On 28/07/2017 10:24, Mike Galbraith wrote: > On Fri, 2017-07-28 at 09:03 +0200, Paolo Bonzini wrote: >> With these two patches, KVM does not blindly pass the exit interruption >> info and exit qualification from the vmcs02 and vmcs12 when injecting >> an exception. There were two spots where this was done, namely >> nested_vmx_check_exception and vmx_inject_page_fault_nested. >> >> Patch 1 avoids writing the vmcs02's VM_EXIT_INTR_ERROR_CODE field, >> which as Jim noticed isn't possible on pre-Haswell machines, but >> otherwise has no semantic effect. >> >> Patch 2 is the actual bugfix. > > I was looking into oodles of lapic_timer.hv_timer_in_use warnings in > RT, introduced by KVM: LAPIC: Fix lapic timer injection delay, when I > noticed this post. Happily, I can report that they are history. This patch won't fix them. They are a race that comes and goes---but yes, it's fixed. Paolo
