Revisiting this following some recent discussions on another
mailing list,
On 2010/09/23 15:13, Marco Pfatschbacher wrote:
> On Wed, Sep 22, 2010 at 06:11:58PM +0200, Mike Belopuhov wrote:
> > there's nothing special about this driver. as it does usual iic_exec's,
> > i'm not sure what can be actually fixed there. so if nobody has any
> > idea what could be done about it, you have my okay to disable it
> > completely.
>
> It's not the drivers fault, it seems that polling on piixpm(4)
> causes hangs on ServerWorks HT-1000 chips.
>
> This is what we are running:
>
> Index: piixpm.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/pci/piixpm.c,v
> retrieving revision 1.34
> diff -p -u -r1.34 piixpm.c
> --- piixpm.c 8 Apr 2010 00:23:53 -0000 1.34
> +++ piixpm.c 23 Sep 2010 13:08:36 -0000
> @@ -159,6 +159,10 @@ piixpm_attach(struct device *parent, str
> if (sc->sc_poll)
> printf(": polling");
> }
> + if (sc->sc_poll) {
> + printf(" disabled\n");
> + return;
> + }
>
> printf("\n");
>
This seems a bigger hammer than necessary unless there are
similar problems with all piixpm polling devices.
Anyone have a suggestion of how to move forward on this?
For some people the momentary hangs (which can trigger netlivelock
mitigation) are a big problem, for others losing sensors would
be worse...
