Re: [BZ#17090/17620/17621]: fix DTV race, assert, and DTV_SURPLUS Static TLS limit

[Pedro Alves <palves at redhat dot com>]

On 11/20/2014 02:17 AM, Roland McGrath wrote:
> In actual fact, GDB ignores the argument entirely and just does a vanilla
> symbol lookup like you'd get if you said "p &symbol".  That's why it works
> at all now for statically-linked libpthread, where the main executable is
> where the symbols are actually found.  

Indeed.  I once tried to change GDB to actually use the passed in argument,
but found that broke the statically-linked libpthread case.

> That said, since the interface has
> the parameter, I'd still be in favor of passing in the correct object name
> just pro forma (since it's really not hard to implement).

Note that's really not sufficient for getting the statically-linked case
fully right.  Some of the libpthread symbols thread_db wants to looks up,
although static, have names that aren't in the implementation namespace.  As
such, if the statically-linked program happens to have other symbols with the
same name, GDB might well find and return those to nptl_db instead.

For example, it should be perfectly legitimate for a program to define a
symbol called "stack_used", but that ends up confusing nptl_db/gdb,
because that's one of the static libpthread symbols nptl_db looks up:

 /* List of the stacks in use.  */
 static LIST_HEAD (stack_used);

And that's exactly what happened in:

 [Bug 9635 - Cannot find new threads: generic error]
 https://sourceware.org/bugzilla/show_bug.cgi?id=9635

Actually looks like stack_used may be the only one with the problem:

 $ grep -rn DB_GET_SYMBOL -rn
 nptl_db/td_ta_thr_iter.c:157:  err = DB_GET_SYMBOL (list, ta, __stack_user);
 nptl_db/td_ta_thr_iter.c:164:    err = DB_GET_SYMBOL (list, ta, stack_used);
 nptl_db/td_ta_tsd_iter.c:53:  err = DB_GET_SYMBOL (addr, ta, __pthread_keys);
 nptl_db/td_ta_event_addr.c:40:      err = DB_GET_SYMBOL (taddr, ta, __nptl_create_event);
 nptl_db/td_ta_event_addr.c:44:      err = DB_GET_SYMBOL (taddr, ta, __nptl_death_event);
 nptl_db/td_thr_validate.c:65:  err = DB_GET_SYMBOL (list, th->th_ta_p, __stack_user);
 nptl_db/td_thr_validate.c:73:      err = DB_GET_SYMBOL (list, th->th_ta_p, stack_used);
 nptl_db/td_ta_map_lwp2thr.c:190:  td_err_e err = DB_GET_SYMBOL (list, ta, __stack_user);
 nptl_db/td_ta_set_event.c:41:  err = DB_GET_SYMBOL (eventmask, ta, __nptl_threads_events);
 nptl_db/thread_dbP.h:150:#define DB_GET_SYMBOL(var, ta, name)                                         \
 nptl_db/td_thr_event_getmsg.c:71:  err = DB_GET_SYMBOL (prevp, th->th_ta_p, __nptl_last_event);
 nptl_db/td_ta_clear_event.c:41:  err = DB_GET_SYMBOL (eventmask, ta, __nptl_threads_events);

Though renaming __stack_user to __nptl_stack_user too might be good, in case
other parts of the implementation (e.g., libgcc) might end up defining their
own unrelated static symbols with that name too.

Thanks,
Pedro Alves