From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 7524 invoked by alias); 15 Sep 2003 14:39:36 -0000 Mailing-List: contact gdb-patches-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: gdb-patches-owner@sources.redhat.com Received: (qmail 7450 invoked from network); 15 Sep 2003 14:39:34 -0000 Received: from unknown (HELO nevyn.them.org) (66.93.172.17) by sources.redhat.com with SMTP; 15 Sep 2003 14:39:34 -0000 Received: from drow by nevyn.them.org with local (Exim 4.22 #1 (Debian)) id 19yuVp-00067a-ID; Mon, 15 Sep 2003 10:39:33 -0400 Date: Mon, 15 Sep 2003 14:39:00 -0000 From: Daniel Jacobowitz To: gdb-patches@sources.redhat.com Cc: binutils@sources.redhat.com, gcc-patches@gcc.gnu.org Subject: [libiberty and gdb] floatformat_is_valid Message-ID: <20030915143933.GA22129@nevyn.them.org> Mail-Followup-To: gdb-patches@sources.redhat.com, binutils@sources.redhat.com, gcc-patches@gcc.gnu.org Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.1i X-SW-Source: 2003-09/txt/msg00325.txt.bz2 GDB's "struct value" has a union in it. This union, named "aligner", has a long double member as well as several others; its goal is to be a cheap way to ensure that we can load all sorts of values straight from the value structure, without alignment problems. Unfortunately, this means that when debugging GDB we tend to feed random bytes to printf ("%.35Lg"). That's not kosher, and on i386 I've found a whole bunch of numbers which can crash it - the gist is that GMP shifts normalized numbers over until it hits a one bit, and if the explicit integer bit isn't set it walks right out of the number. This could show up in other places too. Anywhere you print floating-point data from the inferior, really. So here's a patch which checks for the problem. I added floatformat_is_valid to libiberty/floatformat.c. Is this OK? [Do I need approval? Floatformat.c is in libiberty, but marked as part of GDB.] Oh, this patch misses some copyright year updates. I'll get them before checking in. -- Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer 2003-09-15 Daniel Jacobowitz * floatformat.h (floatformat_is_valid): Add prototype. 2003-09-15 Daniel Jacobowitz * floatformat.c (floatformat_is_valid): New function. (get_field, put_field): Correct comments. 2003-09-15 Daniel Jacobowitz * values.c (unpack_double): Call floatformat_is_valid. Index: gdb/values.c =================================================================== RCS file: /cvs/src/src/gdb/values.c,v retrieving revision 1.54 diff -u -p -r1.54 values.c --- gdb/values.c 14 Sep 2003 16:32:14 -0000 1.54 +++ gdb/values.c 15 Sep 2003 14:30:58 -0000 @@ -739,7 +739,14 @@ unpack_double (struct type *type, const also not defined either. Oops! Hopefully someone will add both the missing floatformat - definitions and floatformat_is_invalid() function. */ + definitions and the new cases for floatformat_is_valid (). */ + + if (!floatformat_is_valid (floatformat_from_type (type), valaddr)) + { + *invp = 1; + return 0.0; + } + return extract_typed_floating (valaddr, type); } else if (nosign) Index: libiberty/floatformat.c =================================================================== RCS file: /cvs/src/src/libiberty/floatformat.c,v retrieving revision 1.7 diff -u -p -r1.7 floatformat.c --- libiberty/floatformat.c 15 Apr 2003 21:29:34 -0000 1.7 +++ libiberty/floatformat.c 15 Sep 2003 14:30:58 -0000 @@ -149,7 +149,7 @@ static unsigned long get_field PARAMS (( unsigned int, unsigned int)); -/* Extract a field which starts at START and is LEN bytes long. DATA and +/* Extract a field which starts at START and is LEN bits long. DATA and TOTAL_LEN are the thing we are extracting it from, in byteorder ORDER. */ static unsigned long get_field (data, order, total_len, start, len) @@ -273,7 +273,7 @@ static void put_field PARAMS ((unsigned unsigned int, unsigned long)); -/* Set a field which starts at START and is LEN bytes long. DATA and +/* Set a field which starts at START and is LEN bits long. DATA and TOTAL_LEN are the thing we are extracting it from, in byteorder ORDER. */ static void put_field (data, order, total_len, start, len, stuff_to_put) @@ -402,6 +402,39 @@ floatformat_from_double (fmt, from, to) mant_off += mant_bits; mant_bits_left -= mant_bits; } +} + +/* Return non-zero iff the data at FROM is a valid number in format FMT. */ + +int +floatformat_is_valid (fmt, from) + const struct floatformat *fmt; + char *from; +{ + if (fmt == &floatformat_i387_ext) + { + /* In the i387 double-extended format, if the exponent is all + ones, then the integer bit must be set. If the exponent + is neither 0 nor ~0, the intbit must also be set. Only + if the exponent is zero can it be zero, and then it must + be zero. */ + unsigned int exponent, int_bit; + unsigned char *ufrom = (unsigned char *) from; + + exponent = get_field (ufrom, fmt->byteorder, fmt->totalsize, + fmt->exp_start, fmt->exp_len); + int_bit = get_field (ufrom, fmt->byteorder, fmt->totalsize, + fmt->man_start, 1); + + if ((exponent == 0) != (int_bit == 0)) + return 0; + else + return 1; + } + + /* Other formats with invalid representations should be added + here. */ + return 1; } Index: include/floatformat.h =================================================================== RCS file: /cvs/src/src/include/floatformat.h,v retrieving revision 1.5 diff -u -p -r1.5 floatformat.h --- include/floatformat.h 6 Mar 2002 06:29:41 -0000 1.5 +++ include/floatformat.h 15 Sep 2003 14:30:58 -0000 @@ -118,4 +118,9 @@ extern void floatformat_from_double PARAMS ((const struct floatformat *, double *, char *)); +/* Return non-zero iff the data at FROM is a valid number in format FMT. */ + +extern int +floatformat_is_valid PARAMS ((const struct floatformat *fmt, char *from)); + #endif /* defined (FLOATFORMAT_H) */