posix thread: Attempt to set audio threads to real time.

base/threading/platform_thread_linux.cc

Index: base/threading/platform_thread_linux.cc
diff --git a/base/threading/platform_thread_linux.cc b/base/threading/platform_thread_linux.cc
index 42e12c71a554920d94c23c503ff02c80bf8aaad4..8556431a36a9b2826a42e96213d4b278b776ba54 100644
--- a/base/threading/platform_thread_linux.cc
+++ b/base/threading/platform_thread_linux.cc
@@ -76,6 +76,24 @@ void PlatformThread::SetName(const char* name) {
 void PlatformThread::SetThreadPriority(PlatformThreadHandle handle,
                                        ThreadPriority priority) {
 #ifndef OS_NACL

[jar (doing other things), 2013/05/29 02:07:04]

nit: To make this a bit more readable, how about:
#if defined(OS_NACL)
  return;
#endif  // OS_NACL

I didn't like having to read the nested ifdefs.... especially as the  range has
grown.

[dgreid, 2013/05/29 18:52:34]

agreed, it would be better, but it won't compile, most of the below functions
don't exist there.

[jar (doing other things), 2013/05/30 18:15:25]

OK...
Please add comments after each #endif to indicate what the matching ifdef was.

[dgreid, 2013/05/30 18:59:22]

Done.

+  if (priority == kThreadPriority_RealtimeAudio) {
+    const int kRealTimePrio = 8;
+
+    struct sched_param sched_param;
+    memset(&sched_param, 0, sizeof(sched_param));
+    sched_param.sched_priority = kRealTimePrio;
+
+    if (pthread_setschedparam(pthread_self(), SCHED_RR, &sched_param) == 0) {
+      // Got real time priority, no need to set nice level.
+      return;
+    }
+
+    LOG(INFO) << "Failed to set audio thread to real time.";

[jar (doing other things), 2013/05/29 02:07:04]

We've tended to avoid doing LOG() in favor of DLOG().  Most (all?) of the time,
no one pays attention to this, and we we blowing a ton of memory with these
strings.  Brett went through the code removing pretty much all he could find...
so it is probably better to stick with that plan.

[dgreid, 2013/05/29 18:52:34]

I'm OK removing this, it's probably not worth the string since it shouldn't ever
fail.  It would leave us without the any insight into the failure, but I can
investigate a way to list realtime threads in chromeos feedback that would be a
good indicator.  Sound like a good solution?

[jar (doing other things), 2013/05/30 18:15:25]

WFM

[dgreid, 2013/05/30 18:59:22]

Great, I've also found a way to test that this and the below code never run on
ChromeOS.

+#if defined(OS_CHROMEOS)
+    return;  // RT prio should always work on ChromeOS. Never set nice level.

[jar (doing other things), 2013/05/29 02:07:04]

This seems strange.  Are you saying the request should fail... and return an
error on line 86, but you should ignore that error and act like it succeeded?

[dgreid, 2013/05/29 18:52:34]

On linux the request will fail unless the user has explicitly allowed chrome RT
thread access then attempt to set the nice level as a fallback.  Per comments on
patch set one, the fallback behavior wasn't desired for chrome os we bail here
instead.

I'm not attached to this part at all.  This is code that should never run.  Do
you think it would be better without it?

[jar (doing other things), 2013/05/30 18:15:25]

If you think it will never run... less code seems better.

...unless there is a catastrophic problem with the "default" behavior.   It
sounds like they have problems when they reach this point... and adding ifdefs
to little advantage seems like a waste.

+#endif
+  }
+
   // setpriority(2) will set a thread's priority if it is passed a tid as
   // the 'process identifier', not affecting the rest of the threads in the
   // process. Setting this priority will only succeed if the user has been