[RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Lin Ming]

Add group scheduling transactional APIs to struct pmu.
These APIs will be implemented in arch code, based on Peter's idea as
below.

> the idea behind hw_perf_group_sched_in() is to not perform
> schedulability tests on each event in the group, but to add the group
as
> a whole and then perform one test.
>
> Of course, when that test fails, you'll have to roll-back the whole
> group again.
>
> So start_txn (or a better name) would simply toggle a flag in the pmu
> implementation that will make pmu::enable() not perform the
> schedulablilty test.
>
> Then commit_txn() will perform the schedulability test (so note the
> method has to have a !void return value, my mistake in the earlier
> email).
>
> This will allow us to use the regular
> kernel/perf_event.c::group_sched_in() and all the rollback code.
> Currently each hw_perf_group_sched_in() implementation duplicates all
> the rolllback code (with various bugs).


Reviewed-by: Stephane Eranian <eranian@google.com>
Reviewed-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
---
 include/linux/perf_event.h |    8 +++++---
 kernel/perf_event.c        |   29 ++++++++++++++++-------------
 2 files changed, 21 insertions(+), 16 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index ace31fb..b16cfba 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -532,6 +532,8 @@ struct hw_perf_event {
 
 struct perf_event;
 
+#define PERF_EVENT_TRAN_STARTED 1
+
 /**
  * struct pmu - generic performance monitoring unit
  */
@@ -542,6 +544,9 @@ struct pmu {
 	void (*stop)			(struct perf_event *event);
 	void (*read)			(struct perf_event *event);
 	void (*unthrottle)		(struct perf_event *event);
+	void (*start_txn)		(const struct pmu *pmu);
+	void (*stop_txn)		(const struct pmu *pmu);
+	int (*commit_txn)		(const struct pmu *pmu);
 };
 
 /**
@@ -807,9 +812,6 @@ extern void perf_disable(void);
 extern void perf_enable(void);
 extern int perf_event_task_disable(void);
 extern int perf_event_task_enable(void);
-extern int hw_perf_group_sched_in(struct perf_event *group_leader,
-	       struct perf_cpu_context *cpuctx,
-	       struct perf_event_context *ctx);
 extern void perf_event_update_userpage(struct perf_event *event);
 extern int perf_event_release_kernel(struct perf_event *event);
 extern struct perf_event *
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 9dbe8cd..ffd2360 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -83,14 +83,6 @@ extern __weak const struct pmu *hw_perf_event_init(struct perf_event *event)
 void __weak hw_perf_disable(void)		{ barrier(); }
 void __weak hw_perf_enable(void)		{ barrier(); }
 
-int __weak
-hw_perf_group_sched_in(struct perf_event *group_leader,
-	       struct perf_cpu_context *cpuctx,
-	       struct perf_event_context *ctx)
-{
-	return 0;
-}
-
 void __weak perf_event_print_debug(void)	{ }
 
 static DEFINE_PER_CPU(int, perf_disable_count);
@@ -641,15 +633,15 @@ group_sched_in(struct perf_event *group_event,
 	       struct perf_cpu_context *cpuctx,
 	       struct perf_event_context *ctx)
 {
-	struct perf_event *event, *partial_group;
+	struct perf_event *event, *partial_group = NULL;
+	const struct pmu *pmu = group_event->pmu;
 	int ret;
 
 	if (group_event->state == PERF_EVENT_STATE_OFF)
 		return 0;
 
-	ret = hw_perf_group_sched_in(group_event, cpuctx, ctx);
-	if (ret)
-		return ret < 0 ? ret : 0;
+	if (pmu->start_txn)
+		pmu->start_txn(pmu);
 
 	if (event_sched_in(group_event, cpuctx, ctx))
 		return -EAGAIN;
@@ -664,9 +656,20 @@ group_sched_in(struct perf_event *group_event,
 		}
 	}
 
-	return 0;
+	if (pmu->commit_txn) {
+		ret = pmu->commit_txn(pmu);
+		if (!ret) {
+			if (pmu->stop_txn)
+				pmu->stop_txn(pmu);
+
+			return 0;
+		}
+	}
 
 group_error:
+	if (pmu->stop_txn)
+		pmu->stop_txn(pmu);
+
 	/*
 	 * Groups can be scheduled in as one unit only, so undo any
 	 * partial group before returning:

Re: [RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Peter Zijlstra]

On Thu, 2010-04-22 at 15:51 +0800, Lin Ming wrote:
> +       if (pmu->commit_txn) {
> +               ret = pmu->commit_txn(pmu);
> +               if (!ret) {
> +                       if (pmu->stop_txn)
> +                               pmu->stop_txn(pmu);
> +
> +                       return 0;
> +               }
> +       } 

I think we can mandate that if one of the _txn methods is available,
they all are, it would be weird otherwise.

Re: [RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Lin Ming]

On Thu, 2010-04-22 at 16:24 +0800, Peter Zijlstra wrote:
> On Thu, 2010-04-22 at 15:51 +0800, Lin Ming wrote:
> > +       if (pmu->commit_txn) {
> > +               ret = pmu->commit_txn(pmu);
> > +               if (!ret) {
> > +                       if (pmu->stop_txn)
> > +                               pmu->stop_txn(pmu);
> > +
> > +                       return 0;
> > +               }
> > +       } 
> 
> I think we can mandate that if one of the _txn methods is available,
> they all are, it would be weird otherwise.
> 

How about below?

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index ffd2360..72ea25c 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -635,12 +635,16 @@ group_sched_in(struct perf_event *group_event,
 {
 	struct perf_event *event, *partial_group = NULL;
 	const struct pmu *pmu = group_event->pmu;
+	bool txn = false;
 	int ret;
 
 	if (group_event->state == PERF_EVENT_STATE_OFF)
 		return 0;
 
 	if (pmu->start_txn)
+		txn = true;
+
+	if (txn)
 		pmu->start_txn(pmu);
 
 	if (event_sched_in(group_event, cpuctx, ctx))
@@ -656,18 +660,17 @@ group_sched_in(struct perf_event *group_event,
 		}
 	}
 
-	if (pmu->commit_txn) {
+	if (txn) {
 		ret = pmu->commit_txn(pmu);
 		if (!ret) {
-			if (pmu->stop_txn)
-				pmu->stop_txn(pmu);
+			pmu->stop_txn(pmu);
 
 			return 0;
 		}
 	}
 
 group_error:
-	if (pmu->stop_txn)
+	if (txn)
 		pmu->stop_txn(pmu);
 
 	/*

Re: [RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Frederic Weisbecker]

On Thu, Apr 22, 2010 at 03:51:02PM +0800, Lin Ming wrote:
> Add group scheduling transactional APIs to struct pmu.
> These APIs will be implemented in arch code, based on Peter's idea as
> below.
> 
> > the idea behind hw_perf_group_sched_in() is to not perform
> > schedulability tests on each event in the group, but to add the group
> as
> > a whole and then perform one test.
> >
> > Of course, when that test fails, you'll have to roll-back the whole
> > group again.
> >
> > So start_txn (or a better name) would simply toggle a flag in the pmu
> > implementation that will make pmu::enable() not perform the
> > schedulablilty test.
> >
> > Then commit_txn() will perform the schedulability test (so note the
> > method has to have a !void return value, my mistake in the earlier
> > email).
> >
> > This will allow us to use the regular
> > kernel/perf_event.c::group_sched_in() and all the rollback code.
> > Currently each hw_perf_group_sched_in() implementation duplicates all
> > the rolllback code (with various bugs).
> 
> 
> Reviewed-by: Stephane Eranian <eranian@google.com>
> Reviewed-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Lin Ming <ming.m.lin@intel.com>
> ---
>  include/linux/perf_event.h |    8 +++++---
>  kernel/perf_event.c        |   29 ++++++++++++++++-------------
>  2 files changed, 21 insertions(+), 16 deletions(-)
> 
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index ace31fb..b16cfba 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -532,6 +532,8 @@ struct hw_perf_event {
>  
>  struct perf_event;
>  
> +#define PERF_EVENT_TRAN_STARTED 1
> +
>  /**
>   * struct pmu - generic performance monitoring unit
>   */
> @@ -542,6 +544,9 @@ struct pmu {
>  	void (*stop)			(struct perf_event *event);
>  	void (*read)			(struct perf_event *event);
>  	void (*unthrottle)		(struct perf_event *event);
> +	void (*start_txn)		(const struct pmu *pmu);
> +	void (*stop_txn)		(const struct pmu *pmu);
> +	int (*commit_txn)		(const struct pmu *pmu);


Please add a few comments that briefly explain what these
*_txn callbacks are supposed to mean.

Unless txn is an acronym that most kernel developers are used to.

Thanks.

Re: [RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Lin Ming]

On Fri, 2010-04-23 at 01:31 +0800, Frederic Weisbecker wrote:
> On Thu, Apr 22, 2010 at 03:51:02PM +0800, Lin Ming wrote:
> > Add group scheduling transactional APIs to struct pmu.
> > These APIs will be implemented in arch code, based on Peter's idea as
> > below.
> > 
> > > the idea behind hw_perf_group_sched_in() is to not perform
> > > schedulability tests on each event in the group, but to add the group
> > as
> > > a whole and then perform one test.
> > >
> > > Of course, when that test fails, you'll have to roll-back the whole
> > > group again.
> > >
> > > So start_txn (or a better name) would simply toggle a flag in the pmu
> > > implementation that will make pmu::enable() not perform the
> > > schedulablilty test.
> > >
> > > Then commit_txn() will perform the schedulability test (so note the
> > > method has to have a !void return value, my mistake in the earlier
> > > email).
> > >
> > > This will allow us to use the regular
> > > kernel/perf_event.c::group_sched_in() and all the rollback code.
> > > Currently each hw_perf_group_sched_in() implementation duplicates all
> > > the rolllback code (with various bugs).
> > 
> > 
> > Reviewed-by: Stephane Eranian <eranian@google.com>
> > Reviewed-by: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Lin Ming <ming.m.lin@intel.com>
> > ---
> >  include/linux/perf_event.h |    8 +++++---
> >  kernel/perf_event.c        |   29 ++++++++++++++++-------------
> >  2 files changed, 21 insertions(+), 16 deletions(-)
> > 
> > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> > index ace31fb..b16cfba 100644
> > --- a/include/linux/perf_event.h
> > +++ b/include/linux/perf_event.h
> > @@ -532,6 +532,8 @@ struct hw_perf_event {
> >  
> >  struct perf_event;
> >  
> > +#define PERF_EVENT_TRAN_STARTED 1
> > +
> >  /**
> >   * struct pmu - generic performance monitoring unit
> >   */
> > @@ -542,6 +544,9 @@ struct pmu {
> >  	void (*stop)			(struct perf_event *event);
> >  	void (*read)			(struct perf_event *event);
> >  	void (*unthrottle)		(struct perf_event *event);
> > +	void (*start_txn)		(const struct pmu *pmu);
> > +	void (*stop_txn)		(const struct pmu *pmu);
> > +	int (*commit_txn)		(const struct pmu *pmu);
> 
> 
> Please add a few comments that briefly explain what these
> *_txn callbacks are supposed to mean.
> 
> Unless txn is an acronym that most kernel developers are used to.

How about below changes?

Thanks for review.

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index b16cfba..bba4c60 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -544,9 +544,21 @@ struct pmu {
 	void (*stop)			(struct perf_event *event);
 	void (*read)			(struct perf_event *event);
 	void (*unthrottle)		(struct perf_event *event);
-	void (*start_txn)		(const struct pmu *pmu);
-	void (*stop_txn)		(const struct pmu *pmu);
-	int (*commit_txn)		(const struct pmu *pmu);
+
+	/*
+	 * group events scheduling is treated as a transaction,
+	 * add group events as a whole and perform one schedulability test.
+	 * If test fails, roll back the whole group
+	 */
+
+	/* start group events transaction  */
+	void (*start_group_trans)	(const struct pmu *pmu);
+
+	/* stop group events transaction  */
+	void (*stop_group_trans)	(const struct pmu *pmu);
+
+	/* commit group events transaction */
+	int (*commit_group_trans)	(const struct pmu *pmu);
 };
 
 /**

Re: [RFC][PATCH 1/4] perf: core, add group scheduling transactional APIs

[Frederic Weisbecker]

On Fri, Apr 23, 2010 at 10:08:25AM +0800, Lin Ming wrote:
> On Fri, 2010-04-23 at 01:31 +0800, Frederic Weisbecker wrote:
> > On Thu, Apr 22, 2010 at 03:51:02PM +0800, Lin Ming wrote:
> > > Add group scheduling transactional APIs to struct pmu.
> > > These APIs will be implemented in arch code, based on Peter's idea as
> > > below.
> > > 
> > > > the idea behind hw_perf_group_sched_in() is to not perform
> > > > schedulability tests on each event in the group, but to add the group
> > > as
> > > > a whole and then perform one test.
> > > >
> > > > Of course, when that test fails, you'll have to roll-back the whole
> > > > group again.
> > > >
> > > > So start_txn (or a better name) would simply toggle a flag in the pmu
> > > > implementation that will make pmu::enable() not perform the
> > > > schedulablilty test.
> > > >
> > > > Then commit_txn() will perform the schedulability test (so note the
> > > > method has to have a !void return value, my mistake in the earlier
> > > > email).
> > > >
> > > > This will allow us to use the regular
> > > > kernel/perf_event.c::group_sched_in() and all the rollback code.
> > > > Currently each hw_perf_group_sched_in() implementation duplicates all
> > > > the rolllback code (with various bugs).
> > > 
> > > 
> > > Reviewed-by: Stephane Eranian <eranian@google.com>
> > > Reviewed-by: Peter Zijlstra <peterz@infradead.org>
> > > Signed-off-by: Lin Ming <ming.m.lin@intel.com>
> > > ---
> > >  include/linux/perf_event.h |    8 +++++---
> > >  kernel/perf_event.c        |   29 ++++++++++++++++-------------
> > >  2 files changed, 21 insertions(+), 16 deletions(-)
> > > 
> > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> > > index ace31fb..b16cfba 100644
> > > --- a/include/linux/perf_event.h
> > > +++ b/include/linux/perf_event.h
> > > @@ -532,6 +532,8 @@ struct hw_perf_event {
> > >  
> > >  struct perf_event;
> > >  
> > > +#define PERF_EVENT_TRAN_STARTED 1
> > > +
> > >  /**
> > >   * struct pmu - generic performance monitoring unit
> > >   */
> > > @@ -542,6 +544,9 @@ struct pmu {
> > >  	void (*stop)			(struct perf_event *event);
> > >  	void (*read)			(struct perf_event *event);
> > >  	void (*unthrottle)		(struct perf_event *event);
> > > +	void (*start_txn)		(const struct pmu *pmu);
> > > +	void (*stop_txn)		(const struct pmu *pmu);
> > > +	int (*commit_txn)		(const struct pmu *pmu);
> > 
> > 
> > Please add a few comments that briefly explain what these
> > *_txn callbacks are supposed to mean.
> > 
> > Unless txn is an acronym that most kernel developers are used to.
> 
> How about below changes?
> 
> Thanks for review.
> 
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index b16cfba..bba4c60 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -544,9 +544,21 @@ struct pmu {
>  	void (*stop)			(struct perf_event *event);
>  	void (*read)			(struct perf_event *event);
>  	void (*unthrottle)		(struct perf_event *event);
> -	void (*start_txn)		(const struct pmu *pmu);
> -	void (*stop_txn)		(const struct pmu *pmu);
> -	int (*commit_txn)		(const struct pmu *pmu);
> +
> +	/*
> +	 * group events scheduling is treated as a transaction,
> +	 * add group events as a whole and perform one schedulability test.
> +	 * If test fails, roll back the whole group
> +	 */
> +
> +	/* start group events transaction  */
> +	void (*start_group_trans)	(const struct pmu *pmu);
> +
> +	/* stop group events transaction  */
> +	void (*stop_group_trans)	(const struct pmu *pmu);
> +
> +	/* commit group events transaction */
> +	int (*commit_group_trans)	(const struct pmu *pmu);
>  };


Looks good!

Thanks.