lkml.org 
[lkml]   [2011]   [Jun]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [patch 4/4] sched: Distangle worker accounting from rq->lock
    Hello, again.

    So, let's get it correct first.

    On Wed, Jun 22, 2011 at 05:52:15PM -0000, Thomas Gleixner wrote:
    > The worker accounting for cpu bound workers is plugged into the core
    > scheduler code and the wakeup code. This is not a hard requirement and
    > can be avoided by keeping track of the state in the workqueue code
    > itself.
    > Keep track of the sleeping state in the worker itself and call the
    > notifier before entering the core scheduler. There might be false
    > positives when the task is woken between that call and actually
    > scheduling, but that's not really different from scheduling and being
    > woken immediately after switching away. There is also no harm from
    > updating nr_running when the task returns from scheduling instead of
    > accounting it in the wakeup code.

    I think false positives on schedule() should be safe. As said
    earlier, the gap between ttwu and actually running is a bit worrisome
    but it might be nothing, but please at least describe the behavior
    change.

    > Index: linux-2.6/kernel/workqueue.c
    > ===================================================================
    > --- linux-2.6.orig/kernel/workqueue.c
    > +++ linux-2.6/kernel/workqueue.c
    > @@ -137,6 +137,7 @@ struct worker {
    > unsigned int flags; /* X: flags */
    > int id; /* I: worker id */
    > struct work_struct rebind_work; /* L: rebind worker to cpu */
    > + int sleeping; /* None */

    bool?

    > -struct task_struct *wq_worker_sleeping(struct task_struct *task,
    > - unsigned int cpu)
    > +void wq_worker_sleeping(struct task_struct *task)
    > {
    > - struct worker *worker = kthread_data(task), *to_wakeup = NULL;
    > - struct global_cwq *gcwq = get_gcwq(cpu);
    > - atomic_t *nr_running = get_gcwq_nr_running(cpu);
    > + struct worker *worker = kthread_data(task);
    > + struct global_cwq *gcwq;
    > + int cpu;
    >
    > if (worker->flags & WORKER_NOT_RUNNING)
    > - return NULL;
    > + return;

    This doesn't look safe. It can race with trustee_thread() setting
    WORKER_ROGUE. Let's just grab gcwq->lock on entry to
    wq_worker_sleeping() for now; then, the schedule() trickery in
    trustee_thread() can go away too. This also means we can remove the
    weird sync rules from ->flags and ->idle_list and just use simple
    gcwq->lock for those, which is pretty nice.

    > - /* this can only happen on the local cpu */
    > - BUG_ON(cpu != raw_smp_processor_id());
    > + if (WARN_ON_ONCE(worker->sleeping))
    > + return;

    Re-entrance is prevented by the scheduler hook being called only for
    non-premption schedule(). Maybe it's better to explain that in the
    function comment?

    Hmmm... Also, I think worker->sleeping should be cleared by
    trustee_thread() when WORKER_ROGUE is set for the worker; otherwise,
    it can get out of sync and the above WARN_ON_ONCE() will trigger.

    Thanks.

    --
    tejun


    \
     
     \ /
      Last update: 2011-06-23 17:09    [from the cache]
    ©2003-2014 Jasper Spaans. Advertise on this site