142457 2015-03-08 12:09:08 -0700 [iOS] Sweep all collected objects on critical memory pressure 2015-03-11 14:34:41 -0700 1 1 1 Unclassified WebKit JavaScriptCore 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED http://jsfiddle.net/fvyw4ba0/3/ InRadar P2 Normal --- 142595 142593 1 cdumez cdumez barraclough commit-queue ggaren kling sam oldest_to_newest 1075390 0 cdumez 2015-03-08 12:09:08 -0700 On memory pressure, we don't have much time to free-up memory. If we are not fast enough, the process can end up getting killed. The mobile version of cnn.com constructs a lot of canvases that use up a lot of memory. On memory pressure, we were calling gcController().garbageCollectSoon(). However, the garbage collection was not triggering fast enough and the WebContent process would often end up being killed. If we call gcController().garbageCollectNow() on memory pressure instead, cnn.com no longer crashes as the canvases get collected in time. Radar: <rdar://problem/20044440> 1075391 1 248193 cdumez 2015-03-08 12:11:55 -0700 Created attachment 248193 Patch 1075392 2 248193 cdumez 2015-03-08 12:16:56 -0700 Comment on attachment 248193 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=248193&action=review > Source/WebCore/platform/cocoa/MemoryPressureHandlerCocoa.mm:-84 > - gcController().garbageCollectSoon(); Maybe we could call garbageCollectSoon() on non-critical memory pressure still? This would decrease the likelihood of ending up under critical memory pressure. 1075395 3 248195 cdumez 2015-03-08 12:25:13 -0700 Created attachment 248195 Patch 1075406 4 248195 sam 2015-03-08 13:37:20 -0700 Comment on attachment 248195 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=248195&action=review > Source/WebCore/platform/cocoa/MemoryPressureHandlerCocoa.mm:88 > - if (isUnderMemoryPressure()) { > - gcController().garbageCollectSoon(); > - } else { > - // If we're not under memory pressure, that means we're here due to impending process suspension. > - // Do a full GC since this is our last chance to run any code. > + { > ReliefLogger log("Collecting JavaScript garbage"); > - gcController().garbageCollectNow(); > + if (critical) > + gcController().garbageCollectNow(); > + else > + gcController().garbageCollectSoon(); It seems like you are changing the behavior of this code getting called when not under memory pressure (but rather "due to impending process suspension"), where it used to call garbageCollectNow() and now will call seemingly call garbageCollectSoon(). Is that intentional? Does an call to this in that scenario always have the critical bit set? 1075409 5 248195 cdumez 2015-03-08 13:40:00 -0700 Comment on attachment 248195 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=248195&action=review >> Source/WebCore/platform/cocoa/MemoryPressureHandlerCocoa.mm:88 >> + gcController().garbageCollectSoon(); > > It seems like you are changing the behavior of this code getting called when not under memory pressure (but rather "due to impending process suspension"), where it used to call garbageCollectNow() and now will call seemingly call garbageCollectSoon(). Is that intentional? Does an call to this in that scenario always have the critical bit set? No, I checked and on suspension we always set the critical flag to true. From WebProcess.cpp: void WebProcess::processWillSuspend() { MemoryPressureHandler::singleton().releaseMemory(true); ... } true is for the critical argument. 1075411 6 cdumez 2015-03-08 13:41:00 -0700 And setting the "critical" flag to true upon suspension makes sense as we want to free-up as much memory as possible in this case. 1075457 7 248195 ggaren 2015-03-08 16:01:45 -0700 Comment on attachment 248195 Patch A few thoughts about this patch: (1) Did anybody figure out why the JSC::Heap::reportExtraMemoryCost API was insufficient to collect the canvases at cnn.com? (2) What is the normal GC timer period at cnn.com, and why is it insufficient to collect the canvases? (3) It is preferable to fix (1) and/or (2) rather than respond to their failure after the fact in a low memory warning handler, since the low memory warning warning handler will have the negative side effect of dumping useful caches. (4) Is there any documented limit on the number of memory warnings we'll get, or the rate at which we'll get them? In the past, I have seen cases where there was no limit, and so we hung the browser and also out-competed useful responses to memory warning that would have reclaimed memory more effectively than garbage collection. That was a terrible bug, and this patch might introduce a regression in those cases, along with a ticking time bomb that might go off any time the implementation details of memory warnings change. I'd rather not reintroduce a well-known very bad performance pathology -- especially without clear evidence showing that no better solution is available. 1075463 8 cdumez 2015-03-08 16:19:59 -0700 (In reply to comment #7) > Comment on attachment 248195 [details] > Patch > > A few thoughts about this patch: > > (1) Did anybody figure out why the JSC::Heap::reportExtraMemoryCost API was > insufficient to collect the canvases at cnn.com? > > (2) What is the normal GC timer period at cnn.com, and why is it > insufficient to collect the canvases? > > (3) It is preferable to fix (1) and/or (2) rather than respond to their > failure after the fact in a low memory warning handler, since the low memory > warning warning handler will have the negative side effect of dumping useful > caches. > > (4) Is there any documented limit on the number of memory warnings we'll > get, or the rate at which we'll get them? > > In the past, I have seen cases where there was no limit, and so we hung the > browser and also out-competed useful responses to memory warning that would > have reclaimed memory more effectively than garbage collection. That was a > terrible bug, and this patch might introduce a regression in those cases, > along with a ticking time bomb that might go off any time the implementation > details of memory warnings change. > > I'd rather not reintroduce a well-known very bad performance pathology -- > especially without clear evidence showing that no better solution is > available. Ok, thanks for the feedback. I'll try to take a look at (1) and (2). 1075481 9 cdumez 2015-03-08 17:41:12 -0700 (In reply to comment #8) > (In reply to comment #7) > > Comment on attachment 248195 [details] > > Patch > > > > A few thoughts about this patch: > > > > (1) Did anybody figure out why the JSC::Heap::reportExtraMemoryCost API was > > insufficient to collect the canvases at cnn.com? > > > > (2) What is the normal GC timer period at cnn.com, and why is it > > insufficient to collect the canvases? I added some logging when loading cnn.com, I see: Mar 8 17:34:44 Chris-iPhone com.apple.WebKit.WebContent[3342] <Notice>: CHRIS: 0x107f78c80 - Time since last gc: 2044.99 ms Mar 8 17:34:49 Chris-iPhone com.apple.WebKit.WebContent[3342] <Notice>: CHRIS: 0x107f78c80 - Time since last gc: 4855.09 ms Mar 8 17:34:50 Chris-iPhone com.apple.WebKit.WebContent[3342] <Notice>: CHRIS: 0x107f78c80 - Time since last gc: 918.94 ms Mar 8 17:34:51 Chris-iPhone com.apple.WebKit.WebContent[3342] <Notice>: CHRIS: 0x107f78c80 - Time since last gc: 1246 ms Mar 8 17:34:57 Chris-iPhone com.apple.WebKit.WebContent[3342] <Notice>: CHRIS: Memory pressure, critical: 1 Mar 8 17:35:01 Chris-iPhone MobileSafari[3339] <Notice>: CHRIS: WebProcess CRASH As you can see, no gc is happening between 17:34:51 and the critical memory pressure signal at 17:34:57. The WebContent process killed is then killed at 17:35:01 and gc() still hasn't happened. Here is the code I used to do the logging (please check I put the logging in the right place as I am not familiar with JSC): http://pastebin.com/VD06nEfv 1075486 10 cdumez 2015-03-08 17:48:53 -0700 I added logging in EdenGCCallback::doCollection() as well: Mar 8 17:47:11 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107f78c80 - FullGCActivityCallback::doCollection() Mar 8 17:47:12 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:13 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107f78c80 - FullGCActivityCallback::doCollection() Mar 8 17:47:14 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:16 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:18 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:18 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:19 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107f78c80 - FullGCActivityCallback::doCollection() Mar 8 17:47:19 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:23 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:24 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:26 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: Memory pressure, critical: 1 Mar 8 17:47:26 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:26 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:27 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: Memory pressure, critical: 1 Mar 8 17:47:27 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107fd8688 - EdenGCActivityCallback::doCollection() Mar 8 17:47:28 Chris-iPhone MobileSafari[4251] <Notice>: CHRIS: WebProcess CRASH 1075487 11 cdumez 2015-03-08 18:03:53 -0700 Here is logging with the heap size before and after the call to collect: http://pastebin.com/1tZ37uju It claims the heap size is ~60MB at the point it crashes. However, I can see there are ~380 MB of IOSurfaces at the point it crashes. 1075489 12 cdumez 2015-03-08 18:16:20 -0700 (In reply to comment #11) > Here is logging with the heap size before and after the call to collect: > http://pastebin.com/1tZ37uju > > It claims the heap size is ~60MB at the point it crashes. However, I can see > there are ~380 MB of IOSurfaces at the point it crashes. We are correctly calling reportExtraMemoryCost(2700000) in HTMLCanvasElement::createImageBuffer() for each Canvas construction. That's about 2.57MB which is very close to the actual size of the IOSurfaces (2.61MB): http://pastebin.com/x3Pj32vw 1075493 13 cdumez 2015-03-08 18:51:50 -0700 Additionally logging when HTMLCanvasElement are constructed / destroyed: http://pastebin.com/JCVDgS0v 1075495 14 cdumez 2015-03-08 19:01:42 -0700 At the point of the crash, there are 157 alive HTMLCanvasElements. 1075532 15 ggaren 2015-03-08 23:07:02 -0700 > Here is the code I used to do the logging (please check I put the logging in > the right place as I am not familiar with JSC): http://pastebin.com/VD06nEfv This patch will log full GCs triggered by the full GC timer. (And it looks like you've got logging from the eden GC timer too, not included in this patch.) One thing we'll miss here is synchronous GCs. You can catch those by logging inside Heap::collect. (Unfortunately, then you'll see double logging for every timer-based GC: one in the timer, and one in Heap::collect.) 1075535 16 ggaren 2015-03-08 23:32:31 -0700 > At the point of the crash, there are 157 alive HTMLCanvasElements. Crazy! > Mar 8 17:47:19 Chris-iPhone com.apple.WebKit.WebContent[4254] <Notice>: CHRIS: 0x107f78c80 - FullGCActivityCallback::doCollection() > ... > Mar 8 17:47:28 Chris-iPhone MobileSafari[4251] <Notice>: CHRIS: WebProcess CRASH So, we're doing eden GCs at a rate of about 1 per second, but 9 seconds have passed since the last full GC. A few thoughts about this: (1) So many canvases surviving GC seems like a bug. It would be interesting to log the number of live canvases at the point of the last full collection, and to use the tool Andreas has for GC backtracing to discover why they survived. Same with the last eden collection. (2) If there's some good reason for the canvases to survive -- something about the design of cnn.com that requires them to stay alive -- it might be a good idea to teach canvases to deallocate their IOSurfaces when they're not in the document and/or painting. This is probably not a substitute for fixing the GC issue, but it's probably desirable for those cases when a website's design simply requires keeping lots of canvases alive. (3) Can you log how long the full GC timer was scheduled for at the time of the crash? We know it was at least 9 seconds, but we don't know how long it was. It seems like there might be a bug in that timer setting. (4) Currently, extra cost only hastens the next eden collection. Perhaps it should hasten the next eden collection and the next full collection, in case the large objects happen to survive eden collection at a pathologically high rate. (5) Are these canvases actually surviving garbage collection, or have we just neglected to sweep them yet? Perhaps the low memory warning should eagerly run all outstanding destructors. The IncrementalSweeper can do this for us. 1075643 17 cdumez 2015-03-09 10:27:54 -0700 (In reply to comment #15) > > Here is the code I used to do the logging (please check I put the logging in > > the right place as I am not familiar with JSC): http://pastebin.com/VD06nEfv > > This patch will log full GCs triggered by the full GC timer. (And it looks > like you've got logging from the eden GC timer too, not included in this > patch.) All the logging I provided was WITHOUT my patch, just vanilla WebKit ToT. 1075645 18 cdumez 2015-03-09 10:32:46 -0700 (In reply to comment #16) > > At the point of the crash, there are 157 alive HTMLCanvasElements. > > Crazy! Over 500 are created so a good amount is getting destroyed by GC, just not enough or fast enough. 1075705 19 ggaren 2015-03-09 12:17:58 -0700 Actually, it looks like the logs imply that canvas objects are being identified as garbage by GC, and just not swept promptly enough. Let's do an experiment to force sweeping inside the low memory handler, and see if that helps. It's seems like a problem that (a) Eden collection does not start the incremental sweeper (b) The incremental sweeper does not hasten itself after lots of extra cost has been reported. 1075719 20 cdumez 2015-03-09 12:33:05 -0700 Crash reproduction case: http://jsfiddle.net/fvyw4ba0/ 1075728 21 248195 darin 2015-03-09 13:17:45 -0700 Comment on attachment 248195 Patch OK. 1075775 22 cdumez 2015-03-09 15:11:48 -0700 Geoff, is this the right way to trigger a full sweep on memory pressure? http://pastebin.com/0Wjz8muv If so, I'm afraid it doesn't help. The jsfiddle demo is still crashing for me. 1075786 23 ggaren 2015-03-09 15:43:06 -0700 > Geoff, is this the right way to trigger a full sweep on memory pressure? > http://pastebin.com/0Wjz8muv Yes, that's right. > If so, I'm afraid it doesn't help. The jsfiddle demo is still crashing for > me. I think this means that the real bug here is that eden collection doesn't register its finalizers with the sweeper. Side note: I think it would be worthwhile to post your eager finalizer code as a stand-alone patch. It is a good thing to do -- just not for this test case. 1075797 24 cdumez 2015-03-09 16:03:44 -0700 (In reply to comment #22) > Geoff, is this the right way to trigger a full sweep on memory pressure? > http://pastebin.com/0Wjz8muv > > If so, I'm afraid it doesn't help. The jsfiddle demo is still crashing for > me. I was an async repro case: http://jsfiddle.net/fvyw4ba0/3/ Full sweep on memory pressure also doesn't seem to help in this case. A gcController().garbageCollectNow() on memory pressure does prevent this reproduction case from crashing though. 1075822 25 248295 cdumez 2015-03-09 17:24:11 -0700 Created attachment 248295 Patch 1075823 26 cdumez 2015-03-09 17:24:59 -0700 This is not enough to fix the crash on cnn.com but this is a step in the right direction. 1075891 27 248295 ggaren 2015-03-09 21:49:20 -0700 Comment on attachment 248295 Patch r=me 1075896 28 248295 commit-queue 2015-03-09 22:33:18 -0700 Comment on attachment 248295 Patch Clearing flags on attachment: 248295 Committed r181310: <http://trac.webkit.org/changeset/181310> 1075897 29 commit-queue 2015-03-09 22:33:22 -0700 All reviewed patches have been landed. Closing bug. 248193 2015-03-08 12:11:55 -0700 2015-03-08 12:25:08 -0700 Patch bug-142457-20150308121136.patch text/plain 2369 cdumez U3VidmVyc2lvbiBSZXZpc2lvbjogMTgxMjQ1CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggMWNiMzI5Yzg1YzRmNTZj M2YxOWZmODIyZTc3ZTM1ODg5ODM5OGFkZi4uYWIwNzA0ZjVkNGNlZDc0YjZhZDhlZmRhNmU2ODJk NTJjYjg1YTk3ZSAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDI2IEBACisyMDE1LTAzLTA4ICBDaHJp cyBEdW1leiAgPGNkdW1lekBhcHBsZS5jb20+CisKKyAgICAgICAgW2lPU10gSHVycnkgSlMgZ2Fy YmFnZSBjb2xsZWN0aW9uIG9uIG1lbW9yeSBwcmVzc3VyZQorICAgICAgICBodHRwczovL2J1Z3Mu d2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MTQyNDU3CisgICAgICAgIDxyZGFyOi8vcHJvYmxl bS8yMDA0NDQ0MD4KKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAg ICAgICBPbiBtZW1vcnkgcHJlc3N1cmUsIHdlIGRvbid0IGhhdmUgbXVjaCB0aW1lIHRvIGZyZWUt dXAgbWVtb3J5LiBJZiB3ZSBhcmUgbm90IGZhc3QKKyAgICAgICAgZW5vdWdoLCB0aGUgcHJvY2Vz cyBjYW4gZW5kIHVwIGdldHRpbmcga2lsbGVkLgorCisgICAgICAgIFRoZSBtb2JpbGUgdmVyc2lv biBvZiBjbm4uY29tIGNvbnN0cnVjdHMgYSBsb3Qgb2YgY2FudmFzZXMgdGhhdCB1c2UgdXAgYSBs b3Qgb2YKKyAgICAgICAgbWVtb3J5LiBPbiBtZW1vcnkgcHJlc3N1cmUsIHdlIHdlcmUgY2FsbGlu ZyBnY0NvbnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdFNvb24oKS4KKyAgICAgICAgSG93ZXZlciwg dGhlIGdhcmJhZ2UgY29sbGVjdGlvbiB3YXMgbm90IHRyaWdnZXJpbmcgZmFzdCBlbm91Z2ggYW5k IHRoZSBXZWJDb250ZW50CisgICAgICAgIHByb2Nlc3Mgd291bGQgb2Z0ZW4gZW5kIHVwIGJlaW5n IGtpbGxlZC4KKworICAgICAgICBUaGlzIHBhdGNoIHVwZGF0ZXMgdGhlIG1lbW9yeSBwcmVzc3Vy ZSBoYW5kbGVyIGluc3RlYWQgdG8gY2FsbAorICAgICAgICBnY0NvbnRyb2xsZXIoKS5nYXJiYWdl Q29sbGVjdE5vdygpIGluc3RlYWQgb24gbWVtb3J5IHByZXNzdXJlLiBBcyBhIHJlc3VsdCwgY25u LmNvbQorICAgICAgICBubyBsb25nZXIgY3Jhc2hlcyBhcyB0aGUgY2FudmFzZXMgZ2V0IGNvbGxl Y3RlZCBpbiB0aW1lLgorCisgICAgICAgICogcGxhdGZvcm0vY29jb2EvTWVtb3J5UHJlc3N1cmVI YW5kbGVyQ29jb2EubW06CisgICAgICAgIChXZWJDb3JlOjpNZW1vcnlQcmVzc3VyZUhhbmRsZXI6 OnBsYXRmb3JtUmVsZWFzZU1lbW9yeSk6CisKIDIwMTUtMDMtMDggIFNpbW9uIEZyYXNlciAgPHNp bW9uLmZyYXNlckBhcHBsZS5jb20+CiAKICAgICAgICAgSW4gUmVuZGVyTGF5ZXJDb21wb3NpdG9y LCB0cmFjayBsYXllciBib3VuZHMgYW5kIHRoZSBoYXZlQ29tcHV0ZWRCb3VuZHMgZmxhZyB0b2dl dGhlciBpbiBhIHN0cnVjdApkaWZmIC0tZ2l0IGEvU291cmNlL1dlYkNvcmUvcGxhdGZvcm0vY29j b2EvTWVtb3J5UHJlc3N1cmVIYW5kbGVyQ29jb2EubW0gYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9y bS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRsZXJDb2NvYS5tbQppbmRleCAxNzg1OGM0NGUzNjYy YTFlZDIyYTU4ODc2MGM0N2JiMDY0ZTgzZDFlLi4xMDg3ZDY0OGMxZjBlMmRkZmM1NmMzM2MzOGM4 Y2Y4ODg3NWJlZmQ5IDEwMDY0NAotLS0gYS9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9N ZW1vcnlQcmVzc3VyZUhhbmRsZXJDb2NvYS5tbQorKysgYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9y bS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRsZXJDb2NvYS5tbQpAQCAtODAsMTEgKzgwLDcgQEAg dm9pZCBNZW1vcnlQcmVzc3VyZUhhbmRsZXI6OnBsYXRmb3JtUmVsZWFzZU1lbW9yeShib29sIGNy aXRpY2FsKQogI2VuZGlmCiAKICNpZiBQTEFURk9STShJT1MpCi0gICAgaWYgKGlzVW5kZXJNZW1v cnlQcmVzc3VyZSgpKSB7Ci0gICAgICAgIGdjQ29udHJvbGxlcigpLmdhcmJhZ2VDb2xsZWN0U29v bigpOwotICAgIH0gZWxzZSB7Ci0gICAgICAgIC8vIElmIHdlJ3JlIG5vdCB1bmRlciBtZW1vcnkg cHJlc3N1cmUsIHRoYXQgbWVhbnMgd2UncmUgaGVyZSBkdWUgdG8gaW1wZW5kaW5nIHByb2Nlc3Mg c3VzcGVuc2lvbi4KLSAgICAgICAgLy8gRG8gYSBmdWxsIEdDIHNpbmNlIHRoaXMgaXMgb3VyIGxh c3QgY2hhbmNlIHRvIHJ1biBhbnkgY29kZS4KKyAgICB7CiAgICAgICAgIFJlbGllZkxvZ2dlciBs b2coIkNvbGxlY3RpbmcgSmF2YVNjcmlwdCBnYXJiYWdlIik7CiAgICAgICAgIGdjQ29udHJvbGxl cigpLmdhcmJhZ2VDb2xsZWN0Tm93KCk7CiAgICAgfQo= 248195 2015-03-08 12:25:13 -0700 2015-03-09 17:24:07 -0700 Patch bug-142457-20150308122453.patch text/plain 2544 cdumez U3VidmVyc2lvbiBSZXZpc2lvbjogMTgxMjQ1CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggMWNiMzI5Yzg1YzRmNTZj M2YxOWZmODIyZTc3ZTM1ODg5ODM5OGFkZi4uOTUyNzNmMzMyNGZiYWU1ZTFiNTIwYzY0NDI1MmQx YzIyNDM4MmViNSAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDI2IEBACisyMDE1LTAzLTA4ICBDaHJp cyBEdW1leiAgPGNkdW1lekBhcHBsZS5jb20+CisKKyAgICAgICAgW2lPU10gSHVycnkgSlMgZ2Fy YmFnZSBjb2xsZWN0aW9uIG9uIGNyaXRpY2FsIG1lbW9yeSBwcmVzc3VyZQorICAgICAgICBodHRw czovL2J1Z3Mud2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MTQyNDU3CisgICAgICAgIDxyZGFy Oi8vcHJvYmxlbS8yMDA0NDQ0MD4KKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMh KS4KKworICAgICAgICBPbiBjcml0aWNhbCBtZW1vcnkgcHJlc3N1cmUsIHdlIGRvbid0IGhhdmUg bXVjaCB0aW1lIHRvIGZyZWUtdXAgbWVtb3J5LiBJZiB3ZSBhcmUKKyAgICAgICAgbm90IGZhc3Qg ZW5vdWdoLCB0aGUgcHJvY2VzcyBjYW4gZW5kIHVwIGdldHRpbmcga2lsbGVkLgorCisgICAgICAg IFRoZSBtb2JpbGUgdmVyc2lvbiBvZiBjbm4uY29tIGNvbnN0cnVjdHMgYSBsb3Qgb2YgY2FudmFz ZXMgdGhhdCB1c2UgdXAgYSBsb3Qgb2YKKyAgICAgICAgbWVtb3J5LiBPbiBtZW1vcnkgcHJlc3N1 cmUsIHdlIHdlcmUgY2FsbGluZyBnY0NvbnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdFNvb24oKS4K KyAgICAgICAgSG93ZXZlciwgdGhlIGdhcmJhZ2UgY29sbGVjdGlvbiB3YXMgbm90IHRyaWdnZXJp bmcgZmFzdCBlbm91Z2ggYW5kIHRoZSBXZWJDb250ZW50CisgICAgICAgIHByb2Nlc3Mgd291bGQg b2Z0ZW4gZW5kIHVwIGJlaW5nIGtpbGxlZC4KKworICAgICAgICBUaGlzIHBhdGNoIHVwZGF0ZXMg dGhlIG1lbW9yeSBwcmVzc3VyZSBoYW5kbGVyIGluc3RlYWQgdG8gY2FsbAorICAgICAgICBnY0Nv bnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdE5vdygpIGluc3RlYWQgb24gY3JpdGljYWwgbWVtb3J5 IHByZXNzdXJlLiBBcyBhIHJlc3VsdCwKKyAgICAgICAgY25uLmNvbSBubyBsb25nZXIgY3Jhc2hl cyBhcyB0aGUgY2FudmFzZXMgZ2V0IGNvbGxlY3RlZCBpbiB0aW1lLgorCisgICAgICAgICogcGxh dGZvcm0vY29jb2EvTWVtb3J5UHJlc3N1cmVIYW5kbGVyQ29jb2EubW06CisgICAgICAgIChXZWJD b3JlOjpNZW1vcnlQcmVzc3VyZUhhbmRsZXI6OnBsYXRmb3JtUmVsZWFzZU1lbW9yeSk6CisKIDIw MTUtMDMtMDggIFNpbW9uIEZyYXNlciAgPHNpbW9uLmZyYXNlckBhcHBsZS5jb20+CiAKICAgICAg ICAgSW4gUmVuZGVyTGF5ZXJDb21wb3NpdG9yLCB0cmFjayBsYXllciBib3VuZHMgYW5kIHRoZSBo YXZlQ29tcHV0ZWRCb3VuZHMgZmxhZyB0b2dldGhlciBpbiBhIHN0cnVjdApkaWZmIC0tZ2l0IGEv U291cmNlL1dlYkNvcmUvcGxhdGZvcm0vY29jb2EvTWVtb3J5UHJlc3N1cmVIYW5kbGVyQ29jb2Eu bW0gYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRsZXJD b2NvYS5tbQppbmRleCAxNzg1OGM0NGUzNjYyYTFlZDIyYTU4ODc2MGM0N2JiMDY0ZTgzZDFlLi42 OWFkYjkwMzU1M2EyN2ExYjAyYzdjZjcwMzgxNmM3ZjFiOGZhODM4IDEwMDY0NAotLS0gYS9Tb3Vy Y2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRsZXJDb2NvYS5tbQor KysgYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRsZXJD b2NvYS5tbQpAQCAtODAsMTMgKzgwLDEyIEBAIHZvaWQgTWVtb3J5UHJlc3N1cmVIYW5kbGVyOjpw bGF0Zm9ybVJlbGVhc2VNZW1vcnkoYm9vbCBjcml0aWNhbCkKICNlbmRpZgogCiAjaWYgUExBVEZP Uk0oSU9TKQotICAgIGlmIChpc1VuZGVyTWVtb3J5UHJlc3N1cmUoKSkgewotICAgICAgICBnY0Nv bnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdFNvb24oKTsKLSAgICB9IGVsc2UgewotICAgICAgICAv LyBJZiB3ZSdyZSBub3QgdW5kZXIgbWVtb3J5IHByZXNzdXJlLCB0aGF0IG1lYW5zIHdlJ3JlIGhl cmUgZHVlIHRvIGltcGVuZGluZyBwcm9jZXNzIHN1c3BlbnNpb24uCi0gICAgICAgIC8vIERvIGEg ZnVsbCBHQyBzaW5jZSB0aGlzIGlzIG91ciBsYXN0IGNoYW5jZSB0byBydW4gYW55IGNvZGUuCisg ICAgewogICAgICAgICBSZWxpZWZMb2dnZXIgbG9nKCJDb2xsZWN0aW5nIEphdmFTY3JpcHQgZ2Fy YmFnZSIpOwotICAgICAgICBnY0NvbnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdE5vdygpOworICAg ICAgICBpZiAoY3JpdGljYWwpCisgICAgICAgICAgICBnY0NvbnRyb2xsZXIoKS5nYXJiYWdlQ29s bGVjdE5vdygpOworICAgICAgICBlbHNlCisgICAgICAgICAgICBnY0NvbnRyb2xsZXIoKS5nYXJi YWdlQ29sbGVjdFNvb24oKTsKICAgICB9CiAjZW5kaWYKIH0K 248295 2015-03-09 17:24:11 -0700 2015-03-09 22:33:18 -0700 Patch bug-142457-20150309172348.patch text/plain 4476 cdumez U3VidmVyc2lvbiBSZXZpc2lvbjogMTgxMjcwCmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvQ2hhbmdlTG9nIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwppbmRleCA0 ZjY0MDhlMGJkMTc2OWMyMzk3MmZiMDVlYzFiNTg2NzJjMTcwMDk1Li42MDE3ZWQxN2QwZWE3M2Iy YTQ0ZjYzY2VkMzYxOGM5N2QxMmYzNTQ3IDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNjcmlwdENv cmUvQ2hhbmdlTG9nCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKQEAgLTEs MyArMSwxOSBAQAorMjAxNS0wMy0wOSAgQ2hyaXMgRHVtZXogIDxjZHVtZXpAYXBwbGUuY29tPgor CisgICAgICAgIFtpT1NdIFN3ZWVwIGFsbCBjb2xsZWN0ZWQgb2JqZWN0cyBvbiBjcml0aWNhbCBt ZW1vcnkgcHJlc3N1cmUKKyAgICAgICAgaHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcu Y2dpP2lkPTE0MjQ1NworICAgICAgICA8cmRhcjovL3Byb2JsZW0vMjAwNDQ0NDA+CisKKyAgICAg ICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgQWxsIGZ1bGxTd2VlcCgp IEFQSSB0byBJbmNyZW1lbnRhbFN3ZWVwZXIgc28gdGhhdCB3ZSBjYW4gY2FsbCBpdCBpbiB0aGUK KyAgICAgICAgbWVtb3J5IHByZXNzdXJlIGhhbmRsZXIuCisKKyAgICAgICAgKiBoZWFwL0luY3Jl bWVudGFsU3dlZXBlci5jcHA6CisgICAgICAgIChKU0M6OkluY3JlbWVudGFsU3dlZXBlcjo6ZnVs bFN3ZWVwKToKKyAgICAgICAgKiBoZWFwL0luY3JlbWVudGFsU3dlZXBlci5oOgorICAgICAgICAo SlNDOjpJbmNyZW1lbnRhbFN3ZWVwZXI6Omhhc1dvcmspOgorCiAyMDE1LTAzLTA4ICBBbmRyZWFz IEtsaW5nICA8YWtsaW5nQGFwcGxlLmNvbT4KIAogICAgICAgICBKSVRUaHVua3Mga2VlcHMgZmlu YWxpemVkIFdlYWtzIGFyb3VuZCwgcGlubmluZyBXZWFrQmxvY2tzLgpkaWZmIC0tZ2l0IGEvU291 cmNlL1dlYkNvcmUvQ2hhbmdlTG9nIGIvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCmluZGV4IDJl NjUyYWYzZmU1Nzk3ZTRmNjllZmYzYzVmMjRiNjMyMjlhYjllZTkuLjEwNmI5ZDdhOGI3Y2NmNzk1 ZTUxYzhjZTJiZGVjZDQ3MzRlNWZiZDMgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJDb3JlL0NoYW5n ZUxvZworKysgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKQEAgLTEsMyArMSwxNyBAQAorMjAx NS0wMy0wOSAgQ2hyaXMgRHVtZXogIDxjZHVtZXpAYXBwbGUuY29tPgorCisgICAgICAgIFtpT1Nd IFN3ZWVwIGFsbCBjb2xsZWN0ZWQgb2JqZWN0cyBvbiBjcml0aWNhbCBtZW1vcnkgcHJlc3N1cmUK KyAgICAgICAgaHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTE0MjQ1Nwor ICAgICAgICA8cmRhcjovL3Byb2JsZW0vMjAwNDQ0NDA+CisKKyAgICAgICAgUmV2aWV3ZWQgYnkg Tk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgRG8gYSBmdWxsIHN3ZWVwIG9mIG9iamVjdHMgbWFy a2VkIGZvciBkZXN0cnVjdGlvbiBvbiBjcml0aWNhbCBtZW1vcnkKKyAgICAgICAgcHJlc3N1cmUg dG8gZnJlZSB1cCBtZW1vcnkuCisKKyAgICAgICAgKiBwbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVz c3VyZUhhbmRsZXJDb2NvYS5tbToKKyAgICAgICAgKFdlYkNvcmU6Ok1lbW9yeVByZXNzdXJlSGFu ZGxlcjo6cGxhdGZvcm1SZWxlYXNlTWVtb3J5KToKKwogMjAxNS0wMy0wOSAgUm9nZXIgRm9uZyAg PHJvZ2VyX2ZvbmdAYXBwbGUuY29tPgogCiAgICAgICAgIFVucmV2aWV3ZWQuIFJlLWFkZCBDU1Mg cHJvcGVydHkgdGhhdCB3YXMgdW5pbnRlbnRpb25hbGx5IHJlbW92ZWQgaW4gcjE4MDg5MwpkaWZm IC0tZ2l0IGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2hlYXAvSW5jcmVtZW50YWxTd2VlcGVyLmNw cCBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9oZWFwL0luY3JlbWVudGFsU3dlZXBlci5jcHAKaW5k ZXggZGExNTEzN2RkNWI3NzlmZTc5Y2U0ZDIyOTVmYTRlZjM5OWU3MDJkMC4uNDgwOWY2MmEzY2Q2 MWU1ZjNhMWNjNzU4MDBjYjM1ZmI1NGQzOWNjNCAxMDA2NDQKLS0tIGEvU291cmNlL0phdmFTY3Jp cHRDb3JlL2hlYXAvSW5jcmVtZW50YWxTd2VlcGVyLmNwcAorKysgYi9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvaGVhcC9JbmNyZW1lbnRhbFN3ZWVwZXIuY3BwCkBAIC02MCw2ICs2MCwxMiBAQCB2b2lk IEluY3JlbWVudGFsU3dlZXBlcjo6Y2FuY2VsVGltZXIoKQogICAgIENGUnVuTG9vcFRpbWVyU2V0 TmV4dEZpcmVEYXRlKG1fdGltZXIuZ2V0KCksIENGQWJzb2x1dGVUaW1lR2V0Q3VycmVudCgpICsg c19kZWNhZGUpOwogfQogCit2b2lkIEluY3JlbWVudGFsU3dlZXBlcjo6ZnVsbFN3ZWVwKCkKK3sK KyAgICB3aGlsZSAoaGFzV29yaygpKQorICAgICAgICBkb1dvcmsoKTsKK30KKwogdm9pZCBJbmNy ZW1lbnRhbFN3ZWVwZXI6OmRvV29yaygpCiB7CiAgICAgZG9Td2VlcChXVEY6Om1vbm90b25pY2Fs bHlJbmNyZWFzaW5nVGltZSgpKTsKZGlmZiAtLWdpdCBhL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9o ZWFwL0luY3JlbWVudGFsU3dlZXBlci5oIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL2hlYXAvSW5j cmVtZW50YWxTd2VlcGVyLmgKaW5kZXggYWY4MjZjNGIzYWNjY2JmMGEzMjNhMmQzNGQ0OTZiMDQ5 YWVhZjVmNi4uMTk5Zjk0ODBmZGEwYjY5YzFmNjRkZDEwMjNlNjFmM2YyNjNkODJkNiAxMDA2NDQK LS0tIGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2hlYXAvSW5jcmVtZW50YWxTd2VlcGVyLmgKKysr IGIvU291cmNlL0phdmFTY3JpcHRDb3JlL2hlYXAvSW5jcmVtZW50YWxTd2VlcGVyLmgKQEAgLTM4 LDYgKzM4LDcgQEAgY2xhc3MgSW5jcmVtZW50YWxTd2VlcGVyIDogcHVibGljIEhlYXBUaW1lciB7 CiBwdWJsaWM6CiAjaWYgVVNFKENGKQogICAgIEpTX0VYUE9SVF9QUklWQVRFIEluY3JlbWVudGFs U3dlZXBlcihIZWFwKiwgQ0ZSdW5Mb29wUmVmKTsKKyAgICBKU19FWFBPUlRfUFJJVkFURSB2b2lk IGZ1bGxTd2VlcCgpOwogI2Vsc2UKICAgICBleHBsaWNpdCBJbmNyZW1lbnRhbFN3ZWVwZXIoVk0q KTsKICNlbmRpZgpAQCAtNTIsNiArNTMsNyBAQCBwcml2YXRlOgogICAgIHZvaWQgZG9Td2VlcChk b3VibGUgc3RhcnRUaW1lKTsKICAgICB2b2lkIHNjaGVkdWxlVGltZXIoKTsKICAgICB2b2lkIGNh bmNlbFRpbWVyKCk7CisgICAgYm9vbCBoYXNXb3JrKCkgY29uc3QgeyByZXR1cm4gIW1fYmxvY2tz VG9Td2VlcC5pc0VtcHR5KCk7IH0KICAgICAKICAgICB1bnNpZ25lZCBtX2N1cnJlbnRCbG9ja1Rv U3dlZXBJbmRleDsKICAgICBWZWN0b3I8TWFya2VkQmxvY2sqPiYgbV9ibG9ja3NUb1N3ZWVwOwpk aWZmIC0tZ2l0IGEvU291cmNlL1dlYkNvcmUvcGxhdGZvcm0vY29jb2EvTWVtb3J5UHJlc3N1cmVI YW5kbGVyQ29jb2EubW0gYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVz c3VyZUhhbmRsZXJDb2NvYS5tbQppbmRleCAxNzg1OGM0NGUzNjYyYTFlZDIyYTU4ODc2MGM0N2Ji MDY0ZTgzZDFlLi4xZWUzYjhmMDQyYzcxNDJiYmZmNTBiNjhkNjUyM2FjMjU5ZTA1Mjk5IDEwMDY0 NAotLS0gYS9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVzc3VyZUhhbmRs ZXJDb2NvYS5tbQorKysgYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9jb2NvYS9NZW1vcnlQcmVz c3VyZUhhbmRsZXJDb2NvYS5tbQpAQCAtMjksMTAgKzI5LDEyIEBACiAjaW1wb3J0ICJEaXNwYXRj aFNQSS5oIgogI2ltcG9ydCAiSU9TdXJmYWNlUG9vbC5oIgogI2ltcG9ydCAiR0NDb250cm9sbGVy LmgiCisjaW1wb3J0ICJKU0RPTVdpbmRvdy5oIgogI2ltcG9ydCAiSlNET01XaW5kb3dCYXNlLmgi CiAjaW1wb3J0ICJMYXllclBvb2wuaCIKICNpbXBvcnQgIkxvZ2dpbmcuaCIKICNpbXBvcnQgIldl YkNvcmVTeXN0ZW1JbnRlcmZhY2UuaCIKKyNpbXBvcnQgPEphdmFTY3JpcHRDb3JlL0luY3JlbWVu dGFsU3dlZXBlci5oPgogI2ltcG9ydCA8bWFjaC9tYWNoLmg+CiAjaW1wb3J0IDxtYWNoL3Rhc2tf aW5mby5oPgogI2ltcG9ydCA8bWFsbG9jL21hbGxvYy5oPgpAQCAtODgsNiArOTAsMTMgQEAgdm9p ZCBNZW1vcnlQcmVzc3VyZUhhbmRsZXI6OnBsYXRmb3JtUmVsZWFzZU1lbW9yeShib29sIGNyaXRp Y2FsKQogICAgICAgICBSZWxpZWZMb2dnZXIgbG9nKCJDb2xsZWN0aW5nIEphdmFTY3JpcHQgZ2Fy YmFnZSIpOwogICAgICAgICBnY0NvbnRyb2xsZXIoKS5nYXJiYWdlQ29sbGVjdE5vdygpOwogICAg IH0KKworICAgIC8vIERvIGEgZnVsbCBzd2VlcCBvZiBjb2xsZWN0ZWQgb2JqZWN0cy4KKyAgICB7 CisgICAgICAgIFJlbGllZkxvZ2dlciBsb2coIkZ1bGwgSmF2YVNjcmlwdCBnYXJiYWdlIHN3ZWVw Iik7CisgICAgICAgIEpTQzo6SlNMb2NrSG9sZGVyIGxvY2soSlNET01XaW5kb3c6OmNvbW1vblZN KCkpOworICAgICAgICBKU0RPTVdpbmRvdzo6Y29tbW9uVk0oKS5oZWFwLnN3ZWVwZXIoKS0+ZnVs bFN3ZWVwKCk7CisgICAgfQogI2VuZGlmCiB9CiAK