Linux Virtualization - Apr 2020 - [virtio-dev] Re: [PATCH] virtio-balloon: Disable free page hinting/reporting if page poison is disabled

On 17.04.20 12:29, Michael S. Tsirkin wrote:
> On Fri, Apr 17, 2020 at 12:26:24PM +0200, David Hildenbrand wrote:
>> On 17.04.20 12:19, Michael S. Tsirkin wrote:
>>> On Fri, Apr 17, 2020 at 12:09:38PM +0200, David Hildenbrand wrote:
>>>>  > What do you call "hinting ends" though? The
fact we put
>>>>> a page in the VQ is not a guarantee that it's been
consumed
>>>>> by the hypervisor.
>>>>>
>>>>
>>>> I'd say hinting ends once the hypervisor sets
FREE_PAGE_REPORT_S_DONE.
>>>
>>> Can't find that one anywhere. what did I miss?
>>
>> Sorry, the QEMU implementation is confusing. FREE_PAGE_REPORT_S_DONE is
>> translated to VIRTIO_BALLOON_CMD_ID_DONE
> 
> Well VIRTIO_BALLOON_CMD_ID_DONE just means "don't give me any
> more hints, I finished migration".
> Guest will stop hinting even without that once it scans all
> free memory.
Yeah, that's the end of the whole process where you can be sure the host
processed all requests definetly.

-- 
Thanks,

David / dhildenb

On Fri, Apr 17, 2020 at 12:31:14PM +0200, David Hildenbrand
wrote:
> On 17.04.20 12:29, Michael S. Tsirkin wrote:
> > On Fri, Apr 17, 2020 at 12:26:24PM +0200, David Hildenbrand wrote:
> >> On 17.04.20 12:19, Michael S. Tsirkin wrote:
> >>> On Fri, Apr 17, 2020 at 12:09:38PM +0200, David Hildenbrand
wrote:
> >>>>  > What do you call "hinting ends" though?
The fact we put
> >>>>> a page in the VQ is not a guarantee that it's been
consumed
> >>>>> by the hypervisor.
> >>>>>
> >>>>
> >>>> I'd say hinting ends once the hypervisor sets
FREE_PAGE_REPORT_S_DONE.
> >>>
> >>> Can't find that one anywhere. what did I miss?
> >>
> >> Sorry, the QEMU implementation is confusing.
FREE_PAGE_REPORT_S_DONE is
> >> translated to VIRTIO_BALLOON_CMD_ID_DONE
> > 
> > Well VIRTIO_BALLOON_CMD_ID_DONE just means "don't give me any
> > more hints, I finished migration".
> > Guest will stop hinting even without that once it scans all
> > free memory.
> 
> Yeah, that's the end of the whole process where you can be sure the
host
> processed all requests definetly.
It's not guaranteed to happen :) Sending an interrupt at the end
of each scan doubles the overhead ...
> -- 
> Thanks,
> 
> David / dhildenb

On 17.04.20 13:02, Michael S. Tsirkin wrote:
> On Fri, Apr 17, 2020 at 12:31:14PM +0200, David Hildenbrand wrote:
>> On 17.04.20 12:29, Michael S. Tsirkin wrote:
>>> On Fri, Apr 17, 2020 at 12:26:24PM +0200, David Hildenbrand wrote:
>>>> On 17.04.20 12:19, Michael S. Tsirkin wrote:
>>>>> On Fri, Apr 17, 2020 at 12:09:38PM +0200, David Hildenbrand
wrote:
>>>>>>  > What do you call "hinting ends" though?
The fact we put
>>>>>>> a page in the VQ is not a guarantee that it's
been consumed
>>>>>>> by the hypervisor.
>>>>>>>
>>>>>>
>>>>>> I'd say hinting ends once the hypervisor sets
FREE_PAGE_REPORT_S_DONE.
>>>>>
>>>>> Can't find that one anywhere. what did I miss?
>>>>
>>>> Sorry, the QEMU implementation is confusing.
FREE_PAGE_REPORT_S_DONE is
>>>> translated to VIRTIO_BALLOON_CMD_ID_DONE
>>>
>>> Well VIRTIO_BALLOON_CMD_ID_DONE just means "don't give me
any
>>> more hints, I finished migration".
>>> Guest will stop hinting even without that once it scans all
>>> free memory.
>>
>> Yeah, that's the end of the whole process where you can be sure the
host
>> processed all requests definetly.
> 
> It's not guaranteed to happen :) Sending an interrupt at the end
> of each scan doubles the overhead ...
Yeah, AFAIKs you either get a VIRTIO_BALLOON_CMD_ID_STOP or a
VIRTIO_BALLOON_CMD_ID_DONE at the end. And AFAIK both will guarantee
that all previous hints were handled.

... but reconstructing this protocol from code is probably too much for
a friday afternoon, lol

-- 
Thanks,

David / dhildenb