>>> On 11.10.17 at 15:59, <rcojoc...@bitdefender.com> wrote: > Now that "x86/hvm: implement hvmemul_write() using real mappings" is in, > we can start working again on implementing hvmemul_cmpxchg() with a real > CMPXCHG, and finally fix the SMP emulation race upstream.
That's my plan, but this will end up being complete only with the RMW patch I had also posted, and which now can be made use of the hvm/emulate.c. I'm intending to get to this as soon as time permits. > However, in order to do that we would need X86EMUL_CMPXCHG_FAILED which > has been dropped by Andrew here: > > https://patchwork.kernel.org/patch/9449339/ > > and re-contributed by Jan here: > > https://patchwork.kernel.org/patch/9651613/ > > (the patch is a combination of Jan's patch and my fumbling with CMPXCHG). > > However, I remember Jan saying that his patch is no longer the way to go > here. How should we proceed? Well, the patch will need some re-working, so that the end result will fit with the RMW one (in particular I also want to eliminate the multiple reads issue we have), but right now I don't recall any reason why it wouldn't be suitable (anymore) at all. But this is all work targeted at 4.11, so not really a need to rush. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org https://lists.xen.org/xen-devel
