Re: [PATCH v2 4/4] accel/mshv: break the vCPU run loop on exit_request

[Magnus Kulke <magnuskulke@linux.microsoft.com>]

On Fri, Jul 31, 2026 at 09:46:42PM +0300, Doru Blânzeanu wrote:
> Ignored VM exits can cause mshv_cpu_exec() to re-enter MSHV_RUN_VP
> without returning to qemu_process_cpu_events(). A busy vCPU may
> prevent pause_all_vcpus() from completing.
> 
> Check exit_request before re-entering MSHV_RUN_VP and return
> EXCP_INTERRUPT so pending stop requests can be processed.
> 
> Signed-off-by: Doru Blânzeanu <dblanzeanu@linux.microsoft.com>
> ---
>  accel/mshv/mshv-all.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/accel/mshv/mshv-all.c b/accel/mshv/mshv-all.c
> index afcb87b3b8..b7c1126fe0 100644
> --- a/accel/mshv/mshv-all.c
> +++ b/accel/mshv/mshv-all.c
> @@ -780,6 +780,16 @@ static int mshv_cpu_exec(CPUState *cpu)
>              cpu->vcpu_dirty = false;
>          }
>  
> +        /*
> +         * A kick/stop can't reliably break MSHV_RUN_VP out, so leave the run
> +         * loop here too; otherwise a busy vCPU never reaches
> +         * qemu_process_cpu_events() and pause_all_vcpus() hangs under the BQL.
> +         */
> +        if (cpu->stop || qatomic_load_acquire(&cpu->exit_request)) {
> +            ret = EXCP_INTERRUPT;
> +            break;
> +        }
> +
>          ret = mshv_run_vcpu(mshv_state->vm, cpu, &mshv_msg, &exit_reason);
>          if (ret < 0) {
>              error_report("Failed to run on vcpu %d", cpu->cpu_index);
> -- 
> 2.53.0

This is already fixed in upstream. checking the cpu->stop flag doesn't
seem to be required and other accelerators do not seem to check it
either. We agreed to drop this patch for now.