NEW 153713
[GTK] 22 new JSC test failures on the GTK+ ARM buildbot after r195233. 
https://bugs.webkit.org/show_bug.cgi?id=153713
Summary [GTK] 22 new JSC test failures on the GTK+ ARM buildbot after r195233.
Carlos Alberto Lopez Perez
Reported 2016-01-30 12:58:38 PST
This is a clean build of r195232 https://build.webkit.org/builders/GTK%20Linux%20ARM%20Release/builds/10106 This is a clean build of r195233 https://build.webkit.org/builders/GTK%20Linux%20ARM%20Release/builds/10107 As you see, it seems that r195233 <http://trac.webkit.org/r195233> has caused 22 new JSC test failures on the GTK+ ARM buildbot. Looking at the log https://build.webkit.org/builders/GTK%20Linux%20ARM%20Release/builds/10107/steps/jscore-test/logs/stdio it seems the tests are failing because of "Illegal instruction". I don't know what is going on here, but I doubt r195233 has directly caused this regression. Any idea?
Attachments
Add attachment proposed patch, testcase, etc.
Csaba Osztrogonác
Comment 1 2016-02-01 10:49:00 PST
JSCTEST_timeout is 60 seconds by default (set by run-javascriptcore-tests) and is doubled by run-jsc-stress-tests if FTL JIT is enabled and then multiplied with sqrt(number_of_processors). r194601 disabled timeout accidentally, I only enabled it again in r195233. Maybe your hardware needs more timeout than the default. Could you try if setting bigger timeout via JSCTEST_timeout env fixes this issue? "Illegal instruction ERROR: Unexpected exit code: 132" We should see "Timed out after xx.xxxxxxx seconds!" message too (dumped by jsc.cpp), but it seems it is disabled on your bot.
Carlos Alberto Lopez Perez
Comment 2 2016-02-05 04:37:11 PST
I don't manage the GTK+ ARM buildbot. Kov (Gustavo) is the admin. What I wonder is how the illegal instructions can be raised if the tests were working before? :\
Csaba Osztrogonác
Comment 3 2016-02-05 04:50:18 PST
(In reply to comment #2) > I don't manage the GTK+ ARM buildbot. Kov (Gustavo) is the admin. > > What I wonder is how the illegal instructions can be raised if the tests > were working before? :\ It's normal. When timeout happens, CRASH() calls __builtin_trap() which is an illegal instruction. Kov, could you try to set JSCTEST_timeout env to a bigger number?
Gustavo Noronha (kov)
Comment 4 2016-02-08 04:16:43 PST
Sure, I'll do that later today (I have to poke the sysadmins to replace my ssh key on the server, the old one was removed while I was on vacation).
Gustavo Noronha (kov)
Comment 5 2016-02-08 05:15:49 PST
Just did it. I double it to 120, let's see if it works.
Carlos Alberto Lopez Perez
Comment 6 2016-02-11 17:53:32 PST
(In reply to comment #3) > (In reply to comment #2) > > What I wonder is how the illegal instructions can be raised if the tests > > were working before? :\ > > It's normal. When timeout happens, CRASH() calls __builtin_trap() > which is an illegal instruction. > But it don't prints any "Timed out after xxx seconds! message as it should do because of http://trac.webkit.org/browser/trunk/Source/JavaScriptCore/jsc.cpp?rev=195233#L1675 If i run " JSCTEST_timeout=1 Tools/Scripts/run-javascriptcore-tests --gtk --release --no-build " on my laptop I get the timeouts printed when they happen. ??? (In reply to comment #5) > Just did it. I double it to 120, let's see if it works. Seems there still many tests crashing with illegal instruction. Kov, can you do a " sudo apt-get install ruby-json ruby-highline " ? it is complaining about this missing libraries on the top of the log. Thanks
Carlos Alberto Lopez Perez
Comment 7 2016-04-04 09:12:30 PDT
(In reply to comment #6) > Kov, can you do a " sudo apt-get install ruby-json ruby-highline " ? it is > complaining about this missing libraries on the top of the log. Kov ping?
Icenowy Zheng
Comment 8 2016-06-11 05:42:08 PDT
I've built the latest release (2.12.3) of webkit2gtk on my arm machine, and even run 1+1 in jsc will cause SIGILL...
Carlos Alberto Lopez Perez
Comment 9 2016-06-11 13:01:51 PDT
(In reply to comment #8) > I've built the latest release (2.12.3) of webkit2gtk on my arm machine, and > even run 1+1 in jsc will cause SIGILL... Could you generate a coredump and get a backtrace?
Carlos Alberto Lopez Perez
Comment 10 2016-06-11 13:08:47 PDT
(In reply to comment #8) > I've built the latest release (2.12.3) of webkit2gtk on my arm machine, and > even run 1+1 in jsc will cause SIGILL... I cannot reproduce that on an ARMv7 machine (a wandboard) with Debian Jessie using webkit2gtk+ from debian-backports (2.12.0-2~bpo8+1) $ dpkg -S $(which jsc) libjavascriptcoregtk-4.0-bin: /usr/bin/jsc $ apt-cache policy libjavascriptcoregtk-4.0-bin libjavascriptcoregtk-4.0-bin: Installed: 2.12.0-2~bpo8+1 Candidate: 2.12.0-2~bpo8+1 Version table: *** 2.12.0-2~bpo8+1 0 100 http://ftp.debian.org/debian/ jessie-backports/main armhf Packages 100 /var/lib/dpkg/status 2.6.2+dfsg1-4 0 500 http://ftp.pt.debian.org/debian/ jessie/main armhf Packages $ jsc -e "print(1 + 1)" 2
Note You need to log in before you can comment on or make changes to this bug.