Re: [PATCH v8 6/9] sched/fair: Add sched group latency support

[Patrick Bellasi <patrick.bellasi@matbug.net>]

Hi Vincent,

On 10-Nov 18:50, Vincent Guittot wrote:

[...]

> diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
> index be4a77baf784..a4866cd4e58c 100644
> --- a/Documentation/admin-guide/cgroup-v2.rst
> +++ b/Documentation/admin-guide/cgroup-v2.rst
> @@ -1095,6 +1095,16 @@ All time durations are in microseconds.
>          values similar to the sched_setattr(2). This maximum utilization
>          value is used to clamp the task specific maximum utilization clamp.
>  
> +  cpu.latency.nice
> +	A read-write single value file which exists on non-root
> +	cgroups.  The default is "0".
> +
> +	The nice value is in the range [-20, 19].
> +
> +	This interface file allows reading and setting latency using the
> +	same values used by sched_setattr(2). The latency_nice of a group is
> + used to limit the impact of the latency_nice of a task outside the
> + group.

This control model is not clear to me.

It does not seem matching what we have for uclamp, where the cgroup values are
used to restrict how much a task can ask or give (in terms of latency here).

in the uclamp's requested-vs-effective values model:

A) a task can "request" (or give up) latency as much as it likes

B) the cgroup in which the task is in any moment limits wthe maximum
   latency a task can "request" (or give up)

C) the system wide knob set the "request" limit for the root cgroup an any task
   not in a cgroup.

This model seems to be what we should use here too.

IOW, for each task compute an "effective" latency_nice value which is defined
starting for a task "requested" latency value and by restricting this value
based on the (B) cgroup value and the (C) system wide value.

That's what we do in uclamp_eff_get():
   https://elixir.bootlin.com/linux/v6.0/source/kernel/sched/core.c#L1484

Why such a model cannot be used for latency_nice too?
Am I missing something?


Best,
patrick

-- 
#include <best/regards.h>

Patrick Bellasi