In a late sprint, developers have added support for Fermi graphics chips to the nouveau KMS driver, following the acquisition of a Fermi card by one of the main developers (see e. g. 1, 2, 3, 4, 5, 6, 7). Fermi GPUs are used on the GeForce GTX 470 and 480, released in March, as well as some more recent 400 series models. The kernel developers also improved suspend and resume support for GeForce cards up to and including the 7000 series as well as DisplayPort and hotplug support for GeForce 8000 to 300 series cards.
The KMS driver for Radeon cards now allows underscan to be enabled to allow better adjustment of image size on digitally controlled monitors which automatically activate overscan. The driver now also offers functions for reading the temperature sensor integrated into newer Radeon GPUs, has improved support for laptop function keys for activating monitor connections and supports audio output via HDMI on RS600, RS690 and RS740 motherboard chipsets.
On R300 and R500 graphics chips, the driver now provides everything needed to use HyperZ acceleration functionality for dealing with depth information. Tiling support for R600, R700 and R800 series (2000 series and later) graphics chips should improve the driver's 2D and 3D performance (e. g. 1, 2, 3).
Optimised locking and new debugging options
In his main git pull request, DRM subsystem maintainer Dave Airlie mentions a number of further changes. For example, from 2.6.36 onwards, the big kernel lock (BKL) will only be used by driver code for some obsolete Intel 800 family graphics chips. This should improve code scalability and performance.
KDB developer Jason Wessel has also added support for the KGDB debuggers and KDB (kernel debugger, first integrated into the kernel in 2.6.35) to the kernel and the KMS Intel driver. Together with changes to the KGDB/KDB code, this allows the Intel driver to switch to a debugger shell for analysing X Server crashes, etc., where it is no longer possible to switch to a text console and there is no serial console configured. The technology is demonstrated in an old YouTube video.