Date: Thu, 30 Oct 2003 08:01:23 -0800 From: Sam Leffler <sam@errno.com> To: Pawel Jakub Dawidek <nick@garage.freebsd.pl> Cc: freebsd-current@freebsd.org Subject: Re: HEADSUP: MPSAFE network drivers Message-ID: <200310300801.23878.sam@errno.com> In-Reply-To: <20031030092238.GY84474@garage.freebsd.pl> References: <200310291052.48617.sam@errno.com> <20031030092238.GY84474@garage.freebsd.pl>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday 30 October 2003 01:22 am, Pawel Jakub Dawidek wrote: > On Wed, Oct 29, 2003 at 11:52:48AM -0700, Sam Leffler wrote: > +> I'm committing changes to mark various network drivers' interrupt > handlers +> MPSAFE. To insure folks have a way to backout if they hit > problems I've also +> added a tunable that lets you disable this w/o > rebuilding your kernel. By +> default all network drivers that register an > interrupt handler INTR_MPSAFE +> are setup to run their ISR w/o Giant. If > you want to defeat this w/o +> changing the code you can set > +> > +> debug.mpsafenet=0 > +> > +> from the loader when booting and the MPSAFE bit will automatically be > removed. +> I plan to use this to also control forthcoming changes for > registering MPSAFE +> netisrs. > +> > +> The following drivers are marked MPSAFE: > +> > +> ath, em, ep, fxp, sn, wi, sis > +> > +> I've got changes coming for bge. Other drivers probably can be marked > MPSAFE +> but I'm only doing it for those drivers that I can test. > > Because there is so many drivers, maybe you could prepare some regression > tests designed to check changed things. This will allow people to test your > changes - it is not very easy now if we don't know what we're looking for > exactly PLUS those drivers aren't marked MPSAFE by default. Unfortunately there is no easy way to decide if the locking in a driver is correct; otherwise I'd simply test them and not provide a fallback as a I did. Before I commit any driver I run with it for at least a few weeks (in some cases months) on a variety of machines (workstation, server, laptop, firewall). If there are no problems then I commit the change. The driver changes I committed yesterday I've been running for >4 months. Likewise, the next round of locking changes to push Giant up have been running for ~2 months. Otherwise the main safeguard I use are numerous assertions to validate assumptions. Sam
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200310300801.23878.sam>