This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA] Patch to fix reverse-debug recursion function tail bug


Hui Zhu wrote:
PING

Thanks for the reminder.


I added some comment and changed the order of evaluation a bit,
hoping to reduce the performance impact on normal debugging.
And I ran the testsuites, before and after.

Modified patch is attached -- is this OK with you guys?
Mark, can you confirm that it fixes your original bug?

Thx again,
Michael


On Mon, May 11, 2009 at 15:07, Hui Zhu<teawater@gmail.com> wrote:
PING

On Wed, May 6, 2009 at 15:23, Hui Zhu <teawater@gmail.com> wrote:
Hi Michael,

Like the prev patch I send to you, this issue still affect cvs-head
and the patch can fix it.
Please help me review it.

The attachment is the new patch follow cvs-head.

2009-05-06 Hui Zhu <teawater@gmail.com>

      * infrun.c (handle_inferior_event): Check frame_id when
      check range in reverse debug mode.

Thanks,
Hui

On Sat, Mar 21, 2009 at 16:52, Hui Zhu <teawater@gmail.com> wrote:
Hi,

This patch is for bug report by Marc in
http://sourceware.org/ml/gdb/2009-03/msg00127.html.

This bug in "handle_inferior_event" deal with recursion function tail
in reverse debug.
infrun: infwait_normal_state
infrun: TARGET_WAITKIND_STOPPED
infrun: stop_pc = 0x8048457
infrun: stepping inside range [0x8048457-0x804845a]
infrun: stop_stepping
factorial (x=4) at b.cc:5

Inferior already step into another frame. But because this is a
recursion function call, And 0x8048457 is in
ecs->event_thread->step_range_start and
ecs->event_thread->step_range_start.

So gdb run in:

if (stop_pc >= ecs->event_thread->step_range_start
     && stop_pc < ecs->event_thread->step_range_end)
   {

This code is in front of:
 if (!frame_id_eq (get_frame_id (get_current_frame ()),
                   ecs->event_thread->step_frame_id)
     && (frame_id_eq (frame_unwind_id (get_current_frame ()),
                      ecs->event_thread->step_frame_id)
         || execution_direction == EXEC_REVERSE))

So gdb check range without check frame_id.

So I make a patch to check frame_id when check range in reverse debug mode.

2008-03-21 Hui Zhu <teawater@gmail.com>

       * infrun.c (handle_inferior_event): Check frame_id when
       check range in reverse debug mode.





Actually, there is another thing, when gdb begin reverse-debug, it's range is:
 8048439:       8b 45 08                mov    0x8(%ebp),%eax
 804843c:       83 e8 01                sub    $0x1,%eax
 804843f:       89 04 24                mov    %eax,(%esp)
 8048442:       e8 dd ff ff ff          call   8048424 <_Z9factoriali>
 8048447:       0f af 45 08             imul   0x8(%ebp),%eax
 804844b:       89 45 fc                mov    %eax,-0x4(%ebp)
Why is changed to infrun: stepping inside range [0x8048457-0x804845a]?
That is because when inferior step at:
 8048458:       c3                      ret
In this address, $ebp is same with high level function and this
function is factorial too.
So the gdb can't found inferior step into another frame.  It will run to:
 ecs->event_thread->step_range_start = stop_pc_sal.pc;
 ecs->event_thread->step_range_end = stop_pc_sal.end;
 ecs->event_thread->step_frame_id = get_frame_id (get_current_frame ());
 ecs->event_thread->current_line = stop_pc_sal.line;
 ecs->event_thread->current_symtab = stop_pc_sal.symtab;

 if (debug_infrun)
    fprintf_unfiltered (gdb_stdlog, "infrun: keep going\n");
 keep_going (ecs);
}
So ecs->event_thread->step_range_start and ecs->event_thread->step_range_end.

I don't find that it affect the reverse debug or something. So I didn't fix it.




Thanks, Hui




2009-06-14  Hui Zhu  <teawater@gmail.com>
	    Michael Snyder  <msnyder@vmware.com>

	* infrun.c (handle_inferior_event): Improve reverse stepping
	through function epilogue.

Index: infrun.c
===================================================================
RCS file: /cvs/src/src/gdb/infrun.c,v
retrieving revision 1.387
diff -u -p -r1.387 infrun.c
--- infrun.c	11 Jun 2009 11:57:46 -0000	1.387
+++ infrun.c	15 Jun 2009 00:45:17 -0000
@@ -3623,9 +3623,17 @@ infrun: not switching back to stepped th
 
      Note that step_range_end is the address of the first instruction
      beyond the step range, and NOT the address of the last instruction
-     within it! */
+     within it!
+
+     Note also that during reverse execution, we may be stepping
+     through a function epilogue and therefore must detect when
+     the current-frame changes in the middle of a line.  */
+
   if (stop_pc >= ecs->event_thread->step_range_start
-      && stop_pc < ecs->event_thread->step_range_end)
+      && stop_pc < ecs->event_thread->step_range_end
+      && (execution_direction != EXEC_REVERSE
+	  || frame_id_eq (get_frame_id (get_current_frame ()),
+			  ecs->event_thread->step_frame_id)))
     {
       if (debug_infrun)
 	fprintf_unfiltered (gdb_stdlog, "infrun: stepping inside range [0x%s-0x%s]\n",

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]