Debugging on unavailable hardware

Hi all,

In revision 223114 I committed a fix to clang project. It caused fail on clang-native-arm-cortex-a9: http://lab.llvm.org:8011/builders/clang-native-arm-cortex-a9/builds/23599. According to logs it was a problem with memory allocated by malloc. Reverting the fix in 223120 made the bot green. The changed code does nothing with heap, none of other bots failed, so it looks like a platform specific problem.

Could somebody advice how I can investigate the problem? I don’t have appropriate hardware and running tests in emulated environment does not shows any problem, all tests pass.

Thank you!

–Serge

In revision 223114 I committed a fix to clang project. It caused fail on
clang-native-arm-cortex-a9:
http://lab.llvm.org:8011/builders/clang-native-arm-cortex-a9/builds/23599.
According to logs it was a problem with memory allocated by malloc.
Reverting the fix in 223120 made the bot green. The changed code does
nothing with heap, none of other bots failed, so it looks like a platform
specific problem.

Hi Serge,

We're seeing this behaviour, too, and on unrelated commits. We're
working to make that bot more stable.

That same commit passed on another ARM bot:

http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15/builds/2123

which means that it probably had nothing to do with your commit, but
the bot's instability. I suggest that, for the time being, if you see
a failure in one ARM bot, check the others to make sure the problem
can be reproduced across the board.

If all fail, revert your patch and we'll try to debug. If only one
fail but not the other, wait until the next build finishes. If the
build is still broken the same way, revert. If not, ignore.

Could somebody advice how I can investigate the problem? I don't have
appropriate hardware and running tests in emulated environment does not
shows any problem, all tests pass.

If the problem persists and you need to debug the problem, let me know
and I could give you temporary access to an ARM board, or even run
that for you.

Alternatively, there is a new cloud service that I'm testing
(https://cloud.online.net/) that provides remote *real* ARM boards for
testing. They have a 15 minute trial that you could run some examples
and retrieve the logs.

cheers,
--renato

Hi Renato,

Thank you very much for the directions, I am going to recommit my fix.

What are hardware used in buildbots? Are these common boards like PandaBoard or some thing special? What is RAM installed?

The one that failed is a Panda (Cortex-A9) board. The ones that pass
are more stable Cortex-A15 boards. The Pandas were never meant to run
for a huge amount of time, especially not at that kind of hammering
level...

cheers,
--renato

OK, thank you!