Debug problem when step debug float point instruction that raise an float point exception

Hi All,

I have a program which use _control87 and signal function to set handle
function for float point exception. The program runs fine if it’s not in
debug environment.
However, In Windows 98, If I debug the program in VC IDE, It always
cause an 0x0E exception when I step the line “r = n1 / n2” which cause a
divid zero exception, after that, I must restart Windows 98 to recover
the system. While if I don’t step this line and just go through it, the
program runs fine and both the program and VC debugger can handle the
float point exception correctlly.

The problem doesn’t exist in Windows NT.

The 0x0E exception is occurred exactlly when I step over the float point
instruction FDIV in VC debugger. According to some documentation, the
float point exception aren’t raised until the next float point
instruction after the actual bad instruction. I supposed that kernel
debug interrupt routine may use some float point instruction, hence CPU
will raise an float point exception inside debug interrupt handler, the
nested interrupt may cause VC debugger work incorrectly.

At fiste, I suspend that the problem may caused by setjmp/longjmp, but
even if I use structure exception handler to handle float point
exception, the problem still exist.

Is it a VC debugger problem or Windows 98 system limitation (because it
doesn’t handle debug interrupt gracefully). Could any one kindly help me
regarding this question?

Best regards,

The following is my code.

/* FPRESET.C: This program uses signal to set up a
* routine for handling floating-point errors.
*/

#include <stdio.h>
#include <signal.h>
#include <setjmp.h>
#include <stdlib.h>
#include <float.h>
#include <math.h>
#include <string.h>

jmp_buf mark; /* Address for long jump to jump to /
int fperr; /
Global error number /

void __cdecl fphandler( int sig, int num ); /
Prototypes /
void fpcheck( void );

void main( void )
{
double n1, n2, r;
int jmpret;
/
Unmask all floating-point exceptions. /
_control87( 0, _MCW_EM );
/
Set up floating-point error handler. The compiler
* will generate a warning because it expects
* signal-handling functions to take only one argument.
/
if( signal( SIGFPE, fphandler ) == SIG_ERR )

{
fprintf( stderr, “Couldn’t set SIGFPE\n” );
abort(); }

/
Save stack environment for return in case of error. First
* time through, jmpret is 0, so true conditional is executed.
* If an error occurs, jmpret will be set to -1 and false
* conditional will be executed.
/
jmpret = setjmp( mark );
if( jmpret == 0 )
{
printf( "Test for invalid operation - " );
printf( "enter two numbers: " );
scanf( “%lf %lf”, &n1, &n2 );
r = n1 / n2;
/
This won’t be reached if error occurs. /
printf( “\n\n%4.3g / %4.3g = %4.3g\n”, n1, n2, r );

r = n1 * n2;
/
This won’t be reached if error occurs. /
printf( “\n\n%4.3g * %4.3g = %4.3g\n”, n1, n2, r );
}
else
fpcheck();
}
/
fphandler handles SIGFPE (floating-point error) interrupt. Note
* that this prototype accepts two arguments and that the
* prototype for signal in the run-time library expects a signal
* handler to have only one argument.

* The second argument in this signal handler allows processing of
* _FPE_INVALID, _FPE_OVERFLOW, _FPE_UNDERFLOW, and
* _FPE_ZERODIVIDE, all of which are Microsoft-specific symbols
* that augment the information provided by SIGFPE. The compiler
* will generate a warning, which is harmless and expected.

/
void fphandler( int sig, int num )
{
/* Set global for outside check since we don’t want
* to do I/O in the handler.
/
fperr = num;
/
Initialize floating-point package. /
_fpreset();
/
Restore calling environment and jump back to setjmp. Return
* -1 so that setjmp will return false for conditional test.
*/
longjmp( mark, -1 );
}
void fpcheck( void )
{
char fpstr[30];
switch( fperr )
{
case _FPE_INVALID:
strcpy( fpstr, “Invalid number” );
break;
case _FPE_OVERFLOW:
strcpy( fpstr, “Overflow” );

break;
case _FPE_UNDERFLOW:
strcpy( fpstr, “Underflow” );
break;
case _FPE_ZERODIVIDE:
strcpy( fpstr, “Divide by zero” );
break;
default:
strcpy( fpstr, “Other floating point error” );
break;
}
printf( “Error %d: %s\n”, fperr, fpstr );
}</string.h></math.h></float.h></stdlib.h></setjmp.h></signal.h></stdio.h>

Hi All,

I have a program which use _control87 and signal function to set handle
function for float point exception. The program runs fine if it’s not in
debug environment.
However, In Windows 98, If I debug the program in VC IDE, It always
cause an 0x0E exception when I step the line “r = n1 / n2” which cause a
divid zero exception, after that, I must restart Windows 98 to recover
the system. While if I don’t step this line and just go through it, the
program runs fine and both the program and VC debugger can handle the
float point exception correctlly.

The problem doesn’t exist in Windows NT.

The 0x0E exception is occurred exactlly when I step over the float point
instruction FDIV in VC debugger. According to some documentation, the
float point exception aren’t raised until the next float point
instruction after the actual bad instruction. I supposed that kernel
debug interrupt routine may use some float point instruction, hence CPU
will raise an float point exception inside debug interrupt handler, the
nested interrupt may cause VC debugger work incorrectly.

At fiste, I suspend that the problem may caused by setjmp/longjmp, but
even if I use structure exception handler to handle float point
exception, the problem still exist.

Is it a VC debugger problem or Windows 98 system limitation (because it
doesn’t handle debug interrupt gracefully). Could any one kindly help me
regarding this question?

Best regards,

Guangjun Zhou

The following is my code.

/* FPRESET.C: This program uses signal to set up a
* routine for handling floating-point errors.
*/

#include <stdio.h>
#include <signal.h>
#include <setjmp.h>
#include <stdlib.h>
#include <float.h>
#include <math.h>
#include <string.h>

jmp_buf mark; /* Address for long jump to jump to /
int fperr; /
Global error number /

void __cdecl fphandler( int sig, int num ); /
Prototypes /
void fpcheck( void );

void main( void )
{
double n1, n2, r;
int jmpret;
/
Unmask all floating-point exceptions. /
_control87( 0, _MCW_EM );
/
Set up floating-point error handler. The compiler
* will generate a warning because it expects
* signal-handling functions to take only one argument.
/
if( signal( SIGFPE, fphandler ) == SIG_ERR )

{
fprintf( stderr, “Couldn’t set SIGFPE\n” );
abort(); }

/
Save stack environment for return in case of error. First
* time through, jmpret is 0, so true conditional is executed.
* If an error occurs, jmpret will be set to -1 and false
* conditional will be executed.
/
jmpret = setjmp( mark );
if( jmpret == 0 )
{
printf( "Test for invalid operation - " );
printf( "enter two numbers: " );
scanf( “%lf %lf”, &n1, &n2 );
r = n1 / n2;
/
This won’t be reached if error occurs. /
printf( “\n\n%4.3g / %4.3g = %4.3g\n”, n1, n2, r );

r = n1 * n2;
/
This won’t be reached if error occurs. /
printf( “\n\n%4.3g * %4.3g = %4.3g\n”, n1, n2, r );
}
else
fpcheck();
}
/
fphandler handles SIGFPE (floating-point error) interrupt. Note
* that this prototype accepts two arguments and that the
* prototype for signal in the run-time library expects a signal
* handler to have only one argument.

* The second argument in this signal handler allows processing of
* _FPE_INVALID, _FPE_OVERFLOW, _FPE_UNDERFLOW, and
* _FPE_ZERODIVIDE, all of which are Microsoft-specific symbols
* that augment the information provided by SIGFPE. The compiler
* will generate a warning, which is harmless and expected.

/
void fphandler( int sig, int num )
{
/* Set global for outside check since we don’t want
* to do I/O in the handler.
/
fperr = num;
/
Initialize floating-point package. /
_fpreset();
/
Restore calling environment and jump back to setjmp. Return
* -1 so that setjmp will return false for conditional test.
*/
longjmp( mark, -1 );
}
void fpcheck( void )
{
char fpstr[30];
switch( fperr )
{
case _FPE_INVALID:
strcpy( fpstr, “Invalid number” );
break;
case _FPE_OVERFLOW:
strcpy( fpstr, “Overflow” );

break;
case _FPE_UNDERFLOW:
strcpy( fpstr, “Underflow” );
break;
case _FPE_ZERODIVIDE:
strcpy( fpstr, “Divide by zero” );
break;
default:
strcpy( fpstr, “Other floating point error” );
break;
}
printf( “Error %d: %s\n”, fperr, fpstr );
}</string.h></math.h></float.h></stdlib.h></setjmp.h></signal.h></stdio.h>