i915: disable GEN6_MBCTL write in gen6_init_clock_gating

[emaste]

Summary

This came from Linux commit b4ae3f22d238617ca11610b29fde16cf8c0bc6e0 which is implicated in Sandy Bridge power consumption issues (albeit under different conditions).

On my Thinkpad X220 at rS296910 I see the following power consumption numbers (on a memstick image with a few services running and no packages installed).

• After booting and logging in as root: 11W
• After kldload i915kms: 21W
• After suspend + resume: 11W

With this patch I see 11W in all three cases.

[emaste]

John, can you give this a try on your X220? You reported no issue w/o the patch but I want to make sure this doesn't introduce some other regression.

sys/dev/drm2/i915/intel_pm.c

3702–3704 ↗

(On Diff #14399)

will delete these mixed up comment lines

[emaste]

Diff is also available at https://people.freebsd.org/~emaste/D5665.diff

[dumbbell]

That's great news, thank you!

I added to minor comments inline.

sys/dev/drm2/i915/intel_pm.c

3676 ↗

(On Diff #14399)

Instead of #if 0, I use #ifdef FREEBSD_WIP to comment out code specifically on FreeBSD. For example:
https://github.com/freebsd/freebsd-base-graphics/blob/master/sys/dev/drm2/i915/i915_gem.c#L1380-L1388

3677 ↗

(On Diff #14399)

Could you please prefix the first line of your comment with NOTE Linux<->FreeBSD: ?

I use that to really identify code which differs from Linux on purpose. For instance:
https://github.com/freebsd/freebsd-base-graphics/blob/master/sys/dev/drm2/i915/i915_gem.c#L1101-L1103

[emaste]

Use #ifdef FREEBSD_WIP and NOTE Linux<->FreeBSD: as suggested by @dumbbell.

[emaste]

Thanks for the review. I'll commit in a day or two if there are no reports of regressions (and hopefully confirmation that it fixes the issue for others).