From freefall.cdrom.com!owner-freebsd-hackers Tue Aug 23 20:16:03 1994 Return-Path: Received: from freefall.cdrom.com by tfs.com (smail3.1.28.1) with SMTP id m0qd8cD-0003xsa; Tue, 23 Aug 94 20:03 PDT Received: (from root@localhost) by freefall.cdrom.com (8.6.8/8.6.6) id UAA06820 for freebsd-hackers-outgoing; Tue, 23 Aug 1994 20:00:31 -0700 Received: from Root.COM (implode.Root.COM [198.145.90.241]) by freefall.cdrom.com (8.6.8/8.6.6) with ESMTP id UAA06814 for ; Tue, 23 Aug 1994 20:00:24 -0700 Received: from corbin.Root.COM (corbin.Root.COM [198.145.90.2]) by Root.COM (8.6.8/8.6.5) with ESMTP id UAA00122; Tue, 23 Aug 1994 20:00:39 -0700 Received: from localhost (localhost [127.0.0.1]) by corbin.Root.COM (8.6.9/8.6.5) with SMTP id TAA00449; Tue, 23 Aug 1994 19:59:59 -0700 Message-Id: <199408240259.TAA00449@corbin.Root.COM> X-Authentication-Warning: corbin.Root.COM: Host localhost didn't use HELO protocol To: batie@agora.rdrop.com, marc@dev.com cc: freebsd-hackers@freefall.cdrom.com Subject: apply this patch file From: David Greenman Reply-To: davidg@Root.COM Date: Tue, 23 Aug 1994 19:59:58 -0700 Sender: freebsd-hackers-owner@freefall.cdrom.com Precedence: bulk Status: RO 1) sio.c Disables automatic fifo trigger level adjustment. I don't think this was ever a good idea. This is the likely cause of the sio serial problems on puptent. I think the auto adjustment code is evil. Here's the scenario: You get one or two sio overflows on one or two of the ports. The auto adjustment code reduces the trigger level basically in half on those ports. Since you have lots of ports on the machine, this increases the load just enough to cause a few more overflows. This starts to reduce the trigger level toward one (completely disabling the fifo). This causes the load to increase dramatically (14 times higher) which causes overflows to occur on other ports. These port's triggers are decreased to 1 and then the whole damn thing goes south. With 4 or 5 active ports, this whole process can take less than 2 seconds. 2) kern_time.c Fixes a bug that causes NTP to not properly synchronize. 3) spec_vnops.c Fixes a bug that will cause a panic if the timing of a tty close is just right. This was the cause of agora's crashes on the first day of operation under 1.1.5. -DG Index: src/sys/i386/isa/sio.c =================================================================== RCS file: /home/cvs/386BSD/src/sys/i386/isa/sio.c,v retrieving revision 1.56 diff -c -r1.56 sio.c *** 1.56 1994/06/16 08:08:44 --- sio.c 1994/08/24 02:30:38 *************** *** 1389,1394 **** --- 1389,1395 ---- "sio%d: %u more %s%s (total %lu)\n", unit, delta, error_desc[errnum], delta == 1 ? "" : "s", total); + #if 0 if (errnum == CE_OVERRUN && com->hasfifo && com->ftl > FIFO_TRIGGER_1) { static u_char ftl_in_bytes[] = *************** *** 1406,1411 **** --- 1407,1413 ---- ftl_in_bytes[com->ftl / FIFO_TRIGGER_DELTA]); } + #endif } } if (com->state & CS_ODONE) { Index: src/sys/kern/kern_time.c =================================================================== RCS file: /home/cvs/386BSD/src/sys/kern/kern_time.c,v retrieving revision 1.5 diff -c -r1.5 kern_time.c *** 1.5 1994/06/22 05:52:47 --- kern_time.c 1994/08/04 12:49:27 *************** *** 103,111 **** return (error); /* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */ boottime.tv_sec += atv.tv_sec - time.tv_sec; ! s = splclock(); ! time.tv_sec = atv.tv_sec; /* XXX avoid skew in tv_usec */ ! splx(s); resettodr(); } if (uap->tzp && (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, --- 103,109 ---- return (error); /* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */ boottime.tv_sec += atv.tv_sec - time.tv_sec; ! s = splhigh(); time = atv; splx(s); resettodr(); } if (uap->tzp && (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, Index: src/sys/kern/spec_vnops.c =================================================================== RCS file: /home/cvs/386BSD/src/sys/kern/spec_vnops.c,v retrieving revision 1.4 diff -c -r1.4 spec_vnops.c *** 1.4 1994/05/30 03:20:08 --- spec_vnops.c 1994/08/24 02:26:18 *************** *** 137,145 **** if ((u_int)maj >= nchrdev) return (ENXIO); VOP_UNLOCK(vp); ! vp->v_opencount++; error = (*cdevsw[maj].d_open)(dev, mode, S_IFCHR, p); ! --vp->v_opencount; VOP_LOCK(vp); return (error); --- 137,147 ---- if ((u_int)maj >= nchrdev) return (ENXIO); VOP_UNLOCK(vp); ! if (vp->v_specinfo) ! vp->v_opencount++; error = (*cdevsw[maj].d_open)(dev, mode, S_IFCHR, p); ! if (vp->v_specinfo) ! --vp->v_opencount; VOP_LOCK(vp); return (error);