[Bug libfortran/118935] Segmentation fault in 'libgomp.fortran/rwlock_1.f90' when compiling libgfortran with '-O0'

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118935

--- Comment #15 from huangpei at loongson dot cn <huangpei at loongson dot cn> ---
Same issue on GCC 15.1.0, Tsan says:

==================
WARNING: ThreadSanitizer: data race (pid=3090994)
  Read of size 8 at 0x7fffee107808 by thread T3 (mutexes: read M0):
    #0 find_file0 ../../../gcc-master/libgfortran/io/unix.c:1721
(libgfortran.so.5+0x36bbfc)
    #1 find_file0 ../../../gcc-master/libgfortran/io/unix.c:1744
(libgfortran.so.5+0x36bcec)
    #2 find_file0 ../../../gcc-master/libgfortran/io/unix.c:1744
(libgfortran.so.5+0x36bcec)
    #3 find_file0 ../../../gcc-master/libgfortran/io/unix.c:1744
(libgfortran.so.5+0x36bcec)
    #4 find_file0 ../../../gcc-master/libgfortran/io/unix.c:1744
(libgfortran.so.5+0x36bcec)
    #5 _gfortrani_find_file ../../../gcc-master/libgfortran/io/unix.c:1778
(libgfortran.so.5+0x36bdd8)
    #6 _gfortran_st_inquire ../../../gcc-master/libgfortran/io/inquire.c:821
(libgfortran.so.5+0x33bc34)
    #7 MAIN__._omp_fn.0 <null> (rwlock_1.f90.exe+0x1200014d0)
    #8 gomp_thread_start ../../../gcc-master/libgomp/team.c:129
(libgomp.so.1+0x25f04)

  Previous write of size 8 at 0x7fffee107808 by thread T2 (mutexes: write M1):
    #0 _gfortrani_new_unit ../../../gcc-master/libgfortran/io/open.c:588
(libgfortran.so.5+0x34d824)
    #1 _gfortran_st_open ../../../gcc-master/libgfortran/io/open.c:931
(libgfortran.so.5+0x34ed5c)
    #2 MAIN__._omp_fn.0 <null> (rwlock_1.f90.exe+0x1200013f4)
    #3 gomp_thread_start ../../../gcc-master/libgomp/team.c:129
(libgomp.so.1+0x25f04)

  Location is heap block of size 752 at 0x7fffee107800 allocated by thread T2:
    #0 calloc
../../../../gcc-master/libsanitizer/tsan/tsan_interceptors_posix.cpp:698
(libtsan.so.2+0x966ac)
    #1 _gfortrani_xcalloc ../../../gcc-master/libgfortran/runtime/memory.c:78
(libgfortran.so.5+0x29c58)
    #2 insert_unit ../../../gcc-master/libgfortran/io/unit.c:239
(libgfortran.so.5+0x365434)
    #3 get_gfc_unit ../../../gcc-master/libgfortran/io/unit.c:391
(libgfortran.so.5+0x365b40)
    #4 _gfortrani_find_or_create_unit
../../../gcc-master/libgfortran/io/unit.c:456 (libgfortran.so.5+0x365db0)
    #5 _gfortran_st_open ../../../gcc-master/libgfortran/io/open.c:928
(libgfortran.so.5+0x34ed28)
    #6 MAIN__._omp_fn.0 <null> (rwlock_1.f90.exe+0x1200013f4)
    #7 gomp_thread_start ../../../gcc-master/libgomp/team.c:129
(libgomp.so.1+0x25f04)

  Mutex M0 (0x7ffff29d18a0) created at:
    #0 pthread_rwlock_wrlock
../../../../gcc-master/libsanitizer/tsan/tsan_interceptors_posix.cpp:1552
(libtsan.so.2+0x5f630)
    #1 __gthread_rwlock_wrlock ../libgcc/gthr-default.h:957
(libgfortran.so.5+0x364ccc)
    #2 _gfortrani_newunit_alloc ../../../gcc-master/libgfortran/io/unit.c:944
(libgfortran.so.5+0x367928)
    #3 _gfortrani_get_unit ../../../gcc-master/libgfortran/io/unit.c:590
(libgfortran.so.5+0x366810)
    #4 data_transfer_init ../../../gcc-master/libgfortran/io/transfer.c:3144
(libgfortran.so.5+0x35de08)
    #5 _gfortran_st_write ../../../gcc-master/libgfortran/io/transfer.c:4714
(libgfortran.so.5+0x363a10)
    #6 MAIN__._omp_fn.0 <null> (rwlock_1.f90.exe+0x12000127c)
    #7 GOMP_parallel ../../../gcc-master/libgomp/parallel.c:178
(libgomp.so.1+0x1ab20)
    #8 main <null> (rwlock_1.f90.exe+0x12000110c)

  Mutex M1 (0x7fffee1078e0) created at:
    #0 pthread_mutex_lock
../../../../gcc-master/libsanitizer/tsan/tsan_interceptors_posix.cpp:1371
(libtsan.so.2+0x60998)
    #1 __gthread_mutex_lock ../libgcc/gthr-default.h:795
(libgfortran.so.5+0x364b2c)
    #2 insert_unit ../../../gcc-master/libgfortran/io/unit.c:250
(libgfortran.so.5+0x3654dc)
    #3 get_gfc_unit ../../../gcc-master/libgfortran/io/unit.c:391
(libgfortran.so.5+0x365b40)
    #4 _gfortrani_find_or_create_unit
../../../gcc-master/libgfortran/io/unit.c:456 (libgfortran.so.5+0x365db0)
    #5 _gfortran_st_open ../../../gcc-master/libgfortran/io/open.c:928
(libgfortran.so.5+0x34ed28)
    #6 MAIN__._omp_fn.0 <null> (rwlock_1.f90.exe+0x1200013f4)
    #7 gomp_thread_start ../../../gcc-master/libgomp/team.c:129
(libgomp.so.1+0x25f04)

  Thread T3 (tid=3090999, running) created by main thread at:
    #0 pthread_create
../../../../gcc-master/libsanitizer/tsan/tsan_interceptors_posix.cpp:1041
(libtsan.so.2+0x613a8)
    #1 gomp_team_start ../../../gcc-master/libgomp/team.c:859
(libgomp.so.1+0x26834)
    #2 main <null> (rwlock_1.f90.exe+0x12000110c)

  Thread T2 (tid=3090998, running) created by main thread at:
    #0 pthread_create
../../../../gcc-master/libsanitizer/tsan/tsan_interceptors_posix.cpp:1041
(libtsan.so.2+0x613a8)
    #1 gomp_team_start ../../../gcc-master/libgomp/team.c:859
(libgomp.so.1+0x26834)
    #2 main <null> (rwlock_1.f90.exe+0x12000110c)

SUMMARY: ThreadSanitizer: data race
../../../gcc-master/libgfortran/io/unix.c:1721 in find_file0
==================================================

I guess this is what happened:

================================================================================
thread T3 (find_file0() <= find_file())    thread T3 (new_unit()  <=
get_unit())
───────────────────────────────         ──────────────────────────────────────
                                        hold/release unit_rwlock : u =
find_or_create_unit() <= st_open()

rdlock(unit_rwlock) M0

#3**(SARAROoO load u->s && s->st_dev → SEGV)

                                        #1 if (u->s == NULL)
                                                u = new_unit(); u->s = s  M1



#2 u->s != NULL 
#3reload u->s && s->st_dev
===============================================================================


Data race is clear, since neither M0 nor M1 cover this case; However, Loong64
is
week ordered( even same address read after read is out of order), #3 is
reordered by cpu and see u->s is NULL before #1;


Use aotmic operartion to fix this problem and the patch is on atttachment