(No comment was entered for this change.)

I don't see this on other platforms, including some that don't have NEON either (that 
we also build with -mfpu=vfpv3-d16).
Is it a recent regression ? We slightly changed the way we pass the mfpu flag to gcc, 
you can specify explicitly what you want by adding e.g. "armv7=1 arm_neon=0 
arm_fpu=vfpv3-d16" to the GYP_DEFINES.

(No comment was entered for this change.)

(No comment was entered for this change.)

As I mentioned, I can't reproduce this on other similar platforms and I don't have 
that one handy.

ok ... i will try the flags you suggested early next week when i am back where i have
that board

I built chromium on Dove board. See http://pastebin.ubuntu.com/448257/ for debdiff

Basically:

diff -u chromium-browser-5.0.342.9~r43360/debian/rules chromium-browser-5.0.342.9~r43360/debian/rules
--- chromium-browser-5.0.342.9~r43360/debian/rules
+++ chromium-browser-5.0.342.9~r43360/debian/rules
@@ -39,6 +39,8 @@
 	target_arch=arm \
 	disable_nacl=1 \
 	linux_use_tcmalloc=0 \
+	armv7=1 \
+	arm_neon=0 \
 	$(NULL)
 WANT_LZMA_DEBS := 0
 WANT_TESTS     := 0

The browser dies with "Illegal Instruction"

When NEON is disabled, it defaults to vfpv3. You may want to try adding arm_fpu=vfpv3-d16 which I believe is the default on Ubuntu.

some more information: when I opened page in  chromium, showed source( looked like junk) copy, paste in ...html file and open that file in firefox, page looked reasonable. So looks like just a rendering on the screen is screwed up.

Also I rebuilt chromium yesterday with:

diff -u chromium-browser-5.0.342.9~r43360/debian/changelog chromium-browser-5.0.342.9~r43360/debian/changelog
--- chromium-browser-5.0.342.9~r43360/debian/changelog
+++ chromium-browser-5.0.342.9~r43360/debian/changelog
@@ -1,3 +1,12 @@
+chromium-browser (5.0.342.9~r43360-0ubuntu2common0) lucid; urgency=low
+
+  [ Alex Wolfson ]
+  * Changed build option in GYP_DEFINES
+   https://code.google.com/p/chromium/issues/detail?id=38957
+   (LP: #539245) 
+
+ -- root <root@ubuntu-laptop>  Fri, 11 Jun 2010 14:42:12 +0000
+
 chromium-browser (5.0.342.9~r43360-0ubuntu2) lucid; urgency=low
 
   [ Fabien Tassin <fta@ubuntu.com> ]
diff -u chromium-browser-5.0.342.9~r43360/debian/rules chromium-browser-5.0.342.9~r43360/debian/rules
--- chromium-browser-5.0.342.9~r43360/debian/rules
+++ chromium-browser-5.0.342.9~r43360/debian/rules
@@ -21,7 +21,7 @@
 WANT_TESTS       ?= 1
 WANT_SANDBOXING  ?= 1
 USE_SYSTEM_V8    ?= 0
-VERBOSE          ?= 0
+VERBOSE          ?= 1
 PROCESSORS       ?= $(shell grep -ic ^processor /proc/cpuinfo)
 ifeq (0,$(PROCESSORS))
 PROCESSORS := 1
@@ -39,10 +39,13 @@
 	target_arch=arm \
 	disable_nacl=1 \
 	linux_use_tcmalloc=0 \
+	armv7=1 \
+	arm_neon=0 \
 	$(NULL)
 WANT_LZMA_DEBS := 0
 WANT_TESTS     := 0
-ifneq (,$(filter unstable development,$(DEBIAN_DIST)))
+#TODO:change that ifneq, so it will work for upcomming releases of ubuntu
+ifneq (,$(filter unstable development 10.04,$(DEBIAN_DIST)))
 # on $(DEB_BUILD_ARCH) and $(DEBIAN_DIST), enforce an armv7 with thumb build (or not)
 GYP_DEFINES += \
 	armv7=0 \

It does not crash, but renders garbage. I did not setup arm_fpu=vfpv3-d16 at that time

I finally managed to cross build chromium for ARM

see: https://wiki.linaro.org/DeveloperDocs/CrossBuildingChromiumBrowser

Build runs on Marvell's Dove board. 
Can somebody look at the wiki pages to sanity check options that I used?
And the main question - How to proceed with actual debugging. Any suggestions?

Your options look reasonable, however I've never had the need to add -idirafter to the CFLAGS if you specify sysroot= correctly in the GYP_DEFINES. If it looks like it's pulling includes out of /usr/include (outside of the sysroot) it usually means you're missing .pc files in your sysroot (note: they need to be present when you run gyp).
Also I haven't had the need to generate a new config.h for ffmpeg, the default arm/non-neon one works on another vfpv3-d16 arch.

Do you have suggestions how to start debugging? I never dealt with chromium before

Some suggestions, trace the data path:
- check if test_shell works or not (you can build it with make test_shell). If it doesn't it's likely the bug is in WebKit and/or Skia.
- If it does, then debug chrome (the renderer process - find the PID from the process manager), check whether the renderer painted correctly. You'll want to check in chrome/renderer/render_widget.cc all the users of PaintRect (mostly DoDeferredUpdate) to see if the contents looks right.
- If that looked right, then debug the browser process, check in chrome/browser/renderer_host/backing_store_x.cc if PaintToBackingStore gets the right data and if it paints the backing store correctly.
- If that was right too, then debug the XShowRect/CairoShowRect/PaintToRect functions in the same file, that copies from the backing store to the window.

I built test_shell and am running it now. Looks like it will take a very long time, but 592 tests already failed and screen looked messed up like in chromium itself.

BTW does it exist a way to copy tests from build PC to the board "make install" equivalent or something? Looks like I copied much more then was necessary, which took a long time. 

I built a tests for skia. They all passed. 
Any suggestions - how to debug WebKit? 
It seems that most likely cause of a bug is  Webkit skia binding.

The chromium built with export "GYP_DEFINES="release_optimize=1...." works properly on the same board. Speed of lunching chromium is not distinguishable visually, comparing to the default optimization (I believe it is -02). Debugging build was visible slower, but rendered properly as well.

(No comment was entered for this change.)

Looks like the problem disappeared downstream.

This issue has been closed for some time. No one will pay attention to new comments.
If you are seeing this bug or have new data, please click New Issue to start a new bug.

(No comment was entered for this change.)