[Android]: Fix hang in CreateChildCrash() on Android.

[Android]: Fix hang in CreateChildCrash() on Android.

After r1299, the LinuxCoreDumperTest::VerifyDumpWithMultipleThreads and
ElfCoreDumpTest::ValidCoreFile would both hang on Android.  This appears to be
due to the tkill
signal not being recieved by the thread which is meant to crash, even though
tkill returns 0.
This CL retries sending the tkill signal multiple times, which prevents the
Hang.

BUG=579
R=thestig@chromium.org

Committed: 1313

[rmcilroy]

thestig: Could you please review this CL.

The CL - https://breakpad.appspot.com/1304005 causes the breakpad_unittests to
hang in the sleep(10*60) call on Android, which is preventing a roll of Breakpad
to Chromium (see https://codereview.chromium.org/234843004/).  Repeating the
call to tkill() seems to avoid this hang, and shouldn't cause any other issues
with the tests.  PTAL.

[Lei Zhang (chromium)]

lgtm, though we should not have to do this. Can you file a bug and investigate
why the signal is not being delivered successfully?

The tkill() man page says you may target the wrong thread if the thread id
passed to tkill() has been recycled, but assuming thread_function() succeeded,
then the thread waiting to be killed is just sitting there, so that should not
be an issue.

https://breakpad.appspot.com/1524002/diff/1/common/linux/tests/crash_generato...
File common/linux/tests/crash_generator.cc (right):

https://breakpad.appspot.com/1524002/diff/1/common/linux/tests/crash_generato...
common/linux/tests/crash_generator.cc:205: for(int i = 0; i < 60; i++) {
nit: "for ("

[rmcilroy]

Thanks Lei: Filed https://code.google.com/p/google-breakpad/issues/detail?id=579
for this.

https://breakpad.appspot.com/1524002/diff/1/common/linux/tests/crash_generato...
File common/linux/tests/crash_generator.cc (right):

https://breakpad.appspot.com/1524002/diff/1/common/linux/tests/crash_generato...
common/linux/tests/crash_generator.cc:205: for(int i = 0; i < 60; i++) {
On 2014/04/15 03:28:49, Lei Zhang (chromium) wrote:
> nit: "for ("

Done.

[rmcilroy]

Committed patchset #1 manually as r1313 (presubmit successful).

[vapier]

my guess is the android threading implementation is limited.  iirc, it's
all in userspace.  sending tkill twice is wrong and probably introduces
flakiness in other targets.

try turning the long sleep into a loop that calls sleep(1) many times.
-mike

[rmcilroy]

On 2014/04/15 13:33:40, vapier wrote:
> my guess is the android threading implementation is limited.  iirc, it's
> all in userspace.  sending tkill twice is wrong and probably introduces
> flakiness in other targets.
> 
> try turning the long sleep into a loop that calls sleep(1) many times.
> -mike

I tried turning the long sleep into many sleep(1) calls as suggested, however
this doesn't work unfortunately.  I've cc'ed you on the bug in case you have any
other suggestions.

[vapier]

then at least bracket the logic with an android ifdef.  we shouldn't send out
tkill more than once on a sane system.