Intel's GEM-Free UMS Driver Still Doesn't Fix Things
Once Intel had pushed the Graphics Execution Manager into the mainline Linux kernel followed by the kernel mode-setting (KMS) infrastructure, they were very quick to kill off their EXA 2D acceleration and user-space mode-setting driver support in favor of their GEM-using UXA acceleration and kernel-space mode-setting, respectively. Intel's open-source driver stack originally had some nasty regressions following this invasive work, but Intel's newest stack is fairly stable and on-par with their older code -- unless you happen to be using an older i8xx series chipset where the support has fallen behind and is really no longer a focus of Intel or that of most distribution vendors.
Chris Wilson worked on adding back a GEM-free UMS code-path for those older Intel customers (though it should also work with newer Intel chipsets too) in hopes that this would take care of the stability and freezing problems that has become a common occurrence to those using the newer driver stack such as what's found in recent distributions like Ubuntu 10.04 LTS. This work hasn't yet been pushed into the mainline xf86-video-intel DDX driver, but it has been packaged for Ubuntu in a PPA and Ubuntu developers have encouraged those with Intel hardware to try this driver. A few weeks have passed since this call-for-testing and the results seem to be mixed.
With the UMS driver, one user reported X freezing with a blank screen on boot using an i855GM chipset. With the mainline driver and its kernel mode-setting paths, the X Server boots and works for this user until it comes time to use X-Video, at which point it freezes. Another user also with an i855GM is able to start the X.Org server properly, but when attempting video playback the system will crash and result in a black screen.
Some users, such as one with a Brookdale 845G chipset has been using this UMS driver for multiple days without issues. Another i845G owner running Ubuntu 10.04 LTS has suffered an unresponsive black screen when starting the X.Org Server with this yet-to-be-mainlined driver. Someone that had tested out this UMS driver with a newer Intel 945GM IGP reported no apparent regressions. The last person to report i8xx problems was this evening and for her the driver from Chris Wilson had resulted in the system locking up in less than 20 minutes.
While this driver work is doing good for some i8xx users, it's not helping out everyone or even most users, which puts this reintroduced UMS support into question and perhaps will not even make it into Intel's mainline X.Org driver. Next month at the X Developers' Summit in Toulouse they may just decide to kill it with fire and recommend users of this vintage Intel hardware to stick to a more dated driver / distribution.