Re: [PATCH 2/2] virtio-balloon: Reject qmp_balloon during hot-unplug

[David Hildenbrand]

On 23.01.20 16:46, Julia Suvorova wrote:
> On Thu, Jan 23, 2020 at 3:17 PM David Hildenbrand <address@hidden> wrote:
>>
>> On 23.01.20 15:08, Julia Suvorova wrote:
>>> On Thu, Jan 16, 2020 at 1:36 PM David Hildenbrand <address@hidden> wrote:
>>>>
>>>> On 15.01.20 23:40, Julia Suvorova wrote:
>>>>> Hot-unplug takes some time due to communication with the guest.
>>>>> Do not change the device while freeing up resources.
>>>>>
>>>>> Signed-off-by: Julia Suvorova <address@hidden>
>>>>> ---
>>>>>  balloon.c                  | 2 +-
>>>>>  hw/virtio/virtio-balloon.c | 9 ++++++++-
>>>>>  include/sysemu/balloon.h   | 2 +-
>>>>>  3 files changed, 10 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/balloon.c b/balloon.c
>>>>> index f104b42961..998ec53a0f 100644
>>>>> --- a/balloon.c
>>>>> +++ b/balloon.c
>>>>> @@ -119,5 +119,5 @@ void qmp_balloon(int64_t target, Error **errp)
>>>>>      }
>>>>>
>>>>>      trace_balloon_event(balloon_opaque, target);
>>>>> -    balloon_event_fn(balloon_opaque, target);
>>>>> +    balloon_event_fn(balloon_opaque, target, errp);
>>>>>  }
>>>>> diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c
>>>>> index 57f3b9f22d..0fa4e4454b 100644
>>>>> --- a/hw/virtio/virtio-balloon.c
>>>>> +++ b/hw/virtio/virtio-balloon.c
>>>>> @@ -717,12 +717,19 @@ static void virtio_balloon_stat(void *opaque, 
>>>>> BalloonInfo *info)
>>>>>                                               VIRTIO_BALLOON_PFN_SHIFT);
>>>>>  }
>>>>>
>>>>> -static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
>>>>> +static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
>>>>> +                                     Error **errp)
>>>>>  {
>>>>> +    DeviceState *bus_dev = qdev_get_bus_device(DEVICE(opaque));
>>>>>      VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
>>>>>      VirtIODevice *vdev = VIRTIO_DEVICE(dev);
>>>>>      ram_addr_t vm_ram_size = get_current_ram_size();
>>>>>
>>>>> +    if (bus_dev && bus_dev->pending_deleted_event) {
>>>>> +        error_setg(errp, "Hot-unplug of %s is in progress", vdev->name);
>>>>> +        return;
>>>>> +    }
>>>>> +
>>>>
>>>> How exactly does this help? The guest is free to inflate/deflate
>>>> whenever it wants.
>>>
>>> Guest is aware of hot-unplug start, and virtio driver should not
>>> initiate any operations. This patch just restricts issuing commands
>>> from qmp monitor.
>>
>> Why shouldn't the guest driver inflate/deflate while memory hotplug is
>> going on?
>>
>> Simple balloon compaction in a Linux guest -> deflate/inflate triggered
>> in the hypervisor.
> 
> QEMU crashes if inflate happens after powering-off PCI slot. Guest is
> unable to interact with virtio-balloon device then, driver is
> unloaded. But inflate can still happen if initiated from qmp.

Ah guest can inflate/deflate whenever it wants. What am I missing?

IMHO, the right approach is to discard inflate/deflate requests on these
areas instead.


-- 
Thanks,

David / dhildenb