[v3] perf: Convert mmap() related reference counts to refcount_t

[PATCH v3 13/15] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by Peter Zijlstra 6 months ago

Mostly just re-indent noise.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 kernel/events/core.c |   32 ++++++++++++++------------------
 1 file changed, 14 insertions(+), 18 deletions(-)

--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -7143,27 +7143,23 @@ static int perf_mmap(struct file *file,
 	if (vma_size != PAGE_SIZE * nr_pages)
 		return -EINVAL;
 
-	mutex_lock(&event->mmap_mutex);
-	ret = -EINVAL;
+	scoped_guard (mutex, &event->mmap_mutex) {
+		/*
+		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
+		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
+		 * will detach the rb created here.
+		 */
+		if (event->state <= PERF_EVENT_STATE_REVOKED) {
+			ret = -ENODEV;
+			break;
+		}
 
-	/*
-	 * This relies on __pmu_detach_event() taking mmap_mutex after marking
-	 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
-	 * will detach the rb created here.
-	 */
-	if (event->state <= PERF_EVENT_STATE_REVOKED) {
-		ret = -ENODEV;
-		goto unlock;
+		if (vma->vm_pgoff == 0)
+			ret = perf_mmap_rb(vma, event, nr_pages);
+		else
+			ret = perf_mmap_aux(vma, event, nr_pages);
 	}
 
-	if (vma->vm_pgoff == 0)
-		ret = perf_mmap_rb(vma, event, nr_pages);
-	else
-		ret = perf_mmap_aux(vma, event, nr_pages);
-
-unlock:
-	mutex_unlock(&event->mmap_mutex);
-
 	if (ret)
 		return ret;

Re: [PATCH v3 13/15] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by Lorenzo Stoakes 5 months, 4 weeks ago

On Tue, Aug 12, 2025 at 12:39:11PM +0200, Peter Zijlstra wrote:
> Mostly just re-indent noise.
>
> Suggested-by: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

LGTM, so:

Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>

> ---
>  kernel/events/core.c |   32 ++++++++++++++------------------
>  1 file changed, 14 insertions(+), 18 deletions(-)
>
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -7143,27 +7143,23 @@ static int perf_mmap(struct file *file,
>  	if (vma_size != PAGE_SIZE * nr_pages)
>  		return -EINVAL;
>
> -	mutex_lock(&event->mmap_mutex);

Yeah getting rid of this lock goto stuff is great... mm is at risk of some
more Lorenzo churn to use this stuff over there... :)

> -	ret = -EINVAL;

You see, I'm getting in a pattern here with comment on prior patch and you
reading my mind and fixing it on next ;)

> +	scoped_guard (mutex, &event->mmap_mutex) {
> +		/*
> +		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
> +		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
> +		 * will detach the rb created here.
> +		 */
> +		if (event->state <= PERF_EVENT_STATE_REVOKED) {
> +			ret = -ENODEV;
> +			break;

I don't absolutely love this break-for-what-is-not-obviously-a-for-loop
formulation (I know scoped_guard in practice _is_ a for loop, but obviously
that's hidden by macro), but I guess hey it's C, and we have to do what we
have to do :)

> +		}
>
> -	/*
> -	 * This relies on __pmu_detach_event() taking mmap_mutex after marking
> -	 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
> -	 * will detach the rb created here.
> -	 */
> -	if (event->state <= PERF_EVENT_STATE_REVOKED) {
> -		ret = -ENODEV;
> -		goto unlock;
> +		if (vma->vm_pgoff == 0)
> +			ret = perf_mmap_rb(vma, event, nr_pages);
> +		else
> +			ret = perf_mmap_aux(vma, event, nr_pages);
>  	}
>
> -	if (vma->vm_pgoff == 0)
> -		ret = perf_mmap_rb(vma, event, nr_pages);
> -	else
> -		ret = perf_mmap_aux(vma, event, nr_pages);
> -
> -unlock:
> -	mutex_unlock(&event->mmap_mutex);
> -
>  	if (ret)
>  		return ret;
>
>
>

Re: [PATCH v3 13/15] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by Peter Zijlstra 5 months, 4 weeks ago

On Wed, Aug 13, 2025 at 07:42:41AM +0100, Lorenzo Stoakes wrote:

> > +	scoped_guard (mutex, &event->mmap_mutex) {
> > +		/*
> > +		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
> > +		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
> > +		 * will detach the rb created here.
> > +		 */
> > +		if (event->state <= PERF_EVENT_STATE_REVOKED) {
> > +			ret = -ENODEV;
> > +			break;
> 
> I don't absolutely love this break-for-what-is-not-obviously-a-for-loop
> formulation (I know scoped_guard in practice _is_ a for loop, but obviously
> that's hidden by macro), but I guess hey it's C, and we have to do what we
> have to do :)

Right, don't love it either, but the alternative was a goto and that's
arguably worse, so meh.

> > +		}
> >
> > +		if (vma->vm_pgoff == 0)
> > +			ret = perf_mmap_rb(vma, event, nr_pages);
> > +		else
> > +			ret = perf_mmap_aux(vma, event, nr_pages);
> >  	}
> >
> >  	if (ret)
> >  		return ret;
> >
> >
> >

Re: [PATCH v3 13/15] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by Peter Zijlstra 5 months, 4 weeks ago

On Wed, Aug 13, 2025 at 10:32:44AM +0200, Peter Zijlstra wrote:
> On Wed, Aug 13, 2025 at 07:42:41AM +0100, Lorenzo Stoakes wrote:
> 
> > > +	scoped_guard (mutex, &event->mmap_mutex) {
> > > +		/*
> > > +		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
> > > +		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
> > > +		 * will detach the rb created here.
> > > +		 */
> > > +		if (event->state <= PERF_EVENT_STATE_REVOKED) {
> > > +			ret = -ENODEV;
> > > +			break;
> > 
> > I don't absolutely love this break-for-what-is-not-obviously-a-for-loop
> > formulation (I know scoped_guard in practice _is_ a for loop, but obviously
> > that's hidden by macro), but I guess hey it's C, and we have to do what we
> > have to do :)
> 
> Right, don't love it either, but the alternative was a goto and that's
> arguably worse, so meh.
> 
> > > +		}
> > >
> > > +		if (vma->vm_pgoff == 0)
> > > +			ret = perf_mmap_rb(vma, event, nr_pages);
> > > +		else
> > > +			ret = perf_mmap_aux(vma, event, nr_pages);
> > >  	}
> > >
> > >  	if (ret)
> > >  		return ret;
> > >

Nah, I'm an idiot.. How's this?

---

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 20189a3354f2..4b82f8ed6b4e 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -7151,20 +7151,17 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
 		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
 		 * will detach the rb created here.
 		 */
-		if (event->state <= PERF_EVENT_STATE_REVOKED) {
-			ret = -ENODEV;
-			break;
-		}
+		if (event->state <= PERF_EVENT_STATE_REVOKED)
+			return -ENODEV;
 
 		if (vma->vm_pgoff == 0)
 			ret = perf_mmap_rb(vma, event, nr_pages);
 		else
 			ret = perf_mmap_aux(vma, event, nr_pages);
+		if (ret)
+			return ret;
 	}
 
-	if (ret)
-		return ret;
-
 	/*
 	 * Since pinned accounting is per vm we cannot allow fork() to copy our
 	 * vma.

Re: [PATCH v3 13/15] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by Lorenzo Stoakes 5 months, 4 weeks ago

On Wed, Aug 13, 2025 at 10:44:48AM +0200, Peter Zijlstra wrote:
> On Wed, Aug 13, 2025 at 10:32:44AM +0200, Peter Zijlstra wrote:
> > On Wed, Aug 13, 2025 at 07:42:41AM +0100, Lorenzo Stoakes wrote:
> >
> > > > +	scoped_guard (mutex, &event->mmap_mutex) {
> > > > +		/*
> > > > +		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
> > > > +		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
> > > > +		 * will detach the rb created here.
> > > > +		 */
> > > > +		if (event->state <= PERF_EVENT_STATE_REVOKED) {
> > > > +			ret = -ENODEV;
> > > > +			break;
> > >
> > > I don't absolutely love this break-for-what-is-not-obviously-a-for-loop
> > > formulation (I know scoped_guard in practice _is_ a for loop, but obviously
> > > that's hidden by macro), but I guess hey it's C, and we have to do what we
> > > have to do :)
> >
> > Right, don't love it either, but the alternative was a goto and that's
> > arguably worse, so meh.
> >
> > > > +		}
> > > >
> > > > +		if (vma->vm_pgoff == 0)
> > > > +			ret = perf_mmap_rb(vma, event, nr_pages);
> > > > +		else
> > > > +			ret = perf_mmap_aux(vma, event, nr_pages);
> > > >  	}
> > > >
> > > >  	if (ret)
> > > >  		return ret;
> > > >
>
> Nah, I'm an idiot.. How's this?

Right yeah, you could just return instead of break :))

[tip: perf/core] perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Posted by tip-bot2 for Peter Zijlstra 5 months, 3 weeks ago

The following commit has been merged into the perf/core branch of tip:

Commit-ID:     d23a6dbc0a71741eb7b141fdc04e31360fba46ef
Gitweb:        https://git.kernel.org/tip/d23a6dbc0a71741eb7b141fdc04e31360fba46ef
Author:        Peter Zijlstra <peterz@infradead.org>
AuthorDate:    Tue, 12 Aug 2025 12:39:11 +02:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Fri, 15 Aug 2025 13:13:01 +02:00

perf: Use scoped_guard() for mmap_mutex in perf_mmap()

Mostly just re-indent noise.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Link: https://lore.kernel.org/r/20250812104019.838047976@infradead.org
---
 kernel/events/core.c | 35 ++++++++++++++---------------------
 1 file changed, 14 insertions(+), 21 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 3a5fd2b..41941df 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -7146,30 +7146,23 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
 	if (vma_size != PAGE_SIZE * nr_pages)
 		return -EINVAL;
 
-	mutex_lock(&event->mmap_mutex);
-	ret = -EINVAL;
+	scoped_guard (mutex, &event->mmap_mutex) {
+		/*
+		 * This relies on __pmu_detach_event() taking mmap_mutex after marking
+		 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
+		 * will detach the rb created here.
+		 */
+		if (event->state <= PERF_EVENT_STATE_REVOKED)
+			return -ENODEV;
 
-	/*
-	 * This relies on __pmu_detach_event() taking mmap_mutex after marking
-	 * the event REVOKED. Either we observe the state, or __pmu_detach_event()
-	 * will detach the rb created here.
-	 */
-	if (event->state <= PERF_EVENT_STATE_REVOKED) {
-		ret = -ENODEV;
-		goto unlock;
+		if (vma->vm_pgoff == 0)
+			ret = perf_mmap_rb(vma, event, nr_pages);
+		else
+			ret = perf_mmap_aux(vma, event, nr_pages);
+		if (ret)
+			return ret;
 	}
 
-	if (vma->vm_pgoff == 0)
-		ret = perf_mmap_rb(vma, event, nr_pages);
-	else
-		ret = perf_mmap_aux(vma, event, nr_pages);
-
-unlock:
-	mutex_unlock(&event->mmap_mutex);
-
-	if (ret)
-		return ret;
-
 	/*
 	 * Since pinned accounting is per vm we cannot allow fork() to copy our
 	 * vma.