Per the codes, this pair of rcu_read_{,un}lock() protects nothing.
Let's remove it.
Further, the variable "pmu" is safe as it is in the SRCU read-side
critical scope of "pmus_srcu" and perf_pmu_unregister() delete it
after calling synchronize_srcu.
Signed-off-by: Yanfei Xu <[email protected]>
---
kernel/events/core.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 950b25c3f210..36b0df6feab4 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -11296,9 +11296,7 @@ static struct pmu *perf_init_event(struct perf_event *event)
}
again:
- rcu_read_lock();
pmu = idr_find(&pmu_idr, type);
- rcu_read_unlock();
if (pmu) {
if (event->attr.type != type && type != PERF_TYPE_RAW &&
!(pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE))
--
2.32.0
On Wed, Jun 08, 2022 at 05:09:38PM +0800, Yanfei Xu wrote:
> Per the codes, this pair of rcu_read_{,un}lock() protects nothing.
> Let's remove it.
>
> Further, the variable "pmu" is safe as it is in the SRCU read-side
> critical scope of "pmus_srcu" and perf_pmu_unregister() delete it
> after calling synchronize_srcu.
>
> Signed-off-by: Yanfei Xu <[email protected]>
> ---
> kernel/events/core.c | 2 --
> 1 file changed, 2 deletions(-)
>
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 950b25c3f210..36b0df6feab4 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -11296,9 +11296,7 @@ static struct pmu *perf_init_event(struct perf_event *event)
> }
>
> again:
> - rcu_read_lock();
> pmu = idr_find(&pmu_idr, type);
> - rcu_read_unlock();
You're mistaken, this is required for the radix tree internal nodes.
Without it we can't safely traverse the radix tree without full
serialization against the modifiers.
> -----Original Message-----
> From: Peter Zijlstra <[email protected]>
> Sent: Wednesday, June 8, 2022 5:43 PM
> To: Xu, Yanfei <[email protected]>
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: Re: [PATCH] perf/core: remove a meaningless pair of
> rcu_read_{,un}lock()
>
> On Wed, Jun 08, 2022 at 05:09:38PM +0800, Yanfei Xu wrote:
> > Per the codes, this pair of rcu_read_{,un}lock() protects nothing.
> > Let's remove it.
> >
> > Further, the variable "pmu" is safe as it is in the SRCU read-side
> > critical scope of "pmus_srcu" and perf_pmu_unregister() delete it
> > after calling synchronize_srcu.
> >
> > Signed-off-by: Yanfei Xu <[email protected]>
> > ---
> > kernel/events/core.c | 2 --
> > 1 file changed, 2 deletions(-)
> >
> > diff --git a/kernel/events/core.c b/kernel/events/core.c index
> > 950b25c3f210..36b0df6feab4 100644
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -11296,9 +11296,7 @@ static struct pmu *perf_init_event(struct
> perf_event *event)
> > }
> >
> > again:
> > - rcu_read_lock();
> > pmu = idr_find(&pmu_idr, type);
> > - rcu_read_unlock();
>
> You're mistaken, this is required for the radix tree internal nodes.
> Without it we can't safely traverse the radix tree without full serialization
> against the modifiers.
Thanks your reminder! And after digging and learning these commits: f9c46d6ea5ce ("idr: make idr_find rcu-safe ") and 452a68d0ef34("KVM: hyperv: idr_find needs RCU protection "). Now I understand it.
Thanks,
Yanfei