12520 Enable IPv6 tso in igb, ixgbe, and i40e

Reviewed by: Patrick Mooney <pmooney@pfmooney.com>
Approved by: Dan McDonald <danmcd@joyent.com>

8 files changed, 106 insertions, 41 deletions

diff --git a/usr/src/uts/common/io/i40e/i40e_gld.c b/usr/src/uts/common/io/i40e/i40e_gld.c
index ccf814be0b..1bf40bbce3 100644
--- a/usr/src/uts/common/io/i40e/i40e_gld.c
+++ b/usr/src/uts/common/io/i40e/i40e_gld.c
@@ -733,8 +733,10 @@ i40e_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
 		mac_capab_lso_t *cap_lso = cap_data;
 
 		if (i40e->i40e_tx_lso_enable == B_TRUE) {
-			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4;
+			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4 |
+			    LSO_TX_BASIC_TCP_IPV6;
 			cap_lso->lso_basic_tcp_ipv4.lso_max = I40E_LSO_MAXLEN;
+			cap_lso->lso_basic_tcp_ipv6.lso_max = I40E_LSO_MAXLEN;
 		} else {
 			return (B_FALSE);
 		}
diff --git a/usr/src/uts/common/io/i40e/i40e_transceiver.c b/usr/src/uts/common/io/i40e/i40e_transceiver.c
index 9662cb58f5..e324957625 100644
--- a/usr/src/uts/common/io/i40e/i40e_transceiver.c
+++ b/usr/src/uts/common/io/i40e/i40e_transceiver.c
@@ -1782,7 +1782,8 @@ i40e_tx_context(i40e_t *i40e, i40e_trqpair_t *itrq, mblk_t *mp,
 		 * LSO requires that checksum offloads are enabled.  If for
 		 * some reason they're not we bail out with an error.
 		 */
-		if ((chkflags & HCK_IPV4_HDRCKSUM) == 0 ||
+		if ((meo->meoi_l3proto == ETHERTYPE_IP &&
+		    (chkflags & HCK_IPV4_HDRCKSUM) == 0) ||
 		    (chkflags & HCK_PARTIALCKSUM) == 0) {
 			txs->itxs_lso_nohck.value.ui64++;
 			return (-1);
diff --git a/usr/src/uts/common/io/igb/igb_gld.c b/usr/src/uts/common/io/igb/igb_gld.c
index 2bb4f99d6f..f68c4b73e9 100644
--- a/usr/src/uts/common/io/igb/igb_gld.c
+++ b/usr/src/uts/common/io/igb/igb_gld.c
@@ -965,8 +965,10 @@ igb_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
 		mac_capab_lso_t *cap_lso = cap_data;
 
 		if (igb->lso_enable) {
-			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4;
+			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4 |
+			    LSO_TX_BASIC_TCP_IPV6;
 			cap_lso->lso_basic_tcp_ipv4.lso_max = IGB_LSO_MAXLEN;
+			cap_lso->lso_basic_tcp_ipv6.lso_max = IGB_LSO_MAXLEN;
 			break;
 		} else {
 			return (B_FALSE);
@@ -1645,7 +1647,7 @@ igb_priv_prop_info(igb_t *igb, const char *pr_name, mac_prop_info_handle_t prh)
 		value = DEFAULT_RX_COPY_THRESHOLD;
 	} else if (strcmp(pr_name, "_rx_limit_per_intr") == 0) {
 		value = DEFAULT_RX_LIMIT_PER_INTR;
-	} else 	if (strcmp(pr_name, "_intr_throttling") == 0) {
+	} else if (strcmp(pr_name, "_intr_throttling") == 0) {
 		value = igb->capab->def_intr_throttle;
 	} else {
 		return;
diff --git a/usr/src/uts/common/io/igb/igb_sw.h b/usr/src/uts/common/io/igb/igb_sw.h
index f689c44b14..2e329d6fdc 100644
--- a/usr/src/uts/common/io/igb/igb_sw.h
+++ b/usr/src/uts/common/io/igb/igb_sw.h
@@ -227,7 +227,7 @@ extern "C" {
 
 enum ioc_reply {
 	IOC_INVAL = -1,	/* bad, NAK with EINVAL */
-	IOC_DONE, 	/* OK, reply sent */
+	IOC_DONE,	/* OK, reply sent */
 	IOC_ACK,	/* OK, just send ACK */
 	IOC_REPLY	/* OK, just send reply */
 };
@@ -355,6 +355,7 @@ typedef struct tx_context {
 	uint32_t		hcksum_flags;
 	uint32_t		ip_hdr_len;
 	uint32_t		mac_hdr_len;
+	uint32_t		l3_proto;
 	uint32_t		l4_proto;
 	uint32_t		mss;
 	uint32_t		l4_hdr_len;
@@ -558,7 +559,7 @@ typedef struct igb_rx_group {
 } igb_rx_group_t;
 
 typedef struct igb {
-	int 			instance;
+	int			instance;
 	mac_handle_t		mac_hdl;
 	dev_info_t		*dip;
 	struct e1000_hw		hw;
@@ -606,7 +607,7 @@ typedef struct igb {
 	boolean_t		tx_ring_init;
 	boolean_t		tx_head_wb_enable; /* Tx head wrtie-back */
 	boolean_t		tx_hcksum_enable; /* Tx h/w cksum offload */
-	boolean_t 		lso_enable; 	/* Large Segment Offload */
+	boolean_t		lso_enable;	/* Large Segment Offload */
 	uint32_t		tx_copy_thresh;	/* Tx copy threshold */
 	uint32_t		tx_recycle_thresh; /* Tx recycle threshold */
 	uint32_t		tx_overload_thresh; /* Tx overload threshold */
diff --git a/usr/src/uts/common/io/igb/igb_tx.c b/usr/src/uts/common/io/igb/igb_tx.c
index 0496bf59fe..5245312adc 100644
--- a/usr/src/uts/common/io/igb/igb_tx.c
+++ b/usr/src/uts/common/io/igb/igb_tx.c
@@ -598,6 +598,7 @@ igb_get_tx_context(mblk_t *mp, tx_context_t *ctx)
 	uint32_t start;
 	uint32_t flags;
 	uint32_t lso_flag;
+	uint32_t lso_cksum;
 	uint32_t mss;
 	uint32_t len;
 	uint32_t size;
@@ -622,19 +623,6 @@ igb_get_tx_context(mblk_t *mp, tx_context_t *ctx)
 	ctx->mss = mss;
 	ctx->lso_flag = (lso_flag == HW_LSO);
 
-	/*
-	 * LSO relies on tx h/w checksum, so here the packet will be
-	 * dropped if the h/w checksum flags are not set.
-	 */
-	if (ctx->lso_flag) {
-		if (!((ctx->hcksum_flags & HCK_PARTIALCKSUM) &&
-		    (ctx->hcksum_flags & HCK_IPV4_HDRCKSUM))) {
-			igb_log(NULL, IGB_LOG_INFO, "igb_tx: h/w "
-			    "checksum flags are not set for LSO");
-			return (TX_CXT_E_LSO_CSUM);
-		}
-	}
-
 	etype = 0;
 	mac_hdr_len = 0;
 	l4_proto = 0;
@@ -679,6 +667,8 @@ igb_get_tx_context(mblk_t *mp, tx_context_t *ctx)
 	 * Here we assume the IP(V6) header is fully included in one
 	 * mblk fragment.
 	 */
+	lso_cksum = HCK_PARTIALCKSUM;
+	ctx->l3_proto = etype;
 	switch (etype) {
 	case ETHERTYPE_IP:
 		offset = mac_hdr_len;
@@ -703,11 +693,27 @@ igb_get_tx_context(mblk_t *mp, tx_context_t *ctx)
 			 * with zero. Currently the tcp/ip stack has done
 			 * these.
 			 */
+			lso_cksum |= HCK_IPV4_HDRCKSUM;
 		}
 
 		l4_proto = *(uint8_t *)(pos + offsetof(ipha_t, ipha_protocol));
 		break;
 	case ETHERTYPE_IPV6:
+		/*
+		 * We need to zero out the length in the header.
+		 */
+		if (ctx->lso_flag) {
+			offset = offsetof(ip6_t, ip6_plen) + mac_hdr_len;
+			while (size <= offset) {
+				mp = mp->b_cont;
+				ASSERT(mp != NULL);
+				len = MBLKL(mp);
+				size += len;
+			}
+			pos = mp->b_rptr + offset + len - size;
+			*((uint16_t *)(uintptr_t)(pos)) = 0;
+		}
+
 		offset = offsetof(ip6_t, ip6_nxt) + mac_hdr_len;
 		while (size <= offset) {
 			mp = mp->b_cont;
@@ -727,6 +733,18 @@ igb_get_tx_context(mblk_t *mp, tx_context_t *ctx)
 	}
 
 	if (ctx->lso_flag) {
+		/*
+		 * LSO relies on tx h/w checksum, so here the packet will be
+		 * dropped if the h/w checksum flags are not set.
+		 */
+		if ((ctx->hcksum_flags & lso_cksum) != lso_cksum) {
+			igb_log(NULL, IGB_LOG_INFO, "igb_tx: h/w "
+			    "checksum flags are not set for LSO, found "
+			    "0x%x, needed bits 0x%x", ctx->hcksum_flags,
+			    lso_cksum);
+			return (TX_CXT_E_LSO_CSUM);
+		}
+
 		offset = mac_hdr_len + start;
 		while (size <= offset) {
 			mp = mp->b_cont;
@@ -771,6 +789,7 @@ igb_check_tx_context(igb_tx_ring_t *tx_ring, tx_context_t *ctx)
 	 * need to be checked are:
 	 *	hcksum_flags
 	 *	l4_proto
+	 *	l3_proto
 	 *	mss (only check for LSO)
 	 *	l4_hdr_len (only check for LSO)
 	 *	ip_hdr_len
@@ -783,6 +802,7 @@ igb_check_tx_context(igb_tx_ring_t *tx_ring, tx_context_t *ctx)
 	if (ctx->hcksum_flags != 0) {
 		if ((ctx->hcksum_flags != last->hcksum_flags) ||
 		    (ctx->l4_proto != last->l4_proto) ||
+		    (ctx->l3_proto != last->l3_proto) ||
 		    (ctx->lso_flag && ((ctx->mss != last->mss) ||
 		    (ctx->l4_hdr_len != last->l4_hdr_len))) ||
 		    (ctx->ip_hdr_len != last->ip_hdr_len) ||
@@ -814,10 +834,19 @@ igb_fill_tx_context(struct e1000_adv_tx_context_desc *ctx_tbd,
 	ctx_tbd->type_tucmd_mlhl =
 	    E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
 
-	if (ctx->hcksum_flags & HCK_IPV4_HDRCKSUM)
-		ctx_tbd->type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
+	/*
+	 * When we have a TX context set up, we enforce that the ethertype is
+	 * either IPv4 or IPv6 in igb_get_tx_context().
+	 */
+	if (ctx->lso_flag || ctx->hcksum_flags & HCK_IPV4_HDRCKSUM) {
+		if (ctx->l3_proto == ETHERTYPE_IP) {
+			ctx_tbd->type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
+		} else {
+			ctx_tbd->type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6;
+		}
+	}
 
-	if (ctx->hcksum_flags & HCK_PARTIALCKSUM) {
+	if (ctx->lso_flag || ctx->hcksum_flags & HCK_PARTIALCKSUM) {
 		switch (ctx->l4_proto) {
 		case IPPROTO_TCP:
 			ctx_tbd->type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
diff --git a/usr/src/uts/common/io/ixgbe/ixgbe_gld.c b/usr/src/uts/common/io/ixgbe/ixgbe_gld.c
index ea888220fa..8508e32b82 100644
--- a/usr/src/uts/common/io/ixgbe/ixgbe_gld.c
+++ b/usr/src/uts/common/io/ixgbe/ixgbe_gld.c
@@ -304,8 +304,10 @@ ixgbe_m_getcapab(void *arg, mac_capab_t cap, void *cap_data)
 		mac_capab_lso_t *cap_lso = cap_data;
 
 		if (ixgbe->lso_enable) {
-			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4;
+			cap_lso->lso_flags = LSO_TX_BASIC_TCP_IPV4 |
+			    LSO_TX_BASIC_TCP_IPV6;
 			cap_lso->lso_basic_tcp_ipv4.lso_max = IXGBE_LSO_MAXLEN;
+			cap_lso->lso_basic_tcp_ipv6.lso_max = IXGBE_LSO_MAXLEN;
 			break;
 		} else {
 			return (B_FALSE);
diff --git a/usr/src/uts/common/io/ixgbe/ixgbe_sw.h b/usr/src/uts/common/io/ixgbe/ixgbe_sw.h
index cfd987787a..8bf93afe9c 100644
--- a/usr/src/uts/common/io/ixgbe/ixgbe_sw.h
+++ b/usr/src/uts/common/io/ixgbe/ixgbe_sw.h
@@ -408,6 +408,7 @@ typedef struct ixgbe_tx_context {
 	uint32_t		hcksum_flags;
 	uint32_t		ip_hdr_len;
 	uint32_t		mac_hdr_len;
+	uint32_t		l3_proto;
 	uint32_t		l4_proto;
 	uint32_t		mss;
 	uint32_t		l4_hdr_len;
diff --git a/usr/src/uts/common/io/ixgbe/ixgbe_tx.c b/usr/src/uts/common/io/ixgbe/ixgbe_tx.c
index 9191aad058..b6926fc3b7 100644
--- a/usr/src/uts/common/io/ixgbe/ixgbe_tx.c
+++ b/usr/src/uts/common/io/ixgbe/ixgbe_tx.c
@@ -697,6 +697,7 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 	uint32_t start;
 	uint32_t hckflags;
 	uint32_t lsoflags;
+	uint32_t lsocksum;
 	uint32_t mss;
 	uint32_t len;
 	uint32_t size;
@@ -722,19 +723,6 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 	ctx->mss = mss;
 	ctx->lso_flag = (lsoflags == HW_LSO);
 
-	/*
-	 * LSO relies on tx h/w checksum, so here will drop the package
-	 * if h/w checksum flag is not declared.
-	 */
-	if (ctx->lso_flag) {
-		if (!((ctx->hcksum_flags & HCK_PARTIALCKSUM) &&
-		    (ctx->hcksum_flags & HCK_IPV4_HDRCKSUM))) {
-			IXGBE_DEBUGLOG_0(NULL, "ixgbe_tx: h/w "
-			    "checksum flags are not specified when doing LSO");
-			return (-1);
-		}
-	}
-
 	etype = 0;
 	mac_hdr_len = 0;
 	l4_proto = 0;
@@ -779,6 +767,8 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 	 * Here we don't assume the IP(V6) header is fully included in
 	 * one mblk fragment.
 	 */
+	lsocksum = HCK_PARTIALCKSUM;
+	ctx->l3_proto = etype;
 	switch (etype) {
 	case ETHERTYPE_IP:
 		if (ctx->lso_flag) {
@@ -810,6 +800,7 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 			 * (ip_source_addr, ip_destination_addr, l4_proto)
 			 * Currently the tcp/ip stack has done it.
 			 */
+			lsocksum |= HCK_IPV4_HDRCKSUM;
 		}
 
 		offset = offsetof(ipha_t, ipha_protocol) + mac_hdr_len;
@@ -824,6 +815,21 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 		l4_proto = *(uint8_t *)pos;
 		break;
 	case ETHERTYPE_IPV6:
+		/*
+		 * We need to zero out the length in the header.
+		 */
+		if (ctx->lso_flag) {
+			offset = offsetof(ip6_t, ip6_plen) + mac_hdr_len;
+			while (size <= offset) {
+				mp = mp->b_cont;
+				ASSERT(mp != NULL);
+				len = MBLKL(mp);
+				size += len;
+			}
+			pos = mp->b_rptr + offset + len - size;
+			*((uint16_t *)(uintptr_t)(pos)) = 0;
+		}
+
 		offset = offsetof(ip6_t, ip6_nxt) + mac_hdr_len;
 		while (size <= offset) {
 			mp = mp->b_cont;
@@ -842,6 +848,18 @@ ixgbe_get_context(mblk_t *mp, ixgbe_tx_context_t *ctx)
 	}
 
 	if (ctx->lso_flag) {
+		/*
+		 * LSO relies on tx h/w checksum, so here will drop the packet
+		 * if h/w checksum flag is not declared.
+		 */
+		if ((ctx->hcksum_flags & lsocksum) != lsocksum) {
+			IXGBE_DEBUGLOG_2(NULL, "ixgbe_tx: h/w checksum flags "
+			    "are not set for LSO, found 0x%x, needed bits 0x%x",
+			    ctx->hcksum_flags, lsocksum);
+			return (-1);
+		}
+
+
 		offset = mac_hdr_len + start;
 		while (size <= offset) {
 			mp = mp->b_cont;
@@ -898,6 +916,7 @@ ixgbe_check_context(ixgbe_tx_ring_t *tx_ring, ixgbe_tx_context_t *ctx)
 
 	if ((ctx->hcksum_flags != last->hcksum_flags) ||
 	    (ctx->l4_proto != last->l4_proto) ||
+	    (ctx->l3_proto != last->l3_proto) ||
 	    (ctx->mac_hdr_len != last->mac_hdr_len) ||
 	    (ctx->ip_hdr_len != last->ip_hdr_len) ||
 	    (ctx->lso_flag != last->lso_flag) ||
@@ -928,11 +947,19 @@ ixgbe_fill_context(struct ixgbe_adv_tx_context_desc *ctx_tbd,
 
 	ctx_tbd->type_tucmd_mlhl =
 	    IXGBE_ADVTXD_DCMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
+	/*
+	 * When we have a TX context set up, we enforce that the ethertype is
+	 * either IPv4 or IPv6 in ixgbe_get_tx_context().
+	 */
+	if (ctx->lso_flag || ctx->hcksum_flags & HCK_IPV4_HDRCKSUM) {
+		if (ctx->l3_proto == ETHERTYPE_IP) {
+			ctx_tbd->type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
+		} else {
+			ctx_tbd->type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV6;
+		}
+	}
 
-	if (ctx->hcksum_flags & HCK_IPV4_HDRCKSUM)
-		ctx_tbd->type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
-
-	if (ctx->hcksum_flags & HCK_PARTIALCKSUM) {
+	if (ctx->lso_flag || ctx->hcksum_flags & HCK_PARTIALCKSUM) {
 		switch (ctx->l4_proto) {
 		case IPPROTO_TCP:
 			ctx_tbd->type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;