Date: Sun, 26 Jul 1998 11:58:03 -0500 (EST) From: "John S. Dyson" <dyson@iquest.net> To: toasty@home.dragondata.com (Kevin Day) Cc: tlambert@primenet.com, dg@root.com, current@FreeBSD.ORG Subject: Re: panic: page fault while in kernel mode - from gcc Message-ID: <199807261658.LAA01185@dyson.iquest.net> In-Reply-To: <199807231959.OAA24934@home.dragondata.com> from Kevin Day at "Jul 23, 98 02:59:08 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
Kevin Day said: > > > >> This seems to indicate that the new process's page directory either hasn't > > > >> been allocated yet or went away for some reason. Apparantly one of several > > > >> new bugs that John has left us with. > > > >> > > > > > > > >Is there any reason why someone doesn't back out his last few batches of > > > >large vm changes? Apparently he was half done with some of it, and what we > > > >had before seemed more stable than what we have now. > > > > > > That may ultimately be what we have to do, but the fixes also fixed some > > > serious 'leak' style problems with the Mach derived VM system, so I'd rather > > > that we find the bugs and fix them rather than going back to the previous > > > code. > > > > FreeBSD doesn't currently include the batch of patches that Elvind > > saved and that John had in his home directory when it was removed, > > does it? > > > > What is the behaviour with those patches added in? > > > > Right now, with my -current server, I get a 'panic: page fault while in > kernel mode' about once a week.... Dual Processor, P/200 with 256M of ram. > Most of the problems seem to occur when the system is starting to swap.. > (obviously) > Ahha!!! There is a *severe* bug in the current SMP pmap code, that it is important to preemptively stop all processors before updating any of the shared pmap entries (actually, that isn't totally necessary, but is a solution to the problem.) It is a combo of P6/PII errata, and the fact that lazy TLB updates are almost impossible to get right. I think that my patches as distributed might have the non-lazy TLB update code in them. If someone is bona-fide going to fix the problem, I can pass them the relevent fixes (I can help with that one -- since it is just distilling out existant fixes, and won't require the endless boot/reboot cycle :-)). One caveat: removing the lazy TLB updates appears to open up other bugs elsewhere. -- John | Never try to teach a pig to sing, dyson@iquest.net | it makes one look stupid jdyson@nc.com | and it irritates the pig. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199807261658.LAA01185>