add hme(4) RX TCP/UDP checksum offload

From a diff posted to sparc@ by Pyun YongHyeon

ok dlg@, tested by mickey@ "i get some badass improvement here." and
a few other users.

2 files changed, 114 insertions, 4 deletions

diff --git a/sys/dev/ic/hme.c b/sys/dev/ic/hme.c
index 2f5e41f1602..5c082d70c02 100644
--- a/sys/dev/ic/hme.c
+++ b/sys/dev/ic/hme.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: hme.c,v 1.34 2005/01/15 05:24:11 brad Exp $	*/
+/*	$OpenBSD: hme.c,v 1.35 2005/02/04 05:02:38 brad Exp $	*/
 /*	$NetBSD: hme.c,v 1.21 2001/07/07 15:59:37 thorpej Exp $	*/
 
 /*-
@@ -67,6 +67,8 @@
 #include <netinet/in_var.h>
 #include <netinet/ip.h>
 #include <netinet/if_ether.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
 #endif
 
 #if NBPFILTER > 0
@@ -112,6 +114,8 @@ void		hme_mediastatus(struct ifnet *, struct ifmediareq *);
 int		hme_eint(struct hme_softc *, u_int);
 int		hme_rint(struct hme_softc *);
 int		hme_tint(struct hme_softc *);
+/* TCP/UDP checksum offloading support */
+void 		hme_rxcksum(struct mbuf *, u_int32_t);
 
 void
 hme_config(sc)
@@ -481,7 +485,7 @@ hme_init(sc)
 	bus_space_handle_t erx = sc->sc_erx;
 	bus_space_handle_t mac = sc->sc_mac;
 	u_int8_t *ea;
-	u_int32_t v;
+	u_int32_t v, n;
 
 	/*
 	 * Initialization sequence. The numbered steps below correspond
@@ -590,6 +594,10 @@ hme_init(sc)
 #endif
 	/* Enable DMA */
 	v |= HME_ERX_CFG_DMAENABLE | (HME_RX_OFFSET << 3);
+	/* RX TCP/UDP cksum offset */
+	n = (ETHER_HDR_LEN + sizeof(struct ip)) / 2;
+	n = (n << HME_ERX_CFG_CSUM_SHIFT) & HME_ERX_CFG_CSUMSTART;
+	v |= n;
 	bus_space_write_4(t, erx, HME_ERXI_CFG, v);
 
 	/* step 11. XIF Configuration */
@@ -727,6 +735,105 @@ hme_tint(sc)
 }
 
 /*
+ * XXX layering violation
+ *
+ * If we can have additional csum data member in 'struct pkthdr' for
+ * these incomplete checksum offload capable hardware, things would be
+ * much simpler. That member variable will carry partial checksum
+ * data and it may be evaluated in TCP/UDP input handler after
+ * computing pseudo header checksumming.
+ */
+void
+hme_rxcksum(struct mbuf *m, u_int32_t flags)
+{
+	struct ether_header *eh;
+	struct ip *ip;
+	struct udphdr *uh;
+	int32_t hlen, len, pktlen;
+	u_int16_t cksum, flag_bad, flag_ok, *opts;
+	u_int32_t temp32;
+	union pseudoh {
+		struct hdr {
+			u_int16_t len;
+			u_int8_t ttl;
+			u_int8_t proto;
+			u_int32_t src;
+			u_int32_t dst;
+		} h;
+		u_int16_t w[6];
+	} ph;
+
+	pktlen = m->m_pkthdr.len;
+	if (pktlen < sizeof(struct ether_header))
+		return;
+	eh = mtod(m, struct ether_header *);
+	if (eh->ether_type != htons(ETHERTYPE_IP))
+		return;
+	ip = (struct ip *)(eh + 1);
+	if (ip->ip_v != IPVERSION)
+		return;
+
+	hlen = ip->ip_hl << 2;
+	pktlen -= sizeof(struct ether_header);
+	if (hlen < sizeof(struct ip))
+		return;
+	if (ntohs(ip->ip_len) < hlen)
+		return;
+	if (ntohs(ip->ip_len) != pktlen) 
+		return;
+	if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
+		return;	/* can't handle fragmented packet */
+
+	switch (ip->ip_p) {
+	case IPPROTO_TCP:
+		if (pktlen < (hlen + sizeof(struct tcphdr)))
+			return;
+		flag_ok = M_TCP_CSUM_IN_OK;
+		flag_bad = M_TCP_CSUM_IN_BAD;
+		break;
+	case IPPROTO_UDP:
+		if (pktlen < (hlen + sizeof(struct udphdr)))
+			return;
+		uh = (struct udphdr *)((caddr_t)ip + hlen);
+		if (uh->uh_sum == 0)
+			return; /* no checksum */
+		flag_ok = M_UDP_CSUM_IN_OK;
+		flag_bad = M_UDP_CSUM_IN_BAD;
+		break;
+	default:
+		return;
+	}
+
+	cksum = ~(flags & HME_XD_RXCKSUM);
+	/* cksum fixup for IP options */
+	len = hlen - sizeof(struct ip);
+	if (len > 0) {
+		opts = (u_int16_t *)(ip + 1);
+		for (; len > 0; len -= sizeof(u_int16_t), opts++) {
+			temp32 = cksum - *opts;
+			temp32 = (temp32 >> 16) + (temp32 & 65535);
+			cksum = temp32 & 65535;
+		}
+	}
+	/* cksum fixup for pseudo-header, replace with in_cksum_phdr()? */
+	ph.h.len = htons(ntohs(ip->ip_len) - hlen);
+	ph.h.ttl = 0;
+	ph.h.proto = ip->ip_p;
+	ph.h.src = ip->ip_src.s_addr;
+	ph.h.dst = ip->ip_dst.s_addr;
+	temp32 = cksum;
+	opts = &ph.w[0];
+	temp32 += opts[0] + opts[1] + opts[2] + opts[3] + opts[4] + opts[5];
+	temp32 = (temp32 >> 16) + (temp32 & 65535);
+	temp32 += (temp32 >> 16);
+	cksum = ~temp32;
+	if (cksum != 0)
+		m->m_pkthdr.csum |= flag_bad;
+	else
+		m->m_pkthdr.csum |= flag_ok;
+}
+
+/*
  * Receive interrupt.
  */
 int
@@ -769,6 +876,9 @@ hme_rint(sc)
 			goto again;
 		}
 
+		ifp->if_ipackets++;
+		hme_rxcksum(m, flags);
+
 #if NBPFILTER > 0
 		if (ifp->if_bpf) {
 			m->m_pkthdr.len = m->m_len = len;
@@ -776,7 +886,6 @@ hme_rint(sc)
 		}
 #endif
 
-		ifp->if_ipackets++;
 		ether_input_mbuf(ifp, m);
 
 again:
diff --git a/sys/dev/ic/hmereg.h b/sys/dev/ic/hmereg.h
index c37e138adf2..22f270d6bb9 100644
--- a/sys/dev/ic/hmereg.h
+++ b/sys/dev/ic/hmereg.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: hmereg.h,v 1.7 2004/09/28 00:21:23 brad Exp $	*/
+/*	$OpenBSD: hmereg.h,v 1.8 2005/02/04 05:02:38 brad Exp $	*/
 /*	$NetBSD: hmereg.h,v 1.8 2001/04/30 12:22:42 bouyer Exp $	*/
 
 /*-
@@ -164,6 +164,7 @@
 #define HME_ERX_CFG_RINGSIZE128	0x00000400	/* Descriptor ring size: 128 */
 #define HME_ERX_CFG_RINGSIZE256	0x00000600	/* Descriptor ring size: 256 */
 #define HME_ERX_CFG_CSUMSTART	0x007f0000	/* cksum offset */
+#define HME_ERX_CFG_CSUM_SHIFT	16
 
 /*
  * HME MAC-core register offsets