Issue 628493  has been merged into this issue.

Adding RB Label as this issue happens on NTP and other URLs like youtube,google, etc.,
Please remove blocker label if not necessary.
Thank You.

Good build:54.0.2793.0
Bad build:54.0.2794.0

CHANGELOG URL:
 https://chromium.googlesource.com/chromium/src/+log/a408d3aea3e7f270266fbf5e483a5b2e708ae553..ccfd78962de023613e2e281b392fdc9982718136

Suspecting https://chromium.googlesource.com/chromium/src/+/35ec773b64c07f58b0e59377d917f6424edda189 from changelog.

@cwallez: Please help in re-assigning if it is not related to your change.

I will take a look. Can you provide more information about the GPU / driver combination this reproduces on?

I can repro on a Linux Intel Mesa 11.2 and can confirm that the bug appeared with this CL. I was not able to reproduce on a Linux NVIDIA machine with, or without a core GL profile.

A faster repro is to start Chrome, remove the eventual yellow warning bar, and press Fn + F11 twice. What seems to happen is that the tab bar is drawn too low, causing garbage to appear where it should have been. Then the page content is drawn, occluding parts of the misplaced tab bar.

It seems like it might be a bug in the browser compositor that was unveiled with the move to core profile.

CCing piman@ and enne@: I'm not sure how to get useful information about the browser compositor operation and the X11 events causing it to invalidate tiles. Can you provide pointers?

Window::SchedulePaintInRect calls are the same between the core and compatibility profiles so the problem doesn't seem related to missing invalidation.

cwallez@, piman@, enne@: Can we get an update on this Dev Blocker issue.

Is it that it's drawn in the wrong place, or is it that when the window is resized, it just picks up garbage (which often is the previous contents in the window with some random transformations) until we draw a new frame?

Is this just flashing garbage, or does it stay visible until it gets patched by redrawing due to UI changes?

It stays visible until it is redrawn due to UI changes. What happens in that the tab bar is drawn offset ~20 pixels down, then the page is drawn at the correct place. This means that the ~20 top pixels of the window show garbage, and that the ~20 pixels at the bottom of the tab bar are overdrawn by the page content.

Are partial swaps getting enabled for some reason? They should be disabled on mesa+GLX. The behavior you describe suggests it's on...

Or the other possibility is some bad interaction between viewport and window resize in the driver, we were having those when virtual contexts were enabled.

Thanks Antoine for the pointer! Partial swap was indeed enabled because "Mesa" isn't in the "Intel Open Source Technology Center" GL_VENDOR string. Weirdly this is the case in the original report in #339493, so that driver bug entry might have been incorrect since it was modified to target only Mesa (vs. all linuxes).

Put up https://codereview.chromium.org/2156993004 that fixes it.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/6bcbbd2886e61691f69ec14c27e62b69f81a271b

commit 6bcbbd2886e61691f69ec14c27e62b69f81a271b
Author: cwallez <cwallez@chromium.org>
Date: Tue Jul 19 02:45:20 2016

gpu driver bug list: correctly backlist partial swap on Mesa

This entry previously used "gl_vendor": "Mesa" when the GL_VENDOR string
is "Intel Open Source Technology Center". Try to match "Mesa" in the
GL_RENDERER string instead.

BUG= 339493 
BUG= 628168 
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2156993004
Cr-Commit-Position: refs/heads/master@{#406194}

[modify] https://crrev.com/6bcbbd2886e61691f69ec14c27e62b69f81a271b/gpu/config/gpu_driver_bug_list_json.cc

 Issue 629032  has been merged into this issue.

Hello I am still having the issue on Linux with Gallium drivers.

It is simply caused by the GL_RENDERER being quite different there:
GL_RENDERER	Gallium 0.4 on AMD CAICOS (DRM 2.45.0 / 4.7.4-2-default, LLVM 3.8.1)

While the driver vendor indeed is Mesa.

Maybe the check should be Mesa || Gallium?

Downstream bugreport:
https://bugzilla.suse.com/show_bug.cgi?id=1000019

Simple fix, running on commandline:
chromium --ui-disable-partial-swap

Thanks for the report Tomas, this is indeed the problem. It was reported for nouveau already, at  https://crbug.com/649296  I'm looking at it.