Re: [rfc, rfa/doc] Multi-threaded watchpoint improvements

[Luis Machado <luisgpm@linux.vnet.ibm.com>]

Ok,

This is the event log without the patch applied. I can see that it does
have additional creation events occuring in the background, different
than what occurs with the patch. Same source code, same compile flags.

== code ==
(gdb) set debug event 1
(gdb) start
breakpoint_create_event
Breakpoint 1 at 0x10000738: file testsuite/gdb.base/watch-thread_num.c,
line 39.
Starting program: /home/luis/src/gdb/gdb-head/git/gdb/test
[Thread debugging using libthread_db enabled]
[New Thread 0x400000257c0 (LWP 4236)]
[Switching to Thread 0x400000257c0 (LWP 4236)]
breakpoint_delete_event
main () at testsuite/gdb.base/watch-thread_num.c:39
39          for (i = 0; i < NUM; i++)
(gdb) watch i
During symbol reading, incomplete CFI data; unspecified registers (e.g.,
r0) at 0x10000738.
breakpoint_create_event
Hardware watchpoint 2: i
(gdb) n
41              res = pthread_create(&threads[i],
(gdb)
[New Thread 0x40000a26240 (LWP 4239)]
breakpoint_delete_event
39          for (i = 0; i < NUM; i++)
(gdb)
breakpoint_create_event
Hardware watchpoint 2: i

Old value = 0
New value = 1
0x0000000010000788 in main () at
testsuite/gdb.base/watch-thread_num.c:39
39          for (i = 0; i < NUM; i++)
(gdb)
41              res = pthread_create(&threads[i],
(gdb)
[New Thread 0x40001226240 (LWP 4240)]
breakpoint_delete_event
39          for (i = 0; i < NUM; i++)
(gdb)
breakpoint_create_event
Hardware watchpoint 2: i

Old value = 1
New value = 2
0x0000000010000788 in main () at
testsuite/gdb.base/watch-thread_num.c:39
39          for (i = 0; i < NUM; i++)
(gdb) n
41              res = pthread_create(&threads[i],
(gdb)
[New Thread 0x40001a26240 (LWP 4241)]
breakpoint_delete_event
39          for (i = 0; i < NUM; i++)
(gdb)
breakpoint_create_event
Hardware watchpoint 2: i

Old value = 2
New value = 3
0x0000000010000788 in main () at
testsuite/gdb.base/watch-thread_num.c:39
39          for (i = 0; i < NUM; i++)
(gdb) 


On Wed, 2007-10-24 at 12:47 -0400, Daniel Jacobowitz wrote:
> On Wed, Oct 24, 2007 at 02:22:00PM -0300, Luis Machado wrote:
> > Hi Daniel,
> > 
> > This is what i get while trying to use a hardware watchpoint to monitor
> > an int variable "i" that changes during a for statement (attached). Upon
> > further investigation, it seems GDB is only calling
> > "breakpoint_delete_event". I don't see GDB calling
> > "breakpoint_create_event" anywhere except in the momment i created the
> > watchpoint up there, so i presume that's the reason it's not triggering.
> 
> That's the only time it is supposed to be called; it's the event at
> the creation of a breakpoint.  THe user creates it, it lives until it
> is deleted.
> 
> > (gdb) watch i
> > During symbol reading, incomplete CFI data; unspecified registers (e.g.,
> > r0) at 0x10000738.
> > breakpoint_create_event (CREATED!)
> > Hardware watchpoint 2: i
> > (gdb) n
> > 41              res = pthread_create(&threads[i],
> > (gdb)
> > [New Thread 0x40000a26240 (LWP 3290)]
> > breakpoint_delete_event (DELETED! After this point, no more creations.)
> 
> GDB has stopped for some reason and checked whether the watchpoint on
> i is still in scope.  It thinks it isn't.  This may be an unwinder
> problem.  Since the scope is gone, the watchpoint is deleted.
> 
> When you have a problem with watchpoints it's useful to remember that
> watchpoints on local variables are by far the most complicated kind.
> See if they still work on a global variable or a memory location.
> 
-- 
Luis Machado
IBM Linux Technology Center
e-mail: luisgpm@linux.vnet.ibm.com