Re: [PATCH v2 02/12] sched/core: uclamp: map TASK's clamp values into CPU's clamp groups

[Patrick Bellasi <patrick.bellasi@arm.com>]

On 20-Jul 17:25, Suren Baghdasaryan wrote:

[...]

> > ---8<---
> >
> > /* Uclamp flags */
> > #define SCHED_FLAG_UTIL_CLAMP_STRICT    0x11 /* Roll-back on failure */
> > #define SCHED_FLAG_UTIL_CLAMP_MIN       0x12 /* Update util_min */
> > #define SCHED_FLAG_UTIL_CLAMP_MAX       0x14 /* Update util_max */
> > #define SCHED_FLAG_UTIL_CLAMP           ( \
> >                 SCHED_FLAG_UTIL_CLAMP_MIN | SCHED_FLAG_UTIL_CLAMP_MAX)
> >
> 
> Having ability to update only min or only max this way might be indeed
> very useful.
> Instead of rolling back on failure I would suggest to check both
> inputs first to make sure there won't be any error before updating.
> This would remove the need for SCHED_FLAG_UTIL_CLAMP_STRICT (which I
> think any user would want to set to 1 anyway).
> Looks like uclamp_group_get() can fail only if uclamp_group_find()
> fails to find a slot for uclamp_value or a free slot. So one way to do
> this search before update is to call uclamp_group_find() for both
> UCLAMP_MIN and UCLAMP_MAX beforehand and if they succeed then pass
> obtained next_group_ids into uclamp_group_get() to avoid doing the
> same search twice. This requires some refactoring of
> uclamp_group_get() but I think the end result would be a cleaner and
> more predictable solution.

Yes, that sound possible... provided we check all the groups under the
same uclamp_mutex, it should be possible to find the group_ids before
actually increasing the refcount.

... will look into this for the next reposting.

-- 
#include <best/regards.h>

Patrick Bellasi