Power trace event cleanup by still providing old interface for some time
Subsystems (6)
and another one -> did not include lkml on last changes and corrected Jean's email address to what he'd like to see in patches.
Thomas
PATCH 1/3 PERF: Do not export power_frequency, but power_start event
power_frequency moved to drivers/cpufreq/cpufreq.c which has to be compiled in, no need to export it.
intel_idle can a be module though...
--- drivers/idle/intel_idle.c | 2 -- kernel/trace/power-traces.c | 2 +- 2 files changed, 1 insertions(+), 3 deletions(-)
drivers/idle/intel_idle.c +0 -2
@@ -220,9 +220,7 @@ static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state) kt_before =ktime_get_real ();stop_critical_timings (); -#ifndef MODULE trace_power_start (POWER_CSTATE , (eax >>4 ) +1 , cpu); -#endif if (!need_resched ()) {__monitor ((void *)¤t_thread_info ()->flags,0 ,0 );
kernel/trace/power-traces.c +1 -1
@@ -13,5 +13,5 @@#define CREATE_TRACE_POINTS #include <trace/events/power.h> -EXPORT_TRACEPOINT_SYMBOL_GPL (power_frequency); +EXPORT_TRACEPOINT_SYMBOL_GPL (power_start);
Signature · 1 line
1.6.3
PATCH 2/3 PERF(kernel): Cleanup power events
In reply to Thomas RenningerRecent changes:
- Enable EVENT_POWER_TRACING_DEPRECATED by default
New power trace events:
power:cpu_idle
power:cpu_frequency
power:machine_suspend
C-state/idle accounting events:
power:power_start
power:power_end
are replaced with:
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.
--- arch/x86/kernel/process.c | 7 +++- arch/x86/kernel/process_32.c | 2 +- arch/x86/kernel/process_64.c | 2 + drivers/cpufreq/cpufreq.c | 1 + drivers/cpuidle/cpuidle.c | 1 + drivers/idle/intel_idle.c | 1 + include/trace/events/power.h | 86 +++++++++++++++++++++++++++++++++++++---- kernel/trace/Kconfig | 15 +++++++ kernel/trace/power-traces.c | 3 + 9 files changed, 107 insertions(+), 11 deletions(-)
arch/x86/kernel/process.c +6 -1
@@ -374,6 +374,7 @@ void default_idle(void) {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
@@ -444,6 +445,7 @@ EXPORT_SYMBOL_GPL(cpu_idle_wait);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 (¤t_cpu_data,X86_FEATURE_CLFLUSH_MONITOR ))clflush ((void *)¤t_thread_info ()->flags);
@@ -460,6 +462,7 @@ static void mwait_idle(void) {if (!need_resched ()) {trace_power_start (POWER_CSTATE ,1 ,smp_processor_id ()); +trace_cpu_idle (1 ,smp_processor_id ());if (cpu_has (¤t_cpu_data,X86_FEATURE_CLFLUSH_MONITOR ))clflush ((void *)¤t_thread_info ()->flags);
@@ -481,10 +484,12 @@ static void mwait_idle(void)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 ()); }/*
arch/x86/kernel/process_32.c +1 -1
@@ -113,8 +113,8 @@ void cpu_idle(void)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 ();
arch/x86/kernel/process_64.c +2 -0
@@ -142,6 +142,8 @@ void cpu_idle(void)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
drivers/cpufreq/cpufreq.c +1 -0
@@ -355,6 +355,7 @@ void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)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))
drivers/cpuidle/cpuidle.c +1 -0
@@ -107,6 +107,7 @@ static void cpuidle_idle_call(void)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 ()); }/**
drivers/idle/intel_idle.c +1 -0
@@ -221,6 +221,7 @@ static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)stop_critical_timings ();trace_power_start (POWER_CSTATE , (eax >>4 ) +1 , cpu); +trace_cpu_idle ((eax >>4 ) +1 , cpu);if (!need_resched ()) {__monitor ((void *)¤t_thread_info ()->flags,0 ,0 );
include/trace/events/power.h +77 -9
106 lines not on this page · raw message
kernel/trace/Kconfig +15 -0
@@ -69,6 +69,21 @@ config EVENT_TRACING 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
kernel/trace/power-traces.c +3 -0
@@ -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);
Signature · 1 line
1.6.3
PATCH 3/3 PERF(userspace): Adjust perf timechart to the new power events
In reply to Thomas RenningerRecent changes:
- Adjust state/cpuid to u32 as done in the kernel
The transition was rather smooth, only part I had to fiddle some time was the check whether a tracepoint/event is supported by the running kernel.
builtin-timechart must only pass -e power:xy events which are supported by the running kernel.
For this I added the tiny helper function:
int is_valid_tracepoint(const char *event_string)
to parse-events.[hc]
which could be more generic as an interface and support hardware/software/... events, not only tracepoints, but someone else could extend that if needed...
--- tools/perf/builtin-timechart.c | 91 ++++++++++++++++++++++++++++++++------- tools/perf/util/parse-events.c | 41 ++++++++++++++++++ tools/perf/util/parse-events.h | 1 + 3 files changed, 116 insertions(+), 17 deletions(-)
tools/perf/builtin-timechart.c +74 -17
@@ -32,6 +32,10 @@#include "util/session.h" #include "util/svghelper.h" +#define SUPPORT_OLD_POWER_EVENTS 1 +#define PWR_EVENT_EXIT -1 + +static char const *input_name ="perf.data" ;static char const *output_name ="output.svg" ;
@@ -298,12 +302,21 @@ struct trace_entry {int lock_depth; }; -struct power_entry { +#if defined(SUPPORT_OLD_POWER_EVENTS ) +static int use_old_power_events; +struct power_entry_old {struct trace_entry te;u64 type;u64 value;u64 cpu_id; }; +#endif + +struct power_processor_entry { +struct trace_entry te; +u32 state; +u32 cpu_id; +};#define TASK_COMM_LEN 16struct wakeup_entry {
@@ -489,29 +502,48 @@ static int process_sample_event(event_t *event, struct perf_session *session) te = (void *)data.raw_data;if (session->sample_type &PERF_SAMPLE_RAW && data.raw_size >0 ) {char *event_str; -struct power_entry *pe; - - pe = (void *)te; +#if defined(SUPPORT_OLD_POWER_EVENTS ) +struct power_entry_old *peo; + peo = (void *)te; +#endif event_str =perf_header__find_event (te->type);if (!event_str)return 0 ; -if (strcmp (event_str,"power:power_start" ) ==0 ) -c_state_start (pe->cpu_id, data.time, pe->value); - -if (strcmp (event_str,"power:power_end" ) ==0 ) -c_state_end (pe->cpu_id, data.time); +if (strcmp (event_str,"power:cpu_idle" ) ==0 ) { +struct power_processor_entry *ppe = (void *)te; +if (ppe->state == (u32 )PWR_EVENT_EXIT ) +c_state_end (ppe->cpu_id, data.time); +else +c_state_start (ppe->cpu_id, data.time, + ppe->state); + } -if (strcmp (event_str,"power:power_frequency" ) ==0 ) -p_state_change (pe->cpu_id, data.time, pe->value); +else if (strcmp (event_str,"power:cpu_frequency" ) ==0 ) { +struct power_processor_entry *ppe = (void *)te; +p_state_change (ppe->cpu_id, data.time, ppe->state); + } -if (strcmp (event_str,"sched:sched_wakeup" ) ==0 ) +else if (strcmp (event_str,"sched:sched_wakeup" ) ==0 )sched_wakeup (data.cpu, data.time, data.pid, te); -if (strcmp (event_str,"sched:sched_switch" ) ==0 ) +else if (strcmp (event_str,"sched:sched_switch" ) ==0 )sched_switch (data.cpu, data.time, te); + +#if defined(SUPPORT_OLD_POWER_EVENTS ) +if (use_old_power_events) { +if (strcmp (event_str,"power:power_start" ) ==0 ) +c_state_start (peo->cpu_id, data.time, peo->value); + +else if (strcmp (event_str,"power:power_end" ) ==0 ) +c_state_end (data.cpu, data.time); + +else if (strcmp (event_str,"power:power_frequency" ) ==0 ) +p_state_change (peo->cpu_id, data.time, peo->value); + } +#endif }return 0 ; }
@@ -968,7 +1000,8 @@ static const char * const timechart_usage[] = {NULL }; -static const char *record_args[] = { +#if defined(SUPPORT_OLD_POWER_EVENTS ) +static const char *record_old_args[] = {"record" ,"-a" , "-R",
@@ -980,16 +1013,40 @@ static const char *record_args[] = {"-e" ,"sched:sched_wakeup" , "-e", "sched:sched_switch", }; +#endif + +static const char *record_new_args[] = { +"record" , +"-a" , +"-R" , +"-f" , +"-c" ,"1" , +"-e" ,"power:cpu_frequency" , +"-e" ,"power:cpu_idle" , +"-e" ,"sched:sched_wakeup" , +"-e" ,"sched:sched_switch" , +};static int __cmd_record (int argc,const char **argv) {unsigned int rec_argc, i, j;const char **rec_argv; - - rec_argc =ARRAY_SIZE (record_args) + argc -1 ; +const char **record_args = record_new_args; +unsigned int record_elems =ARRAY_SIZE (record_new_args); + +#if defined(SUPPORT_OLD_POWER_EVENTS ) +if (!is_valid_tracepoint ("power:cpu_idle" ) && +is_valid_tracepoint ("power:power_start" )) { + use_old_power_events =1 ; + record_args = record_old_args; + record_elems =ARRAY_SIZE (record_old_args); + } +#endif + + rec_argc = record_elems + argc -1 ; rec_argv =calloc (rec_argc +1 ,sizeof (char *)); -for (i =0 ; i <ARRAY_SIZE (record_args); i++) +for (i =0 ; i < record_elems; i++) rec_argv[i] =strdup (record_args[i]);for (j =1 ; j < (unsigned int )argc; j++, i++)
tools/perf/util/parse-events.c +41 -0
47 lines not on this page · raw message
tools/perf/util/parse-events.h +1 -0
@@ -29,6 +29,7 @@ extern int parse_filter(const struct option *opt, const char *str, int unset);#define EVENTS_HELP_MAX (128*1024)extern void print_events (void ); +extern int is_valid_tracepoint (const char *event_string);extern char debugfs_path[];extern int valid_debugfs_mount (const char *debugfs);
Signature · 1 line
1.6.3