Date: Mon, 3 Jun 2013 20:59:45 +0000 From: Sebastian Kuzminsky <S.Kuzminsky@F5.com> To: "freebsd-stable@freebsd.org" <freebsd-stable@freebsd.org> Subject: bce kernel page faults and NMIs (was: Strange reboot since 9.1) Message-ID: <73664E92-986C-4582-9A43-EFA576931318@f5.com>
next in thread | raw e-mail | index | archive | help
Howdy folks, this email is a follow-on to a 3-month-old thread about kernel= page faults from the bce driver[0]. 0: http://lists.freebsd.org/pipermail/freebsd-stable/2013-March/072713.htm= l Sorry to revive such an old thread, but a couple of bits of new information= has come to light here that may be useful for others. The header splitting suggestion that Marius Strobl made[1] did fix the ker= nel page fault rooted in bce_intr() that we were seeing (and that other fol= ks reported in the original thread). I'm no bce expert, but it looks to me= like the bce driver does not apply the same flow control to its page queue= as it does to its receive queue, maybe that's related to the problem? 1: http://lists.freebsd.org/pipermail/freebsd-stable/2013-March/072766.htm= l After disabling bce header splitting we stopped getting kernel page faults,= but we still had problems with this NIC (Broadcom NetXtreme II BCM5716 Gig= abit Ethernet) producing frequent PCI errors and occasional NMIs. I found this thread[2] that suggests that the NIC firmware version may be r= elevant to the NMI problem. The Red Hat people are reporting that firmware= version 6.0.1 is bad and 6.4.5 is good; 9.1 ships with 6.0.17, so who know= s what that means... We ended up reverting to the bce driver from FreeBSD = 7 and that fixed our NMI problems. (The bce driver from FreeBSD 7 also has= header splitting disabled by default: Bonus!) 2: https://bugzilla.redhat.com/show_bug.cgi?id=3D693542 --=20 Sebastian Kuzminsky=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?73664E92-986C-4582-9A43-EFA576931318>