Bug 100540

Summary: [hsw] i915 resume latency 909 ms
Product: DRI Reporter: Jani Nikula <jani.nikula>
Component: DRM/IntelAssignee: Len Brown <lenb>
Status: CLOSED INVALID QA Contact: Intel GFX Bugs mailing list <intel-gfx-bugs>
Severity: normal    
Priority: medium CC: david.weinehall, intel-gfx-bugs, lenb
Version: XOrg git   
Hardware: Other   
OS: All   
URL: https://bugzilla.kernel.org/show_bug.cgi?id=76211
Whiteboard:
i915 platform: HSW i915 features: power/suspend-resume
Attachments:
Description Flags
analyze suspend output from bugzilla.kernel.org none

Description Jani Nikula 2017-04-03 15:26:01 UTC 
Moving from https://bugzilla.kernel.org/show_bug.cgi?id=76211 reported by Len Brown.

The last observation there was:

This problem still exists in Linux-4.8.0, though the delay
on this machine is now 909 ms, rather than 1,200 ms.

Attached, please find the analzye_suspend.py output as run
on the same laptop as above, installed with Ubuntu 16.10
and its 4.8.0-22-generic kernel.

Curiously, i915 resume invokes intel_hdmi_detect FOUR TIMES --
twice in a kthread, and again twice in-line.  This costs 160ms
each time it is invoked.
Comment 1 Jani Nikula 2017-04-03 15:28:12 UTC 
Created attachment 130655 [details]
analyze suspend output from bugzilla.kernel.org

https://bugzilla.kernel.org/attachment.cgi?id=241901
Comment 2 David Weinehall 2017-04-03 16:25:12 UTC 
@Len Brown: Could you provide the output of:

cat /sys/kernel/debug/dri/0/eDP-1/i915_panel_timings

and

analyze_suspend.py -config config/suspend-callgraph.cfg -maxdepth 0 -filter i915

on a recent kernel (preferably drm-tip)?

The callgraph is needed to be able to pinpoint the functions that slow us down, and running on a recent kernel will ensure that the issue hasn't already been fixed.

The multiple invocations of intel_hdmi is most likely triggered by hotplug detection of external displays. That logic is, to say the least, fairly hairy. There have been some fixes to that area since 4.8 though, as well as various other fixes to lower suspend/resume times.
Comment 3 Elizabeth 2017-06-22 18:28:48 UTC 
Hello, 
Is there any advance in this case? If there is any new information please share. Thank you.
Comment 4 Ricardo 2017-08-15 16:48:51 UTC 
can you respond to David's request
Comment 5 Elizabeth 2017-10-16 23:20:04 UTC 
(In reply to David Weinehall from comment #2)
> @Len Brown: Could you provide the output of:
> cat /sys/kernel/debug/dri/0/eDP-1/i915_panel_timings...
> analyze_suspend.py -config config/suspend-callgraph.cfg -maxdepth 0 -filter
> i915...
There haven't been update on this case from the reporter since 2016-10-18 (https://bugzilla.kernel.org/show_bug.cgi?id=76211#c27), is this case going to be keep open or should be closed as invalid?? Thank you.
Comment 6 David Weinehall 2017-10-17 08:06:44 UTC 
I think a year should've been quite enough for the bug reporter to provide more information, especially since the bug report concerned 4.8, while we're currently working on 4.14.

I don't know whether INVALID or WORKSFORME is the relevant status here, but I'm picking INVALID.
Comment 7 Elizabeth 2017-10-26 17:09:57 UTC 
(In reply to David Weinehall from comment #6)
> I think a year should've been quite enough for the bug reporter to provide
> more information, especially since the bug report concerned 4.8, while we're
> currently working on 4.14.
> 
> I don't know whether INVALID or WORKSFORME is the relevant status here, but
> I'm picking INVALID.
Thanks David. Then closing. @Len: If issue exist with latest kernels, please file a new bug.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.