Re: ndis(4) patch to replace obsolete if_watchdog interface

On Thu, May 29, 2008 at 07:45:31PM -0400, Coleman Kane wrote:
> On Thu, 2008-05-29 at 14:46 -0700, Andrew Thompson wrote:
> > On Thu, May 29, 2008 at 04:41:32PM -0400, Coleman Kane wrote:
> > > Hi,
> > > 
> > > I just replaced the obsoleted if_watchdog interface in ndis(4) with a
> > > local implementation. This should remove the obnoxious warning message
> > > on device init. Anyone using -CURRENT with an ndis card, could you send
> > > me success/fails?
> > > 
> > > The patch is here:
> > >   * http://people.freebsd.org/~cokane/patches/if_ndis-new_wd.patch
> > 
> > 
> > This works different to the rest of the network drivers. The existing
> > drivers use a callout tick that runs while the driver is up and an
> > integer counter.
> > 
> >  if (x && --x == 0)
> > 	 ...timeout...
> > 
> > You arm the callout and stop it after each Tx, does this have any
> > perfornace impact?
> > 
> > 
> > Andrew
> > 
> 
> I am attaching a new patch, where I use the method that you suggested
> and do away with the extra callout for the watchdog. A "tick" function
> was already implemented for timeouts related to the NDIS systems.
> 
> I am not sure if the ndis-timeout code using ndis_stat_callout
> duplicates the if_watchdog code (and what I replaced it with), but I
> don't think so as both were implemented side-by-side before...
> 

A few comments inline

>diff --git a/sys/dev/if_ndis/if_ndis.c b/sys/dev/if_ndis/if_ndis.c
>+	NDIS_LOCK(sc);
>+	if(sc->ndis_timer_countdown &&
>+	    (sc->ndis_timer_countdown -=
>+	    sc->ndis_block->nmb_checkforhangsecs) < 1) {
>+		sc->ndis_timer_countdown = 0;
>+		NDIS_UNLOCK(sc);
>+		ndis_watchdog(xsc);
>+	} else {
>+		NDIS_UNLOCK(sc);
>+	}

I think this is harder than it needs to be. It locks, unlocks, and then
locks again in ndis_watchdog. I have attached a version of this diff
that nukes ndis_watchdog() and just merges the code. Its also a pain to
use nmb_checkforhangsecs for the timer period, the ndis driver blob can
set this to whatever it desires and it could be far greater than the
watchdog period of 5 secs.

I introduced two counters, ndis_tx_timer and ndis_hang_timer, that allow
these events to be independent of each other.

> 	return;
> }
> 
>_at__at_ -1888,7 +1904,9 _at__at_ ndis_start(ifp)
> 	/*
> 	 * Set a timeout in case the chip goes out to lunch.
> 	 */
>-	ifp->if_timer = 5;
>+	if(sc->ndis_timer_countdown == 0) {
>+		sc->ndis_timer_countdown = 5;
>+	}

This isnt right. The timer doesnt need to expire in order to be reset,
it should just set it each time. Otherwise if the timer is at 1 and the
callout is _just_ about to fire then you will get a false watchdog
timeout.



cheers,
Andrew