SEO

If you have an sk(4) NIC then please try out this diff..

----- Forwarded message from Brad <brad@comstyle.com> -----

X-Original-To: brad@localhost
Delivered-To: brad@localhost.comstyle.com
X-Original-To: brad@comstyle.com
Delivered-To: brad@comstyle.com
Date: Sun, 10 Jul 2005 21:24:51 -0400
From: Brad <brad@comstyle.com>
To: hackers@openbsd.org
Subject: Re: sk
In-Reply-To: <Pine.BSO.4.61.0504140232310.133@af.pbqrshfvbavf.p bz>
User-Agent: Mutt/1.5.6i
X-Loop: hackers@openbsd.org
Precedence: list

On Thu, Apr 14, 2005 at 02:36:33AM -0400, Ted Unangst wrote:
> run ifconfig sk0 down, ifconfig sk0 up about 20 times. does M_DEVBUF get
> out of hand?
>
> sk_init_tx_ring definitely has problems. i can't find where the txmap
> list is ever freed.

Ya, this is pretty bad..

devbuf 1053 249K 252K 39322K 1899 0 0 16,32,64,128,256,512,1024,2048,4096,8192,16384,327 68

around 10 down/up's

devbuf 11293 2889K 2889K 39322K 12354 0 0 16,32,64,128,256,512,1024,2048,4096,8192,16384,327 68

This seems to do the trick.

Index: if_sk.c
================================================== =================
RCS file: /cvs/src/sys/dev/pci/if_sk.c,v
retrieving revision 1.69
diff -u -p -r1.69 if_sk.c
--- if_sk.c 2 Jul 2005 23:10:11 -0000 1.69
+++ if_sk.c 11 Jul 2005 01:16:23 -0000
@@ -2703,6 +2703,7 @@ sk_stop(struct sk_if_softc *sc_if)
{
struct sk_softc *sc = sc_if->sk_softc;
struct ifnet *ifp = &sc_if->arpcom.ac_if;
+ struct sk_txmap_entry *dma;
int i;

DPRINTFN(2, ("sk_stop\n"));
@@ -2773,7 +2774,16 @@ sk_stop(struct sk_if_softc *sc_if)
if (sc_if->sk_cdata.sk_tx_chain[i].sk_mbuf != NULL) {
m_freem(sc_if->sk_cdata.sk_tx_chain[i].sk_mbuf);
sc_if->sk_cdata.sk_tx_chain[i].sk_mbuf = NULL;
+ SIMPLEQ_INSERT_HEAD(&sc_if->sk_txmap_head,
+ sc_if->sk_cdata.sk_tx_map[i], link);
+ sc_if->sk_cdata.sk_tx_map[i] = 0;
}
+ }
+
+ while ((dma = SIMPLEQ_FIRST(&sc_if->sk_txmap_head))) {
+ SIMPLEQ_REMOVE_HEAD(&sc_if->sk_txmap_head, link);
+ bus_dmamap_destroy(sc->sc_dmatag, dma->dmamap);
+ free(dma, M_DEVBUF);
}
}


----- End forwarded message -----