kill checksum offloading in sk.

it was RX only, nothing on the TX side
rather complex, might eat up (or more) offloading benefits in many cases
the hardware miscomputes the cksums sometimes, so we don't trust it
claiming a cksum is bad and re-do the cksum verification in sw then...
(but we trust it to not mark bad ones good? hmmmmmmmmm.)
diff was on tech for 3 months, nobody disagreed, everybody who spoke up
agrees or doesn't care.
pretty exactly a revert of the cksum offloading addition years ago, tested
by several, the only report i quickly find now is from David Higgs <higgsd
at gmail dot com>, thanks!

1 files changed, 1 insertions, 102 deletions

diff --git a/sys/dev/pci/if_sk.c b/sys/dev/pci/if_sk.c
index 1c0d30d0dd0..9606e386bd8 100644
--- a/sys/dev/pci/if_sk.c
+++ b/sys/dev/pci/if_sk.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_sk.c,v 1.167 2013/12/28 03:36:25 deraadt Exp $	*/
+/*	$OpenBSD: if_sk.c,v 1.168 2014/04/19 18:29:39 henning Exp $	*/
 
 /*
  * Copyright (c) 1997, 1998, 1999, 2000
@@ -180,7 +180,6 @@ void sk_iff_yukon(struct sk_if_softc *);
 
 void sk_tick(void *);
 void sk_yukon_tick(void *);
-void sk_rxcsum(struct ifnet *, struct mbuf *, const u_int16_t, const u_int16_t);
 
 #ifdef SK_DEBUG
 #define DPRINTF(x)	if (skdebug) printf x
@@ -550,9 +549,6 @@ sk_init_rx_ring(struct sk_if_softc *sc_if)
 			nexti = i + 1;
 		cd->sk_rx_chain[i].sk_next = &cd->sk_rx_chain[nexti];
 		rd->sk_rx_ring[i].sk_next = htole32(SK_RX_RING_ADDR(sc_if, nexti));
-		rd->sk_rx_ring[i].sk_csum1_start = htole16(ETHER_HDR_LEN);
-		rd->sk_rx_ring[i].sk_csum2_start = htole16(ETHER_HDR_LEN +
-		    sizeof(struct ip));
 	}
 
 	for (i = 0; i < SK_RX_RING_CNT; i++) {
@@ -1732,7 +1728,6 @@ sk_rxeof(struct sk_if_softc *sc_if)
 	int			i, cur, total_len = 0;
 	u_int32_t		rxstat, sk_ctl;
 	bus_dmamap_t		dmamap;
-	u_int16_t		csum1, csum2;
 
 	DPRINTFN(2, ("sk_rxeof\n"));
 
@@ -1765,9 +1760,6 @@ sk_rxeof(struct sk_if_softc *sc_if)
 		cur_rx->sk_mbuf = NULL;
 		total_len = SK_RXBYTES(letoh32(cur_desc->sk_ctl));
 
-		csum1 = letoh16(sc_if->sk_rdata->sk_rx_ring[i].sk_csum1);
-		csum2 = letoh16(sc_if->sk_rdata->sk_rx_ring[i].sk_csum2);
-
 		SK_INC(i, SK_RX_RING_CNT);
 
 		if ((sk_ctl & (SK_RXCTL_STATUS_VALID | SK_RXCTL_FIRSTFRAG |
@@ -1805,8 +1797,6 @@ sk_rxeof(struct sk_if_softc *sc_if)
 
 		ifp->if_ipackets++;
 
-		sk_rxcsum(ifp, m, csum1, csum2);
-
 #if NBPFILTER > 0
 		if (ifp->if_bpf)
 			bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_IN);
@@ -1818,94 +1808,6 @@ sk_rxeof(struct sk_if_softc *sc_if)
 }
 
 void
-sk_rxcsum(struct ifnet *ifp, struct mbuf *m, const u_int16_t csum1, const u_int16_t csum2)
-{
-	struct ether_header *eh;
-	struct ip *ip;
-	u_int8_t *pp;
-	int hlen, len, plen;
-	u_int16_t iph_csum, ipo_csum, ipd_csum, csum;
-
-	pp = mtod(m, u_int8_t *);
-	plen = m->m_pkthdr.len;
-	if (plen < sizeof(*eh))
-		return;
-	eh = (struct ether_header *)pp;
-	iph_csum = in_cksum_addword(csum1, (~csum2 & 0xffff));
-
-	if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
-		u_int16_t *xp = (u_int16_t *)pp;
-
-		xp = (u_int16_t *)pp;
-		if (xp[1] != htons(ETHERTYPE_IP))
-			return;
-		iph_csum = in_cksum_addword(iph_csum, (~xp[0] & 0xffff));
-		iph_csum = in_cksum_addword(iph_csum, (~xp[1] & 0xffff));
-		xp = (u_int16_t *)(pp + sizeof(struct ip));
-		iph_csum = in_cksum_addword(iph_csum, xp[0]);
-		iph_csum = in_cksum_addword(iph_csum, xp[1]);
-		pp += EVL_ENCAPLEN;
-	} else if (eh->ether_type != htons(ETHERTYPE_IP))
-		return;
-
-	pp += sizeof(*eh);
-	plen -= sizeof(*eh);
-
-	ip = (struct ip *)pp;
-
-	if (ip->ip_v != IPVERSION)
-		return;
-
-	hlen = ip->ip_hl << 2;
-	if (hlen < sizeof(struct ip))
-		return;
-	if (hlen > ntohs(ip->ip_len))
-		return;
-
-	/* Don't deal with truncated or padded packets. */
-	if (plen != ntohs(ip->ip_len))
-		return;
-
-	len = hlen - sizeof(struct ip);
-	if (len > 0) {
-		u_int16_t *p;
-
-		p = (u_int16_t *)(ip + 1);
-		ipo_csum = 0;
-		for (ipo_csum = 0; len > 0; len -= sizeof(*p), p++)
-			ipo_csum = in_cksum_addword(ipo_csum, *p);
-		iph_csum = in_cksum_addword(iph_csum, ipo_csum);
-		ipd_csum = in_cksum_addword(csum2, (~ipo_csum & 0xffff));
-	} else
-		ipd_csum = csum2;
-
-	if (iph_csum != 0xffff)
-		return;
-	m->m_pkthdr.csum_flags |= M_IPV4_CSUM_IN_OK;
-
-	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
-		return;                 /* ip frag, we're done for now */
-
-	pp += hlen;
-
-	/* Only know checksum protocol for udp/tcp */
-	if (ip->ip_p == IPPROTO_UDP) {
-		struct udphdr *uh = (struct udphdr *)pp;
-
-		if (uh->uh_sum == 0)    /* udp with no checksum */
-			return;
-	} else if (ip->ip_p != IPPROTO_TCP)
-		return;
-
-	csum = in_cksum_phdr(ip->ip_src.s_addr, ip->ip_dst.s_addr,
-	    htonl(ntohs(ip->ip_len) - hlen + ip->ip_p) + ipd_csum);
-	if (csum == 0xffff) {
-		m->m_pkthdr.csum_flags |= (ip->ip_p == IPPROTO_TCP) ?
-		    M_TCP_CSUM_IN_OK : M_UDP_CSUM_IN_OK;
-	}
-}
-
-void
 sk_txeof(struct sk_if_softc *sc_if)
 {
 	struct sk_softc		*sc = sc_if->sk_softc;
@@ -2817,9 +2719,6 @@ sk_dump_txdesc(struct sk_tx_desc *desc, int idx)
 	DESC_PRINT(letoh32(desc->sk_data_hi));
 	DESC_PRINT(letoh32(desc->sk_xmac_txstat));
 	DESC_PRINT(letoh16(desc->sk_rsvd0));
-	DESC_PRINT(letoh16(desc->sk_csum_startval));
-	DESC_PRINT(letoh16(desc->sk_csum_startpos));
-	DESC_PRINT(letoh16(desc->sk_csum_writepos));
 	DESC_PRINT(letoh16(desc->sk_rsvd1));
 #undef PRINT
 }