From owner-freebsd-bugs Fri Oct 22 21:18:16 1999 Delivered-To: freebsd-bugs@freebsd.org Received: from sir-alan.chem.indiana.edu (sir-alan.chem.indiana.edu [129.79.137.215]) by hub.freebsd.org (Postfix) with ESMTP id 0EA1B14D28 for ; Fri, 22 Oct 1999 21:18:12 -0700 (PDT) (envelope-from msquires@sir-alan.chem.indiana.edu) Received: (from msquires@localhost) by sir-alan.chem.indiana.edu (8.9.2/8.9.2) id XAA06694 for freebsd-bugs@freebsd.org; Fri, 22 Oct 1999 23:20:05 -0500 (EST) (envelope-from msquires) From: Mike Squires Message-Id: <199910230420.XAA06694@sir-alan.chem.indiana.edu> Subject: samba fails under SMP kernel To: freebsd-bugs@freebsd.org Date: Fri, 22 Oct 1999 23:20:05 -0500 (EST) X-Mailer: ELM [version 2.4ME+ PL43 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-bugs@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org I'm not absolutely positive this is not my mistake, but so far I haven't been able to find another explanation other than a bug in either samba or the SMP kernel on my system. Hardware is Everex PO-6200 dual PPro running 3.3-RELEASE. The config file contains the following, which are the same as the output from mptable: options SMP options APIC_IO options NCPU=2 options NBUS=2 options NAPIC=1 options NINTR=24 The kernel loads and runs fine with everything else (NETATALK, IPX, changer support, ccd/vinum, and bpf are also configured). I have seen a steady 3MB/sec NFS (xperfmon3.0) and 660K/sec on a 1GB file with mars_nwe, no problems. With 1.9, 2.0.2, 2.0.5a with or without MSG_WAITALL, 2.06.pre1, or 2.0.6pre2, and the SMP kernel I can't copy more than 2 - 5MB to the samba server before the connection is dropped by smb. The log files don't show anything useful; they just stop, and eventually the connection is closed. With the single CPU kernel and everything else as is, large files can be copied but they are copied (2.0.5a, 2.0.6pre1/2) fairly slowly, with 2.0.6pre2 being significantly faster - which is the expected behavior. I havne't included other data, since it would be quite large - the full samba logs are 100K at log level 10. I suspect that the input error rate goes up quite a bit when samba is being written to (card is Intel Pro100B; without samba I'm getting zero input errors in 2 million frames, 2 output on 500,000, 100Mbit full duplex) but I haven't been able to prove that, yet. I'm going to run a packet trace next; any other ideas? To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-bugs" in the body of the message