[Bug breakpoints/26096] gdb sets breakpoint at cold clone

["vries at gcc dot gnu.org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=26096

--- Comment #3 from Tom de Vries <vries at gcc dot gnu.org> ---
Another thing I noticed with cold clones:
...
$ gdb -q -batch a.out -ex "maint print psymbols" | egrep "foo|bar"
    `bar', function, 0x400455
    `foo', function, 0x400450
...

The psymbol address of the function bar is the address of "bar() [clone
.cold]".

Presumably we'd want the entry address here.

Looking at the corresponding DWARF, we have:
...
 <1><990>: Abbrev Number: 35 (DW_TAG_subprogram)
    <991>   DW_AT_name        : foo
    <995>   DW_AT_decl_file   : 1
    <996>   DW_AT_decl_line   : 8
    <997>   DW_AT_decl_column : 1
    <998>   DW_AT_type        : <0x39b>
    <99c>   DW_AT_ranges      : 0x40
    <9a0>   DW_AT_frame_base  : 1 byte block: 9c        (DW_OP_call_frame_cfa)
    <9a2>   DW_AT_GNU_all_call_sites: 1
    <9a2>   DW_AT_sibling     : <0x9b4>
...

In DWARF 4 standard 2.18 Entry Address we read:
...
If no DW_AT_entry_pc attribute is present, then the entry address is assumed to
be the same as the value of the DW_AT_low_pc attribute, if present; otherwise,
the entry address is unknown.
...

In this case, we have no DW_AT_entry_pc and no DW_AT_low_pc attribute. 
Arguably, this is a gcc bug.

Either way, in absence of a fix in gcc for this, we could adapt a
interpretation that the entry pc is the start address of the first range. At
least, this works for this exec:
...
    00000070 00000000004005b0 00000000004005d7
    00000070 0000000000400455 000000000040045a
    00000070 <End of list>
...

This also seems to be the current interpretation for full symtabs, AFAIU from
the comment for BLOCK_ENTRY_PC in block.h:
...
/* Define the "entry pc" for a block BL to be the lowest (start) address        
   for the block when all addresses within the block are contiguous.  If        
   non-contiguous, then use the start address for the first range in the        
   block.                                                                       
...

-- 
You are receiving this mail because:
You are on the CC list for the bug.