| Message ID | 1344879528-24897-1-git-send-email-will.deacon@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, 13 Aug 2012, Will Deacon wrote: > Commit a76d7bd96d65 ("ARM: 7467/1: mutex: use generic xchg-based > implementation for ARMv6+") removed the barrier-less, ARM-specific > mutex implementation in favour of the generic xchg-based code. > > Since then, a bug was uncovered in the xchg code when running on SMP > platforms, due to interactions between the locking paths and the > MUTEX_SPIN_ON_OWNER code. This was fixed in 0bce9c46bf3b ("mutex: place > lock in contended state after fastpath_lock failure"), however, the > atomic_dec-based mutex algorithm is now marginally more efficient for > ARM (~0.5% improvement in hackbench scores on dual A15). > > This patch moves ARMv6+ platforms to the atomic_dec-based mutex code. > > Cc: Nicolas Pitre <nico@fluxnic.net> > Signed-off-by: Will Deacon <will.deacon@arm.com> Acked-by: Nicolas Pitre <nico@linaro.org> > --- > arch/arm/include/asm/mutex.h | 8 ++++++-- > 1 files changed, 6 insertions(+), 2 deletions(-) > > diff --git a/arch/arm/include/asm/mutex.h b/arch/arm/include/asm/mutex.h > index b1479fd..d69ebc7 100644 > --- a/arch/arm/include/asm/mutex.h > +++ b/arch/arm/include/asm/mutex.h > @@ -9,8 +9,12 @@ > #define _ASM_MUTEX_H > /* > * On pre-ARMv6 hardware this results in a swp-based implementation, > - * which is the most efficient. For ARMv6+, we emit a pair of exclusive > - * accesses instead. > + * which is the most efficient. For ARMv6+, we have exclusive memory > + * accessors and use atomic_dec to avoid the extra xchg operations > + * on the locking slowpaths. > */ > +#if __LINUX_ARM_ARCH < 6 > #include <asm-generic/mutex-xchg.h> > +#else > +#include <asm-generic/mutex-dec.h> > #endif > -- > 1.7.4.1 >
diff --git a/arch/arm/include/asm/mutex.h b/arch/arm/include/asm/mutex.h index b1479fd..d69ebc7 100644 --- a/arch/arm/include/asm/mutex.h +++ b/arch/arm/include/asm/mutex.h @@ -9,8 +9,12 @@ #define _ASM_MUTEX_H /* * On pre-ARMv6 hardware this results in a swp-based implementation, - * which is the most efficient. For ARMv6+, we emit a pair of exclusive - * accesses instead. + * which is the most efficient. For ARMv6+, we have exclusive memory + * accessors and use atomic_dec to avoid the extra xchg operations + * on the locking slowpaths. */ +#if __LINUX_ARM_ARCH < 6 #include <asm-generic/mutex-xchg.h> +#else +#include <asm-generic/mutex-dec.h> #endif
Commit a76d7bd96d65 ("ARM: 7467/1: mutex: use generic xchg-based implementation for ARMv6+") removed the barrier-less, ARM-specific mutex implementation in favour of the generic xchg-based code. Since then, a bug was uncovered in the xchg code when running on SMP platforms, due to interactions between the locking paths and the MUTEX_SPIN_ON_OWNER code. This was fixed in 0bce9c46bf3b ("mutex: place lock in contended state after fastpath_lock failure"), however, the atomic_dec-based mutex algorithm is now marginally more efficient for ARM (~0.5% improvement in hackbench scores on dual A15). This patch moves ARMv6+ platforms to the atomic_dec-based mutex code. Cc: Nicolas Pitre <nico@fluxnic.net> Signed-off-by: Will Deacon <will.deacon@arm.com> --- arch/arm/include/asm/mutex.h | 8 ++++++-- 1 files changed, 6 insertions(+), 2 deletions(-)