i915: occasional hangs on Baytrail with deeper C-states

[Daniel Bilik]

Hi.

(This is not really a bugreport, just a notice to inform potential
victims. :) AFAICT it's not DragonFly specific issue, see below.)

I'm experiencing occasional freezes running DragonFly desktop on a
Baytrail system (namely Intel Celeron J1900) with machdep.mwait.CX.idle
set to AUTODEEP (which enables C6 and C7 on this system). Limiting it to
C1/1 makes the system stable.

Interestingly, the problem is evolving somewhat... The system was stable
before commit 2c9916c+ (drm/i915: Upgrade to Linux 4.0). After this update
I started to experience the hangs, approximately once a day. I didn't know
about the "C1 workaround" at that time, so I just locally reverted that
commit to get the stable system. Commit 477eb7f+ (drm/i915: Update to
Linux 4.1) somehow fixed it, and system was stable again, even with
machdep.mwait.CX.idle=AUTODEEP. But commit 19c468b+ (drm/i915: Update to
Linux 4.2) brought the hangs back, and this time I've "resolved" it via
machdep.mwait.CX.idle=C1/1.

According to these bugreports from linux users...

https://bugs.freedesktop.org/show_bug.cgi?id=88012
https://bugzilla.kernel.org/show_bug.cgi?id=109051

... it doesn't seem to be DragonFly specific issue but more probably a bug
in shared i915 code. Those discussions also confirm that bug was
introduced somewhere after linux 3.16, then disappeared for a while during
4.1 and was brought back with 4.2 (and remains to 4.5). So far, nobody has
come with a solution, but everyone confirm that "C1 workaround" prevents
from hangs.

HTH.

--
						Dan