[Bug debug/101905] New: Missed debug information for global register variable

[Bug debug/101905] New: Missed debug information for global register variable

[yuri.karlsbrun at broadcom dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

            Bug ID: 101905
           Summary: Missed debug information for global register variable
           Product: gcc
           Version: 9.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: debug
          Assignee: unassigned at gcc dot gnu.org
          Reporter: yuri.karlsbrun at broadcom dot com
  Target Milestone: ---

file evi.c:
register void* regVar asm ("r15");
int
main()
{
  char *str = "Hello";
  regVar = str;
  return 0;
}

% gcc -v
Using built-in specs.
COLLECT_GCC=/tools/oss/packages/x86_64-centos7/gcc/9.3.0/bin/gcc
COLLECT_LTO_WRAPPER=/tools/oss/packages/x86_64-rhel7/gcc/9.3.0/bin/../libexec/gcc/x86_64-pc-linux-gnu/9.3.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../../gcc-9.3.0/configure
--prefix=/tools/oss/packages/x86_64-centos7/gcc/9.3.0 --with-gnu-as
--with-as=/tools/bin/as --with-gnu-ld --with-ld=/tools/bin/ld
--with-mpc=/tools/oss/packages/x86_64-centos7/mpc/1.0.3
--with-mpfr=/tools/oss/packages/x86_64-centos7/mpfr/3.1.2
--enable-languages=c,c++,objc --enable-symvers=gnu
Thread model: posix
gcc version 9.3.0 (GCC) 
% gcc -g3 evi.c -o evi
% gdb evi
(gdb) list
1 register void* regVar asm ("r15");
2 int
3 main()
4 {
5  char *str = "Hello";
6  regVar = str;
7  return 0;
8 }
(gdb) b 7
Breakpoint 1 at 0x400502: file evi.c, line 7.
(gdb) r
Starting program: evi
Breakpoint 1, main () at evi.c:7
7  return 0;
(gdb) p regVar
Missing ELF symbol "regVar".
(gdb) p str
$1 = 0x400594 "Hello"
(gdb) p/x $r15
$2 = 0x400594
(gdb) p (char *)$r15
$3 = 0x400594 "Hello"

I tried it for ARM architecture as well with the same result. 
Thanks,
    Yuri

[Bug debug/101905] Missed debug information for global register variable

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |rguenth at gcc dot gnu.org
   Last reconfirmed|                            |2021-08-16
           Keywords|                            |wrong-debug
      Known to fail|                            |12.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
We get

 <1><2d>: Abbrev Number: 2 (DW_TAG_variable)
    <2e>   DW_AT_name        : (indirect string, offset: 0x39): regVar
    <32>   DW_AT_decl_file   : 1
    <33>   DW_AT_decl_line   : 1
    <34>   DW_AT_decl_column : 16
    <35>   DW_AT_type        : <0x39>
    <39>   DW_AT_external    : 1

that's obviously a bit off.  Not sure if there's a good DWARF representation
for global reg vars.

[Bug debug/101905] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I guess it should have DW_AT_location with DW_OP_reg15
Whether DW_AT_external or not is a question, not sure if consumers will be
happy to see a definition of the var in many TUs, but technically it is a
definition.

[Bug debug/101905] Missed debug information for global register variable

[yuri.karlsbrun at broadcom dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #3 from Yuri Karlsbrun <yuri.karlsbrun at broadcom dot com> ---
With the gcc-5.3 correct debug info for global register variable is written,
and variable is visible to gdb

gcc -v
Using built-in specs.
COLLECT_GCC=/tools/oss/packages/x86_64-centos7/gcc/5.3.0/bin/gcc
COLLECT_LTO_WRAPPER=/tools/oss/packages/x86_64-rhel7/gcc/5.3.0/bin/../libexec/gcc/x86_64-unknown-linux-gnu/5.3.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../../gcc-5.3.0/configure
--prefix=/tools/oss/packages/x86_64-centos7/gcc/5.3.0 --with-gnu-as
--with-as=/tools/oss/packages/x86_64-centos7/binutils/2.25/bin/as --with-gnu-ld
--with-ld=/tools/oss/packages/x86_64-centos7/binutils/2.25/bin/ld
--with-mpc=/tools/oss/packages/x86_64-centos7/mpc/1.0.3
--with-mpfr=/tools/oss/packages/x86_64-centos7/mpfr/3.1.2
--enable-languages=c,c++,objc,java,fortran --enable-symvers=gnu
Thread model: posix
gcc version 5.3.0 (GCC)


< 1><0x00000075>    DW_TAG_variable
                      DW_AT_name                  regVar
                      DW_AT_decl_file             0x00000001
/projects/play/ev.c
                      DW_AT_decl_line             0x00000001
                      DW_AT_type                  <0x00000082>
                      DW_AT_external              yes(1)
                      DW_AT_location              len 0x0001: 5f: DW_OP_reg15

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Missed debug information    |[9/10/11/12 Regression]
                   |for global register         |Missed debug information
                   |variable                    |for global register
                   |                            |variable
   Target Milestone|---                         |9.5

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r6-2867-g5fc6ae7dab2809a4607de11b2c4e48752176a562

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
assemble_decl isn't even called for DECL_HARD_REGISTER global vars:
      if (DECL_HARD_REGISTER (node->decl)
          || DECL_HAS_VALUE_EXPR_P (node->decl))
        continue;
      if (node->definition)
        changed |= node->assemble_decl ();
      else
        assemble_undefined_decl (node->decl);
and even if it would be, it wouldn't get into the spot that calls the debug
hook:
  /* Hard register vars do not need to be output.  */
  if (DECL_HARD_REGISTER (decl))
    return false;
...
      debug_hooks->late_global_decl (decl);
      return true;
So, either we should make sure that those vars get DW_AT_location already
during early debug (IMHO best, as the location should be known then already),
or we need to call late_global_decl on those somewhere else.

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
--- gcc/dwarf2out.c.jj  2021-08-17 09:25:08.196923026 +0200
+++ gcc/dwarf2out.c     2021-08-18 23:18:56.284227245 +0200
@@ -24477,7 +24477,9 @@ gen_variable_die (tree decl, tree origin
     {
       if (early_dwarf)
        add_pubname (decl_or_origin, var_die);
-      else
+      if (!early_dwarf
+         || (DECL_HARD_REGISTER (decl_or_origin)
+             && TREE_STATIC (decl_or_origin)))
        add_location_or_const_value_attribute (var_die, decl_or_origin,
                                               decl == NULL);
     }
isn't sufficient because add_location_or_const_value_attribute has an early
exit for early_dwarf...

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think adding the location early is most sensible in this case.  As you saw
that might need tickling down adjustments or maybe a special-case of
DECL_REGISTER globals, bypassing most of the
add_location_or_const_value_attribute magic.

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 51327
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51327&action=edit
gcc12-pr101905.patch

So like this (untested except for the testcase)?

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #9 from rguenther at suse dot de <rguenther at suse dot de> ---
On Fri, 20 Aug 2021, jakub at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905
> 
> Jakub Jelinek <jakub at gcc dot gnu.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>              Status|NEW                         |ASSIGNED
>            Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
> 
> --- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
> Created attachment 51327
>   --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51327&action=edit
> gcc12-pr101905.patch
> 
> So like this (untested except for the testcase)?

Yes, exactly.

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:b284053bb75661fc1bf13c275f3ba5364bb17608

commit r12-3069-gb284053bb75661fc1bf13c275f3ba5364bb17608
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Mon Aug 23 11:50:14 2021 +0200

    dwarf2out: Emit DW_AT_location for global register vars during early dwarf
[PR101905]

    The following patch emits DW_AT_location for global register variables
    already during early dwarf, since usually late_global_decl hook isn't even
    called for those, as nothing needs to be emitted for them.

    2021-08-23  Jakub Jelinek  <jakub@redhat.com>

            PR debug/101905
            * dwarf2out.c (gen_variable_die): Add DW_AT_location for global
            register variables already during early_dwarf if possible.

            * gcc.dg/guality/pr101905.c: New test.

[Bug debug/101905] [9/10/11/12 Regression] Missed debug information for global register variable

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #11 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:513c543bb5d97cc59ba393f2f0612324a789c50e

commit r11-8893-g513c543bb5d97cc59ba393f2f0612324a789c50e
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Mon Aug 23 11:50:14 2021 +0200

    dwarf2out: Emit DW_AT_location for global register vars during early dwarf
[PR101905]

    The following patch emits DW_AT_location for global register variables
    already during early dwarf, since usually late_global_decl hook isn't even
    called for those, as nothing needs to be emitted for them.

    2021-08-23  Jakub Jelinek  <jakub@redhat.com>

            PR debug/101905
            * dwarf2out.c (gen_variable_die): Add DW_AT_location for global
            register variables already during early_dwarf if possible.

            * gcc.dg/guality/pr101905.c: New test.

    (cherry picked from commit b284053bb75661fc1bf13c275f3ba5364bb17608)

[Bug debug/101905] [9/10 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[9/10/11/12 Regression]     |[9/10 Regression] Missed
                   |Missed debug information    |debug information for
                   |for global register         |global register variable
                   |variable                    |

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for 11.3+ and 12.1+ for now.

[Bug debug/101905] [9/10 Regression] Missed debug information for global register variable

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #13 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:059115826b61c6fb737fc6082a7ce5d4c725c503

commit r10-10640-g059115826b61c6fb737fc6082a7ce5d4c725c503
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Mon Aug 23 11:50:14 2021 +0200

    dwarf2out: Emit DW_AT_location for global register vars during early dwarf
[PR101905]

    The following patch emits DW_AT_location for global register variables
    already during early dwarf, since usually late_global_decl hook isn't even
    called for those, as nothing needs to be emitted for them.

    2021-08-23  Jakub Jelinek  <jakub@redhat.com>

            PR debug/101905
            * dwarf2out.c (gen_variable_die): Add DW_AT_location for global
            register variables already during early_dwarf if possible.

            * gcc.dg/guality/pr101905.c: New test.

    (cherry picked from commit b284053bb75661fc1bf13c275f3ba5364bb17608)

[Bug debug/101905] [9/10 Regression] Missed debug information for global register variable

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

--- Comment #14 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-9 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:6aa2e07b4134bceefe0b8be5cecfe11119466181

commit r9-10097-g6aa2e07b4134bceefe0b8be5cecfe11119466181
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Mon Aug 23 11:50:14 2021 +0200

    dwarf2out: Emit DW_AT_location for global register vars during early dwarf
[PR101905]

    The following patch emits DW_AT_location for global register variables
    already during early dwarf, since usually late_global_decl hook isn't even
    called for those, as nothing needs to be emitted for them.

    2021-08-23  Jakub Jelinek  <jakub@redhat.com>

            PR debug/101905
            * dwarf2out.c (gen_variable_die): Add DW_AT_location for global
            register variables already during early_dwarf if possible.

            * gcc.dg/guality/pr101905.c: New test.

    (cherry picked from commit b284053bb75661fc1bf13c275f3ba5364bb17608)

[Bug debug/101905] [9/10 Regression] Missed debug information for global register variable

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101905

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.