[Bug jit/97867] New: FAIL: test-combination.c.exe test-functions.c.exe test-pr66779.c.exe test-threads.c.exe killed

[Bug jit/97867] New: FAIL: test-combination.c.exe test-functions.c.exe test-pr66779.c.exe test-threads.c.exe killed

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

            Bug ID: 97867
           Summary: FAIL: test-combination.c.exe test-functions.c.exe
                    test-pr66779.c.exe test-threads.c.exe  killed
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: jit
          Assignee: dmalcolm at gcc dot gnu.org
          Reporter: msebor at gcc dot gnu.org
  Target Milestone: ---

Today's build of mine of GCC trunk on x86_64-linux shows failures in the
following tests:

!  FAIL: test-combination.c.exe (1: +1)
!  FAIL: test-functions.c.exe (1: +1)
!  FAIL: test-pr66779.c.exe (1: +1)
!  FAIL: test-threads.c.exe (1: +1)

They all seem to fail with a SEGV:

$ find gcc/testsuite -name "*.log" | xargs grep -e test-combination.c.exe -e
test-functions.c.exe -e test-pr66779.c.exe -e test-threads.c.exe | grep FAIL)
gcc/testsuite/jit/jit.log:FAIL: test-combination.c.exe killed: 23122 exp6 0 0
CHILDKILLED SIGSEGV {segmentation violation}
gcc/testsuite/jit/jit.log:FAIL: test-functions.c.exe killed: 31302 exp6 0 0
CHILDKILLED SIGSEGV {segmentation violation}
gcc/testsuite/jit/jit.log:FAIL: test-pr66779.c.exe killed: 28080 exp6 0 0
CHILDKILLED SIGSEGV {segmentation violation}
gcc/testsuite/jit/jit.log:FAIL: test-threads.c.exe killed: 5980 exp6 0 0
CHILDKILLED SIGABRT SIGABRT

[Bug jit/97867] FAIL: test-combination.c.exe test-functions.c.exe test-pr66779.c.exe test-threads.c.exe killed

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

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

--- Comment #1 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Probably this issue:
  https://gcc.gnu.org/pipermail/gcc-patches/2020-November/558842.html
which was introduced in g:67f3791f7d133214b112bd831ff2876822c665d0 and which
makes the same tests fail for me.

Are the ICEs within thunk_info::release?

Honza: did you get around to fixing that one yet?

[Bug jit/97867] FAIL: test-combination.c.exe test-functions.c.exe test-pr66779.c.exe test-threads.c.exe killed

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

--- Comment #2 from Martin Sebor <msebor at gcc dot gnu.org> ---
The backtrace from test-combination.c.exe shows the SEGV does happen in
thunk_info::release ().  Ditto for test-functions.c.exe.

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff5d74153 in free () from /lib64/libc.so.6
Missing separate debuginfos, use: dnf debuginfo-install gmp-6.1.2-9.fc29.x86_64
libmpc-1.1.0-2.fc29.x86_64 mpfr-3.1.6-2.fc29.x86_64
(gdb) bt
#0  0x00007ffff5d74153 in free () from /lib64/libc.so.6
#1  0x00007ffff654bd62 in thunk_info::release () at
/ssd/test/src/gcc/dump-type-attr/gcc/cgraph.c:4024
#2  cgraph_c_finalize () at /ssd/test/src/gcc/dump-type-attr/gcc/cgraph.c:4017
#3  0x00007ffff69b2229 in toplev::finalize (this=this@entry=0x7fffffffdd4e)
    at /ssd/test/src/gcc/dump-type-attr/gcc/toplev.c:2508
#4  0x00007ffff64a31b0 in gcc::jit::playback::context::compile
(this=this@entry=0x7fffffffdde0)
    at /ssd/test/src/gcc/dump-type-attr/gcc/jit/jit-playback.c:2124
#5  0x00007ffff6498bc6 in gcc::jit::recording::context::compile
(this=this@entry=0x47b3f0)
    at /ssd/test/src/gcc/dump-type-attr/gcc/jit/jit-recording.c:1397
#6  0x00007ffff648a3bb in gcc_jit_context_compile (ctxt=0x47b3f0)
    at /ssd/test/src/gcc/dump-type-attr/gcc/jit/libgccjit.c:2827
#7  0x000000000041be40 in test_jit (
    argv0=0x7fffffffe2fe
"/ssd/test/build/gcc-gcc-dump-type-attr/gcc/testsuite/jit/test-combination.c.exe", 
    user_data=0x0) at
/ssd/test/src/gcc/dump-type-attr/gcc/testsuite/jit.dg/harness.h:371
#8  0x000000000041bf30 in main (argc=1, argv=0x7fffffffdfe8)
    at /ssd/test/src/gcc/dump-type-attr/gcc/testsuite/jit.dg/harness.h:419

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|FAIL:                       |[11 Regression]
                   |test-combination.c.exe      |thunk_info::release breaks
                   |test-functions.c.exe        |function calls in libgccjit
                   |test-pr66779.c.exe          |
                   |test-threads.c.exe  killed  |
           Assignee|dmalcolm at gcc dot gnu.org        |hubicka at gcc dot gnu.org
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-11-24
     Ever confirmed|0                           |1

--- Comment #3 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Thanks Martin.

Looks like this is Honza's thunk_info cleanup stuff that I reported on the
list. 

Reassigning, and setting to P1 as this will likely break all non-trivial usage
of libgccjit.

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

--- Comment #4 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Sorry, I lost track of this, because i still hit the strange linker error with
building libjit

The following ghsould fix it.
diff --git a/gcc/symtab-thunks.h b/gcc/symtab-thunks.h
index 41a684995b3..0dba2217793 100644
--- a/gcc/symtab-thunks.h
+++ b/gcc/symtab-thunks.h
@@ -167,7 +167,7 @@ inline void
 thunk_info::release ()
 { 
   if (symtab->m_thunks)
-    delete (symtab->m_thunks);
+    ggc_delete (symtab->m_thunks);
   symtab->m_thunks = NULL;
 }
 #endif  /* GCC_SYMTAB_THUNKS_H  */

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jan Hubicka <hubicka@gcc.gnu.org>:

https://gcc.gnu.org/g:caea077c17fdf6fe1aa4c350260f25439dfd5066

commit r11-5524-gcaea077c17fdf6fe1aa4c350260f25439dfd5066
Author: Jan Hubicka <jh@suse.cz>
Date:   Sun Nov 29 23:22:24 2020 +0100

    Fix freeing of thunk-info

            PR jit/97867
            * symtab-thunks.h (thunk_info::release): Use ggc_delete.

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

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

--- Comment #6 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Fixed sorry for taking so long

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

David Malcolm <dmalcolm at gcc dot gnu.org> changed:

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

--- Comment #7 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
(In reply to Jan Hubicka from comment #6)
> Fixed sorry for taking so long

Thanks for fixing this.


(In reply to Jan Hubicka from comment #4)
> Sorry, I lost track of this, because i still hit the strange linker error
> with building libjit

FWIW, did you configure with --enable-host-shared ?  Forgetting to enable that
is the usual source of weird errors when building libgccjit.

[Bug jit/97867] [11 Regression] thunk_info::release breaks function calls in libgccjit

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97867

--- Comment #8 from Jan Hubicka <hubicka at ucw dot cz> ---
> 
> FWIW, did you configure with --enable-host-shared ?  Forgetting to enable that
> is the usual source of weird errors when building libgccjit.

I think it does not let you to build it otherwise, but I believe I did.
I just need to find time to update my setup. It is really out of date.

BTW if you build with LTO, I think there should be no performance
penalty for building libbackend with -fpic since we take it away while
producing the final binary.  This may be relevant for those who build
distros (I suppose you do not want to ship main compilers with PIC build
for performance reasons).

Honza