linux-kernel - Re: [PATCH v3] sched/fair: Sync se's load_avg with cfs_rq in reweight

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: <da5b5683-ed22-4e65-9e89-e22ff7f6cf2b@arm.com>
Date: Tue, 23 Jul 2024 23:10:56 +0200
From: Dietmar Eggemann <dietmar.eggemann@....com>
To: Vishal Chourasia <vishalc@...ux.ibm.com>,
 Chuyi Zhou <zhouchuyi@...edance.com>
Cc: mingo@...hat.com, peterz@...radead.org, juri.lelli@...hat.com,
 vincent.guittot@...aro.org, rostedt@...dmis.org, bsegall@...gle.com,
 mgorman@...e.de, vschneid@...hat.com, chengming.zhou@...ux.dev,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3] sched/fair: Sync se's load_avg with cfs_rq in
 reweight_task

On 23/07/2024 17:48, Vishal Chourasia wrote:
> On Tue, Jul 23, 2024 at 07:42:47PM +0800, Chuyi Zhou wrote:
>> In reweight_task(), there are two situations:
>>
>> 1. The task was on_rq, then the task's load_avg is accurate because we
>> synchronized it with cfs_rq through update_load_avg() in dequeue_task().
>>
>> 2. The task is sleeping, its load_avg might not have been updated for some
>> time, which can result in inaccurate dequeue_load_avg() in
>> reweight_entity().
>>
>> This patch solves this by using sync_entity_load_avg() to synchronize the
>> load_avg of se with cfs_rq before dequeue_load_avg() in reweight_entity().
>> For tasks were on_rq, since we already update load_avg to accurate values
>> in dequeue_task(), this change will not have other effects due to the short
>> time interval between the two updates.
>>
>> Suggested-by: Dietmar Eggemann <dietmar.eggemann@....com>
>> Signed-off-by: Chuyi Zhou <zhouchuyi@...edance.com>
>> ---
>> Changes in v3:
>> - use sync_entity_load_avg() rather than update_load_avg() to sync the
>> sleeping task with its cfs_rq suggested by Dietmar.
>> - Link t0 v2: https://lore.kernel.org/lkml/20240720051248.59608-1-zhouchuyi@bytedance.com/
>> Changes in v2:
>> - change the description in commit log.
>> - use update_load_avg() in reweight_task() rather than in reweight_entity
>> suggested by chengming.
>> - Link to v1: https://lore.kernel.org/lkml/20240716150840.23061-1-zhouchuyi@bytedance.com/
>> ---
>>  kernel/sched/fair.c | 43 ++++++++++++++++++++++++-------------------
>>  1 file changed, 24 insertions(+), 19 deletions(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 9057584ec06d..da3cdd86ab2e 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -3669,11 +3669,32 @@ dequeue_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
>>  	cfs_rq->avg.load_sum = max_t(u32, cfs_rq->avg.load_sum,
>>  					  cfs_rq->avg.load_avg * PELT_MIN_DIVIDER);
>>  }
>> +
>> +static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
>> +{
>> +	return u64_u32_load_copy(cfs_rq->avg.last_update_time,
>> +				 cfs_rq->last_update_time_copy);
>> +}
>> +
>> +/*
>> + * Synchronize entity load avg of dequeued entity without locking
>> + * the previous rq.
>> + */
>> +static void sync_entity_load_avg(struct sched_entity *se)
>> +{
>> +	struct cfs_rq *cfs_rq = cfs_rq_of(se);
>> +	u64 last_update_time;
>> +
>> +	last_update_time = cfs_rq_last_update_time(cfs_rq);
>> +	__update_load_avg_blocked_se(last_update_time, se);
>> +}
>> +
> The difference between using update_load_avg() and
> sync_entity_load_avg() is:
> 1. update_load_avg() uses the updated PELT clock value from the rq
>    structure.  
> 2. sync_entity_load_avg() uses the last updated time of
>    the cfs_rq where the scheduling entity (se) is attached.
> 
> Won't this affect the entity load sync?

Not sure what you mean exactly by entity load sync here.

The task has been sleeping for a long time, i.e. its PELT values haven't
been updated or a long time (its last_update_time (lut) value is pretty
old).

In this meantime the task's cfs_rq has potentially seen other PELT
updates due to PELT updates of other non-sleeping tasks related to this
cfs_rq. I.e. the cfs_rq lut is much more recent.

What we want to do here is to sync the sleeping task with its cfs_rq. If
the task was sleeping for more than 1us (1024ns) and we cross a 1ms PELT
period (1024us) when we use cfs_rq's lut as the 'now' value for
__update_load_avg_blocked_se() then we will see the task PELT values decay.

We rely on sync_entity_load_avg() for instance in EAS wakeup where the
task's util_avg influences on which CPU type the task will run next. So
we sync the wakee with its cfs_rq to be able to work with a current task
util_avg.