[edk2-devel] [PATCH v2 2/2] ArmVirtPkg/ArmVirtQemu: Avoid early ID map on ThunderX

[Marc Zyngier]

On Thu, 26 Jan 2023 14:46:13 +0000,
Gerd Hoffmann <kraxel at redhat.com> wrote:
> 
> On Thu, Jan 26, 2023 at 01:48:12PM +0000, Marc Zyngier wrote:
> > On Thu, 26 Jan 2023 12:01:51 +0000,
> > Gerd Hoffmann <kraxel at redhat.com> wrote:
> > > 
> > > > > Thanks for the koji link, Gerd.
> > > > >
> > > > > It works with 6.1.7-200.fc37.aarch64.
> > > > > Perfect.
> > > > >
> > > > > Thanks Ard, thanks Marc!
> > > > >
> > > > 
> > > > Thanks for the report. Up to this point, Marc is the only one who had
> > > > managed to reproduce this afaik, so having another data point where
> > > > the fix works as intended is rather nice.
> > > 
> > > One more data point:  This also gets things working again on the
> > > raspberry pi 3.  Well, sort of.  There still is some timer problem
> > > (it waits forever on the splash screen).  That might also be some
> > > qemu bug though ...
> > 
> > RPi3 as a host? Using the QEMU-provided GIC emulation?
> 
> Yes.
> 
> > I wouldn't be surprised if that broke ages ago as nobody really tests
> > this configuration.
> 
> Understandable.  1G is enough for simple smoke tests but not much
> beyond that.  And it is the odd child in the gang which hasn't a gic.

I suspect that it is that last point that has broken. The kernel
handling of that configuration has always been super sketchy, and it
would take almost nothing for that to break.

I'm almost thinking of writing an in-kernel GICv2 emulation just to
get rid of the rest of the hacks that have proliferated in the timer
and PMU code. But is it worth it for such an odd machine?

The alternative would be to move on and stop supporting KVM on this
wonky setup. :-/

> > If you can pinpoint the last state where this actually worked, I'll
> > try and have a look.
> 
> Don't have that at hand.  Need try downgrade stuff and see what I find.
> Tried with edk2 already back a year or so without success.  Didn't try
> downgrading qemu / kernel yet.

Right. Let me know if you get somewhere with that at some point.

> > But this sounds unrelated to the page fault handling that my patch was
> > fixing.
> 
> Yes, unrelated bug.  But the page fault bug reproduces on the rpi3 too,
> without the fix the firmware hangs before printing the first log line.
> Wanted report that because Ard mentioned it is hard to reproduce.

Right, thanks for that. Very interesting data point indeed.

Cheers,

	M.

-- 
Without deviation from the norm, progress is not possible.


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#99093): https://edk2.groups.io/g/devel/message/99093
Mute This Topic: https://groups.io/mt/96075174/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-