Re: [OpenBIOS] Re: [RFC PATCH 0/2] QEMU/openbios: PPC Software TLB support in the G4 family

[Segher Boessenkool]

Hi!

On Fri, Nov 26, 2021 at 09:34:44AM +0100, Cédric Le Goater wrote:
> On 11/25/21 10:38, Segher Boessenkool wrote:
> >On Thu, Nov 25, 2021 at 01:45:00AM +0100, BALATON Zoltan wrote:
> >>As for guests, those running on the said PowerMac G4 should have support
> >>for these CPUs so maybe you can try some Mac OS X versions (or maybe
> >
> >OSX uses hardware pagetables.
> >
> >>MorphOS but that is not the best for debugging as there's no source
> >>available nor any help from its owners but just to see if it boots it may
> >>be sufficient, it should work on real PowerMac G4).
> >
> >I have no idea what MorphOS uses, but I bet HPT as well.  That is
> >because HPT is fastest in general.  Software TLB reloads are good in
> >special cases only; the most common is real-time OSes, which can use its
> >lower guaranteed latency for some special address spaces (and can have a
> >simpler address map in general).
> 
> The support was added to QEMU knowing that Linux didn't handle soft TLBs.
> And the commit says that it was kept disabled initially. I guess that was
> broken these last years.

Ah :-)  So when was it enabled, do you know?

> $ git show 7dbe11acd807
> commit 7dbe11acd807
> Author: Jocelyn Mayer <l_indien@magic.fr>
> Date:   Mon Oct 1 05:16:57 2007 +0000
> 
>     Handle all MMU models in switches, even if it's just to abort because 
>     of lack
>       of supporting code.
>     Implement 74xx software TLB model.
>     Keep 74xx with software TLB disabled, as Linux is not able to handle 
>     TLB miss
>       on those processors.

This is very specifically for 7450, not 7400, fwiw.  7400 is a nice
core, while 7450 is ugly and asymmetric and unbalanced as hell.  It can
be faster though ;-)


Segher