Peter Zijlstra
2015-Dec-20 19:59 UTC
[Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
On Sun, Dec 20, 2015 at 05:07:19PM +0000, Andrew Cooper wrote:> > Very much +1 for fixing this. > > Those names would be fine, but they do add yet another set of options in > an already-complicated area. > > An alternative might be to have the regular smp_{w,r,}mb() not revert > back to nops if CONFIG_PARAVIRT, or perhaps if pvops have detected a > non-native environment. (I don't know how feasible this suggestion is, > however.)So a regular SMP kernel emits the LOCK prefix and will patch it out with a DS prefix (iirc) when it finds but a single CPU. So for those you could easily do this. However an UP kernel will not emit the LOCK and do no patching. So if you're willing to make CONFIG_PARAVIRT depend on CONFIG_SMP or similar, this is doable. I don't see people going to allow emitting the LOCK prefix (and growing the kernel text size) for UP kernels.
Michael S. Tsirkin
2015-Dec-21 07:10 UTC
[Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
On Sun, Dec 20, 2015 at 08:59:44PM +0100, Peter Zijlstra wrote:> On Sun, Dec 20, 2015 at 05:07:19PM +0000, Andrew Cooper wrote: > > > > Very much +1 for fixing this. > > > > Those names would be fine, but they do add yet another set of options in > > an already-complicated area. > > > > An alternative might be to have the regular smp_{w,r,}mb() not revert > > back to nops if CONFIG_PARAVIRT, or perhaps if pvops have detected a > > non-native environment. (I don't know how feasible this suggestion is, > > however.) > > So a regular SMP kernel emits the LOCK prefix and will patch it out with > a DS prefix (iirc) when it finds but a single CPU. So for those you > could easily do this. > > However an UP kernel will not emit the LOCK and do no patching. > > So if you're willing to make CONFIG_PARAVIRT depend on CONFIG_SMP or > similar, this is doable.One of the uses for virtio is to allow testing an existing kernel on kvm just by loading a module, and this will break this usecase.> I don't see people going to allow emitting the LOCK prefix (and growing > the kernel text size) for UP kernels.Thinking about this more, maybe __smp_*mb is a better set of names. The nice thing about it is that we can then have generic code that does basically #ifdef CONFIG_SMP #define smp_mb() __smp_mb() #else #define smp_mb() barrier() #endif and reuse this on all architectures. So instead of a maintainance burden, we are actually removing code duplication. -- MST
On some architectures smp_store_mb() calls mb() which is stronger than implied by both the name and the documentation. smp_store_mb is only used by core kernel code at the moment, so we know no one mis-uses it for an MMIO barrier. Make it call smp_mb consistently before some arch-specific code uses it as such by mistake. Signed-off-by: Michael S. Tsirkin <mst at redhat.com> --- Note: I'm guessing an ack from arch maintainers will be needed, but I'm working on a bigger cleanup moving a bunch of duplicated code into asm-generic/barrier.h which depends on this, so not Cc'ing widely yet. Please Ack if appropriate but do not merge yet. diff --git a/arch/ia64/include/asm/barrier.h b/arch/ia64/include/asm/barrier.h index df896a1..425552b 100644 --- a/arch/ia64/include/asm/barrier.h +++ b/arch/ia64/include/asm/barrier.h @@ -77,7 +77,7 @@ do { \ ___p1; \ }) -#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); mb(); } while (0) +#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); smp_mb(); } while (0) /* * The group barrier in front of the rsm & ssm are necessary to ensure diff --git a/arch/powerpc/include/asm/barrier.h b/arch/powerpc/include/asm/barrier.h index 0eca6ef..4f0169e 100644 --- a/arch/powerpc/include/asm/barrier.h +++ b/arch/powerpc/include/asm/barrier.h @@ -34,7 +34,7 @@ #define rmb() __asm__ __volatile__ ("sync" : : : "memory") #define wmb() __asm__ __volatile__ ("sync" : : : "memory") -#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); mb(); } while (0) +#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); smp_mb(); } while (0) #ifdef __SUBARCH_HAS_LWSYNC # define SMPWMB LWSYNC diff --git a/arch/s390/include/asm/barrier.h b/arch/s390/include/asm/barrier.h index d68e11e..6c1d8b5 100644 --- a/arch/s390/include/asm/barrier.h +++ b/arch/s390/include/asm/barrier.h @@ -36,7 +36,7 @@ #define smp_mb__before_atomic() smp_mb() #define smp_mb__after_atomic() smp_mb() -#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); mb(); } while (0) +#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); smp_mb(); } while (0) #define smp_store_release(p, v) \ do { \ diff --git a/include/asm-generic/barrier.h b/include/asm-generic/barrier.h index b42afad..0f45f93 100644 --- a/include/asm-generic/barrier.h +++ b/include/asm-generic/barrier.h @@ -93,7 +93,7 @@ #endif /* CONFIG_SMP */ #ifndef smp_store_mb -#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); mb(); } while (0) +#define smp_store_mb(var, value) do { WRITE_ONCE(var, value); smp_mb(); } while (0) #endif #ifndef smp_mb__before_atomic
Reasonably Related Threads
- [Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
- [Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
- [Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
- [Xen-devel] new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)
- new barrier type for paravirt (was Re: [PATCH] virtio_ring: use smp_store_mb)