149621 2015-09-29 03:28:33 -0700 REGRESSION(r190289): It made Speedometer/Full.html performance test fail 2018-01-17 19:11:32 -0800 1 1 1 Unclassified WebKit JavaScriptCore Other Unspecified Unspecified RESOLVED FIXED https://bugs.webkit.org/show_bug.cgi?id=149720 InRadar P2 Normal --- 148664 1 ossy msaboff barraclough basile_clement cdumez fpizlo keith_miller msaboff ossy rniwa oldest_to_newest 1129201 0 ossy 2015-09-29 03:28:33 -0700 https://trac.webkit.org/changeset/190289 made Speedometer/Full.html performance test fail Running Speedometer/Full.html (151 of 151) ERROR: CONSOLE MESSAGE: line 208: Miss the info bar? Run TodoMVC from a server to avoid a cross-origin error. FAILED - EFL perf bot: https://build.webkit.org/builders/EFL%20Linux%2064-bit%20Release%20WK2%20%28Perf%29/builds/6952 - Apple perf bot: https://build.webkit.org/builders/Apple%20Yosemite%20Release%20WK2%20%28Perf%29/builds/3026 The latest known good revision is r190286. ( https://build.webkit.org/builders/Apple%20Yosemite%20Release%20WK2%20%28Perf%29/builds/3025 ) The first failing revision is r190291. ( https://build.webkit.org/builders/EFL%20Linux%2064-bit%20Release%20WK2%20%28Perf%29/builds/6952 ) I tried to bisect this small interval locally, but didn't manage, because this test crashes almost always due to a serious GC related regression - bug149341. ( Unfortunately it seems nobody is interested in fixing crashes/regressions ... ) 1129231 1 ossy 2015-09-29 05:42:34 -0700 r190288 is the last good revision - https://build.webkit.org/builders/Apple%20Yosemite%20Release%20WK2%20%28Perf%29/builds/3031 1129337 2 msaboff 2015-09-29 12:27:26 -0700 I am able to reproduce this in both release and debug builds although the symptoms are a little different. In a debug build, we are OSR exiting from a function due to BadType, but the real reason is that the tag registers (r14 & r15 on X86-64) do not have the tag constants. it appears that they were not properly restored after a call (likely to an FTL function). More debugging. 1129416 3 ap 2015-09-29 14:42:10 -0700 Should this be duped to bug 149647 now (the rollout bug)? 1129418 4 msaboff 2015-09-29 14:44:49 -0700 (In reply to comment #3) > Should this be duped to bug 149647 now (the rollout bug)? The rollout bug is closed. I'll still work on the issue with this defect. I'll add this defect and https://bugs.webkit.org/show_bug.cgi?id=149619 to bug 149647. 1129499 5 msaboff 2015-09-29 22:38:38 -0700 A little more detail. We are calling a function that makes a varargs tail call. When the FTL makes a tail call and it needs to take the slow path, it doesn't properly restore the callee save registers. Moved the restoration of callee saves to common path before deciding to take the fast or slow path. 1129672 6 262192 msaboff 2015-09-30 14:40:05 -0700 Created attachment 262192 Patch 1129689 7 msaboff 2015-09-30 15:30:19 -0700 Landed as part of change set r190370: <http://trac.webkit.org/changeset/190370>. 1129719 8 keith_miller 2015-09-30 16:28:08 -0700 Committed r190372: <http://trac.webkit.org/changeset/190372> 1129722 9 msaboff 2015-09-30 16:35:20 -0700 (In reply to comment #8) > Committed r190372: <http://trac.webkit.org/changeset/190372> This was really a change for an unrelated build fix. 1129887 10 ossy 2015-10-01 02:50:15 -0700 (In reply to comment #7) > Landed as part of change set r190370: > <http://trac.webkit.org/changeset/190370>. Reopen, because r190370 didn't fix the orignal issue, Speedometer/Full.html still fails everywhere. Additionally it can't be an FTL issue, because FTL isn't enabled on Linux platforms at all. - https://build.webkit.org/builders/Apple%20Mavericks%20Release%20WK2%20%28Perf%29/builds/5900 - https://build.webkit.org/builders/EFL%20Linux%2064-bit%20Release%20WK2%20%28Perf%29/builds/6975 To make sure if r190370 is the culprit, I started force builds on perf bots: - r190368 - https://build.webkit.org/builders/Apple%20El%20Capitan%20Release%20WK2%20%28Perf%29/builds/12 - r190370 - https://build.webkit.org/builders/Apple%20Mavericks%20Release%20WK2%20%28Perf%29/builds/5904 (r190369 is an unrelated change) 1130210 11 msaboff 2015-10-02 05:07:25 -0700 (In reply to comment #10) > (In reply to comment #7) > > Landed as part of change set r190370: > > <http://trac.webkit.org/changeset/190370>. > > Reopen, because r190370 didn't fix the orignal issue, Speedometer/Full.html > still fails everywhere. Additionally it can't be an FTL issue, because > FTL isn't enabled on Linux platforms at all. > > - > https://build.webkit.org/builders/ > Apple%20Mavericks%20Release%20WK2%20%28Perf%29/builds/5900 > - > https://build.webkit.org/builders/EFL%20Linux%2064- > bit%20Release%20WK2%20%28Perf%29/builds/6975 > > To make sure if r190370 is the culprit, I started force builds on perf bots: > - r190368 - > https://build.webkit.org/builders/ > Apple%20El%20Capitan%20Release%20WK2%20%28Perf%29/builds/12 > - r190370 - > https://build.webkit.org/builders/ > Apple%20Mavericks%20Release%20WK2%20%28Perf%29/builds/5904 > > (r190369 is an unrelated change) The patch that was landed solves the problem encountered on Mac builds. Is it possible for you to run Speedometer/Full.html from a debug build in a debugger and post stack trace? 1130211 12 ossy 2015-10-02 05:14:36 -0700 I didn't notice crashes (in release mode), but I get this console message locally (EFL Linux x86_64) and you can see it on EFL and Apple Mac performance bots too: ERROR: CONSOLE MESSAGE: line 208: Miss the info bar? Run TodoMVC from a server to avoid a cross-origin error. After grepping a little bit, base.js:140 in SpeedoMeter is responsible for printing this message to console. 1130225 13 cdumez 2015-10-02 08:07:01 -0700 (In reply to comment #12) > I didn't notice crashes (in release mode), but I get this console > message locally (EFL Linux x86_64) and you can see it on EFL and > Apple Mac performance bots too: > > ERROR: CONSOLE MESSAGE: line 208: Miss the info bar? Run TodoMVC from a > server to avoid a cross-origin error. > > After grepping a little bit, base.js:140 in SpeedoMeter > is responsible for printing this message to console. Looks like: https://bugs.webkit.org/show_bug.cgi?id=149720 1130924 14 ossy 2015-10-06 02:46:39 -0700 *** Bug 149720 has been marked as a duplicate of this bug. *** 1131075 15 rniwa 2015-10-06 12:17:04 -0700 *** Bug 149720 has been marked as a duplicate of this bug. *** 1131079 16 rniwa 2015-10-06 12:20:11 -0700 <rdar://problem/22995488> In order to reproduce this failure, revert http://trac.webkit.org/changeset/190535 1131082 17 rniwa 2015-10-06 12:37:04 -0700 The bug is that console.info called in line 140 of base.js (there are a few copies of it inside PerformanceTests/Speedometer/) is reported erroneously at line 208. Given: 138: function getFile(file, callback) { 139: if (!location.host) { 140: return console.info('Miss the info bar? Run TodoMVC from a server to avoid a cross-origin error.'); 141: } 207: redirect(); 208: getFile('learn.json', Learn); 209: })(); I'd imagine that the console.info in "getFile" is tail-call optimized into the caller, and erroneously reporting the line number. 262192 2015-09-30 14:40:05 -0700 2015-09-30 14:47:33 -0700 Patch 149621.patch text/plain 1699 msaboff SW5kZXg6IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkocmV2aXNpb24gMTkwMzY3KQorKysgU291cmNl L0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkod29ya2luZyBjb3B5KQpAQCAtMSwzICsxLDE2IEBA CisyMDE1LTA5LTMwICBNaWNoYWVsIFNhYm9mZiAgPG1zYWJvZmZAYXBwbGUuY29tPgorCisgICAg ICAgIFJFR1JFU1NJT04ocjE5MDI4OSk6IEl0IG1hZGUgU3BlZWRvbWV0ZXIvRnVsbC5odG1sIHBl cmZvcm1hbmNlIHRlc3QgZmFpbAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9zaG93 X2J1Zy5jZ2k/aWQ9MTQ5NjIxCisKKyAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChPT1BTISku CisKKyAgICAgICAgV2UgbmVlZCB0byByZXN0b3JlIGNhbGxlZSBzYXZlcyBmb3IgYm90aCB0aGUg ZmFzdCBhbmQgc2xvdyBwYXRocyBiZWZvcmUgbWFraW5nIGEKKyAgICAgICAgdGFpbCBjYWxsIGlu IHRoZSBGVEwuCisKKyAgICAgICAgKiBmdGwvRlRMSlNDYWxsQmFzZS5jcHA6CisgICAgICAgIChK U0M6OkZUTDo6SlNDYWxsQmFzZTo6ZW1pdCk6CisKIDIwMTUtMDktMzAgIEtlaXRoIE1pbGxlciAg PGtlaXRoX21pbGxlckBhcHBsZS5jb20+CiAKICAgICAgICAgW0VTNl0gQWRkIFR5cGVkQXJyYXku cHJvdG90eXBlIGZ1bmN0aW9uYWxpdHkuCkluZGV4OiBTb3VyY2UvSmF2YVNjcmlwdENvcmUvZnRs L0ZUTEpTQ2FsbEJhc2UuY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9m dGwvRlRMSlNDYWxsQmFzZS5jcHAJKHJldmlzaW9uIDE5MDI4OSkKKysrIFNvdXJjZS9KYXZhU2Ny aXB0Q29yZS9mdGwvRlRMSlNDYWxsQmFzZS5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTUyLDYgKzUy LDkgQEAgdm9pZCBKU0NhbGxCYXNlOjplbWl0KENDYWxsSGVscGVycyYgaml0KQogewogICAgIG1f Y2FsbExpbmtJbmZvID0gaml0LmNvZGVCbG9jaygpLT5hZGRDYWxsTGlua0luZm8oKTsKICAgICAK KyAgICBpZiAoQ2FsbExpbmtJbmZvOjpjYWxsTW9kZUZvcihtX3R5cGUpID09IENhbGxNb2RlOjpU YWlsKQorICAgICAgICBqaXQuZW1pdFJlc3RvcmVDYWxsZWVTYXZlcygpOworCiAgICAgQ0NhbGxI ZWxwZXJzOjpKdW1wIHNsb3dQYXRoID0gaml0LmJyYW5jaFB0cldpdGhQYXRjaCgKICAgICAgICAg Q0NhbGxIZWxwZXJzOjpOb3RFcXVhbCwgR1BSSW5mbzo6cmVnVDAsIG1fdGFyZ2V0VG9DaGVjaywK ICAgICAgICAgQ0NhbGxIZWxwZXJzOjpUcnVzdGVkSW1tUHRyKDApKTsKQEAgLTU5LDcgKzYyLDYg QEAgdm9pZCBKU0NhbGxCYXNlOjplbWl0KENDYWxsSGVscGVycyYgaml0KQogICAgIENDYWxsSGVs cGVyczo6SnVtcCBkb25lOwogCiAgICAgaWYgKENhbGxMaW5rSW5mbzo6Y2FsbE1vZGVGb3IobV90 eXBlKSA9PSBDYWxsTW9kZTo6VGFpbCkgewotICAgICAgICBqaXQuZW1pdFJlc3RvcmVDYWxsZWVT YXZlcygpOwogICAgICAgICBqaXQucHJlcGFyZUZvclRhaWxDYWxsU2xvdygpOwogICAgICAgICBt X2Zhc3RDYWxsID0gaml0Lm5lYXJUYWlsQ2FsbCgpOwogICAgIH0gZWxzZSB7Cg==