Bug 102201 - fbc causes redraw delays with Qt5 applications.
Summary: fbc causes redraw delays with Qt5 applications.
Status: RESOLVED INVALID
Alias: None
Product: Drivers
Classification: Unclassified
Component: Video(DRI - Intel) (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: intel-gfx-bugs@lists.freedesktop.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2015-08-02 03:03 UTC by Francis Herne
Modified: 2015-10-07 14:54 UTC (History)
1 user (show)

See Also:
Kernel Version: 4.1.3
Subsystem:
Regression: No
Bisected commit-id:


Attachments

Description Francis Herne 2015-08-02 03:03:12 UTC
First off - I'm aware that this is an experimental feature that shouldn't be expected to work.

When 'i915.enable_fbc=1' is specified on the kernel cmdline, and a non-compositing window manager is used, Qt5-based applications experience frequent periods (more than half the time) of 5-10 seconds when their windows are not redrawn. Other (non-Qt5-based) apps running at the same time are not affected.

When using non-compositing WMs (either dwm 6.0 or openbox 3.6.1), this occurs for every instance of every Qt5-based application that I've tried, across several reboots, and never for applications using other GUI toolkits including GTK2, FLTK and Qt4.
When using a compositing WM, this never occurs even when fbc is enabled. This includes KWin with either backend or Compton+openbox - in the latter case, the delays immediately cease when compton is started.

This never occurs when framebuffer compression has not been enabled, regardless of toolkit or WM.

I can't see any related output in the system journal, dmesg, stderr of the applications or Xorg log files.

I'm using the standard ArchLinux kernel build 4.1.3-1-ARCH, with Qt 5.5.0, mesa 10.6.3 and Xorg 1.17.2. The hardware is a Thinkpad X201, with an i5-580M (Ironlake) and its integrated graphics and a 1280x800 LVDS display.
Comment 1 Jani Nikula 2015-10-07 11:18:09 UTC
(In reply to Francis Herne from comment #0)
> First off - I'm aware that this is an experimental feature that shouldn't be
> expected to work.

That's right, and we don't support it. Sorry. We'll enable it when we have confidence it works. Closing.

Please file any new bug reports at the freedesktop.org bugzilla [1]. Thank you.

[1] https://bugs.freedesktop.org/enter_bug.cgi?product=DRI&component=DRM/Intel
Comment 2 Francis Herne 2015-10-07 14:54:39 UTC
Well, I was thinking that reporting cases where it doesn't work yet might help when you do get around to fixing it up.

Will report any future issues to the correct location, thanks.

Note You need to log in before you can comment on or make changes to this bug.