168502 2017-02-17 02:29:56 -0800 REGRESSION(r212778): It made 400 tests crash on AArch64 Linux 2017-03-06 15:07:36 -0800 1 1 1 Unclassified WebKit JavaScriptCore Other Unspecified Unspecified RESOLVED FIXED P1 Critical --- 108645 167737 1 ossy ossy commit-queue fpizlo keith_miller mark.lam msaboff ossy saam oldest_to_newest 1278183 0 ossy 2017-02-17 02:29:56 -0800 before: https://build.webkit.org/builders/JSCOnly%20Linux%20AArch64%20Release/builds/4617 (r212462) after: https://build.webkit.org/builders/JSCOnly%20Linux%20AArch64%20Release/builds/4618 (r212482) I bisected manually, and https://trac.webkit.org/changeset/212466 is the culprit. r212465 - https://build.webkit.org/builders/JSCOnly%20Linux%20AArch64%20Release/builds/4627 r212466 - https://build.webkit.org/builders/JSCOnly%20Linux%20AArch64%20Release/builds/4626 1278704 1 ossy 2017-02-19 00:03:56 -0800 Haven't you noticed this regression on internal iOS JSC tester bots too? 1278726 2 fpizlo 2017-02-19 07:37:15 -0800 Passed tests just fine on my iPad. Idea: what happens when you make Linux take the same path as Windows in RegisterStae.h? Basically make it fall through to the setjmp code. 1278766 3 fpizlo 2017-02-19 14:39:57 -0800 Did you get a stack trace? 1279822 4 ossy 2017-02-22 01:23:02 -0800 r212466 was rolled out and relanded in https://trac.webkit.org/changeset/212778 Now we have only 400 crashes on AArch64 Linux. (In reply to comment #2) > Passed tests just fine on my iPad. > > Idea: what happens when you make Linux take the same path as Windows in > RegisterStae.h? Basically make it fall through to the setjmp code. Will check soon. (In reply to comment #3) > Did you get a stack trace? Not yet, I'm going to do a debug build in the near future. 1279941 5 ossy 2017-02-22 10:02:12 -0800 (In reply to comment #2) > Idea: what happens when you make Linux take the same path as Windows in > RegisterStae.h? Basically make it fall through to the setjmp code. This idea fixed this issue. What does it mean? (In reply to comment #3) > Did you get a stack trace? Unfortunately there are no crashes in debug mode. 1282289 6 ossy 2017-03-01 09:07:36 -0800 (In reply to comment #5) > (In reply to comment #2) > > Idea: what happens when you make Linux take the same path as Windows in > > RegisterStae.h? Basically make it fall through to the setjmp code. > > This idea fixed this issue. What does it mean? > > > (In reply to comment #3) > > Did you get a stack trace? > > Unfortunately there are no crashes in debug mode. AArch64 Linux platform is still broken due to this GC bug. Any plan to fix it or should we use the setjmp codepath? 1282308 7 fpizlo 2017-03-01 09:53:41 -0800 (In reply to comment #5) > (In reply to comment #2) > > Idea: what happens when you make Linux take the same path as Windows in > > RegisterStae.h? Basically make it fall through to the setjmp code. > > This idea fixed this issue. What does it mean? It means that this is the fix to this issue. From what I remember setjmp is very fast on Linux so this should be fine. > > > (In reply to comment #3) > > Did you get a stack trace? > > Unfortunately there are no crashes in debug mode. 1283923 8 303550 ossy 2017-03-06 13:48:51 -0800 Created attachment 303550 Patch 1283925 9 303550 fpizlo 2017-03-06 13:56:27 -0800 Comment on attachment 303550 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=303550&action=review > Source/JavaScriptCore/heap/RegisterState.h:32 > -#if !OS(WINDOWS) > +#if !OS(WINDOWS) && !(OS(LINUX) && CPU(ARM64)) This is good. FWIW, it's probably also a bit safer to use setjmp on all Linuxes. I've written this kind of stuff on Linux before and always found that setjmp was the canonical Linux way of getting register state. The only reason why we now do it differently is that Darwin's setjmp does some extra state saving that makes it just a tad bit too slow. I don't think Linux's setjmp has that kind of problem on any architecture. It would even be fine to change this to OS(DARWIN) rather than !OS(WINDOWS) && !OS(LINUX). If you want to make any of those changes then lets do it in another patch. 1283977 10 commit-queue 2017-03-06 14:48:30 -0800 The commit-queue encountered the following flaky tests while processing attachment 303550: media/modern-media-controls/tracks-support/tracks-support-show-panel-after-dragging-controls.html bug 169158 (authors: graouts@apple.com and ryanhaddad@apple.com) The commit-queue is continuing to process your patch. 1283980 11 commit-queue 2017-03-06 14:48:42 -0800 The commit-queue encountered the following flaky tests while processing attachment 303550: media/modern-media-controls/tracks-support/tracks-support-click-track-in-panel.html bug 169159 (authors: graouts@apple.com and ryanhaddad@apple.com) The commit-queue is continuing to process your patch. 1283991 12 303550 commit-queue 2017-03-06 15:07:30 -0800 Comment on attachment 303550 Patch Clearing flags on attachment: 303550 Committed r213472: <http://trac.webkit.org/changeset/213472> 1283992 13 commit-queue 2017-03-06 15:07:36 -0800 All reviewed patches have been landed. Closing bug. 303550 2017-03-06 13:48:51 -0800 2017-03-06 15:07:30 -0800 Patch bug-168502-20170306224849.patch text/plain 1334 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMjEzNDY0CmRpZmYgLS1naXQgYS9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvQ2hhbmdlTG9nIGIvU291cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwppbmRleCBh YjZiMWYzYTg4ZWMzMzEwYTY4ZjRmNWQwNWNlOWM3MzhmNzNkZmQ3Li5kOWQ1NWFhOWE1ZWJjN2Mw YTRhMDg0N2I0MTRmMTM4MThkODEyZWMwIDEwMDY0NAotLS0gYS9Tb3VyY2UvSmF2YVNjcmlwdENv cmUvQ2hhbmdlTG9nCisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKQEAgLTEs MyArMSwxMiBAQAorMjAxNy0wMy0wNiAgQ3NhYmEgT3N6dHJvZ29uw6FjICA8b3NzeUB3ZWJraXQu b3JnPgorCisgICAgICAgIFJFR1JFU1NJT04ocjIxMjc3OCk6IEl0IG1hZGUgNDAwIHRlc3RzIGNy YXNoIG9uIEFBcmNoNjQgTGludXgKKyAgICAgICAgaHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hv d19idWcuY2dpP2lkPTE2ODUwMgorCisgICAgICAgIFJldmlld2VkIGJ5IE5PQk9EWSAoT09QUyEp LgorCisgICAgICAgICogaGVhcC9SZWdpc3RlclN0YXRlLmg6IFVzZSBzZXRqbXAgY29kZSBwYXRo IG9uIEFBcmNoNjQgTGludXggdG9vIHRvIGZpeCBjcmFzaGVzLgorCiAyMDE3LTAzLTA2ICBZdXN1 a2UgU3V6dWtpICA8dXRhdGFuZS50ZWFAZ21haWwuY29tPgogCiAgICAgICAgIFtKU0NdIEFsbG93 IGluZGV4ZWQgbW9kdWxlIG5hbWVzcGFjZSBvYmplY3QgZmllbGRzCmRpZmYgLS1naXQgYS9Tb3Vy Y2UvSmF2YVNjcmlwdENvcmUvaGVhcC9SZWdpc3RlclN0YXRlLmggYi9Tb3VyY2UvSmF2YVNjcmlw dENvcmUvaGVhcC9SZWdpc3RlclN0YXRlLmgKaW5kZXggNjAwNWE5MTQzNTcxYjhjOTZjYmI0OWJm N2QyNmM5Mjg2ZmMzNTU4MC4uZDY4MjkwMzJjNTlkMDEwOTdkN2FhZGYwN2NhZjg0NWM1M2JjNDNl YiAxMDA2NDQKLS0tIGEvU291cmNlL0phdmFTY3JpcHRDb3JlL2hlYXAvUmVnaXN0ZXJTdGF0ZS5o CisrKyBiL1NvdXJjZS9KYXZhU2NyaXB0Q29yZS9oZWFwL1JlZ2lzdGVyU3RhdGUuaApAQCAtMjks NyArMjksNyBAQAogCiBuYW1lc3BhY2UgSlNDIHsKIAotI2lmICFPUyhXSU5ET1dTKQorI2lmICFP UyhXSU5ET1dTKSAmJiAhKE9TKExJTlVYKSAmJiBDUFUoQVJNNjQpKQogCiAvLyBBTExPQ0FURV9B TkRfR0VUX1JFR0lTVEVSX1NUQVRFIGhhcyB0byBlbnN1cmUgdGhhdCB0aGUgR0Mgc2VlcyBjYWxs ZWUtc2F2ZXMuIEl0IGFjaGlldmVzIHRoaXMgYnkKIC8vIGVuc3VyaW5nIHRoYXQgdGhlIGNhbGxl ZS1zYXZlcyBhcmUgZWl0aGVyIHNwaWxsZWQgdG8gdGhlIHN0YWNrIG9yIHNhdmVkIGluIHRoZSBS ZWdpc3RlclN0YXRlLiBUaGUgY29kZQo=