user_mode() tests + * - bp for callchains + * - eflags, for future purposes, just in case + */ +static inline void perf_fetch_caller_regs(struct pt_regs *regs, int skip) +{ + unsigned long ip; + + memset(regs, 0, sizeof(*regs)); + + switch (skip) { + case 1 : + ip = CALLER_ADDR0; + break; + case 2 : + ip = CALLER_ADDR1; + break; + case 3 : + ip = CALLER_ADDR2; + break; + case 4: + ip = CALLER_ADDR3; + break; + /* No need to support further for now */ + default: + ip = 0; + } + + return perf_arch_fetch_caller_regs(regs, ip, skip); +} + extern void __perf_event_mmap(struct vm_area_struct *vma); static inline void perf_event_mmap(struct vm_area_struct *vma) @@ -880,7 +919,8 @@ static inline bool perf_paranoid_kernel(void) } extern void perf_event_init(void); -extern void perf_tp_event(int event_id, u64 addr, u64 count, void *record, int entry_size); +extern void perf_tp_event(int event_id, u64 addr, u64 count, void *record, + int entry_size, struct pt_regs *regs); extern void perf_bp_event(struct perf_event *event, void *data); #ifndef perf_misc_flags -- cgit v1.2.2 From c530665c31c0140b74ca7689e7f836177796e5bd Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Wed, 3 Mar 2010 07:16:16 +0100 Subject: perf: Take a hot regs snapshot for trace events We are taking a wrong regs snapshot when a trace event triggers. Either we use get_irq_regs(), which gives us the interrupted registers if we are in an interrupt, or we use task_pt_regs() which gives us the state before we entered the kernel, assuming we are lucky enough to be no kernel thread, in which case task_pt_regs() returns the initial set of regs when the kernel thread was started. What we want is different. We need a hot snapshot of the regs, so that we can get the instruction pointer to record in the sample, the frame pointer for the callchain, and some other things. Let's use the new perf_fetch_caller_regs() for that. Comparison with perf record -e lock: -R -a -f -g Before: perf [kernel] [k] __do_softirq | --- __do_softirq | |--55.16%-- __open | --44.84%-- __write_nocancel After: perf [kernel] [k] perf_tp_event | --- perf_tp_event | |--41.07%-- lock_acquire | | | |--39.36%-- _raw_spin_lock | | | | | |--7.81%-- hrtimer_interrupt | | | smp_apic_timer_interrupt | | | apic_timer_interrupt The old case was producing unreliable callchains. Now having right frame and instruction pointers, we have the trace we want. Also syscalls and kprobe events already have the right regs, let's use them instead of wasting a retrieval. v2: Follow the rename perf_save_regs() -> perf_fetch_caller_regs() Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Thomas Gleixner Cc: H. Peter Anvin Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Cc: Arnaldo Carvalho de Melo Cc: Masami Hiramatsu Cc: Jason Baron Cc: Archs --- include/linux/ftrace_event.h | 7 +++++-- include/trace/ftrace.h | 6 +++++- 2 files changed, 10 insertions(+), 3 deletions(-) (limited to 'include') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index 6b7c444ab8f6..ac424f18ce63 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -187,6 +187,9 @@ do { \ #ifdef CONFIG_PERF_EVENTS struct perf_event; + +DECLARE_PER_CPU(struct pt_regs, perf_trace_regs); + extern int ftrace_profile_enable(int event_id); extern void ftrace_profile_disable(int event_id); extern int ftrace_profile_set_filter(struct perf_event *event, int event_id, @@ -198,11 +201,11 @@ ftrace_perf_buf_prepare(int size, unsigned short type, int *rctxp, static inline void ftrace_perf_buf_submit(void *raw_data, int size, int rctx, u64 addr, - u64 count, unsigned long irq_flags) + u64 count, unsigned long irq_flags, struct pt_regs *regs) { struct trace_entry *entry = raw_data; - perf_tp_event(entry->type, addr, count, raw_data, size); + perf_tp_event(entry->type, addr, count, raw_data, size, regs); perf_swevent_put_recursion_context(rctx); local_irq_restore(irq_flags); } diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 0804cd594803..f31bb8b9777c 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -764,6 +764,7 @@ ftrace_profile_templ_##call(struct ftrace_event_call *event_call, \ struct ftrace_raw_##call *entry; \ u64 __addr = 0, __count = 1; \ unsigned long irq_flags; \ + struct pt_regs *__regs; \ int __entry_size; \ int __data_size; \ int rctx; \ @@ -784,8 +785,11 @@ ftrace_profile_templ_##call(struct ftrace_event_call *event_call, \ \ { assign; } \ \ + __regs = &__get_cpu_var(perf_trace_regs); \ + perf_fetch_caller_regs(__regs, 2); \ + \ ftrace_perf_buf_submit(entry, __entry_size, rctx, __addr, \ - __count, irq_flags); \ + __count, irq_flags, __regs); \ } #undef DEFINE_EVENT -- cgit v1.2.2 From 97d5a22005f38057b4bc0d95f81cd26510268794 Mon Sep 17 00:00:00 2001 From: Frederic Weisbecker Date: Fri, 5 Mar 2010 05:35:37 +0100 Subject: perf: Drop the obsolete profile naming for trace events Drop the obsolete "profile" naming used by perf for trace events. Perf can now do more than simple events counting, so generalize the API naming. Signed-off-by: Frederic Weisbecker Cc: Ingo Molnar Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Steven Rostedt Cc: Masami Hiramatsu Cc: Jason Baron --- include/linux/ftrace_event.h | 16 ++++++++-------- include/linux/syscalls.h | 24 ++++++++++++------------ include/trace/ftrace.h | 38 +++++++++++++++++++------------------- include/trace/syscall.h | 8 ++++---- 4 files changed, 43 insertions(+), 43 deletions(-) (limited to 'include') diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h index ac424f18ce63..c0f4b364c711 100644 --- a/include/linux/ftrace_event.h +++ b/include/linux/ftrace_event.h @@ -131,12 +131,12 @@ struct ftrace_event_call { void *mod; void *data; - int profile_count; - int (*profile_enable)(struct ftrace_event_call *); - void (*profile_disable)(struct ftrace_event_call *); + int perf_refcount; + int (*perf_event_enable)(struct ftrace_event_call *); + void (*perf_event_disable)(struct ftrace_event_call *); }; -#define FTRACE_MAX_PROFILE_SIZE 2048 +#define PERF_MAX_TRACE_SIZE 2048 #define MAX_FILTER_PRED 32 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */ @@ -190,17 +190,17 @@ struct perf_event; DECLARE_PER_CPU(struct pt_regs, perf_trace_regs); -extern int ftrace_profile_enable(int event_id); -extern void ftrace_profile_disable(int event_id); +extern int perf_trace_enable(int event_id); +extern void perf_trace_disable(int event_id); extern int ftrace_profile_set_filter(struct perf_event *event, int event_id, char *filter_str); extern void ftrace_profile_free_filter(struct perf_event *event); extern void * -ftrace_perf_buf_prepare(int size, unsigned short type, int *rctxp, +perf_trace_buf_prepare(int size, unsigned short type, int *rctxp, unsigned long *irq_flags); static inline void -ftrace_perf_buf_submit(void *raw_data, int size, int rctx, u64 addr, +perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr, u64 count, unsigned long irq_flags, struct pt_regs *regs) { struct trace_entry *entry = raw_data; diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h index 8126f239edf0..51435bcc3460 100644 --- a/include/linux/syscalls.h +++ b/include/linux/syscalls.h @@ -101,18 +101,18 @@ struct perf_event_attr; #ifdef CONFIG_PERF_EVENTS -#define TRACE_SYS_ENTER_PROFILE_INIT(sname) \ - .profile_enable = prof_sysenter_enable, \ - .profile_disable = prof_sysenter_disable, +#define TRACE_SYS_ENTER_PERF_INIT(sname) \ + .perf_event_enable = perf_sysenter_enable, \ + .perf_event_disable = perf_sysenter_disable, -#define TRACE_SYS_EXIT_PROFILE_INIT(sname) \ - .profile_enable = prof_sysexit_enable, \ - .profile_disable = prof_sysexit_disable, +#define TRACE_SYS_EXIT_PERF_INIT(sname) \ + .perf_event_enable = perf_sysexit_enable, \ + .perf_event_disable = perf_sysexit_disable, #else -#define TRACE_SYS_ENTER_PROFILE(sname) -#define TRACE_SYS_ENTER_PROFILE_INIT(sname) -#define TRACE_SYS_EXIT_PROFILE(sname) -#define TRACE_SYS_EXIT_PROFILE_INIT(sname) +#define TRACE_SYS_ENTER_PERF(sname) +#define TRACE_SYS_ENTER_PERF_INIT(sname) +#define TRACE_SYS_EXIT_PERF(sname) +#define TRACE_SYS_EXIT_PERF_INIT(sname) #endif /* CONFIG_PERF_EVENTS */ #ifdef CONFIG_FTRACE_SYSCALLS @@ -149,7 +149,7 @@ struct perf_event_attr; .regfunc = reg_event_syscall_enter, \ .unregfunc = unreg_event_syscall_enter, \ .data = (void *)&__syscall_meta_##sname,\ - TRACE_SYS_ENTER_PROFILE_INIT(sname) \ + TRACE_SYS_ENTER_PERF_INIT(sname) \ } #define SYSCALL_TRACE_EXIT_EVENT(sname) \ @@ -171,7 +171,7 @@ struct perf_event_attr; .regfunc = reg_event_syscall_exit, \ .unregfunc = unreg_event_syscall_exit, \ .data = (void *)&__syscall_meta_##sname,\ - TRACE_SYS_EXIT_PROFILE_INIT(sname) \ + TRACE_SYS_EXIT_PERF_INIT(sname) \ } #define SYSCALL_METADATA(sname, nb) \ diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index f31bb8b9777c..25ab56f75d65 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -401,18 +401,18 @@ static inline notrace int ftrace_get_offsets_##call( \ #undef DEFINE_EVENT #define DEFINE_EVENT(template, name, proto, args) \ \ -static void ftrace_profile_##name(proto); \ +static void perf_trace_##name(proto); \ \ static notrace int \ -ftrace_profile_enable_##name(struct ftrace_event_call *unused) \ +perf_trace_enable_##name(struct ftrace_event_call *unused) \ { \ - return register_trace_##name(ftrace_profile_##name); \ + return register_trace_##name(perf_trace_##name); \ } \ \ static notrace void \ -ftrace_profile_disable_##name(struct ftrace_event_call *unused) \ +perf_trace_disable_##name(struct ftrace_event_call *unused) \ { \ - unregister_trace_##name(ftrace_profile_##name); \ + unregister_trace_##name(perf_trace_##name); \ } #undef DEFINE_EVENT_PRINT @@ -507,12 +507,12 @@ ftrace_profile_disable_##name(struct ftrace_event_call *unused) \ #ifdef CONFIG_PERF_EVENTS -#define _TRACE_PROFILE_INIT(call) \ - .profile_enable = ftrace_profile_enable_##call, \ - .profile_disable = ftrace_profile_disable_##call, +#define _TRACE_PERF_INIT(call) \ + .perf_event_enable = perf_trace_enable_##call, \ + .perf_event_disable = perf_trace_disable_##call, #else -#define _TRACE_PROFILE_INIT(call) +#define _TRACE_PERF_INIT(call) #endif /* CONFIG_PERF_EVENTS */ #undef __entry @@ -638,7 +638,7 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ .unregfunc = ftrace_raw_unreg_event_##call, \ .print_fmt = print_fmt_##template, \ .define_fields = ftrace_define_fields_##template, \ - _TRACE_PROFILE_INIT(call) \ + _TRACE_PERF_INIT(call) \ } #undef DEFINE_EVENT_PRINT @@ -657,18 +657,18 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ .unregfunc = ftrace_raw_unreg_event_##call, \ .print_fmt = print_fmt_##call, \ .define_fields = ftrace_define_fields_##template, \ - _TRACE_PROFILE_INIT(call) \ + _TRACE_PERF_INIT(call) \ } #include TRACE_INCLUDE(TRACE_INCLUDE_FILE) /* - * Define the insertion callback to profile events + * Define the insertion callback to perf events * * The job is very similar to ftrace_raw_event_ except that we don't * insert in the ring buffer but in a perf counter. * - * static void ftrace_profile_(proto) + * static void ftrace_perf_(proto) * { * struct ftrace_data_offsets_ __maybe_unused __data_offsets; * struct ftrace_event_call *event_call = &event_; @@ -757,7 +757,7 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ #undef DECLARE_EVENT_CLASS #define DECLARE_EVENT_CLASS(call, proto, args, tstruct, assign, print) \ static notrace void \ -ftrace_profile_templ_##call(struct ftrace_event_call *event_call, \ +perf_trace_templ_##call(struct ftrace_event_call *event_call, \ proto) \ { \ struct ftrace_data_offsets_##call __maybe_unused __data_offsets;\ @@ -774,10 +774,10 @@ ftrace_profile_templ_##call(struct ftrace_event_call *event_call, \ sizeof(u64)); \ __entry_size -= sizeof(u32); \ \ - if (WARN_ONCE(__entry_size > FTRACE_MAX_PROFILE_SIZE, \ + if (WARN_ONCE(__entry_size > PERF_MAX_TRACE_SIZE, \ "profile buffer not large enough")) \ return; \ - entry = (struct ftrace_raw_##call *)ftrace_perf_buf_prepare( \ + entry = (struct ftrace_raw_##call *)perf_trace_buf_prepare( \ __entry_size, event_call->id, &rctx, &irq_flags); \ if (!entry) \ return; \ @@ -788,17 +788,17 @@ ftrace_profile_templ_##call(struct ftrace_event_call *event_call, \ __regs = &__get_cpu_var(perf_trace_regs); \ perf_fetch_caller_regs(__regs, 2); \ \ - ftrace_perf_buf_submit(entry, __entry_size, rctx, __addr, \ + perf_trace_buf_submit(entry, __entry_size, rctx, __addr, \ __count, irq_flags, __regs); \ } #undef DEFINE_EVENT #define DEFINE_EVENT(template, call, proto, args) \ -static notrace void ftrace_profile_##call(proto) \ +static notrace void perf_trace_##call(proto) \ { \ struct ftrace_event_call *event_call = &event_##call; \ \ - ftrace_profile_templ_##template(event_call, args); \ + perf_trace_templ_##template(event_call, args); \ } #undef DEFINE_EVENT_PRINT diff --git a/include/trace/syscall.h b/include/trace/syscall.h index 0387100752f0..e5e5f48dbfb3 100644 --- a/include/trace/syscall.h +++ b/include/trace/syscall.h @@ -47,10 +47,10 @@ enum print_line_t print_syscall_exit(struct trace_iterator *iter, int flags); #endif #ifdef CONFIG_PERF_EVENTS -int prof_sysenter_enable(struct ftrace_event_call *call); -void prof_sysenter_disable(struct ftrace_event_call *call); -int prof_sysexit_enable(struct ftrace_event_call *call); -void prof_sysexit_disable(struct ftrace_event_call *call); +int perf_sysenter_enable(struct ftrace_event_call *call); +void perf_sysenter_disable(struct ftrace_event_call *call); +int perf_sysexit_enable(struct ftrace_event_call *call); +void perf_sysexit_disable(struct ftrace_event_call *call); #endif #endif /* _TRACE_SYSCALL_H */ -- cgit v1.2.2 From 85cfabbcd10f8d112feee6e2ec64ee78033b6d3c Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Thu, 11 Mar 2010 13:06:56 +0100 Subject: perf, ppc: Fix compile error due to new cpu notifiers Fix: arch/powerpc/kernel/perf_event.c:1334: error: 'power_pmu_notifier' undeclared (first use in this function) arch/powerpc/kernel/perf_event.c:1334: error: (Each undeclared identifier is reported only once arch/powerpc/kernel/perf_event.c:1334: error: for each function it appears in.) arch/powerpc/kernel/perf_event.c:1334: error: implicit declaration of function 'power_pmu_notifier' arch/powerpc/kernel/perf_event.c:1334: error: implicit declaration of function 'register_cpu_notifier' Due to commit 3f6da390 (perf: Rework and fix the arch CPU-hotplug hooks). Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- include/linux/perf_event.h | 1 + 1 file changed, 1 insertion(+) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index 70cffd052c04..95477038a72a 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -453,6 +453,7 @@ enum perf_callchain_context { #include #include #include +#include #include #define PERF_MAX_STACK_DEPTH 255 -- cgit v1.2.2