[Bug testsuite/35677] New: Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[Bug testsuite/35677] New: Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dominiq at lps dot ens dot fr]

For the record, I see an intermitent failure (3 out of 44 tests) of
libgomp.fortran/crayptr2.f90 on i686-apple-darwin9:

rev. 132129: FAIL: libgomp.fortran/crayptr2.f90  -O2  execution test
rev. 133270: FAIL: libgomp.fortran/crayptr2.f90  -O3 -fomit-frame-pointer
-funroll-loops  execution test
rev. 133469: FAIL: libgomp.fortran/crayptr2.f90  -O0  execution test

The first case occured with -m64 and the two others with -m32 (default).


-- 
           Summary: Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: testsuite
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dominiq at lps dot ens dot fr
 GCC build triplet: i686-apple-darwin9
  GCC host triplet: i686-apple-darwin9
GCC target triplet: i686-apple-darwin9


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2008-03-30 20:12 -------
I see them also on powerpc-darwin.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2008-10-14 23:30 -------
I still see this, though it was failing a lot more than it was passing.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #3 from danglin at gcc dot gnu dot org  2008-11-27 02:07 -------
Also see failure on hppa2.0w-hp-hpux11.11.


-- 

danglin at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[howarth at nitro dot med dot uc dot edu]

------- Comment #4 from howarth at nitro dot med dot uc dot edu  2008-12-10 14:11 -------
Shouldn't this be marked as a confirmed and tagged as a regression?
Also, oddly I don't see this on any benchmark runs I do with a 3 GB MacBook
Pro using Core 2 Duo, yet I always see it on a 4 GB MacPro with Xeon 5400
series
processor. Both are building gcc trunk with configure flags: --prefix=/sw
--prefix=/sw/lib/gcc4.4 --mandir=/sw/share/man --infodir=/sw/share/info
--enable-languages=c,c++,fortran,objc,java --with-gmp=/sw
--with-libiconv-prefix=/sw --with-ppl=/sw --with-cloog=/sw --with-system-zlib
--x-includes=/usr/X11R6/include --x-libraries=/usr/X11R6/lib --with-arch=nocona
--with-tune=generic --build=i686-apple-darwin9 --host=i686-apple-darwin9
--target=i686-apple-darwin9


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #5 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-10 15:56 -------
Subject: Re:  Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

> Shouldn't this be marked as a confirmed and tagged as a regression?
> Also, oddly I don't see this on any benchmark runs I do with a 3 GB MacBook
> Pro using Core 2 Duo, yet I always see it on a 4 GB MacPro with Xeon 5400
> series
> processor. Both are building gcc trunk with configure flags: --prefix=/sw
> --prefix=/sw/lib/gcc4.4 --mandir=/sw/share/man --infodir=/sw/share/info
> --enable-languages=c,c++,fortran,objc,java --with-gmp=/sw
> --with-libiconv-prefix=/sw --with-ppl=/sw --with-cloog=/sw --with-system-zlib
> --x-includes=/usr/X11R6/include --x-libraries=/usr/X11R6/lib --with-arch=nocona
> --with-tune=generic --build=i686-apple-darwin9 --host=i686-apple-darwin9
> --target=i686-apple-darwin9

Strangely, I'm not seeing this on my early 2008 Mac Pro.  Possibly, the
difference is specifying "--enable-threads=posix --enable-__cxa_atexit".

Think the failures are more likely on SMP machines.  Do seem to have
more libgomp failures in 4.4 than 4.3 on hppa*-*-hpux*.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

-- 

danglin at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-12-10 16:01:39
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #6 from danglin at gcc dot gnu dot org  2008-12-10 18:10 -------
On hppa64, I see a "random" segfault in crayptr2.f90 compiled at -O0:

(gdb) r
Starting program:
/mnt/gnu/gcc/objdir/hppa64-hp-hpux11.11/libgomp/testsuite/crayptr2.xg0 
warning: Private mapping of shared library text was not specified
by the executable; setting a breakpoint in a shared library which
is not privately mapped will not work.  See the HP-UX 11i v3 chatr
manpage for methods to privately map shared library text.
warning: Loadable segment ".tbss" outside of ELF segments
warning: Loadable segment ".tbss" outside of ELF segments
warning: Loadable segment ".tbss" outside of ELF segments
[New process 28502, lwp 2270431]
[process 28502, lwp 2270431 exited]
[New process 28502, lwp 2270432]
[process 28502, lwp 2270432 exited]
[New process 28502, lwp 2270433]
[process 28502, lwp 2270433 exited]
[New process 28502, lwp 2270434]
[New process 28502, lwp 2270435]
[New process 28502, lwp 2270436]

Program received signal SIGSEGV, Segmentation fault.
[Switching to process 28502, lwp 2270435]
0x4000000000002d08 in MAIN__.omp_fn.0 (.omp_data_i=Cannot access memory at
address 0x0
)
    at /mnt/gnu/gcc/gcc/libgomp/testsuite/libgomp.fortran/crayptr2.f90:23
23        l = p .ne. omp_get_thread_num () + 1
Current language:  auto; currently fortran
(gdb) p/x $pc
$1 = 0x4000000000002d08
(gdb) disass 0x4000000000002cf8 0x4000000000002d18
Dump of assembler code from 0x4000000000002cf8 to 0x4000000000002d18:
0x4000000000002cf8 <MAIN__.omp_fn.0+136>:       b,l 0x4000000000002b2c
<.stub+60>,rp
0x4000000000002cfc <MAIN__.omp_fn.0+140>:       nop
0x4000000000002d00 <MAIN__.omp_fn.0+144>:       copy r4,dp
0x4000000000002d04 <MAIN__.omp_fn.0+148>:       ldd 0(ret0),ret0
0x4000000000002d08 <MAIN__.omp_fn.0+152>:       ldw 0(ret0),ret0
0x4000000000002d0c <MAIN__.omp_fn.0+156>:       extrd,s ret0,63,32,ret0
0x4000000000002d10 <MAIN__.omp_fn.0+160>:       stw r0,18(r3)
0x4000000000002d14 <MAIN__.omp_fn.0+164>:       cmpb,=,n
ret0,r5,0x4000000000002d20 <MAIN__.omp_fn.0+176>
End of assembler dump.
(gdb) p/x $ret0
$2 = 0x0

This results from the code loading indirectly data with a the pointer
returned from a call to __emutls_get_address:

        addil LT'__emutls_v.ip.565,%r27
        copy %r1,%r28
        ldd RT'__emutls_v.ip.565(%r28),%r28
        copy %r28,%r26
        ldo -48(%r30),%r29
        copy %r27,%r4
        b,l __emutls_get_address,%r2
        nop
        copy %r4,%r27
        ldd 0(%r28),%r28
        ldw 0(%r28),%r28

Is darwin also using EMUTLS?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #7 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-10 22:01 -------
Subject: Re:  Intermitent failure "FAIL:
        libgomp.fortran/crayptr2.f90"

On Wed, 10 Dec 2008, howarth at nitro dot med dot uc dot edu wrote:

I need to do some more testing but I believe the attached patch fixes
the failure on hppa*-*-hpux*.

Dave


------- Comment #8 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-10 22:01 -------
Created an attachment (id=16879)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16879&action=view)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[howarth at nitro dot med dot uc dot edu]

------- Comment #9 from howarth at nitro dot med dot uc dot edu  2008-12-11 00:28 -------
On i686-apple-darwin9, configured as...

Using built-in specs.
Target: i686-apple-darwin9
Configured with: ../gcc-4.4-20081209/configure --prefix=/sw
--prefix=/sw/lib/gcc4.4 --mandir=/sw/share/man --infodir=/sw/share/info
--enable-languages=c,c++,fortran,objc,java --with-gmp=/sw
--with-libiconv-prefix=/sw --with-ppl=/sw --with-cloog=/sw --with-system-zlib
--x-includes=/usr/X11R6/include --x-libraries=/usr/X11R6/lib --with-arch=nocona
--with-tune=generic --build=i686-apple-darwin9 --host=i686-apple-darwin9
--target=i686-apple-darwin9
Thread model: posix
gcc version 4.4.0 20081210 (experimental) (GCC) 

I get...

rogram received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000000
[Switching to process 560 thread 0x313]
0x00001ae9 in MAIN__.omp_fn.0 ()
(gdb) bt
#0  0x00001ae9 in MAIN__.omp_fn.0 ()
#1  0x0001a204 in gomp_thread_start (xdata=0xbfffe7f0) at
../../../gcc-4.4-20081209/libgomp/team.c:118
#2  0x910c4095 in _pthread_start ()
#3  0x910c3f52 in thread_start ()

I assume I am using emutls since I see...

nm crayptr2.exe | grep emut
00001cd0 t ___emutls_get_address
00001c90 t ___emutls_register_common
00002014 d ___emutls_v.ip.1499
00001f10 t _emutls_destroy
00001ed0 t _emutls_init
00002090 b _emutls_key
00002040 d _emutls_mutex
00002094 b _emutls_size

I'll try a build of current gcc trunk with your patch and see if it helps.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #10 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-11 00:40 -------
Subject: Re:  Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

> I'll try a build of current gcc trunk with your patch and see if it helps.

Problem looks the same.  You can do a quick check without a full rebuild
by removing the emultls* files from the libgcc directories, apply patch,
do make in top libgcc directory, then rerun libgomp testsuite.

The failures occur because multiple threads read an offset value of 0.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[jakub at gcc dot gnu dot org]

------- Comment #11 from jakub at gcc dot gnu dot org  2008-12-11 00:44 -------
The patch is correct if not obvious.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[howarth at nitro dot med dot uc dot edu]

------- Comment #12 from howarth at nitro dot med dot uc dot edu  2008-12-11 02:02 -------
This patch appears to solve the issue on i686-apple-darwin9 from the couple
testsuite runs I have done in libgomp.

                === libgomp Summary for unix/-m32 ===

# of expected passes            2071
# of unsupported tests          128

                === libgomp Summary for unix/-m64 ===

# of expected passes            2071
# of unsupported tests          128

                === libgomp Summary ===

# of expected passes            4142
# of unsupported tests          256


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #13 from danglin at gcc dot gnu dot org  2008-12-11 18:35 -------
Subject: Bug 35677

Author: danglin
Date: Thu Dec 11 18:33:48 2008
New Revision: 142687

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142687
Log:
        PR testsuite/35677
        * emutls.c (__emutls_get_address): Make sure offset is really zero
        before initializing the object's offset.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/emutls.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #15 from danglin at gcc dot gnu dot org  2008-12-11 23:32 -------
Subject: Bug 35677

Author: danglin
Date: Thu Dec 11 23:30:42 2008
New Revision: 142695

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142695
Log:
        PR testsuite/35677
        * emutls.c (__emutls_get_address): Make sure offset is really zero
        before initializing the object's offset.


Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/emutls.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #14 from danglin at gcc dot gnu dot org  2008-12-11 23:31 -------
Fixed.


-- 

danglin at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #16 from danglin at gcc dot gnu dot org  2008-12-11 23:38 -------
*** Bug 31839 has been marked as a duplicate of this bug. ***


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[danglin at gcc dot gnu dot org]

------- Comment #17 from danglin at gcc dot gnu dot org  2008-12-11 23:39 -------
*** Bug 31838 has been marked as a duplicate of this bug. ***


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #18 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-12 15:04 -------
Subject: Re:  Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

> I assume I am using emutls since I see...
> 
> nm crayptr2.exe | grep emut
> 00001cd0 t ___emutls_get_address
> 00001c90 t ___emutls_register_common
> 00002014 d ___emutls_v.ip.1499
> 00001f10 t _emutls_destroy
> 00001ed0 t _emutls_init
> 00002090 b _emutls_key
> 00002040 d _emutls_mutex
> 00002094 b _emutls_size
> 
> I'll try a build of current gcc trunk with your patch and see if it helps.

I probably was too quick to close this PR.  The emutls support in
darwin's libgcc is botched.  The routines are present in libgcc_eh.a
but not in the shared libraries.

As I understand the situation from the gcc list, the Apple darwin
maintainers oppose the addition of emutls support to the shared libgcc
libraries.  The emutls support is licensed under GPLv2, so that
shouldn't block adoption.  However, Apple may still want to provide
its own TLS support.

On the other hand, Apple is phasing out use of gcc.  So, the issue of
control might be something the steering committee should discuss.
>From my perspective, most of the support and testing of the darwin
port in recent months has come from non-Apple developpers.

I think the only way forward is to disable TLS support on darwin
in gcc 4.3 and 4.4.  Probably, a new PR should be created for this.

Any testcases that use TLS and lack a "dg-require-effective-target tls_runtime"
comment need updating.  Those changes are obvious.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[howarth at nitro dot med dot uc dot edu]

------- Comment #19 from howarth at nitro dot med dot uc dot edu  2008-12-12 17:52 -------
Well, they are phasing it out in the long run perhaps but Apple is just
transitioning to gcc 4.2.1 at the moment so gcc will be in use for awhile
yet on darwin. I am unclear on what the clang/llvm will use for its libgcc
equivalent. Perhaps Chris Lattner can enlighten us on that.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #20 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-12 18:41 -------
Subject: Re:  Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

> Well, they are phasing it out in the long run perhaps but Apple is just
> transitioning to gcc 4.2.1 at the moment so gcc will be in use for awhile
> yet on darwin. I am unclear on what the clang/llvm will use for its libgcc
> equivalent. Perhaps Chris Lattner can enlighten us on that.

In principle, users can install libgcc* wherever they like.  I think
the install path for shared libraries is hardcoded in executables (otool -L).
There's also DYLD_LIBRARY_PATH and DYLD_FALLBACK_LIBRARY_PATH.  As
far as I know, the libgcc libraries shipped with 4.3 and in 4.4 are
backwards compatible with earlier versions.

My impression is that this issue is not confined to libgcc.  The same
problem is present for all the libraries provided by gcc.  Any added
features are going to make the new version incompatible with the system
version, irrespective of ABI changes.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[mrs at apple dot com]

------- Comment #21 from mrs at apple dot com  2008-12-12 18:42 -------
My long term guidance would be to engineer gcc to use its copy of the libgcc_s
dylib and link against it.  This would mean that the newly installed libgcc_s
should be found first, over /usr/lib, and that development and support mirror
linux, in all the usual ways.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677

[Bug testsuite/35677] Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #22 from dave at hiauly1 dot hia dot nrc dot ca  2008-12-12 19:27 -------
Subject: Re:  Intermitent failure "FAIL: libgomp.fortran/crayptr2.f90"

> My long term guidance would be to engineer gcc to use its copy of the libgcc_s
> dylib and link against it.  This would mean that the newly installed libgcc_s
> should be found first, over /usr/lib, and that development and support mirror
> linux, in all the usual ways.

That's certainly reasonable.  However, linux style TLS support needs
help from the assembler, linker and dynamic loader.  So, unless this
is going to happen, the best that can be achieved is that provided by
the emulation layer.

Tracking linux might be straight forward if the kernel supported ELF
executables.  Then, x86 linux tools would work essentially as is.
However, I'm sure there would be issues with syscalls, etc.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35677