Re: Wrong mappin of registers in location lists

[Daniel Berlin <dberlin at dberlin dot org>]

On Fri, 14 Feb 2003, Michal Ludvig wrote:

> Hi again,
> as far as I can see gcc outputs wrong information to .debug_loc on
> x86-64. This is the assembler output:
>
> main:
> .LVL22:
>          pushq   %rbx
>          movl    %edi, %ebx
> .LVL23:
>          leal    1(%rbx), %edi
>          subq    $416, %rsp
> [...]
>
> This is DIE of argc:
>          .uleb128 0x3    # (DIE (0xef) DW_TAG_formal_parameter)
>          .long   .LASF4  # DW_AT_name: "argc"
>          .byte   0x1     # DW_AT_decl_file
>          .byte   0x17    # DW_AT_decl_line
>          .long   0x8f    # DW_AT_type
>          .long   .LLST3-.Ldebug_loc0     # DW_AT_location
>
> And finally location list of argc:
> .LLST3:
>          .quad   .LVL22-.text    # Location list begin address (*.LLST3)
>          .quad   .LVL23-.text    # Location list end address (*.LLST3)
>          .value  0x1     # Location expression size
>          .byte   0x55    # DW_OP_reg5
>          .quad   .LVL23-.text    # Location list begin address (*.LLST3)
>          .quad   .LFE7-.text     # Location list end address (*.LLST3)
>          .value  0x1     # Location expression size
>          .byte   0x53    # DW_OP_reg3
>          .quad   0x0     # Location list terminator begin (*.LLST3)
>          .quad   0x0     # Location list terminator end (*.LLST3)
>
> At the entry of the function (.LVL22), argc is dwarf2 register 5
> (DW_OP_reg5), ie %rdi on x86-64. From .LVL23 on it is saved in %rbx, ie.
> DW_OP_reg1. However debug information says it's in DW_OP_reg3, ie. %rdx.

We just do what the NOTE_VAR_LOCATION's tell us, so it's got to be Josef's
or the dwarf2 numbering.

>
> Something is apparantly wrong - just don't know if it's wrong on Josef's
>   or Daniel's side. Or perhaps somewhere else...
>
> Michal Ludvig
> --
> * SuSE CR, s.r.o     * mludvig@suse.cz
> * (+420) 296.545.373 * http://www.suse.cz
>
>