linux-kernel - Re: [PATCH_V2] sched/fair: updates weight of cfs_rq before update_cfs_group() in enqueue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAKfTPtDhXVc4HVeRJzgf9o1n9nOrXM=SgKSru3cynLngFPdxSg@mail.gmail.com>
Date: Tue, 11 Jun 2024 12:59:18 +0200
From: Vincent Guittot <vincent.guittot@...aro.org>
To: "Gaowei.Pu" <pugaowei@...o.com>
Cc: mingo@...hat.com, peterz@...radead.org, juri.lelli@...hat.com, 
	dietmar.eggemann@....com, rostedt@...dmis.org, bsegall@...gle.com, 
	mgorman@...e.de, bristot@...hat.com, vschneid@...hat.com, 
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH_V2] sched/fair: updates weight of cfs_rq before
 update_cfs_group() in enqueue_entity()

On Mon, 3 Jun 2024 at 11:18, Gaowei.Pu <pugaowei@...o.com> wrote:
>
> From: pugaowei <pugaowei@...o.com>
>
> we should update the weight of cfs_rq before update_cfs_group().

update_cfs_group() updates the weight of se that is about to be
enqueued on cfs_rq so the current order looks good for me:

update_cfs_group()
  reweight_entity(cfs_rq, se, shares); //se->on_rq == 0
    update_load_set(&se->load, weight);
      se->load.weight = weight

account_entity_enqueue(cfs_rq, se);
  update_load_add(&cfs_rq->load, se->load.weight);
    cfs_rq->load.weight += se->load.weight

Have you faced some problems in particular ?

> Ensure that we can get accurate shares of the cfs_rq when its
> weights changes. we can find this work was done correctly in
> dequeue_entity(). so fix it.
>
> patch_V1 :
> https://lore.kernel.org/lkml/20240531030833.3375-1-pugaowei@oppo.com/T/#u
> trigger a warnning below because of the changing order of
> account_entity_enqueue().
>
> [ 0.400603][ T0] ? __warn (kernel/panic.c:693)
>
> [ 0.400603][ T0] ? place_entity (kernel/sched/fair.c:5256 (discriminator 1))
>
> [ 0.400603][ T0] ? report_bug (lib/bug.c:180 lib/bug.c:219)
>
> [ 0.400603][ T0] ? handle_bug (arch/x86/kernel/traps.c:239)
>
> [ 0.400603][ T0] ? exc_invalid_op (arch/x86/kernel/traps.c:260 (discriminator 1))
>
> [ 0.400603][ T0] ? asm_exc_invalid_op (arch/x86/include/asm/idtentry.h:621)
>
> [ 0.400603][ T0] ? place_entity (kernel/sched/fair.c:5256 (discriminator 1))
>
> [ 0.400603][ T0] ? place_entity (kernel/sched/fair.c:5182)
>
> [ 0.400603][ T0] enqueue_entity (kernel/sched/fair.c:5328)
>
> [ 0.400603][ T0] enqueue_task_fair (kernel/sched/fair.c:6785)
>
> V2 fix the warnning and keep the lag without inflating it when it is
> the first sched_entity queued on the cfs_rq.
>
> Signed-off-by: pugaowei <pugaowei@...o.com>
> ---
>  kernel/sched/fair.c | 24 ++++++++++++++++--------
>  1 file changed, 16 insertions(+), 8 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 8a5b1ae0aa55..2fb1fbcfdda3 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -5190,12 +5190,12 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
>          *
>          * EEVDF: placement strategy #1 / #2
>          */
> -       if (sched_feat(PLACE_LAG) && cfs_rq->nr_running) {
> +       if (sched_feat(PLACE_LAG)) {
>                 struct sched_entity *curr = cfs_rq->curr;
> -               unsigned long load;
> +               unsigned long load, se_load;
>
>                 lag = se->vlag;
> -
> +               se_load = scale_load_down(se->load.weight);
>                 /*
>                  * If we want to place a task and preserve lag, we have to
>                  * consider the effect of the new entity on the weighted
> @@ -5252,9 +5252,13 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
>                 if (curr && curr->on_rq)
>                         load += scale_load_down(curr->load.weight);
>
> -               lag *= load + scale_load_down(se->load.weight);
> -               if (WARN_ON_ONCE(!load))
> -                       load = 1;
> +               lag *= load + se_load;
> +               /*
> +                * we just need to keep the lag whithout inflating it when the se is
> +                * the first sched_entity queued on cfs_rq.
> +                */
> +               if (!load)
> +                       load = se_load;
>                 lag = div_s64(lag, load);
>         }
>
> @@ -5304,6 +5308,12 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
>          */
>         update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
>         se_update_runnable(se);
> +
> +       /*
> +        * we should update the weight of cfs_rq before update_cfs_group.
> +        * Ensure we can get accurate shares of the cfs_rq when its weights changes.
> +        */
> +       account_entity_enqueue(cfs_rq, se);
>         /*
>          * XXX update_load_avg() above will have attached us to the pelt sum;
>          * but update_cfs_group() here will re-adjust the weight and have to
> @@ -5318,8 +5328,6 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
>         if (!curr)
>                 place_entity(cfs_rq, se, flags);
>
> -       account_entity_enqueue(cfs_rq, se);
> -
>         /* Entity has migrated, no longer consider this task hot */
>         if (flags & ENQUEUE_MIGRATED)
>                 se->exec_start = 0;
> --
> 2.17.1
>