lkml.org 
[lkml]   [2022]   [Aug]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH v2] sched/fair: Introduce priority load balance to reduce interference from IDLE tasks
Le jeudi 18 août 2022 à 10:46:55 (+0800), Abel Wu a écrit :
> On 8/17/22 8:58 PM, Vincent Guittot Wrote:
> > On Tue, 16 Aug 2022 at 04:53, zhangsong (J) <zhangsong34@huawei.com> wrote:
> > >
> > >

...

> > > Yes, this is usually a corner case, but suppose that some non-idle tasks bounds to CPU 1-2
> > >
> > > and idle tasks bounds to CPU 0-1, so CPU 1 may has many idle tasks and some non-idle
> > >
> > > tasks while idle tasks on CPU 1 can not be pulled to CPU 2, when trigger load balance if
> > >
> > > CPU 2 should pull some tasks from CPU 1, the bad result is idle tasks of CPU 1 cannot be
> > >
> > > migrated and non-idle tasks also cannot be migrated in case of env->loop_max constraint.
> >
> > env->loop_max adds a break but load_balance will continue with next
> > tasks so it also tries to pull your non idle task at the end after
> > several breaks.
>
> Loop will be terminated without LBF_NEED_BREAK if exceeds loop_max :)

Argh yes, my brain is not yet back from vacation
I have been confused by loop_max and loop_break being set to the same value 32

Zhang Song, Could you try the patch below ? If it works, I will prepare a
clean patch with all tags



sched/fair: make sure to try to detach at least one movable task

During load balance we try at most env->loop_max time to move a task. But
it can happen that the LRU tasks (ie tail of the cfs_tasks list) can't
be moved to dst_cpu because of affinity. In this case, loop in the list
until we found at least one.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
kernel/sched/fair.c | 12 +++++++++---
1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index da388657d5ac..02b7b808e186 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8052,8 +8052,12 @@ static int detach_tasks(struct lb_env *env)
p = list_last_entry(tasks, struct task_struct, se.group_node);

env->loop++;
- /* We've more or less seen every task there is, call it quits */
- if (env->loop > env->loop_max)
+ /*
+ * We've more or less seen every task there is, call it quits
+ * unless we haven't found any movable task yet.
+ */
+ if (env->loop > env->loop_max &&
+ !(env->flags & LBF_ALL_PINNED))
break;

/* take a breather every nr_migrate tasks */
@@ -10182,7 +10186,9 @@ static int load_balance(int this_cpu, struct rq *this_rq,

if (env.flags & LBF_NEED_BREAK) {
env.flags &= ~LBF_NEED_BREAK;
- goto more_balance;
+ /* Stop if we tried all running tasks */
+ if (env.loop < busiest->nr_running)
+ goto more_balance;
}

/*
--
2.17.1
>
> >
> > >
> > > This will cause non-idle tasks cannot achieve more CPU utilization.
> >
> > Your problem is not linked to IDLE vs NORMAL tasks but to the large
> > number of pinned tasks that can't migrate on CPU2. You can end with
> > the same behavior without using IDLE tasks but only NORMAL tasks.
>
> I feel the same thing.
>
> Best,
> Abel

\
 
 \ /
  Last update: 2022-08-18 10:32    [W:0.071 / U:0.752 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site