Intel Still Sorting Out Haswell On Linux

Intel should finally have Haswell graphics support on Linux in shape with the Linux 3.8 kernel after earlier admitting they screwed up, not all areas of the open-source GPU driver support are polished ahead of the big Haswell CPU launch in 2013.

Based upon reports from my sources and public code commits, it seems that the Intel Haswell Linux enablement is mostly set. With the Linux 3.8 kernel and Mesa 9.1/10.0 along with the latest xf86-video-intel DDX, it seems mostly everything is set for the Intel Haswell launch next year -- the display output now is mostly in shape, 2D/3D acceleration is working reliably, and as of recently there is video acceleration support.

While the core areas of support are now in place for the much-improved graphics on Haswell processors, in other areas they are still cleaning and fixing up things. Of interest to many will be the power efficiency of Haswell as it's been talked up a lot by Intel, but this is one of the areas still being tackled under Linux.

New on the mailing list are patches to work out dynamic power well support on Haswell.

Daniel recently sent a patch to support disabling the power well, but when reviewing it I discovered that our code to enable/disable the power well - not part of Daniel's patch, only used by it - is broken. So I created a patch to fix this problem and then ported Daniel's patch to be on top of the fix.

With this, we are finally capable of correctly disabling/enabling the power well, but we are also uncovering a huge problem: we now get tons of "unclaimed register" errors, because we read/write the registers inside the "power down well" even when they're not being used.

I think we can try to apply patch 01 now, but before applying patch 02 we should probably try to get rid of all those "unclaimed register" messages that it brings. This will certainly not be a small task. I can volunteer myself to do this, but considering my TODO list I will certainly not be able to address this problem before January. The first step is certainly to avoid calling ivb_disable_plane: this alone will already remove most of the error messages.

Aside from sorting out the dynamic powering up and down of the power well, various other HSW-related patches have been floating around on the mailing list in recent days with the Linux 3.8 merge window opening up in the next week or so.

This will all hopefully be worked out before the Haswell launch in H1'2013, but for some Linux desktop users it may mean needing to pull in newer code than what may be shipping in their distribution of choice at the time of the Ivy Bridge successor's launch.