[Bug 22316] New: Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Summary: Wine should be built with -mincoming-stack-boundary=2
           Product: Wine
           Version: 1.1.42
          Platform: x86
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: build-env
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

GCC 4.4 on Linux assumes the Stack is aligned to 16 bytes and thus assumes it
can use the movdqa instruction when manipulating 64 bit values on the stack.
For x86-32 Windows Microsoft specifies a stack alignment of 4 Bytes.

In my case I had wine compiled with -O3 -march=native and got an "Unhandled
page fault on read access to 0xffffffff" with the disassembly showing "movq
%mm5,%mm3" as the faulting instruction. Running objdump revealed the
instruction was in fact "movdqa %xmm3,-0x88(%ebp)" while ebp % 16 had been 4
since the app entered wine code.

Therefore I propose to compile wine with -mincoming-stack-boundary=2 where it
is available.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #1 from Alexandre Julliard   2010-04-09 05:18:22 ---
That's a gcc bug, see http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40838. The
Linux de-facto ABI is 4-byte alignment.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID

--- Comment #2 from Dmitry Timoshkov   2010-04-09 05:49:06 ---
Invalid then.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #3 from Dmitry Timoshkov   2010-04-09 05:49:19 ---
Closing invalid.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #4 from Dmitry Timoshkov   2010-04-17 04:53:21 ---
*** Bug 22385 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #5 from Alexandre Julliard   2010-04-17 08:52:36 ---
*** Bug 22387 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #6 from Dmitry Timoshkov   2010-04-26 00:51:43 ---
*** Bug 22475 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dan Kegel  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #7 from Dan Kegel  2010-11-07 07:02:01 CST ---
Daniel, does gcc-4.5.1 behave better?

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #8 from Artem S. Tashkinov  2010-11-07 08:01:26 CST ---
Let me have the insolence to say on Daniel's behalf that GCC 4.5.1 is no
different to GCC 4.4 in the regard.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40838 has zero activity, it's not
even on GCC developers radar (bug status: unassigned). GCC 4.6.0 will most
likely have the same problem.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #9 from Dan Kegel  2010-11-07 08:05:45 CST ---
Does the option -mstackrealign help?  From that bug report, and from
Artem's response on bug 22387, I gather that might be the thing to
do when using aggressive optimizations on gcc-4.5.x.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Mike Frysinger  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #10 from Mike Frysinger  2011-07-12 01:09:23 CDT ---
i'm not sure punting this to gcc works anymore.  the Linux x86 ABI requires 16
bytes, so any random Windows programs that unalign below that will cause
problems in code that assumes 16 byte alignment.

ive got at least 2 reports of -mstackrealign fixing things for people.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #11 from Alexandre Julliard  2011-07-12 03:31:46 CDT ---
(In reply to comment #10)
> i'm not sure punting this to gcc works anymore.  the Linux x86 ABI requires 16
> bytes, so any random Windows programs that unalign below that will cause
> problems in code that assumes 16 byte alignment.

The Linux ABI has never required 16 bytes, except maybe on paper. The fact is
that there are many apps out there, Wine among them, that don't align to 16
bytes. Even if Wine starts forcing alignment today, there's nothing we can do
about all the existing binaries. I don't understand how it can be considered
acceptable to break backwards compatibility that way.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #12 from Mike Frysinger  2011-07-12 11:03:34 CDT ---
so even though the standard says "stack must be 16byte aligned", and gcc
assumes the stack is 16byte aligned and will generate code that requires that
(i.e. sse), you still don't care if wine breaks (as reported by various
distros/users) ?

existing apps dont matter ... their stack isnt passed in.  it's created fresh
upon execution, so the change in ABI is irrelevant to them.  if they work with
4byte alignment, and they only use 4byte alignment, then starting off with
16byte alignment changes nothing.

what matters here are newly compiled programs as that is where the code
generation is coming from.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #13 from Dan Kegel  2011-07-12 11:11:11 CDT ---
The standard says 4 bytes.  gcc deviates intentionally from the 
standard because current ia32 works better with 8 or 16 byte alignment.

I filed http://bugs.linuxbase.org/show_bug.cgi?id=3268 about the problem,
let's see what they say.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #14 from Mike Frysinger  2011-07-12 14:01:10 CDT ---
the gcc PR mentioned in comment #1 says the Linux ABI is (being updated to
say?) 16 bytes.  the man page also indicates that 4 byte is legacy and 16 byte
is modern.  and further, the man page says that it realigns the stack only when
needed.  so i dont understand why this would be unacceptable to the wine devs
as it doesnt affect code output if no SSE code is generated that needs 16byte
alignment.

-mstackrealign
    Realign the stack at entry.  On the Intel x86, the -mstackrealign option
    will generate an alternate prologue and epilogue that realigns the runtime
    stack if necessary.  This supports mixing legacy codes that keep a 4-byte
    aligned stack with modern codes that keep a 16-byte stack for SSE
    compatibility.  See also the attribute "force_align_arg_pointer",
    applicable to individual functions.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #15 from Alexandre Julliard  2011-07-12 14:57:30 CDT ---
(In reply to comment #14)
> the gcc PR mentioned in comment #1 says the Linux ABI is (being updated to
> say?) 16 bytes.  the man page also indicates that 4 byte is legacy and 16 byte
> is modern.  and further, the man page says that it realigns the stack only when
> needed.  so i dont understand why this would be unacceptable to the wine devs
> as it doesnt affect code output if no SSE code is generated that needs 16byte
> alignment.
>
> -mstackrealign
>     Realign the stack at entry.  On the Intel x86, the -mstackrealign option
>     will generate an alternate prologue and epilogue that realigns the runtime
>     stack if necessary.  This supports mixing legacy codes that keep a 4-byte
>     aligned stack with modern codes that keep a 16-byte stack for SSE
>     compatibility.  See also the attribute "force_align_arg_pointer",
>     applicable to individual functions.

This always aligns the stack. It's what we use on the Mac, where the ABI truly
is 16 bytes. It has a significant performance cost since it has to be done on
every entry point.

It can certainly be used by people who want to build Wine with aggressive
optimization, though the cost of the constant realignments would probably wipe
out any benefits from the extra optimizations. It's not something we want to do
by default if it can be avoided.

The real problem however is gcc causing new alignment requirements to sneak
into system libraries. That's what breaks compatibility, and the only solution
is for gcc to default to backwards-compatible behavior, and require an explicit
option for the new ABI. That's what that gcc bug is about.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Andrew Eikum  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #16 from Andrew Eikum  2012-02-06 10:56:17 CST ---
*** Bug 29431 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Sylvain Petreolle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

[e-mail]  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |stevenvandenbrandenstift@gm
                   |                            |ail.com

--- Comment #17 from Austin English  2012-07-31 17:45:26 CDT ---
*** Bug 31375 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #18 from Austin English  2012-08-01 17:35:47 CDT ---
*** Bug 31379 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Dan Kegel  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #19 from Dan Kegel  2012-08-11 09:13:03 CDT ---
*** Bug 31444 has been marked as a duplicate of this bug. ***

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Jerome Leclanche  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |http://gcc.gnu.org/bugzilla
                   |                            |/show_bug.cgi?id=40838
         Resolution|INVALID                     |UPSTREAM

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #20 from Dan Kegel  2013-06-15 10:09:50 CDT ---
https://lsbbugs.linuxfoundation.org/show_bug.cgi?id=3268 is now marked
P1 blocker, for what it's worth, but I bet it will be resolved
by defining 16 byte alignment as the standard, since the LSB is mostly about
codifying existing practice.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

http://bugs.winehq.org/show_bug.cgi?id=22316

Anssi Hannula  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

https://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #21 from Jerome Leclanche  ---
Was this fixed upstream? If it's not fixed upstream this bug shouldn't be
CLOSED yet.

[Bug 22316] Wine should be built with -mincoming-stack-boundary=2

https://bugs.winehq.org/show_bug.cgi?id=22316

--- Comment #22 from Anssi Hannula  ---
I don't think this was fixed upstream, but this workaround was applied to wine:
http://source.winehq.org/git/wine.git/commitdiff/4adfb787f4e8c36a37ce1d53a7e...