Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751369AbaLCFO7 (ORCPT ); Wed, 3 Dec 2014 00:14:59 -0500 Received: from rtits2.realtek.com ([60.250.210.242]:50524 "EHLO rtits2.realtek.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750860AbaLCFO5 (ORCPT ); Wed, 3 Dec 2014 00:14:57 -0500 Authenticated-By: X-SpamFilter-By: BOX Solutions SpamTrap 5.49 with qID sB35ErK9004754, This message is accepted by code: ctloc85258 From: Hayes Wang To: CC: , , , Hayes Wang Subject: [PATCH net-next] r8152: reduce memory copy for rx Date: Wed, 3 Dec 2014 13:14:38 +0800 Message-ID: <1394712342-15778-105-Taiwan-albertk@realtek.com> X-Mailer: Microsoft Office Outlook 11 MIME-Version: 1.0 Content-Type: text/plain X-Originating-IP: [172.21.71.44] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org If the data size is more than half of the AGG_BUG_SZ, allocate a new rx buffer and use skb_clone() to avoid the memory copy. The original method is that allocate the memory and copy data for each packet in a rx buffer. The new one is that when the data size for a rx buffer is more than RX_THRESHOLD_CLONED, allocate a new rx buffer and use skb_clone for each packet in the rx buffer. According to the experiment, the new mothod has better performance. Signed-off-by: Hayes Wang --- drivers/net/usb/r8152.c | 110 +++++++++++++++++++++++++++++++++--------------- 1 file changed, 77 insertions(+), 33 deletions(-) diff --git a/drivers/net/usb/r8152.c b/drivers/net/usb/r8152.c index 4a9ece0..e44b9fb 100644 --- a/drivers/net/usb/r8152.c +++ b/drivers/net/usb/r8152.c @@ -26,7 +26,7 @@ #include /* Version Information */ -#define DRIVER_VERSION "v1.07.0 (2014/10/09)" +#define DRIVER_VERSION "v1.08.0 (2014/11/27)" #define DRIVER_AUTHOR "Realtek linux nic maintainers " #define DRIVER_DESC "Realtek RTL8152/RTL8153 Based USB Ethernet Adapters" #define MODULENAME "r8152" @@ -447,6 +447,8 @@ enum rtl_register_content { #define RTL8152_RMS (VLAN_ETH_FRAME_LEN + VLAN_HLEN) #define RTL8153_RMS RTL8153_MAX_PACKET #define RTL8152_TX_TIMEOUT (5 * HZ) +#define AGG_BUF_SZ 16384 /* 16K */ +#define RX_THRESHOLD_CLONED (AGG_BUF_SZ / 2) /* rtl8152 flags */ enum rtl8152_flags { @@ -534,8 +536,7 @@ struct rx_agg { struct list_head list; struct urb *urb; struct r8152 *context; - void *buffer; - void *head; + struct sk_buff *skb; }; struct tx_agg { @@ -605,9 +606,8 @@ enum tx_csum_stat { * The RTL chips use a 64 element hash table based on the Ethernet CRC. */ static const int multicast_filter_limit = 32; -static unsigned int agg_buf_sz = 16384; -#define RTL_LIMITED_TSO_SIZE (agg_buf_sz - sizeof(struct tx_desc) - \ +#define RTL_LIMITED_TSO_SIZE (AGG_BUF_SZ - sizeof(struct tx_desc) - \ VLAN_ETH_HLEN - VLAN_HLEN) static @@ -1210,9 +1210,8 @@ static void free_all_mem(struct r8152 *tp) usb_free_urb(tp->rx_info[i].urb); tp->rx_info[i].urb = NULL; - kfree(tp->rx_info[i].buffer); - tp->rx_info[i].buffer = NULL; - tp->rx_info[i].head = NULL; + dev_kfree_skb(tp->rx_info[i].skb); + tp->rx_info[i].skb = NULL; } for (i = 0; i < RTL8152_MAX_TX; i++) { @@ -1231,6 +1230,31 @@ static void free_all_mem(struct r8152 *tp) tp->intr_buff = NULL; } +static struct sk_buff *rtl_alloc_rx_skb(struct r8152 *tp, gfp_t gfp_mask) +{ + struct net_device *netdev = tp->netdev; + struct sk_buff *skb; + + skb = __netdev_alloc_skb(netdev, AGG_BUF_SZ, gfp_mask); + if (!skb) + goto out1; + + if (skb->data != rx_agg_align(skb->data)) { + int rl; + + dev_kfree_skb_any(skb); + skb = __netdev_alloc_skb(netdev, AGG_BUF_SZ + RX_ALIGN, + gfp_mask); + if (!skb) + goto out1; + + rl = (int)(rx_agg_align(skb->data) - (void *)skb->data); + skb_reserve(skb, rl); + } +out1: + return skb; +} + static int alloc_all_mem(struct r8152 *tp) { struct net_device *netdev = tp->netdev; @@ -1239,7 +1263,6 @@ static int alloc_all_mem(struct r8152 *tp) struct usb_host_endpoint *ep_intr = alt->endpoint + 2; struct urb *urb; int node, i; - u8 *buf; node = netdev->dev.parent ? dev_to_node(netdev->dev.parent) : -1; @@ -1249,39 +1272,33 @@ static int alloc_all_mem(struct r8152 *tp) skb_queue_head_init(&tp->tx_queue); for (i = 0; i < RTL8152_MAX_RX; i++) { - buf = kmalloc_node(agg_buf_sz, GFP_KERNEL, node); - if (!buf) - goto err1; + struct sk_buff *skb; - if (buf != rx_agg_align(buf)) { - kfree(buf); - buf = kmalloc_node(agg_buf_sz + RX_ALIGN, GFP_KERNEL, - node); - if (!buf) - goto err1; - } + skb = rtl_alloc_rx_skb(tp, GFP_KERNEL); + if (!skb) + goto err1; urb = usb_alloc_urb(0, GFP_KERNEL); if (!urb) { - kfree(buf); + dev_kfree_skb(skb); goto err1; } INIT_LIST_HEAD(&tp->rx_info[i].list); tp->rx_info[i].context = tp; tp->rx_info[i].urb = urb; - tp->rx_info[i].buffer = buf; - tp->rx_info[i].head = rx_agg_align(buf); + tp->rx_info[i].skb = skb; } for (i = 0; i < RTL8152_MAX_TX; i++) { - buf = kmalloc_node(agg_buf_sz, GFP_KERNEL, node); + u8 *buf = kmalloc_node(AGG_BUF_SZ, GFP_KERNEL, node); + if (!buf) goto err1; if (buf != tx_agg_align(buf)) { kfree(buf); - buf = kmalloc_node(agg_buf_sz + TX_ALIGN, GFP_KERNEL, + buf = kmalloc_node(AGG_BUF_SZ + TX_ALIGN, GFP_KERNEL, node); if (!buf) goto err1; @@ -1538,7 +1555,7 @@ static int r8152_tx_agg_fill(struct r8152 *tp, struct tx_agg *agg) tx_data = agg->head; agg->skb_num = 0; agg->skb_len = 0; - remain = agg_buf_sz; + remain = AGG_BUF_SZ; while (remain >= ETH_ZLEN + sizeof(struct tx_desc)) { struct tx_desc *tx_desc; @@ -1587,7 +1604,7 @@ static int r8152_tx_agg_fill(struct r8152 *tp, struct tx_agg *agg) dev_kfree_skb_any(skb); - remain = agg_buf_sz - (int)(tx_agg_align(tx_data) - agg->head); + remain = AGG_BUF_SZ - (int)(tx_agg_align(tx_data) - agg->head); } if (!skb_queue_empty(&skb_head)) { @@ -1666,6 +1683,8 @@ static void rx_bottom(struct r8152 *tp) list_for_each_safe(cursor, next, &rx_queue) { struct rx_desc *rx_desc; + struct sk_buff *rx_skb; + bool cloned = false; struct rx_agg *agg; int len_used = 0; struct urb *urb; @@ -1678,10 +1697,21 @@ static void rx_bottom(struct r8152 *tp) if (urb->actual_length < ETH_ZLEN) goto submit; - rx_desc = agg->head; - rx_data = agg->head; + rx_skb = agg->skb; + rx_desc = (struct rx_desc *)rx_skb->data; + rx_data = rx_skb->data; len_used += sizeof(struct rx_desc); + if (!NET_IP_ALIGN && urb->actual_length > RX_THRESHOLD_CLONED) { + struct sk_buff *new_skb; + + new_skb = rtl_alloc_rx_skb(tp, GFP_ATOMIC); + if (new_skb) { + agg->skb = new_skb; + cloned = true; + } + } + while (urb->actual_length > len_used) { struct net_device *netdev = tp->netdev; struct net_device_stats *stats = &netdev->stats; @@ -1699,14 +1729,23 @@ static void rx_bottom(struct r8152 *tp) pkt_len -= CRC_SIZE; rx_data += sizeof(struct rx_desc); - skb = netdev_alloc_skb_ip_align(netdev, pkt_len); + if (cloned) + skb = skb_clone(rx_skb, GFP_ATOMIC); + else + skb = netdev_alloc_skb_ip_align(netdev, + pkt_len); if (!skb) { stats->rx_dropped++; goto find_next_rx; } skb->ip_summed = r8152_rx_csum(tp, rx_desc); - memcpy(skb->data, rx_data, pkt_len); + + if (cloned) + skb_reserve(skb, (int)(rx_data - rx_skb->data)); + else + memcpy(skb->data, rx_data, pkt_len); + skb_put(skb, pkt_len); skb->protocol = eth_type_trans(skb, netdev); rtl_rx_vlan_tag(rx_desc, skb); @@ -1717,10 +1756,14 @@ static void rx_bottom(struct r8152 *tp) find_next_rx: rx_data = rx_agg_align(rx_data + pkt_len + CRC_SIZE); rx_desc = (struct rx_desc *)rx_data; - len_used = (int)(rx_data - (u8 *)agg->head); + len_used = (int)(rx_data - rx_skb->data); len_used += sizeof(struct rx_desc); } + /* free the cloned skb */ + if (cloned) + dev_kfree_skb_any(rx_skb); + submit: r8152_submit_rx(tp, agg, GFP_ATOMIC); } @@ -1789,10 +1832,11 @@ static void bottom_half(unsigned long data) static int r8152_submit_rx(struct r8152 *tp, struct rx_agg *agg, gfp_t mem_flags) { + struct sk_buff *skb = agg->skb; int ret; usb_fill_bulk_urb(agg->urb, tp->udev, usb_rcvbulkpipe(tp->udev, 1), - agg->head, agg_buf_sz, + skb->data, AGG_BUF_SZ, (usb_complete_t)read_bulk_callback, agg); ret = usb_submit_urb(agg->urb, mem_flags); @@ -1951,7 +1995,7 @@ static void set_tx_qlen(struct r8152 *tp) { struct net_device *netdev = tp->netdev; - tp->tx_qlen = agg_buf_sz / (netdev->mtu + VLAN_ETH_HLEN + VLAN_HLEN + + tp->tx_qlen = AGG_BUF_SZ / (netdev->mtu + VLAN_ETH_HLEN + VLAN_HLEN + sizeof(struct tx_desc)); } -- 1.9.3 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/