drivers/net/ethernet/hisilicon/hns/hns_enet.c

   1 /*
   2  * Copyright (c) 2014-2015 Hisilicon Limited.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  */
   9
  10 #include <linux/clk.h>
  11 #include <linux/cpumask.h>
  12 #include <linux/etherdevice.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/interrupt.h>
  15 #include <linux/io.h>
  16 #include <linux/ip.h>
  17 #include <linux/ipv6.h>
  18 #include <linux/module.h>
  19 #include <linux/phy.h>
  20 #include <linux/platform_device.h>
  21 #include <linux/skbuff.h>
  22
  23 #include "hnae.h"
  24 #include "hns_enet.h"
  25 #include "hns_dsaf_mac.h"
  26
  27 #define NIC_MAX_Q_PER_VF 16
  28 #define HNS_NIC_TX_TIMEOUT (5 * HZ)
  29
  30 #define SERVICE_TIMER_HZ (1 * HZ)
  31
  32 #define NIC_TX_CLEAN_MAX_NUM 256
  33 #define NIC_RX_CLEAN_MAX_NUM 64
  34
  35 #define RCB_IRQ_NOT_INITED 0
  36 #define RCB_IRQ_INITED 1
  37 #define HNS_BUFFER_SIZE_2048 2048
  38
  39 #define BD_MAX_SEND_SIZE 8191
  40 #define SKB_TMP_LEN(SKB) \
  41         (((SKB)->transport_header - (SKB)->mac_header) + tcp_hdrlen(SKB))
  42
  43 static void fill_v2_desc(struct hnae_ring *ring, void *priv,
  44                          int size, dma_addr_t dma, int frag_end,
  45                          int buf_num, enum hns_desc_type type, int mtu)
  46 {
  47         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
  48         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
  49         struct iphdr *iphdr;
  50         struct ipv6hdr *ipv6hdr;
  51         struct sk_buff *skb;
  52         __be16 protocol;
  53         u8 bn_pid = 0;
  54         u8 rrcfv = 0;
  55         u8 ip_offset = 0;
  56         u8 tvsvsn = 0;
  57         u16 mss = 0;
  58         u8 l4_len = 0;
  59         u16 paylen = 0;
  60
  61         desc_cb->priv = priv;
  62         desc_cb->length = size;
  63         desc_cb->dma = dma;
  64         desc_cb->type = type;
  65
  66         desc->addr = cpu_to_le64(dma);
  67         desc->tx.send_size = cpu_to_le16((u16)size);
  68
  69         /* config bd buffer end */
  70         hnae_set_bit(rrcfv, HNSV2_TXD_VLD_B, 1);
  71         hnae_set_field(bn_pid, HNSV2_TXD_BUFNUM_M, 0, buf_num - 1);
  72
  73         /* fill port_id in the tx bd for sending management pkts */
  74         hnae_set_field(bn_pid, HNSV2_TXD_PORTID_M,
  75                        HNSV2_TXD_PORTID_S, ring->q->handle->dport_id);
  76
  77         if (type == DESC_TYPE_SKB) {
  78                 skb = (struct sk_buff *)priv;
  79
  80                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
  81                         skb_reset_mac_len(skb);
  82                         protocol = skb->protocol;
  83                         ip_offset = ETH_HLEN;
  84
  85                         if (protocol == htons(ETH_P_8021Q)) {
  86                                 ip_offset += VLAN_HLEN;
  87                                 protocol = vlan_get_protocol(skb);
  88                                 skb->protocol = protocol;
  89                         }
  90
  91                         if (skb->protocol == htons(ETH_P_IP)) {
  92                                 iphdr = ip_hdr(skb);
  93                                 hnae_set_bit(rrcfv, HNSV2_TXD_L3CS_B, 1);
  94                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
  95
  96                                 /* check for tcp/udp header */
  97                                 if (iphdr->protocol == IPPROTO_TCP &&
  98                                     skb_is_gso(skb)) {
  99                                         hnae_set_bit(tvsvsn,
 100                                                      HNSV2_TXD_TSE_B, 1);
 101                                         l4_len = tcp_hdrlen(skb);
 102                                         mss = skb_shinfo(skb)->gso_size;
 103                                         paylen = skb->len - SKB_TMP_LEN(skb);
 104                                 }
 105                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
 106                                 hnae_set_bit(tvsvsn, HNSV2_TXD_IPV6_B, 1);
 107                                 ipv6hdr = ipv6_hdr(skb);
 108                                 hnae_set_bit(rrcfv, HNSV2_TXD_L4CS_B, 1);
 109
 110                                 /* check for tcp/udp header */
 111                                 if (ipv6hdr->nexthdr == IPPROTO_TCP &&
 112                                     skb_is_gso(skb) && skb_is_gso_v6(skb)) {
 113                                         hnae_set_bit(tvsvsn,
 114                                                      HNSV2_TXD_TSE_B, 1);
 115                                         l4_len = tcp_hdrlen(skb);
 116                                         mss = skb_shinfo(skb)->gso_size;
 117                                         paylen = skb->len - SKB_TMP_LEN(skb);
 118                                 }
 119                         }
 120                         desc->tx.ip_offset = ip_offset;
 121                         desc->tx.tse_vlan_snap_v6_sctp_nth = tvsvsn;
 122                         desc->tx.mss = cpu_to_le16(mss);
 123                         desc->tx.l4_len = l4_len;
 124                         desc->tx.paylen = cpu_to_le16(paylen);
 125                 }
 126         }
 127
 128         hnae_set_bit(rrcfv, HNSV2_TXD_FE_B, frag_end);
 129
 130         desc->tx.bn_pid = bn_pid;
 131         desc->tx.ra_ri_cs_fe_vld = rrcfv;
 132
 133         ring_ptr_move_fw(ring, next_to_use);
 134 }
 135
 136 static const struct acpi_device_id hns_enet_acpi_match[] = {
 137         { "HISI00C1", 0 },
 138         { "HISI00C2", 0 },
 139         { },
 140 };
 141 MODULE_DEVICE_TABLE(acpi, hns_enet_acpi_match);
 142
 143 static void fill_desc(struct hnae_ring *ring, void *priv,
 144                       int size, dma_addr_t dma, int frag_end,
 145                       int buf_num, enum hns_desc_type type, int mtu)
 146 {
 147         struct hnae_desc *desc = &ring->desc[ring->next_to_use];
 148         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
 149         struct sk_buff *skb;
 150         __be16 protocol;
 151         u32 ip_offset;
 152         u32 asid_bufnum_pid = 0;
 153         u32 flag_ipoffset = 0;
 154
 155         desc_cb->priv = priv;
 156         desc_cb->length = size;
 157         desc_cb->dma = dma;
 158         desc_cb->type = type;
 159
 160         desc->addr = cpu_to_le64(dma);
 161         desc->tx.send_size = cpu_to_le16((u16)size);
 162
 163         /*config bd buffer end */
 164         flag_ipoffset |= 1 << HNS_TXD_VLD_B;
 165
 166         asid_bufnum_pid |= buf_num << HNS_TXD_BUFNUM_S;
 167
 168         if (type == DESC_TYPE_SKB) {
 169                 skb = (struct sk_buff *)priv;
 170
 171                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 172                         protocol = skb->protocol;
 173                         ip_offset = ETH_HLEN;
 174
 175                         /*if it is a SW VLAN check the next protocol*/
 176                         if (protocol == htons(ETH_P_8021Q)) {
 177                                 ip_offset += VLAN_HLEN;
 178                                 protocol = vlan_get_protocol(skb);
 179                                 skb->protocol = protocol;
 180                         }
 181
 182                         if (skb->protocol == htons(ETH_P_IP)) {
 183                                 flag_ipoffset |= 1 << HNS_TXD_L3CS_B;
 184                                 /* check for tcp/udp header */
 185                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
 186
 187                         } else if (skb->protocol == htons(ETH_P_IPV6)) {
 188                                 /* ipv6 has not l3 cs, check for L4 header */
 189                                 flag_ipoffset |= 1 << HNS_TXD_L4CS_B;
 190                         }
 191
 192                         flag_ipoffset |= ip_offset << HNS_TXD_IPOFFSET_S;
 193                 }
 194         }
 195
 196         flag_ipoffset |= frag_end << HNS_TXD_FE_B;
 197
 198         desc->tx.asid_bufnum_pid = cpu_to_le16(asid_bufnum_pid);
 199         desc->tx.flag_ipoffset = cpu_to_le32(flag_ipoffset);
 200
 201         ring_ptr_move_fw(ring, next_to_use);
 202 }
 203
 204 static void unfill_desc(struct hnae_ring *ring)
 205 {
 206         ring_ptr_move_bw(ring, next_to_use);
 207 }
 208
 209 static int hns_nic_maybe_stop_tx(
 210         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
 211 {
 212         struct sk_buff *skb = *out_skb;
 213         struct sk_buff *new_skb = NULL;
 214         int buf_num;
 215
 216         /* no. of segments (plus a header) */
 217         buf_num = skb_shinfo(skb)->nr_frags + 1;
 218
 219         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
 220                 if (ring_space(ring) < 1)
 221                         return -EBUSY;
 222
 223                 new_skb = skb_copy(skb, GFP_ATOMIC);
 224                 if (!new_skb)
 225                         return -ENOMEM;
 226
 227                 dev_kfree_skb_any(skb);
 228                 *out_skb = new_skb;
 229                 buf_num = 1;
 230         } else if (buf_num > ring_space(ring)) {
 231                 return -EBUSY;
 232         }
 233
 234         *bnum = buf_num;
 235         return 0;
 236 }
 237
 238 static int hns_nic_maybe_stop_tso(
 239         struct sk_buff **out_skb, int *bnum, struct hnae_ring *ring)
 240 {
 241         int i;
 242         int size;
 243         int buf_num;
 244         int frag_num;
 245         struct sk_buff *skb = *out_skb;
 246         struct sk_buff *new_skb = NULL;
 247         struct skb_frag_struct *frag;
 248
 249         size = skb_headlen(skb);
 250         buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
 251
 252         frag_num = skb_shinfo(skb)->nr_frags;
 253         for (i = 0; i < frag_num; i++) {
 254                 frag = &skb_shinfo(skb)->frags[i];
 255                 size = skb_frag_size(frag);
 256                 buf_num += (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
 257         }
 258
 259         if (unlikely(buf_num > ring->max_desc_num_per_pkt)) {
 260                 buf_num = (skb->len + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
 261                 if (ring_space(ring) < buf_num)
 262                         return -EBUSY;
 263                 /* manual split the send packet */
 264                 new_skb = skb_copy(skb, GFP_ATOMIC);
 265                 if (!new_skb)
 266                         return -ENOMEM;
 267                 dev_kfree_skb_any(skb);
 268                 *out_skb = new_skb;
 269
 270         } else if (ring_space(ring) < buf_num) {
 271                 return -EBUSY;
 272         }
 273
 274         *bnum = buf_num;
 275         return 0;
 276 }
 277
 278 static void fill_tso_desc(struct hnae_ring *ring, void *priv,
 279                           int size, dma_addr_t dma, int frag_end,
 280                           int buf_num, enum hns_desc_type type, int mtu)
 281 {
 282         int frag_buf_num;
 283         int sizeoflast;
 284         int k;
 285
 286         frag_buf_num = (size + BD_MAX_SEND_SIZE - 1) / BD_MAX_SEND_SIZE;
 287         sizeoflast = size % BD_MAX_SEND_SIZE;
 288         sizeoflast = sizeoflast ? sizeoflast : BD_MAX_SEND_SIZE;
 289
 290         /* when the frag size is bigger than hardware, split this frag */
 291         for (k = 0; k < frag_buf_num; k++)
 292                 fill_v2_desc(ring, priv,
 293                              (k == frag_buf_num - 1) ?
 294                                         sizeoflast : BD_MAX_SEND_SIZE,
 295                              dma + BD_MAX_SEND_SIZE * k,
 296                              frag_end && (k == frag_buf_num - 1) ? 1 : 0,
 297                              buf_num,
 298                              (type == DESC_TYPE_SKB && !k) ?
 299                                         DESC_TYPE_SKB : DESC_TYPE_PAGE,
 300                              mtu);
 301 }
 302
 303 int hns_nic_net_xmit_hw(struct net_device *ndev,
 304                         struct sk_buff *skb,
 305                         struct hns_nic_ring_data *ring_data)
 306 {
 307         struct hns_nic_priv *priv = netdev_priv(ndev);
 308         struct hnae_ring *ring = ring_data->ring;
 309         struct device *dev = ring_to_dev(ring);
 310         struct netdev_queue *dev_queue;
 311         struct skb_frag_struct *frag;
 312         int buf_num;
 313         int seg_num;
 314         dma_addr_t dma;
 315         int size, next_to_use;
 316         int i;
 317
 318         switch (priv->ops.maybe_stop_tx(&skb, &buf_num, ring)) {
 319         case -EBUSY:
 320                 ring->stats.tx_busy++;
 321                 goto out_net_tx_busy;
 322         case -ENOMEM:
 323                 ring->stats.sw_err_cnt++;
 324                 netdev_err(ndev, "no memory to xmit!\n");
 325                 goto out_err_tx_ok;
 326         default:
 327                 break;
 328         }
 329
 330         /* no. of segments (plus a header) */
 331         seg_num = skb_shinfo(skb)->nr_frags + 1;
 332         next_to_use = ring->next_to_use;
 333
 334         /* fill the first part */
 335         size = skb_headlen(skb);
 336         dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
 337         if (dma_mapping_error(dev, dma)) {
 338                 netdev_err(ndev, "TX head DMA map failed\n");
 339                 ring->stats.sw_err_cnt++;
 340                 goto out_err_tx_ok;
 341         }
 342         priv->ops.fill_desc(ring, skb, size, dma, seg_num == 1 ? 1 : 0,
 343                             buf_num, DESC_TYPE_SKB, ndev->mtu);
 344
 345         /* fill the fragments */
 346         for (i = 1; i < seg_num; i++) {
 347                 frag = &skb_shinfo(skb)->frags[i - 1];
 348                 size = skb_frag_size(frag);
 349                 dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
 350                 if (dma_mapping_error(dev, dma)) {
 351                         netdev_err(ndev, "TX frag(%d) DMA map failed\n", i);
 352                         ring->stats.sw_err_cnt++;
 353                         goto out_map_frag_fail;
 354                 }
 355                 priv->ops.fill_desc(ring, skb_frag_page(frag), size, dma,
 356                                     seg_num - 1 == i ? 1 : 0, buf_num,
 357                                     DESC_TYPE_PAGE, ndev->mtu);
 358         }
 359
 360         /*complete translate all packets*/
 361         dev_queue = netdev_get_tx_queue(ndev, skb->queue_mapping);
 362         netdev_tx_sent_queue(dev_queue, skb->len);
 363
 364         wmb(); /* commit all data before submit */
 365         assert(skb->queue_mapping < priv->ae_handle->q_num);
 366         hnae_queue_xmit(priv->ae_handle->qs[skb->queue_mapping], buf_num);
 367         ring->stats.tx_pkts++;
 368         ring->stats.tx_bytes += skb->len;
 369
 370         return NETDEV_TX_OK;
 371
 372 out_map_frag_fail:
 373
 374         while (ring->next_to_use != next_to_use) {
 375                 unfill_desc(ring);
 376                 if (ring->next_to_use != next_to_use)
 377                         dma_unmap_page(dev,
 378                                        ring->desc_cb[ring->next_to_use].dma,
 379                                        ring->desc_cb[ring->next_to_use].length,
 380                                        DMA_TO_DEVICE);
 381                 else
 382                         dma_unmap_single(dev,
 383                                          ring->desc_cb[next_to_use].dma,
 384                                          ring->desc_cb[next_to_use].length,
 385                                          DMA_TO_DEVICE);
 386         }
 387
 388 out_err_tx_ok:
 389
 390         dev_kfree_skb_any(skb);
 391         return NETDEV_TX_OK;
 392
 393 out_net_tx_busy:
 394
 395         netif_stop_subqueue(ndev, skb->queue_mapping);
 396
 397         /* Herbert's original patch had:
 398          *  smp_mb__after_netif_stop_queue();
 399          * but since that doesn't exist yet, just open code it.
 400          */
 401         smp_mb();
 402         return NETDEV_TX_BUSY;
 403 }
 404
 405 /**
 406  * hns_nic_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
 407  * @data: pointer to the start of the headers
 408  * @max: total length of section to find headers in
 409  *
 410  * This function is meant to determine the length of headers that will
 411  * be recognized by hardware for LRO, GRO, and RSC offloads.  The main
 412  * motivation of doing this is to only perform one pull for IPv4 TCP
 413  * packets so that we can do basic things like calculating the gso_size
 414  * based on the average data per packet.
 415  **/
 416 static unsigned int hns_nic_get_headlen(unsigned char *data, u32 flag,
 417                                         unsigned int max_size)
 418 {
 419         unsigned char *network;
 420         u8 hlen;
 421
 422         /* this should never happen, but better safe than sorry */
 423         if (max_size < ETH_HLEN)
 424                 return max_size;
 425
 426         /* initialize network frame pointer */
 427         network = data;
 428
 429         /* set first protocol and move network header forward */
 430         network += ETH_HLEN;
 431
 432         /* handle any vlan tag if present */
 433         if (hnae_get_field(flag, HNS_RXD_VLAN_M, HNS_RXD_VLAN_S)
 434                 == HNS_RX_FLAG_VLAN_PRESENT) {
 435                 if ((typeof(max_size))(network - data) > (max_size - VLAN_HLEN))
 436                         return max_size;
 437
 438                 network += VLAN_HLEN;
 439         }
 440
 441         /* handle L3 protocols */
 442         if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
 443                 == HNS_RX_FLAG_L3ID_IPV4) {
 444                 if ((typeof(max_size))(network - data) >
 445                     (max_size - sizeof(struct iphdr)))
 446                         return max_size;
 447
 448                 /* access ihl as a u8 to avoid unaligned access on ia64 */
 449                 hlen = (network[0] & 0x0F) << 2;
 450
 451                 /* verify hlen meets minimum size requirements */
 452                 if (hlen < sizeof(struct iphdr))
 453                         return network - data;
 454
 455                 /* record next protocol if header is present */
 456         } else if (hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S)
 457                 == HNS_RX_FLAG_L3ID_IPV6) {
 458                 if ((typeof(max_size))(network - data) >
 459                     (max_size - sizeof(struct ipv6hdr)))
 460                         return max_size;
 461
 462                 /* record next protocol */
 463                 hlen = sizeof(struct ipv6hdr);
 464         } else {
 465                 return network - data;
 466         }
 467
 468         /* relocate pointer to start of L4 header */
 469         network += hlen;
 470
 471         /* finally sort out TCP/UDP */
 472         if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
 473                 == HNS_RX_FLAG_L4ID_TCP) {
 474                 if ((typeof(max_size))(network - data) >
 475                     (max_size - sizeof(struct tcphdr)))
 476                         return max_size;
 477
 478                 /* access doff as a u8 to avoid unaligned access on ia64 */
 479                 hlen = (network[12] & 0xF0) >> 2;
 480
 481                 /* verify hlen meets minimum size requirements */
 482                 if (hlen < sizeof(struct tcphdr))
 483                         return network - data;
 484
 485                 network += hlen;
 486         } else if (hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S)
 487                 == HNS_RX_FLAG_L4ID_UDP) {
 488                 if ((typeof(max_size))(network - data) >
 489                     (max_size - sizeof(struct udphdr)))
 490                         return max_size;
 491
 492                 network += sizeof(struct udphdr);
 493         }
 494
 495         /* If everything has gone correctly network should be the
 496          * data section of the packet and will be the end of the header.
 497          * If not then it probably represents the end of the last recognized
 498          * header.
 499          */
 500         if ((typeof(max_size))(network - data) < max_size)
 501                 return network - data;
 502         else
 503                 return max_size;
 504 }
 505
 506 static void hns_nic_reuse_page(struct sk_buff *skb, int i,
 507                                struct hnae_ring *ring, int pull_len,
 508                                struct hnae_desc_cb *desc_cb)
 509 {
 510         struct hnae_desc *desc;
 511         int truesize, size;
 512         int last_offset;
 513         bool twobufs;
 514
 515         twobufs = ((PAGE_SIZE < 8192) &&
 516                 hnae_buf_size(ring) == HNS_BUFFER_SIZE_2048);
 517
 518         desc = &ring->desc[ring->next_to_clean];
 519         size = le16_to_cpu(desc->rx.size);
 520
 521         if (twobufs) {
 522                 truesize = hnae_buf_size(ring);
 523         } else {
 524                 truesize = ALIGN(size, L1_CACHE_BYTES);
 525                 last_offset = hnae_page_size(ring) - hnae_buf_size(ring);
 526         }
 527
 528         skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
 529                         size - pull_len, truesize - pull_len);
 530
 531          /* avoid re-using remote pages,flag default unreuse */
 532         if (unlikely(page_to_nid(desc_cb->priv) != numa_node_id()))
 533                 return;
 534
 535         if (twobufs) {
 536                 /* if we are only owner of page we can reuse it */
 537                 if (likely(page_count(desc_cb->priv) == 1)) {
 538                         /* flip page offset to other buffer */
 539                         desc_cb->page_offset ^= truesize;
 540
 541                         desc_cb->reuse_flag = 1;
 542                         /* bump ref count on page before it is given*/
 543                         get_page(desc_cb->priv);
 544                 }
 545                 return;
 546         }
 547
 548         /* move offset up to the next cache line */
 549         desc_cb->page_offset += truesize;
 550
 551         if (desc_cb->page_offset <= last_offset) {
 552                 desc_cb->reuse_flag = 1;
 553                 /* bump ref count on page before it is given*/
 554                 get_page(desc_cb->priv);
 555         }
 556 }
 557
 558 static void get_v2rx_desc_bnum(u32 bnum_flag, int *out_bnum)
 559 {
 560         *out_bnum = hnae_get_field(bnum_flag,
 561                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S) + 1;
 562 }
 563
 564 static void get_rx_desc_bnum(u32 bnum_flag, int *out_bnum)
 565 {
 566         *out_bnum = hnae_get_field(bnum_flag,
 567                                    HNS_RXD_BUFNUM_M, HNS_RXD_BUFNUM_S);
 568 }
 569
 570 static void hns_nic_rx_checksum(struct hns_nic_ring_data *ring_data,
 571                                 struct sk_buff *skb, u32 flag)
 572 {
 573         struct net_device *netdev = ring_data->napi.dev;
 574         u32 l3id;
 575         u32 l4id;
 576
 577         /* check if RX checksum offload is enabled */
 578         if (unlikely(!(netdev->features & NETIF_F_RXCSUM)))
 579                 return;
 580
 581         /* In hardware, we only support checksum for the following protocols:
 582          * 1) IPv4,
 583          * 2) TCP(over IPv4 or IPv6),
 584          * 3) UDP(over IPv4 or IPv6),
 585          * 4) SCTP(over IPv4 or IPv6)
 586          * but we support many L3(IPv4, IPv6, MPLS, PPPoE etc) and L4(TCP,
 587          * UDP, GRE, SCTP, IGMP, ICMP etc.) protocols.
 588          *
 589          * Hardware limitation:
 590          * Our present hardware RX Descriptor lacks L3/L4 checksum "Status &
 591          * Error" bit (which usually can be used to indicate whether checksum
 592          * was calculated by the hardware and if there was any error encountered
 593          * during checksum calculation).
 594          *
 595          * Software workaround:
 596          * We do get info within the RX descriptor about the kind of L3/L4
 597          * protocol coming in the packet and the error status. These errors
 598          * might not just be checksum errors but could be related to version,
 599          * length of IPv4, UDP, TCP etc.
 600          * Because there is no-way of knowing if it is a L3/L4 error due to bad
 601          * checksum or any other L3/L4 error, we will not (cannot) convey
 602          * checksum status for such cases to upper stack and will not maintain
 603          * the RX L3/L4 checksum counters as well.
 604          */
 605
 606         l3id = hnae_get_field(flag, HNS_RXD_L3ID_M, HNS_RXD_L3ID_S);
 607         l4id = hnae_get_field(flag, HNS_RXD_L4ID_M, HNS_RXD_L4ID_S);
 608
 609         /*  check L3 protocol for which checksum is supported */
 610         if ((l3id != HNS_RX_FLAG_L3ID_IPV4) && (l3id != HNS_RX_FLAG_L3ID_IPV6))
 611                 return;
 612
 613         /* check for any(not just checksum)flagged L3 protocol errors */
 614         if (unlikely(hnae_get_bit(flag, HNS_RXD_L3E_B)))
 615                 return;
 616
 617         /* we do not support checksum of fragmented packets */
 618         if (unlikely(hnae_get_bit(flag, HNS_RXD_FRAG_B)))
 619                 return;
 620
 621         /*  check L4 protocol for which checksum is supported */
 622         if ((l4id != HNS_RX_FLAG_L4ID_TCP) &&
 623             (l4id != HNS_RX_FLAG_L4ID_UDP) &&
 624             (l4id != HNS_RX_FLAG_L4ID_SCTP))
 625                 return;
 626
 627         /* check for any(not just checksum)flagged L4 protocol errors */
 628         if (unlikely(hnae_get_bit(flag, HNS_RXD_L4E_B)))
 629                 return;
 630
 631         /* now, this has to be a packet with valid RX checksum */
 632         skb->ip_summed = CHECKSUM_UNNECESSARY;
 633 }
 634
 635 static int hns_nic_poll_rx_skb(struct hns_nic_ring_data *ring_data,
 636                                struct sk_buff **out_skb, int *out_bnum)
 637 {
 638         struct hnae_ring *ring = ring_data->ring;
 639         struct net_device *ndev = ring_data->napi.dev;
 640         struct hns_nic_priv *priv = netdev_priv(ndev);
 641         struct sk_buff *skb;
 642         struct hnae_desc *desc;
 643         struct hnae_desc_cb *desc_cb;
 644         unsigned char *va;
 645         int bnum, length, i;
 646         int pull_len;
 647         u32 bnum_flag;
 648
 649         desc = &ring->desc[ring->next_to_clean];
 650         desc_cb = &ring->desc_cb[ring->next_to_clean];
 651
 652         prefetch(desc);
 653
 654         va = (unsigned char *)desc_cb->buf + desc_cb->page_offset;
 655
 656         /* prefetch first cache line of first page */
 657         prefetch(va);
 658 #if L1_CACHE_BYTES < 128
 659         prefetch(va + L1_CACHE_BYTES);
 660 #endif
 661
 662         skb = *out_skb = napi_alloc_skb(&ring_data->napi,
 663                                         HNS_RX_HEAD_SIZE);
 664         if (unlikely(!skb)) {
 665                 netdev_err(ndev, "alloc rx skb fail\n");
 666                 ring->stats.sw_err_cnt++;
 667                 return -ENOMEM;
 668         }
 669
 670         prefetchw(skb->data);
 671         length = le16_to_cpu(desc->rx.pkt_len);
 672         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
 673         priv->ops.get_rxd_bnum(bnum_flag, &bnum);
 674         *out_bnum = bnum;
 675
 676         if (length <= HNS_RX_HEAD_SIZE) {
 677                 memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
 678
 679                 /* we can reuse buffer as-is, just make sure it is local */
 680                 if (likely(page_to_nid(desc_cb->priv) == numa_node_id()))
 681                         desc_cb->reuse_flag = 1;
 682                 else /* this page cannot be reused so discard it */
 683                         put_page(desc_cb->priv);
 684
 685                 ring_ptr_move_fw(ring, next_to_clean);
 686
 687                 if (unlikely(bnum != 1)) { /* check err*/
 688                         *out_bnum = 1;
 689                         goto out_bnum_err;
 690                 }
 691         } else {
 692                 ring->stats.seg_pkt_cnt++;
 693
 694                 pull_len = hns_nic_get_headlen(va, bnum_flag, HNS_RX_HEAD_SIZE);
 695                 memcpy(__skb_put(skb, pull_len), va,
 696                        ALIGN(pull_len, sizeof(long)));
 697
 698                 hns_nic_reuse_page(skb, 0, ring, pull_len, desc_cb);
 699                 ring_ptr_move_fw(ring, next_to_clean);
 700
 701                 if (unlikely(bnum >= (int)MAX_SKB_FRAGS)) { /* check err*/
 702                         *out_bnum = 1;
 703                         goto out_bnum_err;
 704                 }
 705                 for (i = 1; i < bnum; i++) {
 706                         desc = &ring->desc[ring->next_to_clean];
 707                         desc_cb = &ring->desc_cb[ring->next_to_clean];
 708
 709                         hns_nic_reuse_page(skb, i, ring, 0, desc_cb);
 710                         ring_ptr_move_fw(ring, next_to_clean);
 711                 }
 712         }
 713
 714         /* check except process, free skb and jump the desc */
 715         if (unlikely((!bnum) || (bnum > ring->max_desc_num_per_pkt))) {
 716 out_bnum_err:
 717                 *out_bnum = *out_bnum ? *out_bnum : 1; /* ntc moved,cannot 0*/
 718                 netdev_err(ndev, "invalid bnum(%d,%d,%d,%d),%016llx,%016llx\n",
 719                            bnum, ring->max_desc_num_per_pkt,
 720                            length, (int)MAX_SKB_FRAGS,
 721                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
 722                 ring->stats.err_bd_num++;
 723                 dev_kfree_skb_any(skb);
 724                 return -EDOM;
 725         }
 726
 727         bnum_flag = le32_to_cpu(desc->rx.ipoff_bnum_pid_flag);
 728
 729         if (unlikely(!hnae_get_bit(bnum_flag, HNS_RXD_VLD_B))) {
 730                 netdev_err(ndev, "no valid bd,%016llx,%016llx\n",
 731                            ((u64 *)desc)[0], ((u64 *)desc)[1]);
 732                 ring->stats.non_vld_descs++;
 733                 dev_kfree_skb_any(skb);
 734                 return -EINVAL;
 735         }
 736
 737         if (unlikely((!desc->rx.pkt_len) ||
 738                      hnae_get_bit(bnum_flag, HNS_RXD_DROP_B))) {
 739                 ring->stats.err_pkt_len++;
 740                 dev_kfree_skb_any(skb);
 741                 return -EFAULT;
 742         }
 743
 744         if (unlikely(hnae_get_bit(bnum_flag, HNS_RXD_L2E_B))) {
 745                 ring->stats.l2_err++;
 746                 dev_kfree_skb_any(skb);
 747                 return -EFAULT;
 748         }
 749
 750         ring->stats.rx_pkts++;
 751         ring->stats.rx_bytes += skb->len;
 752
 753         /* indicate to upper stack if our hardware has already calculated
 754          * the RX checksum
 755          */
 756         hns_nic_rx_checksum(ring_data, skb, bnum_flag);
 757
 758         return 0;
 759 }
 760
 761 static void
 762 hns_nic_alloc_rx_buffers(struct hns_nic_ring_data *ring_data, int cleand_count)
 763 {
 764         int i, ret;
 765         struct hnae_desc_cb res_cbs;
 766         struct hnae_desc_cb *desc_cb;
 767         struct hnae_ring *ring = ring_data->ring;
 768         struct net_device *ndev = ring_data->napi.dev;
 769
 770         for (i = 0; i < cleand_count; i++) {
 771                 desc_cb = &ring->desc_cb[ring->next_to_use];
 772                 if (desc_cb->reuse_flag) {
 773                         ring->stats.reuse_pg_cnt++;
 774                         hnae_reuse_buffer(ring, ring->next_to_use);
 775                 } else {
 776                         ret = hnae_reserve_buffer_map(ring, &res_cbs);
 777                         if (ret) {
 778                                 ring->stats.sw_err_cnt++;
 779                                 netdev_err(ndev, "hnae reserve buffer map failed.\n");
 780                                 break;
 781                         }
 782                         hnae_replace_buffer(ring, ring->next_to_use, &res_cbs);
 783                 }
 784
 785                 ring_ptr_move_fw(ring, next_to_use);
 786         }
 787
 788         wmb(); /* make all data has been write before submit */
 789         writel_relaxed(i, ring->io_base + RCB_REG_HEAD);
 790 }
 791
 792 /* return error number for error or number of desc left to take
 793  */
 794 static void hns_nic_rx_up_pro(struct hns_nic_ring_data *ring_data,
 795                               struct sk_buff *skb)
 796 {
 797         struct net_device *ndev = ring_data->napi.dev;
 798
 799         skb->protocol = eth_type_trans(skb, ndev);
 800         (void)napi_gro_receive(&ring_data->napi, skb);
 801 }
 802
 803 static int hns_desc_unused(struct hnae_ring *ring)
 804 {
 805         int ntc = ring->next_to_clean;
 806         int ntu = ring->next_to_use;
 807
 808         return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
 809 }
 810
 811 static int hns_nic_rx_poll_one(struct hns_nic_ring_data *ring_data,
 812                                int budget, void *v)
 813 {
 814         struct hnae_ring *ring = ring_data->ring;
 815         struct sk_buff *skb;
 816         int num, bnum;
 817 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
 818         int recv_pkts, recv_bds, clean_count, err;
 819         int unused_count = hns_desc_unused(ring);
 820
 821         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
 822         rmb(); /* make sure num taken effect before the other data is touched */
 823
 824         recv_pkts = 0, recv_bds = 0, clean_count = 0;
 825         num -= unused_count;
 826
 827         while (recv_pkts < budget && recv_bds < num) {
 828                 /* reuse or realloc buffers */
 829                 if (clean_count + unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
 830                         hns_nic_alloc_rx_buffers(ring_data,
 831                                                  clean_count + unused_count);
 832                         clean_count = 0;
 833                         unused_count = hns_desc_unused(ring);
 834                 }
 835
 836                 /* poll one pkt */
 837                 err = hns_nic_poll_rx_skb(ring_data, &skb, &bnum);
 838                 if (unlikely(!skb)) /* this fault cannot be repaired */
 839                         goto out;
 840
 841                 recv_bds += bnum;
 842                 clean_count += bnum;
 843                 if (unlikely(err)) {  /* do jump the err */
 844                         recv_pkts++;
 845                         continue;
 846                 }
 847
 848                 /* do update ip stack process*/
 849                 ((void (*)(struct hns_nic_ring_data *, struct sk_buff *))v)(
 850                                                         ring_data, skb);
 851                 recv_pkts++;
 852         }
 853
 854 out:
 855         /* make all data has been write before submit */
 856         if (clean_count + unused_count > 0)
 857                 hns_nic_alloc_rx_buffers(ring_data,
 858                                          clean_count + unused_count);
 859
 860         return recv_pkts;
 861 }
 862
 863 static bool hns_nic_rx_fini_pro(struct hns_nic_ring_data *ring_data)
 864 {
 865         struct hnae_ring *ring = ring_data->ring;
 866         int num = 0;
 867
 868         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
 869
 870         /* for hardware bug fixed */
 871         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
 872
 873         if (num > 0) {
 874                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
 875                         ring_data->ring, 1);
 876
 877                 return false;
 878         } else {
 879                 return true;
 880         }
 881 }
 882
 883 static bool hns_nic_rx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
 884 {
 885         struct hnae_ring *ring = ring_data->ring;
 886         int num;
 887
 888         num = readl_relaxed(ring->io_base + RCB_REG_FBDNUM);
 889
 890         if (!num)
 891                 return true;
 892         else
 893                 return false;
 894 }
 895
 896 static inline void hns_nic_reclaim_one_desc(struct hnae_ring *ring,
 897                                             int *bytes, int *pkts)
 898 {
 899         struct hnae_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
 900
 901         (*pkts) += (desc_cb->type == DESC_TYPE_SKB);
 902         (*bytes) += desc_cb->length;
 903         /* desc_cb will be cleaned, after hnae_free_buffer_detach*/
 904         hnae_free_buffer_detach(ring, ring->next_to_clean);
 905
 906         ring_ptr_move_fw(ring, next_to_clean);
 907 }
 908
 909 static int is_valid_clean_head(struct hnae_ring *ring, int h)
 910 {
 911         int u = ring->next_to_use;
 912         int c = ring->next_to_clean;
 913
 914         if (unlikely(h > ring->desc_num))
 915                 return 0;
 916
 917         assert(u > 0 && u < ring->desc_num);
 918         assert(c > 0 && c < ring->desc_num);
 919         assert(u != c && h != c); /* must be checked before call this func */
 920
 921         return u > c ? (h > c && h <= u) : (h > c || h <= u);
 922 }
 923
 924 /* netif_tx_lock will turn down the performance, set only when necessary */
 925 #ifdef CONFIG_NET_POLL_CONTROLLER
 926 #define NETIF_TX_LOCK(ring) spin_lock(&(ring)->lock)
 927 #define NETIF_TX_UNLOCK(ring) spin_unlock(&(ring)->lock)
 928 #else
 929 #define NETIF_TX_LOCK(ring)
 930 #define NETIF_TX_UNLOCK(ring)
 931 #endif
 932
 933 /* reclaim all desc in one budget
 934  * return error or number of desc left
 935  */
 936 static int hns_nic_tx_poll_one(struct hns_nic_ring_data *ring_data,
 937                                int budget, void *v)
 938 {
 939         struct hnae_ring *ring = ring_data->ring;
 940         struct net_device *ndev = ring_data->napi.dev;
 941         struct netdev_queue *dev_queue;
 942         struct hns_nic_priv *priv = netdev_priv(ndev);
 943         int head;
 944         int bytes, pkts;
 945
 946         NETIF_TX_LOCK(ring);
 947
 948         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
 949         rmb(); /* make sure head is ready before touch any data */
 950
 951         if (is_ring_empty(ring) || head == ring->next_to_clean) {
 952                 NETIF_TX_UNLOCK(ring);
 953                 return 0; /* no data to poll */
 954         }
 955
 956         if (!is_valid_clean_head(ring, head)) {
 957                 netdev_err(ndev, "wrong head (%d, %d-%d)\n", head,
 958                            ring->next_to_use, ring->next_to_clean);
 959                 ring->stats.io_err_cnt++;
 960                 NETIF_TX_UNLOCK(ring);
 961                 return -EIO;
 962         }
 963
 964         bytes = 0;
 965         pkts = 0;
 966         while (head != ring->next_to_clean) {
 967                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
 968                 /* issue prefetch for next Tx descriptor */
 969                 prefetch(&ring->desc_cb[ring->next_to_clean]);
 970         }
 971
 972         NETIF_TX_UNLOCK(ring);
 973
 974         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
 975         netdev_tx_completed_queue(dev_queue, pkts, bytes);
 976
 977         if (unlikely(priv->link && !netif_carrier_ok(ndev)))
 978                 netif_carrier_on(ndev);
 979
 980         if (unlikely(pkts && netif_carrier_ok(ndev) &&
 981                      (ring_space(ring) >= ring->max_desc_num_per_pkt * 2))) {
 982                 /* Make sure that anybody stopping the queue after this
 983                  * sees the new next_to_clean.
 984                  */
 985                 smp_mb();
 986                 if (netif_tx_queue_stopped(dev_queue) &&
 987                     !test_bit(NIC_STATE_DOWN, &priv->state)) {
 988                         netif_tx_wake_queue(dev_queue);
 989                         ring->stats.restart_queue++;
 990                 }
 991         }
 992         return 0;
 993 }
 994
 995 static bool hns_nic_tx_fini_pro(struct hns_nic_ring_data *ring_data)
 996 {
 997         struct hnae_ring *ring = ring_data->ring;
 998         int head;
 999
1000         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1001
1002         head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1003
1004         if (head != ring->next_to_clean) {
1005                 ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1006                         ring_data->ring, 1);
1007
1008                 return false;
1009         } else {
1010                 return true;
1011         }
1012 }
1013
1014 static bool hns_nic_tx_fini_pro_v2(struct hns_nic_ring_data *ring_data)
1015 {
1016         struct hnae_ring *ring = ring_data->ring;
1017         int head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1018
1019         if (head == ring->next_to_clean)
1020                 return true;
1021         else
1022                 return false;
1023 }
1024
1025 static void hns_nic_tx_clr_all_bufs(struct hns_nic_ring_data *ring_data)
1026 {
1027         struct hnae_ring *ring = ring_data->ring;
1028         struct net_device *ndev = ring_data->napi.dev;
1029         struct netdev_queue *dev_queue;
1030         int head;
1031         int bytes, pkts;
1032
1033         NETIF_TX_LOCK(ring);
1034
1035         head = ring->next_to_use; /* ntu :soft setted ring position*/
1036         bytes = 0;
1037         pkts = 0;
1038         while (head != ring->next_to_clean)
1039                 hns_nic_reclaim_one_desc(ring, &bytes, &pkts);
1040
1041         NETIF_TX_UNLOCK(ring);
1042
1043         dev_queue = netdev_get_tx_queue(ndev, ring_data->queue_index);
1044         netdev_tx_reset_queue(dev_queue);
1045 }
1046
1047 static int hns_nic_common_poll(struct napi_struct *napi, int budget)
1048 {
1049         int clean_complete = 0;
1050         struct hns_nic_ring_data *ring_data =
1051                 container_of(napi, struct hns_nic_ring_data, napi);
1052         struct hnae_ring *ring = ring_data->ring;
1053
1054 try_again:
1055         clean_complete += ring_data->poll_one(
1056                                 ring_data, budget - clean_complete,
1057                                 ring_data->ex_process);
1058
1059         if (clean_complete < budget) {
1060                 if (ring_data->fini_process(ring_data)) {
1061                         napi_complete(napi);
1062                         ring->q->handle->dev->ops->toggle_ring_irq(ring, 0);
1063                 } else {
1064                         goto try_again;
1065                 }
1066         }
1067
1068         return clean_complete;
1069 }
1070
1071 static irqreturn_t hns_irq_handle(int irq, void *dev)
1072 {
1073         struct hns_nic_ring_data *ring_data = (struct hns_nic_ring_data *)dev;
1074
1075         ring_data->ring->q->handle->dev->ops->toggle_ring_irq(
1076                 ring_data->ring, 1);
1077         napi_schedule(&ring_data->napi);
1078
1079         return IRQ_HANDLED;
1080 }
1081
1082 /**
1083  *hns_nic_adjust_link - adjust net work mode by the phy stat or new param
1084  *@ndev: net device
1085  */
1086 static void hns_nic_adjust_link(struct net_device *ndev)
1087 {
1088         struct hns_nic_priv *priv = netdev_priv(ndev);
1089         struct hnae_handle *h = priv->ae_handle;
1090         int state = 1;
1091
1092         if (ndev->phydev) {
1093                 h->dev->ops->adjust_link(h, ndev->phydev->speed,
1094                                          ndev->phydev->duplex);
1095                 state = ndev->phydev->link;
1096         }
1097         state = state && h->dev->ops->get_status(h);
1098
1099         if (state != priv->link) {
1100                 if (state) {
1101                         netif_carrier_on(ndev);
1102                         netif_tx_wake_all_queues(ndev);
1103                         netdev_info(ndev, "link up\n");
1104                 } else {
1105                         netif_carrier_off(ndev);
1106                         netdev_info(ndev, "link down\n");
1107                 }
1108                 priv->link = state;
1109         }
1110 }
1111
1112 /**
1113  *hns_nic_init_phy - init phy
1114  *@ndev: net device
1115  *@h: ae handle
1116  * Return 0 on success, negative on failure
1117  */
1118 int hns_nic_init_phy(struct net_device *ndev, struct hnae_handle *h)
1119 {
1120         struct phy_device *phy_dev = h->phy_dev;
1121         int ret;
1122
1123         if (!h->phy_dev)
1124                 return 0;
1125
1126         if (h->phy_if != PHY_INTERFACE_MODE_XGMII) {
1127                 phy_dev->dev_flags = 0;
1128
1129                 ret = phy_connect_direct(ndev, phy_dev, hns_nic_adjust_link,
1130                                          h->phy_if);
1131         } else {
1132                 ret = phy_attach_direct(ndev, phy_dev, 0, h->phy_if);
1133         }
1134         if (unlikely(ret))
1135                 return -ENODEV;
1136
1137         phy_dev->supported &= h->if_support;
1138         phy_dev->advertising = phy_dev->supported;
1139
1140         if (h->phy_if == PHY_INTERFACE_MODE_XGMII)
1141                 phy_dev->autoneg = false;
1142
1143         return 0;
1144 }
1145
1146 static int hns_nic_ring_open(struct net_device *netdev, int idx)
1147 {
1148         struct hns_nic_priv *priv = netdev_priv(netdev);
1149         struct hnae_handle *h = priv->ae_handle;
1150
1151         napi_enable(&priv->ring_data[idx].napi);
1152
1153         enable_irq(priv->ring_data[idx].ring->irq);
1154         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 0);
1155
1156         return 0;
1157 }
1158
1159 static int hns_nic_net_set_mac_address(struct net_device *ndev, void *p)
1160 {
1161         struct hns_nic_priv *priv = netdev_priv(ndev);
1162         struct hnae_handle *h = priv->ae_handle;
1163         struct sockaddr *mac_addr = p;
1164         int ret;
1165
1166         if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1167                 return -EADDRNOTAVAIL;
1168
1169         ret = h->dev->ops->set_mac_addr(h, mac_addr->sa_data);
1170         if (ret) {
1171                 netdev_err(ndev, "set_mac_address fail, ret=%d!\n", ret);
1172                 return ret;
1173         }
1174
1175         memcpy(ndev->dev_addr, mac_addr->sa_data, ndev->addr_len);
1176
1177         return 0;
1178 }
1179
1180 void hns_nic_update_stats(struct net_device *netdev)
1181 {
1182         struct hns_nic_priv *priv = netdev_priv(netdev);
1183         struct hnae_handle *h = priv->ae_handle;
1184
1185         h->dev->ops->update_stats(h, &netdev->stats);
1186 }
1187
1188 /* set mac addr if it is configed. or leave it to the AE driver */
1189 static void hns_init_mac_addr(struct net_device *ndev)
1190 {
1191         struct hns_nic_priv *priv = netdev_priv(ndev);
1192
1193         if (!device_get_mac_address(priv->dev, ndev->dev_addr, ETH_ALEN)) {
1194                 eth_hw_addr_random(ndev);
1195                 dev_warn(priv->dev, "No valid mac, use random mac %pM",
1196                          ndev->dev_addr);
1197         }
1198 }
1199
1200 static void hns_nic_ring_close(struct net_device *netdev, int idx)
1201 {
1202         struct hns_nic_priv *priv = netdev_priv(netdev);
1203         struct hnae_handle *h = priv->ae_handle;
1204
1205         h->dev->ops->toggle_ring_irq(priv->ring_data[idx].ring, 1);
1206         disable_irq(priv->ring_data[idx].ring->irq);
1207
1208         napi_disable(&priv->ring_data[idx].napi);
1209 }
1210
1211 static int hns_nic_init_affinity_mask(int q_num, int ring_idx,
1212                                       struct hnae_ring *ring, cpumask_t *mask)
1213 {
1214         int cpu;
1215
1216         /* Diffrent irq banlance between 16core and 32core.
1217          * The cpu mask set by ring index according to the ring flag
1218          * which indicate the ring is tx or rx.
1219          */
1220         if (q_num == num_possible_cpus()) {
1221                 if (is_tx_ring(ring))
1222                         cpu = ring_idx;
1223                 else
1224                         cpu = ring_idx - q_num;
1225         } else {
1226                 if (is_tx_ring(ring))
1227                         cpu = ring_idx * 2;
1228                 else
1229                         cpu = (ring_idx - q_num) * 2 + 1;
1230         }
1231
1232         cpumask_clear(mask);
1233         cpumask_set_cpu(cpu, mask);
1234
1235         return cpu;
1236 }
1237
1238 static int hns_nic_init_irq(struct hns_nic_priv *priv)
1239 {
1240         struct hnae_handle *h = priv->ae_handle;
1241         struct hns_nic_ring_data *rd;
1242         int i;
1243         int ret;
1244         int cpu;
1245
1246         for (i = 0; i < h->q_num * 2; i++) {
1247                 rd = &priv->ring_data[i];
1248
1249                 if (rd->ring->irq_init_flag == RCB_IRQ_INITED)
1250                         break;
1251
1252                 snprintf(rd->ring->ring_name, RCB_RING_NAME_LEN,
1253                          "%s-%s%d", priv->netdev->name,
1254                          (is_tx_ring(rd->ring) ? "tx" : "rx"), rd->queue_index);
1255
1256                 rd->ring->ring_name[RCB_RING_NAME_LEN - 1] = '\0';
1257
1258                 ret = request_irq(rd->ring->irq,
1259                                   hns_irq_handle, 0, rd->ring->ring_name, rd);
1260                 if (ret) {
1261                         netdev_err(priv->netdev, "request irq(%d) fail\n",
1262                                    rd->ring->irq);
1263                         return ret;
1264                 }
1265                 disable_irq(rd->ring->irq);
1266
1267                 cpu = hns_nic_init_affinity_mask(h->q_num, i,
1268                                                  rd->ring, &rd->mask);
1269
1270                 if (cpu_online(cpu))
1271                         irq_set_affinity_hint(rd->ring->irq,
1272                                               &rd->mask);
1273
1274                 rd->ring->irq_init_flag = RCB_IRQ_INITED;
1275         }
1276
1277         return 0;
1278 }
1279
1280 static int hns_nic_net_up(struct net_device *ndev)
1281 {
1282         struct hns_nic_priv *priv = netdev_priv(ndev);
1283         struct hnae_handle *h = priv->ae_handle;
1284         int i, j;
1285         int ret;
1286
1287         ret = hns_nic_init_irq(priv);
1288         if (ret != 0) {
1289                 netdev_err(ndev, "hns init irq failed! ret=%d\n", ret);
1290                 return ret;
1291         }
1292
1293         for (i = 0; i < h->q_num * 2; i++) {
1294                 ret = hns_nic_ring_open(ndev, i);
1295                 if (ret)
1296                         goto out_has_some_queues;
1297         }
1298
1299         ret = h->dev->ops->set_mac_addr(h, ndev->dev_addr);
1300         if (ret)
1301                 goto out_set_mac_addr_err;
1302
1303         ret = h->dev->ops->start ? h->dev->ops->start(h) : 0;
1304         if (ret)
1305                 goto out_start_err;
1306
1307         if (ndev->phydev)
1308                 phy_start(ndev->phydev);
1309
1310         clear_bit(NIC_STATE_DOWN, &priv->state);
1311         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
1312
1313         return 0;
1314
1315 out_start_err:
1316         netif_stop_queue(ndev);
1317 out_set_mac_addr_err:
1318 out_has_some_queues:
1319         for (j = i - 1; j >= 0; j--)
1320                 hns_nic_ring_close(ndev, j);
1321
1322         set_bit(NIC_STATE_DOWN, &priv->state);
1323
1324         return ret;
1325 }
1326
1327 static void hns_nic_net_down(struct net_device *ndev)
1328 {
1329         int i;
1330         struct hnae_ae_ops *ops;
1331         struct hns_nic_priv *priv = netdev_priv(ndev);
1332
1333         if (test_and_set_bit(NIC_STATE_DOWN, &priv->state))
1334                 return;
1335
1336         (void)del_timer_sync(&priv->service_timer);
1337         netif_tx_stop_all_queues(ndev);
1338         netif_carrier_off(ndev);
1339         netif_tx_disable(ndev);
1340         priv->link = 0;
1341
1342         if (ndev->phydev)
1343                 phy_stop(ndev->phydev);
1344
1345         ops = priv->ae_handle->dev->ops;
1346
1347         if (ops->stop)
1348                 ops->stop(priv->ae_handle);
1349
1350         netif_tx_stop_all_queues(ndev);
1351
1352         for (i = priv->ae_handle->q_num - 1; i >= 0; i--) {
1353                 hns_nic_ring_close(ndev, i);
1354                 hns_nic_ring_close(ndev, i + priv->ae_handle->q_num);
1355
1356                 /* clean tx buffers*/
1357                 hns_nic_tx_clr_all_bufs(priv->ring_data + i);
1358         }
1359 }
1360
1361 void hns_nic_net_reset(struct net_device *ndev)
1362 {
1363         struct hns_nic_priv *priv = netdev_priv(ndev);
1364         struct hnae_handle *handle = priv->ae_handle;
1365
1366         while (test_and_set_bit(NIC_STATE_RESETTING, &priv->state))
1367                 usleep_range(1000, 2000);
1368
1369         (void)hnae_reinit_handle(handle);
1370
1371         clear_bit(NIC_STATE_RESETTING, &priv->state);
1372 }
1373
1374 void hns_nic_net_reinit(struct net_device *netdev)
1375 {
1376         struct hns_nic_priv *priv = netdev_priv(netdev);
1377
1378         netif_trans_update(priv->netdev);
1379         while (test_and_set_bit(NIC_STATE_REINITING, &priv->state))
1380                 usleep_range(1000, 2000);
1381
1382         hns_nic_net_down(netdev);
1383         hns_nic_net_reset(netdev);
1384         (void)hns_nic_net_up(netdev);
1385         clear_bit(NIC_STATE_REINITING, &priv->state);
1386 }
1387
1388 static int hns_nic_net_open(struct net_device *ndev)
1389 {
1390         struct hns_nic_priv *priv = netdev_priv(ndev);
1391         struct hnae_handle *h = priv->ae_handle;
1392         int ret;
1393
1394         if (test_bit(NIC_STATE_TESTING, &priv->state))
1395                 return -EBUSY;
1396
1397         priv->link = 0;
1398         netif_carrier_off(ndev);
1399
1400         ret = netif_set_real_num_tx_queues(ndev, h->q_num);
1401         if (ret < 0) {
1402                 netdev_err(ndev, "netif_set_real_num_tx_queues fail, ret=%d!\n",
1403                            ret);
1404                 return ret;
1405         }
1406
1407         ret = netif_set_real_num_rx_queues(ndev, h->q_num);
1408         if (ret < 0) {
1409                 netdev_err(ndev,
1410                            "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
1411                 return ret;
1412         }
1413
1414         ret = hns_nic_net_up(ndev);
1415         if (ret) {
1416                 netdev_err(ndev,
1417                            "hns net up fail, ret=%d!\n", ret);
1418                 return ret;
1419         }
1420
1421         return 0;
1422 }
1423
1424 static int hns_nic_net_stop(struct net_device *ndev)
1425 {
1426         hns_nic_net_down(ndev);
1427
1428         return 0;
1429 }
1430
1431 static void hns_tx_timeout_reset(struct hns_nic_priv *priv);
1432 static void hns_nic_net_timeout(struct net_device *ndev)
1433 {
1434         struct hns_nic_priv *priv = netdev_priv(ndev);
1435
1436         hns_tx_timeout_reset(priv);
1437 }
1438
1439 static int hns_nic_do_ioctl(struct net_device *netdev, struct ifreq *ifr,
1440                             int cmd)
1441 {
1442         struct phy_device *phy_dev = netdev->phydev;
1443
1444         if (!netif_running(netdev))
1445                 return -EINVAL;
1446
1447         if (!phy_dev)
1448                 return -ENOTSUPP;
1449
1450         return phy_mii_ioctl(phy_dev, ifr, cmd);
1451 }
1452
1453 /* use only for netconsole to poll with the device without interrupt */
1454 #ifdef CONFIG_NET_POLL_CONTROLLER
1455 void hns_nic_poll_controller(struct net_device *ndev)
1456 {
1457         struct hns_nic_priv *priv = netdev_priv(ndev);
1458         unsigned long flags;
1459         int i;
1460
1461         local_irq_save(flags);
1462         for (i = 0; i < priv->ae_handle->q_num * 2; i++)
1463                 napi_schedule(&priv->ring_data[i].napi);
1464         local_irq_restore(flags);
1465 }
1466 #endif
1467
1468 static netdev_tx_t hns_nic_net_xmit(struct sk_buff *skb,
1469                                     struct net_device *ndev)
1470 {
1471         struct hns_nic_priv *priv = netdev_priv(ndev);
1472         int ret;
1473
1474         assert(skb->queue_mapping < ndev->ae_handle->q_num);
1475         ret = hns_nic_net_xmit_hw(ndev, skb,
1476                                   &tx_ring_data(priv, skb->queue_mapping));
1477         if (ret == NETDEV_TX_OK) {
1478                 netif_trans_update(ndev);
1479                 ndev->stats.tx_bytes += skb->len;
1480                 ndev->stats.tx_packets++;
1481         }
1482         return (netdev_tx_t)ret;
1483 }
1484
1485 static void hns_nic_drop_rx_fetch(struct hns_nic_ring_data *ring_data,
1486                                   struct sk_buff *skb)
1487 {
1488         dev_kfree_skb_any(skb);
1489 }
1490
1491 #define HNS_LB_TX_RING  0
1492 static struct sk_buff *hns_assemble_skb(struct net_device *ndev)
1493 {
1494         struct sk_buff *skb;
1495         struct ethhdr *ethhdr;
1496         int frame_len;
1497
1498         /* allocate test skb */
1499         skb = alloc_skb(64, GFP_KERNEL);
1500         if (!skb)
1501                 return NULL;
1502
1503         skb_put(skb, 64);
1504         skb->dev = ndev;
1505         memset(skb->data, 0xFF, skb->len);
1506
1507         /* must be tcp/ip package */
1508         ethhdr = (struct ethhdr *)skb->data;
1509         ethhdr->h_proto = htons(ETH_P_IP);
1510
1511         frame_len = skb->len & (~1ul);
1512         memset(&skb->data[frame_len / 2], 0xAA,
1513                frame_len / 2 - 1);
1514
1515         skb->queue_mapping = HNS_LB_TX_RING;
1516
1517         return skb;
1518 }
1519
1520 static int hns_enable_serdes_lb(struct net_device *ndev)
1521 {
1522         struct hns_nic_priv *priv = netdev_priv(ndev);
1523         struct hnae_handle *h = priv->ae_handle;
1524         struct hnae_ae_ops *ops = h->dev->ops;
1525         int speed, duplex;
1526         int ret;
1527
1528         ret = ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 1);
1529         if (ret)
1530                 return ret;
1531
1532         ret = ops->start ? ops->start(h) : 0;
1533         if (ret)
1534                 return ret;
1535
1536         /* link adjust duplex*/
1537         if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1538                 speed = 1000;
1539         else
1540                 speed = 10000;
1541         duplex = 1;
1542
1543         ops->adjust_link(h, speed, duplex);
1544
1545         /* wait h/w ready */
1546         mdelay(300);
1547
1548         return 0;
1549 }
1550
1551 static void hns_disable_serdes_lb(struct net_device *ndev)
1552 {
1553         struct hns_nic_priv *priv = netdev_priv(ndev);
1554         struct hnae_handle *h = priv->ae_handle;
1555         struct hnae_ae_ops *ops = h->dev->ops;
1556
1557         ops->stop(h);
1558         ops->set_loopback(h, MAC_INTERNALLOOP_SERDES, 0);
1559 }
1560
1561 /**
1562  *hns_nic_clear_all_rx_fetch - clear the chip fetched descriptions. The
1563  *function as follows:
1564  *    1. if one rx ring has found the page_offset is not equal 0 between head
1565  *       and tail, it means that the chip fetched the wrong descs for the ring
1566  *       which buffer size is 4096.
1567  *    2. we set the chip serdes loopback and set rss indirection to the ring.
1568  *    3. construct 64-bytes ip broadcast packages, wait the associated rx ring
1569  *       recieving all packages and it will fetch new descriptions.
1570  *    4. recover to the original state.
1571  *
1572  *@ndev: net device
1573  */
1574 static int hns_nic_clear_all_rx_fetch(struct net_device *ndev)
1575 {
1576         struct hns_nic_priv *priv = netdev_priv(ndev);
1577         struct hnae_handle *h = priv->ae_handle;
1578         struct hnae_ae_ops *ops = h->dev->ops;
1579         struct hns_nic_ring_data *rd;
1580         struct hnae_ring *ring;
1581         struct sk_buff *skb;
1582         u32 *org_indir;
1583         u32 *cur_indir;
1584         int indir_size;
1585         int head, tail;
1586         int fetch_num;
1587         int i, j;
1588         bool found;
1589         int retry_times;
1590         int ret = 0;
1591
1592         /* alloc indir memory */
1593         indir_size = ops->get_rss_indir_size(h) * sizeof(*org_indir);
1594         org_indir = kzalloc(indir_size, GFP_KERNEL);
1595         if (!org_indir)
1596                 return -ENOMEM;
1597
1598         /* store the orginal indirection */
1599         ops->get_rss(h, org_indir, NULL, NULL);
1600
1601         cur_indir = kzalloc(indir_size, GFP_KERNEL);
1602         if (!cur_indir) {
1603                 ret = -ENOMEM;
1604                 goto cur_indir_alloc_err;
1605         }
1606
1607         /* set loopback */
1608         if (hns_enable_serdes_lb(ndev)) {
1609                 ret = -EINVAL;
1610                 goto enable_serdes_lb_err;
1611         }
1612
1613         /* foreach every rx ring to clear fetch desc */
1614         for (i = 0; i < h->q_num; i++) {
1615                 ring = &h->qs[i]->rx_ring;
1616                 head = readl_relaxed(ring->io_base + RCB_REG_HEAD);
1617                 tail = readl_relaxed(ring->io_base + RCB_REG_TAIL);
1618                 found = false;
1619                 fetch_num = ring_dist(ring, head, tail);
1620
1621                 while (head != tail) {
1622                         if (ring->desc_cb[head].page_offset != 0) {
1623                                 found = true;
1624                                 break;
1625                         }
1626
1627                         head++;
1628                         if (head == ring->desc_num)
1629                                 head = 0;
1630                 }
1631
1632                 if (found) {
1633                         for (j = 0; j < indir_size / sizeof(*org_indir); j++)
1634                                 cur_indir[j] = i;
1635                         ops->set_rss(h, cur_indir, NULL, 0);
1636
1637                         for (j = 0; j < fetch_num; j++) {
1638                                 /* alloc one skb and init */
1639                                 skb = hns_assemble_skb(ndev);
1640                                 if (!skb)
1641                                         goto out;
1642                                 rd = &tx_ring_data(priv, skb->queue_mapping);
1643                                 hns_nic_net_xmit_hw(ndev, skb, rd);
1644
1645                                 retry_times = 0;
1646                                 while (retry_times++ < 10) {
1647                                         mdelay(10);
1648                                         /* clean rx */
1649                                         rd = &rx_ring_data(priv, i);
1650                                         if (rd->poll_one(rd, fetch_num,
1651                                                          hns_nic_drop_rx_fetch))
1652                                                 break;
1653                                 }
1654
1655                                 retry_times = 0;
1656                                 while (retry_times++ < 10) {
1657                                         mdelay(10);
1658                                         /* clean tx ring 0 send package */
1659                                         rd = &tx_ring_data(priv,
1660                                                            HNS_LB_TX_RING);
1661                                         if (rd->poll_one(rd, fetch_num, NULL))
1662                                                 break;
1663                                 }
1664                         }
1665                 }
1666         }
1667
1668 out:
1669         /* restore everything */
1670         ops->set_rss(h, org_indir, NULL, 0);
1671         hns_disable_serdes_lb(ndev);
1672 enable_serdes_lb_err:
1673         kfree(cur_indir);
1674 cur_indir_alloc_err:
1675         kfree(org_indir);
1676
1677         return ret;
1678 }
1679
1680 static int hns_nic_change_mtu(struct net_device *ndev, int new_mtu)
1681 {
1682         struct hns_nic_priv *priv = netdev_priv(ndev);
1683         struct hnae_handle *h = priv->ae_handle;
1684         bool if_running = netif_running(ndev);
1685         int ret;
1686
1687         /* MTU < 68 is an error and causes problems on some kernels */
1688         if (new_mtu < 68)
1689                 return -EINVAL;
1690
1691         /* MTU no change */
1692         if (new_mtu == ndev->mtu)
1693                 return 0;
1694
1695         if (!h->dev->ops->set_mtu)
1696                 return -ENOTSUPP;
1697
1698         if (if_running) {
1699                 (void)hns_nic_net_stop(ndev);
1700                 msleep(100);
1701         }
1702
1703         if (priv->enet_ver != AE_VERSION_1 &&
1704             ndev->mtu <= BD_SIZE_2048_MAX_MTU &&
1705             new_mtu > BD_SIZE_2048_MAX_MTU) {
1706                 /* update desc */
1707                 hnae_reinit_all_ring_desc(h);
1708
1709                 /* clear the package which the chip has fetched */
1710                 ret = hns_nic_clear_all_rx_fetch(ndev);
1711
1712                 /* the page offset must be consist with desc */
1713                 hnae_reinit_all_ring_page_off(h);
1714
1715                 if (ret) {
1716                         netdev_err(ndev, "clear the fetched desc fail\n");
1717                         goto out;
1718                 }
1719         }
1720
1721         ret = h->dev->ops->set_mtu(h, new_mtu);
1722         if (ret) {
1723                 netdev_err(ndev, "set mtu fail, return value %d\n",
1724                            ret);
1725                 goto out;
1726         }
1727
1728         /* finally, set new mtu to netdevice */
1729         ndev->mtu = new_mtu;
1730
1731 out:
1732         if (if_running) {
1733                 if (hns_nic_net_open(ndev)) {
1734                         netdev_err(ndev, "hns net open fail\n");
1735                         ret = -EINVAL;
1736                 }
1737         }
1738
1739         return ret;
1740 }
1741
1742 static int hns_nic_set_features(struct net_device *netdev,
1743                                 netdev_features_t features)
1744 {
1745         struct hns_nic_priv *priv = netdev_priv(netdev);
1746
1747         switch (priv->enet_ver) {
1748         case AE_VERSION_1:
1749                 if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1750                         netdev_info(netdev, "enet v1 do not support tso!\n");
1751                 break;
1752         default:
1753                 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) {
1754                         priv->ops.fill_desc = fill_tso_desc;
1755                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
1756                         /* The chip only support 7*4096 */
1757                         netif_set_gso_max_size(netdev, 7 * 4096);
1758                 } else {
1759                         priv->ops.fill_desc = fill_v2_desc;
1760                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
1761                 }
1762                 break;
1763         }
1764         netdev->features = features;
1765         return 0;
1766 }
1767
1768 static netdev_features_t hns_nic_fix_features(
1769                 struct net_device *netdev, netdev_features_t features)
1770 {
1771         struct hns_nic_priv *priv = netdev_priv(netdev);
1772
1773         switch (priv->enet_ver) {
1774         case AE_VERSION_1:
1775                 features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
1776                                 NETIF_F_HW_VLAN_CTAG_FILTER);
1777                 break;
1778         default:
1779                 break;
1780         }
1781         return features;
1782 }
1783
1784 static int hns_nic_uc_sync(struct net_device *netdev, const unsigned char *addr)
1785 {
1786         struct hns_nic_priv *priv = netdev_priv(netdev);
1787         struct hnae_handle *h = priv->ae_handle;
1788
1789         if (h->dev->ops->add_uc_addr)
1790                 return h->dev->ops->add_uc_addr(h, addr);
1791
1792         return 0;
1793 }
1794
1795 static int hns_nic_uc_unsync(struct net_device *netdev,
1796                              const unsigned char *addr)
1797 {
1798         struct hns_nic_priv *priv = netdev_priv(netdev);
1799         struct hnae_handle *h = priv->ae_handle;
1800
1801         if (h->dev->ops->rm_uc_addr)
1802                 return h->dev->ops->rm_uc_addr(h, addr);
1803
1804         return 0;
1805 }
1806
1807 /**
1808  * nic_set_multicast_list - set mutl mac address
1809  * @netdev: net device
1810  * @p: mac address
1811  *
1812  * return void
1813  */
1814 void hns_set_multicast_list(struct net_device *ndev)
1815 {
1816         struct hns_nic_priv *priv = netdev_priv(ndev);
1817         struct hnae_handle *h = priv->ae_handle;
1818         struct netdev_hw_addr *ha = NULL;
1819
1820         if (!h) {
1821                 netdev_err(ndev, "hnae handle is null\n");
1822                 return;
1823         }
1824
1825         if (h->dev->ops->clr_mc_addr)
1826                 if (h->dev->ops->clr_mc_addr(h))
1827                         netdev_err(ndev, "clear multicast address fail\n");
1828
1829         if (h->dev->ops->set_mc_addr) {
1830                 netdev_for_each_mc_addr(ha, ndev)
1831                         if (h->dev->ops->set_mc_addr(h, ha->addr))
1832                                 netdev_err(ndev, "set multicast fail\n");
1833         }
1834 }
1835
1836 void hns_nic_set_rx_mode(struct net_device *ndev)
1837 {
1838         struct hns_nic_priv *priv = netdev_priv(ndev);
1839         struct hnae_handle *h = priv->ae_handle;
1840
1841         if (h->dev->ops->set_promisc_mode) {
1842                 if (ndev->flags & IFF_PROMISC)
1843                         h->dev->ops->set_promisc_mode(h, 1);
1844                 else
1845                         h->dev->ops->set_promisc_mode(h, 0);
1846         }
1847
1848         hns_set_multicast_list(ndev);
1849
1850         if (__dev_uc_sync(ndev, hns_nic_uc_sync, hns_nic_uc_unsync))
1851                 netdev_err(ndev, "sync uc address fail\n");
1852 }
1853
1854 static void hns_nic_get_stats64(struct net_device *ndev,
1855                                 struct rtnl_link_stats64 *stats)
1856 {
1857         int idx = 0;
1858         u64 tx_bytes = 0;
1859         u64 rx_bytes = 0;
1860         u64 tx_pkts = 0;
1861         u64 rx_pkts = 0;
1862         struct hns_nic_priv *priv = netdev_priv(ndev);
1863         struct hnae_handle *h = priv->ae_handle;
1864
1865         for (idx = 0; idx < h->q_num; idx++) {
1866                 tx_bytes += h->qs[idx]->tx_ring.stats.tx_bytes;
1867                 tx_pkts += h->qs[idx]->tx_ring.stats.tx_pkts;
1868                 rx_bytes += h->qs[idx]->rx_ring.stats.rx_bytes;
1869                 rx_pkts += h->qs[idx]->rx_ring.stats.rx_pkts;
1870         }
1871
1872         stats->tx_bytes = tx_bytes;
1873         stats->tx_packets = tx_pkts;
1874         stats->rx_bytes = rx_bytes;
1875         stats->rx_packets = rx_pkts;
1876
1877         stats->rx_errors = ndev->stats.rx_errors;
1878         stats->multicast = ndev->stats.multicast;
1879         stats->rx_length_errors = ndev->stats.rx_length_errors;
1880         stats->rx_crc_errors = ndev->stats.rx_crc_errors;
1881         stats->rx_missed_errors = ndev->stats.rx_missed_errors;
1882
1883         stats->tx_errors = ndev->stats.tx_errors;
1884         stats->rx_dropped = ndev->stats.rx_dropped;
1885         stats->tx_dropped = ndev->stats.tx_dropped;
1886         stats->collisions = ndev->stats.collisions;
1887         stats->rx_over_errors = ndev->stats.rx_over_errors;
1888         stats->rx_frame_errors = ndev->stats.rx_frame_errors;
1889         stats->rx_fifo_errors = ndev->stats.rx_fifo_errors;
1890         stats->tx_aborted_errors = ndev->stats.tx_aborted_errors;
1891         stats->tx_carrier_errors = ndev->stats.tx_carrier_errors;
1892         stats->tx_fifo_errors = ndev->stats.tx_fifo_errors;
1893         stats->tx_heartbeat_errors = ndev->stats.tx_heartbeat_errors;
1894         stats->tx_window_errors = ndev->stats.tx_window_errors;
1895         stats->rx_compressed = ndev->stats.rx_compressed;
1896         stats->tx_compressed = ndev->stats.tx_compressed;
1897 }
1898
1899 static u16
1900 hns_nic_select_queue(struct net_device *ndev, struct sk_buff *skb,
1901                      void *accel_priv, select_queue_fallback_t fallback)
1902 {
1903         struct ethhdr *eth_hdr = (struct ethhdr *)skb->data;
1904         struct hns_nic_priv *priv = netdev_priv(ndev);
1905
1906         /* fix hardware broadcast/multicast packets queue loopback */
1907         if (!AE_IS_VER1(priv->enet_ver) &&
1908             is_multicast_ether_addr(eth_hdr->h_dest))
1909                 return 0;
1910         else
1911                 return fallback(ndev, skb);
1912 }
1913
1914 static const struct net_device_ops hns_nic_netdev_ops = {
1915         .ndo_open = hns_nic_net_open,
1916         .ndo_stop = hns_nic_net_stop,
1917         .ndo_start_xmit = hns_nic_net_xmit,
1918         .ndo_tx_timeout = hns_nic_net_timeout,
1919         .ndo_set_mac_address = hns_nic_net_set_mac_address,
1920         .ndo_change_mtu = hns_nic_change_mtu,
1921         .ndo_do_ioctl = hns_nic_do_ioctl,
1922         .ndo_set_features = hns_nic_set_features,
1923         .ndo_fix_features = hns_nic_fix_features,
1924         .ndo_get_stats64 = hns_nic_get_stats64,
1925 #ifdef CONFIG_NET_POLL_CONTROLLER
1926         .ndo_poll_controller = hns_nic_poll_controller,
1927 #endif
1928         .ndo_set_rx_mode = hns_nic_set_rx_mode,
1929         .ndo_select_queue = hns_nic_select_queue,
1930 };
1931
1932 static void hns_nic_update_link_status(struct net_device *netdev)
1933 {
1934         struct hns_nic_priv *priv = netdev_priv(netdev);
1935
1936         struct hnae_handle *h = priv->ae_handle;
1937
1938         if (h->phy_dev) {
1939                 if (h->phy_if != PHY_INTERFACE_MODE_XGMII)
1940                         return;
1941
1942                 (void)genphy_read_status(h->phy_dev);
1943         }
1944         hns_nic_adjust_link(netdev);
1945 }
1946
1947 /* for dumping key regs*/
1948 static void hns_nic_dump(struct hns_nic_priv *priv)
1949 {
1950         struct hnae_handle *h = priv->ae_handle;
1951         struct hnae_ae_ops *ops = h->dev->ops;
1952         u32 *data, reg_num, i;
1953
1954         if (ops->get_regs_len && ops->get_regs) {
1955                 reg_num = ops->get_regs_len(priv->ae_handle);
1956                 reg_num = (reg_num + 3ul) & ~3ul;
1957                 data = kcalloc(reg_num, sizeof(u32), GFP_KERNEL);
1958                 if (data) {
1959                         ops->get_regs(priv->ae_handle, data);
1960                         for (i = 0; i < reg_num; i += 4)
1961                                 pr_info("0x%08x: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1962                                         i, data[i], data[i + 1],
1963                                         data[i + 2], data[i + 3]);
1964                         kfree(data);
1965                 }
1966         }
1967
1968         for (i = 0; i < h->q_num; i++) {
1969                 pr_info("tx_queue%d_next_to_clean:%d\n",
1970                         i, h->qs[i]->tx_ring.next_to_clean);
1971                 pr_info("tx_queue%d_next_to_use:%d\n",
1972                         i, h->qs[i]->tx_ring.next_to_use);
1973                 pr_info("rx_queue%d_next_to_clean:%d\n",
1974                         i, h->qs[i]->rx_ring.next_to_clean);
1975                 pr_info("rx_queue%d_next_to_use:%d\n",
1976                         i, h->qs[i]->rx_ring.next_to_use);
1977         }
1978 }
1979
1980 /* for resetting subtask */
1981 static void hns_nic_reset_subtask(struct hns_nic_priv *priv)
1982 {
1983         enum hnae_port_type type = priv->ae_handle->port_type;
1984
1985         if (!test_bit(NIC_STATE2_RESET_REQUESTED, &priv->state))
1986                 return;
1987         clear_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
1988
1989         /* If we're already down, removing or resetting, just bail */
1990         if (test_bit(NIC_STATE_DOWN, &priv->state) ||
1991             test_bit(NIC_STATE_REMOVING, &priv->state) ||
1992             test_bit(NIC_STATE_RESETTING, &priv->state))
1993                 return;
1994
1995         hns_nic_dump(priv);
1996         netdev_info(priv->netdev, "try to reset %s port!\n",
1997                     (type == HNAE_PORT_DEBUG ? "debug" : "service"));
1998
1999         rtnl_lock();
2000         /* put off any impending NetWatchDogTimeout */
2001         netif_trans_update(priv->netdev);
2002
2003         if (type == HNAE_PORT_DEBUG) {
2004                 hns_nic_net_reinit(priv->netdev);
2005         } else {
2006                 netif_carrier_off(priv->netdev);
2007                 netif_tx_disable(priv->netdev);
2008         }
2009         rtnl_unlock();
2010 }
2011
2012 /* for doing service complete*/
2013 static void hns_nic_service_event_complete(struct hns_nic_priv *priv)
2014 {
2015         WARN_ON(!test_bit(NIC_STATE_SERVICE_SCHED, &priv->state));
2016         /* make sure to commit the things */
2017         smp_mb__before_atomic();
2018         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2019 }
2020
2021 static void hns_nic_service_task(struct work_struct *work)
2022 {
2023         struct hns_nic_priv *priv
2024                 = container_of(work, struct hns_nic_priv, service_task);
2025         struct hnae_handle *h = priv->ae_handle;
2026
2027         hns_nic_update_link_status(priv->netdev);
2028         h->dev->ops->update_led_status(h);
2029         hns_nic_update_stats(priv->netdev);
2030
2031         hns_nic_reset_subtask(priv);
2032         hns_nic_service_event_complete(priv);
2033 }
2034
2035 static void hns_nic_task_schedule(struct hns_nic_priv *priv)
2036 {
2037         if (!test_bit(NIC_STATE_DOWN, &priv->state) &&
2038             !test_bit(NIC_STATE_REMOVING, &priv->state) &&
2039             !test_and_set_bit(NIC_STATE_SERVICE_SCHED, &priv->state))
2040                 (void)schedule_work(&priv->service_task);
2041 }
2042
2043 static void hns_nic_service_timer(unsigned long data)
2044 {
2045         struct hns_nic_priv *priv = (struct hns_nic_priv *)data;
2046
2047         (void)mod_timer(&priv->service_timer, jiffies + SERVICE_TIMER_HZ);
2048
2049         hns_nic_task_schedule(priv);
2050 }
2051
2052 /**
2053  * hns_tx_timeout_reset - initiate reset due to Tx timeout
2054  * @priv: driver private struct
2055  **/
2056 static void hns_tx_timeout_reset(struct hns_nic_priv *priv)
2057 {
2058         /* Do the reset outside of interrupt context */
2059         if (!test_bit(NIC_STATE_DOWN, &priv->state)) {
2060                 set_bit(NIC_STATE2_RESET_REQUESTED, &priv->state);
2061                 netdev_warn(priv->netdev,
2062                             "initiating reset due to tx timeout(%llu,0x%lx)\n",
2063                             priv->tx_timeout_count, priv->state);
2064                 priv->tx_timeout_count++;
2065                 hns_nic_task_schedule(priv);
2066         }
2067 }
2068
2069 static int hns_nic_init_ring_data(struct hns_nic_priv *priv)
2070 {
2071         struct hnae_handle *h = priv->ae_handle;
2072         struct hns_nic_ring_data *rd;
2073         bool is_ver1 = AE_IS_VER1(priv->enet_ver);
2074         int i;
2075
2076         if (h->q_num > NIC_MAX_Q_PER_VF) {
2077                 netdev_err(priv->netdev, "too much queue (%d)\n", h->q_num);
2078                 return -EINVAL;
2079         }
2080
2081         priv->ring_data = kzalloc(h->q_num * sizeof(*priv->ring_data) * 2,
2082                                   GFP_KERNEL);
2083         if (!priv->ring_data)
2084                 return -ENOMEM;
2085
2086         for (i = 0; i < h->q_num; i++) {
2087                 rd = &priv->ring_data[i];
2088                 rd->queue_index = i;
2089                 rd->ring = &h->qs[i]->tx_ring;
2090                 rd->poll_one = hns_nic_tx_poll_one;
2091                 rd->fini_process = is_ver1 ? hns_nic_tx_fini_pro :
2092                         hns_nic_tx_fini_pro_v2;
2093
2094                 netif_napi_add(priv->netdev, &rd->napi,
2095                                hns_nic_common_poll, NIC_TX_CLEAN_MAX_NUM);
2096                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2097         }
2098         for (i = h->q_num; i < h->q_num * 2; i++) {
2099                 rd = &priv->ring_data[i];
2100                 rd->queue_index = i - h->q_num;
2101                 rd->ring = &h->qs[i - h->q_num]->rx_ring;
2102                 rd->poll_one = hns_nic_rx_poll_one;
2103                 rd->ex_process = hns_nic_rx_up_pro;
2104                 rd->fini_process = is_ver1 ? hns_nic_rx_fini_pro :
2105                         hns_nic_rx_fini_pro_v2;
2106
2107                 netif_napi_add(priv->netdev, &rd->napi,
2108                                hns_nic_common_poll, NIC_RX_CLEAN_MAX_NUM);
2109                 rd->ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2110         }
2111
2112         return 0;
2113 }
2114
2115 static void hns_nic_uninit_ring_data(struct hns_nic_priv *priv)
2116 {
2117         struct hnae_handle *h = priv->ae_handle;
2118         int i;
2119
2120         for (i = 0; i < h->q_num * 2; i++) {
2121                 netif_napi_del(&priv->ring_data[i].napi);
2122                 if (priv->ring_data[i].ring->irq_init_flag == RCB_IRQ_INITED) {
2123                         (void)irq_set_affinity_hint(
2124                                 priv->ring_data[i].ring->irq,
2125                                 NULL);
2126                         free_irq(priv->ring_data[i].ring->irq,
2127                                  &priv->ring_data[i]);
2128                 }
2129
2130                 priv->ring_data[i].ring->irq_init_flag = RCB_IRQ_NOT_INITED;
2131         }
2132         kfree(priv->ring_data);
2133 }
2134
2135 static void hns_nic_set_priv_ops(struct net_device *netdev)
2136 {
2137         struct hns_nic_priv *priv = netdev_priv(netdev);
2138         struct hnae_handle *h = priv->ae_handle;
2139
2140         if (AE_IS_VER1(priv->enet_ver)) {
2141                 priv->ops.fill_desc = fill_desc;
2142                 priv->ops.get_rxd_bnum = get_rx_desc_bnum;
2143                 priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2144         } else {
2145                 priv->ops.get_rxd_bnum = get_v2rx_desc_bnum;
2146                 if ((netdev->features & NETIF_F_TSO) ||
2147                     (netdev->features & NETIF_F_TSO6)) {
2148                         priv->ops.fill_desc = fill_tso_desc;
2149                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tso;
2150                         /* This chip only support 7*4096 */
2151                         netif_set_gso_max_size(netdev, 7 * 4096);
2152                 } else {
2153                         priv->ops.fill_desc = fill_v2_desc;
2154                         priv->ops.maybe_stop_tx = hns_nic_maybe_stop_tx;
2155                 }
2156                 /* enable tso when init
2157                  * control tso on/off through TSE bit in bd
2158                  */
2159                 h->dev->ops->set_tso_stats(h, 1);
2160         }
2161 }
2162
2163 static int hns_nic_try_get_ae(struct net_device *ndev)
2164 {
2165         struct hns_nic_priv *priv = netdev_priv(ndev);
2166         struct hnae_handle *h;
2167         int ret;
2168
2169         h = hnae_get_handle(&priv->netdev->dev,
2170                             priv->fwnode, priv->port_id, NULL);
2171         if (IS_ERR_OR_NULL(h)) {
2172                 ret = -ENODEV;
2173                 dev_dbg(priv->dev, "has not handle, register notifier!\n");
2174                 goto out;
2175         }
2176         priv->ae_handle = h;
2177
2178         ret = hns_nic_init_phy(ndev, h);
2179         if (ret) {
2180                 dev_err(priv->dev, "probe phy device fail!\n");
2181                 goto out_init_phy;
2182         }
2183
2184         ret = hns_nic_init_ring_data(priv);
2185         if (ret) {
2186                 ret = -ENOMEM;
2187                 goto out_init_ring_data;
2188         }
2189
2190         hns_nic_set_priv_ops(ndev);
2191
2192         ret = register_netdev(ndev);
2193         if (ret) {
2194                 dev_err(priv->dev, "probe register netdev fail!\n");
2195                 goto out_reg_ndev_fail;
2196         }
2197         return 0;
2198
2199 out_reg_ndev_fail:
2200         hns_nic_uninit_ring_data(priv);
2201         priv->ring_data = NULL;
2202 out_init_phy:
2203 out_init_ring_data:
2204         hnae_put_handle(priv->ae_handle);
2205         priv->ae_handle = NULL;
2206 out:
2207         return ret;
2208 }
2209
2210 static int hns_nic_notifier_action(struct notifier_block *nb,
2211                                    unsigned long action, void *data)
2212 {
2213         struct hns_nic_priv *priv =
2214                 container_of(nb, struct hns_nic_priv, notifier_block);
2215
2216         assert(action == HNAE_AE_REGISTER);
2217
2218         if (!hns_nic_try_get_ae(priv->netdev)) {
2219                 hnae_unregister_notifier(&priv->notifier_block);
2220                 priv->notifier_block.notifier_call = NULL;
2221         }
2222         return 0;
2223 }
2224
2225 static int hns_nic_dev_probe(struct platform_device *pdev)
2226 {
2227         struct device *dev = &pdev->dev;
2228         struct net_device *ndev;
2229         struct hns_nic_priv *priv;
2230         u32 port_id;
2231         int ret;
2232
2233         ndev = alloc_etherdev_mq(sizeof(struct hns_nic_priv), NIC_MAX_Q_PER_VF);
2234         if (!ndev)
2235                 return -ENOMEM;
2236
2237         platform_set_drvdata(pdev, ndev);
2238
2239         priv = netdev_priv(ndev);
2240         priv->dev = dev;
2241         priv->netdev = ndev;
2242
2243         if (dev_of_node(dev)) {
2244                 struct device_node *ae_node;
2245
2246                 if (of_device_is_compatible(dev->of_node,
2247                                             "hisilicon,hns-nic-v1"))
2248                         priv->enet_ver = AE_VERSION_1;
2249                 else
2250                         priv->enet_ver = AE_VERSION_2;
2251
2252                 ae_node = of_parse_phandle(dev->of_node, "ae-handle", 0);
2253                 if (IS_ERR_OR_NULL(ae_node)) {
2254                         ret = PTR_ERR(ae_node);
2255                         dev_err(dev, "not find ae-handle\n");
2256                         goto out_read_prop_fail;
2257                 }
2258                 priv->fwnode = &ae_node->fwnode;
2259         } else if (is_acpi_node(dev->fwnode)) {
2260                 struct acpi_reference_args args;
2261
2262                 if (acpi_dev_found(hns_enet_acpi_match[0].id))
2263                         priv->enet_ver = AE_VERSION_1;
2264                 else if (acpi_dev_found(hns_enet_acpi_match[1].id))
2265                         priv->enet_ver = AE_VERSION_2;
2266                 else
2267                         return -ENXIO;
2268
2269                 /* try to find port-idx-in-ae first */
2270                 ret = acpi_node_get_property_reference(dev->fwnode,
2271                                                        "ae-handle", 0, &args);
2272                 if (ret) {
2273                         dev_err(dev, "not find ae-handle\n");
2274                         goto out_read_prop_fail;
2275                 }
2276                 priv->fwnode = acpi_fwnode_handle(args.adev);
2277         } else {
2278                 dev_err(dev, "cannot read cfg data from OF or acpi\n");
2279                 return -ENXIO;
2280         }
2281
2282         ret = device_property_read_u32(dev, "port-idx-in-ae", &port_id);
2283         if (ret) {
2284                 /* only for old code compatible */
2285                 ret = device_property_read_u32(dev, "port-id", &port_id);
2286                 if (ret)
2287                         goto out_read_prop_fail;
2288                 /* for old dts, we need to caculate the port offset */
2289                 port_id = port_id < HNS_SRV_OFFSET ? port_id + HNS_DEBUG_OFFSET
2290                         : port_id - HNS_SRV_OFFSET;
2291         }
2292         priv->port_id = port_id;
2293
2294         hns_init_mac_addr(ndev);
2295
2296         ndev->watchdog_timeo = HNS_NIC_TX_TIMEOUT;
2297         ndev->priv_flags |= IFF_UNICAST_FLT;
2298         ndev->netdev_ops = &hns_nic_netdev_ops;
2299         hns_ethtool_set_ops(ndev);
2300
2301         ndev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2302                 NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2303                 NETIF_F_GRO;
2304         ndev->vlan_features |=
2305                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | NETIF_F_RXCSUM;
2306         ndev->vlan_features |= NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
2307
2308         /* MTU range: 68 - 9578 (v1) or 9706 (v2) */
2309         ndev->min_mtu = MAC_MIN_MTU;
2310         switch (priv->enet_ver) {
2311         case AE_VERSION_2:
2312                 ndev->features |= NETIF_F_TSO | NETIF_F_TSO6;
2313                 ndev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2314                         NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2315                         NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6;
2316                 ndev->max_mtu = MAC_MAX_MTU_V2 -
2317                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2318                 break;
2319         default:
2320                 ndev->max_mtu = MAC_MAX_MTU -
2321                                 (ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN);
2322                 break;
2323         }
2324
2325         SET_NETDEV_DEV(ndev, dev);
2326
2327         if (!dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64)))
2328                 dev_dbg(dev, "set mask to 64bit\n");
2329         else
2330                 dev_err(dev, "set mask to 64bit fail!\n");
2331
2332         /* carrier off reporting is important to ethtool even BEFORE open */
2333         netif_carrier_off(ndev);
2334
2335         setup_timer(&priv->service_timer, hns_nic_service_timer,
2336                     (unsigned long)priv);
2337         INIT_WORK(&priv->service_task, hns_nic_service_task);
2338
2339         set_bit(NIC_STATE_SERVICE_INITED, &priv->state);
2340         clear_bit(NIC_STATE_SERVICE_SCHED, &priv->state);
2341         set_bit(NIC_STATE_DOWN, &priv->state);
2342
2343         if (hns_nic_try_get_ae(priv->netdev)) {
2344                 priv->notifier_block.notifier_call = hns_nic_notifier_action;
2345                 ret = hnae_register_notifier(&priv->notifier_block);
2346                 if (ret) {
2347                         dev_err(dev, "register notifier fail!\n");
2348                         goto out_notify_fail;
2349                 }
2350                 dev_dbg(dev, "has not handle, register notifier!\n");
2351         }
2352
2353         return 0;
2354
2355 out_notify_fail:
2356         (void)cancel_work_sync(&priv->service_task);
2357 out_read_prop_fail:
2358         free_netdev(ndev);
2359         return ret;
2360 }
2361
2362 static int hns_nic_dev_remove(struct platform_device *pdev)
2363 {
2364         struct net_device *ndev = platform_get_drvdata(pdev);
2365         struct hns_nic_priv *priv = netdev_priv(ndev);
2366
2367         if (ndev->reg_state != NETREG_UNINITIALIZED)
2368                 unregister_netdev(ndev);
2369
2370         if (priv->ring_data)
2371                 hns_nic_uninit_ring_data(priv);
2372         priv->ring_data = NULL;
2373
2374         if (ndev->phydev)
2375                 phy_disconnect(ndev->phydev);
2376
2377         if (!IS_ERR_OR_NULL(priv->ae_handle))
2378                 hnae_put_handle(priv->ae_handle);
2379         priv->ae_handle = NULL;
2380         if (priv->notifier_block.notifier_call)
2381                 hnae_unregister_notifier(&priv->notifier_block);
2382         priv->notifier_block.notifier_call = NULL;
2383
2384         set_bit(NIC_STATE_REMOVING, &priv->state);
2385         (void)cancel_work_sync(&priv->service_task);
2386
2387         free_netdev(ndev);
2388         return 0;
2389 }
2390
2391 static const struct of_device_id hns_enet_of_match[] = {
2392         {.compatible = "hisilicon,hns-nic-v1",},
2393         {.compatible = "hisilicon,hns-nic-v2",},
2394         {},
2395 };
2396
2397 MODULE_DEVICE_TABLE(of, hns_enet_of_match);
2398
2399 static struct platform_driver hns_nic_dev_driver = {
2400         .driver = {
2401                 .name = "hns-nic",
2402                 .of_match_table = hns_enet_of_match,
2403                 .acpi_match_table = ACPI_PTR(hns_enet_acpi_match),
2404         },
2405         .probe = hns_nic_dev_probe,
2406         .remove = hns_nic_dev_remove,
2407 };
2408
2409 module_platform_driver(hns_nic_dev_driver);
2410
2411 MODULE_DESCRIPTION("HISILICON HNS Ethernet driver");
2412 MODULE_AUTHOR("Hisilicon, Inc.");
2413 MODULE_LICENSE("GPL");
2414 MODULE_ALIAS("platform:hns-nic");