On Sun, Aug 16, 2009 at 10:48:31PM +0200, Ingo Molnar wrote: > * Heiko Carstens <heiko.carstens@xxxxxxxxxx> wrote: > > On Sun, Aug 16, 2009 at 08:06:31PM +0200, Ingo Molnar wrote: > > > * Heiko Carstens <heiko.carstens@xxxxxxxxxx> wrote: > > > > > #define __spin_lock_is_small > > > > > > > > +unsigned long __lockfunc _spin_lock_irqsave_nested(spinlock_t *lock, int subclass) > > > > > + __acquires(lock); > > > > > + > > > > > +#ifdef __spin_lock_is_small > > > > > +#define _spin_lock(lock) __spin_lock(lock) > > > > > +#else > > > > > +void __lockfunc _spin_lock(spinlock_t *lock) __acquires(lock); > > > > > +#endif > > > > > > Dunno - i'm somewhat wary of introducing a 2^28 variability here. > > > (although the number of real variations is much lower - but still). > > > > > > What's the current situation on s390, precisely which of the 28 lock > > > functions are a win to be inlined and which ones are a loss? Do you > > > have a list/table perhaps? > > > > No list unfortunately. [...] > > Well, if you dont know the functions you want to inline, how will > you make intelligent use of this facility then in s390? What I tried to say: in general we want to have all locking functions inlined. This is because all our measurements show if anything gets inlined performance improves. This is even true for large functions. > Btw., i just noticed that s390 has CONFIG_PREEMPT turned on by > default in its defconfig. Have you made your measurements with > CONFIG_PREEMPT? If yes then the current inlining rules in spinlock.h > will turn all the locking APIs into functions. All measurements were done with CONFIG_PREEMPT off. -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
