Potential problem between non-stop and linux-thread-db

Potential problem between non-stop and linux-thread-db

[Michael Snyder]

Just something I noticed just now...

linux-thread-db depends on callback functions provided by 
gdb_proc_service.c.  Some of those callbacks are for instance

/* Stop the lightweight process LWPID within the target process PH.  */

ps_err_e
ps_lstop (gdb_ps_prochandle_t ph, lwpid_t lwpid)
{
   /* All lightweight processes are stopped when under control of GDB.  */
   return PS_OK;
}

/* Resume the lightweight process (LWP) LWPID within the target
    process PH.  */

ps_err_e
ps_lcontinue (gdb_ps_prochandle_t ph, lwpid_t lwpid)
{
   /* Pretend we did successfully continue LWPID.  GDB will take care
      of it later on.  */
   return PS_OK;
}

These were written under the "all-stop" convention, but the comments
are no longer true, is it not so?

So, async/nonstop guys -- do we need to do anything about this?

Michael

Re: Potential problem between non-stop and linux-thread-db

[Pedro Alves]

On Tuesday 06 October 2009 01:42:24, Michael Snyder wrote:

> These were written under the "all-stop" convention, but the comments
> are no longer true, is it not so?
> 
> So, async/nonstop guys -- do we need to do anything about this?

AFAIK, these callbacks weren't really used by glibc.  Has that
changed recently?

-- 
Pedro Alves

Re: Potential problem between non-stop and linux-thread-db

[Michael Snyder]

Pedro Alves wrote:
> On Tuesday 06 October 2009 01:42:24, Michael Snyder wrote:
> 
>> These were written under the "all-stop" convention, but the comments
>> are no longer true, is it not so?
>>
>> So, async/nonstop guys -- do we need to do anything about this?
> 
> AFAIK, these callbacks weren't really used by glibc.  Has that
> changed recently?

Probably not.  I was just looking at the code, not actually
running or debugging it.

Re: Potential problem between non-stop and linux-thread-db

[Pedro Alves]

On Tuesday 06 October 2009 18:30:58, Michael Snyder wrote:
> Pedro Alves wrote:
> > On Tuesday 06 October 2009 01:42:24, Michael Snyder wrote:
> > 
> > AFAIK, these callbacks weren't really used by glibc.  Has that
> > changed recently?
> 
> Probably not.  I was just looking at the code, not actually
> running or debugging it.

Feel free to put a comment there, or let me know if you'd
like me to.  In any case, in non-stop mode, at least the current
lwp is always stopped already when we call into thread_db.

It's likely that solaris' thread_db does make use of such
callbacks.  I didn't check, but well, they've invented the
interface.  We never did merge sol-threads.c and proc-service.c
though, and solaris doesn't support non-stop presently anyway.

-- 
Pedro Alves