On 1/11/24 15:15, Andrey Drobyshev wrote:
> On 11/1/23 18:13, Denis V. Lunev wrote:
>> On 11/1/23 16:23, Andrey Drobyshev wrote:
>>> Currently we emit GUEST_PANICKED event in case kvm_vcpu_ioctl() returns
>>> KVM_EXIT_SYSTEM_EVENT with the event type KVM_SYSTEM_EVENT_CRASH. Let's
>>> extend this scenario and emit GUEST_PANICKED in case of an abnormal KVM
>>> exit. That's a natural thing to do since in this case guest is no
>>> longer operational anyway.
>>>
>>> Signed-off-by: Andrey Drobyshev <andrey.drobys...@virtuozzo.com>
>>> Acked-by: Denis V. Lunev <d...@virtuozzo.com>
>>> ---
>>> accel/kvm/kvm-all.c | 19 +++++++++++++++----
>>> 1 file changed, 15 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
>>> index e39a810a4e..d74b3f0b0e 100644
>>> --- a/accel/kvm/kvm-all.c
>>> +++ b/accel/kvm/kvm-all.c
>>> @@ -2816,6 +2816,14 @@ static void kvm_eat_signals(CPUState *cpu)
>>> } while (sigismember(&chkset, SIG_IPI));
>>> }
>>> +static void kvm_emit_guest_crash(CPUState *cpu)
>>> +{
>>> + kvm_cpu_synchronize_state(cpu);
>>> + qemu_mutex_lock_iothread();
>>> + qemu_system_guest_panicked(cpu_get_crash_info(cpu));
>>> + qemu_mutex_unlock_iothread();
>>> +}
>>> +
>>> int kvm_cpu_exec(CPUState *cpu)
>>> {
>>> struct kvm_run *run = cpu->kvm_run;
>>> @@ -2969,21 +2977,24 @@ int kvm_cpu_exec(CPUState *cpu)
>>> ret = EXCP_INTERRUPT;
>>> break;
>>> case KVM_SYSTEM_EVENT_CRASH:
>>> - kvm_cpu_synchronize_state(cpu);
>>> - qemu_mutex_lock_iothread();
>>> - qemu_system_guest_panicked(cpu_get_crash_info(cpu));
>>> - qemu_mutex_unlock_iothread();
>>> + kvm_emit_guest_crash(cpu);
>>> ret = 0;
>>> break;
>>> default:
>>> DPRINTF("kvm_arch_handle_exit\n");
>>> ret = kvm_arch_handle_exit(cpu, run);
>>> + if (ret < 0) {
>>> + kvm_emit_guest_crash(cpu);
>>> + }
>>> break;
>>> }
>>> break;
>>> default:
>>> DPRINTF("kvm_arch_handle_exit\n");
>>> ret = kvm_arch_handle_exit(cpu, run);
>>> + if (ret < 0) {
>>> + kvm_emit_guest_crash(cpu);
>>> + }
>>> break;
>>> }
>>> } while (ret == 0);
>> This allows to gracefully handle this problem in production
>> and reset the guest using on_crash action in libvirt.
>
> Ping
Ping
