On 12/12/18 2:26 AM, Jan Beulich wrote: >>>> On 11.12.18 at 19:16, <ashok....@intel.com> wrote: >> BTW: Apart from the fact its ugly and take a loooong time to complete, do you >> have any practical isssues you want to highlight? maybe that can >> help upstream as well. > The situation for a kernel and a hypervisor might be different here > (but in the Linux case may then be more hypervisor-like when > considering KVM): The hypervisor needs to make sure in particular > time management within guests won't break. Stopping the > machine for an extended period of time may not be helpful there. > For processes in an OS the constraints might not be as tight, but > I could imaging problems even there in some less common cases. > > That said, I'm not convinced at all it is a good idea to load new > ucode while _any_ guests are running, but we're talking about a > last resort approach here anyway. >
BTW, one thing I meant to mention about this patch earlier: we observed that updating microcode from only one sibling resulted in guest reading stale version value from another thread. This causes at least some Windows versions to crash. We ended up executing "wrmsr(0x8b, 0); cpuid_eax(1);" on all threads after the update. (Another alternative could be to intercept the reads) -boris _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
