Date: Fri, 14 Dec 2001 11:13:17 +0900 From: Dan Kogai <dankogai@dan.co.jp> To: <stable@FreeBSD.ORG> Subject: rl driver panics w/ large file transfer (Was: Re: 4.4-STABLE crashes...) Message-ID: <B83F8F4D.5190%dankogai@dan.co.jp> In-Reply-To: <20011212214455.C10147@ldc.ro>
next in thread | previous in thread | raw e-mail | index | archive | help
on 01.12.13 4:44 AM, Alex Popa at razor@ldc.ro wrote: > During the bonnie++ runs iostat showed some 22M/s activity on each disk, > so I am beginning to suspect there has to be a correlation between heavy > disk I/O and heavy network I/O (my initial crashes occured when someone > was making a large backup over SMB to the server, at about 9M/s disk > activity) in order to get the crashes. I had a similar simptom when I tried to backup a huge chunk of files (about 30GB) over NFS. The box panic seems to panic at random but it always crashed at the same point. > Dec 13 19:28:57 gm0 /kernel: Fatal trap 12: page fault while in kernel mode > Dec 13 19:28:57 gm0 /kernel: fault virtual address = 0x3a > Dec 13 19:28:57 gm0 /kernel: fault code = supervisor write, page not > present > Dec 13 19:28:57 gm0 /kernel: instruction pointer = 0x8:0xc02d1c56 > Dec 13 19:28:57 gm0 /kernel: stack pointer = 0x10:0xc03a5974 > Dec 13 19:28:57 gm0 /kernel: frame pointer = 0x10:0xc03a5a14 > Dec 13 19:28:57 gm0 /kernel: code segment = base 0x0, limit > 0xfffff, type 0x1b > Dec 13 19:28:57 gm0 /kernel: = DPL 0, pres 1, def32 1, gran 1 > Dec 13 19:28:57 gm0 /kernel: processor eflags = interrupt enabled, resume, > IOPL = 0 > Dec 13 19:28:57 gm0 /kernel: current process = Idle > Dec 13 19:28:57 gm0 /kernel: interrupt mask = net tty > Dec 13 19:28:57 gm0 /kernel: trap number = 12 > Dec 13 19:28:57 gm0 /kernel: panic: page fault I checked the disk, I check the source. I tried both 4.4-RELEASE kernel and 4.4-STABLE. Same result. In a desperate search I changed the NIC from on-baord rl <RealTek 8139 10/100BaseTX> to vr card <VIA VT6102 Rhine II 10/100BaseTX> and voila! it's gone! It seems that rl crashes when the throughput passes 20MBps or 20% of theoretical max. Well, the source says rl sucks as follows. /* * The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is * probably the worst PCI ethernet controller ever made, with the possible * exception of the FEAST chip made by SMC. The 8139 supports bus-master * DMA, but it has a terrible interface that nullifies any performance * gains that bus-master DMA usually offers. * [snip] */ So I said no wonder but I still hate to see the panic in the middle of large file transfer and long fsck to come. You should rather choke the throughput instead.... Dan the Man with too Many NICs to deal with -- _____ Dan Kogai __/ ____ CEO, DAN co. ltd. /__ /-+-/ 2-8-14-418 Shiomi Koto-ku Tokyo 135-0052 Japan /--/--- mailto: dankogai@dan.co.jp / http://www.dan.co.jp/ --------- __/ / Tel:+81 3-5665-6131 Fax:+81 3-5665-6132 PGP Key: http://www.dan.co.jp/‾dankogai/dankogai.pgp.asc To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?B83F8F4D.5190%dankogai>