User breakpoint ignored if software-single-step at same location
authorJoel Brobecker <brobecker@adacore.com>
Tue, 3 Jun 2014 16:42:19 +0000 (17:42 +0100)
committerPedro Alves <palves@redhat.com>
Tue, 3 Jun 2014 16:42:19 +0000 (17:42 +0100)
with the following code...

    12    Nested;   -- break #1
    13    return I; -- break #2
    14  end;

(line 12 is a call to function Nested)

... we have noticed the following errorneous behavior on ppc-aix,
where, after having inserted a breakpoint at line 12 and line 13,
and continuing from the breakpoint at line 12, the program never
stops at line 13, running away until the program terminates:

    % gdb -q func
    (gdb) b func.adb:12
    Breakpoint 1 at 0x10000a24: file func.adb, line 12.
    (gdb) b func.adb:13
    Breakpoint 2 at 0x10000a28: file func.adb, line 13.
    (gdb) run
    Starting program: /[...]/func

    Breakpoint 1, func () at func.adb:12
    12        Nested;   -- break #1
    (gdb) c
    Continuing.
    [Inferior 1 (process 4128872) exited with code 02]

When resuming from the first breakpoint, GDB first tries to step out
of that first breakpoint.  We rely on software single-stepping on this
platform, and it just so happens that the address of the first
software single-step breakpoint is the same as the user's breakpoint
#2 (0x10000a28).  So, with infrun and target traces turned on (but
uninteresting traces snip'ed off), the "continue" operation looks like
this:

    (gdb) c
    ### First, we insert the user breakpoints (the second one is an internal
    ### breakpoint on __pthread_init). The first user breakpoint is not
    ### inserted as we need to step out of it first.
    target_insert_breakpoint (0x0000000010000a28, xxx) = 0
    target_insert_breakpoint (0x00000000d03f3800, xxx) = 0
    ### Then we proceed with the step-out-of-breakpoint...
    infrun: resume (step=1, signal=GDB_SIGNAL_0), trap_expected=1, current thread [process 15335610] at 0x10000a24
    ### That's when we insert the SSS breakpoints...
    target_insert_breakpoint (0x0000000010000a28, xxx) = 0
    target_insert_breakpoint (0x00000000100009ac, xxx) = 0
    ### ... then let the inferior resume...
    target_resume (15335610, continue, 0)
    infrun: wait_for_inferior ()
    target_wait (-1, status, options={}) = 15335610,   status->kind = stopped, signal = GDB_SIGNAL_TRAP
    infrun: target_wait (-1, status) =
    infrun:   15335610 [process 15335610],
    infrun:   status->kind = stopped, signal = GDB_SIGNAL_TRAP
    infrun: infwait_normal_state
    infrun: TARGET_WAITKIND_STOPPED
    infrun: stop_pc = 0x100009ac
    ### At this point, we stopped at the second SSS breakpoint...
    target_stopped_by_watchpoint () = 0
    ### We remove the SSS breakpoints...
    target_remove_breakpoint (0x0000000010000a28, xxx) = 0
    target_remove_breakpoint (0x00000000100009ac, xxx) = 0
    target_stopped_by_watchpoint () = 0
    ### We find that we're not done, so we resume....
    infrun: no stepping, continue
    ### And thus insert the user breakpoints again, except we're not
    ### inserting the second breakpoint?!?
    target_insert_breakpoint (0x0000000010000a24, xxx) = 0
    infrun: resume (step=0, signal=GDB_SIGNAL_0), trap_expected=0, current thread [process 15335610] at 0x100009ac
    target_resume (-1, continue, 0)
    infrun: prepare_to_wait
    target_wait (-1, status, options={}) = 15335610,   status->kind = exited, status = 2

What happens is that the removal of the software single-step
breakpoints effectively removed the breakpoint instruction from
inferior memory.  But because such breakpoints are inserted directly
as raw breakpoints rather than through the normal chain of
breakpoints, we fail to notice that one of the user breakpoints points
to the same address and that this user breakpoint is therefore
effectively un-inserted.  When resuming after the single-step, GDB
thinks that the user breakpoint is still inserted and therefore does
not need to insert it again.

This patch teaches the insert and remove routines of both regular and
raw breakpoints to be aware of each other.  Special care needs to be
applied in case the target supports evaluation of breakpoint
conditions or commands.

gdb/ChangeLog:

PR breakpoints/17000
* breakpoint.c (find_non_raw_software_breakpoint_inserted_here):
New function, extracted from software_breakpoint_inserted_here_p.
(software_breakpoint_inserted_here_p): Replace factored out code
by call to find_non_raw_software_breakpoint_inserted_here.
(bp_target_info_copy_insertion_state): New function.
(bkpt_insert_location): Handle the case of a single-step
breakpoint already inserted at the same address.
(bkpt_remove_location): Handle the case of a single-step
breakpoint still inserted at the same address.
(deprecated_insert_raw_breakpoint): Handle the case of non-raw
breakpoint already inserted at the same address.
(deprecated_remove_raw_breakpoint): Handle the case of a
non-raw breakpoint still inserted at the same address.
(find_single_step_breakpoint): New function, extracted from
single_step_breakpoint_inserted_here_p.
(find_single_step_breakpoint): New function,
factored out from single_step_breakpoint_inserted_here_p.
(single_step_breakpoint_inserted_here_p): Reimplement.

gdb/testsuite/ChangeLog:

PR breakpoints/17000
* gdb.base/sss-bp-on-user-bp.exp: Remove kfail.
* gdb.base/sss-bp-on-user-bp-2.exp: Remove kfail.

Tested on ppc-aix with AdaCore's testsuite.  Tested on x86_64-linux,
(native and gdbserver) with the official testsuite.  Also tested on
x86_64-linux through Pedro's branch enabling software single-stepping
on that platform (native and gdbserver).

gdb/ChangeLog
gdb/breakpoint.c
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.base/sss-bp-on-user-bp-2.exp
gdb/testsuite/gdb.base/sss-bp-on-user-bp.exp

index 698c15c37c41528460b4a02b60e8c0aba5c9053f..2b2a9f48c121d034b7550638e1fd17672d2f2fc0 100644 (file)
@@ -1,3 +1,26 @@
+2014-06-03  Joel Brobecker  <brobecker@adacore.com>
+           Pedro Alves  <palves@redhat.com>
+
+       PR breakpoints/17000
+       * breakpoint.c (find_non_raw_software_breakpoint_inserted_here):
+       New function, extracted from software_breakpoint_inserted_here_p.
+       (software_breakpoint_inserted_here_p): Replace factored out code
+       by call to find_non_raw_software_breakpoint_inserted_here.
+       (bp_target_info_copy_insertion_state): New function.
+       (bkpt_insert_location): Handle the case of a single-step
+       breakpoint already inserted at the same address.
+       (bkpt_remove_location): Handle the case of a single-step
+       breakpoint still inserted at the same address.
+       (deprecated_insert_raw_breakpoint): Handle the case of non-raw
+       breakpoint already inserted at the same address.
+       (deprecated_remove_raw_breakpoint): Handle the case of a
+       non-raw breakpoint still inserted at the same address.
+       (find_single_step_breakpoint): New function, extracted from
+       single_step_breakpoint_inserted_here_p.
+       (find_single_step_breakpoint): New function,
+       factored out from single_step_breakpoint_inserted_here_p.
+       (single_step_breakpoint_inserted_here_p): Reimplement.
+
 2014-06-03  Brad Mouring  <bmouring@ni.com>  (tiny patch)
 
        Pushed by Joel Brobecker  <brobecker@adacore.com>
index 4e6c627d913abcb4e7108f71b8787b3c81832df2..23c88957ad2810c44caca3700f4485d3e9009f77 100644 (file)
@@ -229,6 +229,9 @@ static void tcatch_command (char *arg, int from_tty);
 
 static void detach_single_step_breakpoints (void);
 
+static int find_single_step_breakpoint (struct address_space *aspace,
+                                       CORE_ADDR pc);
+
 static void free_bp_location (struct bp_location *loc);
 static void incref_bp_location (struct bp_location *loc);
 static void decref_bp_location (struct bp_location **loc);
@@ -4165,12 +4168,12 @@ breakpoint_inserted_here_p (struct address_space *aspace, CORE_ADDR pc)
   return 0;
 }
 
-/* This function returns non-zero iff there is a software breakpoint
-   inserted at PC.  */
+/* Ignoring deprecated raw breakpoints, return non-zero iff there is a
+   software breakpoint inserted at PC.  */
 
-int
-software_breakpoint_inserted_here_p (struct address_space *aspace,
-                                    CORE_ADDR pc)
+static struct bp_location *
+find_non_raw_software_breakpoint_inserted_here (struct address_space *aspace,
+                                               CORE_ADDR pc)
 {
   struct bp_location *bl, **blp_tmp;
 
@@ -4188,10 +4191,23 @@ software_breakpoint_inserted_here_p (struct address_space *aspace,
              && !section_is_mapped (bl->section))
            continue;           /* unmapped overlay -- can't be a match */
          else
-           return 1;
+           return bl;
        }
     }
 
+  return NULL;
+}
+
+/* This function returns non-zero iff there is a software breakpoint
+   inserted at PC.  */
+
+int
+software_breakpoint_inserted_here_p (struct address_space *aspace,
+                                    CORE_ADDR pc)
+{
+  if (find_non_raw_software_breakpoint_inserted_here (aspace, pc) != NULL)
+    return 1;
+
   /* Also check for software single-step breakpoints.  */
   if (single_step_breakpoint_inserted_here_p (aspace, pc))
     return 1;
@@ -13112,6 +13128,19 @@ bkpt_re_set (struct breakpoint *b)
   breakpoint_re_set_default (b);
 }
 
+/* Copy SRC's shadow buffer and whatever else we'd set if we actually
+   inserted DEST, so we can remove it later, in case SRC is removed
+   first.  */
+
+static void
+bp_target_info_copy_insertion_state (struct bp_target_info *dest,
+                                    const struct bp_target_info *src)
+{
+  dest->shadow_len = src->shadow_len;
+  memcpy (dest->shadow_contents, src->shadow_contents, src->shadow_len);
+  dest->placed_size = src->placed_size;
+}
+
 static int
 bkpt_insert_location (struct bp_location *bl)
 {
@@ -13119,8 +13148,25 @@ bkpt_insert_location (struct bp_location *bl)
     return target_insert_hw_breakpoint (bl->gdbarch,
                                        &bl->target_info);
   else
-    return target_insert_breakpoint (bl->gdbarch,
-                                    &bl->target_info);
+    {
+      struct bp_target_info *bp_tgt = &bl->target_info;
+      int ret;
+      int sss_slot;
+
+      /* There is no need to insert a breakpoint if an unconditional
+        raw/sss breakpoint is already inserted at that location.  */
+      sss_slot = find_single_step_breakpoint (bp_tgt->placed_address_space,
+                                             bp_tgt->placed_address);
+      if (sss_slot >= 0)
+       {
+         struct bp_target_info *sss_bp_tgt = single_step_breakpoints[sss_slot];
+
+         bp_target_info_copy_insertion_state (bp_tgt, sss_bp_tgt);
+         return 0;
+       }
+
+      return target_insert_breakpoint (bl->gdbarch, bp_tgt);
+    }
 }
 
 static int
@@ -13129,7 +13175,19 @@ bkpt_remove_location (struct bp_location *bl)
   if (bl->loc_type == bp_loc_hardware_breakpoint)
     return target_remove_hw_breakpoint (bl->gdbarch, &bl->target_info);
   else
-    return target_remove_breakpoint (bl->gdbarch, &bl->target_info);
+    {
+      struct bp_target_info *bp_tgt = &bl->target_info;
+      struct address_space *aspace = bp_tgt->placed_address_space;
+      CORE_ADDR address = bp_tgt->placed_address;
+
+      /* Only remove the breakpoint if there is no raw/sss breakpoint
+        still inserted at this location.  Otherwise, we would be
+        effectively disabling the raw/sss breakpoint.  */
+      if (single_step_breakpoint_inserted_here_p (aspace, address))
+       return 0;
+
+      return target_remove_breakpoint (bl->gdbarch, bp_tgt);
+    }
 }
 
 static int
@@ -15138,12 +15196,30 @@ deprecated_insert_raw_breakpoint (struct gdbarch *gdbarch,
                                  struct address_space *aspace, CORE_ADDR pc)
 {
   struct bp_target_info *bp_tgt;
+  struct bp_location *bl;
 
   bp_tgt = XCNEW (struct bp_target_info);
 
   bp_tgt->placed_address_space = aspace;
   bp_tgt->placed_address = pc;
 
+  /* If an unconditional non-raw breakpoint is already inserted at
+     that location, there's no need to insert another.  However, with
+     target-side evaluation of breakpoint conditions, if the
+     breakpoint that is currently inserted on the target is
+     conditional, we need to make it unconditional.  Note that a
+     breakpoint with target-side commands is not reported even if
+     unconditional, so we need to remove the commands from the target
+     as well.  */
+  bl = find_non_raw_software_breakpoint_inserted_here (aspace, pc);
+  if (bl != NULL
+      && VEC_empty (agent_expr_p, bl->target_info.conditions)
+      && VEC_empty (agent_expr_p, bl->target_info.tcommands))
+    {
+      bp_target_info_copy_insertion_state (bp_tgt, &bl->target_info);
+      return bp_tgt;
+    }
+
   if (target_insert_breakpoint (gdbarch, bp_tgt) != 0)
     {
       /* Could not insert the breakpoint.  */
@@ -15161,9 +15237,30 @@ int
 deprecated_remove_raw_breakpoint (struct gdbarch *gdbarch, void *bp)
 {
   struct bp_target_info *bp_tgt = bp;
+  struct address_space *aspace = bp_tgt->placed_address_space;
+  CORE_ADDR address = bp_tgt->placed_address;
+  struct bp_location *bl;
   int ret;
 
-  ret = target_remove_breakpoint (gdbarch, bp_tgt);
+  bl = find_non_raw_software_breakpoint_inserted_here (aspace, address);
+
+  /* Only remove the raw breakpoint if there are no other non-raw
+     breakpoints still inserted at this location.  Otherwise, we would
+     be effectively disabling those breakpoints.  */
+  if (bl == NULL)
+    ret = target_remove_breakpoint (gdbarch, bp_tgt);
+  else if (!VEC_empty (agent_expr_p, bl->target_info.conditions)
+          || !VEC_empty (agent_expr_p, bl->target_info.tcommands))
+    {
+      /* The target is evaluating conditions, and when we inserted the
+        software single-step breakpoint, we had made the breakpoint
+        unconditional and command-less on the target side.  Reinsert
+        to restore the conditions/commands.  */
+      ret = target_insert_breakpoint (bl->gdbarch, &bl->target_info);
+    }
+  else
+    ret = 0;
+
   xfree (bp_tgt);
 
   return ret;
@@ -15269,12 +15366,12 @@ detach_single_step_breakpoints (void)
                                single_step_breakpoints[i]);
 }
 
-/* Check whether a software single-step breakpoint is inserted at
-   PC.  */
+/* Find the software single-step breakpoint that inserted at PC.
+   Returns its slot if found, and -1 if not found.  */
 
-int
-single_step_breakpoint_inserted_here_p (struct address_space *aspace, 
-                                       CORE_ADDR pc)
+static int
+find_single_step_breakpoint (struct address_space *aspace,
+                            CORE_ADDR pc)
 {
   int i;
 
@@ -15285,10 +15382,20 @@ single_step_breakpoint_inserted_here_p (struct address_space *aspace,
          && breakpoint_address_match (bp_tgt->placed_address_space,
                                       bp_tgt->placed_address,
                                       aspace, pc))
-       return 1;
+       return i;
     }
 
-  return 0;
+  return -1;
+}
+
+/* Check whether a software single-step breakpoint is inserted at
+   PC.  */
+
+int
+single_step_breakpoint_inserted_here_p (struct address_space *aspace,
+                                       CORE_ADDR pc)
+{
+  return find_single_step_breakpoint (aspace, pc) >= 0;
 }
 
 /* Returns 0 if 'bp' is NOT a syscall catchpoint,
index 40518b6730738be962cff43796bcf8b789a00dab..36358e0703214b4a49d1115dd12585aa4f952afd 100644 (file)
@@ -1,3 +1,10 @@
+2014-06-03  Joel Brobecker  <brobecker@adacore.com>
+           Pedro Alves  <palves@redhat.com>
+
+       PR breakpoints/17000
+       * gdb.base/sss-bp-on-user-bp.exp: Remove kfail.
+       * gdb.base/sss-bp-on-user-bp-2.exp: Remove kfail.
+
 2014-06-03  Brad Mouring  <bmouring@ni.com>  (tiny patch)
 
        * gdb.base/subst.exp: Add tests to verify partial path matching
index b41f86e2b2049d3d4ff74571bb81c3ad534dd79c..87fa5f81be568d293517ff3add8e508588e6c1c1 100644 (file)
@@ -121,7 +121,6 @@ gdb_test_multiple $test $test {
 
 set command "stepi_del_break"
 set test $command
-setup_kfail "breakpoints/17000" "*-*-*"
 gdb_test_multiple $command $test {
     -re "^$command\r\n$gdb_prompt " {
        # Note no end anchor, because "si&" finishes and prints the
index 62415e7e523201a167c4ef28b3a6d9b8b61f3f3e..2a12ad6f6fc079c0abd21b0c5298b88c76442369 100644 (file)
@@ -47,6 +47,5 @@ gdb_test "si" "Breakpoint .* bar break .*"
 
 # If the breakpoint is still correctly inserted, then this jump should
 # re-trigger it.  Otherwise, GDB will lose control and the program
-# will exit.
-setup_kfail "breakpoints/17000" "*-*-*"
+# will exit.  See PR breakpoints/17000.
 gdb_test "jump *\$pc" "Breakpoint .* bar break .*"