canonical-x team mailing list archive

[Bryce Harrington <bryce@xxxxxxxxxxxxx>]

----- Forwarded message from "Dodonov, Eugeni" <eugeni.dodonov@xxxxxxxxx> -----

Date: Wed, 22 Feb 2012 17:54:01 +0000
From: "Dodonov, Eugeni" <eugeni.dodonov@xxxxxxxxx>
Subject: RE: OSV / Intel Graphics Update - 02.22.2012

Today meeting minutes:

  *   Next Intel Linux Graphics stack release expected to be released this week. Its major components are Linux Kernel 3.2.x and Mesa 8.0.x. The other components (2D driver, cairo, vaapi) will not be updated in this release. The most notable features are: stable Ivy Bridge support, OpenGL 3.0 and GLSL 1.30 compatibility.
  *   Among the most notable changes on the kernel side for the past month, we have 4 fixes for Ivy Bridge hard-hangs under some GL workloads. They are already included in drm-intel-fixes (which will go into 3.3 kernel), and when the first 3.3-rc kernel with those patches comes out, we'll send them to inclusion into the 3.2.x kernel as well. With those patches, we are not aware of any Ivy Bridge hard-hangs in kernel at the moment.
  *   We have also figured out some issues on Sandy Bridge with RC6 enabled, which were solved by disabling Deep RC6. However, according to Canonical testing, even with plain RC6, there are still some issues on Sandy Bridge when RC6 is enabled - mostly with Samsung machines. So far, we suspect that the Bios is responsible for those - apparently, on such machines, the voltage thresholds for RC6 are set too aggressively, so when RC6 stage is entered and machine returns to the normal execution state, different corruption are in place or, in the worst case, the machine shuts down.
     *   Eugeni to continue investigating RC6 issues on Sandy Bridge and provide feedback on the next meeting.
  *   On Mesa project, Mesa 8.0 and 8.0.1 were released, with OpenGL 3.0 and GLSL 1.30 support as major enhancement. Mesa 8.0.1 has reworked HiZ support and multiple regression fixes on top of 8.0. The HiZ rework could potentially fix the random issues related to previous HiZ implementation in 8.0 observed by Canonical, as it isolates the way HiZ works in Mesa. Previously, it was implemented as a meta-op, so its operations could potentially occur in unexpected places. With this new version, HiZ was rewritten in a form of a specific batch which should not interfere with other parts of Mesa.
  *   Finally, a question was raised about the status of support for older Intel GPUs. Currently, our QA team does basic testing for pre-Ironlake architectures as part of their usual testing process unless a patch is specifically targeting an older architecture; and does a full round of testing for G33, GM965, G45, GM45, GMA 3150, Ironlake and newer GPUs for major stack releases. In any case, if you have issues related to older GPUs, those should be reported to us as usual.

Our next meeting will happen on March 14th. Among the proposed topics for that meetings are:

  *   Discuss the features to be included into our next major stack release (currently targeting the May 2012 time frame).

As always, if you have any additional topics you'd like to discuss or any opens not covered by our meetings, feel free to contact me!

Thanks!
Eugeni


----- End forwarded message -----