Messages in this thread | | | Date | Wed, 11 Oct 2006 16:03:28 -0700 | From | Andrew Morton <> | Subject | Re: [PATCH] i386 Time: Avoid PIT SMP lockups |
| |
On Wed, 11 Oct 2006 15:48:31 -0700 john stultz <johnstul@us.ibm.com> wrote:
> > Wouldn't it be better to fix the livelock? What's causing it? > > I spent a few days trying to narrow this down, and I haven't been able > to do so to my satisfaction. > > At this point, my suspicion is that because the PIT io-read is very slow > (~18us), and done while holding a lock. It would be possible that one > cpu calling gettimeofday would do the following: > > grab xtime sequence read lock > grab i8253 spin lock > do port io (very slow) > release i8253 spin lock > realize xtime has been grabed and repeat > > While another cpu does the following after in a timer interrupt: > Grabs xtime sequence write lock > spins trying to grab i8253 spin lock > > Assuming the first thread can reacquire the i8253 lock before the > second, you could have both threads potentially spinning forever.
Is there any actual need to hold xtime_lock while doing the port IO? I'd have thought it would suffice to do
temp = port_io write_seqlock(xtime_lock); xtime = muck_with(temp); write_sequnlock(xtime_lock);
? - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |