Date: Fri, 17 Sep 2004 20:38:51 +0200 From: "Raphael H. Becker" <rabe@p-i-n.com> To: freebsd-current@freebsd.org Subject: Re: de0 doze off Message-ID: <20040917203851.F55054@p-i-n.com> In-Reply-To: <Pine.NEB.3.96L.1040917093737.33867D-100000@fledge.watson.org>; from rwatson@freebsd.org on Fri, Sep 17, 2004 at 09:38:46AM -0400 References: <20040917065926.D55054@p-i-n.com> <Pine.NEB.3.96L.1040917093737.33867D-100000@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Sep 17, 2004 at 09:38:46AM -0400, Robert Watson wrote:
> On Fri, 17 Sep 2004, Raphael H. Becker wrote:
> > on one of my older boxes my D-Link ("tulip") card dozes off when
> > transfering some megabytes of data eg scp.
> >
> > When stalled, the interface can be brought back by "ifconfig down/up".
> >
> > The box is actually running BETA4/GENERIC from last night, this appeared
> > after BETA1 which was the last stable for this.
>
> If you set debug.mpsafenet=0 in loader.conf, does the problem go away?
Yes, just tested with around 300MB via FTP: 10.6MBytes/sec
> We might be looking at a race condition in the task queue handoff for
> IFF_NEEDSGIANT.
> Another possible testing avenue (which might be slightly unstable but
> would illustrate the point) would be to leave debug.mpsafenet=1,
ok
> disable preemption,
Sorry, how exactly do I do this?
> and remove the IFF_NEEDSGIANT flag from the if_de driver flags.
edited as following
--- if_de.c.orig Fri Sep 17 18:53:41 2004
+++ if_de.c Fri Sep 17 18:54:00 2004
@@ -4758,7 +4758,7 @@
/* XXX: driver name/unit should be set some other way */
ifp->if_dname = "de";
ifp->if_dunit = sc->tulip_unit;
- ifp->if_flags = IFF_BROADCAST|IFF_SIMPLEX|IFF_MULTICAST|IFF_NEEDSGIANT;
+ ifp->if_flags = IFF_BROADCAST|IFF_SIMPLEX|IFF_MULTICAST;
ifp->if_ioctl = tulip_ifioctl;
ifp->if_start = tulip_ifstart;
ifp->if_watchdog = tulip_ifwatchdog;
Regards
Raphael Becker
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040917203851.F55054>