aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAndrew Vagin <avagin@openvz.org>2012-07-11 10:14:58 -0400
committerIngo Molnar <mingo@kernel.org>2012-07-31 11:02:05 -0400
commite6dab5ffab59e910ec0e3355f4a6f29f7a7be474 (patch)
tree87acf0fb071b8d09794ac7d834cb256de030cceb
parentd07bdfd322d307789f15b427dbcc39257665356f (diff)
perf/trace: Add ability to set a target task for events
A few events are interesting not only for a current task. For example, sched_stat_* events are interesting for a task which wakes up. For this reason, it will be good if such events will be delivered to a target task too. Now a target task can be set by using __perf_task(). The original idea and a draft patch belongs to Peter Zijlstra. I need these events for profiling sleep times. sched_switch is used for getting callchains and sched_stat_* is used for getting time periods. These events are combined in user space, then it can be analyzed by perf tools. Inspired-by: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Arun Sharma <asharma@fb.com> Signed-off-by: Andrew Vagin <avagin@openvz.org> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1342016098-213063-1-git-send-email-avagin@openvz.org Signed-off-by: Ingo Molnar <mingo@kernel.org>
-rw-r--r--include/linux/ftrace_event.h5
-rw-r--r--include/linux/perf_event.h3
-rw-r--r--include/trace/events/sched.h4
-rw-r--r--include/trace/ftrace.h6
-rw-r--r--kernel/events/callchain.c9
-rw-r--r--kernel/events/core.c30
-rw-r--r--kernel/events/internal.h3
-rw-r--r--kernel/trace/trace_event_perf.c2
-rw-r--r--kernel/trace/trace_kprobe.c6
-rw-r--r--kernel/trace/trace_syscalls.c4
-rw-r--r--kernel/trace/trace_uprobe.c2
11 files changed, 60 insertions, 14 deletions
diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index af961d6f7ab1..642928cf57b4 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -306,9 +306,10 @@ extern void *perf_trace_buf_prepare(int size, unsigned short type,
306 306
307static inline void 307static inline void
308perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr, 308perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
309 u64 count, struct pt_regs *regs, void *head) 309 u64 count, struct pt_regs *regs, void *head,
310 struct task_struct *task)
310{ 311{
311 perf_tp_event(addr, count, raw_data, size, regs, head, rctx); 312 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
312} 313}
313#endif 314#endif
314 315
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 76c5c8b724a7..7602ccb3f40e 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1272,7 +1272,8 @@ static inline bool perf_paranoid_kernel(void)
1272extern void perf_event_init(void); 1272extern void perf_event_init(void);
1273extern void perf_tp_event(u64 addr, u64 count, void *record, 1273extern void perf_tp_event(u64 addr, u64 count, void *record,
1274 int entry_size, struct pt_regs *regs, 1274 int entry_size, struct pt_regs *regs,
1275 struct hlist_head *head, int rctx); 1275 struct hlist_head *head, int rctx,
1276 struct task_struct *task);
1276extern void perf_bp_event(struct perf_event *event, void *data); 1277extern void perf_bp_event(struct perf_event *event, void *data);
1277 1278
1278#ifndef perf_misc_flags 1279#ifndef perf_misc_flags
diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
index ea7a2035456d..5a8671e8a67f 100644
--- a/include/trace/events/sched.h
+++ b/include/trace/events/sched.h
@@ -73,6 +73,9 @@ DECLARE_EVENT_CLASS(sched_wakeup_template,
73 __entry->prio = p->prio; 73 __entry->prio = p->prio;
74 __entry->success = success; 74 __entry->success = success;
75 __entry->target_cpu = task_cpu(p); 75 __entry->target_cpu = task_cpu(p);
76 )
77 TP_perf_assign(
78 __perf_task(p);
76 ), 79 ),
77 80
78 TP_printk("comm=%s pid=%d prio=%d success=%d target_cpu=%03d", 81 TP_printk("comm=%s pid=%d prio=%d success=%d target_cpu=%03d",
@@ -325,6 +328,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
325 ) 328 )
326 TP_perf_assign( 329 TP_perf_assign(
327 __perf_count(delay); 330 __perf_count(delay);
331 __perf_task(tsk);
328 ), 332 ),
329 333
330 TP_printk("comm=%s pid=%d delay=%Lu [ns]", 334 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h
index c6bc2faaf261..a763888a36f9 100644
--- a/include/trace/ftrace.h
+++ b/include/trace/ftrace.h
@@ -712,6 +712,9 @@ __attribute__((section("_ftrace_events"))) *__event_##call = &event_##call
712#undef __perf_count 712#undef __perf_count
713#define __perf_count(c) __count = (c) 713#define __perf_count(c) __count = (c)
714 714
715#undef __perf_task
716#define __perf_task(t) __task = (t)
717
715#undef TP_perf_assign 718#undef TP_perf_assign
716#define TP_perf_assign(args...) args 719#define TP_perf_assign(args...) args
717 720
@@ -725,6 +728,7 @@ perf_trace_##call(void *__data, proto) \
725 struct ftrace_raw_##call *entry; \ 728 struct ftrace_raw_##call *entry; \
726 struct pt_regs __regs; \ 729 struct pt_regs __regs; \
727 u64 __addr = 0, __count = 1; \ 730 u64 __addr = 0, __count = 1; \
731 struct task_struct *__task = NULL; \
728 struct hlist_head *head; \ 732 struct hlist_head *head; \
729 int __entry_size; \ 733 int __entry_size; \
730 int __data_size; \ 734 int __data_size; \
@@ -752,7 +756,7 @@ perf_trace_##call(void *__data, proto) \
752 \ 756 \
753 head = this_cpu_ptr(event_call->perf_events); \ 757 head = this_cpu_ptr(event_call->perf_events); \
754 perf_trace_buf_submit(entry, __entry_size, rctx, __addr, \ 758 perf_trace_buf_submit(entry, __entry_size, rctx, __addr, \
755 __count, &__regs, head); \ 759 __count, &__regs, head, __task); \
756} 760}
757 761
758/* 762/*
diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 6581a040f399..98d4597f43d6 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -153,7 +153,8 @@ put_callchain_entry(int rctx)
153 put_recursion_context(__get_cpu_var(callchain_recursion), rctx); 153 put_recursion_context(__get_cpu_var(callchain_recursion), rctx);
154} 154}
155 155
156struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) 156struct perf_callchain_entry *
157perf_callchain(struct perf_event *event, struct pt_regs *regs)
157{ 158{
158 int rctx; 159 int rctx;
159 struct perf_callchain_entry *entry; 160 struct perf_callchain_entry *entry;
@@ -178,6 +179,12 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
178 } 179 }
179 180
180 if (regs) { 181 if (regs) {
182 /*
183 * Disallow cross-task user callchains.
184 */
185 if (event->ctx->task && event->ctx->task != current)
186 goto exit_put;
187
181 perf_callchain_store(entry, PERF_CONTEXT_USER); 188 perf_callchain_store(entry, PERF_CONTEXT_USER);
182 perf_callchain_user(entry, regs); 189 perf_callchain_user(entry, regs);
183 } 190 }
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f1cf0edeb39a..b7935fcec7d9 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4039,7 +4039,7 @@ void perf_prepare_sample(struct perf_event_header *header,
4039 if (sample_type & PERF_SAMPLE_CALLCHAIN) { 4039 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
4040 int size = 1; 4040 int size = 1;
4041 4041
4042 data->callchain = perf_callchain(regs); 4042 data->callchain = perf_callchain(event, regs);
4043 4043
4044 if (data->callchain) 4044 if (data->callchain)
4045 size += data->callchain->nr; 4045 size += data->callchain->nr;
@@ -5209,7 +5209,8 @@ static int perf_tp_event_match(struct perf_event *event,
5209} 5209}
5210 5210
5211void perf_tp_event(u64 addr, u64 count, void *record, int entry_size, 5211void perf_tp_event(u64 addr, u64 count, void *record, int entry_size,
5212 struct pt_regs *regs, struct hlist_head *head, int rctx) 5212 struct pt_regs *regs, struct hlist_head *head, int rctx,
5213 struct task_struct *task)
5213{ 5214{
5214 struct perf_sample_data data; 5215 struct perf_sample_data data;
5215 struct perf_event *event; 5216 struct perf_event *event;
@@ -5228,6 +5229,31 @@ void perf_tp_event(u64 addr, u64 count, void *record, int entry_size,
5228 perf_swevent_event(event, count, &data, regs); 5229 perf_swevent_event(event, count, &data, regs);
5229 } 5230 }
5230 5231
5232 /*
5233 * If we got specified a target task, also iterate its context and
5234 * deliver this event there too.
5235 */
5236 if (task && task != current) {
5237 struct perf_event_context *ctx;
5238 struct trace_entry *entry = record;
5239
5240 rcu_read_lock();
5241 ctx = rcu_dereference(task->perf_event_ctxp[perf_sw_context]);
5242 if (!ctx)
5243 goto unlock;
5244
5245 list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
5246 if (event->attr.type != PERF_TYPE_TRACEPOINT)
5247 continue;
5248 if (event->attr.config != entry->type)
5249 continue;
5250 if (perf_tp_event_match(event, &data, regs))
5251 perf_swevent_event(event, count, &data, regs);
5252 }
5253unlock:
5254 rcu_read_unlock();
5255 }
5256
5231 perf_swevent_put_recursion_context(rctx); 5257 perf_swevent_put_recursion_context(rctx);
5232} 5258}
5233EXPORT_SYMBOL_GPL(perf_tp_event); 5259EXPORT_SYMBOL_GPL(perf_tp_event);
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index b0b107f90afc..a096c19f2c2a 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -101,7 +101,8 @@ __output_copy(struct perf_output_handle *handle,
101} 101}
102 102
103/* Callchain handling */ 103/* Callchain handling */
104extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs); 104extern struct perf_callchain_entry *
105perf_callchain(struct perf_event *event, struct pt_regs *regs);
105extern int get_callchain_buffers(void); 106extern int get_callchain_buffers(void);
106extern void put_callchain_buffers(void); 107extern void put_callchain_buffers(void);
107 108
diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c
index fee3752ae8f6..8a6d2ee2086c 100644
--- a/kernel/trace/trace_event_perf.c
+++ b/kernel/trace/trace_event_perf.c
@@ -281,7 +281,7 @@ perf_ftrace_function_call(unsigned long ip, unsigned long parent_ip)
281 281
282 head = this_cpu_ptr(event_function.perf_events); 282 head = this_cpu_ptr(event_function.perf_events);
283 perf_trace_buf_submit(entry, ENTRY_SIZE, rctx, 0, 283 perf_trace_buf_submit(entry, ENTRY_SIZE, rctx, 0,
284 1, &regs, head); 284 1, &regs, head, NULL);
285 285
286#undef ENTRY_SIZE 286#undef ENTRY_SIZE
287} 287}
diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
index b31d3d5699fe..1a2117043bb1 100644
--- a/kernel/trace/trace_kprobe.c
+++ b/kernel/trace/trace_kprobe.c
@@ -1002,7 +1002,8 @@ static __kprobes void kprobe_perf_func(struct kprobe *kp,
1002 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize); 1002 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003 1003
1004 head = this_cpu_ptr(call->perf_events); 1004 head = this_cpu_ptr(call->perf_events);
1005 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head); 1005 perf_trace_buf_submit(entry, size, rctx,
1006 entry->ip, 1, regs, head, NULL);
1006} 1007}
1007 1008
1008/* Kretprobe profile handler */ 1009/* Kretprobe profile handler */
@@ -1033,7 +1034,8 @@ static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1033 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize); 1034 store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1034 1035
1035 head = this_cpu_ptr(call->perf_events); 1036 head = this_cpu_ptr(call->perf_events);
1036 perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head); 1037 perf_trace_buf_submit(entry, size, rctx,
1038 entry->ret_ip, 1, regs, head, NULL);
1037} 1039}
1038#endif /* CONFIG_PERF_EVENTS */ 1040#endif /* CONFIG_PERF_EVENTS */
1039 1041
diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
index 96fc73369099..60e4d7875672 100644
--- a/kernel/trace/trace_syscalls.c
+++ b/kernel/trace/trace_syscalls.c
@@ -532,7 +532,7 @@ static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
532 (unsigned long *)&rec->args); 532 (unsigned long *)&rec->args);
533 533
534 head = this_cpu_ptr(sys_data->enter_event->perf_events); 534 head = this_cpu_ptr(sys_data->enter_event->perf_events);
535 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head); 535 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
536} 536}
537 537
538int perf_sysenter_enable(struct ftrace_event_call *call) 538int perf_sysenter_enable(struct ftrace_event_call *call)
@@ -608,7 +608,7 @@ static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
608 rec->ret = syscall_get_return_value(current, regs); 608 rec->ret = syscall_get_return_value(current, regs);
609 609
610 head = this_cpu_ptr(sys_data->exit_event->perf_events); 610 head = this_cpu_ptr(sys_data->exit_event->perf_events);
611 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head); 611 perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
612} 612}
613 613
614int perf_sysexit_enable(struct ftrace_event_call *call) 614int perf_sysexit_enable(struct ftrace_event_call *call)
diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
index 2b36ac68549e..03003cd7dd96 100644
--- a/kernel/trace/trace_uprobe.c
+++ b/kernel/trace/trace_uprobe.c
@@ -670,7 +670,7 @@ static void uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
670 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset); 670 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
671 671
672 head = this_cpu_ptr(call->perf_events); 672 head = this_cpu_ptr(call->perf_events);
673 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head); 673 perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head, NULL);
674 674
675 out: 675 out:
676 preempt_enable(); 676 preempt_enable();