These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / net / ethernet / hisilicon / hns / hns_enet.c
diff --git a/kernel/drivers/net/ethernet/hisilicon/hns/hns_enet.c b/kernel/drivers/net/ethernet/hisilicon/hns/hns_enet.c
new file mode 100644 (file)
index 0000000..08cef0d
--- /dev/null
@@ -0,0 +1,1642 @@
+/*
+ * Copyright (c) 2014-2015 Hisilicon Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#include <linux/clk.h>
+#include <linux/cpumask.h>
+#include <linux/etherdevice.h>
+#include <linux/if_vlan.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/module.h>
+#include <linux/phy.h>
+#include <linux/platform_device.h>
+#include <linux/skbuff.h>
+
+#include "hnae.h"
+#include "hns_enet.h"
+
+#define NIC_MAX_Q_PER_VF 16
+#define HNS_NIC_TX_TIMEOUT (5 * HZ)
+
+#define SERVICE_TIMER_HZ (1 * HZ)
+
+#define NIC_TX_CLEAN_MAX_NUM 256
+#define NIC_RX_CLEAN_MAX_NUM 64
+
+#define RCB_IRQ_NOT_INITED 0
+#define RCB_IRQ_INITED 1
+
+static void fill_desc(struct hnae_ring *ring, void *priv,
+                     int size, dma_addr_t dma, int frag_end,
+                     int buf_num, enum hns_desc_type type)
+{
+       struct hnae_desc *desc = &ring->desc[ring->next_to_use];
+       struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
+       struct sk_buff *skb;
+       __be16 protocol;
+       u32 ip_offset;
+       u32 asid_bufnum_pid = 0;
+       u32 flag_ipoffset = 0;
+
+       desc_cb->priv = priv;
+       desc_cb->length = size;
+       desc_cb->dma = dma;
+       desc_cb->type = type;
+
+       desc->addr = cpu_to_le64(dma);
+       desc->tx.send_size = cpu_to_le16((u16)size);
+
+       /*config bd buffer end */
+       flag_ipoffset |= 1 << HNS_TXD_VLD_B;
+
+       asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
+
+       if (type == DESC_TYPE_SKB) {
+               skb = (struct sk_buff *)priv;
+
+               if (skb->ip_summed == CHECKSUM_PARTIAL) {
+                       protocol = skb->protocol;
+                       ip_offset = ETH_HLEN;
+
+                       /*if it is a SW VLAN check the next protocol*/
+                       if (protocol == htons(ETH_P_8021Q)) {
+                               ip_offset += VLAN_HLEN;
+                               protocol = vlan_get_protocol(skb);
+                               skb->protocol = protocol;
+                       }
+
+                       if (skb->protocol == htons(ETH_P_IP)) {
+                               flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
+                               /* check for tcp/udp header */
+                               flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
+
+                       } else if (skb->protocol == htons(ETH_P_IPV6)) {
+                               /* ipv6 has not l3 cs, check for L4 header */
+                               flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
+                       }
+
+                       flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
+               }
+       }
+
+       flag_ipoffset |= frag_end << HNS_TXD_FE_B;
+
+       desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
+       desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
+
+       ring_ptr_move_fw(ring, next_to_use);
+}
+
+static void unfill_desc(struct hnae_ring *ring)
+{
+       ring_ptr_move_bw(ring, next_to_use);
+}
+
+int hns_nic_net_xmit_hw(struct net_device *ndev,
+                       struct sk_buff *skb,
+                       struct hns_nic_ring_data *ring_data)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct device *dev = priv->dev;
+       struct hnae_ring *ring = ring_data->ring;
+       struct netdev_queue *dev_queue;
+       struct skb_frag_struct *frag;
+       int buf_num;
+       dma_addr_t dma;
+       int size, next_to_use;
+       int i, j;
+       struct sk_buff *new_skb;
+
+       assert(ring->max_desc_num_per_pkt <= ring->desc_num);
+
+       /* no. of segments (plus a header) */
+       buf_num = skb_shinfo(skb)->nr_frags + 1;
+
+       if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
+               if (ring_space(ring) < 1) {
+                       ring->stats.tx_busy++;
+                       goto out_net_tx_busy;
+               }
+
+               new_skb = skb_copy(skb, GFP_ATOMIC);
+               if (!new_skb) {
+                       ring->stats.sw_err_cnt++;
+                       netdev_err(ndev, "no memory to xmit!\n");
+                       goto out_err_tx_ok;
+               }
+
+               dev_kfree_skb_any(skb);
+               skb = new_skb;
+               buf_num = 1;
+               assert(skb_shinfo(skb)->nr_frags == 1);
+       } else if (buf_num > ring_space(ring)) {
+               ring->stats.tx_busy++;
+               goto out_net_tx_busy;
+       }
+       next_to_use = ring->next_to_use;
+
+       /* fill the first part */
+       size = skb_headlen(skb);
+       dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
+       if (dma_mapping_error(dev, dma)) {
+               netdev_err(ndev, "TX head DMA map failed\n");
+               ring->stats.sw_err_cnt++;
+               goto out_err_tx_ok;
+       }
+       fill_desc(ring, skb, size, dma, buf_num == 1 ? 1 : 0, buf_num,
+                 DESC_TYPE_SKB);
+
+       /* fill the fragments */
+       for (i = 1; i < buf_num; i++) {
+               frag = &skb_shinfo(skb)->frags[i - 1];
+               size = skb_frag_size(frag);
+               dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
+               if (dma_mapping_error(dev, dma)) {
+                       netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
+                       ring->stats.sw_err_cnt++;
+                       goto out_map_frag_fail;
+               }
+               fill_desc(ring, skb_frag_page(frag), size, dma,
+                         buf_num - 1 == i ? 1 : 0, buf_num, DESC_TYPE_PAGE);
+       }
+
+       /*complete translate all packets*/
+       dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
+       netdev_tx_sent_queue(dev_queue, skb->len);
+
+       wmb(); /* commit all data before submit */
+       assert(skb->queue_mapping < priv->ae_handle->q_num);
+       hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
+       ring->stats.tx_pkts++;
+       ring->stats.tx_bytes += skb->len;
+
+       return NETDEV_TX_OK;
+
+out_map_frag_fail:
+
+       for (j = i - 1; j > 0; j--) {
+               unfill_desc(ring);
+               next_to_use = ring->next_to_use;
+               dma_unmap_page(dev, ring->desc_cb[next_to_use].dma,
+                              ring->desc_cb[next_to_use].length,
+                              DMA_TO_DEVICE);
+       }
+
+       unfill_desc(ring);
+       next_to_use = ring->next_to_use;
+       dma_unmap_single(dev, ring->desc_cb[next_to_use].dma,
+                        ring->desc_cb[next_to_use].length, DMA_TO_DEVICE);
+
+out_err_tx_ok:
+
+       dev_kfree_skb_any(skb);
+       return NETDEV_TX_OK;
+
+out_net_tx_busy:
+
+       netif_stop_subqueue(ndev, skb->queue_mapping);
+
+       /* Herbert's original patch had:
+        *  smp_mb__after_netif_stop_queue();
+        * but since that doesn't exist yet, just open code it.
+        */
+       smp_mb();
+       return NETDEV_TX_BUSY;
+}
+
+/**
+ * hns_nic_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
+ * @data: pointer to the start of the headers
+ * @max: total length of section to find headers in
+ *
+ * This function is meant to determine the length of headers that will
+ * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
+ * motivation of doing this is to only perform one pull for IPv4 TCP
+ * packets so that we can do basic things like calculating the gso_size
+ * based on the average data per packet.
+ **/
+static unsigned int hns_nic_get_headlen(unsigned char *data, u32 flag,
+                                       unsigned int max_size)
+{
+       unsigned char *network;
+       u8 hlen;
+
+       /* this should never happen, but better safe than sorry */
+       if (max_size < ETH_HLEN)
+               return max_size;
+
+       /* initialize network frame pointer */
+       network = data;
+
+       /* set first protocol and move network header forward */
+       network += ETH_HLEN;
+
+       /* handle any vlan tag if present */
+       if (hnae_get_field(flag, HNS_RXD_VLAN_M, HNS_RXD_VLAN_S)
+               == HNS_RX_FLAG_VLAN_PRESENT) {
+               if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN))
+                       return max_size;
+
+               network += VLAN_HLEN;
+       }
+
+       /* handle L3 protocols */
+       if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
+               == HNS_RX_FLAG_L3ID_IPV4) {
+               if ((typeof(max_size))(network - data) >
+                   (max_size - sizeof(struct iphdr)))
+                       return max_size;
+
+               /* access ihl as a u8 to avoid unaligned access on ia64 */
+               hlen = (network[0] & 0x0F) << 2;
+
+               /* verify hlen meets minimum size requirements */
+               if (hlen < sizeof(struct iphdr))
+                       return network - data;
+
+               /* record next protocol if header is present */
+       } else if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
+               == HNS_RX_FLAG_L3ID_IPV6) {
+               if ((typeof(max_size))(network - data) >
+                   (max_size - sizeof(struct ipv6hdr)))
+                       return max_size;
+
+               /* record next protocol */
+               hlen = sizeof(struct ipv6hdr);
+       } else {
+               return network - data;
+       }
+
+       /* relocate pointer to start of L4 header */
+       network += hlen;
+
+       /* finally sort out TCP/UDP */
+       if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
+               == HNS_RX_FLAG_L4ID_TCP) {
+               if ((typeof(max_size))(network - data) >
+                   (max_size - sizeof(struct tcphdr)))
+                       return max_size;
+
+               /* access doff as a u8 to avoid unaligned access on ia64 */
+               hlen = (network[12] & 0xF0) >> 2;
+
+               /* verify hlen meets minimum size requirements */
+               if (hlen < sizeof(struct tcphdr))
+                       return network - data;
+
+               network += hlen;
+       } else if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
+               == HNS_RX_FLAG_L4ID_UDP) {
+               if ((typeof(max_size))(network - data) >
+                   (max_size - sizeof(struct udphdr)))
+                       return max_size;
+
+               network += sizeof(struct udphdr);
+       }
+
+       /* If everything has gone correctly network should be the
+        * data section of the packet and will be the end of the header.
+        * If not then it probably represents the end of the last recognized
+        * header.
+        */
+       if ((typeof(max_size))(network - data) < max_size)
+               return network - data;
+       else
+               return max_size;
+}
+
+static void
+hns_nic_reuse_page(struct hnae_desc_cb *desc_cb, int tsize, int last_offset)
+{
+        /* avoid re-using remote pages,flag default unreuse */
+       if (likely(page_to_nid(desc_cb->priv) == numa_node_id())) {
+               /* move offset up to the next cache line */
+               desc_cb->page_offset += tsize;
+
+               if (desc_cb->page_offset <= last_offset) {
+                       desc_cb->reuse_flag = 1;
+                       /* bump ref count on page before it is given*/
+                       get_page(desc_cb->priv);
+               }
+       }
+}
+
+static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
+                              struct sk_buff **out_skb, int *out_bnum)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       struct net_device *ndev = ring_data->napi.dev;
+       struct sk_buff *skb;
+       struct hnae_desc *desc;
+       struct hnae_desc_cb *desc_cb;
+       unsigned char *va;
+       int bnum, length, size, i, truesize, last_offset;
+       int pull_len;
+       u32 bnum_flag;
+
+       last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
+       desc = &ring->desc[ring->next_to_clean];
+       desc_cb = &ring->desc_cb[ring->next_to_clean];
+       length = le16_to_cpu(desc->rx.pkt_len);
+       bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
+       bnum = hnae_get_field(bnum_flag, HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
+       *out_bnum = bnum;
+       va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
+
+       skb = *out_skb = napi_alloc_skb(&ring_data->napi, HNS_RX_HEAD_SIZE);
+       if (unlikely(!skb)) {
+               netdev_err(ndev, "alloc rx skb fail\n");
+               ring->stats.sw_err_cnt++;
+               return -ENOMEM;
+       }
+
+       if (length <= HNS_RX_HEAD_SIZE) {
+               memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
+
+               /* we can reuse buffer as-is, just make sure it is local */
+               if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
+                       desc_cb->reuse_flag = 1;
+               else /* this page cannot be reused so discard it */
+                       put_page(desc_cb->priv);
+
+               ring_ptr_move_fw(ring, next_to_clean);
+
+               if (unlikely(bnum != 1)) { /* check err*/
+                       *out_bnum = 1;
+                       goto out_bnum_err;
+               }
+       } else {
+               ring->stats.seg_pkt_cnt++;
+
+               pull_len = hns_nic_get_headlen(va, bnum_flag, HNS_RX_HEAD_SIZE);
+               memcpy(__skb_put(skb, pull_len), va,
+                      ALIGN(pull_len, sizeof(long)));
+
+               size = le16_to_cpu(desc->rx.size);
+               truesize = ALIGN(size, L1_CACHE_BYTES);
+               skb_add_rx_frag(skb, 0, desc_cb->priv,
+                               desc_cb->page_offset + pull_len,
+                               size - pull_len, truesize - pull_len);
+
+               hns_nic_reuse_page(desc_cb, truesize, last_offset);
+               ring_ptr_move_fw(ring, next_to_clean);
+
+               if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
+                       *out_bnum = 1;
+                       goto out_bnum_err;
+               }
+               for (i = 1; i < bnum; i++) {
+                       desc = &ring->desc[ring->next_to_clean];
+                       desc_cb = &ring->desc_cb[ring->next_to_clean];
+                       size = le16_to_cpu(desc->rx.size);
+                       truesize = ALIGN(size, L1_CACHE_BYTES);
+                       skb_add_rx_frag(skb, i, desc_cb->priv,
+                                       desc_cb->page_offset,
+                                       size, truesize);
+
+                       hns_nic_reuse_page(desc_cb, truesize, last_offset);
+                       ring_ptr_move_fw(ring, next_to_clean);
+               }
+       }
+
+       /* check except process, free skb and jump the desc */
+       if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
+out_bnum_err:
+               *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
+               netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
+                          bnum, ring->max_desc_num_per_pkt,
+                          length, (int)MAX_SKB_FRAGS,
+                          ((u64 *)desc)[0], ((u64 *)desc)[1]);
+               ring->stats.err_bd_num++;
+               dev_kfree_skb_any(skb);
+               return -EDOM;
+       }
+
+       bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
+
+       if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
+               netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
+                          ((u64 *)desc)[0], ((u64 *)desc)[1]);
+               ring->stats.non_vld_descs++;
+               dev_kfree_skb_any(skb);
+               return -EINVAL;
+       }
+
+       if (unlikely((!desc->rx.pkt_len) ||
+                    hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
+               ring->stats.err_pkt_len++;
+               dev_kfree_skb_any(skb);
+               return -EFAULT;
+       }
+
+       if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
+               ring->stats.l2_err++;
+               dev_kfree_skb_any(skb);
+               return -EFAULT;
+       }
+
+       ring->stats.rx_pkts++;
+       ring->stats.rx_bytes += skb->len;
+
+       if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L3E_B) ||
+                    hnae_get_bit(bnum_flag, HNS_RXD_L4E_B))) {
+               ring->stats.l3l4_csum_err++;
+               return 0;
+       }
+
+       skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+       return 0;
+}
+
+static void
+hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
+{
+       int i, ret;
+       struct hnae_desc_cb res_cbs;
+       struct hnae_desc_cb *desc_cb;
+       struct hnae_ring *ring = ring_data->ring;
+       struct net_device *ndev = ring_data->napi.dev;
+
+       for (i = 0; i < cleand_count; i++) {
+               desc_cb = &ring->desc_cb[ring->next_to_use];
+               if (desc_cb->reuse_flag) {
+                       ring->stats.reuse_pg_cnt++;
+                       hnae_reuse_buffer(ring, ring->next_to_use);
+               } else {
+                       ret = hnae_reserve_buffer_map(ring, &res_cbs);
+                       if (ret) {
+                               ring->stats.sw_err_cnt++;
+                               netdev_err(ndev, "hnae reserve buffer map failed.\n");
+                               break;
+                       }
+                       hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
+               }
+
+               ring_ptr_move_fw(ring, next_to_use);
+       }
+
+       wmb(); /* make all data has been write before submit */
+       writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
+}
+
+/* return error number for error or number of desc left to take
+ */
+static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
+                             struct sk_buff *skb)
+{
+       struct net_device *ndev = ring_data->napi.dev;
+
+       skb->protocol = eth_type_trans(skb, ndev);
+       (void)napi_gro_receive(&ring_data->napi, skb);
+       ndev->last_rx = jiffies;
+}
+
+static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
+                              int budget, void *v)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       struct sk_buff *skb;
+       int num, bnum, ex_num;
+#define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
+       int recv_pkts, recv_bds, clean_count, err;
+
+       num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
+       rmb(); /* make sure num taken effect before the other data is touched */
+
+       recv_pkts = 0, recv_bds = 0, clean_count = 0;
+recv:
+       while (recv_pkts < budget && recv_bds < num) {
+               /* reuse or realloc buffers*/
+               if (clean_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
+                       hns_nic_alloc_rx_buffers(ring_data, clean_count);
+                       clean_count = 0;
+               }
+
+               /* poll one pkg*/
+               err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
+               if (unlikely(!skb)) /* this fault cannot be repaired */
+                       break;
+
+               recv_bds += bnum;
+               clean_count += bnum;
+               if (unlikely(err)) {  /* do jump the err */
+                       recv_pkts++;
+                       continue;
+               }
+
+               /* do update ip stack process*/
+               ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
+                                                       ring_data, skb);
+               recv_pkts++;
+       }
+
+       /* make all data has been write before submit */
+       if (clean_count > 0) {
+               hns_nic_alloc_rx_buffers(ring_data, clean_count);
+               clean_count = 0;
+       }
+
+       if (recv_pkts < budget) {
+               ex_num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
+               rmb(); /*complete read rx ring bd number*/
+               if (ex_num > 0) {
+                       num += ex_num;
+                       goto recv;
+               }
+       }
+
+       return recv_pkts;
+}
+
+static void hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       int num = 0;
+
+       /* for hardware bug fixed */
+       num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
+
+       if (num > 0) {
+               ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
+                       ring_data->ring, 1);
+
+               napi_schedule(&ring_data->napi);
+       }
+}
+
+static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
+                                           int *bytes, int *pkts)
+{
+       struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
+
+       (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
+       (*bytes) += desc_cb->length;
+       /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
+       hnae_free_buffer_detach(ring, ring->next_to_clean);
+
+       ring_ptr_move_fw(ring, next_to_clean);
+}
+
+static int is_valid_clean_head(struct hnae_ring *ring, int h)
+{
+       int u = ring->next_to_use;
+       int c = ring->next_to_clean;
+
+       if (unlikely(h > ring->desc_num))
+               return 0;
+
+       assert(u > 0 && u < ring->desc_num);
+       assert(c > 0 && c < ring->desc_num);
+       assert(u != c && h != c); /* must be checked before call this func */
+
+       return u > c ? (h > c && h <= u) : (h > c || h <= u);
+}
+
+/* netif_tx_lock will turn down the performance, set only when necessary */
+#ifdef CONFIG_NET_POLL_CONTROLLER
+#define NETIF_TX_LOCK(ndev) netif_tx_lock(ndev)
+#define NETIF_TX_UNLOCK(ndev) netif_tx_unlock(ndev)
+#else
+#define NETIF_TX_LOCK(ndev)
+#define NETIF_TX_UNLOCK(ndev)
+#endif
+/* reclaim all desc in one budget
+ * return error or number of desc left
+ */
+static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
+                              int budget, void *v)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       struct net_device *ndev = ring_data->napi.dev;
+       struct netdev_queue *dev_queue;
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       int head;
+       int bytes, pkts;
+
+       NETIF_TX_LOCK(ndev);
+
+       head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
+       rmb(); /* make sure head is ready before touch any data */
+
+       if (is_ring_empty(ring) || head == ring->next_to_clean) {
+               NETIF_TX_UNLOCK(ndev);
+               return 0; /* no data to poll */
+       }
+
+       if (!is_valid_clean_head(ring, head)) {
+               netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
+                          ring->next_to_use, ring->next_to_clean);
+               ring->stats.io_err_cnt++;
+               NETIF_TX_UNLOCK(ndev);
+               return -EIO;
+       }
+
+       bytes = 0;
+       pkts = 0;
+       while (head != ring->next_to_clean)
+               hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
+
+       NETIF_TX_UNLOCK(ndev);
+
+       dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
+       netdev_tx_completed_queue(dev_queue, pkts, bytes);
+
+       if (unlikely(pkts && netif_carrier_ok(ndev) &&
+                    (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
+               /* Make sure that anybody stopping the queue after this
+                * sees the new next_to_clean.
+                */
+               smp_mb();
+               if (netif_tx_queue_stopped(dev_queue) &&
+                   !test_bit(NIC_STATE_DOWN, &priv->state)) {
+                       netif_tx_wake_queue(dev_queue);
+                       ring->stats.restart_queue++;
+               }
+       }
+       return 0;
+}
+
+static void hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       int head = ring->next_to_clean;
+
+       /* for hardware bug fixed */
+       head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
+
+       if (head != ring->next_to_clean) {
+               ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
+                       ring_data->ring, 1);
+
+               napi_schedule(&ring_data->napi);
+       }
+}
+
+static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
+{
+       struct hnae_ring *ring = ring_data->ring;
+       struct net_device *ndev = ring_data->napi.dev;
+       struct netdev_queue *dev_queue;
+       int head;
+       int bytes, pkts;
+
+       NETIF_TX_LOCK(ndev);
+
+       head = ring->next_to_use; /* ntu :soft setted ring position*/
+       bytes = 0;
+       pkts = 0;
+       while (head != ring->next_to_clean)
+               hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
+
+       NETIF_TX_UNLOCK(ndev);
+
+       dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
+       netdev_tx_reset_queue(dev_queue);
+}
+
+static int hns_nic_common_poll(struct napi_struct *napi, int budget)
+{
+       struct hns_nic_ring_data *ring_data =
+               container_of(napi, struct hns_nic_ring_data, napi);
+       int clean_complete = ring_data->poll_one(
+                               ring_data, budget, ring_data->ex_process);
+
+       if (clean_complete >= 0 && clean_complete < budget) {
+               napi_complete(napi);
+               ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
+                       ring_data->ring, 0);
+
+               ring_data->fini_process(ring_data);
+       }
+
+       return clean_complete;
+}
+
+static irqreturn_t hns_irq_handle(int irq, void *dev)
+{
+       struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
+
+       ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
+               ring_data->ring, 1);
+       napi_schedule(&ring_data->napi);
+
+       return IRQ_HANDLED;
+}
+
+/**
+ *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
+ *@ndev: net device
+ */
+static void hns_nic_adjust_link(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       h->dev->ops->adjust_link(h, ndev->phydev->speed, ndev->phydev->duplex);
+}
+
+/**
+ *hns_nic_init_phy - init phy
+ *@ndev: net device
+ *@h: ae handle
+ * Return 0 on success, negative on failure
+ */
+int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct phy_device *phy_dev = NULL;
+
+       if (!h->phy_node)
+               return 0;
+
+       if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
+               phy_dev = of_phy_connect(ndev, h->phy_node,
+                                        hns_nic_adjust_link, 0, h->phy_if);
+       else
+               phy_dev = of_phy_attach(ndev, h->phy_node, 0, h->phy_if);
+
+       if (unlikely(!phy_dev) || IS_ERR(phy_dev))
+               return !phy_dev ? -ENODEV : PTR_ERR(phy_dev);
+
+       phy_dev->supported &= h->if_support;
+       phy_dev->advertising = phy_dev->supported;
+
+       if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
+               phy_dev->autoneg = false;
+
+       priv->phy = phy_dev;
+
+       return 0;
+}
+
+static int hns_nic_ring_open(struct net_device *netdev, int idx)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       napi_enable(&priv->ring_data[idx].napi);
+
+       enable_irq(priv->ring_data[idx].ring->irq);
+       h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
+
+       return 0;
+}
+
+static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+       struct sockaddr *mac_addr = p;
+       int ret;
+
+       if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
+               return -EADDRNOTAVAIL;
+
+       ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
+       if (ret) {
+               netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
+               return ret;
+       }
+
+       memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
+
+       return 0;
+}
+
+void hns_nic_update_stats(struct net_device *netdev)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       h->dev->ops->update_stats(h, &netdev->stats);
+}
+
+/* set mac addr if it is configed. or leave it to the AE driver */
+static void hns_init_mac_addr(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct device_node *node = priv->dev->of_node;
+       const void *mac_addr_temp;
+
+       mac_addr_temp = of_get_mac_address(node);
+       if (mac_addr_temp && is_valid_ether_addr(mac_addr_temp)) {
+               memcpy(ndev->dev_addr, mac_addr_temp, ndev->addr_len);
+       } else {
+               eth_hw_addr_random(ndev);
+               dev_warn(priv->dev, "No valid mac, use random mac %pM",
+                        ndev->dev_addr);
+       }
+}
+
+static void hns_nic_ring_close(struct net_device *netdev, int idx)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
+       disable_irq(priv->ring_data[idx].ring->irq);
+
+       napi_disable(&priv->ring_data[idx].napi);
+}
+
+static int hns_nic_init_irq(struct hns_nic_priv *priv)
+{
+       struct hnae_handle *h = priv->ae_handle;
+       struct hns_nic_ring_data *rd;
+       int i;
+       int ret;
+       int cpu;
+       cpumask_t mask;
+
+       for (i = 0; i < h->q_num * 2; i++) {
+               rd = &priv->ring_data[i];
+
+               if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
+                       break;
+
+               snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
+                        "%s-%s%d", priv->netdev->name,
+                        (i < h->q_num ? "tx" : "rx"), rd->queue_index);
+
+               rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
+
+               ret = request_irq(rd->ring->irq,
+                                 hns_irq_handle, 0, rd->ring->ring_name, rd);
+               if (ret) {
+                       netdev_err(priv->netdev, "request irq(%d) fail\n",
+                                  rd->ring->irq);
+                       return ret;
+               }
+               disable_irq(rd->ring->irq);
+               rd->ring->irq_init_flag = RCB_IRQ_INITED;
+
+               /*set cpu affinity*/
+               if (cpu_online(rd->queue_index)) {
+                       cpumask_clear(&mask);
+                       cpu = rd->queue_index;
+                       cpumask_set_cpu(cpu, &mask);
+                       irq_set_affinity_hint(rd->ring->irq, &mask);
+               }
+       }
+
+       return 0;
+}
+
+static int hns_nic_net_up(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+       int i, j, k;
+       int ret;
+
+       ret = hns_nic_init_irq(priv);
+       if (ret != 0) {
+               netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
+               return ret;
+       }
+
+       for (i = 0; i < h->q_num * 2; i++) {
+               ret = hns_nic_ring_open(ndev, i);
+               if (ret)
+                       goto out_has_some_queues;
+       }
+
+       for (k = 0; k < h->q_num; k++)
+               h->dev->ops->toggle_queue_status(h->qs[k], 1);
+
+       ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
+       if (ret)
+               goto out_set_mac_addr_err;
+
+       ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
+       if (ret)
+               goto out_start_err;
+
+       if (priv->phy)
+               phy_start(priv->phy);
+
+       clear_bit(NIC_STATE_DOWN, &priv->state);
+       (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
+
+       return 0;
+
+out_start_err:
+       netif_stop_queue(ndev);
+out_set_mac_addr_err:
+       for (k = 0; k < h->q_num; k++)
+               h->dev->ops->toggle_queue_status(h->qs[k], 0);
+out_has_some_queues:
+       for (j = i - 1; j >= 0; j--)
+               hns_nic_ring_close(ndev, j);
+
+       set_bit(NIC_STATE_DOWN, &priv->state);
+
+       return ret;
+}
+
+static void hns_nic_net_down(struct net_device *ndev)
+{
+       int i;
+       struct hnae_ae_ops *ops;
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+
+       if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
+               return;
+
+       (void)del_timer_sync(&priv->service_timer);
+       netif_tx_stop_all_queues(ndev);
+       netif_carrier_off(ndev);
+       netif_tx_disable(ndev);
+       priv->link = 0;
+
+       if (priv->phy)
+               phy_stop(priv->phy);
+
+       ops = priv->ae_handle->dev->ops;
+
+       if (ops->stop)
+               ops->stop(priv->ae_handle);
+
+       netif_tx_stop_all_queues(ndev);
+
+       for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
+               hns_nic_ring_close(ndev, i);
+               hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
+
+               /* clean tx buffers*/
+               hns_nic_tx_clr_all_bufs(priv->ring_data + i);
+       }
+}
+
+void hns_nic_net_reset(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *handle = priv->ae_handle;
+
+       while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
+               usleep_range(1000, 2000);
+
+       (void)hnae_reinit_handle(handle);
+
+       clear_bit(NIC_STATE_RESETTING, &priv->state);
+}
+
+void hns_nic_net_reinit(struct net_device *netdev)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+
+       priv->netdev->trans_start = jiffies;
+       while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
+               usleep_range(1000, 2000);
+
+       hns_nic_net_down(netdev);
+       hns_nic_net_reset(netdev);
+       (void)hns_nic_net_up(netdev);
+       clear_bit(NIC_STATE_REINITING, &priv->state);
+}
+
+static int hns_nic_net_open(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+       int ret;
+
+       if (test_bit(NIC_STATE_TESTING, &priv->state))
+               return -EBUSY;
+
+       priv->link = 0;
+       netif_carrier_off(ndev);
+
+       ret = netif_set_real_num_tx_queues(ndev, h->q_num);
+       if (ret < 0) {
+               netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
+                          ret);
+               return ret;
+       }
+
+       ret = netif_set_real_num_rx_queues(ndev, h->q_num);
+       if (ret < 0) {
+               netdev_err(ndev,
+                          "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
+               return ret;
+       }
+
+       ret = hns_nic_net_up(ndev);
+       if (ret) {
+               netdev_err(ndev,
+                          "hns net up fail, ret=%d!\n", ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+static int hns_nic_net_stop(struct net_device *ndev)
+{
+       hns_nic_net_down(ndev);
+
+       return 0;
+}
+
+static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
+static void hns_nic_net_timeout(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+
+       hns_tx_timeout_reset(priv);
+}
+
+static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
+                           int cmd)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+       struct phy_device *phy_dev = priv->phy;
+
+       if (!netif_running(netdev))
+               return -EINVAL;
+
+       if (!phy_dev)
+               return -ENOTSUPP;
+
+       return phy_mii_ioctl(phy_dev, ifr, cmd);
+}
+
+/* use only for netconsole to poll with the device without interrupt */
+#ifdef CONFIG_NET_POLL_CONTROLLER
+void hns_nic_poll_controller(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       unsigned long flags;
+       int i;
+
+       local_irq_save(flags);
+       for (i = 0; i < priv->ae_handle->q_num * 2; i++)
+               napi_schedule(&priv->ring_data[i].napi);
+       local_irq_restore(flags);
+}
+#endif
+
+static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
+                                   struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       int ret;
+
+       assert(skb->queue_mapping < ndev->ae_handle->q_num);
+       ret = hns_nic_net_xmit_hw(ndev, skb,
+                                 &tx_ring_data(priv, skb->queue_mapping));
+       if (ret == NETDEV_TX_OK) {
+               ndev->trans_start = jiffies;
+               ndev->stats.tx_bytes += skb->len;
+               ndev->stats.tx_packets++;
+       }
+       return (netdev_tx_t)ret;
+}
+
+static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+       int ret;
+
+       /* MTU < 68 is an error and causes problems on some kernels */
+       if (new_mtu < 68)
+               return -EINVAL;
+
+       if (!h->dev->ops->set_mtu)
+               return -ENOTSUPP;
+
+       if (netif_running(ndev)) {
+               (void)hns_nic_net_stop(ndev);
+               msleep(100);
+
+               ret = h->dev->ops->set_mtu(h, new_mtu);
+               if (ret)
+                       netdev_err(ndev, "set mtu fail, return value %d\n",
+                                  ret);
+
+               if (hns_nic_net_open(ndev))
+                       netdev_err(ndev, "hns net open fail\n");
+       } else {
+               ret = h->dev->ops->set_mtu(h, new_mtu);
+       }
+
+       if (!ret)
+               ndev->mtu = new_mtu;
+
+       return ret;
+}
+
+/**
+ * nic_set_multicast_list - set mutl mac address
+ * @netdev: net device
+ * @p: mac address
+ *
+ * return void
+ */
+void hns_set_multicast_list(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+       struct netdev_hw_addr *ha = NULL;
+
+       if (!h) {
+               netdev_err(ndev, "hnae handle is null\n");
+               return;
+       }
+
+       if (h->dev->ops->set_mc_addr) {
+               netdev_for_each_mc_addr(ha, ndev)
+                       if (h->dev->ops->set_mc_addr(h, ha->addr))
+                               netdev_err(ndev, "set multicast fail\n");
+       }
+}
+
+void hns_nic_set_rx_mode(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       if (h->dev->ops->set_promisc_mode) {
+               if (ndev->flags & IFF_PROMISC)
+                       h->dev->ops->set_promisc_mode(h, 1);
+               else
+                       h->dev->ops->set_promisc_mode(h, 0);
+       }
+
+       hns_set_multicast_list(ndev);
+}
+
+struct rtnl_link_stats64 *hns_nic_get_stats64(struct net_device *ndev,
+                                             struct rtnl_link_stats64 *stats)
+{
+       int idx = 0;
+       u64 tx_bytes = 0;
+       u64 rx_bytes = 0;
+       u64 tx_pkts = 0;
+       u64 rx_pkts = 0;
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h = priv->ae_handle;
+
+       for (idx = 0; idx < h->q_num; idx++) {
+               tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
+               tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
+               rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
+               rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
+       }
+
+       stats->tx_bytes = tx_bytes;
+       stats->tx_packets = tx_pkts;
+       stats->rx_bytes = rx_bytes;
+       stats->rx_packets = rx_pkts;
+
+       stats->rx_errors = ndev->stats.rx_errors;
+       stats->multicast = ndev->stats.multicast;
+       stats->rx_length_errors = ndev->stats.rx_length_errors;
+       stats->rx_crc_errors = ndev->stats.rx_crc_errors;
+       stats->rx_missed_errors = ndev->stats.rx_missed_errors;
+
+       stats->tx_errors = ndev->stats.tx_errors;
+       stats->rx_dropped = ndev->stats.rx_dropped;
+       stats->tx_dropped = ndev->stats.tx_dropped;
+       stats->collisions = ndev->stats.collisions;
+       stats->rx_over_errors = ndev->stats.rx_over_errors;
+       stats->rx_frame_errors = ndev->stats.rx_frame_errors;
+       stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
+       stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
+       stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
+       stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
+       stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
+       stats->tx_window_errors = ndev->stats.tx_window_errors;
+       stats->rx_compressed = ndev->stats.rx_compressed;
+       stats->tx_compressed = ndev->stats.tx_compressed;
+
+       return stats;
+}
+
+static const struct net_device_ops hns_nic_netdev_ops = {
+       .ndo_open = hns_nic_net_open,
+       .ndo_stop = hns_nic_net_stop,
+       .ndo_start_xmit = hns_nic_net_xmit,
+       .ndo_tx_timeout = hns_nic_net_timeout,
+       .ndo_set_mac_address = hns_nic_net_set_mac_address,
+       .ndo_change_mtu = hns_nic_change_mtu,
+       .ndo_do_ioctl = hns_nic_do_ioctl,
+       .ndo_get_stats64 = hns_nic_get_stats64,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller = hns_nic_poll_controller,
+#endif
+       .ndo_set_rx_mode = hns_nic_set_rx_mode,
+};
+
+static void hns_nic_update_link_status(struct net_device *netdev)
+{
+       struct hns_nic_priv *priv = netdev_priv(netdev);
+
+       struct hnae_handle *h = priv->ae_handle;
+       int state = 1;
+
+       if (priv->phy) {
+               if (!genphy_update_link(priv->phy))
+                       state = priv->phy->link;
+               else
+                       state = 0;
+       }
+       state = state && h->dev->ops->get_status(h);
+
+       if (state != priv->link) {
+               if (state) {
+                       netif_carrier_on(netdev);
+                       netif_tx_wake_all_queues(netdev);
+                       netdev_info(netdev, "link up\n");
+               } else {
+                       netif_carrier_off(netdev);
+                       netdev_info(netdev, "link down\n");
+               }
+               priv->link = state;
+       }
+}
+
+/* for dumping key regs*/
+static void hns_nic_dump(struct hns_nic_priv *priv)
+{
+       struct hnae_handle *h = priv->ae_handle;
+       struct hnae_ae_ops *ops = h->dev->ops;
+       u32 *data, reg_num, i;
+
+       if (ops->get_regs_len && ops->get_regs) {
+               reg_num = ops->get_regs_len(priv->ae_handle);
+               reg_num = (reg_num + 3ul) & ~3ul;
+               data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
+               if (data) {
+                       ops->get_regs(priv->ae_handle, data);
+                       for (i = 0; i < reg_num; i += 4)
+                               pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
+                                       i, data[i], data[i + 1],
+                                       data[i + 2], data[i + 3]);
+                       kfree(data);
+               }
+       }
+
+       for (i = 0; i < h->q_num; i++) {
+               pr_info("tx_queue%d_next_to_clean:%d\n",
+                       i, h->qs[i]->tx_ring.next_to_clean);
+               pr_info("tx_queue%d_next_to_use:%d\n",
+                       i, h->qs[i]->tx_ring.next_to_use);
+               pr_info("rx_queue%d_next_to_clean:%d\n",
+                       i, h->qs[i]->rx_ring.next_to_clean);
+               pr_info("rx_queue%d_next_to_use:%d\n",
+                       i, h->qs[i]->rx_ring.next_to_use);
+       }
+}
+
+/* for resetting suntask*/
+static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
+{
+       enum hnae_port_type type = priv->ae_handle->port_type;
+
+       if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
+               return;
+       clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
+
+       /* If we're already down, removing or resetting, just bail */
+       if (test_bit(NIC_STATE_DOWN, &priv->state) ||
+           test_bit(NIC_STATE_REMOVING, &priv->state) ||
+           test_bit(NIC_STATE_RESETTING, &priv->state))
+               return;
+
+       hns_nic_dump(priv);
+       netdev_info(priv->netdev, "Reset %s port\n",
+                   (type == HNAE_PORT_DEBUG ? "debug" : "business"));
+
+       rtnl_lock();
+       /* put off any impending NetWatchDogTimeout */
+       priv->netdev->trans_start = jiffies;
+
+       if (type == HNAE_PORT_DEBUG)
+               hns_nic_net_reinit(priv->netdev);
+       rtnl_unlock();
+}
+
+/* for doing service complete*/
+static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
+{
+       assert(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
+
+       smp_mb__before_atomic();
+       clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
+}
+
+static void hns_nic_service_task(struct work_struct *work)
+{
+       struct hns_nic_priv *priv
+               = container_of(work, struct hns_nic_priv, service_task);
+       struct hnae_handle *h = priv->ae_handle;
+
+       hns_nic_update_link_status(priv->netdev);
+       h->dev->ops->update_led_status(h);
+       hns_nic_update_stats(priv->netdev);
+
+       hns_nic_reset_subtask(priv);
+       hns_nic_service_event_complete(priv);
+}
+
+static void hns_nic_task_schedule(struct hns_nic_priv *priv)
+{
+       if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
+           !test_bit(NIC_STATE_REMOVING, &priv->state) &&
+           !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
+               (void)schedule_work(&priv->service_task);
+}
+
+static void hns_nic_service_timer(unsigned long data)
+{
+       struct hns_nic_priv *priv = (struct hns_nic_priv *)data;
+
+       (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
+
+       hns_nic_task_schedule(priv);
+}
+
+/**
+ * hns_tx_timeout_reset - initiate reset due to Tx timeout
+ * @priv: driver private struct
+ **/
+static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
+{
+       /* Do the reset outside of interrupt context */
+       if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
+               set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
+               netdev_warn(priv->netdev,
+                           "initiating reset due to tx timeout(%llu,0x%lx)\n",
+                           priv->tx_timeout_count, priv->state);
+               priv->tx_timeout_count++;
+               hns_nic_task_schedule(priv);
+       }
+}
+
+static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
+{
+       struct hnae_handle *h = priv->ae_handle;
+       struct hns_nic_ring_data *rd;
+       int i;
+
+       if (h->q_num > NIC_MAX_Q_PER_VF) {
+               netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
+               return -EINVAL;
+       }
+
+       priv->ring_data = kzalloc(h->q_num * sizeof(*priv->ring_data) * 2,
+                                 GFP_KERNEL);
+       if (!priv->ring_data)
+               return -ENOMEM;
+
+       for (i = 0; i < h->q_num; i++) {
+               rd = &priv->ring_data[i];
+               rd->queue_index = i;
+               rd->ring = &h->qs[i]->tx_ring;
+               rd->poll_one = hns_nic_tx_poll_one;
+               rd->fini_process = hns_nic_tx_fini_pro;
+
+               netif_napi_add(priv->netdev, &rd->napi,
+                              hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
+               rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
+       }
+       for (i = h->q_num; i < h->q_num * 2; i++) {
+               rd = &priv->ring_data[i];
+               rd->queue_index = i - h->q_num;
+               rd->ring = &h->qs[i - h->q_num]->rx_ring;
+               rd->poll_one = hns_nic_rx_poll_one;
+               rd->ex_process = hns_nic_rx_up_pro;
+               rd->fini_process = hns_nic_rx_fini_pro;
+
+               netif_napi_add(priv->netdev, &rd->napi,
+                              hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
+               rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
+       }
+
+       return 0;
+}
+
+static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
+{
+       struct hnae_handle *h = priv->ae_handle;
+       int i;
+
+       for (i = 0; i < h->q_num * 2; i++) {
+               netif_napi_del(&priv->ring_data[i].napi);
+               if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
+                       irq_set_affinity_hint(priv->ring_data[i].ring->irq,
+                                             NULL);
+                       free_irq(priv->ring_data[i].ring->irq,
+                                &priv->ring_data[i]);
+               }
+
+               priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
+       }
+       kfree(priv->ring_data);
+}
+
+static int hns_nic_try_get_ae(struct net_device *ndev)
+{
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+       struct hnae_handle *h;
+       int ret;
+
+       h = hnae_get_handle(&priv->netdev->dev,
+                           priv->ae_name, priv->port_id, NULL);
+       if (IS_ERR_OR_NULL(h)) {
+               ret = PTR_ERR(h);
+               dev_dbg(priv->dev, "has not handle, register notifier!\n");
+               goto out;
+       }
+       priv->ae_handle = h;
+
+       ret = hns_nic_init_phy(ndev, h);
+       if (ret) {
+               dev_err(priv->dev, "probe phy device fail!\n");
+               goto out_init_phy;
+       }
+
+       ret = hns_nic_init_ring_data(priv);
+       if (ret) {
+               ret = -ENOMEM;
+               goto out_init_ring_data;
+       }
+
+       ret = register_netdev(ndev);
+       if (ret) {
+               dev_err(priv->dev, "probe register netdev fail!\n");
+               goto out_reg_ndev_fail;
+       }
+       return 0;
+
+out_reg_ndev_fail:
+       hns_nic_uninit_ring_data(priv);
+       priv->ring_data = NULL;
+out_init_phy:
+out_init_ring_data:
+       hnae_put_handle(priv->ae_handle);
+       priv->ae_handle = NULL;
+out:
+       return ret;
+}
+
+static int hns_nic_notifier_action(struct notifier_block *nb,
+                                  unsigned long action, void *data)
+{
+       struct hns_nic_priv *priv =
+               container_of(nb, struct hns_nic_priv, notifier_block);
+
+       assert(action == HNAE_AE_REGISTER);
+
+       if (!hns_nic_try_get_ae(priv->netdev)) {
+               hnae_unregister_notifier(&priv->notifier_block);
+               priv->notifier_block.notifier_call = NULL;
+       }
+       return 0;
+}
+
+static int hns_nic_dev_probe(struct platform_device *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct net_device *ndev;
+       struct hns_nic_priv *priv;
+       struct device_node *node = dev->of_node;
+       int ret;
+
+       ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
+       if (!ndev)
+               return -ENOMEM;
+
+       platform_set_drvdata(pdev, ndev);
+
+       priv = netdev_priv(ndev);
+       priv->dev = dev;
+       priv->netdev = ndev;
+
+       if (of_device_is_compatible(node, "hisilicon,hns-nic-v2"))
+               priv->enet_ver = AE_VERSION_2;
+       else
+               priv->enet_ver = AE_VERSION_1;
+
+       ret = of_property_read_string(node, "ae-name", &priv->ae_name);
+       if (ret)
+               goto out_read_string_fail;
+
+       ret = of_property_read_u32(node, "port-id", &priv->port_id);
+       if (ret)
+               goto out_read_string_fail;
+
+       hns_init_mac_addr(ndev);
+
+       ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
+       ndev->priv_flags |= IFF_UNICAST_FLT;
+       ndev->netdev_ops = &hns_nic_netdev_ops;
+       hns_ethtool_set_ops(ndev);
+       ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
+               NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
+               NETIF_F_GRO;
+       ndev->vlan_features |=
+               NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
+       ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
+
+       SET_NETDEV_DEV(ndev, dev);
+
+       if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
+               dev_dbg(dev, "set mask to 64bit\n");
+       else
+               dev_err(dev, "set mask to 32bit fail!\n");
+
+       /* carrier off reporting is important to ethtool even BEFORE open */
+       netif_carrier_off(ndev);
+
+       setup_timer(&priv->service_timer, hns_nic_service_timer,
+                   (unsigned long)priv);
+       INIT_WORK(&priv->service_task, hns_nic_service_task);
+
+       set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
+       clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
+       set_bit(NIC_STATE_DOWN, &priv->state);
+
+       if (hns_nic_try_get_ae(priv->netdev)) {
+               priv->notifier_block.notifier_call = hns_nic_notifier_action;
+               ret = hnae_register_notifier(&priv->notifier_block);
+               if (ret) {
+                       dev_err(dev, "register notifier fail!\n");
+                       goto out_notify_fail;
+               }
+               dev_dbg(dev, "has not handle, register notifier!\n");
+       }
+
+       return 0;
+
+out_notify_fail:
+       (void)cancel_work_sync(&priv->service_task);
+out_read_string_fail:
+       free_netdev(ndev);
+       return ret;
+}
+
+static int hns_nic_dev_remove(struct platform_device *pdev)
+{
+       struct net_device *ndev = platform_get_drvdata(pdev);
+       struct hns_nic_priv *priv = netdev_priv(ndev);
+
+       if (ndev->reg_state != NETREG_UNINITIALIZED)
+               unregister_netdev(ndev);
+
+       if (priv->ring_data)
+               hns_nic_uninit_ring_data(priv);
+       priv->ring_data = NULL;
+
+       if (priv->phy)
+               phy_disconnect(priv->phy);
+       priv->phy = NULL;
+
+       if (!IS_ERR_OR_NULL(priv->ae_handle))
+               hnae_put_handle(priv->ae_handle);
+       priv->ae_handle = NULL;
+       if (priv->notifier_block.notifier_call)
+               hnae_unregister_notifier(&priv->notifier_block);
+       priv->notifier_block.notifier_call = NULL;
+
+       set_bit(NIC_STATE_REMOVING, &priv->state);
+       (void)cancel_work_sync(&priv->service_task);
+
+       free_netdev(ndev);
+       return 0;
+}
+
+static const struct of_device_id hns_enet_of_match[] = {
+       {.compatible = "hisilicon,hns-nic-v1",},
+       {.compatible = "hisilicon,hns-nic-v2",},
+       {},
+};
+
+MODULE_DEVICE_TABLE(of, hns_enet_of_match);
+
+static struct platform_driver hns_nic_dev_driver = {
+       .driver = {
+               .name = "hns-nic",
+               .of_match_table = hns_enet_of_match,
+       },
+       .probe = hns_nic_dev_probe,
+       .remove = hns_nic_dev_remove,
+};
+
+module_platform_driver(hns_nic_dev_driver);
+
+MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
+MODULE_AUTHOR("Hisilicon, Inc.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("platform:hns-nic");