Bug 106471

Summary: Release FastMalloc thread caches on memory warning
Product: WebKit Reporter: Antti Koivisto <koivisto>
Component: WebCore Misc.Assignee: Nobody <webkit-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: barraclough, benjamin, kling, levin+threading, ojan.autocc, philn, psolanki, webkit-bug-importer, webkit-ews, webkit.review.bot, xan.lopez
Priority: P2 Keywords: InRadar
Version: 528+ (Nightly build)   
Hardware: Unspecified   
OS: Unspecified   
Attachments:
Description Flags
patch
webkit-ews: commit-queue-
patch2 ggaren: review+, buildbot: commit-queue-

Description Antti Koivisto 2013-01-09 10:31:01 PST 
FastMalloc keeps some memory in per-thread caches (currently 2MB each). We currently flush these caches on memory warning for the main thread only. We should do it for other WebKit threads that use FastMalloc too.
Comment 1 Radar WebKit Bug Importer 2013-01-09 10:32:13 PST 
<rdar://problem/12982020>
Comment 2 Antti Koivisto 2013-01-09 10:41:25 PST 
Created attachment 181944 [details]
patch
Comment 3 Early Warning System Bot 2013-01-09 10:46:44 PST 
Comment on attachment 181944 [details]
patch

Attachment 181944 [details] did not pass qt-ews (qt):
Output: http://queues.webkit.org/results/15761561
Comment 4 Early Warning System Bot 2013-01-09 10:47:03 PST 
Comment on attachment 181944 [details]
patch

Attachment 181944 [details] did not pass qt-wk2-ews (qt):
Output: http://queues.webkit.org/results/15775422
Comment 5 Antti Koivisto 2013-01-09 10:57:52 PST 
Created attachment 181950 [details]
patch2
Comment 6 Geoffrey Garen 2013-01-09 11:07:44 PST 
Comment on attachment 181950 [details]
patch2

I think you also want Cleanup() to call signalScavenger(). Otherwise, the memory will be in free lists in the central cache, but it won't necessarily get unmapped.
Comment 7 Build Bot 2013-01-09 11:34:27 PST 
Comment on attachment 181950 [details]
patch2

Attachment 181950 [details] did not pass win-ews (win):
Output: http://queues.webkit.org/results/15760567
Comment 8 Antti Koivisto 2013-01-09 12:10:11 PST 
http://trac.webkit.org/changeset/139218
Comment 9 Antti Koivisto 2013-01-09 12:12:20 PST 
(In reply to comment #6)
> (From update of attachment 181950 [details])
> I think you also want Cleanup() to call signalScavenger(). Otherwise, the memory will be in free lists in the central cache, but it won't necessarily get unmapped.

That shouldn't be necessary. On memory warning we forcibly free the memory in central cache afterwards. Under normal circumstances just returning the spans should do sufficient amount of tickling.