i915 update to Linux 3.8 ready!

[am_dxer at fastmail.fm]

On 9 March 2016 at 22:02, Ed Maste <emaste at freebsd.org> wrote:
>>
>> I found reports that Linux introduced a regression on Sandy Bridge
>> power consumption sometime around Linux 3.6 that was fixed after 3.8.
>> Those reports are specifically for an increase after a suspend-resume
>> cycle; something gets confused and causes the GPU to never enter rc6.
>> It's not exactly the same symptom, as I see increased consumption
>> after reboot and loading i915 (before starting X), but certainly seems
>> related.
>>
>> Anyhow, with that patch applied after a suspend-resume cycle I'm back
>> to a reasonable power consumption -- around 15W in X with Firefox
>> running. The patch is in https://reviews.freebsd.org/D5599.

> Some further testing suggests the power consumption's back to normal
> after a suspend+resume cycle, without the patch in D5599. I may have
> mixed up something in my experiment yesterday or perhaps there's some
> intermittent behaviour.

> If anyone else has Sandy Bridge laptops (Thinkpad X220, T420 etc.) I'd
> really appreciate hearing about test results.

On an HP EliteBook 2560p with the Sandy Bridge Core i7-2620M processor,
I have about a 5 watt power increase when loading the KMS driver from
head as opposed to when using vesa. I have enabled the rc6 states in
loader.conf and dmesg output shows them as being enabled. I am testing
from within the Mate Desktop environment. Doing any tests on consoles as
you have done is a bit challenging for me because I am totally blind and
use a screen reader called Orca that only works within a graphical
environment.