From owner-freebsd-smp Wed Dec 4 08:50:33 1996 Return-Path: owner-smp Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id IAA29068 for smp-outgoing; Wed, 4 Dec 1996 08:50:33 -0800 (PST) Received: from bacall.lodgenet.com (bacall.lodgenet.com [205.138.147.242]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id IAA29059 for ; Wed, 4 Dec 1996 08:50:29 -0800 (PST) Received: (from mail@localhost) by bacall.lodgenet.com (8.6.12/8.6.12) id KAA04604 for ; Wed, 4 Dec 1996 10:50:03 -0600 Received: from garbo.lodgenet.com(204.124.123.250) by bacall via smap (V1.3) id sma004586; Wed Dec 4 10:49:41 1996 Received: from jake.lodgenet.com (jake.lodgenet.com [10.0.11.30]) by garbo.lodgenet.com (8.6.12/8.6.9) with ESMTP id KAA24650 for ; Wed, 4 Dec 1996 10:49:48 -0600 Received: from jake.lodgenet.com (localhost [127.0.0.1]) by jake.lodgenet.com (8.8.3/8.6.12) with ESMTP id KAA18601 for ; Wed, 4 Dec 1996 10:49:59 -0600 (CST) Message-Id: <199612041649.KAA18601@jake.lodgenet.com> X-Mailer: exmh version 1.6.9 8/22/96 To: smp@freebsd.org Subject: INVLTLB and stuff Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Date: Wed, 04 Dec 1996 10:49:59 -0600 From: "Eric L. Hernes" Sender: owner-smp@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Hi guys, I decided to go for it all and test with APIC_IO and SMP_INVLTLB first, then back off if necessary. during a kernel compile: (ttyp0@jake)$ vmstat -i interrupt total rate clk0 irq2 744193 431 rtc0 irq8 214594 124 pci irq11 17629 10 sc0 irq1 804 0 sio1 irq3 32319 18 lpt0 irq7 1 0 ed0 irq5 14927 8 sb0 irq9 321 0 stray irq13 1 0 Total 1024789 594 and the tail of dmesg: DEVFS: ready to run Enabled INTs: 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, imen: 0x00ffd401 SMP: AP CPU #1 LAUNCHED!! Starting Scheduling... sio1: 1 more silo overflow (total 1) sio1: 1 more silo overflow (total 2) sio1: 1 more silo overflow (total 3) sio1: 1 more silo overflow (total 4) sio1: 25 more tty-level buffer overflows (total 25) sio1: 92 more tty-level buffer overflows (total 117) sio1: 25 more tty-level buffer overflows (total 142) sio1: 12 more tty-level buffer overflows (total 154) sio1: 11 more tty-level buffer overflows (total 165) sio1: 10 more tty-level buffer overflows (total 175) sio1: 1 more silo overflow (total 5) sio1: 1 more silo overflow (total 6) ed0: device timeout ed0: device timeout ed0: device timeout ed0: device timeout sio1 is the mouse. system responsiveness is very choppy while anything significant is running, even a `make depend' So I removed the SMP_INVLTLB. Is it legal to just comment out #define SMP_INVLTLB 1 in opt_smp_invltlb.h? Or do I need to re-config? kernel without SMP_INVLTLB: (ttyp1@jake)$ vmstat -i interrupt total rate clk0 irq2 1244898 1144 ^^^^ Holy Christmas, what's up with that? rtc0 irq8 139347 128 pci irq11 25834 23 fdc0 irq6 2962 2 sc0 irq1 3278 3 sio1 irq3 53092 48 ed0 irq5 34118 31 sb0 irq9 372 0 stray irq13 1 0 Total 1503902 1382 but no lost interrupts, and the network timeouts are gone. kernel compile with j1: 1265.95 real 728.45 user 280.91 sys kernel compile with j8: 658.73 real 866.27 user 256.26 sys the machine was doing other compiles and work at the same time, so this isn't a very great indication of performance, but it's looking better! eric. -- erich@lodgenet.com http://rrnet.com/~erich erich@rrnet.com