Re: [PATCH-for-9.2? v2 1/2] hw/display/vga: Do not reset 'big_endian_fb' in vga_common_reset()

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Fri, 2024-12-06 at 10:28 +0100, Gerd Hoffmann wrote:
> > 
> > OPAL (well skiboot) doesn't display anything anyways (or at least
> > it
> > didn't when I wrote it :-). It just boots Linux as a bootloader. So
> > as
> > long as Linux itself sets the register it should be fine.
> 
> Oh, mixed up the firmware names, it's SLOF not OPAL.
> 
> sorry for the confusion,

Looking at SLOF source, it doesn't know about the endian register at
all indeed, it also doesn't put an endian propery in the device-node,
and probably expects the fb to be BE on reboot. Gosh I also wrote that
code... a looong time ago, I had forgotten all about it.

On SPAPR where endian *is* switched, qemu will also switch the endian
of the framebuffer on an explicit mode switch via the H_SET_MODE
hypercall. But afaik that does NOT include a reboot (at least from 10mn
browsing the code).

So I think the patch will have the effect of breaking the framebuffer
in SLOF on reboot when using SPAPR with an LE OS.

Something like this might work around it:

--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -1737,6 +1737,9 @@ static void spapr_machine_reset(MachineState *machine, ResetType type)
 
     spapr_clear_pending_events(spapr);
 
+    /* Switch VGA to big endian */
+    spapr_pci_switch_vga(spapr, true);
+
     /*
      * We place the device tree just below either the top of the RMA,
      * or just below 2GB, whichever is lower, so that it can be

I can't test right now, I might later this week, ping me if you don't
hear from me.

Cheers,
Ben.