x86 cpufreq, perf: Make trace_power_frequency cpufreq driver independent
authorThomas Renninger <trenn@suse.de>
Tue, 20 Jul 2010 23:59:34 +0000 (16:59 -0700)
committerIngo Molnar <mingo@elte.hu>
Thu, 22 Jul 2010 10:08:27 +0000 (12:08 +0200)
and fix the broken case if a core's frequency depends on others.

trace_power_frequency was only implemented in a rather ungeneric
way in acpi-cpufreq driver's target() function only.

-> Move the call to trace_power_frequency to
   cpufreq.c:cpufreq_notify_transition() where CPUFREQ_POSTCHANGE
   notifier is triggered.
   This will support power frequency tracing by all cpufreq
   drivers.

trace_power_frequency did not trace frequency changes correctly
when the userspace governor was used or when CPU cores'
frequency depend on each other.

-> Moving this into the CPUFREQ_POSTCHANGE notifier and pass the cpu
   which gets switched automatically fixes this.

Robert Schoene provided some important fixes on top of my
initial quick shot version which are integrated in this patch:
- Forgot some changes in power_end trace (TP_printk/variable names)
- Variable dummy in power_end must now be cpu_id
- Use static 64 bit variable instead of unsigned int for cpu_id

[akpm@linux-foundation.org: build fix]
Signed-off-by: Thomas Renninger <trenn@suse.de>
Cc: davej@codemonkey.org.uk
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Dave Jones <davej@codemonkey.org.uk>
Acked-by: Arjan van de Ven <arjan@infradead.org>
Cc: Robert Schoene <robert.schoene@tu-dresden.de>
Tested-by: Robert Schoene <robert.schoene@tu-dresden.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c
arch/x86/kernel/process.c
drivers/cpufreq/cpufreq.c
drivers/cpuidle/cpuidle.c
drivers/idle/intel_idle.c
include/trace/events/power.h
tools/perf/builtin-timechart.c

index 1d3cdda..cee5263 100644 (file)
@@ -34,7 +34,6 @@
 #include <linux/compiler.h>
 #include <linux/dmi.h>
 #include <linux/slab.h>
-#include <trace/events/power.h>
 
 #include <linux/acpi.h>
 #include <linux/io.h>
@@ -324,8 +323,6 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
                }
        }
 
-       trace_power_frequency(POWER_PSTATE, data->freq_table[next_state].frequency);
-
        switch (data->cpu_feature) {
        case SYSTEM_INTEL_MSR_CAPABLE:
                cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
index e7e3521..787572d 100644 (file)
@@ -371,7 +371,7 @@ static inline int hlt_use_halt(void)
 void default_idle(void)
 {
        if (hlt_use_halt()) {
-               trace_power_start(POWER_CSTATE, 1);
+               trace_power_start(POWER_CSTATE, 1, smp_processor_id());
                current_thread_info()->status &= ~TS_POLLING;
                /*
                 * TS_POLLING-cleared state must be visible before we
@@ -441,7 +441,7 @@ 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);
+       trace_power_start(POWER_CSTATE, (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);
@@ -457,7 +457,7 @@ void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 static void mwait_idle(void)
 {
        if (!need_resched()) {
-               trace_power_start(POWER_CSTATE, 1);
+               trace_power_start(POWER_CSTATE, 1, smp_processor_id());
                if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
                        clflush((void *)&current_thread_info()->flags);
 
@@ -478,7 +478,7 @@ static void mwait_idle(void)
  */
 static void poll_idle(void)
 {
-       trace_power_start(POWER_CSTATE, 0);
+       trace_power_start(POWER_CSTATE, 0, smp_processor_id());
        local_irq_enable();
        while (!need_resched())
                cpu_relax();
index 063b218..4ed6657 100644 (file)
@@ -29,6 +29,8 @@
 #include <linux/completion.h>
 #include <linux/mutex.h>
 
+#include <trace/events/power.h>
+
 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
                                                "cpufreq-core", msg)
 
@@ -354,6 +356,7 @@ void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 
        case CPUFREQ_POSTCHANGE:
                adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
+                trace_power_frequency(POWER_PSTATE, freqs->new, freqs->cpu);
                srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
                                CPUFREQ_POSTCHANGE, freqs);
                if (likely(policy) && likely(policy->cpu == freqs->cpu))
index 1994885..dbefe15 100644 (file)
@@ -95,7 +95,7 @@ static void cpuidle_idle_call(void)
        /* give the governor an opportunity to reflect on the outcome */
        if (cpuidle_curr_governor->reflect)
                cpuidle_curr_governor->reflect(dev);
-       trace_power_end(0);
+       trace_power_end(smp_processor_id());
 }
 
 /**
index 54f0fb4..03d202b 100755 (executable)
@@ -231,7 +231,7 @@ static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
 
        stop_critical_timings();
 #ifndef MODULE
-       trace_power_start(POWER_CSTATE, (eax >> 4) + 1);
+       trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu);
 #endif
        if (!need_resched()) {
 
index c4efe9b..35a2a6e 100644 (file)
@@ -18,52 +18,55 @@ enum {
 
 DECLARE_EVENT_CLASS(power,
 
-       TP_PROTO(unsigned int type, unsigned int state),
+       TP_PROTO(unsigned int type, unsigned int state, unsigned int cpu_id),
 
-       TP_ARGS(type, state),
+       TP_ARGS(type, state, cpu_id),
 
        TP_STRUCT__entry(
                __field(        u64,            type            )
                __field(        u64,            state           )
+               __field(        u64,            cpu_id          )
        ),
 
        TP_fast_assign(
                __entry->type = type;
                __entry->state = state;
+               __entry->cpu_id = cpu_id;
        ),
 
-       TP_printk("type=%lu state=%lu", (unsigned long)__entry->type, (unsigned long)__entry->state)
+       TP_printk("type=%lu state=%lu cpu_id=%lu", (unsigned long)__entry->type,
+               (unsigned long)__entry->state, (unsigned long)__entry->cpu_id)
 );
 
 DEFINE_EVENT(power, power_start,
 
-       TP_PROTO(unsigned int type, unsigned int state),
+       TP_PROTO(unsigned int type, unsigned int state, unsigned int cpu_id),
 
-       TP_ARGS(type, state)
+       TP_ARGS(type, state, cpu_id)
 );
 
 DEFINE_EVENT(power, power_frequency,
 
-       TP_PROTO(unsigned int type, unsigned int state),
+       TP_PROTO(unsigned int type, unsigned int state, unsigned int cpu_id),
 
-       TP_ARGS(type, state)
+       TP_ARGS(type, state, cpu_id)
 );
 
 TRACE_EVENT(power_end,
 
-       TP_PROTO(int dummy),
+       TP_PROTO(unsigned int cpu_id),
 
-       TP_ARGS(dummy),
+       TP_ARGS(cpu_id),
 
        TP_STRUCT__entry(
-               __field(        u64,            dummy           )
+               __field(        u64,            cpu_id          )
        ),
 
        TP_fast_assign(
-               __entry->dummy = 0xffff;
+               __entry->cpu_id = cpu_id;
        ),
 
-       TP_printk("dummy=%lu", (unsigned long)__entry->dummy)
+       TP_printk("cpu_id=%lu", (unsigned long)__entry->cpu_id)
 
 );
 
index 5a52ed9..5161619 100644 (file)
@@ -300,8 +300,9 @@ struct trace_entry {
 
 struct power_entry {
        struct trace_entry te;
-       s64     type;
-       s64     value;
+       u64     type;
+       u64     value;
+       u64     cpu_id;
 };
 
 #define TASK_COMM_LEN 16
@@ -498,13 +499,13 @@ static int process_sample_event(event_t *event, struct perf_session *session)
                        return 0;
 
                if (strcmp(event_str, "power:power_start") == 0)
-                       c_state_start(data.cpu, data.time, pe->value);
+                       c_state_start(pe->cpu_id, data.time, pe->value);
 
                if (strcmp(event_str, "power:power_end") == 0)
-                       c_state_end(data.cpu, data.time);
+                       c_state_end(pe->cpu_id, data.time);
 
                if (strcmp(event_str, "power:power_frequency") == 0)
-                       p_state_change(data.cpu, data.time, pe->value);
+                       p_state_change(pe->cpu_id, data.time, pe->value);
 
                if (strcmp(event_str, "sched:sched_wakeup") == 0)
                        sched_wakeup(data.cpu, data.time, data.pid, te);