Intel's Alan Cox has pushed forward an Intel GMA 500 DRM driver for mainline kernel inclusion that already has been pushed into the staging-next tree. It's been a few days since the actual patch/driver was proposed, but for some reason it slipped under my radar until now when it was mentioned in the forums.
Before getting too excited, however, there are two unfortunate downsides as this open-source staging DRM driver stands right now:
- There's no acceleration support. If you're hoping to use your Poulsbo hardware with VA-API video or OpenGL/3D acceleration, you're still stuck playing with the proprietary bits. This staging DRM driver doesn't support the necessary interfaces for hardware acceleration.
- While the GMA 500 is derived from the PowerVR SGX core, as do many netbook/embedded graphics processors, especially ARM-based mobile devices, this driver is only specific to the GMA 500. The driver in its current form is mostly about mode-setting and bringing up the display, and not about acceleration which is the area common to the PowerVR cores, so it's mostly Intel-specific code within this driver. In fact, much of this driver's mode-setting code is similar to that of Intel's i915 open-source driver. Much of the code could be shared between this new staging driver and Intel's popular open-source driver, but based upon comments by Alan Cox, this has been rejected by Keith Packard's team.
So what we basically have with this new staging driver is a basic mode-setting driver without acceleration. It's a step forward comparing to using an Intel Poulsbo system with the VESA driver, but still it's open-source hardware acceleration that is really needed to please end-users. Alan Cox will also be pushing out new user-space bits based upon xf86-video-psb to support this kernel driver code.
The discussion surrounding this work can be found in the GMA500 staging driver thread. "This is an initial staging driver for the GMA500. It's been stripped out of the PVR drivers and crunched together from various bits of code and different kernels."
This code living in the staging driver is in need of some more TTM love and other work before it will leave the staging area of the kernel. David Airlie also mentions hooking it up to a dumb ioctl for trivial front-buffer allocation so that a dumb KMS driver and KMS library can work atop this code, to basically fake kernel mode-setting.
This is a step forward, but there's still much work left before Poulsbo could possibly lose its reputation of being a bloody mess under Linux. At the same time, the Free Software Foundation has deemed creating a reverse-engineered PowerVR driver with a Gallium3D driver as being a high priority, which could potentially provide the acceleration needs for this new GMA 500 driver. Imagination might also provide an open-source PowerVR driver in Q3'2011.