Re: [RfC PATCH] drm: fourcc byteorder: brings header file comments in line with reality.

From: Gerd Hoffmann <kraxel@redhat.com>
To: Pekka Paalanen <ppaalanen@gmail.com>
Cc: dri-devel@lists.freedesktop.org,
	open list <linux-kernel@vger.kernel.org>,
	amd-gfx@lists.freedesktop.org,
	Daniel Vetter <daniel.vetter@intel.com>,
	Ilia Mirkin <imirkin@alum.mit.edu>
Subject: Re: [RfC PATCH] drm: fourcc byteorder: brings header file comments in line with reality.
Date: Tue, 18 Apr 2017 22:50:18 +0200	[thread overview]
Message-ID: <1492548618.27392.72.camel@redhat.com> (raw)
In-Reply-To: <20170418170134.0a60d340@eldfell>

  Hi,

> Right. Very nice if we can trust the virtual machine at least getting
> things right, gives some chance for people to test anything. Except...
> that's a question of what kind of hardware the virtual machine
> emulates. The display device defines what endianess it uses on
> framebuffers, not the CPU, right?

The display device supports switching the endianess for the framebuffer,
at least with kernel 3.19+ and qemu 2.2+.  Default endianness depends on
the machine type, i.e. ppc64 guests get a bigendian framebuffer and
pretty much everything else a little endian framebuffer on reset.

The bochs-drm driver switches the display into native endian mode, i.e.
big endian for ppc64 and little endian for ppc64le kernels.

See commit 9ecdb039b7517dc10b8c3e6dbeb40859178ac28e

> > Well, I mean color glitches.  But it isn't consistent.  As if some
> > operations operate with the correct byteorder and some don't.
> > alpha/blue being swapped is a problem in some areas.
> > 
> > https://www.kraxel.org/tmp/
> 
> Ooh, yeah, that's definitely bonkers.
> 
> Maybe the 100% blue things are supposed to be a transparent blended
> overlays, like highlights.
> 
> The icons look somehow... not completely right to me. Somehow washed
> out?
> 
> Opaque gray shades are hard to tell right from wrong.
> 
> gnome-terminal and the wallpaper look right, but those might be the
> only things.
> 
> Having a compositing manager complicates things.

In some way yes, in some way no.  Tried wayland meanwhile (using
"gnome-shell --wayland").  Looks pretty much the same.  Window
decorations look a bit different (good on xorg, broken on wayland),
probably because window decorations work completely different.
Otherwise it is bug compatible to xorg.  Probably because gnome-shell
composites everything using llvmpipe, so it's largely the same code
running on both xorg and wayland, which then finally scans out to a dumb
framebuffer.

cheers,
  Gerd