Bug 79448

Summary:	Texture corruption when one context gets used on multiple windows
Product:	Mesa	Reporter:	Martin Flöser <mgraesslin>
Component:	Drivers/DRI/i965	Assignee:	Kristian Høgsberg <krh>
Status:	RESOLVED DUPLICATE	QA Contact:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Severity:	normal
Priority:	medium	CC:	kde
Version:	10.1
Hardware:	Other
OS:	All
Whiteboard:
i915 platform:		i915 features:
Attachments:	Apitrace showing the problem corrupted texture

Description Martin Flöser 2014-05-30 09:44:36 UTC

Created attachment 100153 [details]
Apitrace showing the problem

Downstream bug with a qml test: https://bugreports.qt-project.org/browse/QTBUG-39182

This happens if there are:
* one context
* at least two windows
* a texture used in each window
* texture loaded for each window in same frame

Attached is an apitrace run of the example and a corrupted texture.

If you need any more information from me, please let me know. This highly affects the upcoming Plasma release and we will try as best as possible to help fix this bug.

Comment 1 Martin Flöser 2014-05-30 09:45:32 UTC

Created attachment 100154 [details]
corrupted texture

Comment 2 David Edmundson 2014-06-12 11:57:06 UTC

This has now been 12 days without comment.

This bug is rather major as it makes a lot of displays unreadable and affects a lot of users.

Can you let us know if there is anything else we can provide to help?

Comment 3 Tapani Pälli 2014-06-18 12:11:29 UTC

I've bisected this, corruption (and some qapitrace crashes with this trace) start to happen afer following commit.

--- 8< ---
11baad35088dfd4bdabc1710df650dbfb413e7a3 is the first bad commit
commit 11baad35088dfd4bdabc1710df650dbfb413e7a3
Author: Kristian Høgsberg <krh@bitplanet.net>
Date:   Tue Jan 21 12:17:03 2014 -0800

    intel: Fix initial MakeCurrent for single-buffer drawables
    
    Commit 05da4a7a5e7d5bd988cb31f94ed8e1f053d9ee39 attempts to eliminate the
    call to intel_update_renderbuffer() in the case where we already have a
    drawbuffer for the drawable.  Unfortunately this only checks the
    back left renderbuffer, which breaks in case of single buffer drawables.
    
    This means that the initial viewport will not be set in that case.  Instead,
    we now check whether the initial viewport has not been set, in which case
    we call out to intel_update_renderbuffer().
    
    https://bugs.freedesktop.org/show_bug.cgi?id=73862
    
    Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>

Comment 4 Tapani Pälli 2014-06-18 12:15:14 UTC

FYI there's bugs #74005 and #74083 that are related to this issue.

Comment 5 Tapani Pälli 2014-06-18 13:04:29 UTC

IMO the reason why this fails is that first makecurrent calls sets ctx.ViewportInitialized to true. Further makecurrent calls with different drawables do not trigger intel_prepare_render() call to get buffers.

Comment 6 Kristian Høgsberg 2014-06-18 17:19:59 UTC

Duplicate of 74005.  Fix for that pushed to master and scheduled for upcoming 10.1 and 10.2 releases.

*** This bug has been marked as a duplicate of bug 74005 ***

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.