Util-clamp places tasks in different buckets based on their clamp values
for performance reasons. However, the size of buckets is currently
computed using a rounding division, which can lead to an off-by-one
error in some configurations.
For instance, with 20 buckets, the bucket size will be 1024/20=51.2,
rounded to the closest value: 51. Now, a task with a clamp of 1024 (as
is the default for the min clamp of RT tasks) will be mapped to bucket
id 1024/51=20 as we're now using a standard integer division. Sadly,
correct indexes are in range [0,19], hence leading to an out of bound
memory access.
Fix this by using a rounding-up division when computing the bucket size.
Fixes: 69842cba9ace ("sched/uclamp: Add CPU's clamp buckets refcounting")
Suggested-by: Qais Yousef <[email protected]>
Signed-off-by: Quentin Perret <[email protected]>
---
This was found thanks to the SCHED_WARN_ON() in uclamp_rq_dec_id() which
indicated a broken state while running with 20 buckets on Android.
Big thanks to Qais for the help with this one.
---
kernel/sched/core.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 98191218d891..ec175909e8b0 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -920,8 +920,7 @@ static struct uclamp_se uclamp_default[UCLAMP_CNT];
*/
DEFINE_STATIC_KEY_FALSE(sched_uclamp_used);
-/* Integer rounded range for each bucket */
-#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
+#define UCLAMP_BUCKET_DELTA DIV_ROUND_UP(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
#define for_each_clamp_id(clamp_id) \
for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
--
2.31.1.498.g6c1eba8ee3d-goog
On 28/04/2021 19:27, Quentin Perret wrote:
> Util-clamp places tasks in different buckets based on their clamp values
> for performance reasons. However, the size of buckets is currently
> computed using a rounding division, which can lead to an off-by-one
> error in some configurations.
>
> For instance, with 20 buckets, the bucket size will be 1024/20=51.2,
> rounded to the closest value: 51. Now, a task with a clamp of 1024 (as
> is the default for the min clamp of RT tasks) will be mapped to bucket
> id 1024/51=20 as we're now using a standard integer division. Sadly,
> correct indexes are in range [0,19], hence leading to an out of bound
> memory access.
>
> Fix this by using a rounding-up division when computing the bucket size.
But in case you use e.g. 16 buckets, wouldn't you still end up with this
task mapped into bucket_id=16?
1024/16=64
1024/64=16
>
> Fixes: 69842cba9ace ("sched/uclamp: Add CPU's clamp buckets refcounting")
> Suggested-by: Qais Yousef <[email protected]>
> Signed-off-by: Quentin Perret <[email protected]>
>
> ---
>
> This was found thanks to the SCHED_WARN_ON() in uclamp_rq_dec_id() which
> indicated a broken state while running with 20 buckets on Android.
>
> Big thanks to Qais for the help with this one.
> ---
> kernel/sched/core.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 98191218d891..ec175909e8b0 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -920,8 +920,7 @@ static struct uclamp_se uclamp_default[UCLAMP_CNT];
> */
> DEFINE_STATIC_KEY_FALSE(sched_uclamp_used);
>
> -/* Integer rounded range for each bucket */
> -#define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
> +#define UCLAMP_BUCKET_DELTA DIV_ROUND_UP(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS)
>
> #define for_each_clamp_id(clamp_id) \
> for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++)
>
On Thursday 29 Apr 2021 at 14:34:14 (+0200), Dietmar Eggemann wrote:
> On 28/04/2021 19:27, Quentin Perret wrote:
> > Util-clamp places tasks in different buckets based on their clamp values
> > for performance reasons. However, the size of buckets is currently
> > computed using a rounding division, which can lead to an off-by-one
> > error in some configurations.
> >
> > For instance, with 20 buckets, the bucket size will be 1024/20=51.2,
> > rounded to the closest value: 51. Now, a task with a clamp of 1024 (as
> > is the default for the min clamp of RT tasks) will be mapped to bucket
> > id 1024/51=20 as we're now using a standard integer division. Sadly,
> > correct indexes are in range [0,19], hence leading to an out of bound
> > memory access.
> >
> > Fix this by using a rounding-up division when computing the bucket size.
>
> But in case you use e.g. 16 buckets, wouldn't you still end up with this
> task mapped into bucket_id=16?
>
> 1024/16=64
>
> 1024/64=16
Hrmpf, you're right ...
So I guess the following will do:
#define UCLAMP_BUCKET_DELTA (SCHED_CAPACITY_SCALE / UCLAMP_BUCKETS + 1)
Thanks,
Quentin