amd64/73360: sk0: watchdog timeout

[Bjoern A. Zeeb]

On Sun, 31 Oct 2004, John-Mark Gurney wrote:

> Wolfgang S. Rupprecht wrote this message on Sun, Oct 31, 2004 at 14:17 -0800:
[patch]
>
> Hey, thanks for pointing this patch out...  I've slightly modified it
> to better handle IFF_OACTIVE since it's possible to clear the flag when
> it shouldn't be...
>
> If people see this patch improve things, I'll commit it... I haven't
> tested w/ my sk card yet though...

it didn't improve the situtation here but made it worse:(

that patch itself is not compileable that way anyway. sc is not
defined in txeof.


> Index: if_sk.c
> ===================================================================
> RCS file: /home/ncvs/src/sys/pci/if_sk.c,v
> retrieving revision 1.86
> diff -u -r1.86 if_sk.c
> --- if_sk.c	20 Aug 2004 06:22:04 -0000	1.86
> +++ if_sk.c	1 Nov 2004 00:54:20 -0000
> @@ -1816,11 +1816,13 @@
>  	}
>
>  	/* Transmit */
> -	sc_if->sk_cdata.sk_tx_prod = idx;
> -	CSR_WRITE_4(sc, sc_if->sk_tx_bmu, SK_TXBMU_TX_START);
> +	if (idx != sc_if->sk_cdata.sk_tx_prod) {
> +		sc_if->sk_cdata.sk_tx_prod = idx;
> +		CSR_WRITE_4(sc, sc_if->sk_tx_bmu, SK_TXBMU_TX_START);
>
> -	/* Set a timeout in case the chip goes out to lunch. */
> -	ifp->if_timer = 5;
> +		/* Set a timeout in case the chip goes out to lunch. */
> +		ifp->if_timer = 5;
> +	}
>  	SK_IF_UNLOCK(sc_if);
>
>  	return;

[pasted in]

@@ -1937,10 +1940,12 @@ static void sk_txeof(sc_if)
        struct sk_if_softc      *sc_if;
 {
+       struct sk_softc         *sc;
        struct sk_tx_desc       *cur_tx = NULL;
        struct ifnet            *ifp;
        u_int32_t               idx;

+       sc = sc_if->sk_softc;
        ifp = &sc_if->arpcom.ac_if;

        /*

> @@ -1960,13 +1962,15 @@
>  		}
>  		sc_if->sk_cdata.sk_tx_cnt--;
>  		SK_INC(idx, SK_TX_RING_CNT);
> -		ifp->if_timer = 0;
>  	}
>
> -	sc_if->sk_cdata.sk_tx_cons = idx;
> -
> -	if (cur_tx != NULL)
> +	if (sc_if->sk_cdata.sk_tx_cnt == 0) {
> +		ifp->if_timer = 0;
>  		ifp->if_flags &= ~IFF_OACTIVE;
> +	} else /* nudge chip to keep tx ring moving */
> +		CSR_WRITE_4(sc, sc_if->sk_tx_bmu, SK_TXBMU_TX_START);
> +
> +	sc_if->sk_cdata.sk_tx_cons = idx;
>
>  	return;
>  }


if I add two printfs to each rxeof,txeof to print i resp. idx this
slows down processing to about 120KB/s and make the thing live for
about 3-6 MB transferred before "hanging". The output of the time the
hanging starts looks like that:

legend:
tx = txeof
rx = rxeof
s = start
e = end
i = i,idx

output:
rx s i=47
rx e i=54
tx s i=104
tx e i=126
rx s i=54
rx e i=60
tx s i=126
tx e i=150
rx s i=60
rx e i=67
tx s i=150
tx e i=161
rx s i=67
rx e i=71
tx s i=161
tx e i=163
tx s i=163
tx e i=190
rx s i=71
rx e i=78
tx s i=190
tx e i=217
rx s i=78
rx e i=82
tx s i=217
tx e i=230
tx s i=230
tx e i=232
tx s i=232
tx e i=235
tx s i=235
tx e i=237
tx s i=237
tx e i=240
tx s i=240
tx e i=242
tx s i=242
tx e i=244
tx s i=244
tx e i=247
tx s i=247
tx e i=249
tx s i=249
tx e i=251
tx s i=251
tx e i=254
tx s i=254
tx e i=256
tx s i=256
tx e i=258
tx s i=258
tx e i=260
tx s i=260
tx e i=263
tx s i=263
tx e i=265
tx s i=265
tx e i=267
tx s i=267
tx e i=270
tx s i=270
tx e i=272
tx s i=272
tx e i=275
tx s i=275
tx e i=277
tx s i=277
tx e i=280


The funny thing is if I go into the "also hanging" xterm where I
logged in on that amd64 with ssh and started the ftp put from
and then hit some keys like enter I will see Rx intr fired again:

rx s i=82
rx e i=83
tx s i=297
tx e i=298
rx s i=83
rx e i=84
tx s i=298
tx e i=301
rx s i=84
rx e i=85
tx s i=301
tx e i=304
rx s i=85
rx e i=86
tx s i=304
tx e i=305
tx s i=305
tx e i=308
tx s i=308
tx e i=311
rx s i=86
rx e i=87
tx s i=311
tx e i=313
rx s i=87
rx e i=88
tx s i=313
tx e i=314


-- 
Bjoern A. Zeeb				bzeeb at Zabbadoz dot NeT