Oh dang, sounds like we need to resurrect forcewake hysteresis.

Chris, do you still have the patches handy for the delayed put from mmio writes? Mika, could you perhaps refresh those?

Note: Since the is a w/a we can't just revert the commit for obvious reasons ...

http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?id=b6890154a3d8c43216d04ca31a7e41f1ce163afb

http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?id=2149aeaa4f7803ff35314e5f062a0b47bef76653

http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?id=59b393880dd235015100caa6893666a89b3f80c6

Hello Chris,
Would you pls gudie me which branch kernel commit should I based on to apply you pasted three patches.

I tried to apply your patches based on drm-intel-next-queued branch(this bug's first bad commit), or drm-intel-next-queued branch latest commit 
(commit 89a35ecdc6aa5a88165313ca5cfd52b8e8e7fbbd
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Thu Nov 20 13:24:13 2014 -0800

    drm/i915/g4x: fix g4x infoframe readout

    Need to check the port too.

    Reported-by: Daniel Vetter <daniel@ffwll.ch>
    Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
    Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
) 

or drm-intel-nighly branch latest commit(commit 18748be7c96accc27327423c384f86a8fae99c35
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Nov 20 22:59:05 2014 +0100

    drm-intel-nightly: 2014y-11m-20d-21h-58m-44s UTC integration manifest

)

apply patches all fail because of conflict, especially for 2nd and 3rd patches.

Thanks.

Wendy,

http://cgit.freedesktop.org/~miku/drm-intel/log/?h=uncore_cleanup

Work in progress but in case you want to try out.

(In reply to Mika Kuoppala from comment #5)
> Wendy,
> 
> http://cgit.freedesktop.org/~miku/drm-intel/log/?h=uncore_cleanup
> 
> Work in progress but in case you want to try out.

Hello Mika, 
Shall we checkout "bdw_forcewake" branch?

(In reply to wendy.wang from comment #6)
> (In reply to Mika Kuoppala from comment #5)
> > Wendy,
> > 
> > http://cgit.freedesktop.org/~miku/drm-intel/log/?h=uncore_cleanup
> > 
> > Work in progress but in case you want to try out.
> 
> Hello Mika, 
> Shall we checkout "bdw_forcewake" branch?

uncore_cleanup

(In reply to Mika Kuoppala from comment #7)
> (In reply to wendy.wang from comment #6)
> > (In reply to Mika Kuoppala from comment #5)
> > > Wendy,
> > > 
> > > http://cgit.freedesktop.org/~miku/drm-intel/log/?h=uncore_cleanup
> > > 
> > > Work in progress but in case you want to try out.
> > 
> > Hello Mika, 
> > Shall we checkout "bdw_forcewake" branch?
> 
> uncore_cleanup

Hello Mika, your uncore_cleanup branch commit(59d3f39e57b3e40f7fd249598589910c39270d94)  can fix this regression bug: both suspend and suspend resume responsiveness can back to the values before regression.

verfied on drm-intel-nightly
	
commit 63df8858af984ee11e32579a5d3cf2db17baf942
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Dec 2 16:43:51 2014 +0100

    drm-intel-nightly: 2014y-12m-02d-15h-43m-26s UTC integration manifest


bug has fixed ,close it

No patches from the uncore_cleanup branch was posted on the mailing list when this got fixed.

So we have absolutely no idea what was the commit that fixed this. Would be very important to figure out why the symptoms disappeared.

So could we get a reverse bisect pointing to a commit that fixed the symptomps. Thanks.

(In reply to Mika Kuoppala from comment #10)
> No patches from the uncore_cleanup branch was posted on the mailing list
> when this got fixed.
> 
> So we have absolutely no idea what was the commit that fixed this. Would be
> very important to figure out why the symptoms disappeared.
> 
> So could we get a reverse bisect pointing to a commit that fixed the
> symptomps. Thanks.

Hello Mika,

We've try to bisect which commit fixed this bug on 2014-12-05, but when we try to restore 2014-12-05 test environment onto the same machine, suspend entry and suspend resume responsive time regression bug still exist, cannot find the good commit any more, though our Tinderbox responsiveness history trend data did show that day has normal values, but we cannot reproduce today after checked all the testing conditions, so re-open this bug.

Verified BDW with latest kernel on 2014-12-25 drm-intel-nightly branch,
this responsiveness bug has been fixed, all the data recovered to the values before regression.

2014-12-25 GFX SW stack
 Libdrm:		(master)libdrm-2.4.58-19-gf99522e678dbbaffeca9462a8edcbe900574dc12
 Mesa:		(master)f332cf92b69e52de3cb7c3088ad1efd2e291bb88
 Xserver:		(master)xorg-server-1.16.99.901-102-g826e7c2b36f192fbbe7ddff37eb559f4d6301146
 Xf86_video_intel:		(master)2.99.917
 Cairo:		(master)8e11a42e3e9b679dce97ac45cd8b47322536a253
 Libva:		(master)8986ec692b19d8dd6bd2aa118b5dffbd05a8f909
 Libva_intel_driver:		(master)b5d6d9d425a6d539b27d22992bda05f79d1a0622
 Kernel:   (drm-intel-nightly)4fa23142a15526f4a4b5df61f26eacdd558a849a

Closing old verified.