From owner-freebsd-stable Fri Jan 8 11:29:38 1999 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id LAA12370 for freebsd-stable-outgoing; Fri, 8 Jan 1999 11:29:38 -0800 (PST) (envelope-from owner-freebsd-stable@FreeBSD.ORG) Received: from versa.eng.comsat.com (versa.eng.comsat.com [134.133.169.5]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id LAA12364 for ; Fri, 8 Jan 1999 11:29:33 -0800 (PST) (envelope-from marc@versa.eng.comsat.com) Received: (from marc@localhost) by versa.eng.comsat.com (8.8.8/8.7.3) id PAA22328; Fri, 8 Jan 1999 15:12:25 -0500 (EST) Message-ID: X-Mailer: XFMail 1.1 [p0] on FreeBSD Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 8bit MIME-Version: 1.0 In-Reply-To: <19990108094015.A10590@wattres.Watt.COM> Date: Fri, 08 Jan 1999 14:57:26 -0500 (EST) Organization: Comsat Mobile Communications From: Marc Giannoni To: Steve Watt Subject: Re: IRQ entropy causes panics? Cc: freebsd-stable@FreeBSD.ORG, Mark Murray , Chris Dillon Sender: owner-freebsd-stable@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.ORG Hooking interrupts is a "real time" issue. There is no way an OS, however will written, can tolerate flagrant additions to it's interrupt latencies. ("j. random sysadmin run-amok") I'm suprised that the randomizer code can hook interrupts at all and still keep the OS stable. This stuff is some pretty clever code. You are actually modifying the kernel's core execution by enabling these hooks! Trust me when I say that there probably is no "Underlying Problem". It's pretty amazing that this thing even exists. On 08-Jan-99 Steve Watt wrote: >On Fri, Jan 08, 1999 at 11:10:52AM -0600, Chris Dillon wrote: >> On Fri, 8 Jan 1999, Mark Murray wrote: >> >> > Steve Watt wrote: >> > > [ serial irq running 800int/sec ] >> > >> > That could cause problems; I'll look into it. >> > >> > > What it sounds like is that I shouldn't use "high rate" (for some >> > > definition thereof -- more than the clock irqs?) things to feed >> > > the entropy pool. That probably ought to be documented somewhere. >> > >> > I'll fix that too :-). > >I think that it would be better to fix the underlying problem, if possible; >during a news expire with NFS archiving, the PCI irq can run as high as >300-400/s for most of an hour. I suppose the safest things to hook >would be the keyboard irq (1) and the mouse (12 -- pms). But if there >really is some kind of race condition in there, which is how it smells, >then for max stability none is the right answer. > >> If I recall correctly, this _was_ documented before we moved from >> /etc/sysconfic to /etc/rc.conf (basically, becuse we went from >> "sections" with long descriptions for each option to single lines for >> each option). It also specifically mentioned not to hook interrupts >> attached to serial ports or the clock, I think. > >Interesting. It still seems that it oughta be safe for j. random >(pun half-intended) sysadmin to hook whatever they like. The timers >aren't good sources of entropy simply because of their periodicity. >The same really *isn't* true, even for high-bandwidth serial ports, >because there will always be the vagaries of network timing, non-busy >periods, etc. Seems like a pretty good source, except that it makes >the box fall over. > >To Unsubscribe: send mail to majordomo@FreeBSD.org >with "unsubscribe freebsd-stable" in the body of the message ---------------------------------- E-Mail: Marc Giannoni Date: 08-Jan-99 Time: 14:57:28 This message was sent by XF-Mail ---------------------------------- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message