<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">Crap, yeah indeed that needs to be
      protected by some lock.<br>
      <br>
      Going to prepare a patch for that,<br>
      Christian.<br>
      <br>
      Am 09.08.2018 um 21:49 schrieb Andrey Grodzovsky:<br>
    </div>
    <blockquote type="cite"
      cite="mid:54621fc1-7246-f1bf-26bb-a16c4daf249f@amd.com">
      <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
      <p>Reviewed-by: Andrey Grodzovsky <a
          class="moz-txt-link-rfc2396E"
          href="mailto:andrey.grodzovsky@amd.com" moz-do-not-send="true"><andrey.grodzovsky@amd.com></a></p>
      <p><br>
      </p>
      <p>But I still  have questions about entity->last_user (didn't
        notice this before) - <br>
      </p>
      <p>Looks to me there is a race condition with it's current usage,
        let's say process A was preempted after doing
        drm_sched_entity_flush->cmpxchg(...)</p>
      <p>now process B working on same entity (forked) is inside
        drm_sched_entity_push_job, he writes his PID to
        entity->last_user and also</p>
      <p>executes drm_sched_rq_add_entity. Now process A runs again and
        execute drm_sched_rq_remove_entity inadvertently causing process
        B removal</p>
      <p>from it's scheduler rq.</p>
      <p>Looks to me like instead we should lock together
        entity->last_user accesses and adds/removals of entity to the
        rq.</p>
      <p>Andrey<br>
      </p>
      <br>
      <div class="moz-cite-prefix">On 08/06/2018 10:18 AM, Nayan
        Deshmukh wrote:<br>
      </div>
      <blockquote type="cite"
cite="mid:CAFd4ddzyvHPHepAgs=mjyWVj0WDV_pQbE9x7aHwNZ_zcME6fqQ@mail.gmail.com">
        <meta http-equiv="Content-Type" content="text/html;
          charset=utf-8">
        <div dir="ltr">
          <div>
            <div>I forgot about this since we started discussing
              possible scenarios of processes and threads.<br>
              <br>
            </div>
            In any case, this check is redundant. Acked-by: Nayan
            Deshmukh <<a href="mailto:nayan26deshmukh@gmail.com"
              moz-do-not-send="true">nayan26deshmukh@gmail.com</a>><br>
            <br>
          </div>
          Nayan<br>
        </div>
        <br>
        <div class="gmail_quote">
          <div dir="ltr">On Mon, Aug 6, 2018 at 7:43 PM Christian König
            <<a href="mailto:ckoenig.leichtzumerken@gmail.com"
              moz-do-not-send="true">ckoenig.leichtzumerken@gmail.com</a>>
            wrote:<br>
          </div>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">Ping. Any
            objections to that?<br>
            <br>
            Christian.<br>
            <br>
            Am 03.08.2018 um 13:08 schrieb Christian König:<br>
            > That is superflous now.<br>
            ><br>
            > Signed-off-by: Christian König <<a
              href="mailto:christian.koenig@amd.com" target="_blank"
              moz-do-not-send="true">christian.koenig@amd.com</a>><br>
            > ---<br>
            >   drivers/gpu/drm/scheduler/gpu_scheduler.c | 5 -----<br>
            >   1 file changed, 5 deletions(-)<br>
            ><br>
            > diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c
            b/drivers/gpu/drm/scheduler/gpu_scheduler.c<br>
            > index 85908c7f913e..65078dd3c82c 100644<br>
            > --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c<br>
            > +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c<br>
            > @@ -590,11 +590,6 @@ void
            drm_sched_entity_push_job(struct drm_sched_job *sched_job,<br>
            >       if (first) {<br>
            >               /* Add the entity to the run queue */<br>
            >               spin_lock(&entity->rq_lock);<br>
            > -             if (!entity->rq) {<br>
            > -                     DRM_ERROR("Trying to push to a
            killed entity\n");<br>
            > -                   
             spin_unlock(&entity->rq_lock);<br>
            > -                     return;<br>
            > -             }<br>
            >               drm_sched_rq_add_entity(entity->rq,
            entity);<br>
            >               spin_unlock(&entity->rq_lock);<br>
            >             
             drm_sched_wakeup(entity->rq->sched);<br>
            <br>
          </blockquote>
        </div>
      </blockquote>
      <br>
    </blockquote>
    <br>
  </body>
</html>