GDB TLS testing [actually running some tests finally: a success with -pthread used]

[John Baldwin]

On 1/23/19 12:19 AM, Mark Millard wrote:
> 
> 
> On 2019-Jan-22, at 22:53, Mark Millard <marklmi at yahoo.com> wrote:
> 
> 
> 
>> On 2019-Jan-22, at 19:19, Mark Millard <marklmi at yahoo.com> wrote:
>>
>>
>>
>>> On 2019-Jan-22, at 18:32, Mark Millard <marklm at yahoo.com> wrote:
>>>
>>>
>>>
>>>> On 2019-Jan-22, at 17:06, Mark Millard <marklmi at yahoo.com> wrote:
>>>>
>>>>
>>>>
>>>> . . .
>>>> So I'm trying:
>>>>
>>>> # git clean -f
>>>> # rm */config.cache */*/config.cache
>>>> # env CPATH=/usr/local/include ./configure
>>>> . . .
>>>> # env CPATH=/usr/local/include gmake
>>>> . . .
>>>>
>>>> in order to try to add paths after the command line -I paths.
>>>>
>>>> . . . This looks like it built. I've not used the build yet.
>>>>
>>>
>>> Looking at a *.core did not go well for my context:
>>>
>>> # cc -g -O2 tls_gdb_test.c 
>>> tls_gdb_test.c:16:2: warning: indirection of non-volatile null pointer will be deleted, not trap [-Wnull-dereference]
>>>       *(char *)NULL = 1;
>>>       ^~~~~~~~~~~~~
>>> tls_gdb_test.c:16:2: note: consider using __builtin_trap() or qualifying pointer with 'volatile'
>>> 1 warning generated.

For me, builtin_trap didn't DTRT on ppc before, but that was probably using
gcc rather than clang (and possibly using gcc4.2).

>>> . . .
>>
>>
>> So far all tried-combinations of using gcc versions for build
>> the test program and/or building the gdb used do not work for
>> "p id" and "p &id" in doing the test. clang is not essential
>> to the behavior observed.
> 
> Using "-g -O2 -pthread" to build the test program via system
> clang or gcc8 (for example) did lead to the likes of:
> 
> (gdb) run 
> Starting program: /root/c_tests/a.out 
> main: PID 15350
> id = 15350 (0x810055020)
> 
> ^C
> Program received signal SIGINT, Interrupt.
> main (ac=<optimized out>, av=<optimized out>) at tls_gdb_test.c:16
> 16		*(char *)NULL = 1;
> (gdb) bt
> #0  main (ac=<optimized out>, av=<optimized out>) at tls_gdb_test.c:16
> (gdb) info threads
>   Id   Target Id                   Frame 
> * 1    LWP 100324 of process 15350 main (ac=<optimized out>, av=<optimized out>) at tls_gdb_test.c:16
> (gdb) p id
> $1 = 15350
> (gdb) p &id
> $2 = (int *) 0x810055020
> 
> So it appears one branch of:
> 
> static void
> fbsd_fetch_rtld_offsets (struct gdbarch *gdbarch, struct fbsd_pspace_data *data)
> {
>   TRY
>     {
>       /* Fetch offsets from debug symbols in rtld.  */
>       data->off_linkmap = parse_and_eval_long ("&((Obj_Entry *)0)->linkmap");
>       data->off_tlsindex = parse_and_eval_long ("&((Obj_Entry *)0)->tlsindex");
>       data->rtld_offsets_valid = true;
>       return;
>     }
>   CATCH (e, RETURN_MASK_ERROR)
>     {
>       data->off_linkmap = -1;
>     }
>   END_CATCH
> 
>   TRY
>     {
>       /* Fetch offsets from global variables in libthr.  Note that
>          this does not work for single-threaded processes that are not
>          linked against libthr.  */
>       data->off_linkmap = fbsd_read_integer_by_name(gdbarch,
>                                                     "_thread_off_linkmap");
>       data->off_tlsindex = fbsd_read_integer_by_name(gdbarch,
>                                                      "_thread_off_tlsindex");
>       data->rtld_offsets_valid = true;
>       return;
>     }
>   CATCH (e, RETURN_MASK_ERROR)
>     {
>       data->off_linkmap = -1;
>     }
>   END_CATCH
> }
> 
> is working when -pthread is used.

Yes.  The second one only works for programs linked against -lthr.  Otherwise
you need to have built your system with debug symbols (which is the default),
and gdb needs to be able to access /usr/lib/debug/libexec/ld-elf.so.1.debug
to determine the offsets of the two fields in Obj_Entry (this is what the
first TRY clause does).

-- 
John Baldwin