[base] Don't try to decommit pages if they're about to be unmapped.

DiscardableSharedMemory::Purge() attempts to proactively discard pages
to prevent slow-to-respond clients from keeping the pages live longer.

If the instance is in-process then these kernel calls are unnecessary
since the memory is about to be unmapped anyway.

Bug: 1023693
Change-Id: I2bc19eb64f3fdc009c5bf8c4ed6369b2b5081045
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1911487
Auto-Submit: Wez <wez@chromium.org>
Reviewed-by: Daniel Cheng <dcheng@chromium.org>
Commit-Queue: Wez <wez@chromium.org>
Cr-Commit-Position: refs/heads/master@{#714684}

base/memory/discardable_shared_memory.cc

@@ -368,6 +368,15 @@ bool DiscardableSharedMemory::Purge(Time current_time) {
     return false;
   }
 
+  // If |shared_memory_region_| handle is invalid then the memory will be
+  // released by the manager Unmap()ping |this| in response to Purge() returning
+  // true, so there is no point spending time calling discard APIs.
+  // This is the case for in-process instances.
+  if (!shared_memory_region_.IsValid()) {
+    last_known_usage_ = Time();
+    return true;
+  }
+
 // The next section will release as much resource as can be done
 // from the purging process, until the client process notices the
 // purge and releases its own references.