Skip to contentSkip to conversation

kvm: x86: fix a race condition result to lost INIT

PATCH Wanpeng Li 1 participant 2 messages Jul 30, 2017 1 message not in the archive KVM

lore
Share & tools
lore

Wanpeng Li

author
#1 lore raw
In reply to a message not in the archive

2017-07-30 19:33 GMT+08:00 <[email protected]>:

Quoted 44 lines
Quoted 44 lines

2017-07-30 18:42 GMT+08:00 <[email protected]>:

Quoted 38 lines
Quoted 37 lines

2017-07-31 0:24 GMT+08:00 Peng Hao <[email protected]>:

Quoted 24 lines

when SMP VM start, AP may lost INIT because of receiving INIT between kvm_vcpu_ioctl_x86_get/set_vcpu_events.

vcpu 0 vcpu 1

                               kvm_vcpu_ioctl_x86_get_vcpu_events
                                       events->smi.latched_init=0

send INIT to vcpu1
set vcpu1's pending_events
kvm_vcpu_ioctl_x86_set_vcpu_events
events->smi.latched_init == 0
clear INIT in pending_events
I don't think it need set/clear kernel state according to userspace's info.

---
 arch/x86/kvm/x86.c | 6 ------
 1 file changed, 6 deletions(-)
arch/x86/kvm/x86.c +0 -0
index 6c7266f..393a7b7 100644

arch/x86/kvm/x86.c static int · line 3161

Quoted 8 lines
arch/x86/kvm/x86.c +0 -3
@@ -3157,12 +3157,6 @@ static int
 kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
                        vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
                else
                        vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
-               if (lapic_in_kernel(vcpu)) {
-                       if (events->smi.latched_init)
-                               set_bit(KVM_APIC_INIT,
 &vcpu->arch.apic->pending_events);

This is not correct, you will lose the INIT after live migration. I just send out another patch to fix it. Thanks for the report.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

yes, you're right. I should modify on qemu not kernel.

There is a patch here to fix it in kvm, https://lkml.org/lkml/2017/7/30/72

Quoted 1 line

I was affected by sipi, kernel never report valid sipi to user space,

How you afftected by SIPI? In addition, you reported the lost of INIT. Is it observed by your workload or by code review?

Regards,
Wanpeng Li

Quoted 11 lines

we don't care whether sipi is lost after migration. Why?

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
arch/x86/kvm/x86.c +0 -3
-                       else
-                               clear_bit(KVM_APIC_INIT,
 &vcpu->arch.apic->pending_events);
-               }
        }
author
#2 lore raw
In reply to a message not in the archive

2017-07-30 19:33 GMT+08:00 <[email protected]>:

Quoted 46 lines
Quoted 45 lines

2017-07-30 18:42 GMT+08:00 <[email protected]>:

Quoted 39 lines
Quoted 21 lines

2017-07-31 0:24 GMT+08:00 Peng Hao <[email protected]>:

Quoted 19 lines

when SMP VM start, AP may lost INIT because of receiving INIT between kvm_vcpu_ioctl_x86_get/set_vcpu_events.

vcpu 0 vcpu 1

                               kvm_vcpu_ioctl_x86_get_vcpu_events
                                       events->smi.latched_init=0

send INIT to vcpu1
set vcpu1's pending_events
kvm_vcpu_ioctl_x86_set_vcpu_events
events->smi.latched_init == 0
clear INIT in pending_events
I don't think it need set/clear kernel state according to userspace's info.

---
 arch/x86/kvm/x86.c | 6 ------
 1 file changed, 6 deletions(-)
Quoted 16 lines

arch/x86/kvm/x86.c static int · line 3161

Quoted 13 lines
arch/x86/kvm/x86.c +0 -3
index 6c7266f..393a7b7 100644
@@ -3157,12 +3157,6 @@ static int
 kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
                        vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
                else
                        vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
-               if (lapic_in_kernel(vcpu)) {
-                       if (events->smi.latched_init)
-                               set_bit(KVM_APIC_INIT,
 &vcpu->arch.apic->pending_events);

This is not correct, you will lose the INIT after live migration. I just send out another patch to fix it. Thanks for the report.

yes, you're right. I should modify on qemu not kernel.

Quoted 7 lines
Quoted 3 lines

There is a patch here to fix it in kvm, https://lkml.org/lkml/2017/7/30/72

Quoted 1 line

I was affected by sipi, kernel never report valid sipi to user space,

we don't care whether sipi is lost after migration. Why?

Please refer to Paolo's suggestion,
https://www.mail-archive.com/[email protected]/msg464882.html

Regards,
Wanpeng Li

Quoted 9 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
arch/x86/kvm/x86.c +0 -3
-                       else
-                               clear_bit(KVM_APIC_INIT,
 &vcpu->arch.apic->pending_events);
-               }
        }