bnx2x: Multiple concurrent l2 traffic classes

Overview:
 Support mapping of priorities to traffic classes and
 traffic classes to transmission queues ranges in the net device.
 The queue ranges are (count, offset) pairs relating to the txq
 array.
 This can be done via DCBX negotiation or by kernel.
 As a result Enhanced Transmission Selection (ETS) and Priority Flow
 Control (PFC) are supported between L2 network traffic classes.

 Mapping:
 This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and
 netdev_set_tc_queue functions to map priorities to traffic classes
 and traffic classes to transmission queue ranges.
 This mapping is performed by bnx2x_setup_tc function which is
 connected to the ndo_setup_tc.
 This function is always called at nic load where by default it
 maps all priorities to tc 0, and it may also be called by the
 kernel or by the bnx2x upon DCBX negotiation to modify the mapping.

 rtnl lock:
 When the ndo_setup_tc is called at nic load or by kernel the rtnl
 lock is already taken. However, when DCBX negotiation takes place
 the lock is not taken. The work is therefore scheduled to be
 handled by the sp_rtnl task.

 Fastpath:
 The fastpath structure of the bnx2x which was previously used
 to hold the information of one tx queue and one rx queue was
 redesigned to represent multiple tx queues, one for each traffic
 class.
 The transmission queue supplied in the skb by the kernel can no
 longer be interpreted as a straightforward index into the fastpath
 structure array, but it must rather be decoded to the appropriate
 fastpath index and the tc within that fastpath.

 Slowpath:
 The bnx2x's queue object was redesigned to accommodate multiple
 transmission queues. The queue object's state machine was enhanced
 to allow opening multiple transmission-only connections on top of
 the regular tx-rx connection.

 Firmware:
 This feature relies on the tx-only queue feature introduced in the
 bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi
 cos support.

 Signed-off-by: Ariel Elior <ariele@broadcom.com>
 Signed-off-by: Eilon Greenstein <eilong@broadcom.com>

Signed-off-by: David S. Miller <davem@davemloft.net>

11 files changed, 1316 insertions, 524 deletions

diff --git a/drivers/net/bnx2x/bnx2x.h b/drivers/net/bnx2x/bnx2x.h
index 16dc2c9df8b7..53fa8ea983e9 100644
--- a/drivers/net/bnx2x/bnx2x.h
+++ b/drivers/net/bnx2x/bnx2x.h
@@ -120,6 +120,7 @@ do {								 \
 
 
 #ifdef BNX2X_STOP_ON_ERROR
+void bnx2x_int_disable(struct bnx2x *bp);
 #define bnx2x_panic() do { \
 		bp->panic = 1; \
 		BNX2X_ERR("driver assert\n"); \
@@ -240,21 +241,21 @@ do {								 \
  */
 /* iSCSI L2 */
 #define BNX2X_ISCSI_ETH_CL_ID_IDX	1
-#define BNX2X_ISCSI_ETH_CID		17
+#define BNX2X_ISCSI_ETH_CID		49
 
 /* FCoE L2 */
 #define BNX2X_FCOE_ETH_CL_ID_IDX	2
-#define BNX2X_FCOE_ETH_CID		18
+#define BNX2X_FCOE_ETH_CID		50
 
 /** Additional rings budgeting */
 #ifdef BCM_CNIC
-#define CNIC_CONTEXT_USE		1
-#define FCOE_CONTEXT_USE		1
+#define CNIC_PRESENT			1
+#define FCOE_PRESENT			1
 #else
-#define CNIC_CONTEXT_USE		0
-#define FCOE_CONTEXT_USE		0
+#define CNIC_PRESENT			0
+#define FCOE_PRESENT			0
 #endif /* BCM_CNIC */
-#define NONE_ETH_CONTEXT_USE	(FCOE_CONTEXT_USE)
+#define NON_ETH_CONTEXT_USE	(FCOE_PRESENT)
 
 #define AEU_IN_ATTN_BITS_PXPPCICLOCKCLIENT_PARITY_ERROR \
 	AEU_INPUTS_ATTN_BITS_PXPPCICLOCKCLIENT_PARITY_ERROR
@@ -262,8 +263,35 @@ do {								 \
 #define SM_RX_ID			0
 #define SM_TX_ID			1
 
-/* fast path */
+/* defines for multiple tx priority indices */
+#define FIRST_TX_ONLY_COS_INDEX		1
+#define FIRST_TX_COS_INDEX		0
+
+/* defines for decodeing the fastpath index and the cos index out of the
+ * transmission queue index
+ */
+#define MAX_TXQS_PER_COS	FP_SB_MAX_E1x
+
+#define TXQ_TO_FP(txq_index)	((txq_index) % MAX_TXQS_PER_COS)
+#define TXQ_TO_COS(txq_index)	((txq_index) / MAX_TXQS_PER_COS)
+
+/* rules for calculating the cids of tx-only connections */
+#define CID_TO_FP(cid)		((cid) % MAX_TXQS_PER_COS)
+#define CID_COS_TO_TX_ONLY_CID(cid, cos)	(cid + cos * MAX_TXQS_PER_COS)
+
+/* fp index inside class of service range */
+#define FP_COS_TO_TXQ(fp, cos)    ((fp)->index + cos * MAX_TXQS_PER_COS)
+
+/*
+ * 0..15 eth cos0
+ * 16..31 eth cos1 if applicable
+ * 32..47 eth cos2 If applicable
+ * fcoe queue follows eth queues (16, 32, 48 depending on cos)
+ */
+#define MAX_ETH_TXQ_IDX(bp)	(MAX_TXQS_PER_COS * (bp)->max_cos)
+#define FCOE_TXQ_IDX(bp)	(MAX_ETH_TXQ_IDX(bp))
 
+/* fast path */
 struct sw_rx_bd {
 	struct sk_buff	*skb;
 	DEFINE_DMA_UNMAP_ADDR(mapping);
@@ -388,6 +416,29 @@ struct bnx2x_agg_info {
 #define Q_STATS_OFFSET32(stat_name) \
 			(offsetof(struct bnx2x_eth_q_stats, stat_name) / 4)
 
+struct bnx2x_fp_txdata {
+
+	struct sw_tx_bd		*tx_buf_ring;
+
+	union eth_tx_bd_types	*tx_desc_ring;
+	dma_addr_t		tx_desc_mapping;
+
+	u32			cid;
+
+	union db_prod		tx_db;
+
+	u16			tx_pkt_prod;
+	u16			tx_pkt_cons;
+	u16			tx_bd_prod;
+	u16			tx_bd_cons;
+
+	unsigned long		tx_pkt;
+
+	__le16			*tx_cons_sb;
+
+	int			txq_index;
+};
+
 struct bnx2x_fastpath {
 	struct bnx2x		*bp; /* parent */
 
@@ -404,10 +455,8 @@ struct bnx2x_fastpath {
 
 	dma_addr_t		status_blk_mapping;
 
-	struct sw_tx_bd		*tx_buf_ring;
-
-	union eth_tx_bd_types	*tx_desc_ring;
-	dma_addr_t		tx_desc_mapping;
+	u8			max_cos; /* actual number of active tx coses */
+	struct bnx2x_fp_txdata	txdata[BNX2X_MULTI_TX_COS];
 
 	struct sw_rx_bd		*rx_buf_ring;	/* BDs mappings ring */
 	struct sw_rx_page	*rx_page_ring;	/* SGE pages mappings ring */
@@ -426,20 +475,13 @@ struct bnx2x_fastpath {
 
 	u32			cid;
 
+	__le16			fp_hc_idx;
+
 	u8			index;		/* number in fp array */
 	u8			cl_id;		/* eth client id */
 	u8			cl_qzone_id;
 	u8			fw_sb_id;	/* status block number in FW */
 	u8			igu_sb_id;	/* status block number in HW */
-	union db_prod		tx_db;
-
-	u16			tx_pkt_prod;
-	u16			tx_pkt_cons;
-	u16			tx_bd_prod;
-	u16			tx_bd_cons;
-	__le16			*tx_cons_sb;
-
-	__le16			fp_hc_idx;
 
 	u16			rx_bd_prod;
 	u16			rx_bd_cons;
@@ -449,8 +491,7 @@ struct bnx2x_fastpath {
 	/* The last maximal completed SGE */
 	u16			last_max_sge;
 	__le16			*rx_cons_sb;
-	unsigned long		tx_pkt,
-				rx_pkt,
+	unsigned long		rx_pkt,
 				rx_calls;
 
 	/* TPA related */
@@ -489,8 +530,12 @@ struct bnx2x_fastpath {
 #define FCOE_IDX			BNX2X_NUM_ETH_QUEUES(bp)
 #define bnx2x_fcoe_fp(bp)		(&bp->fp[FCOE_IDX])
 #define bnx2x_fcoe(bp, var)		(bnx2x_fcoe_fp(bp)->var)
+#define bnx2x_fcoe_tx(bp, var)		(bnx2x_fcoe_fp(bp)-> \
+						txdata[FIRST_TX_COS_INDEX].var)
 
 
+#define IS_ETH_FP(fp)			(fp->index < \
+					 BNX2X_NUM_ETH_QUEUES(fp->bp))
 #ifdef BCM_CNIC
 #define IS_FCOE_FP(fp)			(fp->index == FCOE_IDX)
 #define IS_FCOE_IDX(idx)		((idx) == FCOE_IDX)
@@ -649,18 +694,23 @@ struct bnx2x_fastpath {
 
 #define HC_INDEX_TOE_TX_CQ_CONS		4 /* Formerly Cstorm TOE CQ index   */
 					  /* (HC_INDEX_C_TOE_TX_CQ_CONS)    */
-#define HC_INDEX_ETH_TX_CQ_CONS		5 /* Formerly Cstorm ETH CQ index   */
+#define HC_INDEX_ETH_TX_CQ_CONS_COS0	5 /* Formerly Cstorm ETH CQ index   */
+					  /* (HC_INDEX_C_ETH_TX_CQ_CONS)    */
+#define HC_INDEX_ETH_TX_CQ_CONS_COS1	6 /* Formerly Cstorm ETH CQ index   */
+					  /* (HC_INDEX_C_ETH_TX_CQ_CONS)    */
+#define HC_INDEX_ETH_TX_CQ_CONS_COS2	7 /* Formerly Cstorm ETH CQ index   */
 					  /* (HC_INDEX_C_ETH_TX_CQ_CONS)    */
 
-#define U_SB_ETH_RX_CQ_INDEX		HC_INDEX_ETH_RX_CQ_CONS
-#define U_SB_ETH_RX_BD_INDEX		HC_INDEX_ETH_RX_BD_CONS
-#define C_SB_ETH_TX_CQ_INDEX		HC_INDEX_ETH_TX_CQ_CONS
+#define HC_INDEX_ETH_FIRST_TX_CQ_CONS	HC_INDEX_ETH_TX_CQ_CONS_COS0
+
 
 #define BNX2X_RX_SB_INDEX \
 	(&fp->sb_index_values[HC_INDEX_ETH_RX_CQ_CONS])
 
-#define BNX2X_TX_SB_INDEX \
-	(&fp->sb_index_values[C_SB_ETH_TX_CQ_INDEX])
+#define BNX2X_TX_SB_INDEX_BASE BNX2X_TX_SB_INDEX_COS0
+
+#define BNX2X_TX_SB_INDEX_COS0 \
+	(&fp->sb_index_values[HC_INDEX_ETH_TX_CQ_CONS_COS0])
 
 /* end of fast path */
 
@@ -845,25 +895,6 @@ extern struct workqueue_struct *bnx2x_wq;
 /* fast-path interrupt contexts E2 */
 #define FP_SB_MAX_E2		HC_SB_MAX_SB_E2
 
-/*
- * cid_cnt paramter below refers to the value returned by
- * 'bnx2x_get_l2_cid_count()' routine
- */
-
-/*
- * The number of FP context allocated by the driver == max number of regular
- * L2 queues + 1 for the FCoE L2 queue
- */
-#define L2_FP_COUNT(cid_cnt)	((cid_cnt) - FCOE_CONTEXT_USE)
-
-/*
- * The number of FP-SB allocated by the driver == max number of regular L2
- * queues + 1 for the CNIC which also consumes an FP-SB
- */
-#define FP_SB_COUNT(cid_cnt)	((cid_cnt) - CNIC_CONTEXT_USE)
-#define NUM_IGU_SB_REQUIRED(cid_cnt) \
-				(FP_SB_COUNT(cid_cnt) - NONE_ETH_CONTEXT_USE)
-
 union cdu_context {
 	struct eth_context eth;
 	char pad[1024];
@@ -871,7 +902,7 @@ union cdu_context {
 
 /* CDU host DB constants */
 #define CDU_ILT_PAGE_SZ_HW	3
-#define CDU_ILT_PAGE_SZ		(4096 << CDU_ILT_PAGE_SZ_HW) /* 32K */
+#define CDU_ILT_PAGE_SZ		(8192 << CDU_ILT_PAGE_SZ_HW) /* 64K */
 #define ILT_PAGE_CIDS		(CDU_ILT_PAGE_SZ / sizeof(union cdu_context))
 
 #ifdef BCM_CNIC
@@ -1048,6 +1079,7 @@ struct bnx2x_fw_stats_data {
 
 /* Public slow path states */
 enum {
+	BNX2X_SP_RTNL_SETUP_TC,
 	BNX2X_SP_RTNL_TX_TIMEOUT,
 };
 
@@ -1226,6 +1258,10 @@ struct bnx2x {
 #define BNX2X_STATE_ERROR		0xf000
 
 	int			multi_mode;
+#define BNX2X_MAX_PRIORITY		8
+#define BNX2X_MAX_ENTRIES_PER_PRI	16
+#define BNX2X_MAX_COS			3
+#define BNX2X_MAX_TX_COS		2
 	int			num_queues;
 	int			disable_tpa;
 
@@ -1275,11 +1311,21 @@ struct bnx2x {
 	struct bnx2x_ilt	*ilt;
 #define BP_ILT(bp)		((bp)->ilt)
 #define ILT_MAX_LINES		256
+/*
+ * Maximum supported number of RSS queues: number of IGU SBs minus one that goes
+ * to CNIC.
+ */
+#define BNX2X_MAX_RSS_COUNT(bp)	((bp)->igu_sb_cnt - CNIC_PRESENT)
 
-	int			l2_cid_count;
-#define L2_ILT_LINES(bp)	(DIV_ROUND_UP((bp)->l2_cid_count, \
-				 ILT_PAGE_CIDS))
-#define BNX2X_DB_SIZE(bp)	((bp)->l2_cid_count * (1 << BNX2X_DB_SHIFT))
+/*
+ * Maximum CID count that might be required by the bnx2x:
+ * Max Tss * Max_Tx_Multi_Cos + CNIC L2 Clients (FCoE and iSCSI related)
+ */
+#define BNX2X_L2_CID_COUNT(bp)	(MAX_TXQS_PER_COS * BNX2X_MULTI_TX_COS +\
+					NON_ETH_CONTEXT_USE + CNIC_PRESENT)
+#define L2_ILT_LINES(bp)	(DIV_ROUND_UP(BNX2X_L2_CID_COUNT(bp),\
+					ILT_PAGE_CIDS))
+#define BNX2X_DB_SIZE(bp)	(BNX2X_L2_CID_COUNT(bp) * (1 << BNX2X_DB_SHIFT))
 
 	int			qm_cid_count;
 
@@ -1421,16 +1467,24 @@ struct bnx2x {
 	u32					dcbx_remote_flags;
 #endif
 	u32					pending_max;
+
+	/* multiple tx classes of service */
+	u8					max_cos;
+
+	/* priority to cos mapping */
+	u8					prio_to_cos[8];
 };
 
 /* Tx queues may be less or equal to Rx queues */
 extern int num_queues;
 #define BNX2X_NUM_QUEUES(bp)	(bp->num_queues)
-#define BNX2X_NUM_ETH_QUEUES(bp) (BNX2X_NUM_QUEUES(bp) - NONE_ETH_CONTEXT_USE)
+#define BNX2X_NUM_ETH_QUEUES(bp) (BNX2X_NUM_QUEUES(bp) - NON_ETH_CONTEXT_USE)
+#define BNX2X_NUM_RX_QUEUES(bp)	BNX2X_NUM_QUEUES(bp)
 
 #define is_multi(bp)		(BNX2X_NUM_QUEUES(bp) > 1)
 
-#define BNX2X_MAX_QUEUES(bp)	(bp->igu_sb_cnt - CNIC_CONTEXT_USE)
+#define BNX2X_MAX_QUEUES(bp)	BNX2X_MAX_RSS_COUNT(bp)
+/* #define is_eth_multi(bp)	(BNX2X_NUM_ETH_QUEUES(bp) > 1) */
 
 #define RSS_IPV4_CAP_MASK						\
 	TSTORM_ETH_FUNCTION_COMMON_CONFIG_RSS_IPV4_CAPABILITY
@@ -1465,35 +1519,40 @@ struct bnx2x_func_init_params {
 };
 
 #define for_each_eth_queue(bp, var) \
-	for (var = 0; var < BNX2X_NUM_ETH_QUEUES(bp); var++)
+	for ((var) = 0; (var) < BNX2X_NUM_ETH_QUEUES(bp); (var)++)
 
 #define for_each_nondefault_eth_queue(bp, var) \
-	for (var = 1; var < BNX2X_NUM_ETH_QUEUES(bp); var++)
+	for ((var) = 1; (var) < BNX2X_NUM_ETH_QUEUES(bp); (var)++)
 
 #define for_each_queue(bp, var) \
-	for (var = 0; var < BNX2X_NUM_QUEUES(bp); var++) \
+	for ((var) = 0; (var) < BNX2X_NUM_QUEUES(bp); (var)++) \
 		if (skip_queue(bp, var))	\
 			continue;		\
 		else
 
+/* Skip forwarding FP */
 #define for_each_rx_queue(bp, var) \
-	for (var = 0; var < BNX2X_NUM_QUEUES(bp); var++) \
+	for ((var) = 0; (var) < BNX2X_NUM_QUEUES(bp); (var)++) \
 		if (skip_rx_queue(bp, var))	\
 			continue;		\
 		else
 
+/* Skip OOO FP */
 #define for_each_tx_queue(bp, var) \
-	for (var = 0; var < BNX2X_NUM_QUEUES(bp); var++) \
+	for ((var) = 0; (var) < BNX2X_NUM_QUEUES(bp); (var)++) \
 		if (skip_tx_queue(bp, var))	\
 			continue;		\
 		else
 
 #define for_each_nondefault_queue(bp, var) \
-	for (var = 1; var < BNX2X_NUM_QUEUES(bp); var++) \
+	for ((var) = 1; (var) < BNX2X_NUM_QUEUES(bp); (var)++) \
 		if (skip_queue(bp, var))	\
 			continue;		\
 		else
 
+#define for_each_cos_in_tx_queue(fp, var) \
+	for ((var) = 0; (var) < (fp)->max_cos; (var)++)
+
 /* skip rx queue
  * if FCOE l2 support is disabled and this is the fcoe L2 queue
  */
diff --git a/drivers/net/bnx2x/bnx2x_cmn.c b/drivers/net/bnx2x/bnx2x_cmn.c
index 8763625b09d0..e5fac6244f4a 100644
--- a/drivers/net/bnx2x/bnx2x_cmn.c
+++ b/drivers/net/bnx2x/bnx2x_cmn.c
@@ -47,6 +47,25 @@ static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
 
 	/* Restore the NAPI object as it has been already initialized */
 	fp->napi = orig_napi;
+
+	fp->bp = bp;
+	fp->index = index;
+	if (IS_ETH_FP(fp))
+		fp->max_cos = bp->max_cos;
+	else
+		/* Special queues support only one CoS */
+		fp->max_cos = 1;
+
+	/*
+	 * set the tpa flag for each queue. The tpa flag determines the queue
+	 * minimal size so it must be set prior to queue memory allocation
+	 */
+	fp->disable_tpa = ((bp->flags & TPA_ENABLE_FLAG) == 0);
+
+#ifdef BCM_CNIC
+	/* We don't want TPA on FCoE, FWD and OOO L2 rings */
+	bnx2x_fcoe(bp, disable_tpa) = 1;
+#endif
 }
 
 /**
@@ -77,10 +96,10 @@ int load_count[2][3] = { {0} }; /* per-path: 0-common, 1-port0, 2-port1 */
 /* free skb in the packet ring at pos idx
  * return idx of last bd freed
  */
-static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
+static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata,
 			     u16 idx)
 {
-	struct sw_tx_bd *tx_buf = &fp->tx_buf_ring[idx];
+	struct sw_tx_bd *tx_buf = &txdata->tx_buf_ring[idx];
 	struct eth_tx_start_bd *tx_start_bd;
 	struct eth_tx_bd *tx_data_bd;
 	struct sk_buff *skb = tx_buf->skb;
@@ -91,11 +110,11 @@ static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 	prefetch(&skb->end);
 
 	DP(BNX2X_MSG_FP, "fp[%d]: pkt_idx %d  buff @(%p)->skb %p\n",
-	   fp->index, idx, tx_buf, skb);
+	   txdata->txq_index, idx, tx_buf, skb);
 
 	/* unmap first bd */
 	DP(BNX2X_MSG_OFF, "free bd_idx %d\n", bd_idx);
-	tx_start_bd = &fp->tx_desc_ring[bd_idx].start_bd;
+	tx_start_bd = &txdata->tx_desc_ring[bd_idx].start_bd;
 	dma_unmap_single(&bp->pdev->dev, BD_UNMAP_ADDR(tx_start_bd),
 			 BD_UNMAP_LEN(tx_start_bd), DMA_TO_DEVICE);
 
@@ -126,7 +145,7 @@ static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 	while (nbd > 0) {
 
 		DP(BNX2X_MSG_OFF, "free frag bd_idx %d\n", bd_idx);
-		tx_data_bd = &fp->tx_desc_ring[bd_idx].reg_bd;
+		tx_data_bd = &txdata->tx_desc_ring[bd_idx].reg_bd;
 		dma_unmap_page(&bp->pdev->dev, BD_UNMAP_ADDR(tx_data_bd),
 			       BD_UNMAP_LEN(tx_data_bd), DMA_TO_DEVICE);
 		if (--nbd)
@@ -142,20 +161,19 @@ static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
 	return new_cons;
 }
 
-int bnx2x_tx_int(struct bnx2x_fastpath *fp)
+int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
 {
-	struct bnx2x *bp = fp->bp;
 	struct netdev_queue *txq;
-	u16 hw_cons, sw_cons, bd_cons = fp->tx_bd_cons;
+	u16 hw_cons, sw_cons, bd_cons = txdata->tx_bd_cons;
 
 #ifdef BNX2X_STOP_ON_ERROR
 	if (unlikely(bp->panic))
 		return -1;
 #endif
 
-	txq = netdev_get_tx_queue(bp->dev, fp->index);
-	hw_cons = le16_to_cpu(*fp->tx_cons_sb);
-	sw_cons = fp->tx_pkt_cons;
+	txq = netdev_get_tx_queue(bp->dev, txdata->txq_index);
+	hw_cons = le16_to_cpu(*txdata->tx_cons_sb);
+	sw_cons = txdata->tx_pkt_cons;
 
 	while (sw_cons != hw_cons) {
 		u16 pkt_cons;
@@ -164,14 +182,14 @@ int bnx2x_tx_int(struct bnx2x_fastpath *fp)
 
 		DP(NETIF_MSG_TX_DONE, "queue[%d]: hw_cons %u  sw_cons %u "
 				      " pkt_cons %u\n",
-		   fp->index, hw_cons, sw_cons, pkt_cons);
+		   txdata->txq_index, hw_cons, sw_cons, pkt_cons);
 
-		bd_cons = bnx2x_free_tx_pkt(bp, fp, pkt_cons);
+		bd_cons = bnx2x_free_tx_pkt(bp, txdata, pkt_cons);
 		sw_cons++;
 	}
 
-	fp->tx_pkt_cons = sw_cons;
-	fp->tx_bd_cons = bd_cons;
+	txdata->tx_pkt_cons = sw_cons;
+	txdata->tx_bd_cons = bd_cons;
 
 	/* Need to make the tx_bd_cons update visible to start_xmit()
 	 * before checking for netif_tx_queue_stopped().  Without the
@@ -199,7 +217,7 @@ int bnx2x_tx_int(struct bnx2x_fastpath *fp)
 
 		if ((netif_tx_queue_stopped(txq)) &&
 		    (bp->state == BNX2X_STATE_OPEN) &&
-		    (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
+		    (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3))
 			netif_tx_wake_queue(txq);
 
 		__netif_tx_unlock(txq);
@@ -777,6 +795,7 @@ static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
 {
 	struct bnx2x_fastpath *fp = fp_cookie;
 	struct bnx2x *bp = fp->bp;
+	u8 cos;
 
 	DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB "
 			 "[fp %d fw_sd %d igusb %d]\n",
@@ -790,7 +809,10 @@ static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
 
 	/* Handle Rx and Tx according to MSI-X vector */
 	prefetch(fp->rx_cons_sb);
-	prefetch(fp->tx_cons_sb);
+
+	for_each_cos_in_tx_queue(fp, cos)
+		prefetch(fp->txdata[cos].tx_cons_sb);
+
 	prefetch(&fp->sb_running_index[SM_RX_ID]);
 	napi_schedule(&bnx2x_fp(bp, fp->index, napi));
 
@@ -1060,17 +1082,22 @@ void bnx2x_init_rx_rings(struct bnx2x *bp)
 static void bnx2x_free_tx_skbs(struct bnx2x *bp)
 {
 	int i;
+	u8 cos;
 
 	for_each_tx_queue(bp, i) {
 		struct bnx2x_fastpath *fp = &bp->fp[i];
+		for_each_cos_in_tx_queue(fp, cos) {
+			struct bnx2x_fp_txdata *txdata = &fp->txdata[cos];
 
-		u16 bd_cons = fp->tx_bd_cons;
-		u16 sw_prod = fp->tx_pkt_prod;
-		u16 sw_cons = fp->tx_pkt_cons;
+			u16 bd_cons = txdata->tx_bd_cons;
+			u16 sw_prod = txdata->tx_pkt_prod;
+			u16 sw_cons = txdata->tx_pkt_cons;
 
-		while (sw_cons != sw_prod) {
-			bd_cons = bnx2x_free_tx_pkt(bp, fp, TX_BD(sw_cons));
-			sw_cons++;
+			while (sw_cons != sw_prod) {
+				bd_cons = bnx2x_free_tx_pkt(bp, txdata,
+							    TX_BD(sw_cons));
+				sw_cons++;
+			}
 		}
 	}
 }
@@ -1174,7 +1201,7 @@ void bnx2x_free_irq(struct bnx2x *bp)
 {
 	if (bp->flags & USING_MSIX_FLAG)
 		bnx2x_free_msix_irqs(bp, BNX2X_NUM_ETH_QUEUES(bp) +
-				     CNIC_CONTEXT_USE + 1);
+				     CNIC_PRESENT + 1);
 	else if (bp->flags & USING_MSI_FLAG)
 		free_irq(bp->pdev->irq, bp->dev);
 	else
@@ -1196,6 +1223,7 @@ int bnx2x_enable_msix(struct bnx2x *bp)
 	   bp->msix_table[msix_vec].entry, bp->msix_table[msix_vec].entry);
 	msix_vec++;
 #endif
+	/* We need separate vectors for ETH queues only (not FCoE) */
 	for_each_eth_queue(bp, i) {
 		bp->msix_table[msix_vec].entry = msix_vec;
 		DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
@@ -1203,7 +1231,7 @@ int bnx2x_enable_msix(struct bnx2x *bp)
 		msix_vec++;
 	}
 
-	req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_CONTEXT_USE + 1;
+	req_cnt = BNX2X_NUM_ETH_QUEUES(bp) + CNIC_PRESENT + 1;
 
 	rc = pci_enable_msix(bp->pdev, &bp->msix_table[0], req_cnt);
 
@@ -1278,7 +1306,7 @@ static int bnx2x_req_msix_irqs(struct bnx2x *bp)
 	}
 
 	i = BNX2X_NUM_ETH_QUEUES(bp);
-	offset = 1 + CNIC_CONTEXT_USE;
+	offset = 1 + CNIC_PRESENT;
 	netdev_info(bp->dev, "using MSI-X  IRQs: sp %d  fp[%d] %d"
 	       " ... fp[%d] %d\n",
 	       bp->msix_table[0].vector,
@@ -1393,13 +1421,12 @@ u16 bnx2x_select_queue(struct net_device *dev, struct sk_buff *skb)
 
 		/* If ethertype is FCoE or FIP - use FCoE ring */
 		if ((ether_type == ETH_P_FCOE) || (ether_type == ETH_P_FIP))
-			return bnx2x_fcoe(bp, index);
+			return bnx2x_fcoe_tx(bp, txq_index);
 	}
 #endif
 	/* Select a none-FCoE queue:  if FCoE is enabled, exclude FCoE L2 ring
 	 */
-	return __skb_tx_hash(dev, skb,
-			dev->real_num_tx_queues - FCOE_CONTEXT_USE);
+	return __skb_tx_hash(dev, skb, BNX2X_NUM_ETH_QUEUES(bp));
 }
 
 void bnx2x_set_num_queues(struct bnx2x *bp)
@@ -1418,20 +1445,38 @@ void bnx2x_set_num_queues(struct bnx2x *bp)
 	}
 
 	/* Add special queues */
-	bp->num_queues += NONE_ETH_CONTEXT_USE;
+	bp->num_queues += NON_ETH_CONTEXT_USE;
 }
 
 static inline int bnx2x_set_real_num_queues(struct bnx2x *bp)
 {
-	int rc, num = bp->num_queues;
+	int rc, tx, rx;
 
-#ifdef BCM_CNIC
-	if (NO_FCOE(bp))
-		num -= FCOE_CONTEXT_USE;
+	tx = MAX_TXQS_PER_COS * bp->max_cos;
+	rx = BNX2X_NUM_ETH_QUEUES(bp);
 
+/* account for fcoe queue */
+#ifdef BCM_CNIC
+	if (!NO_FCOE(bp)) {
+		rx += FCOE_PRESENT;
+		tx += FCOE_PRESENT;
+	}
 #endif
-	netif_set_real_num_tx_queues(bp->dev, num);
-	rc = netif_set_real_num_rx_queues(bp->dev, num);
+
+	rc = netif_set_real_num_tx_queues(bp->dev, tx);
+	if (rc) {
+		BNX2X_ERR("Failed to set real number of Tx queues: %d\n", rc);
+		return rc;
+	}
+	rc = netif_set_real_num_rx_queues(bp->dev, rx);
+	if (rc) {
+		BNX2X_ERR("Failed to set real number of Rx queues: %d\n", rc);
+		return rc;
+	}
+
+	DP(NETIF_MSG_DRV, "Setting real num queues to (tx, rx) (%d, %d)\n",
+			  tx, rx);
+
 	return rc;
 }
 
@@ -1661,28 +1706,18 @@ int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
 	/* must be called before memory allocation and HW init */
 	bnx2x_ilt_set_info(bp);
 
-	/* zero fastpath structures preserving invariants like napi which are
-	 * allocated only once
+	/*
+	 * Zero fastpath structures preserving invariants like napi, which are
+	 * allocated only once, fp index, max_cos, bp pointer.
+	 * Also set fp->disable_tpa.
 	 */
 	for_each_queue(bp, i)
 		bnx2x_bz_fp(bp, i);
 
+
 	/* Set the receive queues buffer size */
 	bnx2x_set_rx_buf_size(bp);
 
-	/*
-	 * set the tpa flag for each queue. The tpa flag determines the queue
-	 * minimal size so it must be set prior to queue memory allocation
-	 */
-	for_each_queue(bp, i)
-		bnx2x_fp(bp, i, disable_tpa) =
-					((bp->flags & TPA_ENABLE_FLAG) == 0);
-
-#ifdef BCM_CNIC
-	/* We don't want TPA on FCoE L2 ring */
-	bnx2x_fcoe(bp, disable_tpa) = 1;
-#endif
-
 	if (bnx2x_alloc_mem(bp))
 		return -ENOMEM;
 
@@ -1696,6 +1731,12 @@ int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
 		LOAD_ERROR_EXIT(bp, load_error0);
 	}
 
+	/* configure multi cos mappings in kernel.
+	 * this configuration may be overriden by a multi class queue discipline
+	 * or by a dcbx negotiation result.
+	 */
+	bnx2x_setup_tc(bp->dev, bp->max_cos);
+
 	bnx2x_napi_enable(bp);
 
 	/* Send LOAD_REQUEST command to MCP
@@ -1747,6 +1788,7 @@ int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
 		queue_delayed_work(bnx2x_wq, &bp->period_task, 0);
 	} else
 		bp->port.pmf = 0;
+
 	DP(NETIF_MSG_LINK, "pmf %d\n", bp->port.pmf);
 
 	/* Init Function state controlling object */
@@ -2089,6 +2131,7 @@ int bnx2x_set_power_state(struct bnx2x *bp, pci_power_t state)
 int bnx2x_poll(struct napi_struct *napi, int budget)
 {
 	int work_done = 0;
+	u8 cos;
 	struct bnx2x_fastpath *fp = container_of(napi, struct bnx2x_fastpath,
 						 napi);
 	struct bnx2x *bp = fp->bp;
@@ -2101,8 +2144,10 @@ int bnx2x_poll(struct napi_struct *napi, int budget)
 		}
 #endif
 
-		if (bnx2x_has_tx_work(fp))
-			bnx2x_tx_int(fp);
+		for_each_cos_in_tx_queue(fp, cos)
+			if (bnx2x_tx_queue_has_work(&fp->txdata[cos]))
+				bnx2x_tx_int(bp, &fp->txdata[cos]);
+
 
 		if (bnx2x_has_rx_work(fp)) {
 			work_done += bnx2x_rx_int(fp, budget - work_done);
@@ -2164,7 +2209,7 @@ int bnx2x_poll(struct napi_struct *napi, int budget)
  * in Other Operating Systems(TM)
  */
 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
-				   struct bnx2x_fastpath *fp,
+				   struct bnx2x_fp_txdata *txdata,
 				   struct sw_tx_bd *tx_buf,
 				   struct eth_tx_start_bd **tx_bd, u16 hlen,
 				   u16 bd_prod, int nbd)
@@ -2185,7 +2230,7 @@ static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
 	/* now get a new data BD
 	 * (after the pbd) and fill it */
 	bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
-	d_tx_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
+	d_tx_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
 
 	mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
 			   le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
@@ -2481,8 +2526,10 @@ static inline u8 bnx2x_set_pbd_csum(struct bnx2x *bp, struct sk_buff *skb,
 netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct bnx2x *bp = netdev_priv(dev);
+
 	struct bnx2x_fastpath *fp;
 	struct netdev_queue *txq;
+	struct bnx2x_fp_txdata *txdata;
 	struct sw_tx_bd *tx_buf;
 	struct eth_tx_start_bd *tx_start_bd, *first_bd;
 	struct eth_tx_bd *tx_data_bd, *total_pkt_bd = NULL;
@@ -2490,7 +2537,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct eth_tx_parse_bd_e2 *pbd_e2 = NULL;
 	u32 pbd_e2_parsing_data = 0;
 	u16 pkt_prod, bd_prod;
-	int nbd, fp_index;
+	int nbd, txq_index, fp_index, txdata_index;
 	dma_addr_t mapping;
 	u32 xmit_type = bnx2x_xmit_type(bp, skb);
 	int i;
@@ -2504,12 +2551,43 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		return NETDEV_TX_BUSY;
 #endif
 
-	fp_index = skb_get_queue_mapping(skb);
-	txq = netdev_get_tx_queue(dev, fp_index);
+	txq_index = skb_get_queue_mapping(skb);
+	txq = netdev_get_tx_queue(dev, txq_index);
+
+	BUG_ON(txq_index >= MAX_ETH_TXQ_IDX(bp) + FCOE_PRESENT);
+
+	/* decode the fastpath index and the cos index from the txq */
+	fp_index = TXQ_TO_FP(txq_index);
+	txdata_index = TXQ_TO_COS(txq_index);
+
+#ifdef BCM_CNIC
+	/*
+	 * Override the above for the FCoE queue:
+	 *   - FCoE fp entry is right after the ETH entries.
+	 *   - FCoE L2 queue uses bp->txdata[0] only.
+	 */
+	if (unlikely(!NO_FCOE(bp) && (txq_index ==
+				      bnx2x_fcoe_tx(bp, txq_index)))) {
+		fp_index = FCOE_IDX;
+		txdata_index = 0;
+	}
+#endif
+
+	/* enable this debug print to view the transmission queue being used
+	DP(BNX2X_MSG_FP, "indices: txq %d, fp %d, txdata %d",
+	   txq_index, fp_index, txdata_index); */
 
+	/* locate the fastpath and the txdata */
 	fp = &bp->fp[fp_index];
+	txdata = &fp->txdata[txdata_index];
+
+	/* enable this debug print to view the tranmission details
+	DP(BNX2X_MSG_FP,"transmitting packet cid %d fp index %d txdata_index %d"
+			" tx_data ptr %p fp pointer %p",
+	   txdata->cid, fp_index, txdata_index, txdata, fp); */
 
-	if (unlikely(bnx2x_tx_avail(fp) < (skb_shinfo(skb)->nr_frags + 3))) {
+	if (unlikely(bnx2x_tx_avail(bp, txdata) <
+		     (skb_shinfo(skb)->nr_frags + 3))) {
 		fp->eth_q_stats.driver_xoff++;
 		netif_tx_stop_queue(txq);
 		BNX2X_ERR("BUG! Tx ring full when queue awake!\n");
@@ -2518,7 +2596,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	DP(NETIF_MSG_TX_QUEUED, "queue[%d]: SKB: summed %x  protocol %x  "
 				"protocol(%x,%x) gso type %x  xmit_type %x\n",
-	   fp_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
+	   txq_index, skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
 	   ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
 
 	eth = (struct ethhdr *)skb->data;
@@ -2567,15 +2645,15 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	/* get current pkt produced now - advance it just before sending packet
 	 * since mapping of pages may fail and cause packet to be dropped
 	 */
-	pkt_prod = fp->tx_pkt_prod;
-	bd_prod = TX_BD(fp->tx_bd_prod);
+	pkt_prod = txdata->tx_pkt_prod;
+	bd_prod = TX_BD(txdata->tx_bd_prod);
 
 	/* get a tx_buf and first BD
 	 * tx_start_bd may be changed during SPLIT,
 	 * but first_bd will always stay first
 	 */
-	tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
-	tx_start_bd = &fp->tx_desc_ring[bd_prod].start_bd;
+	tx_buf = &txdata->tx_buf_ring[TX_BD(pkt_prod)];
+	tx_start_bd = &txdata->tx_desc_ring[bd_prod].start_bd;
 	first_bd = tx_start_bd;
 
 	tx_start_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_START_BD;
@@ -2586,13 +2664,13 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	SET_FLAG(tx_start_bd->general_data, ETH_TX_START_BD_HDR_NBDS, 1);
 
 	/* remember the first BD of the packet */
-	tx_buf->first_bd = fp->tx_bd_prod;
+	tx_buf->first_bd = txdata->tx_bd_prod;
 	tx_buf->skb = skb;
 	tx_buf->flags = 0;
 
 	DP(NETIF_MSG_TX_QUEUED,
 	   "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
-	   pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_start_bd);
+	   pkt_prod, tx_buf, txdata->tx_pkt_prod, bd_prod, tx_start_bd);
 
 	if (vlan_tx_tag_present(skb)) {
 		tx_start_bd->vlan_or_ethertype =
@@ -2609,7 +2687,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		bnx2x_set_sbd_csum(bp, skb, tx_start_bd, xmit_type);
 
 	if (!CHIP_IS_E1x(bp)) {
-		pbd_e2 = &fp->tx_desc_ring[bd_prod].parse_bd_e2;
+		pbd_e2 = &txdata->tx_desc_ring[bd_prod].parse_bd_e2;
 		memset(pbd_e2, 0, sizeof(struct eth_tx_parse_bd_e2));
 		/* Set PBD in checksum offload case */
 		if (xmit_type & XMIT_CSUM)
@@ -2631,7 +2709,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 					      eth->h_dest);
 		}
 	} else {
-		pbd_e1x = &fp->tx_desc_ring[bd_prod].parse_bd_e1x;
+		pbd_e1x = &txdata->tx_desc_ring[bd_prod].parse_bd_e1x;
 		memset(pbd_e1x, 0, sizeof(struct eth_tx_parse_bd_e1x));
 		/* Set PBD in checksum offload case */
 		if (xmit_type & XMIT_CSUM)
@@ -2663,8 +2741,9 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
 
 		if (unlikely(skb_headlen(skb) > hlen))
-			bd_prod = bnx2x_tx_split(bp, fp, tx_buf, &tx_start_bd,
-						 hlen, bd_prod, ++nbd);
+			bd_prod = bnx2x_tx_split(bp, txdata, tx_buf,
+						 &tx_start_bd, hlen,
+						 bd_prod, ++nbd);
 		if (!CHIP_IS_E1x(bp))
 			bnx2x_set_pbd_gso_e2(skb, &pbd_e2_parsing_data,
 					     xmit_type);
@@ -2698,14 +2777,15 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 			 * before call to bnx2x_free_tx_pkt
 			 */
 			first_bd->nbd = cpu_to_le16(nbd);
-			bnx2x_free_tx_pkt(bp, fp, TX_BD(fp->tx_pkt_prod));
+			bnx2x_free_tx_pkt(bp, txdata,
+					  TX_BD(txdata->tx_pkt_prod));
 			return NETDEV_TX_OK;
 		}
 
 		bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
-		tx_data_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
+		tx_data_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
 		if (total_pkt_bd == NULL)
-			total_pkt_bd = &fp->tx_desc_ring[bd_prod].reg_bd;
+			total_pkt_bd = &txdata->tx_desc_ring[bd_prod].reg_bd;
 
 		tx_data_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
 		tx_data_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
@@ -2759,7 +2839,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		   pbd_e2->parsing_data);
 	DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
 
-	fp->tx_pkt_prod++;
+	txdata->tx_pkt_prod++;
 	/*
 	 * Make sure that the BD data is updated before updating the producer
 	 * since FW might read the BD right after the producer is updated.
@@ -2769,16 +2849,16 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	 */
 	wmb();
 
-	fp->tx_db.data.prod += nbd;
+	txdata->tx_db.data.prod += nbd;
 	barrier();
 
-	DOORBELL(bp, fp->cid, fp->tx_db.raw);
+	DOORBELL(bp, txdata->cid, txdata->tx_db.raw);
 
 	mmiowb();
 
-	fp->tx_bd_prod += nbd;
+	txdata->tx_bd_prod += nbd;
 
-	if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
+	if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 3)) {
 		netif_tx_stop_queue(txq);
 
 		/* paired memory barrier is in bnx2x_tx_int(), we have to keep
@@ -2787,14 +2867,81 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		smp_mb();
 
 		fp->eth_q_stats.driver_xoff++;
-		if (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3)
+		if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3)
 			netif_tx_wake_queue(txq);
 	}
-	fp->tx_pkt++;
+	txdata->tx_pkt++;
 
 	return NETDEV_TX_OK;
 }
 
+/**
+ * bnx2x_setup_tc - routine to configure net_device for multi tc
+ *
+ * @netdev: net device to configure
+ * @tc: number of traffic classes to enable
+ *
+ * callback connected to the ndo_setup_tc function pointer
+ */
+int bnx2x_setup_tc(struct net_device *dev, u8 num_tc)
+{
+	int cos, prio, count, offset;
+	struct bnx2x *bp = netdev_priv(dev);
+
+	/* setup tc must be called under rtnl lock */
+	ASSERT_RTNL();
+
+	/* no traffic classes requested. aborting */
+	if (!num_tc) {
+		netdev_reset_tc(dev);
+		return 0;
+	}
+
+	/* requested to support too many traffic classes */
+	if (num_tc > bp->max_cos) {
+		DP(NETIF_MSG_TX_ERR, "support for too many traffic classes"
+				     " requested: %d. max supported is %d",
+				     num_tc, bp->max_cos);
+		return -EINVAL;
+	}
+
+	/* declare amount of supported traffic classes */
+	if (netdev_set_num_tc(dev, num_tc)) {
+		DP(NETIF_MSG_TX_ERR, "failed to declare %d traffic classes",
+				     num_tc);
+		return -EINVAL;
+	}
+
+	/* configure priority to traffic class mapping */
+	for (prio = 0; prio &