Re: Bug in memset() causing quake3 to crash.

[Andreas Jaeger]

Ulrich Drepper <address@hidden> writes:

> Han <address@hidden> writes:
> 
>> Hello,
>> 
>> In recent versions of the kernel a bug apeared which caused quake3 to
>> segfault during startup. It was discussed on this mailing list and
>> recognized by Alan Cox:
>> 
>> http://opensource.creative.com/pipermail/emu10k1-devel/2001-September/004884.html
> 
> There is no desription of a problem.  And there will be no bug found
> in memset.  It's most definitely just another incapable programmer who
> assumes that memory can be write-only.

The report reads:

   aha.  I solved the quake issue.  so, quake3 mmap()'s the sound device with
   PROT_WRITE, then calls memset(mem, 0, size) .. and on the i586 and i686 the
   memset() routine in glibc (I'm using redhat's .i686.rpm glibc rpm) does
   a read on the memory because of caching issues.  This was fixed by doing an
   LD_PRELOAD of a little dummy memset.so containing a simple C implementation
   of memset().  The bug has been reported to the iD linux people.

And therefore I agree with Ulrich - the memory cannot be write-only.

Andreas
-- 
 Andreas Jaeger
  SuSE Labs address@hidden
   private address@hidden
    http://www.suse.de/~aj