[Bug 76211] New: BISECTED REGRESSION: i915 resume latency 1200ms
bugzilla-daemon at bugzilla.kernel.org
bugzilla-daemon at bugzilla.kernel.org
Wed May 14 11:09:40 PDT 2014
https://bugzilla.kernel.org/show_bug.cgi?id=76211
Bug ID: 76211
Summary: BISECTED REGRESSION: i915 resume latency 1200ms
Product: Drivers
Version: 2.5
Kernel Version: 3.15-rc1
Hardware: All
OS: Linux
Tree: Mainline
Status: NEW
Severity: normal
Priority: P1
Component: Video(DRI - Intel)
Assignee: intel-gfx-bugs at lists.freedesktop.org
Reporter: lenb at kernel.org
CC: intel-gfx-bugs at lists.freedesktop.org
Regression: No
Created attachment 136101
--> https://bugzilla.kernel.org/attachment.cgi?id=136101&action=edit
0001-drm-i915-don-t-wait-for-power-cycle-when-waiting-for.patch
On a Dell XPS11 convertable (Intel HSW-YLT),
i915 resume from S3 got slower in 3.15-rc1
Linux 3.14: 400ms
Linux 3.15-rc1: 1200ms
If the attached patch is reverted, 400ms resume time is restored.
Unfortunately, the attached patch also had the dramatic
_benefit_ of reducing i915 suspend latency on the same XPS11
to under 150ms from over 1300ms.
The numbers are different when measured on a Lenovo Yoga2 Pro HSW-ULT,
but the trend is the same. The patch increases resume time
to about 700ms from under 400ms, and the patch reduces
suspend time to under 300ms from over 1100ms.
Obviously, it would be ideal to capture the suspend speed
benefit of this patch without having the resume cost. However,
if we have to pick between suspend and resume performance,
resume performance is more important.
Note that it is easy to graphically measure resume latency with this tool:
https://01.org/suspendresume
if analyze_suspend.py is invoked with the -f argument, it also
produces an ftrace that can show exactly where the time goes.
In the case of 3.15-rc1 on the Dell XPS11:
intel_edp_panel_on takes > 1000ms
wait_panel_power_cycle takes 800ms
wait_panel_status takes 200ms
--
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.
More information about the intel-gfx-bugs
mailing list