Re: [PATCH 1/1] cpus: resume hotplugged vCPU only when the guest is running

[Dongli Zhang <dongli.zhang@oracle.com>]

Ping?

This can prevent un-predictable symptoms during Libvirt/QEMU live migration,
especially when hotpluggable='yes' for vCPU in Libvirt XML file.

For instance, vCPU hotpluggable='yes' is equivalent to:

1. Create source QEMU with "-smp 1,maxvcpus=4". Keep QEMU in prelaunch state.

2. Cold-plug vCPUs=1,2,3 to source QEMU (prelaunch).

3. Continue source QEMU to running status.

4. Create target QEMU with "-smp 1,maxvpus=4". Target QEMU remains
prelaunch/stopped.

5. Cold plug vCPUs=1,2,3 to target QEMU. They are expected to remain in stopped
status.

Unfortunately, due to the bug, vCPUs=1,2,3 are in running status. There are
chances for them to start running in KVM guest mode before target QEMU is fully
resumed during live migration.

As a result, anything abnormal can happen.

Thank you very much!

Dongli Zhang

On 10/10/25 2:36 PM, Dongli Zhang wrote:
> When a new vCPU is hotplugged, cpu->stopped is unconditionally set to false
> by cpu_common_realizefn().
> 
> However, there are scenarios where the guest is not running, i.e., when the
> guest has been stopped via the HMP 'stop' command, or when the instance is
> a live migration target started with "-incoming defer". In these cases, all
> existing vCPUs have (cpu->stopped == true), except for the newly hotplugged
> vCPU.
> 
> Unpause the hotplugged vCPU only when the guest is running.
> 
> Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
> ---
>  hw/core/cpu-common.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/core/cpu-common.c b/hw/core/cpu-common.c
> index 8c306c89e4..789382cad5 100644
> --- a/hw/core/cpu-common.c
> +++ b/hw/core/cpu-common.c
> @@ -30,6 +30,7 @@
>  #include "qemu/target-info.h"
>  #include "exec/log.h"
>  #include "exec/gdbstub.h"
> +#include "system/runstate.h"
>  #include "system/tcg.h"
>  #include "hw/boards.h"
>  #include "hw/qdev-properties.h"
> @@ -263,7 +264,10 @@ static void cpu_common_realizefn(DeviceState *dev, Error **errp)
>  
>      if (dev->hotplugged) {
>          cpu_synchronize_post_init(cpu);
> -        cpu_resume(cpu);
> +
> +        if (runstate_is_running()) {
> +            cpu_resume(cpu);
> +        }
>      }
>  
>      /* NOTE: latest generic point where the cpu is fully realized */