643760 - Suspected stack misalignment on ARM with methodjit

Status: RESOLVED WONTFIX

(Core :: JavaScript Engine, defect)

Description

[Mike Hommey [:glandium]]

On ARM EABI, stack needs to be 64 bits aligned. While this works most of the time, it can lead to weird problems with 64 bits arguments in called functions much downwards the stack.

I suspect the jaegermonkey trampoline misaligns the stack pointer.
http://mxr.mozilla.org/mozilla-central/source/js/src/methodjit/MethodJIT.cpp#475

Comment 1

[David Mandelin [:dmandelin]]

(In reply to comment #0)
> On ARM EABI, stack needs to be 64 bits aligned. While this works most of the
> time, it can lead to weird problems with 64 bits arguments in called functions
> much downwards the stack.
> 
> I suspect the jaegermonkey trampoline misaligns the stack pointer.
> http://mxr.mozilla.org/mozilla-central/source/js/src/methodjit/MethodJIT.cpp#475

It looks to me like 19 words are being pushed. So we should push one more word?

Comment 2

[Mike Hommey [:glandium]]

(In reply to comment #1)
> (In reply to comment #0)
> > On ARM EABI, stack needs to be 64 bits aligned. While this works most of the
> > time, it can lead to weird problems with 64 bits arguments in called functions
> > much downwards the stack.
> > 
> > I suspect the jaegermonkey trampoline misaligns the stack pointer.
> > http://mxr.mozilla.org/mozilla-central/source/js/src/methodjit/MethodJIT.cpp#475
> 
> It looks to me like 19 words are being pushed. So we should push one more word?

I would say so.

Comment 3

[Mike Hommey [:glandium]]

For reference, bug 545747 is the kind of problems you can run into with a non properly aligned stack. That was subtly breaking nsNavBookmarks::GetFolderIdForItem.

Comment 4

[Jan de Mooij [:jandem]]

JM was removed.