On Wed, Apr 24, 2013 at 4:10 AM, Will Deacon <will.deacon@xxxxxxx> wrote: > On Wed, Apr 24, 2013 at 12:03:10PM +0100, Marc Zyngier wrote: >> On 23/04/13 23:58, Christoffer Dall wrote: >> > I noticed that this doesn't do any cache cleaning. Are the MMU page >> > table walks guaranteed to be coherent with the MMU on arm64? >> >> I suppose you meant the cache. In this case, yes. The hardware page >> table walker must snoop the caches. > > Also, for ARMv7, SMP implies that the hardware walker snoops the cache. I > recently upstreamed some patches for this (see "ARM: 7691/1: mm: kill unused > TLB_CAN_READ_FROM_L1_CACHE and use ALT_SMP instead" in -next), so you might > want to check if there are any remaining, redundant flushes in kvm for ARMv7. > always on SMP? I thought this was implementation specific and you could read this in some feature register? In any case, I'm quite sure we do some unnecessary cleaning in KVM then, so I'll look at it. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
