[PATCH 0/8] bnx2x: fixes

@@ -315,6 +315,14 @@ union db_prod {
 	u32		raw;
 };
 
+/* dropless fc FW/HW related params */
+#define BRB_SIZE(bp)		(CHIP_IS_E3(bp) ? 1024 : 512)
+#define MAX_AGG_QS(bp)		(CHIP_IS_E1(bp) ? \
+					ETH_MAX_AGGREGATION_QUEUES_E1 :\
+					ETH_MAX_AGGREGATION_QUEUES_E1H_E2)
+#define FW_DROP_LEVEL(bp)	(3 + MAX_SPQ_PENDING + MAX_AGG_QS(bp))
+#define FW_PREFETCH_CNT		16
+#define DROPLESS_FC_HEADROOM	100
 
 /* MC hsi */
 #define BCM_PAGE_SHIFT		12

@@ -331,15 +339,35 @@ union db_prod {
 /* SGE ring related macros */
 #define NUM_RX_SGE_PAGES	2
 #define RX_SGE_CNT		(BCM_PAGE_SIZE / sizeof(struct eth_rx_sge))
-#define MAX_RX_SGE_CNT		(RX_SGE_CNT - 2)
+#define NEXT_PAGE_SGE_DESC_CNT	2
+#define MAX_RX_SGE_CNT		(RX_SGE_CNT - NEXT_PAGE_SGE_DESC_CNT)
 /* RX_SGE_CNT is promised to be a power of 2 */
 #define RX_SGE_MASK		(RX_SGE_CNT - 1)
 #define NUM_RX_SGE		(RX_SGE_CNT * NUM_RX_SGE_PAGES)
 #define MAX_RX_SGE		(NUM_RX_SGE - 1)
 #define NEXT_SGE_IDX(x)		((((x) & RX_SGE_MASK) == \
-				  (MAX_RX_SGE_CNT - 1)) ? (x) + 3 : (x) + 1)
+				  (MAX_RX_SGE_CNT - 1)) ? \
+					(x) + 1 + NEXT_PAGE_SGE_DESC_CNT : \
+					(x) + 1)
 #define RX_SGE(x)		((x) & MAX_RX_SGE)
 
+/*
+ * Number of required  SGEs is the sum of two:
+ * 1. Number of possible opened aggregations (next packet for
+ *    these aggregations will probably consume SGE immidiatelly)
+ * 2. Rest of BRB blocks divided by 2 (block will consume new SGE only
+ *    after placement on BD for new TPA aggregation)
+ *
+ * Takes into account NEXT_PAGE_SGE_DESC_CNT "next" elements on each page
+ */
+#define NUM_SGE_REQ		(MAX_AGG_QS(bp) + \
+					(BRB_SIZE(bp) - MAX_AGG_QS(bp)) / 2)
+#define NUM_SGE_PG_REQ		((NUM_SGE_REQ + MAX_RX_SGE_CNT - 1) / \
+						MAX_RX_SGE_CNT)
+#define SGE_TH_LO(bp)		(NUM_SGE_REQ + \
+				 NUM_SGE_PG_REQ * NEXT_PAGE_SGE_DESC_CNT)
+#define SGE_TH_HI(bp)		(SGE_TH_LO(bp) + DROPLESS_FC_HEADROOM)
+
 /* Manipulate a bit vector defined as an array of u64 */
 
 /* Number of bits in one sge_mask array element */

@@ -551,24 +579,43 @@ struct bnx2x_fastpath {
 
 #define NUM_TX_RINGS		16
 #define TX_DESC_CNT		(BCM_PAGE_SIZE / sizeof(union eth_tx_bd_types))
-#define MAX_TX_DESC_CNT		(TX_DESC_CNT - 1)
+#define NEXT_PAGE_TX_DESC_CNT	1
+#define MAX_TX_DESC_CNT		(TX_DESC_CNT - NEXT_PAGE_TX_DESC_CNT)
 #define NUM_TX_BD		(TX_DESC_CNT * NUM_TX_RINGS)
 #define MAX_TX_BD		(NUM_TX_BD - 1)
 #define MAX_TX_AVAIL		(MAX_TX_DESC_CNT * NUM_TX_RINGS - 2)
 #define NEXT_TX_IDX(x)		((((x) & MAX_TX_DESC_CNT) == \
-				  (MAX_TX_DESC_CNT - 1)) ? (x) + 2 : (x) + 1)
+				  (MAX_TX_DESC_CNT - 1)) ? \
+					(x) + 1 + NEXT_PAGE_TX_DESC_CNT : \
+					(x) + 1)
 #define TX_BD(x)		((x) & MAX_TX_BD)
 #define TX_BD_POFF(x)		((x) & MAX_TX_DESC_CNT)
 
 /* The RX BD ring is special, each bd is 8 bytes but the last one is 16 */
 #define NUM_RX_RINGS		8
 #define RX_DESC_CNT		(BCM_PAGE_SIZE / sizeof(struct eth_rx_bd))
-#define MAX_RX_DESC_CNT		(RX_DESC_CNT - 2)
+#define NEXT_PAGE_RX_DESC_CNT	2
+#define MAX_RX_DESC_CNT		(RX_DESC_CNT - NEXT_PAGE_RX_DESC_CNT)
 #define RX_DESC_MASK		(RX_DESC_CNT - 1)
 #define NUM_RX_BD		(RX_DESC_CNT * NUM_RX_RINGS)
 #define MAX_RX_BD		(NUM_RX_BD - 1)
 #define MAX_RX_AVAIL		(MAX_RX_DESC_CNT * NUM_RX_RINGS - 2)
-#define MIN_RX_AVAIL		128
+
+/* dropless fc calculations for BDs
+ *
+ * Number of BDs should as number of buffers in BRB:
+ * Low threshold takes into account NEXT_PAGE_RX_DESC_CNT
+ * "next" elements on each page
+ */
+#define NUM_BD_REQ		BRB_SIZE(bp)
+#define NUM_BD_PG_REQ		((NUM_BD_REQ + MAX_RX_DESC_CNT - 1) / \
+					      MAX_RX_DESC_CNT)
+#define BD_TH_LO(bp)		(NUM_BD_REQ + \
+				 NUM_BD_PG_REQ * NEXT_PAGE_RX_DESC_CNT + \
+				 FW_DROP_LEVEL(bp))
+#define BD_TH_HI(bp)		(BD_TH_LO(bp) + DROPLESS_FC_HEADROOM)
+
+#define MIN_RX_AVAIL		((bp)->dropless_fc ? BD_TH_HI(bp) + 128 : 128)
 
 #define MIN_RX_SIZE_TPA_HW	(CHIP_IS_E1(bp) ? \
 					ETH_MIN_RX_CQES_WITH_TPA_E1 : \

@@ -579,7 +626,9 @@ struct bnx2x_fastpath {
 								MIN_RX_AVAIL))
 
 #define NEXT_RX_IDX(x)		((((x) & RX_DESC_MASK) == \
-				  (MAX_RX_DESC_CNT - 1)) ? (x) + 3 : (x) + 1)
+				  (MAX_RX_DESC_CNT - 1)) ? \
+					(x) + 1 + NEXT_PAGE_RX_DESC_CNT : \
+					(x) + 1)
 #define RX_BD(x)		((x) & MAX_RX_BD)
 
 /*

@@ -589,14 +638,31 @@ struct bnx2x_fastpath {
 #define CQE_BD_REL	(sizeof(union eth_rx_cqe) / sizeof(struct eth_rx_bd))
 #define NUM_RCQ_RINGS		(NUM_RX_RINGS * CQE_BD_REL)
 #define RCQ_DESC_CNT		(BCM_PAGE_SIZE / sizeof(union eth_rx_cqe))
-#define MAX_RCQ_DESC_CNT	(RCQ_DESC_CNT - 1)
+#define NEXT_PAGE_RCQ_DESC_CNT	1
+#define MAX_RCQ_DESC_CNT	(RCQ_DESC_CNT - NEXT_PAGE_RCQ_DESC_CNT)
 #define NUM_RCQ_BD		(RCQ_DESC_CNT * NUM_RCQ_RINGS)
 #define MAX_RCQ_BD		(NUM_RCQ_BD - 1)
 #define MAX_RCQ_AVAIL		(MAX_RCQ_DESC_CNT * NUM_RCQ_RINGS - 2)
 #define NEXT_RCQ_IDX(x)		((((x) & MAX_RCQ_DESC_CNT) == \
-				  (MAX_RCQ_DESC_CNT - 1)) ? (x) + 2 : (x) + 1)
+				  (MAX_RCQ_DESC_CNT - 1)) ? \
+					(x) + 1 + NEXT_PAGE_RCQ_DESC_CNT : \
+					(x) + 1)
 #define RCQ_BD(x)		((x) & MAX_RCQ_BD)
 
+/* dropless fc calculations for RCQs
+ *
+ * Number of RCQs should be as number of buffers in BRB:
+ * Low threshold takes into account NEXT_PAGE_RCQ_DESC_CNT
+ * "next" elements on each page
+ */
+#define NUM_RCQ_REQ		BRB_SIZE(bp)
+#define NUM_RCQ_PG_REQ		((NUM_BD_REQ + MAX_RCQ_DESC_CNT - 1) / \
+					      MAX_RCQ_DESC_CNT)
+#define RCQ_TH_LO(bp)		(NUM_RCQ_REQ + \
+				 NUM_RCQ_PG_REQ * NEXT_PAGE_RCQ_DESC_CNT + \
+				 FW_DROP_LEVEL(bp))
+#define RCQ_TH_HI(bp)		(RCQ_TH_LO(bp) + DROPLESS_FC_HEADROOM)
+
 
 /* This is needed for determining of last_max */
 #define SUB_S16(a, b)		(s16)((s16)(a) - (s16)(b))

@@ -987,8 +987,6 @@ void __bnx2x_link_report(struct bnx2x *bp)
 void bnx2x_init_rx_rings(struct bnx2x *bp)
 {
 	int func = BP_FUNC(bp);
-	int max_agg_queues = CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
-					      ETH_MAX_AGGREGATION_QUEUES_E1H_E2;
 	u16 ring_prod;
 	int i, j;
 

@@ -1001,7 +999,7 @@ void bnx2x_init_rx_rings(struct bnx2x *bp)
 
 		if (!fp->disable_tpa) {
 			/* Fill the per-aggregtion pool */
-			for (i = 0; i < max_agg_queues; i++) {
+			for (i = 0; i < MAX_AGG_QS(bp); i++) {
 				struct bnx2x_agg_info *tpa_info =
 					&fp->tpa_info[i];
 				struct sw_rx_bd *first_buf =

@@ -1041,7 +1039,7 @@ void bnx2x_init_rx_rings(struct bnx2x *bp)
 					bnx2x_free_rx_sge_range(bp, fp,
 								ring_prod);
 					bnx2x_free_tpa_pool(bp, fp,
-							    max_agg_queues);
+							    MAX_AGG_QS(bp));
 					fp->disable_tpa = 1;
 					ring_prod = 0;
 					break;

@@ -1137,9 +1135,7 @@ static void bnx2x_free_rx_skbs(struct bnx2x *bp)
 		bnx2x_free_rx_bds(fp);
 
 		if (!fp->disable_tpa)
-			bnx2x_free_tpa_pool(bp, fp, CHIP_IS_E1(bp) ?
-					    ETH_MAX_AGGREGATION_QUEUES_E1 :
-					    ETH_MAX_AGGREGATION_QUEUES_E1H_E2);
+			bnx2x_free_tpa_pool(bp, fp, MAX_AGG_QS(bp));
 	}
 }
 

@@ -2756,8 +2756,14 @@ static void bnx2x_pf_rx_q_prep(struct bnx2x *bp,
 	u16 tpa_agg_size = 0;
 
 	if (!fp->disable_tpa) {
-		pause->sge_th_hi = 250;
-		pause->sge_th_lo = 150;
+		pause->sge_th_lo = SGE_TH_LO(bp);
+		pause->sge_th_hi = SGE_TH_HI(bp);
+
+		/* validate SGE ring has enough to cross high threshold */
+		WARN_ON(bp->dropless_fc &&
+				pause->sge_th_hi + FW_PREFETCH_CNT >
+				MAX_RX_SGE_CNT * NUM_RX_SGE_PAGES);
+
 		tpa_agg_size = min_t(u32,
 			(min_t(u32, 8, MAX_SKB_FRAGS) *
 			SGE_PAGE_SIZE * PAGES_PER_SGE), 0xffff);

@@ -2771,10 +2777,21 @@ static void bnx2x_pf_rx_q_prep(struct bnx2x *bp,
 
 	/* pause - not for e1 */
 	if (!CHIP_IS_E1(bp)) {
-		pause->bd_th_hi = 350;
-		pause->bd_th_lo = 250;
-		pause->rcq_th_hi = 350;
-		pause->rcq_th_lo = 250;
+		pause->bd_th_lo = BD_TH_LO(bp);
+		pause->bd_th_hi = BD_TH_HI(bp);
+
+		pause->rcq_th_lo = RCQ_TH_LO(bp);
+		pause->rcq_th_hi = RCQ_TH_HI(bp);
+		/*
+		 * validate that rings have enough entries to cross
+		 * high thresholds
+		 */
+		WARN_ON(bp->dropless_fc &&
+				pause->bd_th_hi + FW_PREFETCH_CNT >
+				bp->rx_ring_size);
+		WARN_ON(bp->dropless_fc &&
+				pause->rcq_th_hi + FW_PREFETCH_CNT >
+				NUM_RCQ_RINGS * MAX_RCQ_DESC_CNT);
 
 		pause->pri_map = 1;
 	}

@@ -2802,9 +2819,7 @@ static void bnx2x_pf_rx_q_prep(struct bnx2x *bp,
 	 * For PF Clients it should be the maximum avaliable number.
 	 * VF driver(s) may want to define it to a smaller value.
 	 */
-	rxq_init->max_tpa_queues =
-		(CHIP_IS_E1(bp) ? ETH_MAX_AGGREGATION_QUEUES_E1 :
-		ETH_MAX_AGGREGATION_QUEUES_E1H_E2);
+	rxq_init->max_tpa_queues = MAX_AGG_QS(bp);
 
 	rxq_init->cache_line_log = BNX2X_RX_ALIGN_SHIFT;
 	rxq_init->fw_sb_id = fp->fw_sb_id;

@@ -710,7 +710,8 @@ static int bnx2x_hw_stats_update(struct bnx2x *bp)
 		break;
 
 	case MAC_TYPE_NONE: /* unreached */
-		BNX2X_ERR("stats updated by DMAE but no MAC active\n");
+		DP(BNX2X_MSG_STATS,
+		   "stats updated by DMAE but no MAC active\n");
 		return -1;
 
 	default: /* unreached */

@@ -3091,15 +3091,20 @@ static int bnx2x_alloc_fp_mem_at(struct bnx2x *bp, int index)
 	struct bnx2x_fastpath *fp = &bp->fp[index];
 	int ring_size = 0;
 	u8 cos;
+	int rx_ring_size = 0;
 
 	/* if rx_ring_size specified - use it */
-	int rx_ring_size = bp->rx_ring_size ? bp->rx_ring_size :
-			   MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
+	if (!bp->rx_ring_size) {
 
-	/* allocate at least number of buffers required by FW */
-	rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
-						    MIN_RX_SIZE_TPA,
-				  rx_ring_size);
+		rx_ring_size = MAX_RX_AVAIL/BNX2X_NUM_RX_QUEUES(bp);
+
+		/* allocate at least number of buffers required by FW */
+		rx_ring_size = max_t(int, bp->disable_tpa ? MIN_RX_SIZE_NONTPA :
+				     MIN_RX_SIZE_TPA, rx_ring_size);
+
+		bp->rx_ring_size = rx_ring_size;
+	} else
+		rx_ring_size = bp->rx_ring_size;
 
 	/* Common */
 	sb = &bnx2x_fp(bp, index, status_blk);

@@ -1310,10 +1310,7 @@ static void bnx2x_get_ringparam(struct net_device *dev,
 	if (bp->rx_ring_size)
 		ering->rx_pending = bp->rx_ring_size;
 	else
-		if (bp->state == BNX2X_STATE_OPEN && bp->num_queues)
-			ering->rx_pending = MAX_RX_AVAIL/bp->num_queues;
-		else
-			ering->rx_pending = MAX_RX_AVAIL;
+		ering->rx_pending = MAX_RX_AVAIL;
 
 	ering->rx_mini_pending = 0;
 	ering->rx_jumbo_pending = 0;

@@ -9623,13 +9623,6 @@ static int __devinit bnx2x_get_hwinfo(struct bnx2x *bp)
 	/* port info */
 	bnx2x_get_port_hwinfo(bp);
 
-	if (!BP_NOMCP(bp)) {
-		bp->fw_seq =
-			(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
-			 DRV_MSG_SEQ_NUMBER_MASK);
-		BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
-	}
-
 	/* Get MAC addresses */
 	bnx2x_get_mac_hwinfo(bp);
 

@@ -9795,6 +9788,14 @@ static int __devinit bnx2x_init_bp(struct bnx2x *bp)
 	if (!BP_NOMCP(bp))
 		bnx2x_undi_unload(bp);
 
+	/* init fw_seq after undi_unload! */
+	if (!BP_NOMCP(bp)) {
+		bp->fw_seq =
+			(SHMEM_RD(bp, func_mb[BP_FW_MB_IDX(bp)].drv_mb_header) &
+			 DRV_MSG_SEQ_NUMBER_MASK);
+		BNX2X_DEV_INFO("fw_seq 0x%08x\n", bp->fw_seq);
+	}
+
 	if (CHIP_REV_IS_FPGA(bp))
 		dev_err(&bp->pdev->dev, "FPGA detected\n");
 

@@ -10290,17 +10290,21 @@ static int __devinit bnx2x_init_dev(struct pci_dev *pdev,
 	/* clean indirect addresses */
 	pci_write_config_dword(bp->pdev, PCICFG_GRC_ADDRESS,
 			       PCICFG_VENDOR_ID_OFFSET);
-	/* Clean the following indirect addresses for all functions since it
+	/*
+	 * Clean the following indirect addresses for all functions since it
 	 * is not used by the driver.
 	 */
 	REG_WR(bp, PXP2_REG_PGL_ADDR_88_F0, 0);
 	REG_WR(bp, PXP2_REG_PGL_ADDR_8C_F0, 0);
 	REG_WR(bp, PXP2_REG_PGL_ADDR_90_F0, 0);
 	REG_WR(bp, PXP2_REG_PGL_ADDR_94_F0, 0);
-	REG_WR(bp, PXP2_REG_PGL_ADDR_88_F1, 0);
-	REG_WR(bp, PXP2_REG_PGL_ADDR_8C_F1, 0);
-	REG_WR(bp, PXP2_REG_PGL_ADDR_90_F1, 0);
-	REG_WR(bp, PXP2_REG_PGL_ADDR_94_F1, 0);
+
+	if (CHIP_IS_E1x(bp)) {
+		REG_WR(bp, PXP2_REG_PGL_ADDR_88_F1, 0);
+		REG_WR(bp, PXP2_REG_PGL_ADDR_8C_F1, 0);
+		REG_WR(bp, PXP2_REG_PGL_ADDR_90_F1, 0);
+		REG_WR(bp, PXP2_REG_PGL_ADDR_94_F1, 0);
+	}
 
 	/*
 	 * Enable internal target-read (in case we are probed after PF FLR).

@@ -751,24 +751,17 @@ struct bnx2x_fastpath {
 #define FP_CSB_FUNC_OFF	\
 			offsetof(struct cstorm_status_block_c, func)
 
-#define HC_INDEX_TOE_RX_CQ_CONS		0 /* Formerly Ustorm TOE CQ index */
-					  /* (HC_INDEX_U_TOE_RX_CQ_CONS)  */
-#define HC_INDEX_ETH_RX_CQ_CONS		1 /* Formerly Ustorm ETH CQ index */
-					  /* (HC_INDEX_U_ETH_RX_CQ_CONS)  */
-#define HC_INDEX_ETH_RX_BD_CONS		2 /* Formerly Ustorm ETH BD index */
-					  /* (HC_INDEX_U_ETH_RX_BD_CONS)  */
-
-#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_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 HC_INDEX_ETH_RX_CQ_CONS		1
 
-#define HC_INDEX_ETH_FIRST_TX_CQ_CONS	HC_INDEX_ETH_TX_CQ_CONS_COS0
+#define HC_INDEX_OOO_TX_CQ_CONS		4
+
+#define HC_INDEX_ETH_TX_CQ_CONS_COS0	5
+
+#define HC_INDEX_ETH_TX_CQ_CONS_COS1	6
 
+#define HC_INDEX_ETH_TX_CQ_CONS_COS2	7
+
+#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])

@@ -4828,6 +4828,37 @@ void bnx2x_setup_ndsb_state_machine(struct hc_status_block_sm *hc_sm,
 	hc_sm->time_to_expire = 0xFFFFFFFF;
 }
 
+
+/* allocates state machine ids. */
+static inline
+void bnx2x_map_sb_state_machines(struct hc_index_data *index_data)
+{
+	/* zero out state machine indices */
+	/* rx indices */
+	index_data[HC_INDEX_ETH_RX_CQ_CONS].flags &= ~HC_INDEX_DATA_SM_ID;
+
+	/* tx indices */
+	index_data[HC_INDEX_OOO_TX_CQ_CONS].flags &= ~HC_INDEX_DATA_SM_ID;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS0].flags &= ~HC_INDEX_DATA_SM_ID;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS1].flags &= ~HC_INDEX_DATA_SM_ID;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS2].flags &= ~HC_INDEX_DATA_SM_ID;
+
+	/* map indices */
+	/* rx indices */
+	index_data[HC_INDEX_ETH_RX_CQ_CONS].flags |=
+		SM_RX_ID << HC_INDEX_DATA_SM_ID_SHIFT;
+
+	/* tx indices */
+	index_data[HC_INDEX_OOO_TX_CQ_CONS].flags |=
+		SM_TX_ID << HC_INDEX_DATA_SM_ID_SHIFT;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS0].flags |=
+		SM_TX_ID << HC_INDEX_DATA_SM_ID_SHIFT;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS1].flags |=
+		SM_TX_ID << HC_INDEX_DATA_SM_ID_SHIFT;
+	index_data[HC_INDEX_ETH_TX_CQ_CONS_COS2].flags |=
+		SM_TX_ID << HC_INDEX_DATA_SM_ID_SHIFT;
+}
+
 static void bnx2x_init_sb(struct bnx2x *bp, dma_addr_t mapping, int vfid,
 			  u8 vf_valid, int fw_sb_id, int igu_sb_id)
 {

@@ -4859,6 +4890,7 @@ static void bnx2x_init_sb(struct bnx2x *bp, dma_addr_t mapping, int vfid,
 		hc_sm_p = sb_data_e2.common.state_machine;
 		sb_data_p = (u32 *)&sb_data_e2;
 		data_size = sizeof(struct hc_status_block_data_e2)/sizeof(u32);
+		bnx2x_map_sb_state_machines(sb_data_e2.index_data);
 	} else {
 		memset(&sb_data_e1x, 0,
 		       sizeof(struct hc_status_block_data_e1x));

@@ -4873,6 +4905,7 @@ static void bnx2x_init_sb(struct bnx2x *bp, dma_addr_t mapping, int vfid,
 		hc_sm_p = sb_data_e1x.common.state_machine;
 		sb_data_p = (u32 *)&sb_data_e1x;
 		data_size = sizeof(struct hc_status_block_data_e1x)/sizeof(u32);
+		bnx2x_map_sb_state_machines(sb_data_e1x.index_data);
 	}
 
 	bnx2x_setup_ndsb_state_machine(&hc_sm_p[SM_RX_ID],

@@ -5822,7 +5822,7 @@ static int bnx2x_init_hw_common(struct bnx2x *bp)
 	 * take the UNDI lock to protect undi_unload flow from accessing
 	 * registers while we're resetting the chip
 	 */
-	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
+	bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
 
 	bnx2x_reset_common(bp);
 	REG_WR(bp, GRCBASE_MISC + MISC_REGISTERS_RESET_REG_1_SET, 0xffffffff);

@@ -5834,7 +5834,7 @@ static int bnx2x_init_hw_common(struct bnx2x *bp)
 	}
 	REG_WR(bp, GRCBASE_MISC + MISC_REGISTERS_RESET_REG_2_SET, val);
 
-	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
+	bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
 
 	bnx2x_init_block(bp, BLOCK_MISC, PHASE_COMMON);
 

@@ -8570,10 +8570,12 @@ static void __devinit bnx2x_undi_unload(struct bnx2x *bp)
 	/* Check if there is any driver already loaded */
 	val = REG_RD(bp, MISC_REG_UNPREPARED);
 	if (val == 0x1) {
-		/* Check if it is the UNDI driver
+
+		bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
+		/*
+		 * Check if it is the UNDI driver
 		 * UNDI driver initializes CID offset for normal bell to 0x7
 		 */
-		bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
 		val = REG_RD(bp, DORQ_REG_NORM_CID_OFST);
 		if (val == 0x7) {
 			u32 reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS;

@@ -8611,9 +8613,6 @@ static void __devinit bnx2x_undi_unload(struct bnx2x *bp)
 				bnx2x_fw_command(bp, reset_code, 0);
 			}
 
-			/* now it's safe to release the lock */
-			bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
-
 			bnx2x_undi_int_disable(bp);
 			port = BP_PORT(bp);
 

@@ -8663,8 +8662,10 @@ static void __devinit bnx2x_undi_unload(struct bnx2x *bp)
 			bp->fw_seq =
 			      (SHMEM_RD(bp, func_mb[bp->pf_num].drv_mb_header) &
 				DRV_MSG_SEQ_NUMBER_MASK);
-		} else
-			bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
+		}
+
+		/* now it's safe to release the lock */
+		bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
 	}
 }
 

@@ -9440,6 +9441,10 @@ static int __devinit bnx2x_get_hwinfo(struct bnx2x *bp)
 		bp->igu_base_sb = 0;
 	} else {
 		bp->common.int_block = INT_BLOCK_IGU;
+
+		/* do not allow device reset during IGU info preocessing */
+		bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
+
 		val = REG_RD(bp, IGU_REG_BLOCK_CONFIGURATION);
 
 		if (val & IGU_BLOCK_CONFIGURATION_REG_BACKWARD_COMP_EN) {

@@ -9471,6 +9476,7 @@ static int __devinit bnx2x_get_hwinfo(struct bnx2x *bp)
 
 		bnx2x_get_igu_cam_info(bp);
 
+		bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_RESET);
 	}
 
 	/*

@@ -5766,7 +5766,7 @@
 #define HW_LOCK_RESOURCE_RECOVERY_LEADER_0			 8
 #define HW_LOCK_RESOURCE_RECOVERY_LEADER_1			 9
 #define HW_LOCK_RESOURCE_SPIO					 2
-#define HW_LOCK_RESOURCE_UNDI					 5
+#define HW_LOCK_RESOURCE_RESET					 5
 #define AEU_INPUTS_ATTN_BITS_ATC_HW_INTERRUPT			 (0x1<<4)
 #define AEU_INPUTS_ATTN_BITS_ATC_PARITY_ERROR			 (0x1<<5)
 #define AEU_INPUTS_ATTN_BITS_BRB_PARITY_ERROR			 (0x1<<18)

@@ -1166,11 +1166,12 @@ struct bnx2x {
 #define BP_PORT(bp)			(bp->pfid & 1)
 #define BP_FUNC(bp)			(bp->pfid)
 #define BP_ABS_FUNC(bp)			(bp->pf_num)
-#define BP_E1HVN(bp)			(bp->pfid >> 1)
-#define BP_VN(bp)			(BP_E1HVN(bp)) /*remove when approved*/
-#define BP_L_ID(bp)			(BP_E1HVN(bp) << 2)
-#define BP_FW_MB_IDX(bp)		(BP_PORT(bp) +\
-	  BP_VN(bp) * ((CHIP_IS_E1x(bp) || (CHIP_MODE_IS_4_PORT(bp))) ? 2  : 1))
+#define BP_VN(bp)			((bp)->pfid >> 1)
+#define BP_MAX_VN_NUM(bp)		(CHIP_MODE_IS_4_PORT(bp) ? 2 : 4)
+#define BP_L_ID(bp)			(BP_VN(bp) << 2)
+#define BP_FW_MB_IDX_VN(bp, vn)		(BP_PORT(bp) +\
+	  (vn) * ((CHIP_IS_E1x(bp) || (CHIP_MODE_IS_4_PORT(bp))) ? 2  : 1))
+#define BP_FW_MB_IDX(bp)		BP_FW_MB_IDX_VN(bp, BP_VN(bp))
 
 	struct net_device	*dev;
 	struct pci_dev		*pdev;

@@ -1833,7 +1834,7 @@ static inline u32 reg_poll(struct bnx2x *bp, u32 reg, u32 expected, int ms,
 
 #define MAX_DMAE_C_PER_PORT		8
 #define INIT_DMAE_C(bp)			(BP_PORT(bp) * MAX_DMAE_C_PER_PORT + \
-					 BP_E1HVN(bp))
+					 BP_VN(bp))
 #define PMF_DMAE_C(bp)			(BP_PORT(bp) * MAX_DMAE_C_PER_PORT + \
 					 E1HVN_MAX)
 

@@ -1859,7 +1860,7 @@ static inline u32 reg_poll(struct bnx2x *bp, u32 reg, u32 expected, int ms,
 
 /* must be used on a CID before placing it on a HW ring */
 #define HW_CID(bp, x)			((BP_PORT(bp) << 23) | \
-					 (BP_E1HVN(bp) << BNX2X_SWCID_SHIFT) | \
+					 (BP_VN(bp) << BNX2X_SWCID_SHIFT) | \
 					 (x))
 
 #define SP_DESC_CNT		(BCM_PAGE_SIZE / sizeof(struct eth_spe))

@@ -407,8 +407,8 @@ u32 bnx2x_dmae_opcode(struct bnx2x *bp, u8 src_type, u8 dst_type,
 	opcode |= (DMAE_CMD_SRC_RESET | DMAE_CMD_DST_RESET);
 
 	opcode |= (BP_PORT(bp) ? DMAE_CMD_PORT_1 : DMAE_CMD_PORT_0);
-	opcode |= ((BP_E1HVN(bp) << DMAE_CMD_E1HVN_SHIFT) |
-		   (BP_E1HVN(bp) << DMAE_COMMAND_DST_VN_SHIFT));
+	opcode |= ((BP_VN(bp) << DMAE_CMD_E1HVN_SHIFT) |
+		   (BP_VN(bp) << DMAE_COMMAND_DST_VN_SHIFT));
 	opcode |= (DMAE_COM_SET_ERR << DMAE_COMMAND_ERR_POLICY_SHIFT);
 
 #ifdef __BIG_ENDIAN

@@ -1419,7 +1419,7 @@ static void bnx2x_hc_int_enable(struct bnx2x *bp)
 	if (!CHIP_IS_E1(bp)) {
 		/* init leading/trailing edge */
 		if (IS_MF(bp)) {
-			val = (0xee0f | (1 << (BP_E1HVN(bp) + 4)));
+			val = (0xee0f | (1 << (BP_VN(bp) + 4)));
 			if (bp->port.pmf)
 				/* enable nig and gpio3 attention */
 				val |= 0x1100;

@@ -1471,7 +1471,7 @@ static void bnx2x_igu_int_enable(struct bnx2x *bp)
 
 	/* init leading/trailing edge */
 	if (IS_MF(bp)) {
-		val = (0xee0f | (1 << (BP_E1HVN(bp) + 4)));
+		val = (0xee0f | (1 << (BP_VN(bp) + 4)));
 		if (bp->port.pmf)
 			/* enable nig and gpio3 attention */
 			val |= 0x1100;

@@ -2287,7 +2287,7 @@ static void bnx2x_calc_vn_weight_sum(struct bnx2x *bp)
 	int vn;
 
 	bp->vn_weight_sum = 0;
-	for (vn = VN_0; vn < E1HVN_MAX; vn++) {
+	for (vn = VN_0; vn < BP_MAX_VN_NUM(bp); vn++) {
 		u32 vn_cfg = bp->mf_config[vn];
 		u32 vn_min_rate = ((vn_cfg & FUNC_MF_CFG_MIN_BW_MASK) >>
 				   FUNC_MF_CFG_MIN_BW_SHIFT) * 100;

@@ -2320,12 +2320,18 @@ static void bnx2x_calc_vn_weight_sum(struct bnx2x *bp)
 					CMNG_FLAGS_PER_PORT_FAIRNESS_VN;
 }
 
+/* returns func by VN for current port */
+static inline int func_by_vn(struct bnx2x *bp, int vn)
+{
+	return 2 * vn + BP_PORT(bp);
+}
+
 static void bnx2x_init_vn_minmax(struct bnx2x *bp, int vn)
 {
 	struct rate_shaping_vars_per_vn m_rs_vn;
 	struct fairness_vars_per_vn m_fair_vn;
 	u32 vn_cfg = bp->mf_config[vn];
-	int func = 2*vn + BP_PORT(bp);
+	int func = func_by_vn(bp, vn);
 	u16 vn_min_rate, vn_max_rate;
 	int i;
 

@@ -2422,7 +2428,7 @@ void bnx2x_read_mf_cfg(struct bnx2x *bp)
 	 *
 	 *      and there are 2 functions per port
 	 */
-	for (vn = VN_0; vn < E1HVN_MAX; vn++) {
+	for (vn = VN_0; vn < BP_MAX_VN_NUM(bp); vn++) {
 		int /*abs*/func = n * (2 * vn + BP_PORT(bp)) + BP_PATH(bp);
 
 		if (func >= E1H_FUNC_MAX)

@@ -2454,7 +2460,7 @@ static void bnx2x_cmng_fns_init(struct bnx2x *bp, u8 read_cfg, u8 cmng_type)
 
 		/* calculate and set min-max rate for each vn */
 		if (bp->port.pmf)
-			for (vn = VN_0; vn < E1HVN_MAX; vn++)
+			for (vn = VN_0; vn < BP_MAX_VN_NUM(bp); vn++)
 				bnx2x_init_vn_minmax(bp, vn);
 
 		/* always enable rate shaping and fairness */

@@ -2473,16 +2479,15 @@ static void bnx2x_cmng_fns_init(struct bnx2x *bp, u8 read_cfg, u8 cmng_type)
 
 static inline void bnx2x_link_sync_notify(struct bnx2x *bp)
 {
-	int port = BP_PORT(bp);
 	int func;
 	int vn;
 
 	/* Set the attention towards other drivers on the same port */
-	for (vn = VN_0; vn < E1HVN_MAX; vn++) {
-		if (vn == BP_E1HVN(bp))
+	for (vn = VN_0; vn < BP_MAX_VN_NUM(bp); vn++) {
+		if (vn == BP_VN(bp))
 			continue;
 
-		func = ((vn << 1) | port);
+		func = func_by_vn(bp, vn);
 		REG_WR(bp, MISC_REG_AEU_GENERAL_ATTN_0 +
 		       (LINK_SYNC_ATTENTION_BIT_FUNC_0 + func)*4, 1);
 	}

@@ -2577,7 +2582,7 @@ static void bnx2x_pmf_update(struct bnx2x *bp)
 	bnx2x_dcbx_pmf_update(bp);
 
 	/* enable nig attention */
-	val = (0xff0f | (1 << (BP_E1HVN(bp) + 4)));
+	val = (0xff0f | (1 << (BP_VN(bp) + 4)));
 	if (bp->common.int_block == INT_BLOCK_HC) {
 		REG_WR(bp, HC_REG_TRAILING_EDGE_0 + port*8, val);
 		REG_WR(bp, HC_REG_LEADING_EDGE_0 + port*8, val);

@@ -6686,12 +6691,16 @@ static int bnx2x_init_hw_func(struct bnx2x *bp)
 			if (CHIP_MODE_IS_4_PORT(bp))
 				dsb_idx = BP_FUNC(bp);
 			else
-				dsb_idx = BP_E1HVN(bp);
+				dsb_idx = BP_VN(bp);
 
 			prod_offset = (CHIP_INT_MODE_IS_BC(bp) ?
 				       IGU_BC_BASE_DSB_PROD + dsb_idx :
 				       IGU_NORM_BASE_DSB_PROD + dsb_idx);
 
+			/*
+			 * igu prods come in chunks of E1HVN_MAX (4) -
+			 * does not matters what is the current chip mode
+			 */
 			for (i = 0; i < (num_segs * E1HVN_MAX);
 			     i += E1HVN_MAX) {
 				addr = IGU_REG_PROD_CONS_MEMORY +

@@ -7585,7 +7594,7 @@ u32 bnx2x_send_unload_req(struct bnx2x *bp, int unload_mode)
 		u32 val;
 		/* The mac address is written to entries 1-4 to
 		   preserve entry 0 which is used by the PMF */
-		u8 entry = (BP_E1HVN(bp) + 1)*8;
+		u8 entry = (BP_VN(bp) + 1)*8;
 
 		val = (mac_addr[0] << 8) | mac_addr[1];
 		EMAC_WR(bp, EMAC_REG_EMAC_MAC_MATCH + entry, val);

@@ -8792,13 +8801,13 @@ static void __devinit bnx2x_get_common_hwinfo(struct bnx2x *bp)
 static void __devinit bnx2x_get_igu_cam_info(struct bnx2x *bp)
 {
 	int pfid = BP_FUNC(bp);
-	int vn = BP_E1HVN(bp);
 	int igu_sb_id;
 	u32 val;
 	u8 fid, igu_sb_cnt = 0;
 
 	bp->igu_base_sb = 0xff;
 	if (CHIP_INT_MODE_IS_BC(bp)) {
+		int vn = BP_VN(bp);
 		igu_sb_cnt = bp->igu_sb_cnt;
 		bp->igu_base_sb = (CHIP_MODE_IS_4_PORT(bp) ? pfid : vn) *
 			FP_SB_MAX_E1x;

@@ -9488,7 +9497,7 @@ static int __devinit bnx2x_get_hwinfo(struct bnx2x *bp)
 
 	bp->mf_ov = 0;
 	bp->mf_mode = 0;
-	vn = BP_E1HVN(bp);
+	vn = BP_VN(bp);
 
 	if (!CHIP_IS_E1(bp) && !BP_NOMCP(bp)) {
 		BNX2X_DEV_INFO("shmem2base 0x%x, size %d, mfcfg offset %d\n",

@@ -1392,7 +1392,7 @@ static void bnx2x_port_stats_base_init(struct bnx2x *bp)
 
 static void bnx2x_func_stats_base_init(struct bnx2x *bp)
 {
-	int vn, vn_max = IS_MF(bp) ? E1HVN_MAX : E1VN_MAX;
+	int vn, vn_max = IS_MF(bp) ? BP_MAX_VN_NUM(bp) : E1VN_MAX;
 	u32 func_stx;
 
 	/* sanity */

@@ -1405,7 +1405,7 @@ static void bnx2x_func_stats_base_init(struct bnx2x *bp)
 	func_stx = bp->func_stx;
 
 	for (vn = VN_0; vn < vn_max; vn++) {
-		int mb_idx = CHIP_IS_E1x(bp) ? 2*vn + BP_PORT(bp) : vn;
+		int mb_idx = BP_FW_MB_IDX_VN(bp, vn);
 
 		bp->func_stx = SHMEM_RD(bp, func_mb[mb_idx].fw_mb_param);
 		bnx2x_func_stats_init(bp);