Linux math exception handling

Linux math exception handling

Post by Kirk D. Bene » Thu, 01 Jun 1995 04:00:00



Hi,
        I'm currently porting an interactive software package to Linux that
requires the ability to handle math exceptions as follows:

        - When traps are enabled, the program catches SIGFPE and or
          SIGTRAP, decodes the type of exception that occurred, and then
          allows execution to continue. The result of the computation is
          undefined, although it would be nice if fixups were performed
          to yield the standard IEEE special values (Nan, Inf) for the result.

        - When traps are disabled, no exceptions are reported, and the
          hardware/kernel arrange for the special IEEE values to
          result as necessary. Afterwards, it should be possible to
          query the hardware to find out which exceptions occurred (usually
          from a status word inside the FPU) and reset the status.

This is an area in which Unix systems differ. Typically one enables traps
by calling a library routine (ieee_handler() on a Sun for example) and
registering a signal handler to catch SIGFPE (and sometimes SIGTRAP) and
do the required decoding. Sometimes the signal handler must also advance
the program counter and/or perform fixups of the resulting value to the
IEEE special values.

As an experiment, I used sigaction() to set up a SIGFPE signal handler.
I'm using kernel version 1.1.59 from slackware version 2.1.
Linux calls my handler when exceptions occur, but I see the following
problems:

        - Integer exceptions don't automatically advance the PC past the
          offending computation, so the handler gets called repeatedly in
          an infinite loop.
        - Results don't get the special IEEE values (This is true even
          if no signal handler is set).
        - I'm not sure what extra information is passed to the handler
          by Linux, and am therefore not sure how to decode the reason for
          the exception.

I'm having trouble finding any documentation on the proper way to do
this sort of thing under Linux. In general I have the following questions,
which boil down to "Where can I find out how exceptions work under Linux?":

        [] What is the proper method to use to establish an exception
           handler for math exceptions. Is it as simple as calling sigaction(),
           or is there a library function that should be used?
        [] Once a math exception is detected, what extra information is passed
           to the signal handler and how does one decode it to determine what
           type of exception occurred. (ie divide by 0, underflow...etc)?
        [] If exception handling is off, is there a way to query the system
           to determine the accumulated math errors and then clear them?
               ** Does this have something to do with the __fpu_control word
                  and the __setfpucw() functions in libc.a?
        [] How does one arrange for the results to be fixed up to the
           IEEE special values such as Nan and Inf?
        [] Which exceptions require the signal handler to increment
           the program counter, and what is the proper method to do it?
           (For example, under Solaris, integer exceptions must be manually
           bumped. This is done by setting fields in the context structure
           that is passed to the handler).

Any help would be greatly appreciated.


 
 
 

1. How to handle math exception/SIGFPE?

Hello all,
I'm working on the Linux version of POV-Ray 3.0, and I am unable to
determine how to handle floating point problems under Linux (ie
under/overflow, divide-by-zero, singularity).  With many BSD Unices,
it is possible to define a function called matherr() which is called
if there is a floating point exception.  Inside matherr(), you can
define return values for the math function depending on the type
of error (ie zero for underflow, DBL_MAX for overflow, etc).  Does
anyone know how this is done under Linux?  I have considered trapping
the SIGFPE, but this doesn't allow one to fix the return values.

I don't have these problems on my local machine, as it seems the kernel
was compiled with the FPU exceptions turned off, but other people have
reported a problem to me, and I'm looking for a way to handle the
exceptions without asking the users to re-compile the kernel.

Cheers, Andreas.
--
Andreas Dilger   University of Calgary  \"If a man ate a pound of pasta and
(403) 220-8792   Micronet Research Group \ a pound of antipasto, would they
Dept of Electrical & Computer Engineering \   cancel out, leaving him still
http://www-mddsp.enel.ucalgary.ca/People/adilger/       hungry?" -- Dogbert

2. Jack Schofield talks through his other orifice !

3. ?How can i handle math. exception?

4. Q:PLEASE HELP !!! Routing and Subnet Problem

5. Trapping math exceptions in C / Linux?

6. mars, novel server on linux

7. Machine Check exception handling in linux

8. Linux woes...

9. Porting MS Structured Exception Handling to Linux.

10. Math library exceptions

11. Math coprocessor exception 16 error???

12. Exception handling support in GNU C++ ??

13. Exception Handling with Sun's Workshop Compiler Version 6 Update 2