Bug 155790 – [ARM] There are 400-600 more crashes on ARMv7 Thumb2 with GLIB timers

• Component: JavaScriptCore
• Status: RESOLVED DUPLICATE

[Csaba Osztrogonác]

On ARMv7 Thumb2 Linux there are
-  26 failures with EFL port     - (no available link, but I saved the log previously, see it below)
- 138 failures with JSCOnly port - https://build.webkit.org/builders/JSCOnly%20Linux%20ARMv7%20Thumb2%20Release/builds/13
- 110 failures with GTK port     - https://build.webkit.org/builders/GTK%20Linux%20ARM%20Release/builds/10621

( these results are on the same revision - r198487 )

failures on EFL port:
----------------------
** The following JSC stress test failures have been introduced:
	cdjs-tests.yaml/main.js.eager-no-cjit
	cdjs-tests.yaml/main.js.no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout
	jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-dfg-eager-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-no-llint
	jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout
	jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-dfg-eager-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-no-llint
	jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout
	jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-dfg-eager-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-no-cjit
	jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-no-llint
	jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout
	jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-dfg-eager-no-cjit
	jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-no-cjit
	jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-no-llint
	mozilla-tests.yaml/js1_5/Array/regress-157652.js.mozilla
	mozilla-tests.yaml/js1_5/Array/regress-157652.js.mozilla-baseline
	mozilla-tests.yaml/js1_5/Array/regress-157652.js.mozilla-dfg-eager-no-cjit-validate-phases
	stress/arrowfunction-lexical-bind-superproperty.js.dfg-eager
	stress/arrowfunction-lexical-bind-superproperty.js.dfg-eager-no-cjit-validate
	stress/arrowfunction-lexical-bind-superproperty.js.no-llint
	stress/op_mod.js.no-cjit-validate-phases
	stress/op_mul.js.no-cjit-validate-phases

Results for JSC stress tests:
    26 failures found.

[Csaba Osztrogonác]

note: There are no similar issue on ARMv7 traditional and AArch64.
(at least with JSCOnly port)

[Csaba Osztrogonác]

There are only 25 failures on EFL Thumb2 build, but ~600 on JSCOnly Thumb2
and ~400 on GTK Thumb2:
- https://build.webkit.org/builders/JSCOnly%20Linux%20ARMv7%20Thumb2%20Release/builds/167
- https://build.webkit.org/builders/GTK%20Linux%20ARM%20Release/builds/10782
- EFL results:

** The following JSC stress test failures have been introduced:
        jsc-layout-tests.yaml/fast/regex/script-tests/pcre-test-1.js.layout-dfg-eager-no-cjit
        jsc-layout-tests.yaml/fast/regex/script-tests/pcre-test-1.js.layout-no-llint
        jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout
        jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-dfg-eager-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-collator.js.layout-no-llint
        jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout
        jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-dfg-eager-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-datetimeformat.js.layout-no-llint
        jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout
        jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-dfg-eager-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-no-cjit
        jsc-layout-tests.yaml/js/script-tests/intl-numberformat.js.layout-no-llint
        jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout
        jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-dfg-eager-no-cjit
        jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-no-cjit
        jsc-layout-tests.yaml/js/script-tests/number-toLocaleString.js.layout-no-llint
        mozilla-tests.yaml/js1_5/Array/regress-157652.js.mozilla-baseline
        mozilla-tests.yaml/js1_5/Array/regress-157652.js.mozilla-llint
        stress/arrowfunction-lexical-bind-superproperty.js.dfg-eager
        stress/arrowfunction-lexical-bind-superproperty.js.dfg-eager-no-cjit-validate
        stress/arrowfunction-lexical-bind-superproperty.js.no-llint
        stress/sparse-array-entry-update-144067.js.always-trigger-copy-phase
        stress/v8-earley-boyer-strict.js.no-llint

Results for JSC stress tests:
    25 failures found.

[Csaba Osztrogonác]

It seems there is a serious bug near the GLIB timer or in the GC.

[Csaba Osztrogonác]

*** Bug 156160 has been marked as a duplicate of this bug. ***

[Csaba Osztrogonác]

*** Bug 155784 has been marked as a duplicate of this bug. ***

[Csaba Osztrogonác]

*** Bug 156163 has been marked as a duplicate of this bug. ***

[Yusuke Suzuki]

So many test failures report that there are "Illegal instruction".
I think there is some bug in online assembler. But I cannot investigate it b/c I didn't have any ARM Thumb2 env.......

[Zan Dobersek]

Bug #150083 has a WIP patch that addresses two places where tail call linking wasn't properly working on Thumb2 platforms on Linux. Bug #154857 probably has more in common with that problem, but I was not experiencing a larger set of JSC test failures with GLib timers when testing with r198847 in release or debug builds, apart from a few failures due to internationalization and timezone settings in the testing environment.

[Csaba Osztrogonác]

*** This bug has been marked as a duplicate of bug 150083 ***