Bug 65665

That's more likely to be the read-read bug that I re-enabled in order to check for its existence. A couple of patches have landed since your snaphot for 2 issues found that looks like your bug.

In fact, I am not experiencing those corruptions anymore, after the latest pull.

Thanks!

Thanks! Hopefully we can stop meeting like this. :)

Unfortunately, one more meeting.

Unfortunately, I still see (though probably less often) exactly the same kind of behavior as in the previous screenshot attached on 13-06-12. Now my config is:

Kernel 3.9.6

xserver xorg video intel driver 2.21.9+ 
     (git snapshot taken 13-06-13 - 1f180b...)
libdrm 2.4.45+
     (git snapshot taken 13-06-07 - a0178c...)
mesa 9.2+
     (git snapshot taken 13-06-13 - 6057d7...)

Looks like it is exactly the same issue as 55500, as the issue with the 'sample libreoffice document that almost always shows issues on gen4' is also back.

Can probably be closed as a duplicate.

*** Bug 65948 has been marked as a duplicate of this bug. ***

Created attachment 81096 [details]
Another image showing the issue

Just one note about the fact that many sites that deliver pdf documents are moving not to have the pdfs downloadable, but to rely on 'in web-browser' rendering.

This bug seems to hit quite bad on them, since whole pages disappear, as in the attachment.

Fortunately, the pdf.js renderer incorporated in firefox does not seem to be this troublesome.

Also libreoffice impress (presentation a la powerpoint) custom animations are completely broken now.

When you animate an object, other objects that are on screen and that should not be touched by the animation simply disappear.

This makes giving presentations from the laptop unfeasible unless all custom animations are turned off.  

However, for some reason asking libreoffice to render animations 'with hardware acceleration on' (which was a completely broken on my machine till no long ago) now sort of works. I do not know what 'with hardware acceleration on' means in libreoffice wording, though. Maybe it uses an opengl rendering path rather than the standard 2D one.

Can you prepare an .odg with the animation setup or give me a quick guide on how to recreate such a beast?

Created attachment 81207 [details]
Presentation file showing animation issues

This is fine with 'hardware acceleration on' and troublesome with 'hardware acceleration off'.

To test, load and start the presentation with F5.

On my system, with hardware acceleration off, a blue block appears on the left top corner while playing throught the animations.

To turn acceleration on and off, use tools->options. Note that not all the libreoffice builds let you activate and disactivate the acceleration. But certainly the 4.0.3 build from www.libreoffice.org does.

To make things even more interesting: if you select the text area on the left, remove the background color, run the presentation again, on my system the presentation glitch changes into a new one. No more blue box on the top left corner, but the text appearing from the first animation click now disappears during the second animation click.

I haven't isolated the bug yet, but the hw acceleration option here seems to just vary the amount of X rendering calls made. So it would seem to switch between a client-side rasteriser and server-side, but not enabling an OpenGL path.

I can reproduce the offset blue square with hw-off, so I suspect this is a separate bug.

Ugg. My hack strikes back:

commit 5189ba727fe48990f00d5787c53d7b15fa1838a4
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Sat Jun 22 13:39:47 2013 +0100

    sna: Refine the self-cow copy hack to reduce collateral damage
    
    Sergio Callegari found a case in LibreOffice that copied cloned pixmaps
    onto each other and then modified one of the pair. This was triggering
    the hack put in to speed up firefox scrolling, unfortunately.
    
    Reported-by: Sergio Callegari <sergio.callegari@gmail.com>
    References: https://bugs.freedesktop.org/show_bug.cgi?id=65665
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Sergio, I think I have all the regressions fixed - is there anything else that looks unusual versus bug 55000?

It's bug 55500, not 55000 ;)

I'm not entirely sure what all is covered under this bug vs 55500, but for me using 2e2c448a77ab9dce4807b159708290cd7ad26b5c, page contents in Firefox seem better, but the Firebug icon I have on my toolbar still gets corrupted.  I also see flickering on Chrome tabs as described in 55500.

I think I spoke too soon.  Images on websites rendered in Firefox eventually started appearing corrupted ... although it seemed to take much longer for that to happen than it did before.

Hi, could test only today.

Unfortunately libreoffice file in 55500 (test.odg, i.e. https://bugs.freedesktop.org/attachment.cgi?id=74207) still shows the usual issues. The drawing is rendered incorrectly when zooming in and out.

Furthermore all large bitmaps first or later are rendered incorrectly in libreoffice draw or impress.

Is the pixmap corruption like the examples attached? If so they may be

commit 22fd5ca947b58901927d100d2b1aa0f1672b3435
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Jun 28 16:54:08 2013 +0100

    drm/i915: Only clear write-domains after a successful wait-seqno

which affects v3.7+.

This means that it should be fixed in latest kernels?

Unfortunately for people on ubuntu, this is not completely good news, since newer kernels are only built from the development version of ubuntu (saucy) and since 3.9.8 do not seem to work anymore on the current distro (raring). So it is impossible to test without backporting on an older kernel or anyway building a personal kernel for raring...

We've flagged the fix for the stable channels, it should arrive within an updated raring, quantal, and precise kernel in the next month. *Should*.

I am optimistic I have the coherency bugs resolved now which unfortunately just leaves me with bug55500. Thanks for the help and testing.

Thanks, and thanks also for the effort on 55500. I'm keeping fingers crossed it can be fixed soon and that the drm fix 'Only clear write-domains after a successful wait-seqno' can land in ubuntu raring 3.8 kernel.

For 55500, the attached Libreoffice document is an almost infallible trigger for it (unless its artifacts depend just on missing 'Only clear write-domains after a successful wait-seqno')

Ubuntu has an experimental 64bit raring kernel (3.8) with the fix for the coherency bug. Works great for me.

Can be tested at http://kernel.ubuntu.com/~jsalisbury/lp1200126/