5 files changed, 256 insertions, 187 deletions

diff --git a/usr/src/uts/common/io/dld/dld_proto.c b/usr/src/uts/common/io/dld/dld_proto.c
index 1371fa47c0..596147f4e9 100644
--- a/usr/src/uts/common/io/dld/dld_proto.c
+++ b/usr/src/uts/common/io/dld/dld_proto.c
@@ -1537,13 +1537,23 @@ dld_capab_lso(dld_str_t *dsp, void *data, uint_t flags)
 		 * accordingly.
 		 */
 		if (mac_capab_get(dsp->ds_mh, MAC_CAPAB_LSO, &mac_lso)) {
-			lso->lso_max = mac_lso.lso_basic_tcp_ipv4.lso_max;
+			lso->lso_max_tcpv4 = mac_lso.lso_basic_tcp_ipv4.lso_max;
+			lso->lso_max_tcpv6 = mac_lso.lso_basic_tcp_ipv6.lso_max;
 			lso->lso_flags = 0;
 			/* translate the flag for mac clients */
 			if ((mac_lso.lso_flags & LSO_TX_BASIC_TCP_IPV4) != 0)
 				lso->lso_flags |= DLD_LSO_BASIC_TCP_IPV4;
-			dsp->ds_lso = B_TRUE;
-			dsp->ds_lso_max = lso->lso_max;
+			if ((mac_lso.lso_flags & LSO_TX_BASIC_TCP_IPV6) != 0)
+				lso->lso_flags |= DLD_LSO_BASIC_TCP_IPV6;
+			dsp->ds_lso = lso->lso_flags != 0;
+			/*
+			 * DLS uses this to try and make sure that a raw ioctl
+			 * doesn't send too much data, but doesn't currently
+			 * check the actual SAP that is sending this (or that
+			 * it's TCP). So for now, just use the max value here.
+			 */
+			dsp->ds_lso_max = MAX(lso->lso_max_tcpv4,
+			    lso->lso_max_tcpv6);
 		} else {
 			dsp->ds_lso = B_FALSE;
 			dsp->ds_lso_max = 0;
@@ -1581,10 +1591,18 @@ dld_capab(dld_str_t *dsp, uint_t type, void *data, uint_t flags)
 
 	switch (type) {
 	case DLD_CAPAB_DIRECT:
+		if (dsp->ds_sap == ETHERTYPE_IPV6) {
+			err = ENOTSUP;
+			break;
+		}
 		err = dld_capab_direct(dsp, data, flags);
 		break;
 
 	case DLD_CAPAB_POLL:
+		if (dsp->ds_sap == ETHERTYPE_IPV6) {
+			err = ENOTSUP;
+			break;
+		}
 		err =  dld_capab_poll(dsp, data, flags);
 		break;
 
diff --git a/usr/src/uts/common/io/i40e/i40e_transceiver.c b/usr/src/uts/common/io/i40e/i40e_transceiver.c
index caafa3e102..9662cb58f5 100644
--- a/usr/src/uts/common/io/i40e/i40e_transceiver.c
+++ b/usr/src/uts/common/io/i40e/i40e_transceiver.c
@@ -1663,186 +1663,6 @@ i40e_ring_rx_poll(void *arg, int poll_bytes)
 }
 
 /*
- * This is a structure I wish someone would fill out for me for dorking with the
- * checksums. When we get some more experience with this, we should go ahead and
- * consider adding this to MAC.
- */
-typedef enum mac_ether_offload_flags {
-	MEOI_L2INFO_SET		= 0x01,
-	MEOI_VLAN_TAGGED	= 0x02,
-	MEOI_L3INFO_SET		= 0x04,
-	MEOI_L3CKSUM_SET	= 0x08,
-	MEOI_L4INFO_SET		= 0x10,
-	MEOI_L4CKSUM_SET	= 0x20
-} mac_ether_offload_flags_t;
-
-typedef struct mac_ether_offload_info {
-	mac_ether_offload_flags_t	meoi_flags;
-	uint8_t		meoi_l2hlen;	/* How long is the Ethernet header? */
-	uint16_t	meoi_l3proto;	/* What's the Ethertype */
-	uint8_t		meoi_l3hlen;	/* How long is the header? */
-	uint8_t		meoi_l4proto;	/* What is the payload type? */
-	uint8_t		meoi_l4hlen;	/* How long is the L4 header */
-	mblk_t		*meoi_l3ckmp;	/* Which mblk has the l3 checksum */
-	off_t		meoi_l3ckoff;	/* What's the offset to it */
-	mblk_t		*meoi_l4ckmp;	/* Which mblk has the L4 checksum */
-	off_t		meoi_l4off;	/* What is the offset to it? */
-} mac_ether_offload_info_t;
-
-/*
- * This is something that we'd like to make a general MAC function. Before we do
- * that, we should add support for TSO.
- *
- * We should really keep track of our offset and not walk everything every
- * time. I can't imagine that this will be kind to us at high packet rates;
- * however, for the moment, let's leave that.
- *
- * This walks a message block chain without pulling up to fill in the context
- * information. Note that the data we care about could be hidden across more
- * than one mblk_t.
- */
-static int
-i40e_meoi_get_uint8(mblk_t *mp, off_t off, uint8_t *out)
-{
-	size_t mpsize;
-	uint8_t *bp;
-
-	mpsize = msgsize(mp);
-	/* Check for overflow */
-	if (off + sizeof (uint16_t) > mpsize)
-		return (-1);
-
-	mpsize = MBLKL(mp);
-	while (off >= mpsize) {
-		mp = mp->b_cont;
-		off -= mpsize;
-		mpsize = MBLKL(mp);
-	}
-
-	bp = mp->b_rptr + off;
-	*out = *bp;
-	return (0);
-
-}
-
-static int
-i40e_meoi_get_uint16(mblk_t *mp, off_t off, uint16_t *out)
-{
-	size_t mpsize;
-	uint8_t *bp;
-
-	mpsize = msgsize(mp);
-	/* Check for overflow */
-	if (off + sizeof (uint16_t) > mpsize)
-		return (-1);
-
-	mpsize = MBLKL(mp);
-	while (off >= mpsize) {
-		mp = mp->b_cont;
-		off -= mpsize;
-		mpsize = MBLKL(mp);
-	}
-
-	/*
-	 * Data is in network order. Note the second byte of data might be in
-	 * the next mp.
-	 */
-	bp = mp->b_rptr + off;
-	*out = *bp << 8;
-	if (off + 1 == mpsize) {
-		mp = mp->b_cont;
-		bp = mp->b_rptr;
-	} else {
-		bp++;
-	}
-
-	*out |= *bp;
-	return (0);
-
-}
-
-static int
-mac_ether_offload_info(mblk_t *mp, mac_ether_offload_info_t *meoi)
-{
-	size_t off;
-	uint16_t ether;
-	uint8_t ipproto, iplen, l4len, maclen;
-
-	bzero(meoi, sizeof (mac_ether_offload_info_t));
-
-	off = offsetof(struct ether_header, ether_type);
-	if (i40e_meoi_get_uint16(mp, off, &ether) != 0)
-		return (-1);
-
-	if (ether == ETHERTYPE_VLAN) {
-		off = offsetof(struct ether_vlan_header, ether_type);
-		if (i40e_meoi_get_uint16(mp, off, &ether) != 0)
-			return (-1);
-		meoi->meoi_flags |= MEOI_VLAN_TAGGED;
-		maclen = sizeof (struct ether_vlan_header);
-	} else {
-		maclen = sizeof (struct ether_header);
-	}
-	meoi->meoi_flags |= MEOI_L2INFO_SET;
-	meoi->meoi_l2hlen = maclen;
-	meoi->meoi_l3proto = ether;
-
-	switch (ether) {
-	case ETHERTYPE_IP:
-		/*
-		 * For IPv4 we need to get the length of the header, as it can
-		 * be variable.
-		 */
-		off = offsetof(ipha_t, ipha_version_and_hdr_length) + maclen;
-		if (i40e_meoi_get_uint8(mp, off, &iplen) != 0)
-			return (-1);
-		iplen &= 0x0f;
-		if (iplen < 5 || iplen > 0x0f)
-			return (-1);
-		iplen *= 4;
-		off = offsetof(ipha_t, ipha_protocol) + maclen;
-		if (i40e_meoi_get_uint8(mp, off, &ipproto) == -1)
-			return (-1);
-		break;
-	case ETHERTYPE_IPV6:
-		iplen = 40;
-		off = offsetof(ip6_t, ip6_nxt) + maclen;
-		if (i40e_meoi_get_uint8(mp, off, &ipproto) == -1)
-			return (-1);
-		break;
-	default:
-		return (0);
-	}
-	meoi->meoi_l3hlen = iplen;
-	meoi->meoi_l4proto = ipproto;
-	meoi->meoi_flags |= MEOI_L3INFO_SET;
-
-	switch (ipproto) {
-	case IPPROTO_TCP:
-		off = offsetof(tcph_t, th_offset_and_rsrvd) + maclen + iplen;
-		if (i40e_meoi_get_uint8(mp, off, &l4len) == -1)
-			return (-1);
-		l4len = (l4len & 0xf0) >> 4;
-		if (l4len < 5 || l4len > 0xf)
-			return (-1);
-		l4len *= 4;
-		break;
-	case IPPROTO_UDP:
-		l4len = sizeof (struct udphdr);
-		break;
-	case IPPROTO_SCTP:
-		l4len = sizeof (sctp_hdr_t);
-		break;
-	default:
-		return (0);
-	}
-
-	meoi->meoi_l4hlen = l4len;
-	meoi->meoi_flags |= MEOI_L4INFO_SET;
-	return (0);
-}
-
-/*
  * Attempt to put togther the information we'll need to feed into a descriptor
  * to properly program the hardware for checksum offload as well as the
  * generally required flags.
diff --git a/usr/src/uts/common/io/mac/mac_provider.c b/usr/src/uts/common/io/mac/mac_provider.c
index 0f917cd8ca..7f193f68eb 100644
--- a/usr/src/uts/common/io/mac/mac_provider.c
+++ b/usr/src/uts/common/io/mac/mac_provider.c
@@ -58,6 +58,10 @@
 #include <sys/pattr.h>
 #include <sys/strsun.h>
 #include <sys/vlan.h>
+#include <inet/ip.h>
+#include <inet/tcp.h>
+#include <netinet/udp.h>
+#include <netinet/sctp.h>
 
 /*
  * MAC Provider Interface.
@@ -1653,3 +1657,155 @@ mac_transceiver_info_set_usable(mac_transceiver_info_t *infop,
 {
 	infop->mti_usable = usable;
 }
+
+/*
+ * We should really keep track of our offset and not walk everything every
+ * time. I can't imagine that this will be kind to us at high packet rates;
+ * however, for the moment, let's leave that.
+ *
+ * This walks a message block chain without pulling up to fill in the context
+ * information. Note that the data we care about could be hidden across more
+ * than one mblk_t.
+ */
+static int
+mac_meoi_get_uint8(mblk_t *mp, off_t off, uint8_t *out)
+{
+	size_t mpsize;
+	uint8_t *bp;
+
+	mpsize = msgsize(mp);
+	/* Check for overflow */
+	if (off + sizeof (uint16_t) > mpsize)
+		return (-1);
+
+	mpsize = MBLKL(mp);
+	while (off >= mpsize) {
+		mp = mp->b_cont;
+		off -= mpsize;
+		mpsize = MBLKL(mp);
+	}
+
+	bp = mp->b_rptr + off;
+	*out = *bp;
+	return (0);
+
+}
+
+static int
+mac_meoi_get_uint16(mblk_t *mp, off_t off, uint16_t *out)
+{
+	size_t mpsize;
+	uint8_t *bp;
+
+	mpsize = msgsize(mp);
+	/* Check for overflow */
+	if (off + sizeof (uint16_t) > mpsize)
+		return (-1);
+
+	mpsize = MBLKL(mp);
+	while (off >= mpsize) {
+		mp = mp->b_cont;
+		off -= mpsize;
+		mpsize = MBLKL(mp);
+	}
+
+	/*
+	 * Data is in network order. Note the second byte of data might be in
+	 * the next mp.
+	 */
+	bp = mp->b_rptr + off;
+	*out = *bp << 8;
+	if (off + 1 == mpsize) {
+		mp = mp->b_cont;
+		bp = mp->b_rptr;
+	} else {
+		bp++;
+	}
+
+	*out |= *bp;
+	return (0);
+
+}
+
+
+int
+mac_ether_offload_info(mblk_t *mp, mac_ether_offload_info_t *meoi)
+{
+	size_t off;
+	uint16_t ether;
+	uint8_t ipproto, iplen, l4len, maclen;
+
+	bzero(meoi, sizeof (mac_ether_offload_info_t));
+
+	meoi->meoi_len = msgsize(mp);
+	off = offsetof(struct ether_header, ether_type);
+	if (mac_meoi_get_uint16(mp, off, &ether) != 0)
+		return (-1);
+
+	if (ether == ETHERTYPE_VLAN) {
+		off = offsetof(struct ether_vlan_header, ether_type);
+		if (mac_meoi_get_uint16(mp, off, &ether) != 0)
+			return (-1);
+		meoi->meoi_flags |= MEOI_VLAN_TAGGED;
+		maclen = sizeof (struct ether_vlan_header);
+	} else {
+		maclen = sizeof (struct ether_header);
+	}
+	meoi->meoi_flags |= MEOI_L2INFO_SET;
+	meoi->meoi_l2hlen = maclen;
+	meoi->meoi_l3proto = ether;
+
+	switch (ether) {
+	case ETHERTYPE_IP:
+		/*
+		 * For IPv4 we need to get the length of the header, as it can
+		 * be variable.
+		 */
+		off = offsetof(ipha_t, ipha_version_and_hdr_length) + maclen;
+		if (mac_meoi_get_uint8(mp, off, &iplen) != 0)
+			return (-1);
+		iplen &= 0x0f;
+		if (iplen < 5 || iplen > 0x0f)
+			return (-1);
+		iplen *= 4;
+		off = offsetof(ipha_t, ipha_protocol) + maclen;
+		if (mac_meoi_get_uint8(mp, off, &ipproto) == -1)
+			return (-1);
+		break;
+	case ETHERTYPE_IPV6:
+		iplen = 40;
+		off = offsetof(ip6_t, ip6_nxt) + maclen;
+		if (mac_meoi_get_uint8(mp, off, &ipproto) == -1)
+			return (-1);
+		break;
+	default:
+		return (0);
+	}
+	meoi->meoi_l3hlen = iplen;
+	meoi->meoi_l4proto = ipproto;
+	meoi->meoi_flags |= MEOI_L3INFO_SET;
+
+	switch (ipproto) {
+	case IPPROTO_TCP:
+		off = offsetof(tcph_t, th_offset_and_rsrvd) + maclen + iplen;
+		if (mac_meoi_get_uint8(mp, off, &l4len) == -1)
+			return (-1);
+		l4len = (l4len & 0xf0) >> 4;
+		if (l4len < 5 || l4len > 0xf)
+			return (-1);
+		l4len *= 4;
+		break;
+	case IPPROTO_UDP:
+		l4len = sizeof (struct udphdr);
+		break;
+	case IPPROTO_SCTP:
+		l4len = sizeof (sctp_hdr_t);
+		break;
+	default:
+		return (0);
+	}
+
+	meoi->meoi_l4hlen = l4len;
+	meoi->meoi_flags |= MEOI_L4INFO_SET;
+	return (0);
+}
diff --git a/usr/src/uts/common/io/vioif/vioif.c b/usr/src/uts/common/io/vioif/vioif.c
index e1535182b3..cac90d3073 100644
--- a/usr/src/uts/common/io/vioif/vioif.c
+++ b/usr/src/uts/common/io/vioif/vioif.c
@@ -74,6 +74,7 @@
 #include <sys/random.h>
 #include <sys/containerof.h>
 #include <sys/stream.h>
+#include <inet/tcp.h>
 
 #include <sys/mac.h>
 #include <sys/mac_provider.h>
@@ -1018,7 +1019,7 @@ vioif_send(vioif_t *vif, mblk_t *mp)
 	for (nmp = mp; nmp; nmp = nmp->b_cont)
 		msg_size += MBLKL(nmp);
 
-	if (vif->vif_tx_tso4) {
+	if (vif->vif_tx_tso4 || vif->vif_tx_tso6) {
 		mac_lso_get(mp, &lso_mss, &lso_flags);
 		lso_required = (lso_flags & HW_LSO) != 0;
 	}
@@ -1084,8 +1085,70 @@ vioif_send(vioif_t *vif, mblk_t *mp)
 	 * Setup LSO fields if required.
 	 */
 	if (lso_required) {
-		vnh->vnh_gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+		mac_ether_offload_flags_t needed;
+		mac_ether_offload_info_t meo;
+		uint32_t cksum;
+		size_t len;
+		mblk_t *pullmp = NULL;
+		tcpha_t *tcpha;
+
+		if (mac_ether_offload_info(mp, &meo) != 0) {
+			goto fail;
+		}
+
+		needed = MEOI_L2INFO_SET | MEOI_L3INFO_SET | MEOI_L4INFO_SET;
+		if ((meo.meoi_flags & needed) != needed) {
+			goto fail;
+		}
+
+		if (meo.meoi_l4proto != IPPROTO_TCP) {
+			goto fail;
+		}
+
+		if (meo.meoi_l3proto == ETHERTYPE_IP && vif->vif_tx_tso4) {
+			vnh->vnh_gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+		} else if (meo.meoi_l3proto == ETHERTYPE_IPV6 &&
+		    vif->vif_tx_tso6) {
+			vnh->vnh_gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+		} else {
+			goto fail;
+		}
+
+		/*
+		 * The TCP stack does not include the length in the TCP
+		 * pseudo-header when it is performing LSO since hardware
+		 * generally asks for it to be removed (as it'll change).
+		 * Unfortunately, for virtio, we actually need it. This means we
+		 * need to go through and calculate the actual length and fix
+		 * things up. Because the virtio spec cares about the ECN flag
+		 * and indicating that, at least this means we'll have that
+		 * available as well.
+		 */
+		if (MBLKL(mp) < vnh->vnh_hdr_len) {
+			pullmp = msgpullup(mp, vnh->vnh_hdr_len);
+			if (pullmp == NULL)
+				goto fail;
+			tcpha = (tcpha_t *)(pullmp->b_rptr + meo.meoi_l2hlen +
+			    meo.meoi_l3hlen);
+		} else {
+			tcpha = (tcpha_t *)(mp->b_rptr + meo.meoi_l2hlen +
+			    meo.meoi_l3hlen);
+		}
+
+		len = meo.meoi_len - meo.meoi_l2hlen - meo.meoi_l3hlen;
+		cksum = ntohs(tcpha->tha_sum) + len;
+		cksum = (cksum >> 16) + (cksum & 0xffff);
+		cksum = (cksum >> 16) + (cksum & 0xffff);
+		tcpha->tha_sum = htons(cksum);
+
+		if (tcpha->tha_flags & TH_CWR) {
+			vnh->vnh_gso_type |= VIRTIO_NET_HDR_GSO_ECN;
+		}
 		vnh->vnh_gso_size = (uint16_t)lso_mss;
+		vnh->vnh_hdr_len = meo.meoi_l2hlen + meo.meoi_l3hlen +
+		    meo.meoi_l4hlen;
+
+		freemsg(pullmp);
 	}
 
 	/*
@@ -1453,8 +1516,9 @@ vioif_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
 		}
 
 		mac_capab_lso_t *lso = cap_data;
-		lso->lso_flags = LSO_TX_BASIC_TCP_IPV4;
+		lso->lso_flags = LSO_TX_BASIC_TCP_IPV4 | LSO_TX_BASIC_TCP_IPV6;
 		lso->lso_basic_tcp_ipv4.lso_max = VIOIF_RX_DATA_SIZE;
+		lso->lso_basic_tcp_ipv6.lso_max = VIOIF_RX_DATA_SIZE;
 
 		return (B_TRUE);
 	}
@@ -1556,6 +1620,7 @@ vioif_check_features(vioif_t *vif)
 
 	vif->vif_tx_csum = 0;
 	vif->vif_tx_tso4 = 0;
+	vif->vif_tx_tso6 = 0;
 
 	if (vioif_has_feature(vif, VIRTIO_NET_F_CSUM)) {
 		/*
@@ -1569,6 +1634,7 @@ vioif_check_features(vioif_t *vif)
 		 */
 		boolean_t gso = vioif_has_feature(vif, VIRTIO_NET_F_GSO);
 		boolean_t tso4 = vioif_has_feature(vif, VIRTIO_NET_F_HOST_TSO4);
+		boolean_t tso6 = vioif_has_feature(vif, VIRTIO_NET_F_HOST_TSO6);
 		boolean_t ecn = vioif_has_feature(vif, VIRTIO_NET_F_HOST_ECN);
 
 		/*
@@ -1578,8 +1644,15 @@ vioif_check_features(vioif_t *vif)
 		 * we require the device to support the combination of
 		 * segmentation offload and ECN support.
 		 */
-		if (gso || (tso4 && ecn)) {
+		if (gso) {
 			vif->vif_tx_tso4 = 1;
+			vif->vif_tx_tso6 = 1;
+		}
+		if (tso4 && ecn) {
+			vif->vif_tx_tso4 = 1;
+		}
+		if (tso6 && ecn) {
+			vif->vif_tx_tso6 = 1;
 		}
 	}
 }
diff --git a/usr/src/uts/common/io/vioif/vioif.h b/usr/src/uts/common/io/vioif/vioif.h
index 19d8965bd4..9f750c9b8a 100644
--- a/usr/src/uts/common/io/vioif/vioif.h
+++ b/usr/src/uts/common/io/vioif/vioif.h
@@ -164,6 +164,7 @@ extern "C" {
 #define	VIRTIO_NET_WANTED_FEATURES	(VIRTIO_NET_F_CSUM |		\
 					VIRTIO_NET_F_GSO |		\
 					VIRTIO_NET_F_HOST_TSO4 |	\
+					VIRTIO_NET_F_HOST_TSO6 |	\
 					VIRTIO_NET_F_HOST_ECN |		\
 					VIRTIO_NET_F_MAC |		\
 					VIRTIO_NET_F_MTU)
@@ -356,6 +357,7 @@ struct vioif {
 	 */
 	unsigned int			vif_tx_csum:1;
 	unsigned int			vif_tx_tso4:1;
+	unsigned int			vif_tx_tso6:1;
 
 	/*
 	 * For debugging, it is useful to know whether the MAC address we