google / alioth / 42981753395

44%

main: 44%

Build Type

push

github

Committed by Lencerf

Commit Message

fix(kvm): do not lose vCPU kicks outside of KVM_RUN

`stop_vcpu()` kicks a vCPU with SIGRTMIN, which only interrupts a vCPU
inside KVM_RUN. A vCPU outside KVM_RUN, e.g. one still initializing or
emulating an MMIO access, misses the kick and enters the guest again.
If the guest then halts, the vCPU never notices the request to pause,
shut down, or reboot the VM.

As suggested by the KVM API documentation, let the signal handler set
`immediate_exit` of the vCPU of the current thread, so that the next
KVM_RUN returns immediately, and clear it when KVM_RUN returns EINTR.
The handler finds the vCPU through a thread-local variable, so a vCPU
stays on the thread that created it.

Signed-off-by: Changyuan Lyu <changyuanl@google.com>

Coverage Stats

980 of 1760 branches covered (55.68%)

Branch coverage included in aggregate %.

0 of 27 new or added lines in 2 files covered. (0.0%)

3 existing lines in 2 files now uncovered.

6995 of 16457 relevant lines covered (42.5%)

905.04 hits per line