[Bug regression/61436] New: [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[Bug regression/61436] New: [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

            Bug ID: 61436
           Summary: [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)
           Product: gcc
           Version: 4.10.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: regression
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hp at gcc dot gnu.org
                CC: hubicka at gcc dot gnu.org
              Host: x86_64-unknown-linux-gnu
            Target: mmix-knuth-mmixware

This test previously passed, now it fails.
A patch in the revision range (last_known_working:first_known_failing)
r210000:r211121
exposed or caused these regressions.  Since then (up to and including at least
r211290) it fails as follows:

Running /home/hp/gcctop/tmp/mbgnu0/gcc/gcc/testsuite/g++.dg/tls/tls.exp ...
FAIL: g++.dg/tls/diag-1.C -std=c++98 (internal compiler error)
FAIL: g++.dg/tls/diag-1.C -std=c++98 (test for excess errors)
FAIL: g++.dg/tls/diag-1.C -std=c++11 (internal compiler error)
FAIL: g++.dg/tls/diag-1.C -std=c++11 (test for excess errors)
FAIL: g++.dg/tls/diag-1.C -std=c++1y (internal compiler error)
FAIL: g++.dg/tls/diag-1.C -std=c++1y (test for excess errors)

I'm guessing all emutls targets are affected.

The messages in g++.log are similar:

spawn -ignore SIGHUP
/home/hp/gcctop/tmp/mbase0/gccobj/gcc/testsuite/g++/../../xg++
-B/home/hp/gcctop/tmp/mbase0/gccobj/gcc/testsuite/g++/../../
/home/hp/gcctop/tmp/mbase0/gcc/gcc/testsuite/g++.dg/tls/diag-1.C
-fno-diagnostics-show-caret -fdiagnostics-color=never -nostdinc++
-I/home/hp/gcctop/tmp/mbase0/gccobj/mmix/libstdc++-v3/include/mmix
-I/home/hp/gcctop/tmp/mbase0/gccobj/mmix/libstdc++-v3/include
-I/home/hp/gcctop/tmp/mbase0/gcc/libstdc++-v3/libsupc++
-I/home/hp/gcctop/tmp/mbase0/gcc/libstdc++-v3/include/backward
-I/home/hp/gcctop/tmp/mbase0/gcc/libstdc++-v3/testsuite/util -fmessage-length=0
-std=c++98 -pedantic-errors -Wno-long-long -S -isystem
/home/hp/gcctop/tmp/mbase0/gccobj/mmix/./newlib/targ-include -isystem
/home/hp/gcctop/tmp/mbase0/gcc/newlib/libc/include -o diag-1.s

/home/hp/gcctop/tmp/mbase0/gcc/gcc/testsuite/g++.dg/tls/diag-1.C:31:1: internal
compiler error: in symtab_get_node, at cgraph.h:1035

0xd610c3 symtab_get_node

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraph.h:1032

0xd610c3 varpool_get_node

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraph.h:1074

0xd610c3 varpool_node_for_decl(tree_node*)

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/varpool.c:147

0xd54cb4 make_decl_one_only(tree_node*, tree_node*)

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/varasm.c:5932

0xb2a14e get_emutls_init_templ_addr

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:254

0xb2a14e get_emutls_init_templ_addr

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:222

0xb2aade new_emutls_decl

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:339

0xb2aade create_emultls_var

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:717

0xd625ef varpool_for_node_and_aliases(varpool_node*, bool (*)(varpool_node*,
void*), void*, bool)

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/varpool.c:682

0xb2ba87 ipa_lower_emutls

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:788

0xb2ba87 execute

    /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:849

Please submit a full bug report,

with preprocessed source if appropriate.


Looking at the code, it seems the assert in symtab_get_node at cgraph.h:1035
trigs. This code was changed in r210901, though the assert was only copied.
Still, many changes around that time by the same author, so there's a good
chance the attribution is correct. :)

A gdb session shows, at the gcc_assert:

(gdb) p decl
$15 = (const_tree) 0x7ffff7ea68e8
(gdb) pt
 <var_decl 0x7ffff7ea68e8 __emutls_t__ZN1BIiE1tE
    type <integer_type 0x7ffff7dbd690 int sizes-gimplified public type_6 SI
        size <integer_cst 0x7ffff7dbce28 constant 32>
        unit size <integer_cst 0x7ffff7dbce40 constant 4>
        align 32 symtab 0 alias set -1 canonical type 0x7ffff7dbd690 precision
32 min <integer_cst 0x7ffff7dbcde0 -2147483648> max <integer_cst 0x7ffff7dbcdf8
2147483647>
        pointer_to_this <pointer_type 0x7ffff7dd1738>>
    readonly used public ignored weak SI file
/home/hp/gcctop/tmp/mbase0/gcc/gcc/testsuite/g++.dg/tls/diag-1.C line 25 col 12
size <integer_cst 0x7ffff7dbce28 32> unit size <integer_cst 0x7ffff7dbce40 4>
    align 32 context <record_type 0x7ffff7ea2540 B>>
(gdb) bt
#0  fancy_abort (file=0xfa7550 "/home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraph.h", 
    line=1035, function=0xfa77e0 "symtab_get_node")
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/diagnostic.c:1190
#1  0x0000000000d64ed4 in symtab_get_node (decl=0x7ffff7ea68e8)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraph.h:1032
#2  0x0000000000d64f8b in varpool_get_node (decl=0x7ffff7ea68e8)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraph.h:1074
#3  0x0000000000d655fc in varpool_node_for_decl (decl=
    <var_decl 0x7ffff7ea68e8 __emutls_t__ZN1BIiE1tE>)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/varpool.c:147
#4  0x0000000000d58c85 in make_decl_one_only (decl=0x7ffff7ea68e8, 
    comdat_group=0x7ffff7e9ca50)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/varasm.c:5932
#5  0x0000000000b2e11f in get_emutls_init_templ_addr (decl=0x7ffff7e94da8)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:254
#6  get_emutls_init_templ_addr (decl=0x7ffff7e94da8)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:222
#7  0x0000000000b2eaaf in new_emutls_decl (alias_of=0x0, decl=0x7ffff7e94da8)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:339
#8  create_emultls_var (var=0x7ffff7ea31c0, data=0x7fffffffe888)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:717
#9  0x0000000000d66ccd in varpool_for_node_and_aliases (node=0x7ffff7ea31c0, 
    callback=0xb2e600 <create_emultls_var(varpool_node*, void*)>, data=
    0x7fffffffe888, include_overwritable=true)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/varpool.c:682
#10 0x0000000000b2fa58 in ipa_lower_emutls ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:788
#11 (anonymous namespace)::pass_ipa_lower_emutls::execute (
    this=<optimized out>)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/tree-emutls.c:849
#12 0x0000000000a54cc8 in execute_one_pass (pass=pass@entry=
    <opt_pass* 0x142cf80 "emutls"(47)>)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/passes.c:2180
#13 0x0000000000a5569b in execute_ipa_pass_list (pass=
    <opt_pass* 0x142cf80 "emutls"(47)>)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/passes.c:2571
#14 0x00000000007e8850 in ipa_passes ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraphunit.c:2054
#15 compile () at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraphunit.c:2185
#16 0x00000000007e9095 in finalize_compilation_unit ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cgraphunit.c:2340
#17 0x00000000005cc468 in cp_write_global_declarations ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/cp/decl2.c:4647
#18 0x0000000000ad8485 in compile_file ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/toplev.c:562
#19 0x0000000000ada245 in do_compile ()
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/toplev.c:1918
#20 toplev_main (argc=17, argv=0x7fffffffec38)
    at /home/hp/gcctop/tmp/mbase0/gcc/gcc/toplev.c:1994
#21 0x0000003522e21735 in __libc_start_main () from /lib64/libc.so.6
#22 0x00000000004d2495 in _start ()
(gdb) 

Author of one or more suspect patches in revision range CC:ed.

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|mmix-knuth-mmixware         |mmix-knuth-mmixware,
                   |                            |cris-elf

--- Comment #1 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
Bah, I should've checked my cris-elf autotester, since I suspected an
emutls-related bug.  It confirms the issue, and says the regression was
introduced in (210899:210906], consistent with the previous observations.

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

--- Comment #2 from Jan Hubicka <hubicka at ucw dot cz> ---
Hi,
it seems that get_emutls_init_templ_addr just create the comdat too early
and we trigger sanity check that symbols are static and not automatic
variables.

I have busy day tomorrow, so won't beat if anyone beats me in testing if
this fix works.

Honza

Index: tree-emutls.c
===================================================================
--- tree-emutls.c    (revision 211363)
+++ tree-emutls.c    (working copy)
@@ -250,10 +250,10 @@ get_emutls_init_templ_addr (tree decl)
   DECL_WEAK (to) = DECL_WEAK (decl);
   if (DECL_ONE_ONLY (decl))
     {
-      make_decl_one_only (to, DECL_ASSEMBLER_NAME (to));
       TREE_STATIC (to) = TREE_STATIC (decl);
       TREE_PUBLIC (to) = TREE_PUBLIC (decl);
       DECL_VISIBILITY (to) = DECL_VISIBILITY (decl);
+      make_decl_one_only (to, DECL_ASSEMBLER_NAME (to));
     }
   else
     TREE_STATIC (to) = 1;

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> it seems that get_emutls_init_templ_addr just create the comdat too early
> and we trigger sanity check that symbols are static and not automatic variables.
>
> I have busy day tomorrow, so won't beat if anyone beats me in testing if
> this fix works. ...

The fix works for me (x86_64-apple-darwin13).

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.10.0

[Bug regression/61436] [4.10 Regression]: g++.dg/tls/diag-1.C ICE (emutls)

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61436

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

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

--- Comment #4 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
Fixed, apparently in r211689 (though trunk broken for cris-elf at that time).
Though, the fix may be incomplete as the commit apparently introduced PR61548.