Merge branch 'r8152-modify-rx_bottom'

Hayes Wang says:

====================
r8152: modify rx_bottom

v3:
For patch #1, this patch is replaced. The new patch only break the loop,
and keep that the driver would queue the rx packets.

For patch #2, modify the code depends on patch #1. For work_down < budget,
napi_get_frags() and napi_gro_frags() would be used. For the others,
nothing is changed.

v2:
For patch #1, add comment, update commit message, and add Fixes tag.

v1:
These patches are used to improve rx_bottom().
====================

Link: https://lore.kernel.org/r/20230926111714.9448-432-nic_swsd@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: kuba-moo

drivers/net/usb/r8152.c

@@ -2449,7 +2449,7 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		}
 	}
 
-	if (list_empty(&tp->rx_done))
+	if (list_empty(&tp->rx_done) || work_done >= budget)
 		goto out1;
 
 	clear_bit(RX_EPROTO, &tp->flags);
@@ -2465,6 +2465,15 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		struct urb *urb;
 		u8 *rx_data;
 
+		/* A bulk transfer of USB may contain may packets, so the
+		 * total packets may more than the budget. Deal with all
+		 * packets in current bulk transfer, and stop to handle the
+		 * next bulk transfer until next schedule, if budget is
+		 * exhausted.
+		 */
+		if (work_done >= budget)
+			break;
+
 		list_del_init(cursor);
 
 		agg = list_entry(cursor, struct rx_agg, list);
@@ -2481,12 +2490,11 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		while (urb->actual_length > len_used) {
 			struct net_device *netdev = tp->netdev;
 			struct net_device_stats *stats = &netdev->stats;
-			unsigned int pkt_len, rx_frag_head_sz;
+			unsigned int pkt_len, rx_frag_head_sz, len;
 			struct sk_buff *skb;
+			bool use_frags;
 
-			/* limit the skb numbers for rx_queue */
-			if (unlikely(skb_queue_len(&tp->rx_queue) >= 1000))
-				break;
+			WARN_ON_ONCE(skb_queue_len(&tp->rx_queue) >= 1000);
 
 			pkt_len = le32_to_cpu(rx_desc->opts1) & RX_LEN_MASK;
 			if (pkt_len < ETH_ZLEN)
@@ -2497,45 +2505,77 @@ static int rx_bottom(struct r8152 *tp, int budget)
 				break;
 
 			pkt_len -= ETH_FCS_LEN;
+			len = pkt_len;
 			rx_data += sizeof(struct rx_desc);
 
-			if (!agg_free || tp->rx_copybreak > pkt_len)
-				rx_frag_head_sz = pkt_len;
+			if (!agg_free || tp->rx_copybreak > len)
+				use_frags = false;
 			else
-				rx_frag_head_sz = tp->rx_copybreak;
+				use_frags = true;
+
+			if (use_frags) {
+				/* If the budget is exhausted, the packet
+				 * would be queued in the driver. That is,
+				 * napi_gro_frags() wouldn't be called, so
+				 * we couldn't use napi_get_frags().
+				 */
+				if (work_done >= budget) {
+					rx_frag_head_sz = tp->rx_copybreak;
+					skb = napi_alloc_skb(napi,
+							     rx_frag_head_sz);
+				} else {
+					rx_frag_head_sz = 0;
+					skb = napi_get_frags(napi);
+				}
+			} else {
+				rx_frag_head_sz = 0;
+				skb = napi_alloc_skb(napi, len);
+			}
 
-			skb = napi_alloc_skb(napi, rx_frag_head_sz);
 			if (!skb) {
 				stats->rx_dropped++;
 				goto find_next_rx;
 			}
 
 			skb->ip_summed = r8152_rx_csum(tp, rx_desc);
-			memcpy(skb->data, rx_data, rx_frag_head_sz);
-			skb_put(skb, rx_frag_head_sz);
-			pkt_len -= rx_frag_head_sz;
-			rx_data += rx_frag_head_sz;
-			if (pkt_len) {
+			rtl_rx_vlan_tag(rx_desc, skb);
+
+			if (use_frags) {
+				if (rx_frag_head_sz) {
+					memcpy(skb->data, rx_data,
+					       rx_frag_head_sz);
+					skb_put(skb, rx_frag_head_sz);
+					len -= rx_frag_head_sz;
+					rx_data += rx_frag_head_sz;
+					skb->protocol = eth_type_trans(skb,
+								       netdev);
+				}
+
 				skb_add_rx_frag(skb, 0, agg->page,
 						agg_offset(agg, rx_data),
-						pkt_len,
-						SKB_DATA_ALIGN(pkt_len));
+						len, SKB_DATA_ALIGN(len));
 				get_page(agg->page);
+			} else {
+				memcpy(skb->data, rx_data, len);
+				skb_put(skb, len);
+				skb->protocol = eth_type_trans(skb, netdev);
 			}
 
-			skb->protocol = eth_type_trans(skb, netdev);
-			rtl_rx_vlan_tag(rx_desc, skb);
 			if (work_done < budget) {
+				if (use_frags)
+					napi_gro_frags(napi);
+				else
+					napi_gro_receive(napi, skb);
+
 				work_done++;
 				stats->rx_packets++;
-				stats->rx_bytes += skb->len;
-				napi_gro_receive(napi, skb);
+				stats->rx_bytes += pkt_len;
 			} else {
 				__skb_queue_tail(&tp->rx_queue, skb);
 			}
 
 find_next_rx:
-			rx_data = rx_agg_align(rx_data + pkt_len + ETH_FCS_LEN);
+			rx_data = rx_agg_align(rx_data + len + ETH_FCS_LEN);
 			rx_desc = (struct rx_desc *)rx_data;
 			len_used = agg_offset(agg, rx_data);
 			len_used += sizeof(struct rx_desc);
@@ -2564,9 +2604,10 @@ static int rx_bottom(struct r8152 *tp, int budget)
 		}
 	}
 
+	/* Splice the remained list back to rx_done for next schedule */
 	if (!list_empty(&rx_queue)) {
 		spin_lock_irqsave(&tp->rx_lock, flags);
-		list_splice_tail(&rx_queue, &tp->rx_done);
+		list_splice(&rx_queue, &tp->rx_done);
 		spin_unlock_irqrestore(&tp->rx_lock, flags);
 	}