The stacktrace unit test was flaking on arm32; my theory is that this
happened when the thread whose stack we were dumping was doing a
system call inside `params->deadlock_start_event.Set();` in
ThreadFunction(). (This would be a problem because, according to the
comment at the bottom of the file, "stack traces originating from
kernel space do not include user space stack traces for ARM32.")
Attempt to solve this problem by spinning on an atomic flag instead,
since this involve no system calls. And add a short sleep to the main
thread, to give the other thread time to get from the barrier to the
thing it's actually supposed to deadlock on.
Bug: webrtc:10420
Change-Id: I4c6392157c8a06c64cb11146ffe9368e5bade6fc
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/128340
Commit-Queue: Karl Wiberg <kwiberg@webrtc.org>
Reviewed-by: Magnus Jedvert <magjed@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#27158}
0611a15c29