[RFC] GPU device layer patchset (00/07)

[RFC] GPU device layer patchset (00/07)

[Dave Airlie]

This patchset contains 7 patches to implement the GPU device layer.

0001-drm-remove-local-copies-of-pci-bus-slot-func.txt
0002-drm-remove-pci-domain-local-copy.txt
0003-gpu-Initial-GPU-layer-addition.txt
0004-gpu-radeon-add-a-radeon-lowlevel-GPU-driver.txt
0005-gpu-radeonfb-add-GPU-support-to-radeonfb.txt
0006-gpu-drm-Add-GPU-layer-support-to-generic-DRM.txt
0007-drm-gpu-radeon-Add-radeon-DRM-support-to-use-GPU-layer.txt

The first two patches are changes to the DRM layer that I will submit separately
later but are required for this work.

The GPU layer generic code is in 0003 and 0006, radeon gpu driver 0004,
radeonfb in 0005 and radeon drm in 0007.

The code is also available in a git repo branch: 'gpu-branch'
git://git.kernel.org/pub/scm/linux/kernel/git/airlied/gpu-2.6

This is the initial implementation for review.

Dave.

Re: [RFC] GPU device layer patchset (00/07)

[Jeff Garzik]

Dave Airlie wrote:
> This patchset contains 7 patches to implement the GPU device layer.

Where can I find more info on why this might be nice?
Or simply more info on what this actually does for me?

	Jeff

Re: [RFC] GPU device layer patchset (00/07)

[Dave Airlie]

>
> Where can I find more info on why this might be nice?
> Or simply more info on what this actually does for me?

It does nothing for you yet, it is just step one for getting the DRM and 
framebuffer drivers to co-exist in the driver model world, with the next 
step being allowing a channel of communication between the two layers with 
a view that later I can ask the DRM to disable fbdev or pass info to it..

Why do we not want fbdev and DRM in one driver? as it would break way too 
many existing systems..

It also allows DRM to get called at suspend/resume time.

Dave.

-- 
David Airlie, Software Engineer
http://www.skynet.ie/~airlied / airlied at skynet.ie
Linux kernel - DRI, VAX / pam_smb / ILUG

[RFC] GPU device layer patchset (00/07)

[Louis E Garcia II]

I believe their is some confusion over the use of graphics and video. It
might seem logical to use video for video cards but we also have another
use of video for video capture. The source tree has the drm
under /drivers/video but /sys has class/graphics for fb and
class/drm. /dev has /drm /fb.

Is is possible to clean some of this up? Since you are starting on a
rework can you put the gpu device layer source under /drivers/graphics
and maybe move the drm to /drivers/graphics/drm? Frame buffer could
possibly be moved under graphics too. 

The gpu layer can register /sys/class/graphics and udev can still
create /dev/dri/cardX and /dev/fbX. We can create a new dev node
maybe /dev/graphics/cardX or /dev/graphics/{drmX,fbX} for transition.

I really don't want to get in the middle of this as I am not that
technical but I wanted to share my thoughts.

-Louis

Re: [RFC] GPU device layer patchset (00/07)

[Pavel Machek]

Hi!

> >Where can I find more info on why this might be nice?
> >Or simply more info on what this actually does for me?
> 
> It does nothing for you yet, it is just step one for getting the DRM and 
> framebuffer drivers to co-exist in the driver model world, with the next 
> step being allowing a channel of communication between the two layers with 
> a view that later I can ask the DRM to disable fbdev or pass info to it..
> 
> Why do we not want fbdev and DRM in one driver? as it would break way too 
> many existing systems..
> 
> It also allows DRM to get called at suspend/resume time.

Looks good to me. I guess I should get DRM working...

I actually tried a bit, and got some reasonably-recent Xserver here,
but now I get

root@amd:/home/pavel# glxinfo
name of display: :0.0

ERROR!  sizeof(I830DRIRec) does not match passed size from device
driver
libGL warning: 3D driver returned no fbconfigs.
libGL error: InitDriver failed
libGL error: reverting to (slow) indirect rendering
display: :0  screen: 0
direct rendering: No
server glx vendor string: SGI
server glx version string: 1.2
server glx extensions:

:-(
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html