On 17/08/2017 00:31, Michael S. Tsirkin wrote: > On Wed, Aug 16, 2017 at 11:25:35PM +0200, Paolo Bonzini wrote: >> Yes, I agree. EMULTYPE_SKIP is fine because failed decoding still >> causes an exception to be injected. Maybe it's better to gate the >> EMULTYPE_SKIP emulation on the exit qualification saying this is a write > > I thought it's already limited to writes. I agree that's a reasonable > limitation in any case. > >> and also not a page table walk---just in case. > > I still don't get it, sorry. Let's assume for the sake of argument > that it's a PT walk causing the MMIO access. Just why do you think > that it makes sense to skip the instruction that caused the walk? I think it doesn't. I think in that case it's better to skip the fast write and proceed with full emulation. >>> It's just that this has been there for 3 years and people have built a >>> product around this. >> >> Around 700 clock cycles? > > About 30% the cost of exit, isn't it? There are definitely workloads > where cost of exit gates performance. We didn't work on fast mmio based > on theoretical assumptions. But maybe I am wrong. We'll see. Jason here > volunteered to test your patch and we'll see what comes out of it. If > I'm wrong and it's about 1%, I won't split hairs. Note that we still get the latency benefit from fast MMIO, and maybe we can cut a couple hundred clock cycles more---which would benefit all emulation, not just fast MMIO. Paolo