Testing latest -intel in Ubuntu

[bouazza]

When installing the xorg-edgers xserver 1.6 beta package:
For the moment it is important to install ("downgrade" to) the -evdev and -synaptics from the PPA, not the ones from the main archive which are built for the old xserver.

Thanks, it works now and the glxgears FPS has inscreased to 900 *-* ( it was 400~500 before ) .

Hans on launchpad is talking about 1300~1500 FPS, is that possible ? So that means that we can get those values or less with the official intel/gem drivers ?

( My card is an Intel 965GM ( X3100 ) )

[Biji]

how to instal xorg-server beta? i have added repository and only intel driver is updated

[AliBaba]

Could someone test a 3D intensive application...

To be honest, you don't even need a 3D intensive application to trigger freeze bugs of all kinds.


I've been trying 2.6.27/2.6-git with and without GEM patches, DRM patches, different repositories (arlied, anholt), Mesa from 7.0 to current head and have been able to freeze X and keyboard input as reported in several bugs (e.g: http://bugs.freedesktop.org/show_bug.cgi?id=18640) quite reliably.


The freezing problems do of course only apply to those rare instances where Compiz does not make the whole X display go mad and draw everything white.


Not that I am against development and evolution - I really can't wait for DRI2 and GEM to be stable - but I'd really appreciate at least one driver/mesa/kernel combination that I could fall back to after experimenting. There is none.
Unfortunately I'm stuck to Metacity after being used to Compiz and it's great abilities.



I'm using 4500MHD and GM45, FYI.
[/rant_mode]

[tormod]

how to instal xorg-server beta? i have added repository and only intel driver is updated

You must be using Jaunty to upgrade to xserver 1.6 beta from xorg-edgers.

Note that in general I can not recommend anyone to install beta packages if you're not sure how to uninstall them and revert to official versions or how to recover from a broken X setup.

[tormod]

Not that I am against development and evolution - I really can't wait for DRI2 and GEM to be stable - but I'd really appreciate at least one driver/mesa/kernel combination that I could fall back to after experimenting. There is none.
Unfortunately I'm stuck to Metacity after being used to Compiz and it's great abilities.

Can't you go back to your old configuration where you had Compiz working?

[tormod]

I have come to the understanding that the drm modules from libdrm git (drm-modules-source package) are not up to date, and the stock modules in the 2.6.28 kernel might be better. This means you can skip the compiling modules part (module-assistant). Please uninstall the drm-modules-2.6.28* package if you built and installed it with module-assistant.

If this however make things worse, please report.

[AliBaba]

Can't you go back to your old configuration where you had Compiz working?

Yeah, sorry, I didn't state that correctly. I never had Compiz running on this notebook without issues - which I had to buy after my old one just died recently (r300 & Compiz).



Anyway, to give you a short log of my nightly activities (call me a geek :-)):

Sources:

• drm - git/modesetting-gem
• kernel - linus' git + anholt's drm repository
• xf86-video-intel - git
• mesa - git

Following built successfully (using the debian-way):

• libdrm and stuff
• kernel
• xserver-xorg-video-intel
• libgl1-mesa-dri/glx, lubglu1-mesa

After trying for about 3 hours to get the mesa-stuff to only build what I need (intel, amd64) I finally have a system at hand that does not take a few hours for building all those mesa-packages nobody needs.

2.6.28-git+intel-drm-next boots successfully and module i915 isn't there - as expected.


But, and this is the big caveat, I can't - and never have been able to - build drm-modules-source. It always stops at an implicit declaration of function "pci_read_base" in i915_gem_tiling.c

The latest drm sources don't even get me that far so I had to use "modesetting-gem.12e68f80-0ubuntu0tormod".

[AliBaba]

I have come to the understanding that the drm modules from libdrm git (drm-modules-source package) are not up to date, and the stock modules in the 2.6.28 kernel might be better.

Well, there is only _one_ big problem:

As far as I understood, 2.6.28 does not hold everything needed for kernel-based modesetting. So in order to have it in place, one must use 2.6.28 + anholt's repository (at least this is what Anholt himself wrote, read here).

Well, and here's the funny (Hahaha!) thing about it: The module needed, i915, was taken out of that repository in order to use the one provided in modesetting-gem branch of freedesktop/drm.


So you can either use:
2.6.28 and i915 - no modesetting or
2.6.28+anholt, without i915 - and fail at building i915.


Please correct me, if I'm wrong.

[tormod]

Anyway, to give you a short log of my nightly activities (call me a geek :-)):

Oh, I have the greatest understanding for this

But, and this is the big caveat, I can't - and never have been able to - build drm-modules-source. It always stops at an implicit declaration of function "pci_read_base" in i915_gem_tiling.c

As I noted in my previous post, you should not build drm modules from libdrm git (= drm-modules-source) if you are on 2.6.28.

[tormod]

Well, and here's the funny (Hahaha!) thing about it: The module needed, i915, was taken out of that repository in order to use the one provided in modesetting-gem branch of freedesktop/drm.

Yep, the kernel mode setting is messy nowadays. I will revert to the master trunk for the next drm-snapshot upload. The modesetting branch was tried merged onto master but caused to much brokenness and was reverted. A more conservative patch is still in discussion on the dri-devel ML.