Re: ppc pbr403 vmstate

[Cédric Le Goater]

On 1/17/22 06:52, David Gibson wrote:
> On Fri, Jan 14, 2022 at 08:07:21AM +0100, Cédric le Goater wrote:
> > On 1/14/22 00:41, David Gibson wrote:
> > > On Thu, Jan 13, 2022 at 06:51:56PM +0000, Dr. David Alan Gilbert wrote:
> > > > Hi,
> > > >     Is there any easy way of getting a machine where the pbr403 vmstate
> > > > would be generated?
> > >
> > > The condition in pbr403_needed is...
> > >
> > >       return (pvr & 0xffff0000) == 0x00200000;
> > >
> > > .. which looks to be the PVR for ppc403 models.  That makes sense with
> > > the section name... but not so much with the fact that it's under
> > > cpu/tlb6xx.  The 6xx MMU is basically unrelated to the 40x MMU.  But
> > > it looks like the vmstate_tlbemb might be shared between then, because
> > > of bad ideas of the past.
> > >
> > > But in any case, we already dropped what little 403 support we ever
> > > had - there's nothing with that PVR even listed in
> > > target/ppc/cpu-models.h.
> > >
> > > So I think we should just drop it.
> >
> > yes. But we can not remove env.pb since this would break migration
> > compatibility, correct ?
>
> Only if it appears in a migration section that's actually emitted by a
> supported machine type.  As far as I can tell the only section that
> does that is vmstate_pbr403, which we're also dropping so we should be
> fine.

I sent a patch to remove vmstate_pbr403 first.

 > It is also touched in the *super* old cpu_load_old.  I suspect we
> could probably just drop that completely, since I don't think we
> realistically support migration from a version that old anyway.  But
> even if we don't want to do that right now, we can just replace the
> reads into env->pb with discarding reads and we'll be fine.  We don't
> implement any cpus that actually used those fields, so we can ignore
> them in the migration stream.

I will take a look at this also with follow ups.

Thanks,

C.