Xref: utzoo comp.unix.admin:1400 comp.unix.internals:2424 comp.unix.programmer:1432 Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!auspex!guy From: guy@auspex.auspex.com (Guy Harris) Newsgroups: comp.unix.admin,comp.unix.internals,comp.unix.programmer Subject: Re: How do you make your UNIX crash ??? Message-ID: <6822@auspex.auspex.com> Date: 26 Mar 91 07:59:27 GMT References: <690@tndsyd.oz.au> <513@bria> <3442@engadm3.csd.mot.com> Followup-To: alt.religion.computers Organization: Auspex Systems, Santa Clara Lines: 39 >Awhile back, someone found that executing random data made quite >a few RISC chips sieze up. No, a while back, someone found that executing random data made some operating systems running on machines with RISC chips crash, and immediately charged ahead and, on the basis of that small bit of partial data and a lot of assumptions about RISC chip design methodology, speculated that this was caused by RISC-chip designers not being as rigorous as CISC-chip designers in testing their chips. Subsequent to this: 1) the same program was found to crash the OS on at least one machine running a *CISC* chip (a 80386 machine, I think); 2) at least in one case, the problem was a bug in the *OS*'s code to deal with, as I remember, illegal instructions in the delay slots of illegal branches, or something like that (this was in the MIPS version). At some point, I may dive in and see what caused SunOS to barf; I suspect it's a bug in the floating-point simulation code (which may get invoked even on SPARCs with an FPU, as the FPU may not implement every single floating-point instruction in the architecture). If one wishes to consider the code that broke on the RISC machines to be low-enough-level support code that it "should" be considered as much a part of the architecture's implementation as would the chip itself, you could, I guess, flame RISC - or, at least, the folks doing the software part of the implementation. Of course, given that, you can probably find plenty of microcode bugs to damn CISC as well, if your goal is to bash some particular architecture style (the posting in which the person revealed the results of his test had a bit of a RISC-bashing flavor to it). (Followups directed to "alt.religion.computers", if you really feel you *MUST* make your 2 small monetary units worth known on the RISC vs. CISC topic.)