The UCLAMP_FLAG_IDLE flag is set on a runqueue when dequeueing the last
active task to maintain the last uclamp.max and prevent blocked util
from suddenly becoming visible.
However, there is an asymmetry in how the flag is set and cleared which
can lead to having the flag set whilst there are active task on the rq.
Specifically, the flag is set in the uclamp_rq_inc() path, which is
called at enqueue time, but cleared in the uclamp_rq_dec_id() which is
called both when dequeueing and task _and_ during cgroup migrations.
Fix this by setting the flag in the uclamp_rq_inc_id() path to ensure
things remain symmetrical.
Reported-by: Rick Yiu <[email protected]>
Signed-off-by: Quentin Perret <[email protected]>
---
kernel/sched/core.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 5226cc26a095..3b213402798e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -980,6 +980,7 @@ static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id,
if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE))
return;
+ rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value);
}
@@ -1252,10 +1253,6 @@ static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p)
for_each_clamp_id(clamp_id)
uclamp_rq_inc_id(rq, p, clamp_id);
-
- /* Reset clamp idle holding when there is one RUNNABLE task */
- if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
- rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
}
static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p)
--
2.32.0.272.g935e593368-goog
On Wednesday 09 Jun 2021 at 14:33:39 (+0000), Quentin Perret wrote:
> The UCLAMP_FLAG_IDLE flag is set on a runqueue when dequeueing the last
> active task to maintain the last uclamp.max and prevent blocked util
> from suddenly becoming visible.
>
> However, there is an asymmetry in how the flag is set and cleared which
> can lead to having the flag set whilst there are active task on the rq.
> Specifically, the flag is set in the uclamp_rq_inc() path, which is
> called at enqueue time, but cleared in the uclamp_rq_dec_id() which is
> called both when dequeueing and task _and_ during cgroup migrations.
>
> Fix this by setting the flag in the uclamp_rq_inc_id() path to ensure
> things remain symmetrical.
>
> Reported-by: Rick Yiu <[email protected]>
> Signed-off-by: Quentin Perret <[email protected]>
Argh, and I think this wants
Fixes: e496187da710 ("sched/uclamp: Enforce last task's UCLAMP_MAX")
On 06/09/21 14:33, Quentin Perret wrote:
> The UCLAMP_FLAG_IDLE flag is set on a runqueue when dequeueing the last
> active task to maintain the last uclamp.max and prevent blocked util
> from suddenly becoming visible.
>
> However, there is an asymmetry in how the flag is set and cleared which
> can lead to having the flag set whilst there are active task on the rq.
> Specifically, the flag is set in the uclamp_rq_inc() path, which is
> called at enqueue time, but cleared in the uclamp_rq_dec_id() which is
It is actually the other way around. It is cleared in uclamp_rq_inc() and
set in uclamp_rq_dec_id().
> called both when dequeueing and task _and_ during cgroup migrations.
Is it cgroup migrations or when cgroup uclamp values are updated you mean?
It is worth being direct and mention that uclamp_update_active() will perform
uclamp_rq_dec/inc_id() when the cgroup uclamp values are updated, which would
end up with the flag set but not cleared in this path.
>
> Fix this by setting the flag in the uclamp_rq_inc_id() path to ensure
> things remain symmetrical.
>
> Reported-by: Rick Yiu <[email protected]>
> Signed-off-by: Quentin Perret <[email protected]>
> ---
With the commit message fixed.
Reviewed-by: Qais Yousef <[email protected]>
Thanks!
--
Qais Yousef
> kernel/sched/core.c | 5 +----
> 1 file changed, 1 insertion(+), 4 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 5226cc26a095..3b213402798e 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -980,6 +980,7 @@ static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id,
> if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE))
> return;
>
> + rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
> WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value);
> }
>
> @@ -1252,10 +1253,6 @@ static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p)
>
> for_each_clamp_id(clamp_id)
> uclamp_rq_inc_id(rq, p, clamp_id);
> -
> - /* Reset clamp idle holding when there is one RUNNABLE task */
> - if (rq->uclamp_flags & UCLAMP_FLAG_IDLE)
> - rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE;
> }
>
> static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p)
> --
> 2.32.0.272.g935e593368-goog
>
On Thursday 10 Jun 2021 at 11:00:39 (+0100), Qais Yousef wrote:
> On 06/09/21 14:33, Quentin Perret wrote:
> > The UCLAMP_FLAG_IDLE flag is set on a runqueue when dequeueing the last
> > active task to maintain the last uclamp.max and prevent blocked util
> > from suddenly becoming visible.
> >
> > However, there is an asymmetry in how the flag is set and cleared which
> > can lead to having the flag set whilst there are active task on the rq.
> > Specifically, the flag is set in the uclamp_rq_inc() path, which is
> > called at enqueue time, but cleared in the uclamp_rq_dec_id() which is
>
> It is actually the other way around. It is cleared in uclamp_rq_inc() and
> set in uclamp_rq_dec_id().
>
> > called both when dequeueing and task _and_ during cgroup migrations.
>
> Is it cgroup migrations or when cgroup uclamp values are updated you mean?
>
> It is worth being direct and mention that uclamp_update_active() will perform
> uclamp_rq_dec/inc_id() when the cgroup uclamp values are updated, which would
> end up with the flag set but not cleared in this path.
Yep, that's the problematic path.
> >
> > Fix this by setting the flag in the uclamp_rq_inc_id() path to ensure
> > things remain symmetrical.
> >
> > Reported-by: Rick Yiu <[email protected]>
> > Signed-off-by: Quentin Perret <[email protected]>
> > ---
>
> With the commit message fixed.
Ack, this commit message is a mess, I'll rewrite.
> Reviewed-by: Qais Yousef <[email protected]>
Thanks!
Quentin