Nouveau Companion 31

In this Thanksgiving Day issue of the Nouveau Companion, the Nouveau team goes over what's left to do for a 2D stable driver release, the current work being done on the driver, and an assortment of Gallium3D and RandR 1.2 work. The stable 2D Nouveau driver wasn't announced in this issue, but it's coming soon.

Introduction

Dear audience, welcome back to the newest issue of the TiNDC. The last issue caught the eye of some mainstream tech news sites in Germany. Although it was a bit premature to call 2D mostly done then, we basically have everything done except for mode-setting and thus a better Randr1.2 support. Currently you may be lucky on NV4x and NV3x but earlier cards and NV5x don't work very good. Two German links, Golem and Pro-Linux.

While work on 2D is slowly ceasing (exceptions: see above), attention is shifting to 3D. Darktama and jkolb are working on Gallium and DRM / TTM issues. Marcheu is still fighting the generic Gallium3D frame work for older cards. Ahuillet, pmdata, and p0g are just waiting for them to finish so that they finally can start coding 3D features.

The Current Status

Marcheu was busy the last weeks, he worked with Benh and various testers on PPC bugs. When doing DFS (DownloadFromScreen) / UTS (UploadToScreen) DMA transfers hang but DMA otherwise worked, as long as AGP wasn't activated. IFC (ImageFromCPU) had an error with inverted logic indicating an error when in fact everything went fine. This bug was spotted by Benh who continued helping us to get PPC working for all of our testers. Additionally, Marcheu got online access to a PPC machine for testing. He is currently waiting for some register dumps, which currently do hang and lock the machine unexpectedly. So he needs a person in front of the keyboard of the remote machine.

Those dumps will hopefully enable nouveau to work on chimaera's machine, who still reported some startup hangs with nouveau on PowerPC. After some experimental patches were exchanged, marcheu managed to fix that only to have nouveau hang a little bit later during the startup. It turned out that this was due to some endian bugs in the bios parser. So the bug was handed over to malc0 who did fix some bugs in his parser.

The BIOS parser fixes to honor endianess plus some additional code clean ups (aka deobfuscation) were pushed by malc0. Together with stillunknown they further cleaned up the randr1.2 code and fixed some more bugs.

Stillunknown is still trying to get randr-1.2 working on AndrewR's card. Many test patches were exchanged but still, randr1.2 does not work for AndrewR's NV3x based card.

Although stillunknown did find a lot of errors, he talked with Thunderbird and both compared their wisdom about various registers controlling display output. That helped both of them but still resulting patches from stillunknown didn't work for AndrewR. Will our mode setting experts finally claim their part of nouveau fame? Please don't skip our next TiNDC issue when we will reveal all (or at least more :) ).