Hi,
Here is a patch proposal for adding new trace events for power management.
Comments are welcome on this email address Jean Pihet
<[email protected]>.
Jean
>From b1d6df4fb4ebb72dca6660659d7030736c2a5cb2 Mon Sep 17 00:00:00 2001
From: Jean Pihet <[email protected]>
Date: Mon, 16 Aug 2010 12:31:31 +0200
Subject: [PATCH] tracing, perf: add more power related events
This patch adds new generic events for dynamic power management
tracing:
- clock events class: used for clock enable/disable and for
clock rate change,
- power_domain events class: used for power domains transitions.
The OMAP architecture is using the new events for PM debugging, however
the new events are made generic enough to be used by all platforms.
Signed-off-by: Jean Pihet <[email protected]>
---
include/trace/events/power.h | 84 ++++++++++++++++++++++++++++++++++++++++++
1 files changed, 84 insertions(+), 0 deletions(-)
diff --git a/include/trace/events/power.h b/include/trace/events/power.h
index 35a2a6e..0caa79c 100644
--- a/include/trace/events/power.h
+++ b/include/trace/events/power.h
@@ -16,6 +16,10 @@ enum {
};
#endif
+/*
+ * The power events are used for cpuidle (power_start, power_end) and for
+ * cpufreq (power_frequency)
+ */
DECLARE_EVENT_CLASS(power,
TP_PROTO(unsigned int type, unsigned int state, unsigned int cpu_id),
@@ -70,6 +74,86 @@ TRACE_EVENT(power_end,
);
+/*
+ * The clock events are used for clock enable/disable and for
+ * clock rate change
+ */
+DECLARE_EVENT_CLASS(clock,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id),
+
+ TP_STRUCT__entry(
+ __string( name, name )
+ __field( u64, state )
+ __field( u64, cpu_id )
+ ),
+
+ TP_fast_assign(
+ __assign_str(name, name);
+ __entry->state = state;
+ __entry->cpu_id = cpu_id;
+ ),
+
+ TP_printk("name=%s state=%lu cpu_id=%lu", __get_str(name),
+ (unsigned long)__entry->state, (unsigned long)__entry->cpu_id)
+);
+
+
+DEFINE_EVENT(clock, clock_enable,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id)
+);
+
+DEFINE_EVENT(clock, clock_disable,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id)
+);
+
+DEFINE_EVENT(clock, clock_set_rate,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id)
+);
+
+/*
+ * The power domain events are used for power domains transitions
+ */
+DECLARE_EVENT_CLASS(power_domain,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id),
+
+ TP_STRUCT__entry(
+ __string( name, name )
+ __field( u64, state )
+ __field( u64, cpu_id )
+ ),
+
+ TP_fast_assign(
+ __assign_str(name, name);
+ __entry->state = state;
+ __entry->cpu_id = cpu_id;
+ ),
+
+ TP_printk("name=%s state=%lu cpu_id=%lu", __get_str(name),
+ (unsigned long)__entry->state, (unsigned long)__entry->cpu_id)
+);
+
+DEFINE_EVENT(power_domain, power_domain_transition,
+
+ TP_PROTO(const char *name, unsigned int state, unsigned int cpu_id),
+
+ TP_ARGS(name, state, cpu_id)
+);
+
#endif /* _TRACE_POWER_H */
/* This part must be outside protection */
--
1.7.1
On Mon, 2010-08-16 at 13:09 +0200, Jean Pihet wrote:
> This patch adds new generic events for dynamic power management
> tracing:
> - clock events class: used for clock enable/disable and for
> clock rate change,
> - power_domain events class: used for power domains transitions.
>
> The OMAP architecture is using the new events for PM debugging, however
> the new events are made generic enough to be used by all platforms.
Even though I'm blissfully ignorant on power stuffs, the proposed set of
trace thingies doesn't appear to even start covering x86..
CC'ed people who know more on the subject.
On Mon, Aug 16, 2010 at 1:15 PM, Peter Zijlstra <[email protected]> wrote:
> On Mon, 2010-08-16 at 13:09 +0200, Jean Pihet wrote:
>> This patch adds new generic events for dynamic power management
>> tracing:
>> - clock events class: used for clock enable/disable and for
>> ? ?clock rate change,
>> - power_domain events class: used for power domains transitions.
>>
>> The OMAP architecture is using the new events for PM debugging, however
>> the new events are made generic enough to be used by all platforms.
>
> Even though I'm blissfully ignorant on power stuffs, the proposed set of
> trace thingies doesn't appear to even start covering x86..
Ok, let's discuss it! The intention is to provide new generic events.
>
> CC'ed people who know more on the subject.
>
Thanks!
Hi,
On Mon, Aug 16, 2010 at 1:25 PM, Jean Pihet <[email protected]> wrote:
> On Mon, Aug 16, 2010 at 1:15 PM, Peter Zijlstra <[email protected]> wrote:
>> On Mon, 2010-08-16 at 13:09 +0200, Jean Pihet wrote:
>>> This patch adds new generic events for dynamic power management
>>> tracing:
>>> - clock events class: used for clock enable/disable and for
>>> ? ?clock rate change,
>>> - power_domain events class: used for power domains transitions.
>>>
>>> The OMAP architecture is using the new events for PM debugging, however
>>> the new events are made generic enough to be used by all platforms.
>>
>> Even though I'm blissfully ignorant on power stuffs, the proposed set of
>> trace thingies doesn't appear to even start covering x86..
> Ok, let's discuss it! The intention is to provide new generic events.
>
>>
>> CC'ed people who know more on the subject.
Any news on the topic? Any feedback on the code or on the new events
definition is welcome.
>>
Thanks!