Date: Mon, 7 Apr 2003 08:25:27 +1000 (EST) From: Bruce Evans <bde@zeta.org.au> To: Dag-Erling =?iso-8859-1?q?Sm=F8rgrav?= <des@ofug.org> Cc: current@freebsd.org Subject: Re: weird fxp / timecounter interaction in top-of-tree Message-ID: <20030407081351.V2224@gamplex.bde.org> In-Reply-To: <xzpel4fmqu9.fsf@flood.ping.uio.no> References: <xzpu1dbeflv.fsf@flood.ping.uio.no> <xzpel4fmqu9.fsf@flood.ping.uio.no>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 6 Apr 2003, Dag-Erling [iso-8859-1] Smørgrav wrote:
> des@ofug.org (Dag-Erling Smørgrav) writes:
> > - loading the if_fxp module (and miibus as a dependency) switches the
> > timecounter hardware to PIIX.
>
> This turns out to be caused by a combination of bugs in tc_init() and
> the piix driver. The piix driver initializes the PIIX timecounter in
> piix_attach(), which is called whenever the PCI bus is rescanned (e.g.
> when a driver for a PCI device such as fxp is loaded) instead of
> piix_probe() which is only called once. The other problem is that
> tc_init() will actually install the timecounter being initialized, so
> in effect, unless you explicitly select one after boot, your machine
> will use whichever timecounter was probed and attached last.
>
> I've modified the piix driver to only initialize the timecounter once,
> and tc_init() to use the *first* timecounter it runs across (on i386,
> this is generally the i8254), leaving the admin to pick another one if
> the default does not suit her. See the attached patch.
Do any other attach routines get called on rescan? I would expect most of
them to do the wrong thing.
> Index: sys/kern/kern_tc.c
> ===================================================================
> RCS file: /home/ncvs/src/sys/kern/kern_tc.c,v
> retrieving revision 1.148
> diff -u -r1.148 kern_tc.c
> --- sys/kern/kern_tc.c 18 Mar 2003 08:45:23 -0000 1.148
> +++ sys/kern/kern_tc.c 6 Apr 2003 18:06:38 -0000
> @@ -295,7 +295,8 @@
> printf("\n");
> (void)tc->tc_get_timecount(tc);
> (void)tc->tc_get_timecount(tc);
> - timecounter = tc;
> + if (timecounter == &dummy_timecounter)
> + timecounter = tc;
> }
This is not quite right (see other replies), but tc_init() should be split
up so that it never decides whether to start using the timecounter. Note
that the tc_get_timecounter() calls are part of using it.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030407081351.V2224>