[PATCH v4 5/5] drm/ofdrm: Support big-endian scanout buffers
Benjamin Herrenschmidt
benh at kernel.crashing.org
Mon Oct 17 06:44:05 UTC 2022
On Thu, 2022-10-13 at 09:39 +0200, Javier Martinez Canillas wrote:
> > In absence of such test results I think the most reasonable thing is to
> > keep the logic that nobody complained about for 10+ years.
> >
>
> I agree with Michal and Thomas on this. I don't see a strong reason to not
> use the same heuristic that the offb fbdev driver has been doing for this.
>
> Otherwise if this turns out to be needed, it will cause a regression for a
> user that switches to this driver instead. Specially since both fbdev and
> DRM drivers match against the same "display" OF compatible string.
I agree.
In the end, what it boils down to is, we don't know, we should guess. The
endianness of the kernel is as good a guess as anything here.
If not that, then assume BE.
That code was originally written for old macs. Those could simply not boot
anything other than a BE kernel. OF would always setup a 8bpp fb (so endianness
is irrelevant) but BootX could setup something else.
There's almost no old LE powerpc system out there, and I'm reasonably sure
actually none of any relevance to this code.
That leaves us with newer systems capable of endian switching. Those should just
get the property added.
I don't know of many cases out there. There' SLOS on powerpc which still won't
set it (which is what qemu uses). That could just be fixed. In the meantime
it makes sense for the kernel to follow its existing behaviour.
Cheers,
Ben.
More information about the dri-devel
mailing list