i40e/i40evf: enable hardware feature head write back

The hardware supports a feature to avoid updating the descriptor
ring by marking each descriptor with a DD bit, and instead
writes a memory location with an update to where the driver
should clean up to.  Enable this feature.

Change-ID: I5da4e0681f0b581a6401c950a81808792267fe57
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Kavindya Deegala <kavindya.s.deegala@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

drivers/net/ethernet/intel/i40e/i40e_main.c

@@ -2181,6 +2181,11 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
 	tx_ctx.fd_ena = !!(vsi->back->flags & (I40E_FLAG_FD_SB_ENABLED |
 					       I40E_FLAG_FD_ATR_ENABLED));
 	tx_ctx.timesync_ena = !!(vsi->back->flags & I40E_FLAG_PTP);
+	/* FDIR VSI tx ring can still use RS bit and writebacks */
+	if (vsi->type != I40E_VSI_FDIR)
+		tx_ctx.head_wb_ena = 1;
+	tx_ctx.head_wb_addr = ring->dma +
+			      (ring->count * sizeof(struct i40e_tx_desc));
 
 	/* As part of VSI creation/update, FW allocates certain
 	 * Tx arbitration queue sets for each TC enabled for

drivers/net/ethernet/intel/i40e/i40e_txrx.c

@@ -618,6 +618,20 @@ static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
 	return ret;
 }
 
+/**
+ * i40e_get_head - Retrieve head from head writeback
+ * @tx_ring:  tx ring to fetch head of
+ *
+ * Returns value of Tx ring head based on value stored
+ * in head write-back location
+ **/
+static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
+{
+	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
+
+	return le32_to_cpu(*(volatile __le32 *)head);
+}
+
 /**
  * i40e_clean_tx_irq - Reclaim resources after transmit completes
  * @tx_ring:  tx ring to clean
@@ -629,6 +643,7 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 {
 	u16 i = tx_ring->next_to_clean;
 	struct i40e_tx_buffer *tx_buf;
+	struct i40e_tx_desc *tx_head;
 	struct i40e_tx_desc *tx_desc;
 	unsigned int total_packets = 0;
 	unsigned int total_bytes = 0;
@@ -637,6 +652,8 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 	tx_desc = I40E_TX_DESC(tx_ring, i);
 	i -= tx_ring->count;
 
+	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
+
 	do {
 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 
@@ -647,9 +664,8 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 		/* prevent any other reads prior to eop_desc */
 		read_barrier_depends();
 
-		/* if the descriptor isn't done, no work yet to do */
-		if (!(eop_desc->cmd_type_offset_bsz &
-		      cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE)))
+		/* we have caught up to head, no work left to do */
+		if (tx_head == tx_desc)
 			break;
 
 		/* clear next_to_watch to prevent false hangs */
@@ -905,6 +921,10 @@ int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
 
 	/* round up to nearest 4K */
 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
+	/* add u32 for head writeback, align after this takes care of
+	 * guaranteeing this is at least one cache line in size
+	 */
+	tx_ring->size += sizeof(u32);
 	tx_ring->size = ALIGN(tx_ring->size, 4096);
 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 					   &tx_ring->dma, GFP_KERNEL);
@@ -2042,9 +2062,23 @@ static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
 		tx_bi = &tx_ring->tx_bi[i];
 	}
 
+	/* Place RS bit on last descriptor of any packet that spans across the
+	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
+	 */
+#define WB_STRIDE 0x3
+	if (((i & WB_STRIDE) != WB_STRIDE) &&
+	    (first <= &tx_ring->tx_bi[i]) &&
+	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
 		tx_desc->cmd_type_offset_bsz =
 			build_ctob(td_cmd, td_offset, size, td_tag) |
-		cpu_to_le64((u64)I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT);
+			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
+					 I40E_TXD_QW1_CMD_SHIFT);
+	} else {
+		tx_desc->cmd_type_offset_bsz =
+			build_ctob(td_cmd, td_offset, size, td_tag) |
+			cpu_to_le64((u64)I40E_TXD_CMD <<
+					 I40E_TXD_QW1_CMD_SHIFT);
+	}
 
 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
 						 tx_ring->queue_index),

drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c

@@ -230,6 +230,9 @@ static int i40e_config_vsi_tx_queue(struct i40e_vf *vf, u16 vsi_idx,
 	tx_ctx.qlen = info->ring_len;
 	tx_ctx.rdylist = le16_to_cpu(pf->vsi[vsi_idx]->info.qs_handle[0]);
 	tx_ctx.rdylist_act = 0;
+	tx_ctx.head_wb_ena = 1;
+	tx_ctx.head_wb_addr = info->dma_ring_addr +
+			      (info->ring_len * sizeof(struct i40e_tx_desc));
 
 	/* clear the context in the HMC */
 	ret = i40e_clear_lan_tx_queue_context(hw, pf_queue_id);

drivers/net/ethernet/intel/i40evf/i40e_txrx.c

@@ -169,6 +169,20 @@ static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
 	return ret;
 }
 
+/**
+ * i40e_get_head - Retrieve head from head writeback
+ * @tx_ring:  tx ring to fetch head of
+ *
+ * Returns value of Tx ring head based on value stored
+ * in head write-back location
+ **/
+static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
+{
+	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
+
+	return le32_to_cpu(*(volatile __le32 *)head);
+}
+
 /**
  * i40e_clean_tx_irq - Reclaim resources after transmit completes
  * @tx_ring:  tx ring to clean
@@ -180,6 +194,7 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 {
 	u16 i = tx_ring->next_to_clean;
 	struct i40e_tx_buffer *tx_buf;
+	struct i40e_tx_desc *tx_head;
 	struct i40e_tx_desc *tx_desc;
 	unsigned int total_packets = 0;
 	unsigned int total_bytes = 0;
@@ -188,6 +203,8 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 	tx_desc = I40E_TX_DESC(tx_ring, i);
 	i -= tx_ring->count;
 
+	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
+
 	do {
 		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 
@@ -198,9 +215,8 @@ static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
 		/* prevent any other reads prior to eop_desc */
 		read_barrier_depends();
 
-		/* if the descriptor isn't done, no work yet to do */
-		if (!(eop_desc->cmd_type_offset_bsz &
-		      cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE)))
+		/* we have caught up to head, no work left to do */
+		if (tx_head == tx_desc)
 			break;
 
 		/* clear next_to_watch to prevent false hangs */
@@ -432,6 +448,10 @@ int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
 
 	/* round up to nearest 4K */
 	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
+	/* add u32 for head writeback, align after this takes care of
+	 * guaranteeing this is at least one cache line in size
+	 */
+	tx_ring->size += sizeof(u32);
 	tx_ring->size = ALIGN(tx_ring->size, 4096);
 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 					   &tx_ring->dma, GFP_KERNEL);
@@ -1377,9 +1397,23 @@ static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
 		tx_bi = &tx_ring->tx_bi[i];
 	}
 
+	/* Place RS bit on last descriptor of any packet that spans across the
+	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
+	 */
+#define WB_STRIDE 0x3
+	if (((i & WB_STRIDE) != WB_STRIDE) &&
+	    (first <= &tx_ring->tx_bi[i]) &&
+	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
 		tx_desc->cmd_type_offset_bsz =
 			build_ctob(td_cmd, td_offset, size, td_tag) |
-		cpu_to_le64((u64)I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT);
+			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
+					 I40E_TXD_QW1_CMD_SHIFT);
+	} else {
+		tx_desc->cmd_type_offset_bsz =
+			build_ctob(td_cmd, td_offset, size, td_tag) |
+			cpu_to_le64((u64)I40E_TXD_CMD <<
+					 I40E_TXD_QW1_CMD_SHIFT);
+	}
 
 	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
 						 tx_ring->queue_index),