[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CANaxB-wvWc6kNWmwhXeuN+YQJ6=tVDx9BJ18tOqMUZx5GfR55Q@mail.gmail.com>
Date: Wed, 26 Oct 2022 23:51:18 -0700
From: Andrei Vagin <avagin@...il.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: linux-kernel@...r.kernel.org,
Andy Lutomirski <luto@...capital.net>,
Christian Brauner <brauner@...nel.org>,
Dietmar Eggemann <dietmar.eggemann@....com>,
Kees Cook <keescook@...omium.org>,
Ingo Molnar <mingo@...hat.com>,
Juri Lelli <juri.lelli@...hat.com>,
Peter Oskolkov <posk@...gle.com>,
Tycho Andersen <tycho@...ho.pizza>,
Will Drewry <wad@...omium.org>,
Vincent Guittot <vincent.guittot@...aro.org>
Subject: Re: [PATCH 3/5] sched: add a few helpers to wake up tasks on the
current cpu
On Thu, Oct 20, 2022 at 5:44 PM Andrei Vagin <avagin@...il.com> wrote:
>
> On Thu, Oct 20, 2022 at 04:15:04PM +0200, Peter Zijlstra wrote:
> > On Wed, Oct 19, 2022 at 06:10:46PM -0700, Andrei Vagin wrote:
> > > Add complete_on_current_cpu, wake_up_poll_on_current_cpu helpers to wake
> > > up processes on the current CPU.
> >
> > There is an astounding lack of *why* in this changelog.
>
> I use them in the next patch to handle seccomp user notify requests
> faster.
>
> The seccomp notify mechanism allows less privileged processes to offload
> specific syscalls to more privileged processes. In many cases, the
> workflow is fully synchronous. It means a target process triggers a
> system call, the kernel stops it and wakes up a supervisor process that
> handles the system call and returns controls back to the target process.
> In this context, "synchronous" means that only one process is running
> and another one is waiting.
>
> New helpers advices the scheduler to move the wakee to the current CPU.
> For synchronous workflows like described above, these helpers makes
> context switches a few times faster.
Peter,
I've found that I don't understand why WF_SYNC doesn't work in this
case. The test from the last patch shows performance improvements in the
case of WF_CURRENT_CPU, but WF_SYNC doesn't make any difference. I
looked at the code and found that select_task_rq_fair calls
select_idle_sibling, but it doesn't take into account the sync flag.
Does it make sense to do something like this:
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 4ebe7222664c..c29f758ccfe3 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -7249,7 +7249,8 @@ select_task_rq_fair(struct task_struct *p, int
prev_cpu, int wake_flags)
new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
} else if (wake_flags & WF_TTWU) { /* XXX always ? */
/* Fast path */
- new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
+ if (!(sync && cpu == new_cpu && this_rq()->nr_running == 1))
+ new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
}
rcu_read_unlock();
With this patch, the test shows the same numbers for WF_CURRENT_CPU and WF_SYNC.
Thanks,
Andrei
Powered by blists - more mailing lists