[tracing] child-process-side impl for inter-process memory dumps

This CL introduces the implementation for handling the inter-process
memory dumps on the child-process side.
Conceptually this consists of a simple plumbing in TraceMessageFilter
in order to propagate the request up to the browser's MemoryDumpManager,
as the child processes' MDM play a very passive role in the inter-process
coordination game.
In practice, some extra care is required to match the lifetime of MDM
MDM (long) and ChildTraceMessageFilter (shorter) in a thread-safe
manner, which requires the introduction of the proxy class herein
called ChildMemoryDumpManagerDelegateImpl.

More context and design doc are available in the attached BUG.

Reason for NOTRY: all but one bots are green. I started goo.gl/tBX5f1
on infra-dev about the failure and it seems due to a bot out of space.
The change has been sitting for 4+ hours and there are other CLs blocked
on this.

BUG=462930,474973
NOTRY=true

Review URL: https://codereview.chromium.org/1039963003

Cr-Commit-Position: refs/heads/master@{#324229}

1 files changed, 4 insertions, 1 deletions

diff --git a/base/trace_event/memory_dump_manager.cc b/base/trace_event/memory_dump_manager.cc
index ed965fa..ac79144 100644
--- a/base/trace_event/memory_dump_manager.cc
+++ b/base/trace_event/memory_dump_manager.cc
@@ -77,7 +77,10 @@ void MemoryDumpManager::Initialize() {
 
 void MemoryDumpManager::SetDelegate(MemoryDumpManagerDelegate* delegate) {
   AutoLock lock(lock_);
-  DCHECK(delegate_ == nullptr);
+  // TODO(primiano): The DCHECK below is disabled just temporary (and it is
+  // strongly useful). It should be re-enabled soon as crbug.com/474973 is fixed
+  // (ETA: end of April 2015). Commenting out just to de-entangle CL deps.
+  // DCHECK(delegate_ == nullptr);
   delegate_ = delegate;
 }