gcc 3.1 i386 testing request

Joel Sherrill joel.sherrill at OARcorp.com
Thu Apr 25 20:59:01 UTC 2002


Eric Norum was testing gcc 3.1 preview tarballs on the i386
and we think he may have discovered a bad problem and we
need volunteers to help confirm this.  You would need to
have an i386 target with networking and be willing to
build a gcc 3.1 preview toolset.  Then use that toolset
to build RTEMS with networking and see if it works.

>From Eric Norum...

Does this happen for you, too?  I'd like to know that I'm not chasing a 
ghost.  The symptoms, when running the netdemo program are that you can 
ping the RTEMS target, but you can't telnet to it.  You can also see the 
problem if you run a packet sniffer and type a `u' to the netdemo 
program console -- you'll see a bunch of UDP packets from the RTEMS 
machine, all with bad IP checksums :-(

As best we can tell, gcc 3.1 previews may be generating incorrect
code for the optimized i386 in_cksum routines.

We are trying not to be alarmist and need help in confirming this
and tracking it down.  If this is a real problem in gcc, then it
would potentially impact any x86 target.  So we want to be sure
about the problem, file a good report, and try to get it fixed.

Eric.. didn't you have a framework at one point that ran both
the optimized and unoptimized versions outside the source tree
to compare their results?  Installing the native GNU/Linux x86
3.1 compiler and running that might be a quick doublecheck.

Joel Sherrill, Ph.D.             Director of Research & Development
joel at OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985

More information about the users mailing list