Closed Bug 771320 Opened 8 years ago Closed 8 years ago

Fix endian issues with handles

Categories

(Core :: JavaScript Engine, defect)

15 Branch
PowerPC
macOS
defect
Not set

Tracking

()

RESOLVED FIXED
mozilla16

People

(Reporter: spectre, Assigned: spectre)

References

(Blocks 1 open bug, )

Details

(Whiteboard: [js:t])

Attachments

(1 file, 1 obsolete file)

TenFourFox: Bigger Endian is Better(tm)

Bug 750733 introduced handles widely as part of the improved GC, but the current implementation makes assumptions about the word order, causing crashes with things like argument objects on big-endian platforms. See also https://code.google.com/p/tenfourfox/issues/detail?id=165#c39 (note 39 and following). This small patch should work on both little- and big-endian, and passes all tests with the PowerPC backend.
Attachment #639454 - Attachment is patch: true
Attachment #639454 - Flags: review?(dmandelin)
Comment on attachment 639454 [details] [diff] [review]
Use payloadOf/PAYLOAD_OFFSET when constructing PIC stubs

I think Brian's a better reviewer for this one.
Attachment #639454 - Flags: review?(dmandelin) → review?(bhackett1024)
Attachment #639454 - Flags: review?(bhackett1024) → review+
Thanks, David and Brian!
Assignee: general → spectre
Status: NEW → ASSIGNED
Keywords: checkin-needed
QA Contact: spectre
QA Contact: spectre
https://hg.mozilla.org/integration/mozilla-inbound/rev/2dd36c1cd662

Bonus points next time if your patches include the needed metadata. It makes life easier for checkin monkeys like me :-)
https://developer.mozilla.org/en/Creating_a_patch_that_can_be_checked_in
Flags: in-testsuite-
Keywords: checkin-needed
Target Milestone: --- → mozilla16
http://mozillamemes.tumblr.com/post/19498220636/try-server-takes-the-beatings-so-mozilla-inbound

Compiling is also a nice feature for a patch to have. Backed out due to red.
https://hg.mozilla.org/integration/mozilla-inbound/rev/08fb8d6997bc

https://tbpl.mozilla.org/php/getParsedLog.php?id=13334873&tree=Mozilla-Inbound

PolyIC.cpp
/usr/bin/ccache /tools/gcc-4.5-0moz3/bin/g++ -o PolyIC.o -c  -I./../../dist/system_wrappers_js -include /builds/slave/m-in-lnx64/build/js/src/config/gcc_hidden.h -DMOZ_GLUE_IN_PROGRAM -DEXPORT_JS_API -DJS_HAS_CTYPES -DDLL_PREFIX=\"lib\" -DDLL_SUFFIX=\".so\" -DNO_NSPR_10_SUPPORT -Ictypes/libffi/include -I.  -I/builds/slave/m-in-lnx64/build/js/src/../../mfbt/double-conversion -I/builds/slave/m-in-lnx64/build/js/src -I. -I./../../dist/include  -I/builds/slave/m-in-lnx64/build/obj-firefox/dist/include/nspr      -I/builds/slave/m-in-lnx64/build/js/src -I/builds/slave/m-in-lnx64/build/js/src/assembler -I/builds/slave/m-in-lnx64/build/js/src/yarr  -fPIC  -pedantic -Wall -Wpointer-arith -Woverloaded-virtual -Werror=return-type -Wtype-limits -Wempty-body -Wno-ctor-dtor-privacy -Wno-overlength-strings -Wno-invalid-offsetof -Wno-variadic-macros -Wcast-align -Wno-long-long -fno-rtti -ffunction-sections -fdata-sections -fno-exceptions -pthread -pipe  -DNDEBUG -DTRIMMED -g -O3 -freorder-blocks -finline-limit=50 -fno-omit-frame-pointer  -DUSE_SYSTEM_MALLOC=1 -DENABLE_ASSEMBLER=1 -DENABLE_JIT=1   -DMOZILLA_CLIENT -include ./js-confdefs.h -MD -MF .deps/PolyIC.o.pp /builds/slave/m-in-lnx64/build/js/src/methodjit/PolyIC.cpp
../../../js/src/methodjit/PunboxAssembler.h: In member function 'void js::mjit::GetPropCompiler::generateGetterStub(js::mjit::Assembler&, js::Shape*, jsid, js::mjit::MacroAssemblerTypedefs::Label, js::Vector<JSC::AbstractMacroAssembler<JSC::X86Assembler>::Jump, 8ul>&)':
../../../js/src/methodjit/PunboxAssembler.h:52:44: error: 'const uint32_t js::mjit::PunboxAssembler::PAYLOAD_OFFSET' is private
../../../js/src/methodjit/PolyIC.cpp:1102:80: error: within this context
Target Milestone: mozilla16 → ---
FWIW, all the build failures were 64bit.
error: 'const uint32_t js::mjit::PunboxAssembler::PAYLOAD_OFFSET' is private? Should I make that public also?

Sorry about the metadata; I'll just generate checkin-ready versions separately.
Actually, I take that back -- it seems easier just to make this 32-bit only. Would that be the right approach, Brian?
(In reply to Cameron Kaiser from comment #7)
> Actually, I take that back -- it seems easier just to make this 32-bit only.
> Would that be the right approach, Brian?

I think that making PAYLOAD_OFFSET public in PunboxAssembler.h should be fine (and seems simple?), it'd be nice to avoid any #ifdef JS_BLAH in the IC methods.
Okay, patched PunboxAssembler.h also.

Unfortunately, I do not have access to push to try.
Attachment #639454 - Attachment is obsolete: true
Attachment #640116 - Flags: review?(bhackett1024)
Comment on attachment 640116 [details] [diff] [review]
Use payloadOf/PAYLOAD_OFFSET when constructing PIC stubs (v2)

Pushed to try @ https://tbpl.mozilla.org/?tree=Try&rev=7520fdb6c037
Attachment #640116 - Flags: review?(bhackett1024) → review+
Whiteboard: [js:t]
On bhackett's try, I see a single red mochitest for Android and I don't know if that has anything to do with this, plus some oranges under Linux 64bit. OS X64 and Win64 are green. I don't think this is the cause of the fail/oranges.
https://hg.mozilla.org/mozilla-central/rev/f3145f3e7f44
Status: ASSIGNED → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla16
You need to log in before you can comment on or make changes to this bug.