RE: [PATCH v7 09/10] hisi_acc_vfio_pci: Add support for VFIO live migration

[Shameerali Kolothum Thodi]

> -----Original Message-----
> From: Jason Gunthorpe [mailto:jgg@xxxxxxxxxx]
> Sent: 03 March 2022 13:20
> To: Shameerali Kolothum Thodi <shameerali.kolothum.thodi@xxxxxxxxxx>
> Cc: kvm@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> linux-crypto@xxxxxxxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx;
> alex.williamson@xxxxxxxxxx; cohuck@xxxxxxxxxx; mgurtovoy@xxxxxxxxxx;
> yishaih@xxxxxxxxxx; Linuxarm <linuxarm@xxxxxxxxxx>; liulongfang
> <liulongfang@xxxxxxxxxx>; Zengtao (B) <prime.zeng@xxxxxxxxxxxxx>;
> Jonathan Cameron <jonathan.cameron@xxxxxxxxxx>; Wangzhou (B)
> <wangzhou1@xxxxxxxxxxxxx>
> Subject: Re: [PATCH v7 09/10] hisi_acc_vfio_pci: Add support for VFIO live
> migration
> 
> On Thu, Mar 03, 2022 at 01:17:05PM +0000, Shameerali Kolothum Thodi
> wrote:
> > > Tthere is a scenario that transfers only QM_MATCH_SIZE in stop_copy?
> > > This doesn't seem like a good idea, I think you should transfer a
> > > positive indication 'this device is not ready' instead of truncating
> > > the stream. A truncated stream should not be a valid stream.
> > >
> > > ie always transfer the whole struct.
> >
> > We could add a 'qm_state' and return the whole struct. But the rest
> > of the struct is basically invalid if qm_state = QM_NOT_REDAY.
> 
> This seems like the right thing to do to me
> 
> > > > Looks like setting the total_length = 0 in STOP_COPY is a better
> > > > solution(If there are no other issues with that) as it will avoid
> > > > grabbing the state_mutex as you mentioned above.
> > >
> > > That seems really weird, I wouldn't recommend doing that..
> >
> > Does that mean we don't support a zero data transfer in STOP_COPY?
> 
> total_length should not go backwards
> 
> > The concern is if we always transfer the whole struct, we end up reading
> > and writing the whole thing even if most of the data is invalid.
> 
> Well, you can't know if the device is not ready or not until the
> reaching STOP_COPY, so you have to transfer something to avoid
> truncating the data stream. The state here is tiny, is the extra
> transfer a real worry?

Ok, don't think so. I will introduce the 'qm_state' and handle it as discussed
above then.

Thanks,
Shameer