Zhang Yi <zhang.y...@zte.com.cn> wrote on 2013/10/08 13:59:36:
> Re: Re: Re: [PATCH] futex: Remove the owner check when waking task in > handle_futex_death > > Darren Hart <dvh...@linux.intel.com> wrote on 2013/09/27 23:32:27: > > > > Re: Re: [PATCH] futex: Remove the owner check when waking task in > > handle_futex_death > > > > > > > > The earlier patch cannot solve another problem: > > > The owner wakes the next waiter through normal unlocking which make the > > > futex value as zero, the waked task exits before actually locking the > > > mutex. > > > In this case, the owner doesn't call handle_futex_death() and the waked > > > task > > > doesn't call futex_wake() when they are dying. The rest waiters will still > > > block as the same. > > > > > > This is also the reason that I drop the owner and FUTEX_WAITERS check, > > > because the futex value can be zero at that time. > > > > > > > If the FUTEX_WAITERS bit is not set, there are no waiters, and thus no > > need to wake. I understand why you dropped the OWNER check, but I'm not > > following this one. Where would the futex word be set from having > > waiters to zero when there might still be waiters pending? > > > > > > -- > > Darren Hart > > Intel Open Source Technology Center > > Yocto Project - Linux Kernel > > > > > I have drawn a diagram as below: > > process1 | process2 > ------------------------------------------------------------- > | thread1 | thread2 | thread3 > ------------------------------------------------------------- > t1|pthread_mutex_lock: | | > | __lock=self | | > | | | > t2| |pthread_mutex_lock:| > | |__lock|=FUTEX_WAITERS > | | syscall futex_wait| > | | | > t3| | |pthrea_mutex_lock: > | | |__lock|=FUTEX_WAITERS > | | | syscall futex_wait > | | | > t4|pthread_mutex_unlock:| | > | __lock=0 | | > | syscall futex_wake | waked | > | | | > t5| exit |exit: | > | | handle_futex_death| > --------------------------------------------------------------- > t6| |pthread_mutex_lock:| > | |__lock=self|FUTEX_WAITERS > > 1, At time t4, in the unlocking process of glibc, it clears the FUTEX_WAITERS > bit before > calling futex_wake syscall. > > 2, At time t5, thread2 cannot know if the FUTEX_WAITERS bit was set. > > 3, Time t6 is expected but can never be true. Are there any questions? -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
