MIME-Version: 1.0
References: <1638780405-38026-1-git-send-email-quic_pragalla@quicinc.com>
 <CAJfpegvDfc9JUo6VASRyYXzj1=j3t6oU9W3QGWO08vhfWHf-UA@mail.gmail.com>
 <Ya8ycLODlcvLx4xB@hirez.programming.kicks-ass.net> <CAJfpegsVg2K0CvrPvXGSu=Jz_R3VZZOy49Jw51rThQUJ1_9e6g@mail.gmail.com>
 <Ya86coKm4RuQDmVS@hirez.programming.kicks-ass.net>
In-Reply-To: <Ya86coKm4RuQDmVS@hirez.programming.kicks-ass.net>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Tue, 7 Dec 2021 13:44:49 +0100
Message-ID: <CAJfpegumZ1RQLBCtbrOiOAT9ygDtDThpySwb8yCpWGBu1fRQmw@mail.gmail.com>
Subject: Re: [PATCH V1] fuse: give wakeup hints to the scheduler
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     Ingo Molnar <mingo@redhat.com>, quic_stummala@quicinc.com,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        quic_pkondeti@quicinc.com, quic_sayalil@quicinc.com,
        quic_aiquny@quicinc.com, quic_zljing@quicinc.com,
        quic_blong@quicinc.com, quic_richardp@quicinc.com,
        quic_cdevired@quicinc.com,
        Pradeep P V K <quic_pragalla@quicinc.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, 7 Dec 2021 at 11:42, Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Tue, Dec 07, 2021 at 11:20:59AM +0100, Miklos Szeredi wrote:

> > That may be good for fuse as the data is indeed shared.  It would be
> > even better if the woken task had already a known affinity to this
> > CPU, since that would mean thread local data wouldn't have to be
> > migrated each time...   So I'm not sure sync wakeup alone is worth it,
> > needs real life benchmarking.
>
> Hard affinities have other down-sides.. occasional migrations shouldn't
> be a problem, constant migrations are bad.

Currently fuse readers are sleeping in
wait_event_interruptible_exclusive().  wake_up_interruptible_sync()
will pick the first thread in the queue and wake that up.  That will
likely result in migration, right?

What would be much nicer, is to look at all the threads on the waitq
and pick one that previously ran on the current CPU if there's one.
Could this be implemented?

Thanks,
Miklos