Do you have a question? Post it now! No Registration Necessary
Subject
- Posted on
Divide by zero does not crash
- 02-07-2008
February 7, 2008, 11:02 pm

I noticed that my application does not crash when there is a divide by
zero. For example, the following test code:
int i;
i = 10/0;
printf("i = %d\n", i);
gives:
i = 0
It does not crash.
Is there a way to make the application crash when there is a divide by
zero? I am running Linux 2.4.26 on a powerpc (MPC8248). I compile
with powerpc-linux-gcc.
zero. For example, the following test code:
int i;
i = 10/0;
printf("i = %d\n", i);
gives:
i = 0
It does not crash.
Is there a way to make the application crash when there is a divide by
zero? I am running Linux 2.4.26 on a powerpc (MPC8248). I compile
with powerpc-linux-gcc.

Re: Divide by zero does not crash

This does give a warning at compile time because the constant zero is
there. What concerns me is when the denominator is a variable that
takes the value of zero while the application is running. There is
neither a compilation warning nor any indication that a divide by zero
is occurring when the application is running.

Re: Divide by zero does not crash

The behavior of integer division by 0 is undefined per the C Language
standard and, hence, implementation-dependent. On x86, division by 0
produces a hardware exception. On x86 linux, this exception is presented
to the program as a signal. However, a conforming implementation may
choose to do anything else that is convenient: nothing (leave quotient
unmodified), produce random value for the quotient, kill the process.
GH

Re: Divide by zero does not crash

The compiler could insert code that crashes (or produces a signal or
whatever behaviour is deemed appropriate) at run-time. But it does
not, because the behaviour when not doing an evaluation at
compile-time is not to crash or produce a signal. In contrast, on an
AMD64 box I also get a warning at compile-time, and a SIGFPE at
run-time.

Yes, that's a property of the PowerPC archictecture. AFAIK the theory
of the hardware designers was that the compiler should produce code
equivalent to:
q=n/d;
if (d == 0)
raise(SIGFPE);
The idea is that the check would be performed during the latency of
the division, so it would usually not cost extra time. So one could
make the hardware simpler by not putting in the check there. I may be
confusing PowerPC with another RISC architecture (most likely MIPS)
wrt this aspect, though.
But obviously gcc does not do this (there does not even seem to be an
option to turn this on). Do other C compilers for PowerPC (from IBM,
Apple or Motorola generate such code)?
Followups set to comp.os.linux.powerpc.
- anton
--
M. Anton Ertl Some things have to be seen to be believed
snipped-for-privacy@mips.complang.tuwien.ac.at Most things have to be believed to be seen
M. Anton Ertl Some things have to be seen to be believed
snipped-for-privacy@mips.complang.tuwien.ac.at Most things have to be believed to be seen
We've slightly trimmed the long signature. Click to see the full one.
Site Timeline
- » Call For Presentation: LSM '08 Conference on free software for Embedded Systems
- — Next thread in » Embedded Linux
-
- » [Commercial] Safesquid Version 4.2.2.RC7 released
- — Previous thread in » Embedded Linux
-
- » Crosscompiling for ARM: reloc type R_ARM_ABS32 is not supported for PIC - ...
- — Newest thread in » Embedded Linux
-
- » ottenere alim duale da trafo singolo
- — The site's Newest Thread. Posted in » Electronics Hobby (Italian)
-