RFC: Re: [patch] Fix for 'info threads' crashes if zombie threads exist

[Jan Kratochvil]

Hi Daniel,

On Tue, 20 Jun 2006 19:11:09 +0200, Daniel Jacobowitz wrote:
...
> TD_DEATH events were supported because there are all sorts of things
> which can go wrong when you ask libthread_db about a thread that it
> considers dead.  At that point, as far as the library is concerned, the
> thread is gone.
> After your patch, I strongly suspect there are places where you could hit
> control-c and get mysterious errors from GDB.

OK...  I checked now that my patch may have problems accessing TCB after:
	if (IS_DETACHED (pd)) __free_tcb (pd);
for the detached threads and it even has problems before this point (not
analysed why):
	Program received signal SIGTRAP, Trace/breakpoint trap.
	[Switching to Thread -1208153184 (unknown thread_db state 1)]
	0x00000000 in ?? ()

Proposing the attached reduced patch with only the most important+safe part.
It still catches the initial terminating state with EXITING_BIT
(->TD_THR_ZOMBIE) where I hope most of the crashes/breakpoints may occur.

Still the perfect functionality would require patching libthread_db and it
looks to me a bit as a chicken&egg problem. :-)


> A new thread can be created with the same thread ID - even before this one
> exits.

Not sure of how much are non-Linux platforms a concern for these minor issues.
Apparently on Linux kernel the same LWP id cannot be created until the final
syscall __NR_exit.


Regards,
Jan Kratochvil


> A little patience, please.  It takes a while to review GDB patches,
> especially non-obvious ones - and thread-db support seems to be quite
> complicated.

(Sorry, I only did not want to get the patch lost.)
-------------- next part --------------
Index: linux-thread-db.c
===================================================================
RCS file: /cvs/src/src/gdb/linux-thread-db.c,v
retrieving revision 1.16
diff -u -p -r1.16 linux-thread-db.c
--- linux-thread-db.c	5 May 2006 22:42:43 -0000	1.16
+++ linux-thread-db.c	20 Jun 2006 17:51:01 -0000
@@ -1006,6 +1006,9 @@ thread_db_fetch_registers (int regno)
     }
 
   thread_info = find_thread_pid (inferior_ptid);
+  if (!thread_info)
+    error (_("Stopped at dead thread %ld; should not happen."),
+	   (long) GET_THREAD (inferior_ptid));
   thread_db_map_id2thr (thread_info, 1);
 
   err = td_thr_getgregs_p (&thread_info->private->th, gregset);
@@ -1127,6 +1130,8 @@ thread_db_thread_alive (ptid_t ptid)
       struct thread_info *thread_info;
       thread_info = find_thread_pid (ptid);
 
+      if (!thread_info)
+	return 0;
       thread_db_map_id2thr (thread_info, 0);
       if (!thread_info->private->th_valid)
 	return 0;
@@ -1145,9 +1150,9 @@ thread_db_thread_alive (ptid_t ptid)
 	  thread_info->private->ti_valid = 1;
 	}
 
-      if (thread_info->private->ti.ti_state == TD_THR_UNKNOWN
-	  || thread_info->private->ti.ti_state == TD_THR_ZOMBIE)
-	return 0;		/* A zombie thread.  */
+      /* Never 0 on TD_THR_ZOMBIE to be able to trace pthread_exit(3).  */
+      if (thread_info->private->ti.ti_state == TD_THR_UNKNOWN)
+	return 0;		/* A disappeared thread.  */
 
       return 1;
     }