From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 6823 invoked by alias); 23 Dec 2009 23:43:39 -0000 Received: (qmail 6812 invoked by uid 22791); 23 Dec 2009 23:43:38 -0000 X-SWARE-Spam-Status: No, hits=-2.6 required=5.0 tests=AWL,BAYES_00,SPF_PASS X-Spam-Check-By: sourceware.org Received: from e24smtp01.br.ibm.com (HELO e24smtp01.br.ibm.com) (32.104.18.85) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Wed, 23 Dec 2009 23:43:31 +0000 Received: from d24relay01.br.ibm.com (d24relay01.br.ibm.com [9.8.31.16]) by e24smtp01.br.ibm.com (8.14.3/8.13.1) with ESMTP id nBNNjvcl028207 for ; Wed, 23 Dec 2009 21:45:57 -0200 Received: from d24av03.br.ibm.com (d24av03.br.ibm.com [9.8.31.95]) by d24relay01.br.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id nBNNhSaI4063432 for ; Wed, 23 Dec 2009 20:43:28 -0300 Received: from d24av03.br.ibm.com (loopback [127.0.0.1]) by d24av03.br.ibm.com (8.14.3/8.13.1/NCO v10.0 AVout) with ESMTP id nBNNhRVi005740 for ; Wed, 23 Dec 2009 21:43:27 -0200 Received: from hactar.localnet ([9.18.198.117]) by d24av03.br.ibm.com (8.14.3/8.13.1/NCO v10.0 AVin) with ESMTP id nBNNhRAn005731 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Wed, 23 Dec 2009 21:43:27 -0200 From: Thiago Jung Bauermann To: gdb-patches@sourceware.org Subject: [RFA] Fix verification of changed values for big values. Date: Wed, 23 Dec 2009 23:43:00 -0000 User-Agent: KMail/1.12.2 (Linux/2.6.30-2-686; KDE/4.3.2; i686; ; ) Cc: Luis Machado MIME-Version: 1.0 Content-Type: Multipart/Mixed; boundary="Boundary-00=_srqMLa9aqQ2p7mU" Message-Id: <200912232142.37008.bauerman@br.ibm.com> X-IsSubscribed: yes Mailing-List: contact gdb-patches-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sourceware.org X-SW-Source: 2009-12/txt/msg00368.txt.bz2 --Boundary-00=_srqMLa9aqQ2p7mU Content-Type: Text/Plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Content-length: 1316 Hi, This is a resubmission of a patch I posted long ago: http://sourceware.org/ml/gdb-patches/2009-05/msg00670.html It's good on its own, but is also important for ranged watchpoints in our ppc476 support patches. Here's my original explanation: Right now, GDB calls value_equal when comparing the old and new values of a watchpoint. IMO this is not correct, since that function will call coerce_array and effectively just compare the addresses of arrays being watched. This patch introduces a new value comparison function which works in the mentioned case, and a testcase which fails without the patch and passes with it. Tromey had approved it here: http://sourceware.org/ml/gdb-patches/2009-06/msg00076.html But I don't know if approvals expire or not, so... This version addresses Tromey's comments. -- []'s Thiago Jung Bauermann IBM Linux Technology Center gdb/ * valarith.c (value_equal_contents): New function. * value.h (value_equal_contents): Declare. * breakpoint.c (watchpoint_check): Use value_equal_contents instead of value_equal. gdb/testsuite/ * gdb.base/watchpoint.exp (test_watchpoint_in_big_blob): New function. (top level): Call test_watchpoint_in_big_blob. * gdb.base/watchpoint.c (buf): Change size to value too big for hardware watchpoints. (func3): Write to buf. --Boundary-00=_srqMLa9aqQ2p7mU Content-Type: text/x-patch; charset="UTF-8"; name="equal-contents.diff" Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename="equal-contents.diff" Content-length: 4137 Index: gdb/gdb/breakpoint.c =================================================================== --- gdb.orig/gdb/breakpoint.c 2009-12-23 17:39:55.000000000 -0200 +++ gdb/gdb/breakpoint.c 2009-12-23 17:56:27.000000000 -0200 @@ -3174,7 +3174,6 @@ watchpoints_triggered (struct target_wai #define BP_TEMPFLAG 1 #define BP_HARDWAREFLAG 2 -/* Check watchpoint condition. */ static int watchpoint_check (void *p) @@ -3245,8 +3244,12 @@ watchpoint_check (void *p) struct value *new_val; fetch_watchpoint_value (b->exp, &new_val, NULL, NULL); + + /* We use value_equal_contents instead of value_equal because the latter + coerces an array to a pointer, thus comparing just the address of the + array instead of its contents. This is not what we want. */ if ((b->val != NULL) != (new_val != NULL) - || (b->val != NULL && !value_equal (b->val, new_val))) + || (b->val != NULL && !value_equal_contents (b->val, new_val))) { if (new_val != NULL) { Index: gdb/gdb/testsuite/gdb.base/watchpoint.c =================================================================== --- gdb.orig/gdb/testsuite/gdb.base/watchpoint.c 2009-12-23 17:39:55.000000000 -0200 +++ gdb/gdb/testsuite/gdb.base/watchpoint.c 2009-12-23 17:40:03.000000000 -0200 @@ -30,7 +30,7 @@ int ival2 = -1; int ival3 = -1; int ival4 = -1; int ival5 = -1; -char buf[10]; +char buf[30]; struct foo { int val; @@ -95,6 +95,7 @@ func3 () x = 1; /* second x assignment */ y = 1; y = 2; + buf[26] = 3; } int Index: gdb/gdb/testsuite/gdb.base/watchpoint.exp =================================================================== --- gdb.orig/gdb/testsuite/gdb.base/watchpoint.exp 2009-12-23 17:39:55.000000000 -0200 +++ gdb/gdb/testsuite/gdb.base/watchpoint.exp 2009-12-23 17:40:03.000000000 -0200 @@ -678,6 +678,24 @@ proc test_inaccessible_watchpoint {} { } } +proc test_watchpoint_in_big_blob {} { + global gdb_prompt + + gdb_test "watch buf" ".*atchpoint \[0-9\]+: buf" + send_gdb "cont\n" + gdb_expect { + -re "Continuing.*\[Ww\]atchpoint.*buf.*Old value = .*$gdb_prompt $" { + pass "watchpoint on buf hit" + } + -re "Continuing.*$gdb_prompt $" { + fail "watchpoint on buf hit" + } + -re ".*$gdb_prompt $" { fail "watchpoint on buf hit" ; return } + timeout { fail "watchpoint on buf hit (timeout)" ; return } + eof { fail "watchpoint on buf hit (eof)" ; return } + } +} + # Start with a fresh gdb. gdb_exit @@ -842,6 +860,8 @@ if [initialize] then { } test_watchpoint_and_breakpoint + + test_watchpoint_in_big_blob } # Restore old timeout Index: gdb/gdb/valarith.c =================================================================== --- gdb.orig/gdb/valarith.c 2009-12-23 17:44:23.000000000 -0200 +++ gdb/gdb/valarith.c 2009-12-23 17:53:15.000000000 -0200 @@ -1397,6 +1397,24 @@ value_equal (struct value *arg1, struct } } +/* Compare values based on their raw contents. Useful for arrays since + value_equal coerces them to pointers, thus comparing just the address + of the array instead of its contents. */ + +int +value_equal_contents (struct value *arg1, struct value *arg2) +{ + struct type *type1, *type2; + + type1 = check_typedef (value_type (arg1)); + type2 = check_typedef (value_type (arg2)); + + return (TYPE_CODE (type1) == TYPE_CODE (type2) + && TYPE_LENGTH (type1) == TYPE_LENGTH (type2) + && memcmp (value_contents (arg1), value_contents (arg2), + TYPE_LENGTH (type1)) == 0); +} + /* Simulate the C operator < by returning 1 iff ARG1's contents are less than ARG2's. */ Index: gdb/gdb/value.h =================================================================== --- gdb.orig/gdb/value.h 2009-12-23 17:53:47.000000000 -0200 +++ gdb/gdb/value.h 2009-12-23 17:58:36.000000000 -0200 @@ -563,6 +563,8 @@ extern struct internalvar *lookup_intern extern int value_equal (struct value *arg1, struct value *arg2); +extern int value_equal_contents (struct value *arg1, struct value *arg2); + extern int value_less (struct value *arg1, struct value *arg2); extern int value_logical_not (struct value *arg1); --Boundary-00=_srqMLa9aqQ2p7mU--