Return to BSD News archive
Xref: sserve comp.os.386bsd.bugs:2328 comp.os.386bsd.questions:11527 Newsgroups: comp.os.386bsd.bugs,comp.os.386bsd.questions Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!bunyip.cc.uq.oz.au!munnari.oz.au!spool.mu.edu!agate!howland.reston.ans.net!vixen.cso.uiuc.edu!usenet.ucs.indiana.edu!bigbang.astro.indiana.edu!ahabig From: ahabig@bigbang.astro.indiana.edu (Alec Habig) Subject: Re: Floating point bug in FreeBSD? Message-ID: <Csxrq3.BCC@usenet.ucs.indiana.edu> Sender: news@usenet.ucs.indiana.edu (USENET News System) Nntp-Posting-Host: bigbang.astro.indiana.edu Organization: Indiana University Astrophysics, Bloomington, IN References: <steve.774119810@ichips.intel.com> <301us7$2qp@panix3.panix.com> Date: Thu, 14 Jul 1994 15:15:39 GMT Lines: 19 On this subject - I have an older 386 board that doesn't handle the 387 error IRQ properly. Under DOS and OS/2, my 387 works perfectly, as those OS's (or the software I'm running) don't use the IRQ to check for bad math. However, the FreeBSD kernel npx probe checks for a response on this IRQ at boot time, and if it doesn't get one, uses the emulation instead of the npx device. My question - what Bad Things will happen if I hack npx.c to skip this error check, allowing the kernel to blindly use the math chip I invested my $$ in? It seems that if the software that uses the chip checks for divide by 0's etc. before farming out the calculations (as my DOS software seems to), then life is good. But programs which rely on the IRQ to detect bad math would have problems. Does gcc have such an idiot filter built in? Thanks, Alec