Re: [Qemu-devel] [PATCH] msi: remove return code for msi_init()

[Peter Xu]

On Tue, May 30, 2017 at 04:29:57PM +0200, Paolo Bonzini wrote:
> 
> 
> On 30/05/2017 16:28, Markus Armbruster wrote:
> > Paolo Bonzini <address@hidden> writes:
> > 
> >>> Note that a board that doesn't support MSI can take MSI-capable devices
> >>> just fine.  Only the broken boards can't.
> >>>
> >>> Obviously, broken boards should be fixed.  Once they all are, we can
> >>> (and should!) remove msi_nonbroken.
> >>
> >> That only works if we know what the broken boards are.
> > 
> > Yes.
> > 
> >> Right now, all boards that do not support MSI hide the capability, which
> >> is wrong.
> > 
> > Agreed.
> > 
> >>            I'd prefer to remove msi_nonbroken completely if we don't
> >> know where the problem is.
> > 
> > So you're proposing to (1) remove msi_nonbroken, (2) see which boards
> > burst into flames, and (3) fix them, or perhaps add a less wrong stop
> > gap msi_broken just for them?
> 
> Yes, adding back msi_broken is one "git revert" away.

Not sure whether I got the point here, but... Adding msi_broken is not
a "git revert"? Since there is no msi_broken before, only
msi_supported. And iirc if we want to provide one msi_broken we need
to know exactly all the borads that are broken with MSI, and that
blacklist is something we don't have now (as you mentioned above)?

Please correct me if I misunderstood.

> 
> Of course, this means the edu memory leak should be fixed in a separate
> small patch.
> 
> Thanks,
> 
> Paolo

Thanks,

-- 
Peter Xu