Re: ERESTARTSYS escaping from sem_wait with RTLinux patch

[Thomas Gleixner]

On Tue, 13 Oct 2009, Darren Hart wrote:
> Talking with Thomas a bit more we're not sure the patch that introduced this
> lockless waking actually does any good, as the normal wakeup path doesn't take
> the hb->lock anyway, it's more likely the contention was due to an app like
> this that wakes a task and almost immediately puts it back to sleep on a futex
> before the waker has a chance to drop the hb->lock.

The patch was done before we restructured the wakeup code in mainline
and it's not necessary anymore.
 
> The futex wake-up path is complicated enough as it is, in my personal opinion,
> we are better off dropping the "lockless wake-up" patch and removing the race
> and simplifying the wake-up path at the same time.

Nevertheless we need graceful handling of spurious wakeups in the
waiter code. I'm working on a fix for mainline, which will be in the
next -rt release as well (after I reverted the no longer valid wakeup
optimization)

Thanks folks for debugging that.

       tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/