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

Issue 717687 link

Starred by 7 users

Issue metadata

Status: Untriaged
Owner: ----
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 2
Type: Bug



Sign in to add a comment

kevin: user reports GPU process taking up over 2GB

Project Member Reported by diand...@chromium.org, May 2 2017

Issue description

Chrome Version: 56.0.2924.104, 57.0.2987.146
OS: Chrome OS

Two feedback reports.  Same user:

* Report=54068673739
* Report=58738057593

GPU process shows as taking 2.5 GB.  ...it's causing OOM kills, but the root cause is really that the GPU process seems to be eating like there's no tomorrow.  Here's a small snippet from 58738057593

--

[15144.025997] Out of memory: Kill process 20486 (chrome) score 519 or sacrifice child
[15144.026031] Killed process 20486 (chrome) total-vm:545776kB, anon-rss:1192kB, file-rss:7064kB
[15144.029751] chrome invoked oom-killer: gfp_mask=0x24280c2, order=0, oom_score_adj=-1000
[15144.029763] chrome cpuset=/ mems_allowed=0
[15144.029778] CPU: 4 PID: 1696 Comm: chrome Not tainted 4.4.35-06801-g74bef12 #1
[15144.029782] Hardware name: Google Kevin (DT)
[15144.029786] Call trace:
[15144.029803] [<ffffffc0002087dc>] dump_backtrace+0x0/0x160
[15144.029808] [<ffffffc00020895c>] show_stack+0x20/0x28
[15144.029816] [<ffffffc0004bbea8>] dump_stack+0x90/0xb0
[15144.029823] [<ffffffc00035088c>] dump_header.isra.10+0x64/0x198
[15144.029829] [<ffffffc00030661c>] oom_kill_process+0x98/0x3e8
[15144.029834] [<ffffffc000306c5c>] out_of_memory+0x258/0x2ec
[15144.029840] [<ffffffc00030b9ac>] __alloc_pages_nodemask+0x7d8/0x97c
[15144.029847] [<ffffffc0005d4354>] kbase_mem_pool_alloc_page.isra.15+0x58/0x150
[15144.029852] [<ffffffc0005d4ca8>] kbase_mem_pool_alloc_pages+0xd8/0x148
[15144.029856] [<ffffffc0005d4c80>] kbase_mem_pool_alloc_pages+0xb0/0x148
[15144.029861] [<ffffffc0005b9b70>] kbase_alloc_phy_pages_helper+0xa0/0x15c
[15144.029865] [<ffffffc0005b9eec>] kbase_alloc_phy_pages+0x50/0xa4
[15144.029870] [<ffffffc0005caab4>] kbase_mem_alloc+0x34c/0x5d0
[15144.029874] [<ffffffc0005cf7d8>] kbase_ioctl+0x274/0xa94
[15144.029882] [<ffffffc0003a5e1c>] compat_SyS_ioctl+0x134/0x10b8
[15144.029888] [<ffffffc000203e90>] __sys_trace_return+0x0/0x4

...

[15144.029994] [ pid ]   uid  tgid total_vm      rss nr_ptes nr_pmds swapents oom_score_adj name
[15144.030156] [ 1696]  1000  1696    99899   680154     150       2     8548         -1000 chrome

...

And the screenshot of top confirms that PID has 2.5GB resident.

---

Maybe someone can look at the UI log and see if there's anything worrisome?
 
Cc: vsu...@chromium.org

Comment 2 by cklin@google.com, May 12 2017

Additional feedback reports:

* 59244935859
* 59246310635
* 59332514589
* 59329526708
* 59404534233

All feedback reports are captured on kevin.  I have never been able to reproduce this problem on my other Chromebooks (Acer C720, Chromebook Pixel 2015), so the problem is device specific (though not necessarily specific to kevin).

Video of crosh top showing GPU process rss memory growth:
https://goo.gl/photos/Z4Sa6GGmtyD3jMKK6

The video also suggests a possible renderer memory leak, which should be tracked separately.

See attachment for memory-infra trace captured while reproducing the problem.

I have been using the following repro procedure, which typically succeeds in introducing the problem within an hour of trying.

* Play one of the DailyMotion videos (labeled DM) from http://maplestage.com/episode/p29qO/%E7%84%A6%E6%80%A5%E7%9A%84%E7%BE%85%E6%9B%BC%E5%8F%B2-%E7%AC%AC6%E9%9B%86-My-Secret-Romance-Ep-6

* Watch GPU process RSS memory from top.  If GPU process memory remains below 500MB (as opposed to jumping to around 700MB), the problem is most likely not happening.  Reboot and try again.

* It may help to leave the system off for a few minutes before turning it back on.  (This is getting into cargo-cult territory, but who knows).

trace_2017-05-04_19.18.26.json.gz
3.3 MB Download
Labels: videoshortlist
From the above steps, one thing that's kinda odd is that it happens sometimes but not other times.  I wonder if perhaps it depends on what video quality you end up getting?  This video site appears to default to "Auto" quality, so perhaps depending on the speed of your internet connection you might get 720p, 480p, 380p, 240p, 144p, ...  ...and maybe the problem only reproduces with one of those quality settings?
...or, I suppose, it could also matter what advertisement is served to you.  I wonder if an add blocker would make the problem go away?

Comment 6 by willg...@gmail.com, Jun 1 2017

Comment #5, I see this as a YouTube Red member as well, so ads don't appear to make a difference. 

Just a long YouTube TV or Netflix session with a couple of tabs open has me hitting swap. 
Possibly related: crbug.com/700776.

Comment 8 Deleted

Comment 9 Deleted

Labels: -videoshortlist

Sign in to add a comment