Re: [PATCH v1 2/2] Drivers: virtio: balloon: Report inflated memory

[Alexander Atanasov]

Hi,

On 9.08.22 20:44, Nadav Amit wrote:
> On Aug 9, 2022, at 2:53 AM, Alexander Atanasov <alexander.atanasov@xxxxxxxxxxxxx> wrote:
>
> > Update the value in page_alloc on balloon fill/leak.
>
> Some general comments if this patch goes forward.
>
> Please cc pv-drivers@xxxxxxxxxx in the future.

Ok.

> > Cc: David Hildenbrand <david@xxxxxxxxxx>
> > Cc: Wei Liu <wei.liu@xxxxxxxxxx>
> > Cc: Nadav Amit <namit@xxxxxxxxxx>
> >
> > Signed-off-by: Alexander Atanasov <alexander.atanasov@xxxxxxxxxxxxx>
> > ---
> > drivers/virtio/virtio_balloon.c | 13 +++++++++++++
> > 1 file changed, 13 insertions(+)
> >
> > Firts user, other balloons i will do if it is accepted to avoid too much emails.
> >
> >
> > diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
> > index b9737da6c4dd..e2693ffbd48b 100644
> > --- a/drivers/virtio/virtio_balloon.c
> > +++ b/drivers/virtio/virtio_balloon.c
> > @@ -208,6 +208,16 @@ static void set_page_pfns(struct virtio_balloon *vb,
> > 					  page_to_balloon_pfn(page) + i);
> > }
> >
> > +static void update_meminfo(struct virtio_balloon *vb)
>
> Putting aside the less-than-optimal function name, I would like to ask that

Right, i will think of a better one.

> any new generic balloon logic would go into balloon_compaction.[hc] as much

If it is going to be a place for generic logic may be it should be 
renamed to balloon_common.[ch] ?

> as possible. I made the effort to reuse this infrastructure (which is now
> used by both VMware and virtio), and would prefer to share as much code as
> possible.
>
> For instance, I would appreciate if the update upon inflate would go into
> balloon_page_list_enqueue() and balloon_page_enqueue(). VMware's 2MB pages
> logic is not shared, so it would require a change that is specific for
> VMware code.

I looked at the code and i do not see how i can reuse it since
BALLOON_COMPACTION can be disabled and as you say even for VMWare it 
would require updates on other places. Looks like if i do so i would 
have to handle update from each driver for both cases. I think it is 
better to clearly mark the spots when drivers do their internal 
recalculations and report to the core. I see only VMWare is using
balloon_page_list_enqueue , virtio balloon is using only migration and
i don't see how to hook it there - i haven't checked the rest of the 
balloons but i guess it would be similiar . I agree it is a good to have 
a common place for such logic but it might be better of for a separate 
work in the future.

--
Regards,
Alexander Atanasov