On Di, 2010-11-23 at 15:20 +0100, Peter Zijlstra wrote: > On Tue, 2010-11-23 at 15:12 +0100, Gerald Schaefer wrote: > > On Mo, 2010-11-22 at 12:10 -0800, Andrew Morton wrote: > > > On Mon, 22 Nov 2010 15:47:36 +0100 > > > Gerald Schaefer <gerald.schaefer@xxxxxxxxxx> wrote: > > > > > > > From: Gerald Schaefer <gerald.schaefer@xxxxxxxxxx> > > > > > > > > The spinning mutex implementation uses cpu_relax() in busy loops as a > > > > compiler barrier. Depending on the architecture, cpu_relax() may do more > > > > than needed in this specific mutex spin loops. On System z we also give > > > > up the time slice of the virtual cpu in cpu_relax(), which prevents > > > > effective spinning on the mutex. > > > > > > > > This patch replaces cpu_relax() in the spinning mutex code with > > > > arch_mutex_cpu_relax(), which can be defined by each architecture that > > > > selects HAVE_ARCH_MUTEX_CPU_RELAX. The default is still cpu_relax(), so > > > > this patch should not affect other architectures than System z for now. > > > > > > > > ... > > > > > > > > --- a/include/linux/mutex.h > > > > +++ b/include/linux/mutex.h > > > > @@ -160,4 +160,8 @@ extern int mutex_trylock(struct mutex *l > > > > extern void mutex_unlock(struct mutex *lock); > > > > extern int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock); > > > > > > > > +#ifndef CONFIG_HAVE_ARCH_MUTEX_CPU_RELAX > > > > +#define arch_mutex_cpu_relax() cpu_relax() > > > > +#endif > > > > > > A simpler way of doing this is to remove the CONFIG_ variable > > > altogether and do > > > > > > #ifndef arch_mutex_cpu_relax > > > #define arch_mutex_cpu_relax() cpu_relax() > > > #endif > > > > > > When doing this, one should be clear about _which_ arch file has the > > > responsibility of defining arch_mutex_cpu_relax, and make sure that > > > this arch file is reliably included in the .c file. > > > > Well, I've tried that with my last approach, defining arch_mutex_cpu_relax() > > in <asm/mutex.h> and including that from <linux/mutex.h>. This didn't work > > well because of ugly header file dependencies, and Peter also commented > > that "including "asm/mutex.h" isn't advised". The problem is the following > > code in kernel/mutex.c (after including <linux/mutex.h>) when > > CONFIG_DEBUG_MUTEXES is set: > > > > #ifdef CONFIG_DEBUG_MUTEXES > > # include "mutex-debug.h" > > # include <asm-generic/mutex-null.h> > > #else > > # include "mutex.h" > > # include <asm/mutex.h> > > #endif > > > > So I can only include <asm/mutex.h> from <linux/mutex.h> with an ugly > > "#ifndef CONFIG_DEBUG_MUTEXES" around it, or use a completely different > > or new arch header file (but <asm/mutex.h> seems like the right place > > for this). The CONFIG_ approach avoids all this header file dependency > > mess, or did I miss something (or maybe it's just me and it is not ugly > > at all)? > > Yeah, that all cause massive grief.. I've applied your patch as is, > assuming s390 already has the needed arch_mutex_cpu_relax() > implementation (otherwise I've just broken my s390 build). Thanks, my patch already includes the s390 implementation for arch_mutex_cpu_relax() in arch/s390/include/asm/mutex.h (we use a simple barrier() in this case). We still have HAVE_DEFAULT_NO_SPIN_MUTEXES selected though, this will be removed with one of Martins next patches, once this preparation patch is upstream. -- To unsubscribe from this list: send the line "unsubscribe linux-s390" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
