820898 - Memory reporter for worker's JS runtime is not being called

Status: RESOLVED DUPLICATE of bug 813867

(Core :: DOM: Workers, defect)

Description

[Justin Lebar (not reading bugmail)]

The DMD report in attachment 691386 [details] shows the following set of unreported blocks in the main process with the B2G gallery app open and in the foreground.

> Unreported: 4 blocks in block group 7 of 724
>  458,752 bytes (448,064 requested / 10,688 slop)
>  1.67% of the heap (19.15% cumulative);  3.41% of unreported (39.19% cumulative)
>  Allocated at
>    malloc /Users/jlebar/code/moz/ff-git/src/memory/build/replace_malloc.c:152 (0x4024e2f6 libmozglue.so+0x42f6)
>    js_new<JSRuntime, JSUseHelperThreads> /Users/jlebar/code/moz/B2G/objdir-gecko/dist/include/js/Utility.h:496 (0x41b7c6f2 libxul.so+0xc686f2)
>    CreateJSContextForWorker /Users/jlebar/code/moz/ff-git/src/dom/workers/RuntimeService.cpp:382 (0x41543f04 libxul.so+0x62ff04)
>    nsThread::ProcessNextEvent(bool, bool*) /Users/jlebar/code/moz/ff-git/src/xpcom/threads/nsThread.cpp:627 (0x419679de libxul.so+0xa539de)
>    NS_ProcessNextEvent_P(nsIThread*, bool) /Users/jlebar/code/moz/B2G/objdir-gecko/dist/include/nsError.h:156 (0x41947c2a libxul.so+0xa33c2a)
>    nsThread::ThreadFunc(void*) /Users/jlebar/code/moz/ff-git/src/xpcom/threads/nsThread.h:63 (0x41967e28 libxul.so+0xa53e28)
>    _pt_root /Users/jlebar/code/moz/ff-git/src/nsprpub/pr/src/pthreads/ptthread.c:159 (0x40180a70 libnspr4.so+0x19a70)
>    __thread_entry /Users/jlebar/code/moz/B2G/bionic/libc/bionic/pthread.c:218 (0x400cde18 libc.so+0x12e18)
>    pthread_create /Users/jlebar/code/moz/B2G/bionic/libc/bionic/pthread.c:362 (0x400cd96c libc.so+0x1296c)

Comment 1

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Weird, we have a memory reporter... The only time I know of that the worker runtime won't be included in the report is when the worker is using ctypes, but I don't think that's possible in gaia apps...

Comment 3

[Andrew McCreight [:mccr8]]

jlebar points out that there are 4 blocks being unreported here. Given that there are 4 instances of ballast also being unreported, and there should only be one per worker, that suggests there are four workers being unreported.

Comment 4

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Ah, I misread comment 0. Since this is the main process we can have ChromeWorkers using ctypes which will not get reported. We have at least one of those (RIL), probably more.

Comment 5

[Andrew McCreight [:mccr8]]

Yes, it looks like net_worker.js, ril_worker.js and wifi_worker.js all use ctypes. wifi_worker seems to get spawned twice, so there's our four unreported workers.