Re: Threads not waking on pthread_cond_signal()

[Jakub Jelinek <jakub redhat com>]

On Mon, Jun 23, 2003 at 01:11:00PM +1000, Ian Wienand wrote:
> On Sun, Jun 22, 2003 at 06:42:03PM -0700, Ulrich Drepper wrote:
> > > I am seeing a problem on PowerPC and i386 with NPTL 0.47 where a
> > > thread waiting in pthread_cond_wait() doesn't get woken up on
> > > pthread_cond_signal().
> > 
> > What kernel are you using?
> 
> 2.5.72 -- but I get correct behaviour with NPTL 0.48 so it was
> probably me.
> 
> However, i386 does seem to be a lot slower signalling threads than the
> NPTL included in RedHat 9.
> 
> RUNNING REDHAT 9 WAKEUP TEST
> 718674 wakes ups in 4.99956 sec = 143748 per second
> 861807 wakes ups in 5 sec       = 172361 per second
> 841964 wakes ups in 5.00006 sec = 168391 per second
> 816596 wakes ups in 5.00003 sec = 163318 per second
> 624066 wakes ups in 5.00004 sec = 124812 per second
> 
> RUNNING NPTL 0.48 WAKEUP TEST
> 114766 wakes ups in 4.99935 sec = 22956 per second
> 168681 wakes ups in 4.99996 sec = 33736 per second
> 286987 wakes ups in 4.99985 sec = 57399 per second
> 286391 wakes ups in 4.99962 sec = 57283 per second
> 337517 wakes ups in 5.00024 sec = 67500 per second

Is this 2.5.72 on i386 as well or some other kernel?
On kernels which don't support FUTEX_REQUEUE current NPTL is certainly
expected to be slower (as FUTEX_REQUEUE syscalls are attempted
unconditionally).

	Jakub