perf: Clean up power events by introducing new, more generic ones

Add these new power trace events: power:cpu_idle power:cpu_frequency power:machine_suspend The old C-state/idle accounting events: power:power_start power:power_end Have now a replacement (but we are still keeping the old tracepoints for compatibility): power:cpu_idle and power:power_frequency is replaced with: power:cpu_frequency power:machine_suspend is newly introduced. Jean Pihet has a patch integrated into the generic layer (kernel/power/suspend.c) which will make use of it. the type= field got removed from both, it was never used and the type is differed by the event type itself. perf timechart userspace tool gets adjusted in a separate patch. Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Arjan van de Ven <arjan@linux.intel.com> Acked-by: Jean Pihet <jean.pihet@newoldbits.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: rjw@sisk.pl LKML-Reference: <1294073445-14812-3-git-send-email-trenn@suse.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1290072314-31155-2-git-send-email-trenn@suse.de>

perf: Clean up power events by introducing new, more generic ones
Add these new power trace events: power:cpu_idle power:cpu_frequency power:machine_suspend The old C-state/idle accounting events: power:power_start power:power_end Have now a replacement (but we are still keeping the old tracepoints for compatibility): power:cpu_idle and power:power_frequency is replaced with: power:cpu_frequency power:machine_suspend is newly introduced. Jean Pihet has a patch integrated into the generic layer (kernel/power/suspend.c) which will make use of it. the type= field got removed from both, it was never used and the type is differed by the event type itself. perf timechart userspace tool gets adjusted in a separate patch. Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Arjan van de Ven <arjan@linux.intel.com> Acked-by: Jean Pihet <jean.pihet@newoldbits.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: rjw@sisk.pl LKML-Reference: <1294073445-14812-3-git-send-email-trenn@suse.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> LKML-Reference: <1290072314-31155-2-git-send-email-trenn@suse.de>
Thomas Renninger · Ingo Molnar
1 parent 61a0d49c33
Showing 9 changed files with 119 additions and 11 deletions Side-by-side Diff
arch/x86/kernel/process.c
arch/x86/kernel/process_32.c
arch/x86/kernel/process_64.c
drivers/cpufreq/cpufreq.c
drivers/cpuidle/cpuidle.c
drivers/idle/intel_idle.c
include/trace/events/power.h
kernel/trace/Kconfig
kernel/trace/power-traces.c
@@ -373,6 +373,7 @@
 {
 	if (hlt_use_halt()) {
 		trace_power_start(POWER_CSTATE, 1, smp_processor_id());
+		trace_cpu_idle(1, smp_processor_id());
 		current_thread_info()->status &= ~TS_POLLING;
 		/*
 		 * TS_POLLING-cleared state must be visible before we
@@ -443,6 +444,7 @@
 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 {
 	trace_power_start(POWER_CSTATE, (ax>>4)+1, smp_processor_id());
+	trace_cpu_idle((ax>>4)+1, smp_processor_id());
 	if (!need_resched()) {
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
@@ -459,6 +461,7 @@
 {
 	if (!need_resched()) {
 		trace_power_start(POWER_CSTATE, 1, smp_processor_id());
+		trace_cpu_idle(1, smp_processor_id());
 		if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 			clflush((void *)&current_thread_info()->flags);
  
  
@@ -480,10 +483,12 @@
 static void poll_idle(void)
 {
 	trace_power_start(POWER_CSTATE, 0, smp_processor_id());
+	trace_cpu_idle(0, smp_processor_id());
 	local_irq_enable();
 	while (!need_resched())
 		cpu_relax();
-	trace_power_end(0);
+	trace_power_end(smp_processor_id());
+	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 }
  
 /*
@@ -113,8 +113,8 @@
 			stop_critical_timings();
 			pm_idle();
 			start_critical_timings();
-
 			trace_power_end(smp_processor_id());
+			trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 		}
 		tick_nohz_restart_sched_tick();
 		preempt_enable_no_resched();
@@ -142,6 +142,8 @@
 			start_critical_timings();
  
 			trace_power_end(smp_processor_id());
+			trace_cpu_idle(PWR_EVENT_EXIT,
+				       smp_processor_id());
  
 			/* In many cases the interrupt that ended idle
 			   has already called exit_idle. But some idle
@@ -355,6 +355,7 @@
 		dprintk("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
 			(unsigned long)freqs->cpu);
 		trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
+		trace_cpu_frequency(freqs->new, freqs->cpu);
 		srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 				CPUFREQ_POSTCHANGE, freqs);
 		if (likely(policy) && likely(policy->cpu == freqs->cpu))
@@ -107,6 +107,7 @@
 	if (cpuidle_curr_governor->reflect)
 		cpuidle_curr_governor->reflect(dev);
 	trace_power_end(smp_processor_id());
+	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 }
  
 /**
@@ -221,6 +221,7 @@
  
 	stop_critical_timings();
 	trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu);
+	trace_cpu_idle((eax >> 4) + 1, cpu);
 	if (!need_resched()) {
  
 		__monitor((void *)&current_thread_info()->flags, 0, 0);
@@ -7,16 +7,67 @@
 #include <linux/ktime.h>
 #include <linux/tracepoint.h>
  
-#ifndef _TRACE_POWER_ENUM_
-#define _TRACE_POWER_ENUM_
-enum {
-	POWER_NONE	= 0,
-	POWER_CSTATE	= 1,	/* C-State */
-	POWER_PSTATE	= 2,	/* Fequency change or DVFS */
-	POWER_SSTATE	= 3,	/* Suspend */
-};
+DECLARE_EVENT_CLASS(cpu,
+
+	TP_PROTO(unsigned int state, unsigned int cpu_id),
+
+	TP_ARGS(state, cpu_id),
+
+	TP_STRUCT__entry(
+		__field(	u32,		state		)
+		__field(	u32,		cpu_id		)
+	),
+
+	TP_fast_assign(
+		__entry->state = state;
+		__entry->cpu_id = cpu_id;
+	),
+
+	TP_printk("state=%lu cpu_id=%lu", (unsigned long)__entry->state,
+		  (unsigned long)__entry->cpu_id)
+);
+
+DEFINE_EVENT(cpu, cpu_idle,
+
+	TP_PROTO(unsigned int state, unsigned int cpu_id),
+
+	TP_ARGS(state, cpu_id)
+);
+
+/* This file can get included multiple times, TRACE_HEADER_MULTI_READ at top */
+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING
+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING
+
+#define PWR_EVENT_EXIT -1
 #endif
  
+DEFINE_EVENT(cpu, cpu_frequency,
+
+	TP_PROTO(unsigned int frequency, unsigned int cpu_id),
+
+	TP_ARGS(frequency, cpu_id)
+);
+
+TRACE_EVENT(machine_suspend,
+
+	TP_PROTO(unsigned int state),
+
+	TP_ARGS(state),
+
+	TP_STRUCT__entry(
+		__field(	u32,		state		)
+	),
+
+	TP_fast_assign(
+		__entry->state = state;
+	),
+
+	TP_printk("state=%lu", (unsigned long)__entry->state)
+);
+
+/* This code will be removed after deprecation time exceeded (2.6.41) */
+#ifdef CONFIG_EVENT_POWER_TRACING_DEPRECATED
+
 /*
  * The power events are used for cpuidle & suspend (power_start, power_end)
  *  and for cpufreq (power_frequency)
@@ -75,6 +126,36 @@
  
 );
  
+/* Deprecated dummy functions must be protected against multi-declartion */
+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
+
+enum {
+	POWER_NONE = 0,
+	POWER_CSTATE = 1,
+	POWER_PSTATE = 2,
+};
+#endif /* _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED */
+
+#else /* CONFIG_EVENT_POWER_TRACING_DEPRECATED */
+
+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
+enum {
+       POWER_NONE = 0,
+       POWER_CSTATE = 1,
+       POWER_PSTATE = 2,
+};
+
+/* These dummy declaration have to be ripped out when the deprecated
+   events get removed */
+static inline void trace_power_start(u64 type, u64 state, u64 cpuid) {};
+static inline void trace_power_end(u64 cpuid) {};
+static inline void trace_power_frequency(u64 type, u64 state, u64 cpuid) {};
+#endif /* _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED */
+
+#endif /* CONFIG_EVENT_POWER_TRACING_DEPRECATED */
+
 /*
  * The clock events are used for clock enable/disable and for
  *  clock rate change
@@ -153,7 +234,6 @@
  
 	TP_ARGS(name, state, cpu_id)
 );
-
 #endif /* _TRACE_POWER_H */
  
 /* This part must be outside protection */
@@ -69,6 +69,21 @@
 	select CONTEXT_SWITCH_TRACER
 	bool
  
+config EVENT_POWER_TRACING_DEPRECATED
+	depends on EVENT_TRACING
+	bool "Deprecated power event trace API, to be removed"
+	default y
+	help
+	  Provides old power event types:
+	  C-state/idle accounting events:
+	  power:power_start
+	  power:power_end
+	  and old cpufreq accounting event:
+	  power:power_frequency
+	  This is for userspace compatibility
+	  and will vanish after 5 kernel iterations,
+	  namely 2.6.41.
+
 config CONTEXT_SWITCH_TRACER
 	bool
  
@@ -13,5 +13,8 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/power.h>
  
+#ifdef EVENT_POWER_TRACING_DEPRECATED
 EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
+#endif
+EXPORT_TRACEPOINT_SYMBOL_GPL(cpu_idle);
...	...	@@ -373,6 +373,7 @@
373	373	{
374	374	if (hlt_use_halt()) {
375	375	trace_power_start(POWER_CSTATE, 1, smp_processor_id());
	376	+ trace_cpu_idle(1, smp_processor_id());
376	377	current_thread_info()->status &= ~TS_POLLING;
377	378	/*
378	379	* TS_POLLING-cleared state must be visible before we
...	...	@@ -443,6 +444,7 @@
443	444	void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
444	445	{
445	446	trace_power_start(POWER_CSTATE, (ax>>4)+1, smp_processor_id());
	447	+ trace_cpu_idle((ax>>4)+1, smp_processor_id());
446	448	if (!need_resched()) {
447	449	if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
448	450	clflush((void *)&current_thread_info()->flags);
...	...	@@ -459,6 +461,7 @@
459	461	{
460	462	if (!need_resched()) {
461	463	trace_power_start(POWER_CSTATE, 1, smp_processor_id());
	464	+ trace_cpu_idle(1, smp_processor_id());
462	465	if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
463	466	clflush((void *)&current_thread_info()->flags);
464	467
465	468
...	...	@@ -480,10 +483,12 @@
480	483	static void poll_idle(void)
481	484	{
482	485	trace_power_start(POWER_CSTATE, 0, smp_processor_id());
	486	+ trace_cpu_idle(0, smp_processor_id());
483	487	local_irq_enable();
484	488	while (!need_resched())
485	489	cpu_relax();
486		- trace_power_end(0);
	490	+ trace_power_end(smp_processor_id());
	491	+ trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
487	492	}
488	493
489	494	/*
...	...	@@ -113,8 +113,8 @@
113	113	stop_critical_timings();
114	114	pm_idle();
115	115	start_critical_timings();
116		-
117	116	trace_power_end(smp_processor_id());
	117	+ trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
118	118	}
119	119	tick_nohz_restart_sched_tick();
120	120	preempt_enable_no_resched();
...	...	@@ -142,6 +142,8 @@
142	142	start_critical_timings();
143	143
144	144	trace_power_end(smp_processor_id());
	145	+ trace_cpu_idle(PWR_EVENT_EXIT,
	146	+ smp_processor_id());
145	147
146	148	/* In many cases the interrupt that ended idle
147	149	has already called exit_idle. But some idle
...	...	@@ -355,6 +355,7 @@
355	355	dprintk("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
356	356	(unsigned long)freqs->cpu);
357	357	trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
	358	+ trace_cpu_frequency(freqs->new, freqs->cpu);
358	359	srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
359	360	CPUFREQ_POSTCHANGE, freqs);
360	361	if (likely(policy) && likely(policy->cpu == freqs->cpu))
...	...	@@ -107,6 +107,7 @@
107	107	if (cpuidle_curr_governor->reflect)
108	108	cpuidle_curr_governor->reflect(dev);
109	109	trace_power_end(smp_processor_id());
	110	+ trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
110	111	}
111	112
112	113	/**
...	...	@@ -221,6 +221,7 @@
221	221
222	222	stop_critical_timings();
223	223	trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu);
	224	+ trace_cpu_idle((eax >> 4) + 1, cpu);
224	225	if (!need_resched()) {
225	226
226	227	__monitor((void *)&current_thread_info()->flags, 0, 0);
...	...	@@ -7,16 +7,67 @@
7	7	#include <linux/ktime.h>
8	8	#include <linux/tracepoint.h>
9	9
10		-#ifndef _TRACE_POWER_ENUM_
11		-#define _TRACE_POWER_ENUM_
12		-enum {
13		- POWER_NONE = 0,
14		- POWER_CSTATE = 1, /* C-State */
15		- POWER_PSTATE = 2, /* Fequency change or DVFS */
16		- POWER_SSTATE = 3, /* Suspend */
17		-};
	10	+DECLARE_EVENT_CLASS(cpu,
	11	+
	12	+ TP_PROTO(unsigned int state, unsigned int cpu_id),
	13	+
	14	+ TP_ARGS(state, cpu_id),
	15	+
	16	+ TP_STRUCT__entry(
	17	+ __field( u32, state )
	18	+ __field( u32, cpu_id )
	19	+ ),
	20	+
	21	+ TP_fast_assign(
	22	+ __entry->state = state;
	23	+ __entry->cpu_id = cpu_id;
	24	+ ),
	25	+
	26	+ TP_printk("state=%lu cpu_id=%lu", (unsigned long)__entry->state,
	27	+ (unsigned long)__entry->cpu_id)
	28	+);
	29	+
	30	+DEFINE_EVENT(cpu, cpu_idle,
	31	+
	32	+ TP_PROTO(unsigned int state, unsigned int cpu_id),
	33	+
	34	+ TP_ARGS(state, cpu_id)
	35	+);
	36	+
	37	+/* This file can get included multiple times, TRACE_HEADER_MULTI_READ at top */
	38	+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING
	39	+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING
	40	+
	41	+#define PWR_EVENT_EXIT -1
18	42	#endif
19	43
	44	+DEFINE_EVENT(cpu, cpu_frequency,
	45	+
	46	+ TP_PROTO(unsigned int frequency, unsigned int cpu_id),
	47	+
	48	+ TP_ARGS(frequency, cpu_id)
	49	+);
	50	+
	51	+TRACE_EVENT(machine_suspend,
	52	+
	53	+ TP_PROTO(unsigned int state),
	54	+
	55	+ TP_ARGS(state),
	56	+
	57	+ TP_STRUCT__entry(
	58	+ __field( u32, state )
	59	+ ),
	60	+
	61	+ TP_fast_assign(
	62	+ __entry->state = state;
	63	+ ),
	64	+
	65	+ TP_printk("state=%lu", (unsigned long)__entry->state)
	66	+);
	67	+
	68	+/* This code will be removed after deprecation time exceeded (2.6.41) */
	69	+#ifdef CONFIG_EVENT_POWER_TRACING_DEPRECATED
	70	+
20	71	/*
21	72	* The power events are used for cpuidle & suspend (power_start, power_end)
22	73	* and for cpufreq (power_frequency)
...	...	@@ -75,6 +126,36 @@
75	126
76	127	);
77	128
	129	+/* Deprecated dummy functions must be protected against multi-declartion */
	130	+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
	131	+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
	132	+
	133	+enum {
	134	+ POWER_NONE = 0,
	135	+ POWER_CSTATE = 1,
	136	+ POWER_PSTATE = 2,
	137	+};
	138	+#endif /* _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED */
	139	+
	140	+#else /* CONFIG_EVENT_POWER_TRACING_DEPRECATED */
	141	+
	142	+#ifndef _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
	143	+#define _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED
	144	+enum {
	145	+ POWER_NONE = 0,
	146	+ POWER_CSTATE = 1,
	147	+ POWER_PSTATE = 2,
	148	+};
	149	+
	150	+/* These dummy declaration have to be ripped out when the deprecated
	151	+ events get removed */
	152	+static inline void trace_power_start(u64 type, u64 state, u64 cpuid) {};
	153	+static inline void trace_power_end(u64 cpuid) {};
	154	+static inline void trace_power_frequency(u64 type, u64 state, u64 cpuid) {};
	155	+#endif /* _PWR_EVENT_AVOID_DOUBLE_DEFINING_DEPRECATED */
	156	+
	157	+#endif /* CONFIG_EVENT_POWER_TRACING_DEPRECATED */
	158	+
78	159	/*
79	160	* The clock events are used for clock enable/disable and for
80	161	* clock rate change
...	...	@@ -153,7 +234,6 @@
153	234
154	235	TP_ARGS(name, state, cpu_id)
155	236	);
156		-
157	237	#endif /* _TRACE_POWER_H */
158	238
159	239	/* This part must be outside protection */
...	...	@@ -69,6 +69,21 @@
69	69	select CONTEXT_SWITCH_TRACER
70	70	bool
71	71
	72	+config EVENT_POWER_TRACING_DEPRECATED
	73	+ depends on EVENT_TRACING
	74	+ bool "Deprecated power event trace API, to be removed"
	75	+ default y
	76	+ help
	77	+ Provides old power event types:
	78	+ C-state/idle accounting events:
	79	+ power:power_start
	80	+ power:power_end
	81	+ and old cpufreq accounting event:
	82	+ power:power_frequency
	83	+ This is for userspace compatibility
	84	+ and will vanish after 5 kernel iterations,
	85	+ namely 2.6.41.
	86	+
72	87	config CONTEXT_SWITCH_TRACER
73	88	bool
74	89
...	...	@@ -13,5 +13,8 @@
13	13	#define CREATE_TRACE_POINTS
14	14	#include <trace/events/power.h>
15	15
	16	+#ifdef EVENT_POWER_TRACING_DEPRECATED
16	17	EXPORT_TRACEPOINT_SYMBOL_GPL(power_start);
	18	+#endif
	19	+EXPORT_TRACEPOINT_SYMBOL_GPL(cpu_idle);