bnx2x: fix memory barriers

Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/bnx2x/bnx2x_main.c b/drivers/net/bnx2x/bnx2x_main.c
index 7a89182..e4dea11 100644
--- a/drivers/net/bnx2x/bnx2x_main.c
+++ b/drivers/net/bnx2x/bnx2x_main.c
@@ -2984,8 +2984,12 @@
 {
 	int func = BP_FUNC(bp);
 
-	/* Make sure that BD data is updated before writing the producer */
-	wmb();
+	/*
+	 * Make sure that BD data is updated before writing the producer:
+	 * BD data is written to the memory, the producer is read from the
+	 * memory, thus we need a full memory barrier to ensure the ordering.
+	 */
+	mb();
 
 	REG_WR16(bp, BAR_XSTRORM_INTMEM + XSTORM_SPQ_PROD_OFFSET(func),
 		 bp->spq_prod_idx);
diff --git a/drivers/net/bnx2x/bnx2x_sp.c b/drivers/net/bnx2x/bnx2x_sp.c
index 5a213e4..cce87d5 100644
--- a/drivers/net/bnx2x/bnx2x_sp.c
+++ b/drivers/net/bnx2x/bnx2x_sp.c
@@ -1563,8 +1563,13 @@
 				idx++;
 		}
 
-		/* Commit the data writes towards the memory */
-		mb();
+		/*
+		 *  No need for an explicit memory barrier here as long we would
+		 *  need to ensure the ordering of writing to the SPQ element
+		 *  and updating of the SPQ producer which involves a memory
+		 *  read and we will have to put a full memory barrier there
+		 *  (inside bnx2x_sp_post()).
+		 */
 
 		rc = bnx2x_sp_post(bp, o->ramrod_cmd, r->cid,
 				   U64_HI(r->rdata_mapping),
@@ -2224,8 +2229,13 @@
 			 data->header.rule_cnt, p->rx_accept_flags,
 			 p->tx_accept_flags);
 
-	/* Commit writes towards the memory before sending a ramrod */
-	mb();
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	/* Send a ramrod */
 	rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_FILTER_RULES, p->cid,
@@ -2918,16 +2928,22 @@
 	if (!o->total_pending_num)
 		bnx2x_mcast_refresh_registry_e2(bp, o);
 
-	/* Commit writes towards the memory before sending a ramrod */
-	mb();
-
-	/* If CLEAR_ONLY was requested - don't send a ramrod and clear
+	/*
+	 * If CLEAR_ONLY was requested - don't send a ramrod and clear
 	 * RAMROD_PENDING status immediately.
 	 */
 	if (test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags)) {
 		raw->clear_pending(raw);
 		return 0;
 	} else {
+		/*
+		 *  No need for an explicit memory barrier here as long we would
+		 *  need to ensure the ordering of writing to the SPQ element
+		 *  and updating of the SPQ producer which involves a memory
+		 *  read and we will have to put a full memory barrier there
+		 *  (inside bnx2x_sp_post()).
+		 */
+
 		/* Send a ramrod */
 		rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_MULTICAST_RULES,
 				   raw->cid, U64_HI(raw->rdata_mapping),
@@ -3404,16 +3420,22 @@
 	if (rc)
 		return rc;
 
-	/* Commit writes towards the memory before sending a ramrod */
-	mb();
-
-	/* If CLEAR_ONLY was requested - don't send a ramrod and clear
+	/*
+	 * If CLEAR_ONLY was requested - don't send a ramrod and clear
 	 * RAMROD_PENDING status immediately.
 	 */
 	if (test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags)) {
 		raw->clear_pending(raw);
 		return 0;
 	} else {
+		/*
+		 *  No need for an explicit memory barrier here as long we would
+		 *  need to ensure the ordering of writing to the SPQ element
+		 *  and updating of the SPQ producer which involves a memory
+		 *  read and we will have to put a full memory barrier there
+		 *  (inside bnx2x_sp_post()).
+		 */
+
 		/* Send a ramrod */
 		rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_SET_MAC, raw->cid,
 				   U64_HI(raw->rdata_mapping),
@@ -4038,8 +4060,13 @@
 		data->capabilities |= ETH_RSS_UPDATE_RAMROD_DATA_UPDATE_RSS_KEY;
 	}
 
-	/* Commit writes towards the memory before sending a ramrod */
-	mb();
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	/* Send a ramrod */
 	rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_RSS_UPDATE, r->cid,
@@ -4505,7 +4532,13 @@
 	/* Fill the ramrod data */
 	bnx2x_q_fill_setup_data_cmn(bp, params, rdata);
 
-	mb();
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	return bnx2x_sp_post(bp, ramrod, o->cids[BNX2X_PRIMARY_CID_INDEX],
 			     U64_HI(data_mapping),
@@ -4528,6 +4561,13 @@
 	bnx2x_q_fill_setup_data_cmn(bp, params, rdata);
 	bnx2x_q_fill_setup_data_e2(bp, params, rdata);
 
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	return bnx2x_sp_post(bp, ramrod, o->cids[BNX2X_PRIMARY_CID_INDEX],
 			     U64_HI(data_mapping),
@@ -4665,7 +4705,13 @@
 	/* Fill the ramrod data */
 	bnx2x_q_fill_update_data(bp, o, update_params, rdata);
 
-	mb();
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_CLIENT_UPDATE,
 			     o->cids[cid_index], U64_HI(data_mapping),
@@ -5484,7 +5530,13 @@
 	rdata->path_id       = BP_PATH(bp);
 	rdata->network_cos_mode = start_params->network_cos_mode;
 
-	mb();
+	/*
+	 *  No need for an explicit memory barrier here as long we would
+	 *  need to ensure the ordering of writing to the SPQ element
+	 *  and updating of the SPQ producer which involves a memory
+	 *  read and we will have to put a full memory barrier there
+	 *  (inside bnx2x_sp_post()).
+	 */
 
 	return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_FUNCTION_START, 0,
 			     U64_HI(data_mapping),