More ATI KMS Improvements For Linux 2.6.32

[Louise]

The other videos are available at:

http://streaming.linux-magazin.de/en/archive.htm

http://events.linuxfoundation.org/multimedia/archive

[Louise]

So... check this off as "DONE"

I watched the LinuxConf Roundtable talk, and it was interesting how the core Linux developers look at KMS a bit differently

http://video.linuxfoundation.org/video/1550

Kernel developers are crazy people

Very funny talk btw!

[airlied]

Phoronix: More ATI KMS Improvements For Linux 2.6.32

There's already a fair amount of DRM changes in the Linux 2.6.32 kernel including ATI R600/700 KMS and 3D support, but now nearly half-way into the Linux 2.6.32 development cycle there is a huge pull request of new ATI kernel mode-setting code. With this new code, all ATI GPUs with kernel mode-setting are using a newer and cleaner initialization path while dropping support for the current paths. Lots of this init path work was recently done by Jerome Glisse and others, which is designed to give better suspend-and-resume on most ASICs along with other advantages. This code also has new work for kernel mode-setting with AGP-based R600 graphics cards, a multi-card memory management fix, single CRTC server GPU support, and R600 stability fixes. This code has over 3,000 lines of new code that is landing late in the Linux 2.6.32 cycle, which will likely not please Linus...

http://www.phoronix.com/vr.php?view=NzU5MA

Yet again Michael displays a stunning lack of clue about kernel development processes. If you actually read the pull request you can
see I separate out the non-staging changes and the changes to radeon kms
which are under staging. Linus has no problems with kms development happening in real time in his tree and keeping it as close as possible to what I am shipping in Fedora. When KMS is stable enough we'll move it to non-staging and then he'd get angry if I did this.

Dave.

[bridgman]

AFAIK the hardware-specific portions of the kernel graphics driver are usually built into a separate dynamically loadable module ("modular drm") so the KMS drivers don't contribute much to the size of the core kernel image.

So... check this off as "DONE"

[tball]

i mean the open source drivers, but anyway having a simple support to let you turn on desktop and change to other if you change graphic card

Sounds like a bad plan to me. The only vendor not supporting the inkernel oss support is nvidia. So why make a whole specialkernel just for the sake of nvidia?

Remember that intel is on the kms part also.

[Chewi]

Hardware drivers generally only get loaded if you have the hardware, otherwise they just take up (relatively little) space on your hard drive.

[NomadDemon]

i mean the open source drivers, but anyway having a simple support to let you turn on desktop and change to other if you change graphic card

[bridgman]

The discussion is actually about the open source driver code, which is much smaller than the proprietary drivers.

Kernel portions of the proprietary drivers do not go into the kernel tree; they are always delivered separately.

[NomadDemon]

maybe theres should be 2 kinds of kernels? with A and with N in name A for ATI, N for NVIDIA..to not to load 2 big drivers

[Chewi]

I tried KMS with 2.6.32-rc3 but experienced many freezes. I did see this new stuff waiting in the wings and was really hoping it would make it into rc4. Please, Linus!