vmm: cpu: Add pending removed vcpu check to avoid resize vcpu hang #5668
Conversation
vmm/src/cpu.rs
Outdated
| @@ -518,12 +521,6 @@ impl BusDevice for CpuManager { | |||
| { | |||
| state.inserting = false; | |||
| } | |||
| // Ditto for removal | |||
| if (data[0] & (1 << CPU_REMOVING_FLAG) == 1 << CPU_REMOVING_FLAG) | |||
There was a problem hiding this comment.
Hhm, I think we might still want to keep the ACPI pending removal flag - my only worry is if the kernel gets confused that is has acknowledged the removal and then the removal bit doesn't get cleared when it checks again. Maybe we need to track another bit of state :-(
There was a problem hiding this comment.
Okay. Let me think a better way out.
There was a problem hiding this comment.
I think it's fine to keep the existing removing bool. And add a Arc<AtomicBool>> called pending_removal. Set both on the request to remove, let the guest clear the removing as it sees fit but use the atomic one to check if a subsequent removal should be rejected.
There was a problem hiding this comment.
Thx for ur suggestions, that sounds make sense and I will update the patch.
Add pending removed vcpu check according to VcpuState.removing, which can avoid cloud hypervisor hangup during continual vcpu resize. Fix cloud-hypervisor#5419 Signed-off-by: Yi Wang <foxywang@tencent.com>
Add pending removed vcpu check according to VcpuState.removing, which can avoid cloud hypervisor hangup during continual vcpu resize.
Fix #5419