Re: nptl 0.41

[Jakub Jelinek <jakub redhat com>]

On Thu, May 29, 2003 at 04:44:34PM -0500, Steven Munroe wrote:
> Is nptl 0.41 still "considered more or less x86-only." like nptl-0.40 or 
> is it intended for general use?
> 
> PPC (and I suspect S390 and SH) is broken due to the futex requeue 
> change.  kernel-features.h has __ASSUME_FUTEX_REQUEUE but that is not 
> used in the common ./libc23/nptl/sysdeps/pthread/pthread_cond_*.c 
> source. It seems pthread_cond_wait.c, pthread_cond_signal.c, and 
> pthread_cond_broadcast.c  where changed to use __data.__mutex and 
> lll_futex_requeue without conditioning it on __ASSUME_FUTEX_REQUEUE.
> 
> What is the general intent for this;
> 1) all platforms should assume the new kernel and ignore 
> __ASSUME_FUTEX_REQUEUE,
> 2) or nptl should honor __ASSUME_FUTEX_REQUEUE and continue to support 
> pre 2.5.70 kernels?

NPTL should use FUTEX_REQUEUE always and unless __ASSUME_FUTEX_REQUEUE
fall back to FUTEX_WAKE if FUTEX_REQUEUE is not available.
I have ppc NPTL changes done and they pass make check on pre-FUTEX_REQUEUE
kernel.

	Jakub