Re: [Qemu-devel] [PATCH v4] virtio-scsi: restart DMA after iothread

[Kevin Wolf]

Am 17.06.2019 um 19:23 hat Paolo Bonzini geschrieben:
> On 17/06/19 14:29, Kevin Wolf wrote:
> > But in the end, if Paolo feels strongly that for whatever reason
> > propagating events through the real device tree isn't good, let's get
> > the bug fixed with whatever hack it takes.
> 
> It is actually good, but the implementation in hw/scsi is ugly because
> it singles out virtio-scsi - whereas the rule should simply be that the
> HBA is stopped after the disks and started before.  Having the HBA do
> something special if it cares about the order is the part that I didn't
> like.

The alternative would be to add a VM state change handler to all other
HBAs, too, so that scsi-bus never has to register its own handler. I
don't think we have that many HBAs, so maybe it's actually workable.

Of course, this makes me think that maybe for the actual proper
solution, VM state change handlers should really be a feature that qdev
provides for devices. Then if a HBA doesn't have anything else to do,
the qdev core would just recurse into the children right away; if it has
something to do, it would disable the device after its children, and
re-enable it before the children.

However, I'm afraid we're talking about a major feature then and not
about a simple fix any more. :-(

Kevin