[Bug libgomp/60107] New: libgomp.c/pr58392.c etc. FAIL

[Bug libgomp/60107] New: libgomp.c/pr58392.c etc. FAIL

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

            Bug ID: 60107
           Summary: libgomp.c/pr58392.c etc. FAIL
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libgomp
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
                CC: jakub at gcc dot gnu.org
              Host: i386-pc-solaris2.9
            Target: i386-pc-solaris2.9
             Build: i386-pc-solaris2.9

3 libgomp tests FAIL on Solaris 9/x86 since they were introduced:

FAIL: libgomp.c/pr58392.c execution test
FAIL: libgomp.c/simd-6.c execution test
FAIL: libgomp.c++/simd-8.C execution test

For the first, I find

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 2 (LWP 2)]
0x080509f1 in foo._omp_fn ()
    at /vol/gcc/src/hg/trunk/local/libgomp/testsuite/libgomp.c/pr58392.c:20
20                  c += d[j + l];
(gdb) x/i $pc
=> 0x80509f1 <foo._omp_fn.0+177>:       paddd  0x10(%esp),%xmm0
(gdb) p $esp
$1 = (void *) 0xb5aedf64

truss shows
/2:         Incurred fault #6, FLTBOUNDS  %pc = 0x080509F1
/2:           siginfo: SIGSEGV SEGV_MAPERR addr=0x080509F1
/2:         Received signal #11, SIGSEGV [default]
/2:           siginfo: SIGSEGV SEGV_MAPERR addr=0x080509F1

and <sys/fault.h> has

#define FLTBOUNDS       6       /* Memory bounds (invalid address) */

pmap reveals the following mappings:

08062000      12K rwx--    [ heap ]
B5AED000       4K rwx-R    [ anon ]
B5AF0000     108K rw---    [ anon ]

  Rainer

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Does solaris "properly" align its stack on the entry of main?
0xb5aedf64 + 0x10 isn't aligned for paddd.

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Well, this isn't about main alignment (and, after all, main realigns the stack
anyway), but about stack alignment upon entering pthread_create callbacks.
If the Solaris libraries don't maintain 16-byte stack alignment, then either
you could arrange to compile parts of libgomp with -mincoming-stack-boundary=2
(or -mstackrealign) on Solaris/i?86 32-bit.  Or just make sure the default on
Solaris is -mincoming-stack-boundary=2 or whatever the libraries do guarantee.

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #3 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> ---
> --- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
> Well, this isn't about main alignment (and, after all, main realigns the stack
> anyway), but about stack alignment upon entering pthread_create callbacks.
> If the Solaris libraries don't maintain 16-byte stack alignment, then either
> you could arrange to compile parts of libgomp with -mincoming-stack-boundary=2
> (or -mstackrealign) on Solaris/i?86 32-bit.  Or just make sure the default on
> Solaris is -mincoming-stack-boundary=2 or whatever the libraries do guarantee.

I've checked the Solaris 9 and OpenSolaris sources: on S9, there's just
the 4-byte alignment guaranteed by the i386 psABI.  On S10 and up, they
enforce 16-byte alignment for the benefit of SSE.

IMO, the best (most general) solution is to change STACK_REALIGN_DEFAULT
to 1 on Solaris 9/x86.  While it comes at a performance and code size
penalty, it's guaranteed to be correct.

I've bootstrapped a patch to do just that and as expected, the libgomp
failures are gone.

Unlike I'm overlooking something important, I'd like to check this patch
into mainline for 4.9.0.

    Rainer

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-02-11
   Target Milestone|---                         |4.9.0
     Ever confirmed|0                           |1

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #4 from Rainer Orth <ro at gcc dot gnu.org> ---
Created attachment 32101
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32101&action=edit
proposed patch

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
>From RM POV this is ok for 4.9.

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

--- Comment #6 from Rainer Orth <ro at gcc dot gnu.org> ---
Author: ro
Date: Tue Feb 11 12:56:39 2014
New Revision: 207688

URL: http://gcc.gnu.org/viewcvs?rev=207688&root=gcc&view=rev
Log:
Only assume 4-byte stack alignment on Solaris 9/x86 (PR libgomp/60107)

    PR libgomp/60107
    * config/i386/sol2-9.h: New file.
    * config.gcc (i[34567]86-*-solaris2* | x86_64-*-solaris2.1[0-9]*,
    *-*-solaris2.9*): Use it.

Added:
    trunk/gcc/config/i386/sol2-9.h
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config.gcc

[Bug libgomp/60107] libgomp.c/pr58392.c etc. FAIL

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60107

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED
           Assignee|unassigned at gcc dot gnu.org      |ro at gcc dot gnu.org

--- Comment #7 from Rainer Orth <ro at gcc dot gnu.org> ---
Fixed for 4.9.0.