lan743x: boost performance on cpu archs w/o dma cache snooping

The buffers in the lan743x driver's receive ring are always 9K, even when the largest packet that can be received (the mtu) is much smaller. This performs particularly badly on cpu archs without dma cache snooping (such as ARM): each received packet results in a 9K dma_{map|unmap} operation, which is very expensive because cpu caches need to be invalidated. Careful measurement of the driver rx path on armv7 reveals that the cpu spends the majority of its time waiting for cache invalidation. Optimize by keeping the rx ring buffer size as close as possible to the mtu. This limits the amount of cache that requires invalidation. This optimization would normally force us to re-allocate all ring buffers when the mtu is changed - a disruptive event, because it can only happen when the network interface is down. Remove the need to re-allocate all ring buffers by adding support for multi-buffer frames. Now any combination of mtu and ring buffer size will work. When the mtu changes from mtu1 to mtu2, consumed buffers of size mtu1 are lazily replaced by newly allocated buffers of size mtu2. These optimizations double the rx performance on armv7. Third parties report 3x rx speedup on armv8. Tested with iperf3 on a freescale imx6qp + lan7430, both sides set to mtu 1500 bytes, measure rx performance: Before: [ ID] Interval Transfer Bandwidth Retr [ 4] 0.00-20.00 sec 550 MBytes 231 Mbits/sec 0 After: [ ID] Interval Transfer Bandwidth Retr [ 4] 0.00-20.00 sec 1.33 GBytes 570 Mbits/sec 0 Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com> Reviewed-by: Bryan Whitehead <Bryan.Whitehead@microchip.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2024-12-02 08:34:20 +08:00 · 2021-02-15 20:08:02 -05:00 · 2021-02-15 20:08:02 -05:00 · a8db76d40e
commit a8db76d40e
parent 80fea53dbe
2 changed files with 146 additions and 179 deletions
--- a/drivers/net/ethernet/microchip/lan743x_main.c
+++ b/drivers/net/ethernet/microchip/lan743x_main.c
@ -1926,15 +1926,6 @@ static int lan743x_rx_next_index(struct lan743x_rx *rx, int index)
 	return ((++index) % rx->ring_size);
 }

-static struct sk_buff *lan743x_rx_allocate_skb(struct lan743x_rx *rx)
-{
-	int length = 0;
-
-	length = (LAN743X_MAX_FRAME_SIZE + ETH_HLEN + 4 + RX_HEAD_PADDING);
-	return __netdev_alloc_skb(rx->adapter->netdev,
-				  length, GFP_ATOMIC | GFP_DMA);
-}
-
 static void lan743x_rx_update_tail(struct lan743x_rx *rx, int index)
 {
 	/* update the tail once per 8 descriptors */
@ -1943,36 +1934,40 @@ static void lan743x_rx_update_tail(struct lan743x_rx *rx, int index)
 				  index);
 }

-static int lan743x_rx_init_ring_element(struct lan743x_rx *rx, int index,
-					struct sk_buff *skb)
+static int lan743x_rx_init_ring_element(struct lan743x_rx *rx, int index)
 {
+	struct net_device *netdev = rx->adapter->netdev;
+	struct device *dev = &rx->adapter->pdev->dev;
 	struct lan743x_rx_buffer_info *buffer_info;
 	struct lan743x_rx_descriptor *descriptor;
-	int length = 0;
+	struct sk_buff *skb;
+	dma_addr_t dma_ptr;
+	int length;
+
+	length = netdev->mtu + ETH_HLEN + 4 + RX_HEAD_PADDING;

-	length = (LAN743X_MAX_FRAME_SIZE + ETH_HLEN + 4 + RX_HEAD_PADDING);
 	descriptor = &rx->ring_cpu_ptr[index];
 	buffer_info = &rx->buffer_info[index];
-	buffer_info->skb = skb;
-	if (!(buffer_info->skb))
+	skb = __netdev_alloc_skb(netdev, length, GFP_ATOMIC | GFP_DMA);
+	if (!skb)
 		return -ENOMEM;
-	buffer_info->dma_ptr = dma_map_single(&rx->adapter->pdev->dev,
-					      buffer_info->skb->data,
-					      length,
-					      DMA_FROM_DEVICE);
-	if (dma_mapping_error(&rx->adapter->pdev->dev,
-			      buffer_info->dma_ptr)) {
-		buffer_info->dma_ptr = 0;
+	dma_ptr = dma_map_single(dev, skb->data, length, DMA_FROM_DEVICE);
+	if (dma_mapping_error(dev, dma_ptr)) {
+		dev_kfree_skb_any(skb);
 		return -ENOMEM;
 	}
+	if (buffer_info->dma_ptr)
+		dma_unmap_single(dev, buffer_info->dma_ptr,
+				 buffer_info->buffer_length, DMA_FROM_DEVICE);

+	buffer_info->skb = skb;
+	buffer_info->dma_ptr = dma_ptr;
 	buffer_info->buffer_length = length;
 	descriptor->data1 = cpu_to_le32(DMA_ADDR_LOW32(buffer_info->dma_ptr));
 	descriptor->data2 = cpu_to_le32(DMA_ADDR_HIGH32(buffer_info->dma_ptr));
 	descriptor->data3 = 0;
 	descriptor->data0 = cpu_to_le32((RX_DESC_DATA0_OWN_ |
 			    (length & RX_DESC_DATA0_BUF_LENGTH_MASK_)));
-	skb_reserve(buffer_info->skb, RX_HEAD_PADDING);
 	lan743x_rx_update_tail(rx, index);

 	return 0;
@ -2021,16 +2016,32 @@ static void lan743x_rx_release_ring_element(struct lan743x_rx *rx, int index)
 	memset(buffer_info, 0, sizeof(*buffer_info));
 }

-static int lan743x_rx_process_packet(struct lan743x_rx *rx)
+static struct sk_buff *
+lan743x_rx_trim_skb(struct sk_buff *skb, int frame_length)
+{
+	if (skb_linearize(skb)) {
+		dev_kfree_skb_irq(skb);
+		return NULL;
+	}
+	frame_length = max_t(int, 0, frame_length - RX_HEAD_PADDING - 2);
+	if (skb->len > frame_length) {
+		skb->tail -= skb->len - frame_length;
+		skb->len = frame_length;
+	}
+	return skb;
+}
+
+static int lan743x_rx_process_buffer(struct lan743x_rx *rx)
 {
-	struct skb_shared_hwtstamps *hwtstamps = NULL;
-	int result = RX_PROCESS_RESULT_NOTHING_TO_DO;
 	int current_head_index = le32_to_cpu(*rx->head_cpu_ptr);
+	struct lan743x_rx_descriptor *descriptor, *desc_ext;
+	struct net_device *netdev = rx->adapter->netdev;
+	int result = RX_PROCESS_RESULT_NOTHING_TO_DO;
 	struct lan743x_rx_buffer_info *buffer_info;
-	struct lan743x_rx_descriptor *descriptor;
+	int frame_length, buffer_length;
 	int extension_index = -1;
-	int first_index = -1;
-	int last_index = -1;
+	bool is_last, is_first;
+	struct sk_buff *skb;

 	if (current_head_index < 0 || current_head_index >= rx->ring_size)
 		goto done;
@ -2038,163 +2049,120 @@ static int lan743x_rx_process_packet(struct lan743x_rx *rx)
 	if (rx->last_head < 0 || rx->last_head >= rx->ring_size)
 		goto done;

-	if (rx->last_head != current_head_index) {
-		descriptor = &rx->ring_cpu_ptr[rx->last_head];
-		if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_OWN_)
+	if (rx->last_head == current_head_index)
+		goto done;
+
+	descriptor = &rx->ring_cpu_ptr[rx->last_head];
+	if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_OWN_)
+		goto done;
+	buffer_info = &rx->buffer_info[rx->last_head];
+
+	is_last = le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_LS_;
+	is_first = le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_FS_;
+
+	if (is_last && le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_EXT_) {
+		/* extension is expected to follow */
+		int index = lan743x_rx_next_index(rx, rx->last_head);
+
+		if (index == current_head_index)
+			/* extension not yet available */
 			goto done;
-
-		if (!(le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_FS_))
+		desc_ext = &rx->ring_cpu_ptr[index];
+		if (le32_to_cpu(desc_ext->data0) & RX_DESC_DATA0_OWN_)
+			/* extension not yet available */
 			goto done;
-
-		first_index = rx->last_head;
-		if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_LS_) {
-			last_index = rx->last_head;
-		} else {
-			int index;
-
-			index = lan743x_rx_next_index(rx, first_index);
-			while (index != current_head_index) {
-				descriptor = &rx->ring_cpu_ptr[index];
-				if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_OWN_)
-					goto done;
-
-				if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_LS_) {
-					last_index = index;
-					break;
-				}
-				index = lan743x_rx_next_index(rx, index);
-			}
-		}
-		if (last_index >= 0) {
-			descriptor = &rx->ring_cpu_ptr[last_index];
-			if (le32_to_cpu(descriptor->data0) & RX_DESC_DATA0_EXT_) {
-				/* extension is expected to follow */
-				int index = lan743x_rx_next_index(rx,
-								  last_index);
-				if (index != current_head_index) {
-					descriptor = &rx->ring_cpu_ptr[index];
-					if (le32_to_cpu(descriptor->data0) &
-					    RX_DESC_DATA0_OWN_) {
-						goto done;
-					}
-					if (le32_to_cpu(descriptor->data0) &
-					    RX_DESC_DATA0_EXT_) {
-						extension_index = index;
-					} else {
-						goto done;
-					}
-				} else {
-					/* extension is not yet available */
-					/* prevent processing of this packet */
-					first_index = -1;
-					last_index = -1;
-				}
-			}
-		}
+		if (!(le32_to_cpu(desc_ext->data0) & RX_DESC_DATA0_EXT_))
+			goto move_forward;
+		extension_index = index;
 	}
-	if (first_index >= 0 && last_index >= 0) {
-		int real_last_index = last_index;
-		struct sk_buff *skb = NULL;
-		u32 ts_sec = 0;
-		u32 ts_nsec = 0;

-		/* packet is available */
-		if (first_index == last_index) {
-			/* single buffer packet */
-			struct sk_buff *new_skb = NULL;
-			int packet_length;
+	/* Only the last buffer in a multi-buffer frame contains the total frame
+	 * length. The chip occasionally sends more buffers than strictly
+	 * required to reach the total frame length.
+	 * Handle this by adding all buffers to the skb in their entirety.
+	 * Once the real frame length is known, trim the skb.
+	 */
+	frame_length =
+		RX_DESC_DATA0_FRAME_LENGTH_GET_(le32_to_cpu(descriptor->data0));
+	buffer_length = buffer_info->buffer_length;

-			new_skb = lan743x_rx_allocate_skb(rx);
-			if (!new_skb) {
-				/* failed to allocate next skb.
-				 * Memory is very low.
-				 * Drop this packet and reuse buffer.
-				 */
-				lan743x_rx_reuse_ring_element(rx, first_index);
-				goto process_extension;
-			}
+	netdev_dbg(netdev, "%s%schunk: %d/%d",
+		   is_first ? "first " : "      ",
+		   is_last  ? "last  " : "      ",
+		   frame_length, buffer_length);

-			buffer_info = &rx->buffer_info[first_index];
-			skb = buffer_info->skb;
-			descriptor = &rx->ring_cpu_ptr[first_index];
+	/* save existing skb, allocate new skb and map to dma */
+	skb = buffer_info->skb;
+	if (lan743x_rx_init_ring_element(rx, rx->last_head)) {
+		/* failed to allocate next skb.
+		 * Memory is very low.
+		 * Drop this packet and reuse buffer.
+		 */
+		lan743x_rx_reuse_ring_element(rx, rx->last_head);
+		/* drop packet that was being assembled */
+		dev_kfree_skb_irq(rx->skb_head);
+		rx->skb_head = NULL;
+		goto process_extension;
+	}

-			/* unmap from dma */
-			if (buffer_info->dma_ptr) {
-				dma_unmap_single(&rx->adapter->pdev->dev,
-						 buffer_info->dma_ptr,
-						 buffer_info->buffer_length,
-						 DMA_FROM_DEVICE);
-				buffer_info->dma_ptr = 0;
-				buffer_info->buffer_length = 0;
-			}
-			buffer_info->skb = NULL;
-			packet_length =	RX_DESC_DATA0_FRAME_LENGTH_GET_
-					(le32_to_cpu(descriptor->data0));
-			skb_put(skb, packet_length - 4);
-			skb->protocol = eth_type_trans(skb,
-						       rx->adapter->netdev);
-			lan743x_rx_init_ring_element(rx, first_index, new_skb);
-		} else {
-			int index = first_index;
-
-			/* multi buffer packet not supported */
-			/* this should not happen since
-			 * buffers are allocated to be at least jumbo size
-			 */
-
-			/* clean up buffers */
-			if (first_index <= last_index) {
-				while ((index >= first_index) &&
-				       (index <= last_index)) {
-					lan743x_rx_reuse_ring_element(rx,
-								      index);
-					index = lan743x_rx_next_index(rx,
-								      index);
-				}
-			} else {
-				while ((index >= first_index) ||
-				       (index <= last_index)) {
-					lan743x_rx_reuse_ring_element(rx,
-								      index);
-					index = lan743x_rx_next_index(rx,
-								      index);
-				}
-			}
-		}
+	/* add buffers to skb via skb->frag_list */
+	if (is_first) {
+		skb_reserve(skb, RX_HEAD_PADDING);
+		skb_put(skb, buffer_length - RX_HEAD_PADDING);
+		if (rx->skb_head)
+			dev_kfree_skb_irq(rx->skb_head);
+		rx->skb_head = skb;
+	} else if (rx->skb_head) {
+		skb_put(skb, buffer_length);
+		if (skb_shinfo(rx->skb_head)->frag_list)
+			rx->skb_tail->next = skb;
+		else
+			skb_shinfo(rx->skb_head)->frag_list = skb;
+		rx->skb_tail = skb;
+		rx->skb_head->len += skb->len;
+		rx->skb_head->data_len += skb->len;
+		rx->skb_head->truesize += skb->truesize;
+	} else {
+		/* packet to assemble has already been dropped because one or
+		 * more of its buffers could not be allocated
+		 */
+		netdev_dbg(netdev, "drop buffer intended for dropped packet");
+		dev_kfree_skb_irq(skb);
+	}

 process_extension:
-		if (extension_index >= 0) {
-			descriptor = &rx->ring_cpu_ptr[extension_index];
-			buffer_info = &rx->buffer_info[extension_index];
+	if (extension_index >= 0) {
+		u32 ts_sec;
+		u32 ts_nsec;

-			ts_sec = le32_to_cpu(descriptor->data1);
-			ts_nsec = (le32_to_cpu(descriptor->data2) &
-				  RX_DESC_DATA2_TS_NS_MASK_);
-			lan743x_rx_reuse_ring_element(rx, extension_index);
-			real_last_index = extension_index;
-		}
+		ts_sec = le32_to_cpu(desc_ext->data1);
+		ts_nsec = (le32_to_cpu(desc_ext->data2) &
+			  RX_DESC_DATA2_TS_NS_MASK_);
+		if (rx->skb_head)
+			skb_hwtstamps(rx->skb_head)->hwtstamp =
+				ktime_set(ts_sec, ts_nsec);
+		lan743x_rx_reuse_ring_element(rx, extension_index);
+		rx->last_head = extension_index;
+		netdev_dbg(netdev, "process extension");
+	}

-		if (!skb) {
-			result = RX_PROCESS_RESULT_PACKET_DROPPED;
-			goto move_forward;
-		}
+	if (is_last && rx->skb_head)
+		rx->skb_head = lan743x_rx_trim_skb(rx->skb_head, frame_length);

-		if (extension_index < 0)
-			goto pass_packet_to_os;
-		hwtstamps = skb_hwtstamps(skb);
-		if (hwtstamps)
-			hwtstamps->hwtstamp = ktime_set(ts_sec, ts_nsec);
-
-pass_packet_to_os:
-		/* pass packet to OS */
-		napi_gro_receive(&rx->napi, skb);
-		result = RX_PROCESS_RESULT_PACKET_RECEIVED;
+	if (is_last && rx->skb_head) {
+		rx->skb_head->protocol = eth_type_trans(rx->skb_head,
+							rx->adapter->netdev);
+		netdev_dbg(netdev, "sending %d byte frame to OS",
+			   rx->skb_head->len);
+		napi_gro_receive(&rx->napi, rx->skb_head);
+		rx->skb_head = NULL;
+	}

 move_forward:
-		/* push tail and head forward */
-		rx->last_tail = real_last_index;
-		rx->last_head = lan743x_rx_next_index(rx, real_last_index);
-	}
+	/* push tail and head forward */
+	rx->last_tail = rx->last_head;
+	rx->last_head = lan743x_rx_next_index(rx, rx->last_head);
+	result = RX_PROCESS_RESULT_BUFFER_RECEIVED;
 done:
 	return result;
 }
@ -2213,12 +2181,12 @@ static int lan743x_rx_napi_poll(struct napi_struct *napi, int weight)
 				  DMAC_INT_BIT_RXFRM_(rx->channel_number));
 	}
 	for (count = 0; count < weight; count++) {
-		result = lan743x_rx_process_packet(rx);
+		result = lan743x_rx_process_buffer(rx);
 		if (result == RX_PROCESS_RESULT_NOTHING_TO_DO)
 			break;
 	}
 	rx->frame_count += count;
-	if (count == weight || result == RX_PROCESS_RESULT_PACKET_RECEIVED)
+	if (count == weight || result == RX_PROCESS_RESULT_BUFFER_RECEIVED)
 		return weight;

 	if (!napi_complete_done(napi, count))
@ -2330,9 +2298,7 @@ static int lan743x_rx_ring_init(struct lan743x_rx *rx)

 	rx->last_head = 0;
 	for (index = 0; index < rx->ring_size; index++) {
-		struct sk_buff *new_skb = lan743x_rx_allocate_skb(rx);
-
-		ret = lan743x_rx_init_ring_element(rx, index, new_skb);
+		ret = lan743x_rx_init_ring_element(rx, index);
 		if (ret)
 			goto cleanup;
 	}
--- a/drivers/net/ethernet/microchip/lan743x_main.h
+++ b/drivers/net/ethernet/microchip/lan743x_main.h
@ -699,6 +699,8 @@ struct lan743x_rx {
 	struct napi_struct napi;

 	u32		frame_count;
+
+	struct sk_buff *skb_head, *skb_tail;
 };

 struct lan743x_adapter {
@ -831,8 +833,7 @@ struct lan743x_rx_buffer_info {
 #define LAN743X_RX_RING_SIZE        (65)

 #define RX_PROCESS_RESULT_NOTHING_TO_DO     (0)
-#define RX_PROCESS_RESULT_PACKET_RECEIVED   (1)
-#define RX_PROCESS_RESULT_PACKET_DROPPED    (2)
+#define RX_PROCESS_RESULT_BUFFER_RECEIVED   (1)

 u32 lan743x_csr_read(struct lan743x_adapter *adapter, int offset);
 void lan743x_csr_write(struct lan743x_adapter *adapter, int offset, u32 data);