New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 630373 link

Starred by 3 users

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 3
Type: Bug

Blocking:
issue 698266



Sign in to add a comment

MemoryInfra: report base::SharedMemory in the heap profiler

Project Member Reported by primiano@chromium.org, Jul 21 2016

Issue description

Background context:go/memory-infra: memory profiling in chrome://tracing


This came from the "15 MB reduction in Tumblur!" thread in project-trim@.
Thinking about that, reporting the shared memory totals in general memory-infra counters is a bit tricky (bashi@ is looking into that in  Issue 604726 ).
But plumbing that to the heap profiler should be way easier and I don't see why that should escape the singularity :)
 
Components: Blink>MemoryAllocator

Comment 2 by tasak@google.com, Aug 25 2016

Owner: tasak@google.com
Status: Assigned (was: Untriaged)
Components: Internals>Instrumentation>Memory
Components: -Blink>MemoryAllocator
Blocking: 698266

Comment 6 by ssid@chromium.org, Mar 7 2017

Why do we need to hook this into heap profiler? We will have a dump provider and I assume that most of the shared memory is being reported by discardable and cc. Maybe we should add more dump providers instead of hooking into heap profiler.

Comment 7 by tasak@google.com, May 23 2017

Cc: tasak@google.com
Owner: hajimehoshi@chromium.org
hajimehoshi@ is now working on how to report shared memory usage.

I was wondering what is the difference from crbug/604726. Duplicated?
Cc: erikc...@chromium.org
Labels: -Pri-2 Pri-3
Re #8: no. This is lower-prio specifically only about the heap profiler.  Issue 604726  is about doing the right accounting at a MDP level, so we can use that for general memory-infra usage, UMA and whatnot (and is the more important big).
This one, instead, is an add-on to add stacktraces to base::SharedMemory calls similarly to what we do for {malloc, blinkgc, partition_alloc}.
That would allow us to see which code is causing shared_memory usage. 
 Issue 604726  will just allow us to know the general total, and eventually the break down (eventually : if ownership edges are declared) by clients.

This bug essentially involves plumbing a 4th AllocationRegister to base::SharedMemoryTracker and hooking insert/remove calls when --enable-heap-profiling is enabled.

Sign in to add a comment