octeon_ep: add Tx/Rx processing and interrupt support

Add support to enable MSI-x and register interrupts.
Add support to process Tx and Rx traffic. Includes processing
Tx completions and Rx refill.
Signed-off-by: Veerasenareddy Burru <vburru@marvell.com>
Signed-off-by: Abhijit Ayarekar <aayarekar@marvell.com>
Signed-off-by: Satananda Burla <sburla@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/marvell/octeon_ep/octep_rx.c

@@ -68,6 +68,50 @@ static int octep_oq_fill_ring_buffers(struct octep_oq *oq)
 	return -1;
 }
 
+/**
+ * octep_oq_refill() - refill buffers for used Rx ring descriptors.
+ *
+ * @oct: Octeon device private data structure.
+ * @oq: Octeon Rx queue data structure.
+ *
+ * Return: number of descriptors successfully refilled with receive buffers.
+ */
+static int octep_oq_refill(struct octep_device *oct, struct octep_oq *oq)
+{
+	struct octep_oq_desc_hw *desc_ring = oq->desc_ring;
+	struct page *page;
+	u32 refill_idx, i;
+
+	refill_idx = oq->host_refill_idx;
+	for (i = 0; i < oq->refill_count; i++) {
+		page = dev_alloc_page();
+		if (unlikely(!page)) {
+			dev_err(oq->dev, "refill: rx buffer alloc failed\n");
+			oq->stats.alloc_failures++;
+			break;
+		}
+
+		desc_ring[refill_idx].buffer_ptr = dma_map_page(oq->dev, page, 0,
+								PAGE_SIZE, DMA_FROM_DEVICE);
+		if (dma_mapping_error(oq->dev, desc_ring[refill_idx].buffer_ptr)) {
+			dev_err(oq->dev,
+				"OQ-%d buffer refill: DMA mapping error!\n",
+				oq->q_no);
+			put_page(page);
+			oq->stats.alloc_failures++;
+			break;
+		}
+		oq->buff_info[refill_idx].page = page;
+		refill_idx++;
+		if (refill_idx == oq->max_count)
+			refill_idx = 0;
+	}
+	oq->host_refill_idx = refill_idx;
+	oq->refill_count -= i;
+
+	return i;
+}
+
 /**
  * octep_setup_oq() - Setup a Rx queue.
  *
@@ -262,3 +306,203 @@ void octep_free_oqs(struct octep_device *oct)
 			"Successfully freed OQ(RxQ)-%d.\n", i);
 	}
 }
+
+/**
+ * octep_oq_check_hw_for_pkts() - Check for new Rx packets.
+ *
+ * @oct: Octeon device private data structure.
+ * @oq: Octeon Rx queue data structure.
+ *
+ * Return: packets received after previous check.
+ */
+static int octep_oq_check_hw_for_pkts(struct octep_device *oct,
+				      struct octep_oq *oq)
+{
+	u32 pkt_count, new_pkts;
+
+	pkt_count = readl(oq->pkts_sent_reg);
+	new_pkts = pkt_count - oq->last_pkt_count;
+
+	/* Clear the hardware packets counter register if the rx queue is
+	 * being processed continuously with-in a single interrupt and
+	 * reached half its max value.
+	 * this counter is not cleared every time read, to save write cycles.
+	 */
+	if (unlikely(pkt_count > 0xF0000000U)) {
+		writel(pkt_count, oq->pkts_sent_reg);
+		pkt_count = readl(oq->pkts_sent_reg);
+		new_pkts += pkt_count;
+	}
+	oq->last_pkt_count = pkt_count;
+	oq->pkts_pending += new_pkts;
+	return new_pkts;
+}
+
+/**
+ * __octep_oq_process_rx() - Process hardware Rx queue and push to stack.
+ *
+ * @oct: Octeon device private data structure.
+ * @oq: Octeon Rx queue data structure.
+ * @pkts_to_process: number of packets to be processed.
+ *
+ * Process the new packets in Rx queue.
+ * Packets larger than single Rx buffer arrive in consecutive descriptors.
+ * But, count returned by the API only accounts full packets, not fragments.
+ *
+ * Return: number of packets processed and pushed to stack.
+ */
+static int __octep_oq_process_rx(struct octep_device *oct,
+				 struct octep_oq *oq, u16 pkts_to_process)
+{
+	struct octep_oq_resp_hw_ext *resp_hw_ext = NULL;
+	struct octep_rx_buffer *buff_info;
+	struct octep_oq_resp_hw *resp_hw;
+	u32 pkt, rx_bytes, desc_used;
+	struct sk_buff *skb;
+	u16 data_offset;
+	u32 read_idx;
+
+	read_idx = oq->host_read_idx;
+	rx_bytes = 0;
+	desc_used = 0;
+	for (pkt = 0; pkt < pkts_to_process; pkt++) {
+		buff_info = (struct octep_rx_buffer *)&oq->buff_info[read_idx];
+		dma_unmap_page(oq->dev, oq->desc_ring[read_idx].buffer_ptr,
+			       PAGE_SIZE, DMA_FROM_DEVICE);
+		resp_hw = page_address(buff_info->page);
+		buff_info->page = NULL;
+
+		/* Swap the length field that is in Big-Endian to CPU */
+		buff_info->len = be64_to_cpu(resp_hw->length);
+		if (oct->caps_enabled & OCTEP_CAP_RX_CHECKSUM) {
+			/* Extended response header is immediately after
+			 * response header (resp_hw)
+			 */
+			resp_hw_ext = (struct octep_oq_resp_hw_ext *)
+				      (resp_hw + 1);
+			buff_info->len -= OCTEP_OQ_RESP_HW_EXT_SIZE;
+			/* Packet Data is immediately after
+			 * extended response header.
+			 */
+			data_offset = OCTEP_OQ_RESP_HW_SIZE +
+				      OCTEP_OQ_RESP_HW_EXT_SIZE;
+		} else {
+			/* Data is immediately after
+			 * Hardware Rx response header.
+			 */
+			data_offset = OCTEP_OQ_RESP_HW_SIZE;
+		}
+		rx_bytes += buff_info->len;
+
+		if (buff_info->len <= oq->max_single_buffer_size) {
+			skb = build_skb((void *)resp_hw, PAGE_SIZE);
+			skb_reserve(skb, data_offset);
+			skb_put(skb, buff_info->len);
+			read_idx++;
+			desc_used++;
+			if (read_idx == oq->max_count)
+				read_idx = 0;
+		} else {
+			struct skb_shared_info *shinfo;
+			u16 data_len;
+
+			skb = build_skb((void *)resp_hw, PAGE_SIZE);
+			skb_reserve(skb, data_offset);
+			/* Head fragment includes response header(s);
+			 * subsequent fragments contains only data.
+			 */
+			skb_put(skb, oq->max_single_buffer_size);
+			read_idx++;
+			desc_used++;
+			if (read_idx == oq->max_count)
+				read_idx = 0;
+
+			shinfo = skb_shinfo(skb);
+			data_len = buff_info->len - oq->max_single_buffer_size;
+			while (data_len) {
+				dma_unmap_page(oq->dev, oq->desc_ring[read_idx].buffer_ptr,
+					       PAGE_SIZE, DMA_FROM_DEVICE);
+				buff_info = (struct octep_rx_buffer *)
+					    &oq->buff_info[read_idx];
+				if (data_len < oq->buffer_size) {
+					buff_info->len = data_len;
+					data_len = 0;
+				} else {
+					buff_info->len = oq->buffer_size;
+					data_len -= oq->buffer_size;
+				}
+
+				skb_add_rx_frag(skb, shinfo->nr_frags,
+						buff_info->page, 0,
+						buff_info->len,
+						buff_info->len);
+				buff_info->page = NULL;
+				read_idx++;
+				desc_used++;
+				if (read_idx == oq->max_count)
+					read_idx = 0;
+			}
+		}
+
+		skb->dev = oq->netdev;
+		skb->protocol =  eth_type_trans(skb, skb->dev);
+		if (resp_hw_ext &&
+		    resp_hw_ext->csum_verified == OCTEP_CSUM_VERIFIED)
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
+		else
+			skb->ip_summed = CHECKSUM_NONE;
+		napi_gro_receive(oq->napi, skb);
+	}
+
+	oq->host_read_idx = read_idx;
+	oq->refill_count += desc_used;
+	oq->stats.packets += pkt;
+	oq->stats.bytes += rx_bytes;
+
+	return pkt;
+}
+
+/**
+ * octep_oq_process_rx() - Process Rx queue.
+ *
+ * @oq: Octeon Rx queue data structure.
+ * @budget: max number of packets can be processed in one invocation.
+ *
+ * Check for newly received packets and process them.
+ * Keeps checking for new packets until budget is used or no new packets seen.
+ *
+ * Return: number of packets processed.
+ */
+int octep_oq_process_rx(struct octep_oq *oq, int budget)
+{
+	u32 pkts_available, pkts_processed, total_pkts_processed;
+	struct octep_device *oct = oq->octep_dev;
+
+	pkts_available = 0;
+	pkts_processed = 0;
+	total_pkts_processed = 0;
+	while (total_pkts_processed < budget) {
+		 /* update pending count only when current one exhausted */
+		if (oq->pkts_pending == 0)
+			octep_oq_check_hw_for_pkts(oct, oq);
+		pkts_available = min(budget - total_pkts_processed,
+				     oq->pkts_pending);
+		if (!pkts_available)
+			break;
+
+		pkts_processed = __octep_oq_process_rx(oct, oq,
+						       pkts_available);
+		oq->pkts_pending -= pkts_processed;
+		total_pkts_processed += pkts_processed;
+	}
+
+	if (oq->refill_count >= oq->refill_threshold) {
+		u32 desc_refilled = octep_oq_refill(oct, oq);
+
+		/* flush pending writes before updating credits */
+		wmb();
+		writel(desc_refilled, oq->pkts_credit_reg);
+	}
+
+	return total_pkts_processed;
+}

drivers/net/ethernet/marvell/octeon_ep/octep_tx.c

@@ -24,6 +24,79 @@ static void octep_iq_reset_indices(struct octep_iq *iq)
 	atomic_set(&iq->instr_pending, 0);
 }
 
+/**
+ * octep_iq_process_completions() - Process Tx queue completions.
+ *
+ * @iq: Octeon Tx queue data structure.
+ * @budget: max number of completions to be processed in one invocation.
+ */
+int octep_iq_process_completions(struct octep_iq *iq, u16 budget)
+{
+	u32 compl_pkts, compl_bytes, compl_sg;
+	struct octep_device *oct = iq->octep_dev;
+	struct octep_tx_buffer *tx_buffer;
+	struct skb_shared_info *shinfo;
+	u32 fi = iq->flush_index;
+	struct sk_buff *skb;
+	u8 frags, i;
+
+	compl_pkts = 0;
+	compl_sg = 0;
+	compl_bytes = 0;
+	iq->octep_read_index = oct->hw_ops.update_iq_read_idx(iq);
+
+	while (likely(budget && (fi != iq->octep_read_index))) {
+		tx_buffer = iq->buff_info + fi;
+		skb = tx_buffer->skb;
+
+		fi++;
+		if (unlikely(fi == iq->max_count))
+			fi = 0;
+		compl_bytes += skb->len;
+		compl_pkts++;
+		budget--;
+
+		if (!tx_buffer->gather) {
+			dma_unmap_single(iq->dev, tx_buffer->dma,
+					 tx_buffer->skb->len, DMA_TO_DEVICE);
+			dev_kfree_skb_any(skb);
+			continue;
+		}
+
+		/* Scatter/Gather */
+		shinfo = skb_shinfo(skb);
+		frags = shinfo->nr_frags;
+		compl_sg++;
+
+		dma_unmap_single(iq->dev, tx_buffer->sglist[0].dma_ptr[0],
+				 tx_buffer->sglist[0].len[0], DMA_TO_DEVICE);
+
+		i = 1; /* entry 0 is main skb, unmapped above */
+		while (frags--) {
+			dma_unmap_page(iq->dev, tx_buffer->sglist[i >> 2].dma_ptr[i & 3],
+				       tx_buffer->sglist[i >> 2].len[i & 3], DMA_TO_DEVICE);
+			i++;
+		}
+
+		dev_kfree_skb_any(skb);
+	}
+
+	iq->pkts_processed += compl_pkts;
+	atomic_sub(compl_pkts, &iq->instr_pending);
+	iq->stats.instr_completed += compl_pkts;
+	iq->stats.bytes_sent += compl_bytes;
+	iq->stats.sgentry_sent += compl_sg;
+	iq->flush_index = fi;
+
+	netdev_tx_completed_queue(iq->netdev_q, compl_pkts, compl_bytes);
+
+	if (unlikely(__netif_subqueue_stopped(iq->netdev, iq->q_no)) &&
+	    ((iq->max_count - atomic_read(&iq->instr_pending)) >
+	     OCTEP_WAKE_QUEUE_THRESHOLD))
+		netif_wake_subqueue(iq->netdev, iq->q_no);
+	return !budget;
+}
+
 /**
  * octep_iq_free_pending() - Free Tx buffers for pending completions.
  *