| 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
| ||
|
Date: Tue, 31 May 2022 10:16:05 +0200
From: Dietmar Eggemann <dietmar.eggemann@....com>
To: Vincent Donnefort <vdonnefort@...gle.com>, peterz@...radead.org,
mingo@...hat.com, vincent.guittot@...aro.org
Cc: linux-kernel@...r.kernel.org, morten.rasmussen@....com,
chris.redpath@....com, qperret@...gle.com, tao.zhou@...ux.dev,
kernel-team@...roid.com
Subject: Re: [PATCH v9 2/7] sched/fair: Decay task PELT values during wakeup
migration
- Vincent Donnefort <vincent.donnefort@....com>
On 23/05/2022 17:51, Vincent Donnefort wrote:
> From: Vincent Donnefort <vincent.donnefort@....com>
[...]
> [1] https://lore.kernel.org/all/20190709115759.10451-1-chris.redpath@arm.com/
minor:
I get `WARNING: Possible unwrapped commit description (prefer a maximum
75 chars per line)`. If you use
https://lkml.kernel.org/r/20190709115759.10451-1-chris.redpath@arm.com
this warning disappears.
[...]
> +static inline void migrate_se_pelt_lag(struct sched_entity *se)
> +{
> + u64 throttled = 0, now, lut;
> + struct cfs_rq *cfs_rq;
> + struct rq *rq;
> + bool is_idle;
> +
> + if (load_avg_is_decayed(&se->avg))
> + return;
> +
> + cfs_rq = cfs_rq_of(se);
> + rq = rq_of(cfs_rq);
> +
> + rcu_read_lock();
> + is_idle = is_idle_task(rcu_dereference(rq->curr));
> + rcu_read_unlock();
> +
> + /*
> + * The lag estimation comes with a cost we don't want to pay all the
> + * time. Hence, limiting to the case where the source CPU is idle and
> + * we know we are at the greatest risk to have an outdated clock.
> + */
> + if (!is_idle)
> + return;
> +
> + /*
> + * Estimated "now" is: last_update_time + cfs_idle_lag + rq_idle_lag, where:
> + *
> + * last_update_time (the cfs_rq's last_update_time)
> + * = cfs_rq_clock_pelt()
> + * = rq_clock_pelt() - cfs->throttled_clock_pelt_time
So this line is always:
= rq_clock_pelt()@cfs_rq_idle -
cfs->throttled_clock_pelt_time@..._rq_idle
since we only execute this code when idle. Which then IMHO explains (1)
better.
> + *
> + * cfs_idle_lag (delta between cfs_rq's update and rq's update)
> + * = rq_clock_pelt()@rq_idle - rq_clock_pelt()@cfs_rq_idle
> + *
> + * rq_idle_lag (delta between rq's update and now)
> + * = sched_clock_cpu() - rq_clock()@rq_idle
> + *
> + * The rq_clock_pelt() from last_update_time being the same as
> + * rq_clock_pelt()@cfs_rq_idle, we can write:
--> (1) ^^^
> + *
> + * now = rq_clock_pelt()@rq_idle - cfs->throttled_clock_pelt_time +
> + * sched_clock_cpu() - rq_clock()@rq_idle
> + * Where:
> + * rq_clock_pelt()@rq_idle is rq->clock_pelt_idle
> + * rq_clock()@rq_idle is rq->enter_idle
> + * cfs->throttled_clock_pelt_time is cfs_rq->throttled_pelt_idle
To understand this better:
cfs->throttled_clock_pelt_time@..._rq_idle is
cfs_rq->throttled_pelt_idle
[...]
> + /*
> + * Paired with _update_idle_rq_clock_pelt. It ensures at the worst case
minor:
s/_update_idle_rq_clock_pelt/_update_idle_rq_clock_pelt()
> + * is observed the old clock_pelt_idle value and the new enter_idle,
> + * which lead to an understimation. The opposite would lead to an
s/understimation/underestimation
[...]
> @@ -8114,6 +8212,10 @@ static bool __update_blocked_fair(struct rq *rq, bool *done)
> if (update_cfs_rq_load_avg(cfs_rq_clock_pelt(cfs_rq), cfs_rq)) {
> update_tg_load_avg(cfs_rq);
>
> + /* sync clock_pelt_idle with last update */
update_idle_cfs_rq_clock_pelt() syncs cfs_rq->throttled_pelt_idle with
cfs_rq->throttled_clock_pelt_time. Not sure what `clock_pelt_idle` and
`last update` here mean?
[...]
> +/* The rq is idle, we can sync to clock_task */
> +static inline void _update_idle_rq_clock_pelt(struct rq *rq)
> +{
> + rq->clock_pelt = rq_clock_task(rq);
> +
> + u64_u32_store(rq->enter_idle, rq_clock(rq));
> + /* Paired with smp_rmb in migrate_se_pelt_lag */
minor:
s/migrate_se_pelt_lag/migrate_se_pelt_lag()
[...]
> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index bf4a0ec98678..97bc26e5c8af 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -648,6 +648,10 @@ struct cfs_rq {
> int runtime_enabled;
> s64 runtime_remaining;
>
> + u64 throttled_pelt_idle;
> +#ifndef CONFIG_64BIT
> + u64 throttled_pelt_idle_copy;
> +#endif
> u64 throttled_clock;
> u64 throttled_clock_pelt;
> u64 throttled_clock_pelt_time;
> @@ -1020,6 +1024,12 @@ struct rq {
> u64 clock_task ____cacheline_aligned;
> u64 clock_pelt;
> unsigned long lost_idle_time;
> + u64 clock_pelt_idle;
> + u64 enter_idle;
> +#ifndef CONFIG_64BIT
> + u64 clock_pelt_idle_copy;
> + u64 enter_idle_copy;
> +#endif
>
> atomic_t nr_iowait;
`throttled_pelt_idle`, `clock_pelt_idle` and `enter_idle` are clock
snapshots when cfs_rq resp. rq go idle. But the naming does not really
show this relation. And this makes reading those equations rather difficult.
What about something like `throttled_clock_pelt_time_enter_idle`,
`clock_pelt_enter_idle`, `clock_enter_idle`? Especially the first one is
too long but something which shows that those are clock snapshots when
enter idle would IMHO augment readability in migrate_se_pelt_lag().
Besides these small issues:
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@....com>
Powered by blists - more mailing lists