[PATCH] Fix an issue with the gdb step-over aka. "n" command
Bernd Edlinger
bernd.edlinger@hotmail.de
Sun Dec 15 08:39:00 GMT 2019
On 12/15/19 2:25 AM, Simon Marchi wrote:
> On 2019-11-24 7:17 a.m., Bernd Edlinger wrote:
>> Hi,
>>
>> this fixes an issue with the gdb step-over aka. "n" command.
>>
>> Apologies, the motivation for this patch was from sub-optimal
>> debug experience using some gcc code involving inlined functions,
>> and initially I tried to convince gcc folks that it is in fact a
>> gcc bug, but...
>>
>> It can be seen when you debug an optimized stage-3 cc1
>> it does not affect -O0 code, though.
>>
>> Note: you can use "gcc -S hello.c -wrapper gdb,--args" to invoke cc1 with
>> debugger attached.
>>
>> This example debug session will explain the effect.
>>
>> (gdb) b get_alias_set
>> Breakpoint 5 at 0xa099f0: file ../../gcc-trunk/gcc/alias.c, line 837.
>> (gdb) r
>> Breakpoint 5, get_alias_set (t=t@entry=0x7ffff7ff7ab0) at ../../gcc-trunk/gcc/alias.c:837
>> 837 if (t == error_mark_node
>> (gdb) n
>> 839 && (TREE_TYPE (t) == 0 || TREE_TYPE (t) == error_mark_node)))
>> (gdb) n
>> 3382 return __t; <-- now we have a problem: wrong line info here
>> (gdb) bt
>> #0 get_alias_set (t=t@entry=0x7ffff7ff7ab0) at ../../gcc-trunk/gcc/tree.h:3382
>> #1 0x0000000000b25dfe in set_mem_attributes_minus_bitpos (ref=0x7ffff746f990, t=0x7ffff7ff7ab0, objectp=1, bitpos=...)
>> at ../../gcc-trunk/gcc/emit-rtl.c:1957
>> #2 0x0000000001137a55 in make_decl_rtl (decl=0x7ffff7ff7ab0) at ../../gcc-trunk/gcc/varasm.c:1518
>> #3 0x000000000113b6e8 in assemble_variable (decl=0x7ffff7ff7ab0, top_level=<optimized out>, at_end=<optimized out>,
>> dont_output_data=0) at ../../gcc-trunk/gcc/varasm.c:2246
>> #4 0x000000000113f0ea in varpool_node::assemble_decl (this=0x7ffff745b000) at ../../gcc-trunk/gcc/varpool.c:584
>> #5 0x000000000113fa17 in varpool_node::assemble_decl (this=0x7ffff745b000) at ../../gcc-trunk/gcc/varpool.c:750
>
> I have a hard time understanding what is going wrong and what we should see
> instead. I think it would help if you showed what's $pc at every place where
> you are stopping, as well as the output for readelf --debug-dump=decodedline
> for those regions. It would also help if you provided the same session without
> and with your patch applied, so we could see the difference.
>
Hi Simon,
the issue is there is a line-info from the inline function right at the end of the
inline superblock without any code just variable tacking info there.
Maybe it helps to get the background if you look at my attempt of fixing this as
a gcc bug:
https://gcc.gnu.org/ml/gcc-patches/2019-10/msg01459.html
and Alexandre Oliva's response here:
https://gcc.gnu.org/ml/gcc-patches/2019-11/msg01771.html
So this is a problem in the design of the dwarf line info which
has stmt-type line info at the end of a inlined subroutine.
Those fall outside the inline block, therefore the step over stops
at the line where the inline function ends, and for gdb it just appears
as if this line was in the calling subroutine, which happens since
the line info does not have any connection to the inlined subroutine
range info.
Contents of the .debug_info section:
<2><4f686>: Abbrev Number: 12 (DW_TAG_inlined_subroutine)
<4f687> DW_AT_abstract_origin: <0x53d4e>
<4f68b> DW_AT_entry_pc : 0x7280
<4f693> DW_AT_GNU_entry_view: 1
<4f695> DW_AT_ranges : 0xb480
<4f699> DW_AT_call_file : 8 <- alias.c
<4f69a> DW_AT_call_line : 839
<4f69c> DW_AT_call_column : 8
<4f69d> DW_AT_sibling : <0x4f717>
The File Name Table (offset 0x253):
8 2 0 0 alias.c
10 2 0 0 tree.h
Contents of the .debug_ranges section:
0000b480 0000000000007280 0000000000007291
0000b480 0000000000002764 000000000000277e
0000b480 <End of list>
The problem is at pc=0x7291 in the Line Number Section:
Line Number Statements:
[0x00008826] Special opcode 61: advance Address by 4 to 0x7284 and Line by 0 to 3380
[0x00008827] Set is_stmt to 1
[0x00008828] Special opcode 189: advance Address by 13 to 0x7291 and Line by 2 to 3382 (*)
[0x00008829] Set is_stmt to 0 (**)
[0x0000882a] Copy (view 1)
[0x0000882b] Set File Name to entry 8 in the File Name Table <- back to alias.c
[0x0000882d] Set column to 8
[0x0000882f] Advance Line by -2543 to 839
[0x00008832] Copy (view 2)
[0x00008833] Set column to 27
[0x00008835] Special opcode 61: advance Address by 4 to 0x7295 and Line by 0 to 839
[0x00008836] Set column to 3
[0x00008838] Set is_stmt to 1 <-- next line info counts: alias.c:847
[0x00008839] Special opcode 153: advance Address by 10 to 0x729f and Line by 8 to 847
[0x0000883a] Set column to 7
(*) this line is tree.h:3382, but the program counter is *not* within the subroutine,
but exactly at the first instruction *after* the subroutine according to the debug_ranges.
What makes it worse, is that (**) makes gdb ignore the new location info alias.c:839,
which means, normally the n command would have continued to pc=0x729f, which is at alias.c:847.
What this patch does, is a heuristic, that means when the last satement line number (*)
contains no code, and is followed by a non-statment line number in another file, then
pretend the non-statement (**) was actually a stmt-type line number. By adding the
end of sequence marker here this code in buildsym.c cancels the last line number in the
inline file:
/* Normally, we treat lines as unsorted. But the end of sequence
marker is special. We sort line markers at the same PC by line
number, so end of sequence markers (which have line == 0) appear
first. This is right if the marker ends the previous function,
and there is no padding before the next function. But it is
wrong if the previous line was empty and we are now marking a
switch to a different subfile. We must leave the end of sequence
marker at the end of this group of lines, not sort the empty line
to after the marker. The easiest way to accomplish this is to
delete any empty lines from our table, if they are followed by
end of sequence markers. All we lose is the ability to set
breakpoints at some lines which contain no instructions
anyway. */
if (line == 0 && subfile->line_vector->nitems > 0)
{
e = subfile->line_vector->item + subfile->line_vector->nitems - 1;
while (subfile->line_vector->nitems > 0 && e->pc == pc)
{
e--;
subfile->line_vector->nitems--;
}
}
(That's where I discovered the line number 65535 issue BTW)
Bernd.
More information about the Gdb-patches
mailing list