[Bug other/106355] New: Linux s390x -O2 argument passing miscompile

[Bug other/106355] New: Linux s390x -O2 argument passing miscompile

[sbergman at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

            Bug ID: 106355
           Summary: Linux s390x -O2 argument passing miscompile
           Product: gcc
           Version: 12.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
          Assignee: unassigned at gcc dot gnu.org
          Reporter: sbergman at redhat dot com
  Target Milestone: ---

(I ran into this when trying to build recent LibreOffice 7.5 master on Fedora
36 on s390x causing a crash in one of the executables generated during the
build.  I tracked that down to a member function S::f2(arg1,args2,arg3) calling
S::f1(arg1,arg2,arg3,std::nullopt) but within
S::f1(arg1,arg2,arg3,std::optional<unsigned short> x) then x.has_value()
evaluates to true, or even to some random non-zero value like 163:)

At least with gcc-c++-12.1.1-1.fc36.s390x:

> $ cat test.cc
> #include <cstdlib>
> #include <optional>
> int f1(int, int, int, int, std::optional<unsigned short> x) {
>   if (x) std::abort();
>   return 0;
> }
> int f2(int a, int b, int c, int d) {
>   return f1(a, b, c, d, std::nullopt);
> }

> $ g++ -std=c++17 -O2 -fpic -S test.cc

> $ cat test.s
>         .file   "test.cc"
>         .machinemode zarch
>         .machine "zEC12"
> .text
>         .align  8
>         .align  16
> .globl _Z2f1iiiiSt8optionalItE
>         .type   _Z2f1iiiiSt8optionalItE, @function
> _Z2f1iiiiSt8optionalItE:
> .LFB284:
>         .cfi_startproc
>         tmll    %r6,65280
>         jne     .L7
>         lghi    %r2,0
>         br      %r14
> .L7:
>         stmg    %r14,%r15,112(%r15)
>         .cfi_offset 14, -48
>         .cfi_offset 15, -40
>         lay     %r15,-168(%r15)
>         .cfi_def_cfa_offset 328
>         brasl   %r14,abort@PLT
>         .cfi_endproc
> .LFE284:
>         .size   _Z2f1iiiiSt8optionalItE, .-_Z2f1iiiiSt8optionalItE
>         .align  8
>         .align  16
> .globl _Z2f2iiii
>         .type   _Z2f2iiii, @function
> _Z2f2iiii:
> .LFB287:
>         .cfi_startproc
>         ldgr    %f0,%r15
>         .cfi_register 15, 16
>         lay     %r15,-168(%r15)
>         .cfi_def_cfa_offset 328
>         mvi     166(%r15),0
>         lgdr    %r15,%f0
>         .cfi_restore 15
>         .cfi_def_cfa_offset 160
>         jg      _Z2f1iiiiSt8optionalItE@PLT
>         .cfi_endproc
> .LFE287:
>         .size   _Z2f2iiii, .-_Z2f2iiii
>         .ident  "GCC: (GNU) 12.1.1 20220507 (Red Hat 12.1.1-1)"
>         .section        .note.GNU-stack,"",@progbits

With my limited understanding of s390x assembly, f2 sets the optional's bool
_M_engaged flag byte to 0 on the stack with

        lay     %r15,-168(%r15)
        mvi     166(%r15),0

but then f1 expects to test that byte in %r6 with

        tmll    %r6,65280

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |krebbel at gcc dot gnu.org,
                   |                            |rguenth at gcc dot gnu.org

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Did you see whether this is changed behavior from GCC 11?

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[sbergman at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #2 from Stephan Bergmann <sbergman at redhat dot com> ---
(In reply to Richard Biener from comment #1)
> Did you see whether this is changed behavior from GCC 11?

I didn't check that myself, but the Debian LibreOffice maintainer claims that
he sees the same symptoms when building with some GCC 11 (against some GCC 12
libstdc++, though).  (It appears that this issue only happened to start to hit
builds of LibreOffice after some recent code changes in LibreOffice, so we
don't have any historic data on which versions of GCC would not have had this
issue.)

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[dan at danny dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

Dan Horák <dan at danny dot cz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Dan Horák <dan at danny dot cz> ---
adding Jakub as this affect Fedora ...

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[stefansf at linux dot ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #4 from Stefan Schulze Frielinghaus <stefansf at linux dot ibm.com> ---
The problem is with sibling call optimization where parameters with BLKmode are
not handled correctly. I will prepare a patch and submit it shortly.

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Stefan Schulze Frielinghaus
<stefansf@gcc.gnu.org>:

https://gcc.gnu.org/g:cb994acc08b67f26a54e7c5dc1f4995a2ce24d98

commit r13-3377-gcb994acc08b67f26a54e7c5dc1f4995a2ce24d98
Author: Stefan Schulze Frielinghaus <stefansf@linux.ibm.com>
Date:   Wed Oct 19 14:28:22 2022 +0200

    IBM zSystems: Fix function_ok_for_sibcall [PR106355]

    For a parameter with BLKmode we cannot use REG_NREGS in order to
    determine the number of consecutive registers.  Streamlined this with
    the implementation of s390_function_arg.

    Fix some indentation whitespace, too.

    gcc/ChangeLog:

            PR target/106355
            * config/s390/s390.cc (s390_call_saved_register_used): For a
            parameter with BLKmode fix determining number of consecutive
            registers.

    gcc/testsuite/ChangeLog:

            * gcc.target/s390/pr106355.h: Common code for new tests.
            * gcc.target/s390/pr106355-1.c: New test.
            * gcc.target/s390/pr106355-2.c: New test.
            * gcc.target/s390/pr106355-3.c: New test.

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Stefan Schulze Frielinghaus
<stefansf@gcc.gnu.org>:

https://gcc.gnu.org/g:0737125ec0e9102e139f8f49130b4666be4cc626

commit r12-8865-g0737125ec0e9102e139f8f49130b4666be4cc626
Author: Stefan Schulze Frielinghaus <stefansf@linux.ibm.com>
Date:   Wed Oct 19 14:28:22 2022 +0200

    IBM zSystems: Fix function_ok_for_sibcall [PR106355]

    For a parameter with BLKmode we cannot use REG_NREGS in order to
    determine the number of consecutive registers.  Streamlined this with
    the implementation of s390_function_arg.

    Fix some indentation whitespace, too.

    gcc/ChangeLog:

            PR target/106355
            * config/s390/s390.cc (s390_call_saved_register_used): For a
            parameter with BLKmode fix determining number of consecutive
            registers.

    gcc/testsuite/ChangeLog:

            * gcc.target/s390/pr106355.h: Common code for new tests.
            * gcc.target/s390/pr106355-1.c: New test.
            * gcc.target/s390/pr106355-2.c: New test.
            * gcc.target/s390/pr106355-3.c: New test.

    (cherry picked from commit cb994acc08b67f26a54e7c5dc1f4995a2ce24d98)

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Stefan Schulze Frielinghaus
<stefansf@gcc.gnu.org>:

https://gcc.gnu.org/g:5e41c45fef8262caf20fdcf2a04427f16c295881

commit r11-10336-g5e41c45fef8262caf20fdcf2a04427f16c295881
Author: Stefan Schulze Frielinghaus <stefansf@linux.ibm.com>
Date:   Tue Oct 25 10:37:19 2022 +0200

    IBM zSystems: Fix function_ok_for_sibcall [PR106355]

    For a parameter with BLKmode we cannot use REG_NREGS in order to
    determine the number of consecutive registers.  Streamlined this with
    the implementation of s390_function_arg.

    Fix some indentation whitespace, too.

    gcc/ChangeLog:

            PR target/106355
            * config/s390/s390.c (s390_call_saved_register_used): For a
            parameter with BLKmode fix determining number of consecutive
            registers.

    gcc/testsuite/ChangeLog:

            * gcc.target/s390/pr106355.h: Common code for new tests.
            * gcc.target/s390/pr106355-1.c: New test.
            * gcc.target/s390/pr106355-2.c: New test.
            * gcc.target/s390/pr106355-3.c: New test.

    (cherry picked from commit cb994acc08b67f26a54e7c5dc1f4995a2ce24d98)

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Stefan Schulze Frielinghaus
<stefansf@gcc.gnu.org>:

https://gcc.gnu.org/g:97509e98d9199ff43916250fc0a20f3d1838d067

commit r10-11059-g97509e98d9199ff43916250fc0a20f3d1838d067
Author: Stefan Schulze Frielinghaus <stefansf@linux.ibm.com>
Date:   Tue Oct 25 10:38:23 2022 +0200

    IBM zSystems: Fix function_ok_for_sibcall [PR106355]

    For a parameter with BLKmode we cannot use REG_NREGS in order to
    determine the number of consecutive registers.  Streamlined this with
    the implementation of s390_function_arg.

    Fix some indentation whitespace, too.

    gcc/ChangeLog:

            PR target/106355
            * config/s390/s390.c (s390_call_saved_register_used): For a
            parameter with BLKmode fix determining number of consecutive
            registers.

    gcc/testsuite/ChangeLog:

            * gcc.target/s390/pr106355.h: Common code for new tests.
            * gcc.target/s390/pr106355-1.c: New test.
            * gcc.target/s390/pr106355-2.c: New test.
            * gcc.target/s390/pr106355-3.c: New test.

    (cherry picked from commit cb994acc08b67f26a54e7c5dc1f4995a2ce24d98)

[Bug target/106355] Linux s390x -O2 argument passing miscompile

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106355

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |mpolacek at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #9 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Assuming fixed.