Fix memory leaks from adding GPU memory to task manager

There were two leaks detected here.  

The first is from a preexisting bug. GpuProcessHost::SendOnIO will leak the message if the IO thread is not present.  Note that it will not leak the message if the IO thread is present, but the GPU process is absent.

The second issue is that we would keep around one GpuData-Observer indefinitely.  I changed this to be scoped to the task manager.

On the second issue I have a concern here that there may be a race between destruction of the task manager and the GpuDataManager.  The TaskManagerModel dtor calls into GpuDataManager::RemoveObserver, but is this safe if the GpuDataManager's dtor has been called first?  Do we offer any guarantees of this?  In practice, neither singleton's dtor is called when I just run the browser, and the GpuDataManager's dtor is called second in the unit test.

BUG=143563
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=152847

[ccameron, 2012-08-20 21:20:02]

http://codereview.chromium.org/10861018/diff/1/chrome/browser/task_manager/ta...
File chrome/browser/task_manager/task_manager.h (right):

http://codereview.chromium.org/10861018/diff/1/chrome/browser/task_manager/ta...
chrome/browser/task_manager/task_manager.h:575: 
The dtor for the observer will call into GpuDataManagerImpl::RemoveObserver --
do we have guarantees of dtor order here?

http://codereview.chromium.org/10861018/diff/1/content/browser/gpu/gpu_proces...
File content/browser/gpu/gpu_process_host.cc (right):

http://codereview.chromium.org/10861018/diff/1/content/browser/gpu/gpu_proces...
content/browser/gpu/gpu_process_host.cc:301: &SendGpuProcessMessage, kind,
cause, message))) {
Note that SendGpuProcessMessage will delete message if the GPU is disabled, but
this didn't of we had no IO thread.

[Ken Russell (switch to Gerrit), 2012-08-20 23:22:58]

The content/browser/gpu/ changes LGTM (I double-checked the implementation of
BrowserThreadImpl::PostTaskHelper to verify when it returns false) but Al should
take a look at this too.

[ccameron, 2012-08-22 08:24:42]

Pinging this again.

[yoshiki, 2012-08-22 08:31:29]

chrome/browser/task_manager/* LGTM

[ccameron, 2012-08-22 09:23:19]

Thank you!

Al, did you want to look, too?

[apatrick_chromium, 2012-08-22 19:22:21]

content/browser/gpu LGTM

[ccameron, 2012-08-22 19:25:13]

Thanks!

[commit-bot: I haz the power, 2012-08-22 19:25:39]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ccameron@chromium.org/10861018/1

[commit-bot: I haz the power, 2012-08-22 22:46:08]

Change committed as 152847