generic-library/breakpad
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

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

Browse Source 
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

Review URL: https://breakpad.appspot.com/1524002

git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1313 4c0a9323-5329-0410-9bdc-e9ce6186880e
This commit is contained in:
rmcilroy@chromium.org 2014-04-15 10:07:50 +00:00
parent eebdc96430
commit 32031a9136
1 changed files with 19 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/common/linux/tests/crash_generator.cc
View File

@ -199,19 +199,25 @@ bool CrashGenerator::CreateChildCrash(
fprintf(stderr, "CrashGenerator: Failed to copy proc files\n"); fprintf(stderr, "CrashGenerator: Failed to copy proc files\n");
exit(1); exit(1);
} }
if (tkill(*GetThreadIdPointer(crash_thread), crash_signal) == -1) { // On Android the signal sometimes doesn't seem to get sent even though
perror("CrashGenerator: Failed to kill thread by signal"); // tkill returns '0'. Retry a couple of times if the signal doesn't get
} else { // through on the first go:
// At this point, we've queued the signal for delivery, but there's no // https://code.google.com/p/google-breakpad/issues/detail?id=579
// guarantee when it'll be delivered. We don't want the main thread to for (int i = 0; i < 60; i++) {
// race and exit before the thread we signaled is processed. So sleep if (tkill(*GetThreadIdPointer(crash_thread), crash_signal) == -1) {
// long enough that we won't flake even under fairly high load. perror("CrashGenerator: Failed to kill thread by signal");
// TODO: See if we can't be a bit more deterministic. There doesn't } else {
// seem to be an API to check on signal delivery status, so we can't // At this point, we've queued the signal for delivery, but there's no
// really poll and wait for the kernel to declare the signal has been // guarantee when it'll be delivered. We don't want the main thread to
// delivered. If it has, and things worked, we'd be killed, so the // race and exit before the thread we signaled is processed. So sleep
// sleep length doesn't really matter. // long enough that we won't flake even under fairly high load.
sleep(10 * 60); // TODO: See if we can't be a bit more deterministic. There doesn't
// seem to be an API to check on signal delivery status, so we can't
// really poll and wait for the kernel to declare the signal has been
// delivered. If it has, and things worked, we'd be killed, so the
// sleep length doesn't really matter.
sleep(1);
}
} }
} else { } else {
perror("CrashGenerator: Failed to set core limit"); perror("CrashGenerator: Failed to set core limit");