[lkml]   [2010]   [Mar]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [RFC,PATCH 1/2] cputimers/proc: do_task_stat()->task_times() can race with getrusage()
    On 03/26, Balbir Singh wrote:
    > * Oleg Nesterov <> [2010-03-24 21:45:50]:
    > > do_task_stat()->task_times() can race with getrusage(), they both can
    > > try to update task->prev_Xtime at the same time.
    > >
    > > Remove this bit of d180c5bc "sched: Introduce task_times() to replace
    > > task_{u,s}time()".
    > One of the reasons for adding this accuracy was to avoid sampling
    > based noise and errors that occur with utime and stime.
    > As long as there is no preemption during the assignment, I think we
    > should be OK.

    I don't think preemp_disable() can help. Probably we can use task_lock().

    As for do_task_stat()->thread_group_times(), I think we can make it
    rc-safe without breaking /bin/top.

    1. add spin_lock_irqsave(&sig->cputimer.lock) around
    sig->prev_Xtime = max(...)

    2. Add a couple of barriers into thread_group_cputime()
    and __exit_signal() so that without ->siglock we can
    never overestimate utime/stime if we race with exit.

    If we underestimate these values, this should be fine:

    - the error can't be "systematic", the next read from
    /prod/pid/stat will see the updated values

    - the prev_Xtime logic in thread_group_times() ensures
    the reported time can never go back.

    IOW: at worse, cat /proc/pid/stat can miss the time
    which the exited thread spent on CPU after the previous
    read of /proc/pid/stat. This looks absolutely harmless,
    the next read will see this time.

    Probably we can even detect this case if we look at
    sig->nr_threads and retry.

    I'll try to make patches unless someone has a better idea.

    I just can't accept the fact that we are doing while_each_thread()
    under ->siglock here ;)


     \ /
      Last update: 2010-03-26 22:53    [W:0.020 / U:127.424 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site