From: Anirban Chakraborty Date: Fri, 1 Apr 2011 14:28:11 +0000 (+0000) Subject: qlcnic: Code optimization patch X-Git-Tag: v3.0~749^2~512 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=036d61f05189c9c02de22dd19a1c64a4fd74a914;p=platform%2Fkernel%2Flinux-amlogic.git qlcnic: Code optimization patch Optimized code resulted in achieving lower CPU utilization on transmit path and higher throughput for small packet sizes (64 bytes). Signed-off-by: Anirban Chakraborty Signed-off-by: David S. Miller --- diff --git a/drivers/net/qlcnic/qlcnic.h b/drivers/net/qlcnic/qlcnic.h index 15d950a..a5b28d1 100644 --- a/drivers/net/qlcnic/qlcnic.h +++ b/drivers/net/qlcnic/qlcnic.h @@ -434,50 +434,49 @@ struct qlcnic_adapter_stats { * be one Rcv Descriptor for normal packets, one for jumbo and may be others. */ struct qlcnic_host_rds_ring { - u32 producer; + void __iomem *crb_rcv_producer; + struct rcv_desc *desc_head; + struct qlcnic_rx_buffer *rx_buf_arr; u32 num_desc; + u32 producer; u32 dma_size; u32 skb_size; u32 flags; - void __iomem *crb_rcv_producer; - struct rcv_desc *desc_head; - struct qlcnic_rx_buffer *rx_buf_arr; struct list_head free_list; spinlock_t lock; dma_addr_t phys_addr; -}; +} ____cacheline_internodealigned_in_smp; struct qlcnic_host_sds_ring { u32 consumer; u32 num_desc; void __iomem *crb_sts_consumer; - void __iomem *crb_intr_mask; struct status_desc *desc_head; struct qlcnic_adapter *adapter; struct napi_struct napi; struct list_head free_list[NUM_RCV_DESC_RINGS]; + void __iomem *crb_intr_mask; int irq; dma_addr_t phys_addr; char name[IFNAMSIZ+4]; -}; +} ____cacheline_internodealigned_in_smp; struct qlcnic_host_tx_ring { u32 producer; - __le32 *hw_consumer; u32 sw_consumer; - void __iomem *crb_cmd_producer; u32 num_desc; - - struct netdev_queue *txq; - - struct qlcnic_cmd_buffer *cmd_buf_arr; + void __iomem *crb_cmd_producer; struct cmd_desc_type0 *desc_head; + struct qlcnic_cmd_buffer *cmd_buf_arr; + __le32 *hw_consumer; + dma_addr_t phys_addr; dma_addr_t hw_cons_phys_addr; -}; + struct netdev_queue *txq; +} ____cacheline_internodealigned_in_smp; /* * Receive context. There is one such structure per instance of the @@ -1328,8 +1327,7 @@ static const struct qlcnic_brdinfo qlcnic_boards[] = { static inline u32 qlcnic_tx_avail(struct qlcnic_host_tx_ring *tx_ring) { - smp_mb(); - if (tx_ring->producer < tx_ring->sw_consumer) + if (likely(tx_ring->producer < tx_ring->sw_consumer)) return tx_ring->sw_consumer - tx_ring->producer; else return tx_ring->sw_consumer + tx_ring->num_desc - diff --git a/drivers/net/qlcnic/qlcnic_main.c b/drivers/net/qlcnic/qlcnic_main.c index dde7e44..3b740f55 100644 --- a/drivers/net/qlcnic/qlcnic_main.c +++ b/drivers/net/qlcnic/qlcnic_main.c @@ -1861,6 +1861,7 @@ static void qlcnic_change_filter(struct qlcnic_adapter *adapter, vlan_req->vlan_id = vlan_id; tx_ring->producer = get_next_index(producer, tx_ring->num_desc); + smp_mb(); } #define QLCNIC_MAC_HASH(MAC)\ @@ -1921,58 +1922,122 @@ qlcnic_send_filter(struct qlcnic_adapter *adapter, spin_unlock(&adapter->mac_learn_lock); } -static void -qlcnic_tso_check(struct net_device *netdev, - struct qlcnic_host_tx_ring *tx_ring, +static int +qlcnic_tx_pkt(struct qlcnic_adapter *adapter, struct cmd_desc_type0 *first_desc, struct sk_buff *skb) { - u8 opcode = TX_ETHER_PKT; - __be16 protocol = skb->protocol; - u16 flags = 0; - int copied, offset, copy_len, hdr_len = 0, tso = 0; + u8 opcode = 0, hdr_len = 0; + u16 flags = 0, vlan_tci = 0; + int copied, offset, copy_len; struct cmd_desc_type0 *hwdesc; struct vlan_ethhdr *vh; - struct qlcnic_adapter *adapter = netdev_priv(netdev); + struct qlcnic_host_tx_ring *tx_ring = adapter->tx_ring; + u16 protocol = ntohs(skb->protocol); u32 producer = tx_ring->producer; - __le16 vlan_oob = first_desc->flags_opcode & - cpu_to_le16(FLAGS_VLAN_OOB); + + if (protocol == ETH_P_8021Q) { + vh = (struct vlan_ethhdr *)skb->data; + flags = FLAGS_VLAN_TAGGED; + vlan_tci = vh->h_vlan_TCI; + } else if (vlan_tx_tag_present(skb)) { + flags = FLAGS_VLAN_OOB; + vlan_tci = vlan_tx_tag_get(skb); + } + if (unlikely(adapter->pvid)) { + if (vlan_tci && !(adapter->flags & QLCNIC_TAGGING_ENABLED)) + return -EIO; + if (vlan_tci && (adapter->flags & QLCNIC_TAGGING_ENABLED)) + goto set_flags; + + flags = FLAGS_VLAN_OOB; + vlan_tci = adapter->pvid; + } +set_flags: + qlcnic_set_tx_vlan_tci(first_desc, vlan_tci); + qlcnic_set_tx_flags_opcode(first_desc, flags, opcode); if (*(skb->data) & BIT_0) { flags |= BIT_0; memcpy(&first_desc->eth_addr, skb->data, ETH_ALEN); } - - if ((netdev->features & (NETIF_F_TSO | NETIF_F_TSO6)) && + opcode = TX_ETHER_PKT; + if ((adapter->netdev->features & (NETIF_F_TSO | NETIF_F_TSO6)) && skb_shinfo(skb)->gso_size > 0) { hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb); first_desc->mss = cpu_to_le16(skb_shinfo(skb)->gso_size); first_desc->total_hdr_length = hdr_len; - if (vlan_oob) { + + opcode = (protocol == ETH_P_IPV6) ? TX_TCP_LSO6 : TX_TCP_LSO; + + /* For LSO, we need to copy the MAC/IP/TCP headers into + * the descriptor ring */ + copied = 0; + offset = 2; + + if (flags & FLAGS_VLAN_OOB) { first_desc->total_hdr_length += VLAN_HLEN; first_desc->tcp_hdr_offset = VLAN_HLEN; first_desc->ip_hdr_offset = VLAN_HLEN; /* Only in case of TSO on vlan device */ flags |= FLAGS_VLAN_TAGGED; + + /* Create a TSO vlan header template for firmware */ + + hwdesc = &tx_ring->desc_head[producer]; + tx_ring->cmd_buf_arr[producer].skb = NULL; + + copy_len = min((int)sizeof(struct cmd_desc_type0) - + offset, hdr_len + VLAN_HLEN); + + vh = (struct vlan_ethhdr *)((char *) hwdesc + 2); + skb_copy_from_linear_data(skb, vh, 12); + vh->h_vlan_proto = htons(ETH_P_8021Q); + vh->h_vlan_TCI = htons(vlan_tci); + + skb_copy_from_linear_data_offset(skb, 12, + (char *)vh + 16, copy_len - 16); + + copied = copy_len - VLAN_HLEN; + offset = 0; + + producer = get_next_index(producer, tx_ring->num_desc); } - opcode = (protocol == cpu_to_be16(ETH_P_IPV6)) ? - TX_TCP_LSO6 : TX_TCP_LSO; - tso = 1; + while (copied < hdr_len) { + + copy_len = min((int)sizeof(struct cmd_desc_type0) - + offset, (hdr_len - copied)); + + hwdesc = &tx_ring->desc_head[producer]; + tx_ring->cmd_buf_arr[producer].skb = NULL; + + skb_copy_from_linear_data_offset(skb, copied, + (char *) hwdesc + offset, copy_len); + + copied += copy_len; + offset = 0; + + producer = get_next_index(producer, tx_ring->num_desc); + } + + tx_ring->producer = producer; + smp_mb(); + adapter->stats.lso_frames++; } else if (skb->ip_summed == CHECKSUM_PARTIAL) { u8 l4proto; - if (protocol == cpu_to_be16(ETH_P_IP)) { + if (protocol == ETH_P_IP) { l4proto = ip_hdr(skb)->protocol; if (l4proto == IPPROTO_TCP) opcode = TX_TCP_PKT; else if (l4proto == IPPROTO_UDP) opcode = TX_UDP_PKT; - } else if (protocol == cpu_to_be16(ETH_P_IPV6)) { + } else if (protocol == ETH_P_IPV6) { l4proto = ipv6_hdr(skb)->nexthdr; if (l4proto == IPPROTO_TCP) @@ -1981,63 +2046,11 @@ qlcnic_tso_check(struct net_device *netdev, opcode = TX_UDPV6_PKT; } } - first_desc->tcp_hdr_offset += skb_transport_offset(skb); first_desc->ip_hdr_offset += skb_network_offset(skb); qlcnic_set_tx_flags_opcode(first_desc, flags, opcode); - if (!tso) - return; - - /* For LSO, we need to copy the MAC/IP/TCP headers into - * the descriptor ring - */ - copied = 0; - offset = 2; - - if (vlan_oob) { - /* Create a TSO vlan header template for firmware */ - - hwdesc = &tx_ring->desc_head[producer]; - tx_ring->cmd_buf_arr[producer].skb = NULL; - - copy_len = min((int)sizeof(struct cmd_desc_type0) - offset, - hdr_len + VLAN_HLEN); - - vh = (struct vlan_ethhdr *)((char *)hwdesc + 2); - skb_copy_from_linear_data(skb, vh, 12); - vh->h_vlan_proto = htons(ETH_P_8021Q); - vh->h_vlan_TCI = (__be16)swab16((u16)first_desc->vlan_TCI); - - skb_copy_from_linear_data_offset(skb, 12, - (char *)vh + 16, copy_len - 16); - - copied = copy_len - VLAN_HLEN; - offset = 0; - - producer = get_next_index(producer, tx_ring->num_desc); - } - - while (copied < hdr_len) { - - copy_len = min((int)sizeof(struct cmd_desc_type0) - offset, - (hdr_len - copied)); - - hwdesc = &tx_ring->desc_head[producer]; - tx_ring->cmd_buf_arr[producer].skb = NULL; - - skb_copy_from_linear_data_offset(skb, copied, - (char *)hwdesc + offset, copy_len); - - copied += copy_len; - offset = 0; - - producer = get_next_index(producer, tx_ring->num_desc); - } - - tx_ring->producer = producer; - barrier(); - adapter->stats.lso_frames++; + return 0; } static int @@ -2088,39 +2101,21 @@ out_err: return -ENOMEM; } -static int -qlcnic_check_tx_tagging(struct qlcnic_adapter *adapter, - struct sk_buff *skb, - struct cmd_desc_type0 *first_desc) +static void +qlcnic_unmap_buffers(struct pci_dev *pdev, struct sk_buff *skb, + struct qlcnic_cmd_buffer *pbuf) { - u8 opcode = 0; - u16 flags = 0; - __be16 protocol = skb->protocol; - struct vlan_ethhdr *vh; + struct qlcnic_skb_frag *nf = &pbuf->frag_array[0]; + int nr_frags = skb_shinfo(skb)->nr_frags; + int i; - if (protocol == cpu_to_be16(ETH_P_8021Q)) { - vh = (struct vlan_ethhdr *)skb->data; - protocol = vh->h_vlan_encapsulated_proto; - flags = FLAGS_VLAN_TAGGED; - qlcnic_set_tx_vlan_tci(first_desc, ntohs(vh->h_vlan_TCI)); - } else if (vlan_tx_tag_present(skb)) { - flags = FLAGS_VLAN_OOB; - qlcnic_set_tx_vlan_tci(first_desc, vlan_tx_tag_get(skb)); + for (i = 0; i < nr_frags; i++) { + nf = &pbuf->frag_array[i+1]; + pci_unmap_page(pdev, nf->dma, nf->length, PCI_DMA_TODEVICE); } - if (unlikely(adapter->pvid)) { - if (first_desc->vlan_TCI && - !(adapter->flags & QLCNIC_TAGGING_ENABLED)) - return -EIO; - if (first_desc->vlan_TCI && - (adapter->flags & QLCNIC_TAGGING_ENABLED)) - goto set_flags; - flags = FLAGS_VLAN_OOB; - qlcnic_set_tx_vlan_tci(first_desc, adapter->pvid); - } -set_flags: - qlcnic_set_tx_flags_opcode(first_desc, flags, opcode); - return 0; + nf = &pbuf->frag_array[0]; + pci_unmap_single(pdev, nf->dma, skb_headlen(skb), PCI_DMA_TODEVICE); } static inline void @@ -2144,7 +2139,7 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev) int i, k; u32 producer; - int frag_count, no_of_desc; + int frag_count; u32 num_txd = tx_ring->num_desc; if (!test_bit(__QLCNIC_DEV_UP, &adapter->state)) { @@ -2161,12 +2156,8 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev) frag_count = skb_shinfo(skb)->nr_frags + 1; - /* 4 fragments per cmd des */ - no_of_desc = (frag_count + 3) >> 2; - if (unlikely(qlcnic_tx_avail(tx_ring) <= TX_STOP_THRESH)) { netif_stop_queue(netdev); - smp_mb(); if (qlcnic_tx_avail(tx_ring) > TX_STOP_THRESH) netif_start_queue(netdev); else { @@ -2183,9 +2174,6 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev) first_desc = hwdesc = &tx_ring->desc_head[producer]; qlcnic_clear_cmddesc((u64 *)hwdesc); - if (qlcnic_check_tx_tagging(adapter, skb, first_desc)) - goto drop_packet; - if (qlcnic_map_tx_skb(pdev, skb, pbuf)) { adapter->stats.tx_dma_map_error++; goto drop_packet; @@ -2229,8 +2217,10 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev) } tx_ring->producer = get_next_index(producer, num_txd); + smp_mb(); - qlcnic_tso_check(netdev, tx_ring, first_desc, skb); + if (unlikely(qlcnic_tx_pkt(adapter, first_desc, skb))) + goto unwind_buff; if (qlcnic_mac_learn) qlcnic_send_filter(adapter, tx_ring, first_desc, skb); @@ -2242,6 +2232,8 @@ qlcnic_xmit_frame(struct sk_buff *skb, struct net_device *netdev) return NETDEV_TX_OK; +unwind_buff: + qlcnic_unmap_buffers(pdev, skb, pbuf); drop_packet: adapter->stats.txdropped++; dev_kfree_skb_any(skb);