Re: [Qemu-devel] [PATCH qemu 0/7] KVM: race-free exit from KVM_RUN without POSIX signals

[Paolo Bonzini <pbonzini@redhat.com>]

On 10/02/2017 10:50, Paolo Bonzini wrote:
> The purpose of the KVM_SET_SIGNAL_MASK API is to let userspace "kick"
> a VCPU out of KVM_RUN through a POSIX signal.  A signal is attached
> to a dummy signal handler; by blocking the signal outside KVM_RUN and
> unblocking it inside, this possible race is closed:
> 
>           VCPU thread                     service thread
>    --------------------------------------------------------------
>         check flag
>                                           set flag
>                                           raise signal
>         (signal handler does nothing)
>         KVM_RUN
> 
> However, one issue with KVM_SET_SIGNAL_MASK is that it has to take
> tsk->sighand->siglock on every KVM_RUN.  This lock is often on a
> remote NUMA node, because it is on the node of a thread's creator.
> Taking this lock can be very expensive if there are many userspace
> exits (as is the case for SMP Windows VMs without Hyper-V reference
> time counter).
> 
> As an alternative, we can put the flag directly in kvm_run so that
> KVM can see it:
> 
>           VCPU thread                     service thread
>    --------------------------------------------------------------
>                                           raise signal
>         signal handler
>           set run->immediate_exit
>         KVM_RUN
>           check run->immediate_exit
> 
> This is what the last patch in this series does, together with
> the corresponding kernel API.  The first six patches are a long
> detour in the signal handling code, moving KVM-specific stuff
> from cpus.c to kvm-all.c so that we have a better hook point for
> KVM_CAP_IMMEDIATE_EXIT (patches 1-3, 6).
> 
> Because KVM_SET_SIGNAL_MASK is also unblocking SIGBUS so that
> BUS_MCEERR_AR actions can be delivered via sigwait, we also have
> to rewrite it (patch 4-5) to avoid sigwait, stowing the machine check
> exception as soon as KVM_RUN exits and process it outside the signal
> handler.
> 
> The seventh patch would of course be split between a linux-headers
> update and the rest.
> 
> Paolo
> 
> ps: As an aside, I finally figured out how to test machine check
> forwarding and I hope to write something about it.

This is needed to test these patches with the final version of the KVM part:

diff --git a/kvm-all.c b/kvm-all.c
index 083143f..c79833e 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -1922,6 +1922,7 @@ static void kvm_eat_signals(CPUState *cpu)
     int r;
 
     if (kvm_immediate_exit) {
+        atomic_set(&cpu->kvm_run->immediate_exit, 0);
         return;
     }
 
Paolo