From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gdb-patches-return-68555-listarch-gdb-patches=sources.redhat.com@sourceware.org>
Received: (qmail 17567 invoked by alias); 21 Nov 2009 09:41:01 -0000
Received: (qmail 17532 invoked by uid 22791); 21 Nov 2009 09:40:59 -0000
X-SWARE-Spam-Status: No, hits=-1.7 required=5.0 	tests=AWL,BAYES_00,SARE_MSGID_LONG40,SPF_PASS
X-Spam-Check-By: sourceware.org
Received: from mail-px0-f189.google.com (HELO mail-px0-f189.google.com) (209.85.216.189)     by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Sat, 21 Nov 2009 09:39:55 +0000
Received: by pxi27 with SMTP id 27so2791981pxi.25         for <gdb-patches@sourceware.org>; Sat, 21 Nov 2009 01:39:54 -0800 (PST)
MIME-Version: 1.0
Received: by 10.142.65.24 with SMTP id n24mr50010wfa.312.1258796394099; Sat,  	21 Nov 2009 01:39:54 -0800 (PST)
In-Reply-To: <4B06DAB2.7090409@vmware.com>
References: <4AECE12F.3000704@vmware.com> <4AF31C1F.2000405@vmware.com>  	<200911051909.40796.pedro@codesourcery.com> <200911120027.03426.pedro@codesourcery.com>  	<4B06DAB2.7090409@vmware.com>
From: Hui Zhu <teawater@gmail.com>
Date: Sat, 21 Nov 2009 09:41:00 -0000
Message-ID: <daef60380911210139m43a8df7er487e3da9c0994907@mail.gmail.com>
Subject: Re: [RFA] Fix hw watchpoints in process record.
To: Michael Snyder <msnyder@vmware.com>, Pedro Alves <pedro@codesourcery.com>
Cc: "gdb-patches@sourceware.org" <gdb-patches@sourceware.org>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
X-IsSubscribed: yes
Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm
Precedence: bulk
List-Id: <gdb-patches.sourceware.org>
List-Subscribe: <mailto:gdb-patches-subscribe@sourceware.org>
List-Archive: <http://sourceware.org/ml/gdb-patches/>
List-Post: <mailto:gdb-patches@sourceware.org>
List-Help: <mailto:gdb-patches-help@sourceware.org>, <http://sourceware.org/ml/#faqs>
Sender: gdb-patches-owner@sourceware.org
X-SW-Source: 2009-11/txt/msg00462.txt.bz2

Hi guys,

There are some patches about hw breakpoint are running on maillist.
If you don't mind, I suggest we wait until they OK.

Thanks,
Hui

On Sat, Nov 21, 2009 at 02:06, Michael Snyder <msnyder@vmware.com> wrote:
> OK, I withdraw my patch in favor of this one.
> If Hui likes it, let's check it in.
>
> Thanks for your help, Pedro!
>
> Pedro Alves wrote:
>>
>> On Thursday 05 November 2009 19:09:40, Pedro Alves wrote:
>>
>>> this assumes the target/arch has continuable watchpoints,
>>> like x86/x86_64. =A0You'll see this misbehave on mips, when Hui
>>> contributes the support.
>>>
>>> You're evaluating the watchpoint expression on every event --- I
>>> have a feeling it would make more sense to decouple precord from
>>> that, like all other targets are decoupled. =A0That is, have precord
>>> check for low level watchpoint triggers --- given [addr,length,type]
>>> low level watchpoints --- when memory changes (at record_mem_entry time=
),
>>> instead of evaluating the whole high-level watchpoint expression.
>>>
>>> [IMO, we should mostly think of precord as just another a
>>> simulator that mostly only interfaces gdb's core through
>>> the target_ methods.]
>>
>>
>> Sure enough, trying a simple test like this shows
>> breakage:
>>
>> =A01 =A0 =A0 =A0 #include <stdio.h>
>> =A02
>> =A03 =A0 =A0 =A0 int glob;
>> =A04
>> =A05 =A0 =A0 =A0 int main ()
>> =A06 =A0 =A0 =A0 {
>> =A07 =A0 =A0 =A0 =A0 int loc =3D -1;
>> =A08
>> =A09 =A0 =A0 =A0 =A0 while (1)
>> =A010 =A0 =A0 =A0 =A0 =A0{
>> =A011 =A0 =A0 =A0 =A0 =A0 =A0printf ("glob =3D %d\n", glob);
>> =A012 =A0 =A0 =A0 =A0 =A0 =A0glob++;
>> =A013 =A0 =A0 =A0 =A0 =A0 =A0printf ("loc =3D %d\n", loc);
>> =A014 =A0 =A0 =A0 =A0 =A0 =A0loc++;
>> =A015 =A0 =A0 =A0 =A0 =A0}
>> =A016 =A0 =A0 =A0}
>>
>>
>> =A0(gdb) start
>> =A0Temporary breakpoint 1, main () at test.c:7
>> =A07 =A0 =A0 =A0 =A0 int loc =3D -1;
>> =A0(gdb) record
>> =A0(gdb) watch loc
>> =A0During symbol reading, incomplete CFI data; unspecified registers (e.=
g.,
>> rax) at 0x40049c.
>> =A0Hardware watchpoint 2: loc
>> =A0(gdb) watch glob
>> =A0Hardware watchpoint 3: glob
>> =A0(gdb) n
>> =A0Hardware watchpoint 2: loc
>>
>> =A0Old value =3D 0
>> =A0New value =3D -1
>> =A0main () at test.c:11
>> =A011 =A0 =A0 =A0 =A0 =A0 =A0printf ("glob =3D %d\n", glob);
>> =A0(gdb)
>> =A0glob =3D 0
>> =A012 =A0 =A0 =A0 =A0 =A0 =A0glob++;
>> =A0(gdb)
>> =A0Hardware watchpoint 3: glob
>>
>> =A0<a few more next's, then>
>>
>> =A0(gdb) reverse-continue
>> =A0Continuing.
>> =A0Hardware watchpoint 2: loc
>>
>> =A0Old value =3D 1
>> =A0New value =3D 0
>> =A0main () at test.c:14
>> =A014 =A0 =A0 =A0 =A0 =A0 =A0loc++;
>> =A0(gdb) reverse-continue
>> =A0Continuing.
>>
>> =A0Watchpoint 2 deleted because the program has left the block in
>> =A0which its expression is valid.
>>
>> =A0Watchpoint 2 deleted because the program has left the block in
>> =A0which its expression is valid.
>> =A00x00007ffff7aebd15 in _IO_file_write () from /lib/libc.so.6
>> =A0(gdb)
>>
>> Nope, it hasn't left the block, I was continue'ing in the
>> loop... =A0The trouble relates to the fact that precord forces
>> single-stepping --- which reverse single-steps into the
>> printf call --- and, watchpoint_check wants to check
>> if the watchpoints are in scope.
>>
>>
>> Here's the alternative implementation I was suggesting last
>> week. =A0It ends up being simpler and more efficient, in fact.
>> We checking for watchpoint hits less often, and the check is
>> much cheaper since it doesn't do any extra reads, create
>> values, or evaluate expressions.
>>
>> I also didn't see any need for set_executing calls, which
>> were a sign of trouble.
>>
>> I've added a comment indicating what needs to be done
>> to support targets/archs with non-continuable watchpoints
>> (just check for inserted watchpoints that should trigger,
>> before actually doing any memory change)
>>
>> An alternative to that is to expose better the continuable vs
>> steppable watchpoints at the target layer, so that record can
>> always claim continuable watchpoints while replaying, dispite
>> what the target beneath or the arch supports.
>>
>>
>> This passed the new watch-reverse.exp and watch-precsave.exp.
>> Playing a bit with the small test shown above didn't show
>> any issue. =A0Want to give it a try?
>>
>> --
>> Pedro Alves
>>
>>
>> ---
>> =A0gdb/NEWS =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 | =A0 =A04 +
>> =A0gdb/breakpoint.c =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 | =A0 33 ++++++++++
>> =A0gdb/breakpoint.h =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 | =A0 =A05 +
>> =A0gdb/record.c =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 | =A0 68
>> +++++++++++++++++++---
>> =A0gdb/testsuite/gdb.reverse/watch-precsave.exp | =A0 80
>> ++++++++++++++++++++++++++
>> =A0gdb/testsuite/gdb.reverse/watch-reverse.exp =A0| =A0 82
>> ++++++++++++++++++++++++++-
>> =A06 files changed, 260 insertions(+), 12 deletions(-)
>>
>> Index: src/gdb/breakpoint.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/breakpoint.c =A0 2009-11-12 00:06:31.000000000 +0000
>> +++ src/gdb/breakpoint.c =A0 =A0 =A0 =A02009-11-12 00:06:55.000000000 +0=
000
>> @@ -3063,6 +3063,39 @@ watchpoints_triggered (struct target_wai
>> =A0 return 1;
>> =A0}
>>
>> +int
>> +hardware_watchpoint_inserted_in_range (CORE_ADDR addr, ULONGEST len)
>> +{
>> + =A0struct breakpoint *bpt;
>> +
>> + =A0ALL_BREAKPOINTS (bpt)
>> + =A0 =A0{
>> + =A0 =A0 =A0struct bp_location *loc;
>> +
>> + =A0 =A0 =A0if (bpt->type !=3D bp_hardware_watchpoint
>> + =A0 =A0 =A0 =A0 && bpt->type !=3D bp_access_watchpoint)
>> + =A0 =A0 =A0 continue;
>> +
>> + =A0 =A0 =A0if (!breakpoint_enabled (bpt))
>> + =A0 =A0 =A0 continue;
>> +
>> + =A0 =A0 =A0for (loc =3D bpt->loc; loc; loc =3D loc->next)
>> + =A0 =A0 =A0 if (loc->inserted)
>> + =A0 =A0 =A0 =A0 {
>> + =A0 =A0 =A0 =A0 =A0 CORE_ADDR l, h;
>> +
>> + =A0 =A0 =A0 =A0 =A0 /* Check for interception. =A0*/
>> +
>> + =A0 =A0 =A0 =A0 =A0 l =3D max (loc->address, addr);
>> + =A0 =A0 =A0 =A0 =A0 h =3D min (loc->address + loc->length, addr + len);
>> +
>> + =A0 =A0 =A0 =A0 =A0 if (l < h)
>> + =A0 =A0 =A0 =A0 =A0 =A0 return 1;
>> + =A0 =A0 =A0 =A0 }
>> + =A0 =A0}
>> + =A0return 0;
>> +}
>> +
>> =A0/* Possible return values for watchpoint_check (this can't be an enum
>> =A0 =A0because of check_errors). =A0*/
>> =A0/* The watchpoint has been deleted. =A0*/
>> Index: src/gdb/breakpoint.h
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/breakpoint.h =A0 2009-11-12 00:06:31.000000000 +0000
>> +++ src/gdb/breakpoint.h =A0 =A0 =A0 =A02009-11-12 00:06:55.000000000 +0=
000
>> @@ -978,4 +978,9 @@ extern struct breakpoint *get_tracepoint
>> =A0 =A0is newly allocated; the caller should free when done with it. =A0=
*/
>> =A0extern VEC(breakpoint_p) *all_tracepoints (void);
>>
>> +/* Returns true if there's a hardware watchpoint or access watchpoint
>> + =A0 inserted in the range defined by ADDR and LEN. =A0*/
>> +extern int hardware_watchpoint_inserted_in_range (CORE_ADDR addr,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 ULONGEST len);
>> +
>> =A0#endif /* !defined (BREAKPOINT_H) */
>> Index: src/gdb/record.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/record.c =A0 =A0 =A0 2009-11-12 00:06:31.000000000 +0000
>> +++ src/gdb/record.c =A0 =A02009-11-12 00:06:55.000000000 +0000
>> @@ -224,6 +224,7 @@ static int (*record_beneath_to_insert_br
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 struct bp_target_info
>> *);
>> =A0static int (*record_beneath_to_remove_breakpoint) (struct gdbarch *,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 struct bp_target_info
>> *);
>> +static int (*record_beneath_to_stopped_by_watchpoint) (void);
>>
>> =A0/* Alloc and free functions for record_reg, record_mem, and record_end
>> =A0 =A0entries. =A0*/
>> @@ -673,6 +674,9 @@ record_gdb_operation_disable_set (void)
>> =A0 return old_cleanups;
>> =A0}
>>
>> +/* Flag set to TRUE for target_stopped_by_watchpoint. =A0*/
>> +static int record_hw_watchpoint =3D 0;
>> +
>> =A0/* Execute one instruction from the record log. =A0Each instruction in
>> =A0 =A0the log will be represented by an arbitrary sequence of register
>> =A0 =A0entries and memory entries, followed by an 'end' entry. =A0*/
>> @@ -739,7 +743,21 @@ record_exec_insn (struct regcache *regca
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0entry->u.=
mem.len);
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 else
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0memcpy (record_get_loc (entry), mem=
, entry->u.mem.len);
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 memcpy (record_get_loc (entry), me=
m,
>> entry->u.mem.len);
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* We've changed memory --- check =
if a hardware
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0watchpoint should trap. =A0=
Note that this
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0presently assumes the targe=
t beneath supports
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0continuable watchpoints. =
=A0On non-continuable
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0watchpoints target, we'll w=
ant to check this
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0_before_ actually doing the=
 memory change, and
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0not doing the change at all=
 if the watchpoint
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0traps. =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (hardware_watchpoint_inserted_i=
n_range
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 (entry->u.mem.addr, entry-=
>u.mem.len))
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 record_hw_watchpoint =3D 1;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> =A0 =A0 =A0 =A0 =A0 }
>> =A0 =A0 =A0 }
>> @@ -770,6 +788,7 @@ static int (*tmp_to_insert_breakpoint) (
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0struct bp_target_info *);
>> =A0static int (*tmp_to_remove_breakpoint) (struct gdbarch *,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0struct bp_target_info *);
>> +static int (*tmp_to_stopped_by_watchpoint) (void);
>>
>> =A0static void record_restore (void);
>>
>> @@ -894,6 +913,8 @@ record_open (char *name, int from_tty)
>> =A0 =A0 =A0 =A0tmp_to_insert_breakpoint =3D t->to_insert_breakpoint;
>> =A0 =A0 =A0 if (!tmp_to_remove_breakpoint)
>> =A0 =A0 =A0 =A0tmp_to_remove_breakpoint =3D t->to_remove_breakpoint;
>> + =A0 =A0 =A0if (!tmp_to_stopped_by_watchpoint)
>> + =A0 =A0 =A0 tmp_to_stopped_by_watchpoint =3D t->to_stopped_by_watchpoi=
nt;
>> =A0 =A0 }
>> =A0 if (!tmp_to_xfer_partial)
>> =A0 =A0 error (_("Could not find 'to_xfer_partial' method on the target
>> stack."));
>> @@ -915,6 +936,7 @@ record_open (char *name, int from_tty)
>> =A0 record_beneath_to_xfer_partial =3D tmp_to_xfer_partial;
>> =A0 record_beneath_to_insert_breakpoint =3D tmp_to_insert_breakpoint;
>> =A0 record_beneath_to_remove_breakpoint =3D tmp_to_remove_breakpoint;
>> + =A0record_beneath_to_stopped_by_watchpoint =3D tmp_to_stopped_by_watch=
point;
>>
>> =A0 if (current_target.to_stratum =3D=3D core_stratum)
>> =A0 =A0 record_core_open_1 (name, from_tty);
>> @@ -1069,14 +1091,23 @@ record_wait (struct target_ops *ops,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0{
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0struct regcache *regcache;
>>
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* Yes -- check if there is a breakpoi=
nt. =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* Yes -- this is likely our single-st=
ep finishing,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0but check if there's any reason=
 the core would be
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0interested in the event. =A0*/
>> +
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0registers_changed ();
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0regcache =3D get_current_regcache ();
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0tmp_pc =3D regcache_read_pc (regcache=
);
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (breakpoint_inserted_here_p (get_re=
gcache_aspace
>> (regcache),
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 tmp_pc))
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (target_stopped_by_watchpoint ())
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* Always interested in watchp=
oints. =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 else if (breakpoint_inserted_here_p (g=
et_regcache_aspace
>> (regcache),
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0tmp_pc))
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0{
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* There is a breakpoint. =A0G=
DB will want to stop.
>> =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* There is a breakpoint here.=
 =A0Let the core
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0handle it. =A0*/
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0struct gdbarch *gdbarch =3D g=
et_regcache_arch
>> (regcache);
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0CORE_ADDR decr_pc_after_break
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0=3D gdbarch_decr_pc_after=
_break (gdbarch);
>> @@ -1086,10 +1117,8 @@ record_wait (struct target_ops *ops,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0}
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0else
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0{
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* There is not a breakpoint, =
and gdb is not
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0stepping, therefore gdb=
 will not stop.
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0Therefore we will not r=
eturn to gdb.
>> - =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0Record the insn and res=
ume. =A0*/
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 /* This must be a single-step =
trap. =A0Record the
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0insn and issue another =
step. =A0*/
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0if (!do_record_message (regca=
che, TARGET_SIGNAL_0))
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0break;
>>
>> @@ -1116,6 +1145,7 @@ record_wait (struct target_ops *ops,
>> =A0 =A0 =A0 struct cleanup *old_cleanups =3D make_cleanup (record_wait_c=
leanups,
>> 0);
>> =A0 =A0 =A0 CORE_ADDR tmp_pc;
>>
>> + =A0 =A0 =A0record_hw_watchpoint =3D 0;
>> =A0 =A0 =A0 status->kind =3D TARGET_WAITKIND_STOPPED;
>>
>> =A0 =A0 =A0 /* Check breakpoint when forward execute. =A0*/
>> @@ -1219,6 +1249,14 @@ record_wait (struct target_ops *ops,
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 gdbarch_decr_pc_after_break
>> (gdbarch));
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0continue_flag =3D 0;
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0}
>> +
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (record_hw_watchpoint)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 {
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 if (record_debug)
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 fprintf_unfiltered (gdb_st=
dlog,
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 "Process record: hit hw
>> watchpoint.\n");
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 continue_flag =3D 0;
>> + =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 }
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0/* Check target signal */
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0if (record_list->u.end.sigval !=3D TA=
RGET_SIGNAL_0)
>> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0/* FIXME: better way to check */
>> @@ -1260,6 +1298,16 @@ replay_out:
>> =A0 return inferior_ptid;
>> =A0}
>>
>> +/* to_stopped_by_watchpoint method */
>> +static int
>> +record_stopped_by_watchpoint (void)
>> +{
>> + =A0if (RECORD_IS_REPLAY)
>> + =A0 =A0return record_hw_watchpoint;
>> + =A0else
>> + =A0 =A0return record_beneath_to_stopped_by_watchpoint ();
>> +}
>> +
>> =A0/* "to_disconnect" method for process record target. =A0*/
>>
>> =A0static void
>> @@ -1545,6 +1593,7 @@ init_record_ops (void)
>> =A0 record_ops.to_remove_breakpoint =3D record_remove_breakpoint;
>> =A0 record_ops.to_can_execute_reverse =3D record_can_execute_reverse;
>> =A0 record_ops.to_stratum =3D record_stratum;
>> + =A0record_ops.to_stopped_by_watchpoint =3D record_stopped_by_watchpoin=
t;
>> =A0 record_ops.to_magic =3D OPS_MAGIC;
>> =A0}
>>
>> @@ -1750,6 +1799,7 @@ init_record_core_ops (void)
>> =A0 record_core_ops.to_can_execute_reverse =3D record_can_execute_revers=
e;
>> =A0 record_core_ops.to_has_execution =3D record_core_has_execution;
>> =A0 record_core_ops.to_stratum =3D record_stratum;
>> + =A0record_core_ops.to_stopped_by_watchpoint =3D
>> record_stopped_by_watchpoint;
>> =A0 record_core_ops.to_magic =3D OPS_MAGIC;
>> =A0}
>>
>> Index: src/gdb/NEWS
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/NEWS =A0 2009-11-12 00:06:31.000000000 +0000
>> +++ src/gdb/NEWS =A0 =A0 =A0 =A02009-11-12 00:06:55.000000000 +0000
>> @@ -71,6 +71,10 @@ show follow-exec-mode
>> =A0 creates a new one. =A0This is useful to be able to restart the old
>> =A0 executable after the inferior having done an exec call.
>>
>> +* Bug fixes
>> +
>> +Process record now works correctly with hardware watchpoints.
>> +
>> =A0*** Changes in GDB 7.0
>>
>> =A0* GDB now has an interface for JIT compilation. =A0Applications that
>> Index: src/gdb/testsuite/gdb.reverse/watch-reverse.exp
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/testsuite/gdb.reverse/watch-reverse.exp =A0 =A0 =A0 =A0=
2009-11-12
>> 00:06:31.000000000 +0000
>> +++ src/gdb/testsuite/gdb.reverse/watch-reverse.exp =A0 =A0 2009-11-12
>> 00:10:23.000000000 +0000
>> @@ -38,8 +38,8 @@ if [target_info exists gdb,use_precord]
>> =A0 =A0 # FIXME: command ought to acknowledge, so we can test if it succ=
eeded.
>> =A0}
>>
>> -# Only software watchpoints can be used in reverse
>> -gdb_test "set can-use-hw-watchpoints 0" "" ""
>> +# Test software watchpoints
>> +gdb_test "set can-use-hw-watchpoints 0" "" "disable hw watchpoints"
>>
>> =A0gdb_test "break marker1" \
>> =A0 =A0 "Breakpoint $decimal at $hex: file .*$srcfile, line $decimal.*" \
>> @@ -122,3 +122,81 @@ gdb_test "continue" \
>> =A0gdb_test "continue" \
>> =A0 =A0 ".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D -1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> =A0 =A0 "watchpoint hit in reverse, fifth time"
>> +
>> +gdb_test "set can-use-hw-watchpoints 1" "" "enable hw watchpoints"
>> +
>> +###
>> +###
>> +###
>> +
>> +# FIXME 'set exec-dir' command should give some output so we can test.
>> +gdb_test "set exec-direction forward" "" "set forward"
>> +
>> +# Continue until first change, from -1 to 0
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D -1.*New value =3D 0.*i=
val3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, first time"
>> +
>> +# Continue until the next change, from 0 to 1.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D 1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, second time"
>> +
>> +# Continue until the next change, from 1 to 2.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 1.*New value =3D 2.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, third time"
>> +
>> +# Continue until the next change, from 2 to 3.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 2.*New value =3D 3.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, fourth time"
>> +
>> +# Continue until the next change, from 3 to 4.
>> +# Note that this one is outside the loop.
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 3.*New value =3D 4.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, fifth time"
>> +
>> +# Continue until we hit the finishing marker function.
>> +# Make sure we hit no more watchpoints.
>> +
>> +gdb_test "continue" "marker2 .*" "replay forward to marker2"
>> +
>> +###
>> +###
>> +###
>> +
>> +# FIXME 'set exec-dir' command should give some output so we can test.
>> +gdb_test "set exec-direction reverse" "" "set reverse"
>> +
>> +# Reverse until the previous change, from 4 to 3
>> +# Note that this one is outside the loop
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 4.*New value =3D 3.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, first time"
>> +
>> +# Reverse until the previous change, from 3 to 2.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 3.*New value =3D 2.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, second time"
>> +
>> +# Reverse until the previous change, from 2 to 1.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 2.*New value =3D 1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, third time"
>> +
>> +# Reverse until the previous change, from 1 to 0.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 1.*New value =3D 0.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, fourth time"
>> +
>> +# Reverse until first change, from 0 to -1
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D -1.*i=
val3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, fifth time"
>> +
>> Index: src/gdb/testsuite/gdb.reverse/watch-precsave.exp
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- src.orig/gdb/testsuite/gdb.reverse/watch-precsave.exp =A0 =A0 =A0 20=
09-11-12
>> 00:06:31.000000000 +0000
>> +++ src/gdb/testsuite/gdb.reverse/watch-precsave.exp =A0 =A02009-11-12
>> 00:10:07.000000000 +0000
>> @@ -1,4 +1,4 @@
>> -# Copyright 2008, 2009 Free Software Foundation, Inc.
>> +# Copyright 2009 Free Software Foundation, Inc.
>>
>> =A0# This program is free software; you can redistribute it and/or modify
>> =A0# it under the terms of the GNU General Public License as published by
>> @@ -140,3 +140,81 @@ gdb_test "continue" \
>> =A0gdb_test "continue" \
>> =A0 =A0 ".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D -1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> =A0 =A0 "watchpoint hit in reverse, fifth time"
>> +
>> +gdb_test "set can-use-hw-watchpoints 1" "" "enable hw watchpoints"
>> +
>> +###
>> +###
>> +###
>> +
>> +# FIXME 'set exec-dir' command should give some output so we can test.
>> +gdb_test "set exec-direction forward" "" "set forward"
>> +
>> +# Continue until first change, from -1 to 0
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D -1.*New value =3D 0.*i=
val3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, first time"
>> +
>> +# Continue until the next change, from 0 to 1.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D 1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, second time"
>> +
>> +# Continue until the next change, from 1 to 2.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 1.*New value =3D 2.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, third time"
>> +
>> +# Continue until the next change, from 2 to 3.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 2.*New value =3D 3.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, fourth time"
>> +
>> +# Continue until the next change, from 3 to 4.
>> +# Note that this one is outside the loop.
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 3.*New value =3D 4.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit, forward replay, fifth time"
>> +
>> +# Continue until we hit the finishing marker function.
>> +# Make sure we hit no more watchpoints.
>> +
>> +gdb_test "continue" "marker2 .*" "replay forward to marker2"
>> +
>> +###
>> +###
>> +###
>> +
>> +# FIXME 'set exec-dir' command should give some output so we can test.
>> +gdb_test "set exec-direction reverse" "" "set reverse"
>> +
>> +# Reverse until the previous change, from 4 to 3
>> +# Note that this one is outside the loop
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 4.*New value =3D 3.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, first time"
>> +
>> +# Reverse until the previous change, from 3 to 2.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 3.*New value =3D 2.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, second time"
>> +
>> +# Reverse until the previous change, from 2 to 1.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 2.*New value =3D 1.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, third time"
>> +
>> +# Reverse until the previous change, from 1 to 0.
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 1.*New value =3D 0.*iv=
al3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, fourth time"
>> +
>> +# Reverse until first change, from 0 to -1
>> +
>> +gdb_test "continue" \
>> + =A0 =A0".*\[Ww\]atchpoint.*ival3.*Old value =3D 0.*New value =3D -1.*i=
val3 =3D
>> count; ival4 =3D count;.*" \
>> + =A0 =A0"watchpoint hit in reverse, HW, fifth time"
>> +
>>
>
>