drivers/net/ethernet/sun/sunvnet.c

   1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
   2  *
   3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/module.h>
   9 #include <linux/kernel.h>
  10 #include <linux/types.h>
  11 #include <linux/slab.h>
  12 #include <linux/delay.h>
  13 #include <linux/init.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/ethtool.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/mutex.h>
  18 #include <linux/highmem.h>
  19 #include <linux/if_vlan.h>
  20
  21 #if IS_ENABLED(CONFIG_IPV6)
  22 #include <linux/icmpv6.h>
  23 #endif
  24
  25 #include <net/ip.h>
  26 #include <net/icmp.h>
  27 #include <net/route.h>
  28
  29 #include <asm/vio.h>
  30 #include <asm/ldc.h>
  31
  32 #include "sunvnet.h"
  33
  34 #define DRV_MODULE_NAME         "sunvnet"
  35 #define DRV_MODULE_VERSION      "1.0"
  36 #define DRV_MODULE_RELDATE      "June 25, 2007"
  37
  38 static char version[] =
  39         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
  40 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
  41 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
  42 MODULE_LICENSE("GPL");
  43 MODULE_VERSION(DRV_MODULE_VERSION);
  44
  45 #define VNET_MAX_TXQS           16
  46
  47 /* Heuristic for the number of times to exponentially backoff and
  48  * retry sending an LDC trigger when EAGAIN is encountered
  49  */
  50 #define VNET_MAX_RETRIES        10
  51
  52 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
  53 static void vnet_port_reset(struct vnet_port *port);
  54
  55 /* Ordered from largest major to lowest */
  56 static struct vio_version vnet_versions[] = {
  57         { .major = 1, .minor = 8 },
  58         { .major = 1, .minor = 7 },
  59         { .major = 1, .minor = 6 },
  60         { .major = 1, .minor = 0 },
  61 };
  62
  63 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
  64 {
  65         return vio_dring_avail(dr, VNET_TX_RING_SIZE);
  66 }
  67
  68 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
  69 {
  70         struct vio_msg_tag *pkt = arg;
  71
  72         pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
  73                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
  74         pr_err("Resetting connection\n");
  75
  76         ldc_disconnect(port->vio.lp);
  77
  78         return -ECONNRESET;
  79 }
  80
  81 static int vnet_port_alloc_tx_ring(struct vnet_port *port);
  82
  83 static int vnet_send_attr(struct vio_driver_state *vio)
  84 {
  85         struct vnet_port *port = to_vnet_port(vio);
  86         struct net_device *dev = port->vp->dev;
  87         struct vio_net_attr_info pkt;
  88         int framelen = ETH_FRAME_LEN;
  89         int i, err;
  90
  91         err = vnet_port_alloc_tx_ring(to_vnet_port(vio));
  92         if (err)
  93                 return err;
  94
  95         memset(&pkt, 0, sizeof(pkt));
  96         pkt.tag.type = VIO_TYPE_CTRL;
  97         pkt.tag.stype = VIO_SUBTYPE_INFO;
  98         pkt.tag.stype_env = VIO_ATTR_INFO;
  99         pkt.tag.sid = vio_send_sid(vio);
 100         if (vio_version_before(vio, 1, 2))
 101                 pkt.xfer_mode = VIO_DRING_MODE;
 102         else
 103                 pkt.xfer_mode = VIO_NEW_DRING_MODE;
 104         pkt.addr_type = VNET_ADDR_ETHERMAC;
 105         pkt.ack_freq = 0;
 106         for (i = 0; i < 6; i++)
 107                 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
 108         if (vio_version_after(vio, 1, 3)) {
 109                 if (port->rmtu) {
 110                         port->rmtu = min(VNET_MAXPACKET, port->rmtu);
 111                         pkt.mtu = port->rmtu;
 112                 } else {
 113                         port->rmtu = VNET_MAXPACKET;
 114                         pkt.mtu = port->rmtu;
 115                 }
 116                 if (vio_version_after_eq(vio, 1, 6))
 117                         pkt.options = VIO_TX_DRING;
 118         } else if (vio_version_before(vio, 1, 3)) {
 119                 pkt.mtu = framelen;
 120         } else { /* v1.3 */
 121                 pkt.mtu = framelen + VLAN_HLEN;
 122         }
 123
 124         pkt.cflags = 0;
 125         if (vio_version_after_eq(vio, 1, 7) && port->tso) {
 126                 pkt.cflags |= VNET_LSO_IPV4_CAPAB;
 127                 if (!port->tsolen)
 128                         port->tsolen = VNET_MAXTSO;
 129                 pkt.ipv4_lso_maxlen = port->tsolen;
 130         }
 131
 132         pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
 133
 134         viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
 135                "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
 136                "cflags[0x%04x] lso_max[%u]\n",
 137                pkt.xfer_mode, pkt.addr_type,
 138                (unsigned long long)pkt.addr,
 139                pkt.ack_freq, pkt.plnk_updt, pkt.options,
 140                (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
 141
 142
 143         return vio_ldc_send(vio, &pkt, sizeof(pkt));
 144 }
 145
 146 static int handle_attr_info(struct vio_driver_state *vio,
 147                             struct vio_net_attr_info *pkt)
 148 {
 149         struct vnet_port *port = to_vnet_port(vio);
 150         u64     localmtu;
 151         u8      xfer_mode;
 152
 153         viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
 154                "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
 155                " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
 156                pkt->xfer_mode, pkt->addr_type,
 157                (unsigned long long)pkt->addr,
 158                pkt->ack_freq, pkt->plnk_updt, pkt->options,
 159                (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
 160                pkt->ipv4_lso_maxlen);
 161
 162         pkt->tag.sid = vio_send_sid(vio);
 163
 164         xfer_mode = pkt->xfer_mode;
 165         /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
 166         if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
 167                 xfer_mode = VIO_NEW_DRING_MODE;
 168
 169         /* MTU negotiation:
 170          *      < v1.3 - ETH_FRAME_LEN exactly
 171          *      > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
 172          *                      pkt->mtu for ACK
 173          *      = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
 174          */
 175         if (vio_version_before(vio, 1, 3)) {
 176                 localmtu = ETH_FRAME_LEN;
 177         } else if (vio_version_after(vio, 1, 3)) {
 178                 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
 179                 localmtu = min(pkt->mtu, localmtu);
 180                 pkt->mtu = localmtu;
 181         } else { /* v1.3 */
 182                 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
 183         }
 184         port->rmtu = localmtu;
 185
 186         /* LSO negotiation */
 187         if (vio_version_after_eq(vio, 1, 7))
 188                 port->tso &= !!(pkt->cflags & VNET_LSO_IPV4_CAPAB);
 189         else
 190                 port->tso = false;
 191         if (port->tso) {
 192                 if (!port->tsolen)
 193                         port->tsolen = VNET_MAXTSO;
 194                 port->tsolen = min(port->tsolen, pkt->ipv4_lso_maxlen);
 195                 if (port->tsolen < VNET_MINTSO) {
 196                         port->tso = false;
 197                         port->tsolen = 0;
 198                         pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
 199                 }
 200                 pkt->ipv4_lso_maxlen = port->tsolen;
 201         } else {
 202                 pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
 203                 pkt->ipv4_lso_maxlen = 0;
 204         }
 205
 206         /* for version >= 1.6, ACK packet mode we support */
 207         if (vio_version_after_eq(vio, 1, 6)) {
 208                 pkt->xfer_mode = VIO_NEW_DRING_MODE;
 209                 pkt->options = VIO_TX_DRING;
 210         }
 211
 212         if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
 213             pkt->addr_type != VNET_ADDR_ETHERMAC ||
 214             pkt->mtu != localmtu) {
 215                 viodbg(HS, "SEND NET ATTR NACK\n");
 216
 217                 pkt->tag.stype = VIO_SUBTYPE_NACK;
 218
 219                 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
 220
 221                 return -ECONNRESET;
 222         } else {
 223                 viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
 224                        "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
 225                        "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
 226                        pkt->xfer_mode, pkt->addr_type,
 227                        (unsigned long long)pkt->addr,
 228                        pkt->ack_freq, pkt->plnk_updt, pkt->options,
 229                        (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
 230                        pkt->ipv4_lso_maxlen);
 231
 232                 pkt->tag.stype = VIO_SUBTYPE_ACK;
 233
 234                 return vio_ldc_send(vio, pkt, sizeof(*pkt));
 235         }
 236
 237 }
 238
 239 static int handle_attr_ack(struct vio_driver_state *vio,
 240                            struct vio_net_attr_info *pkt)
 241 {
 242         viodbg(HS, "GOT NET ATTR ACK\n");
 243
 244         return 0;
 245 }
 246
 247 static int handle_attr_nack(struct vio_driver_state *vio,
 248                             struct vio_net_attr_info *pkt)
 249 {
 250         viodbg(HS, "GOT NET ATTR NACK\n");
 251
 252         return -ECONNRESET;
 253 }
 254
 255 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
 256 {
 257         struct vio_net_attr_info *pkt = arg;
 258
 259         switch (pkt->tag.stype) {
 260         case VIO_SUBTYPE_INFO:
 261                 return handle_attr_info(vio, pkt);
 262
 263         case VIO_SUBTYPE_ACK:
 264                 return handle_attr_ack(vio, pkt);
 265
 266         case VIO_SUBTYPE_NACK:
 267                 return handle_attr_nack(vio, pkt);
 268
 269         default:
 270                 return -ECONNRESET;
 271         }
 272 }
 273
 274 static void vnet_handshake_complete(struct vio_driver_state *vio)
 275 {
 276         struct vio_dring_state *dr;
 277
 278         dr = &vio->drings[VIO_DRIVER_RX_RING];
 279         dr->snd_nxt = dr->rcv_nxt = 1;
 280
 281         dr = &vio->drings[VIO_DRIVER_TX_RING];
 282         dr->snd_nxt = dr->rcv_nxt = 1;
 283 }
 284
 285 /* The hypervisor interface that implements copying to/from imported
 286  * memory from another domain requires that copies are done to 8-byte
 287  * aligned buffers, and that the lengths of such copies are also 8-byte
 288  * multiples.
 289  *
 290  * So we align skb->data to an 8-byte multiple and pad-out the data
 291  * area so we can round the copy length up to the next multiple of
 292  * 8 for the copy.
 293  *
 294  * The transmitter puts the actual start of the packet 6 bytes into
 295  * the buffer it sends over, so that the IP headers after the ethernet
 296  * header are aligned properly.  These 6 bytes are not in the descriptor
 297  * length, they are simply implied.  This offset is represented using
 298  * the VNET_PACKET_SKIP macro.
 299  */
 300 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
 301                                            unsigned int len)
 302 {
 303         struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
 304         unsigned long addr, off;
 305
 306         if (unlikely(!skb))
 307                 return NULL;
 308
 309         addr = (unsigned long) skb->data;
 310         off = ((addr + 7UL) & ~7UL) - addr;
 311         if (off)
 312                 skb_reserve(skb, off);
 313
 314         return skb;
 315 }
 316
 317 static inline void vnet_fullcsum(struct sk_buff *skb)
 318 {
 319         struct iphdr *iph = ip_hdr(skb);
 320         int offset = skb_transport_offset(skb);
 321
 322         if (skb->protocol != htons(ETH_P_IP))
 323                 return;
 324         if (iph->protocol != IPPROTO_TCP &&
 325             iph->protocol != IPPROTO_UDP)
 326                 return;
 327         skb->ip_summed = CHECKSUM_NONE;
 328         skb->csum_level = 1;
 329         skb->csum = 0;
 330         if (iph->protocol == IPPROTO_TCP) {
 331                 struct tcphdr *ptcp = tcp_hdr(skb);
 332
 333                 ptcp->check = 0;
 334                 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
 335                 ptcp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 336                                                 skb->len - offset, IPPROTO_TCP,
 337                                                 skb->csum);
 338         } else if (iph->protocol == IPPROTO_UDP) {
 339                 struct udphdr *pudp = udp_hdr(skb);
 340
 341                 pudp->check = 0;
 342                 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
 343                 pudp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 344                                                 skb->len - offset, IPPROTO_UDP,
 345                                                 skb->csum);
 346         }
 347 }
 348
 349 static int vnet_rx_one(struct vnet_port *port, struct vio_net_desc *desc)
 350 {
 351         struct net_device *dev = port->vp->dev;
 352         unsigned int len = desc->size;
 353         unsigned int copy_len;
 354         struct sk_buff *skb;
 355         int maxlen;
 356         int err;
 357
 358         err = -EMSGSIZE;
 359         if (port->tso && port->tsolen > port->rmtu)
 360                 maxlen = port->tsolen;
 361         else
 362                 maxlen = port->rmtu;
 363         if (unlikely(len < ETH_ZLEN || len > maxlen)) {
 364                 dev->stats.rx_length_errors++;
 365                 goto out_dropped;
 366         }
 367
 368         skb = alloc_and_align_skb(dev, len);
 369         err = -ENOMEM;
 370         if (unlikely(!skb)) {
 371                 dev->stats.rx_missed_errors++;
 372                 goto out_dropped;
 373         }
 374
 375         copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
 376         skb_put(skb, copy_len);
 377         err = ldc_copy(port->vio.lp, LDC_COPY_IN,
 378                        skb->data, copy_len, 0,
 379                        desc->cookies, desc->ncookies);
 380         if (unlikely(err < 0)) {
 381                 dev->stats.rx_frame_errors++;
 382                 goto out_free_skb;
 383         }
 384
 385         skb_pull(skb, VNET_PACKET_SKIP);
 386         skb_trim(skb, len);
 387         skb->protocol = eth_type_trans(skb, dev);
 388
 389         if (vio_version_after_eq(&port->vio, 1, 8)) {
 390                 struct vio_net_dext *dext = vio_net_ext(desc);
 391
 392                 skb_reset_network_header(skb);
 393
 394                 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM) {
 395                         if (skb->protocol == ETH_P_IP) {
 396                                 struct iphdr *iph = ip_hdr(skb);
 397
 398                                 iph->check = 0;
 399                                 ip_send_check(iph);
 400                         }
 401                 }
 402                 if ((dext->flags & VNET_PKT_HCK_FULLCKSUM) &&
 403                     skb->ip_summed == CHECKSUM_NONE) {
 404                         if (skb->protocol == htons(ETH_P_IP)) {
 405                                 struct iphdr *iph = ip_hdr(skb);
 406                                 int ihl = iph->ihl * 4;
 407
 408                                 skb_reset_transport_header(skb);
 409                                 skb_set_transport_header(skb, ihl);
 410                                 vnet_fullcsum(skb);
 411                         }
 412                 }
 413                 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM_OK) {
 414                         skb->ip_summed = CHECKSUM_PARTIAL;
 415                         skb->csum_level = 0;
 416                         if (dext->flags & VNET_PKT_HCK_FULLCKSUM_OK)
 417                                 skb->csum_level = 1;
 418                 }
 419         }
 420
 421         skb->ip_summed = port->switch_port ? CHECKSUM_NONE : CHECKSUM_PARTIAL;
 422
 423         dev->stats.rx_packets++;
 424         dev->stats.rx_bytes += len;
 425         napi_gro_receive(&port->napi, skb);
 426         return 0;
 427
 428 out_free_skb:
 429         kfree_skb(skb);
 430
 431 out_dropped:
 432         dev->stats.rx_dropped++;
 433         return err;
 434 }
 435
 436 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
 437                          u32 start, u32 end, u8 vio_dring_state)
 438 {
 439         struct vio_dring_data hdr = {
 440                 .tag = {
 441                         .type           = VIO_TYPE_DATA,
 442                         .stype          = VIO_SUBTYPE_ACK,
 443                         .stype_env      = VIO_DRING_DATA,
 444                         .sid            = vio_send_sid(&port->vio),
 445                 },
 446                 .dring_ident            = dr->ident,
 447                 .start_idx              = start,
 448                 .end_idx                = end,
 449                 .state                  = vio_dring_state,
 450         };
 451         int err, delay;
 452         int retries = 0;
 453
 454         hdr.seq = dr->snd_nxt;
 455         delay = 1;
 456         do {
 457                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 458                 if (err > 0) {
 459                         dr->snd_nxt++;
 460                         break;
 461                 }
 462                 udelay(delay);
 463                 if ((delay <<= 1) > 128)
 464                         delay = 128;
 465                 if (retries++ > VNET_MAX_RETRIES) {
 466                         pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
 467                                 port->raddr[0], port->raddr[1],
 468                                 port->raddr[2], port->raddr[3],
 469                                 port->raddr[4], port->raddr[5]);
 470                         break;
 471                 }
 472         } while (err == -EAGAIN);
 473
 474         if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
 475                 port->stop_rx_idx = end;
 476                 port->stop_rx = true;
 477         } else {
 478                 port->stop_rx_idx = 0;
 479                 port->stop_rx = false;
 480         }
 481
 482         return err;
 483 }
 484
 485 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
 486                                         struct vio_dring_state *dr,
 487                                         u32 index)
 488 {
 489         struct vio_net_desc *desc = port->vio.desc_buf;
 490         int err;
 491
 492         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
 493                                   (index * dr->entry_size),
 494                                   dr->cookies, dr->ncookies);
 495         if (err < 0)
 496                 return ERR_PTR(err);
 497
 498         return desc;
 499 }
 500
 501 static int put_rx_desc(struct vnet_port *port,
 502                        struct vio_dring_state *dr,
 503                        struct vio_net_desc *desc,
 504                        u32 index)
 505 {
 506         int err;
 507
 508         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
 509                                   (index * dr->entry_size),
 510                                   dr->cookies, dr->ncookies);
 511         if (err < 0)
 512                 return err;
 513
 514         return 0;
 515 }
 516
 517 static int vnet_walk_rx_one(struct vnet_port *port,
 518                             struct vio_dring_state *dr,
 519                             u32 index, int *needs_ack)
 520 {
 521         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
 522         struct vio_driver_state *vio = &port->vio;
 523         int err;
 524
 525         BUG_ON(desc == NULL);
 526         if (IS_ERR(desc))
 527                 return PTR_ERR(desc);
 528
 529         if (desc->hdr.state != VIO_DESC_READY)
 530                 return 1;
 531
 532         dma_rmb();
 533
 534         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 535                desc->hdr.state, desc->hdr.ack,
 536                desc->size, desc->ncookies,
 537                desc->cookies[0].cookie_addr,
 538                desc->cookies[0].cookie_size);
 539
 540         err = vnet_rx_one(port, desc);
 541         if (err == -ECONNRESET)
 542                 return err;
 543         desc->hdr.state = VIO_DESC_DONE;
 544         err = put_rx_desc(port, dr, desc, index);
 545         if (err < 0)
 546                 return err;
 547         *needs_ack = desc->hdr.ack;
 548         return 0;
 549 }
 550
 551 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 552                         u32 start, u32 end, int *npkts, int budget)
 553 {
 554         struct vio_driver_state *vio = &port->vio;
 555         int ack_start = -1, ack_end = -1;
 556         bool send_ack = true;
 557
 558         end = (end == (u32) -1) ? vio_dring_prev(dr, start)
 559                                 : vio_dring_next(dr, end);
 560
 561         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
 562
 563         while (start != end) {
 564                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
 565                 if (err == -ECONNRESET)
 566                         return err;
 567                 if (err != 0)
 568                         break;
 569                 (*npkts)++;
 570                 if (ack_start == -1)
 571                         ack_start = start;
 572                 ack_end = start;
 573                 start = vio_dring_next(dr, start);
 574                 if (ack && start != end) {
 575                         err = vnet_send_ack(port, dr, ack_start, ack_end,
 576                                             VIO_DRING_ACTIVE);
 577                         if (err == -ECONNRESET)
 578                                 return err;
 579                         ack_start = -1;
 580                 }
 581                 if ((*npkts) >= budget) {
 582                         send_ack = false;
 583                         break;
 584                 }
 585         }
 586         if (unlikely(ack_start == -1))
 587                 ack_start = ack_end = vio_dring_prev(dr, start);
 588         if (send_ack) {
 589                 port->napi_resume = false;
 590                 return vnet_send_ack(port, dr, ack_start, ack_end,
 591                                      VIO_DRING_STOPPED);
 592         } else  {
 593                 port->napi_resume = true;
 594                 port->napi_stop_idx = ack_end;
 595                 return 1;
 596         }
 597 }
 598
 599 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
 600                    int budget)
 601 {
 602         struct vio_dring_data *pkt = msgbuf;
 603         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
 604         struct vio_driver_state *vio = &port->vio;
 605
 606         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
 607                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
 608
 609         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 610                 return 0;
 611         if (unlikely(pkt->seq != dr->rcv_nxt)) {
 612                 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
 613                        pkt->seq, dr->rcv_nxt);
 614                 return 0;
 615         }
 616
 617         if (!port->napi_resume)
 618                 dr->rcv_nxt++;
 619
 620         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
 621
 622         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
 623                             npkts, budget);
 624 }
 625
 626 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
 627 {
 628         u32 idx = dr->cons;
 629         int found = 0;
 630
 631         while (idx != dr->prod) {
 632                 if (idx == end) {
 633                         found = 1;
 634                         break;
 635                 }
 636                 idx = vio_dring_next(dr, idx);
 637         }
 638         return found;
 639 }
 640
 641 static int vnet_ack(struct vnet_port *port, void *msgbuf)
 642 {
 643         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 644         struct vio_dring_data *pkt = msgbuf;
 645         struct net_device *dev;
 646         struct vnet *vp;
 647         u32 end;
 648         struct vio_net_desc *desc;
 649         struct netdev_queue *txq;
 650
 651         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 652                 return 0;
 653
 654         end = pkt->end_idx;
 655         vp = port->vp;
 656         dev = vp->dev;
 657         netif_tx_lock(dev);
 658         if (unlikely(!idx_is_pending(dr, end))) {
 659                 netif_tx_unlock(dev);
 660                 return 0;
 661         }
 662
 663         /* sync for race conditions with vnet_start_xmit() and tell xmit it
 664          * is time to send a trigger.
 665          */
 666         dr->cons = vio_dring_next(dr, end);
 667         desc = vio_dring_entry(dr, dr->cons);
 668         if (desc->hdr.state == VIO_DESC_READY && !port->start_cons) {
 669                 /* vnet_start_xmit() just populated this dring but missed
 670                  * sending the "start" LDC message to the consumer.
 671                  * Send a "start" trigger on its behalf.
 672                  */
 673                 if (__vnet_tx_trigger(port, dr->cons) > 0)
 674                         port->start_cons = false;
 675                 else
 676                         port->start_cons = true;
 677         } else {
 678                 port->start_cons = true;
 679         }
 680         netif_tx_unlock(dev);
 681
 682         txq = netdev_get_tx_queue(dev, port->q_index);
 683         if (unlikely(netif_tx_queue_stopped(txq) &&
 684                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
 685                 return 1;
 686
 687         return 0;
 688 }
 689
 690 static int vnet_nack(struct vnet_port *port, void *msgbuf)
 691 {
 692         /* XXX just reset or similar XXX */
 693         return 0;
 694 }
 695
 696 static int handle_mcast(struct vnet_port *port, void *msgbuf)
 697 {
 698         struct vio_net_mcast_info *pkt = msgbuf;
 699
 700         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
 701                 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
 702                        port->vp->dev->name,
 703                        pkt->tag.type,
 704                        pkt->tag.stype,
 705                        pkt->tag.stype_env,
 706                        pkt->tag.sid);
 707
 708         return 0;
 709 }
 710
 711 /* Got back a STOPPED LDC message on port. If the queue is stopped,
 712  * wake it up so that we'll send out another START message at the
 713  * next TX.
 714  */
 715 static void maybe_tx_wakeup(struct vnet_port *port)
 716 {
 717         struct netdev_queue *txq;
 718
 719         txq = netdev_get_tx_queue(port->vp->dev, port->q_index);
 720         __netif_tx_lock(txq, smp_processor_id());
 721         if (likely(netif_tx_queue_stopped(txq))) {
 722                 struct vio_dring_state *dr;
 723
 724                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 725                 netif_tx_wake_queue(txq);
 726         }
 727         __netif_tx_unlock(txq);
 728 }
 729
 730 static inline bool port_is_up(struct vnet_port *vnet)
 731 {
 732         struct vio_driver_state *vio = &vnet->vio;
 733
 734         return !!(vio->hs_state & VIO_HS_COMPLETE);
 735 }
 736
 737 static int vnet_event_napi(struct vnet_port *port, int budget)
 738 {
 739         struct vio_driver_state *vio = &port->vio;
 740         int tx_wakeup, err;
 741         int npkts = 0;
 742         int event = (port->rx_event & LDC_EVENT_RESET);
 743
 744 ldc_ctrl:
 745         if (unlikely(event == LDC_EVENT_RESET ||
 746                      event == LDC_EVENT_UP)) {
 747                 vio_link_state_change(vio, event);
 748
 749                 if (event == LDC_EVENT_RESET) {
 750                         vnet_port_reset(port);
 751                         vio_port_up(vio);
 752                 }
 753                 port->rx_event = 0;
 754                 return 0;
 755         }
 756         /* We may have multiple LDC events in rx_event. Unroll send_events() */
 757         event = (port->rx_event & LDC_EVENT_UP);
 758         port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
 759         if (event == LDC_EVENT_UP)
 760                 goto ldc_ctrl;
 761         event = port->rx_event;
 762         if (!(event & LDC_EVENT_DATA_READY))
 763                 return 0;
 764
 765         /* we dont expect any other bits than RESET, UP, DATA_READY */
 766         BUG_ON(event != LDC_EVENT_DATA_READY);
 767
 768         tx_wakeup = err = 0;
 769         while (1) {
 770                 union {
 771                         struct vio_msg_tag tag;
 772                         u64 raw[8];
 773                 } msgbuf;
 774
 775                 if (port->napi_resume) {
 776                         struct vio_dring_data *pkt =
 777                                 (struct vio_dring_data *)&msgbuf;
 778                         struct vio_dring_state *dr =
 779                                 &port->vio.drings[VIO_DRIVER_RX_RING];
 780
 781                         pkt->tag.type = VIO_TYPE_DATA;
 782                         pkt->tag.stype = VIO_SUBTYPE_INFO;
 783                         pkt->tag.stype_env = VIO_DRING_DATA;
 784                         pkt->seq = dr->rcv_nxt;
 785                         pkt->start_idx = vio_dring_next(dr, port->napi_stop_idx);
 786                         pkt->end_idx = -1;
 787                         goto napi_resume;
 788                 }
 789                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 790                 if (unlikely(err < 0)) {
 791                         if (err == -ECONNRESET)
 792                                 vio_conn_reset(vio);
 793                         break;
 794                 }
 795                 if (err == 0)
 796                         break;
 797                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 798                        msgbuf.tag.type,
 799                        msgbuf.tag.stype,
 800                        msgbuf.tag.stype_env,
 801                        msgbuf.tag.sid);
 802                 err = vio_validate_sid(vio, &msgbuf.tag);
 803                 if (err < 0)
 804                         break;
 805 napi_resume:
 806                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 807                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 808                                 if (!port_is_up(port)) {
 809                                         /* failures like handshake_failure()
 810                                          * may have cleaned up dring, but
 811                                          * NAPI polling may bring us here.
 812                                          */
 813                                         err = -ECONNRESET;
 814                                         break;
 815                                 }
 816                                 err = vnet_rx(port, &msgbuf, &npkts, budget);
 817                                 if (npkts >= budget)
 818                                         break;
 819                                 if (npkts == 0)
 820                                         break;
 821                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 822                                 err = vnet_ack(port, &msgbuf);
 823                                 if (err > 0)
 824                                         tx_wakeup |= err;
 825                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 826                                 err = vnet_nack(port, &msgbuf);
 827                         }
 828                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 829                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 830                                 err = handle_mcast(port, &msgbuf);
 831                         else
 832                                 err = vio_control_pkt_engine(vio, &msgbuf);
 833                         if (err)
 834                                 break;
 835                 } else {
 836                         err = vnet_handle_unknown(port, &msgbuf);
 837                 }
 838                 if (err == -ECONNRESET)
 839                         break;
 840         }
 841         if (unlikely(tx_wakeup && err != -ECONNRESET))
 842                 maybe_tx_wakeup(port);
 843         return npkts;
 844 }
 845
 846 static int vnet_poll(struct napi_struct *napi, int budget)
 847 {
 848         struct vnet_port *port = container_of(napi, struct vnet_port, napi);
 849         struct vio_driver_state *vio = &port->vio;
 850         int processed = vnet_event_napi(port, budget);
 851
 852         if (processed < budget) {
 853                 napi_complete(napi);
 854                 port->rx_event &= ~LDC_EVENT_DATA_READY;
 855                 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
 856         }
 857         return processed;
 858 }
 859
 860 static void vnet_event(void *arg, int event)
 861 {
 862         struct vnet_port *port = arg;
 863         struct vio_driver_state *vio = &port->vio;
 864
 865         port->rx_event |= event;
 866         vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
 867         napi_schedule(&port->napi);
 868
 869 }
 870
 871 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
 872 {
 873         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 874         struct vio_dring_data hdr = {
 875                 .tag = {
 876                         .type           = VIO_TYPE_DATA,
 877                         .stype          = VIO_SUBTYPE_INFO,
 878                         .stype_env      = VIO_DRING_DATA,
 879                         .sid            = vio_send_sid(&port->vio),
 880                 },
 881                 .dring_ident            = dr->ident,
 882                 .start_idx              = start,
 883                 .end_idx                = (u32) -1,
 884         };
 885         int err, delay;
 886         int retries = 0;
 887
 888         if (port->stop_rx) {
 889                 err = vnet_send_ack(port,
 890                                     &port->vio.drings[VIO_DRIVER_RX_RING],
 891                                     port->stop_rx_idx, -1,
 892                                     VIO_DRING_STOPPED);
 893                 if (err <= 0)
 894                         return err;
 895         }
 896
 897         hdr.seq = dr->snd_nxt;
 898         delay = 1;
 899         do {
 900                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 901                 if (err > 0) {
 902                         dr->snd_nxt++;
 903                         break;
 904                 }
 905                 udelay(delay);
 906                 if ((delay <<= 1) > 128)
 907                         delay = 128;
 908                 if (retries++ > VNET_MAX_RETRIES)
 909                         break;
 910         } while (err == -EAGAIN);
 911
 912         return err;
 913 }
 914
 915 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 916 {
 917         unsigned int hash = vnet_hashfn(skb->data);
 918         struct hlist_head *hp = &vp->port_hash[hash];
 919         struct vnet_port *port;
 920
 921         hlist_for_each_entry_rcu(port, hp, hash) {
 922                 if (!port_is_up(port))
 923                         continue;
 924                 if (ether_addr_equal(port->raddr, skb->data))
 925                         return port;
 926         }
 927         list_for_each_entry_rcu(port, &vp->port_list, list) {
 928                 if (!port->switch_port)
 929                         continue;
 930                 if (!port_is_up(port))
 931                         continue;
 932                 return port;
 933         }
 934         return NULL;
 935 }
 936
 937 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
 938                                           unsigned *pending)
 939 {
 940         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 941         struct sk_buff *skb = NULL;
 942         int i, txi;
 943
 944         *pending = 0;
 945
 946         txi = dr->prod;
 947         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
 948                 struct vio_net_desc *d;
 949
 950                 --txi;
 951                 if (txi < 0)
 952                         txi = VNET_TX_RING_SIZE-1;
 953
 954                 d = vio_dring_entry(dr, txi);
 955
 956                 if (d->hdr.state == VIO_DESC_READY) {
 957                         (*pending)++;
 958                         continue;
 959                 }
 960                 if (port->tx_bufs[txi].skb) {
 961                         if (d->hdr.state != VIO_DESC_DONE)
 962                                 pr_notice("invalid ring buffer state %d\n",
 963                                           d->hdr.state);
 964                         BUG_ON(port->tx_bufs[txi].skb->next);
 965
 966                         port->tx_bufs[txi].skb->next = skb;
 967                         skb = port->tx_bufs[txi].skb;
 968                         port->tx_bufs[txi].skb = NULL;
 969
 970                         ldc_unmap(port->vio.lp,
 971                                   port->tx_bufs[txi].cookies,
 972                                   port->tx_bufs[txi].ncookies);
 973                 } else if (d->hdr.state == VIO_DESC_FREE)
 974                         break;
 975                 d->hdr.state = VIO_DESC_FREE;
 976         }
 977         return skb;
 978 }
 979
 980 static inline void vnet_free_skbs(struct sk_buff *skb)
 981 {
 982         struct sk_buff *next;
 983
 984         while (skb) {
 985                 next = skb->next;
 986                 skb->next = NULL;
 987                 dev_kfree_skb(skb);
 988                 skb = next;
 989         }
 990 }
 991
 992 static void vnet_clean_timer_expire(unsigned long port0)
 993 {
 994         struct vnet_port *port = (struct vnet_port *)port0;
 995         struct sk_buff *freeskbs;
 996         unsigned pending;
 997
 998         netif_tx_lock(port->vp->dev);
 999         freeskbs = vnet_clean_tx_ring(port, &pending);
1000         netif_tx_unlock(port->vp->dev);
1001
1002         vnet_free_skbs(freeskbs);
1003
1004         if (pending)
1005                 (void)mod_timer(&port->clean_timer,
1006                                 jiffies + VNET_CLEAN_TIMEOUT);
1007          else
1008                 del_timer(&port->clean_timer);
1009 }
1010
1011 static inline int vnet_skb_map(struct ldc_channel *lp, struct sk_buff *skb,
1012                                struct ldc_trans_cookie *cookies, int ncookies,
1013                                unsigned int map_perm)
1014 {
1015         int i, nc, err, blen;
1016
1017         /* header */
1018         blen = skb_headlen(skb);
1019         if (blen < ETH_ZLEN)
1020                 blen = ETH_ZLEN;
1021         blen += VNET_PACKET_SKIP;
1022         blen += 8 - (blen & 7);
1023
1024         err = ldc_map_single(lp, skb->data-VNET_PACKET_SKIP, blen, cookies,
1025                              ncookies, map_perm);
1026         if (err < 0)
1027                 return err;
1028         nc = err;
1029
1030         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1031                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1032                 u8 *vaddr;
1033
1034                 if (nc < ncookies) {
1035                         vaddr = kmap_atomic(skb_frag_page(f));
1036                         blen = skb_frag_size(f);
1037                         blen += 8 - (blen & 7);
1038                         err = ldc_map_single(lp, vaddr + f->page_offset,
1039                                              blen, cookies + nc, ncookies - nc,
1040                                              map_perm);
1041                         kunmap_atomic(vaddr);
1042                 } else {
1043                         err = -EMSGSIZE;
1044                 }
1045
1046                 if (err < 0) {
1047                         ldc_unmap(lp, cookies, nc);
1048                         return err;
1049                 }
1050                 nc += err;
1051         }
1052         return nc;
1053 }
1054
1055 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, int ncookies)
1056 {
1057         struct sk_buff *nskb;
1058         int i, len, pad, docopy;
1059
1060         len = skb->len;
1061         pad = 0;
1062         if (len < ETH_ZLEN) {
1063                 pad += ETH_ZLEN - skb->len;
1064                 len += pad;
1065         }
1066         len += VNET_PACKET_SKIP;
1067         pad += 8 - (len & 7);
1068
1069         /* make sure we have enough cookies and alignment in every frag */
1070         docopy = skb_shinfo(skb)->nr_frags >= ncookies;
1071         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1072                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1073
1074                 docopy |= f->page_offset & 7;
1075         }
1076         if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
1077             skb_tailroom(skb) < pad ||
1078             skb_headroom(skb) < VNET_PACKET_SKIP || docopy) {
1079                 int start = 0, offset;
1080                 __wsum csum;
1081
1082                 len = skb->len > ETH_ZLEN ? skb->len : ETH_ZLEN;
1083                 nskb = alloc_and_align_skb(skb->dev, len);
1084                 if (nskb == NULL) {
1085                         dev_kfree_skb(skb);
1086                         return NULL;
1087                 }
1088                 skb_reserve(nskb, VNET_PACKET_SKIP);
1089
1090                 nskb->protocol = skb->protocol;
1091                 offset = skb_mac_header(skb) - skb->data;
1092                 skb_set_mac_header(nskb, offset);
1093                 offset = skb_network_header(skb) - skb->data;
1094                 skb_set_network_header(nskb, offset);
1095                 offset = skb_transport_header(skb) - skb->data;
1096                 skb_set_transport_header(nskb, offset);
1097
1098                 offset = 0;
1099                 nskb->csum_offset = skb->csum_offset;
1100                 nskb->ip_summed = skb->ip_summed;
1101
1102                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1103                         start = skb_checksum_start_offset(skb);
1104                 if (start) {
1105                         struct iphdr *iph = ip_hdr(nskb);
1106                         int offset = start + nskb->csum_offset;
1107
1108                         if (skb_copy_bits(skb, 0, nskb->data, start)) {
1109                                 dev_kfree_skb(nskb);
1110                                 dev_kfree_skb(skb);
1111                                 return NULL;
1112                         }
1113                         *(__sum16 *)(skb->data + offset) = 0;
1114                         csum = skb_copy_and_csum_bits(skb, start,
1115                                                       nskb->data + start,
1116                                                       skb->len - start, 0);
1117                         if (iph->protocol == IPPROTO_TCP ||
1118                             iph->protocol == IPPROTO_UDP) {
1119                                 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
1120                                                          skb->len - start,
1121                                                          iph->protocol, csum);
1122                         }
1123                         *(__sum16 *)(nskb->data + offset) = csum;
1124
1125                         nskb->ip_summed = CHECKSUM_NONE;
1126                 } else if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1127                         dev_kfree_skb(nskb);
1128                         dev_kfree_skb(skb);
1129                         return NULL;
1130                 }
1131                 (void)skb_put(nskb, skb->len);
1132                 if (skb_is_gso(skb)) {
1133                         skb_shinfo(nskb)->gso_size = skb_shinfo(skb)->gso_size;
1134                         skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
1135                 }
1136                 nskb->queue_mapping = skb->queue_mapping;
1137                 dev_kfree_skb(skb);
1138                 skb = nskb;
1139         }
1140         return skb;
1141 }
1142
1143 static u16
1144 vnet_select_queue(struct net_device *dev, struct sk_buff *skb,
1145                   void *accel_priv, select_queue_fallback_t fallback)
1146 {
1147         struct vnet *vp = netdev_priv(dev);
1148         struct vnet_port *port = __tx_port_find(vp, skb);
1149
1150         if (port == NULL)
1151                 return 0;
1152         return port->q_index;
1153 }
1154
1155 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev);
1156
1157 static int vnet_handle_offloads(struct vnet_port *port, struct sk_buff *skb)
1158 {
1159         struct net_device *dev = port->vp->dev;
1160         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1161         struct sk_buff *segs;
1162         int maclen, datalen;
1163         int status;
1164         int gso_size, gso_type, gso_segs;
1165         int hlen = skb_transport_header(skb) - skb_mac_header(skb);
1166         int proto = IPPROTO_IP;
1167
1168         if (skb->protocol == htons(ETH_P_IP))
1169                 proto = ip_hdr(skb)->protocol;
1170         else if (skb->protocol == htons(ETH_P_IPV6))
1171                 proto = ipv6_hdr(skb)->nexthdr;
1172
1173         if (proto == IPPROTO_TCP)
1174                 hlen += tcp_hdr(skb)->doff * 4;
1175         else if (proto == IPPROTO_UDP)
1176                 hlen += sizeof(struct udphdr);
1177         else {
1178                 pr_err("vnet_handle_offloads GSO with unknown transport "
1179                        "protocol %d tproto %d\n", skb->protocol, proto);
1180                 hlen = 128; /* XXX */
1181         }
1182         datalen = port->tsolen - hlen;
1183
1184         gso_size = skb_shinfo(skb)->gso_size;
1185         gso_type = skb_shinfo(skb)->gso_type;
1186         gso_segs = skb_shinfo(skb)->gso_segs;
1187
1188         if (port->tso && gso_size < datalen)
1189                 gso_segs = DIV_ROUND_UP(skb->len - hlen, datalen);
1190
1191         if (unlikely(vnet_tx_dring_avail(dr) < gso_segs)) {
1192                 struct netdev_queue *txq;
1193
1194                 txq  = netdev_get_tx_queue(dev, port->q_index);
1195                 netif_tx_stop_queue(txq);
1196                 if (vnet_tx_dring_avail(dr) < skb_shinfo(skb)->gso_segs)
1197                         return NETDEV_TX_BUSY;
1198                 netif_tx_wake_queue(txq);
1199         }
1200
1201         maclen = skb_network_header(skb) - skb_mac_header(skb);
1202         skb_pull(skb, maclen);
1203
1204         if (port->tso && gso_size < datalen) {
1205                 if (skb_unclone(skb, GFP_ATOMIC))
1206                         goto out_dropped;
1207
1208                 /* segment to TSO size */
1209                 skb_shinfo(skb)->gso_size = datalen;
1210                 skb_shinfo(skb)->gso_segs = gso_segs;
1211         }
1212         segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1213         if (IS_ERR(segs))
1214                 goto out_dropped;
1215
1216         skb_push(skb, maclen);
1217         skb_reset_mac_header(skb);
1218
1219         status = 0;
1220         while (segs) {
1221                 struct sk_buff *curr = segs;
1222
1223                 segs = segs->next;
1224                 curr->next = NULL;
1225                 if (port->tso && curr->len > dev->mtu) {
1226                         skb_shinfo(curr)->gso_size = gso_size;
1227                         skb_shinfo(curr)->gso_type = gso_type;
1228                         skb_shinfo(curr)->gso_segs =
1229                                 DIV_ROUND_UP(curr->len - hlen, gso_size);
1230                 } else
1231                         skb_shinfo(curr)->gso_size = 0;
1232
1233                 skb_push(curr, maclen);
1234                 skb_reset_mac_header(curr);
1235                 memcpy(skb_mac_header(curr), skb_mac_header(skb),
1236                        maclen);
1237                 curr->csum_start = skb_transport_header(curr) - curr->head;
1238                 if (ip_hdr(curr)->protocol == IPPROTO_TCP)
1239                         curr->csum_offset = offsetof(struct tcphdr, check);
1240                 else if (ip_hdr(curr)->protocol == IPPROTO_UDP)
1241                         curr->csum_offset = offsetof(struct udphdr, check);
1242
1243                 if (!(status & NETDEV_TX_MASK))
1244                         status = vnet_start_xmit(curr, dev);
1245                 if (status & NETDEV_TX_MASK)
1246                         dev_kfree_skb_any(curr);
1247         }
1248
1249         if (!(status & NETDEV_TX_MASK))
1250                 dev_kfree_skb_any(skb);
1251         return status;
1252 out_dropped:
1253         dev->stats.tx_dropped++;
1254         dev_kfree_skb_any(skb);
1255         return NETDEV_TX_OK;
1256 }
1257
1258 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
1259 {
1260         struct vnet *vp = netdev_priv(dev);
1261         struct vnet_port *port = NULL;
1262         struct vio_dring_state *dr;
1263         struct vio_net_desc *d;
1264         unsigned int len;
1265         struct sk_buff *freeskbs = NULL;
1266         int i, err, txi;
1267         unsigned pending = 0;
1268         struct netdev_queue *txq;
1269
1270         rcu_read_lock();
1271         port = __tx_port_find(vp, skb);
1272         if (unlikely(!port)) {
1273                 rcu_read_unlock();
1274                 goto out_dropped;
1275         }
1276
1277         if (skb_is_gso(skb) && skb->len > port->tsolen) {
1278                 err = vnet_handle_offloads(port, skb);
1279                 rcu_read_unlock();
1280                 return err;
1281         }
1282
1283         if (!skb_is_gso(skb) && skb->len > port->rmtu) {
1284                 unsigned long localmtu = port->rmtu - ETH_HLEN;
1285
1286                 if (vio_version_after_eq(&port->vio, 1, 3))
1287                         localmtu -= VLAN_HLEN;
1288
1289                 if (skb->protocol == htons(ETH_P_IP)) {
1290                         struct flowi4 fl4;
1291                         struct rtable *rt = NULL;
1292
1293                         memset(&fl4, 0, sizeof(fl4));
1294                         fl4.flowi4_oif = dev->ifindex;
1295                         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
1296                         fl4.daddr = ip_hdr(skb)->daddr;
1297                         fl4.saddr = ip_hdr(skb)->saddr;
1298
1299                         rt = ip_route_output_key(dev_net(dev), &fl4);
1300                         rcu_read_unlock();
1301                         if (!IS_ERR(rt)) {
1302                                 skb_dst_set(skb, &rt->dst);
1303                                 icmp_send(skb, ICMP_DEST_UNREACH,
1304                                           ICMP_FRAG_NEEDED,
1305                                           htonl(localmtu));
1306                         }
1307                 }
1308 #if IS_ENABLED(CONFIG_IPV6)
1309                 else if (skb->protocol == htons(ETH_P_IPV6))
1310                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1311 #endif
1312                 goto out_dropped;
1313         }
1314
1315         skb = vnet_skb_shape(skb, 2);
1316
1317         if (unlikely(!skb))
1318                 goto out_dropped;
1319
1320         if (skb->ip_summed == CHECKSUM_PARTIAL)
1321                 vnet_fullcsum(skb);
1322
1323         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1324         i = skb_get_queue_mapping(skb);
1325         txq = netdev_get_tx_queue(dev, i);
1326         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1327                 if (!netif_tx_queue_stopped(txq)) {
1328                         netif_tx_stop_queue(txq);
1329
1330                         /* This is a hard error, log it. */
1331                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1332                         dev->stats.tx_errors++;
1333                 }
1334                 rcu_read_unlock();
1335                 return NETDEV_TX_BUSY;
1336         }
1337
1338         d = vio_dring_cur(dr);
1339
1340         txi = dr->prod;
1341
1342         freeskbs = vnet_clean_tx_ring(port, &pending);
1343
1344         BUG_ON(port->tx_bufs[txi].skb);
1345
1346         len = skb->len;
1347         if (len < ETH_ZLEN)
1348                 len = ETH_ZLEN;
1349
1350         err = vnet_skb_map(port->vio.lp, skb, port->tx_bufs[txi].cookies, 2,
1351                            (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1352         if (err < 0) {
1353                 netdev_info(dev, "tx buffer map error %d\n", err);
1354                 goto out_dropped;
1355         }
1356
1357         port->tx_bufs[txi].skb = skb;
1358         skb = NULL;
1359         port->tx_bufs[txi].ncookies = err;
1360
1361         /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1362          * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1363          * the protocol itself does not require it as long as the peer
1364          * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1365          *
1366          * An ACK for every packet in the ring is expensive as the
1367          * sending of LDC messages is slow and affects performance.
1368          */
1369         d->hdr.ack = VIO_ACK_DISABLE;
1370         d->size = len;
1371         d->ncookies = port->tx_bufs[txi].ncookies;
1372         for (i = 0; i < d->ncookies; i++)
1373                 d->cookies[i] = port->tx_bufs[txi].cookies[i];
1374         if (vio_version_after_eq(&port->vio, 1, 7)) {
1375                 struct vio_net_dext *dext = vio_net_ext(d);
1376
1377                 memset(dext, 0, sizeof(*dext));
1378                 if (skb_is_gso(port->tx_bufs[txi].skb)) {
1379                         dext->ipv4_lso_mss = skb_shinfo(port->tx_bufs[txi].skb)
1380                                              ->gso_size;
1381                         dext->flags |= VNET_PKT_IPV4_LSO;
1382                 }
1383                 if (vio_version_after_eq(&port->vio, 1, 8) &&
1384                     !port->switch_port) {
1385                         dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1386                         dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1387                 }
1388         }
1389
1390         /* This has to be a non-SMP write barrier because we are writing
1391          * to memory which is shared with the peer LDOM.
1392          */
1393         dma_wmb();
1394
1395         d->hdr.state = VIO_DESC_READY;
1396
1397         /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1398          * to notify the consumer that some descriptors are READY.
1399          * After that "start" trigger, no additional triggers are needed until
1400          * a DRING_STOPPED is received from the consumer. The dr->cons field
1401          * (set up by vnet_ack()) has the value of the next dring index
1402          * that has not yet been ack-ed. We send a "start" trigger here
1403          * if, and only if, start_cons is true (reset it afterward). Conversely,
1404          * vnet_ack() should check if the dring corresponding to cons
1405          * is marked READY, but start_cons was false.
1406          * If so, vnet_ack() should send out the missed "start" trigger.
1407          *
1408          * Note that the dma_wmb() above makes sure the cookies et al. are
1409          * not globally visible before the VIO_DESC_READY, and that the
1410          * stores are ordered correctly by the compiler. The consumer will
1411          * not proceed until the VIO_DESC_READY is visible assuring that
1412          * the consumer does not observe anything related to descriptors
1413          * out of order. The HV trap from the LDC start trigger is the
1414          * producer to consumer announcement that work is available to the
1415          * consumer
1416          */
1417         if (!port->start_cons)
1418                 goto ldc_start_done; /* previous trigger suffices */
1419
1420         err = __vnet_tx_trigger(port, dr->cons);
1421         if (unlikely(err < 0)) {
1422                 netdev_info(dev, "TX trigger error %d\n", err);
1423                 d->hdr.state = VIO_DESC_FREE;
1424                 skb = port->tx_bufs[txi].skb;
1425                 port->tx_bufs[txi].skb = NULL;
1426                 dev->stats.tx_carrier_errors++;
1427                 goto out_dropped;
1428         }
1429
1430 ldc_start_done:
1431         port->start_cons = false;
1432
1433         dev->stats.tx_packets++;
1434         dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1435
1436         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1437         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1438                 netif_tx_stop_queue(txq);
1439                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1440                         netif_tx_wake_queue(txq);
1441         }
1442
1443         (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1444         rcu_read_unlock();
1445
1446         vnet_free_skbs(freeskbs);
1447
1448         return NETDEV_TX_OK;
1449
1450 out_dropped:
1451         if (pending)
1452                 (void)mod_timer(&port->clean_timer,
1453                                 jiffies + VNET_CLEAN_TIMEOUT);
1454         else if (port)
1455                 del_timer(&port->clean_timer);
1456         if (port)
1457                 rcu_read_unlock();
1458         if (skb)
1459                 dev_kfree_skb(skb);
1460         vnet_free_skbs(freeskbs);
1461         dev->stats.tx_dropped++;
1462         return NETDEV_TX_OK;
1463 }
1464
1465 static void vnet_tx_timeout(struct net_device *dev)
1466 {
1467         /* XXX Implement me XXX */
1468 }
1469
1470 static int vnet_open(struct net_device *dev)
1471 {
1472         netif_carrier_on(dev);
1473         netif_tx_start_all_queues(dev);
1474
1475         return 0;
1476 }
1477
1478 static int vnet_close(struct net_device *dev)
1479 {
1480         netif_tx_stop_all_queues(dev);
1481         netif_carrier_off(dev);
1482
1483         return 0;
1484 }
1485
1486 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1487 {
1488         struct vnet_mcast_entry *m;
1489
1490         for (m = vp->mcast_list; m; m = m->next) {
1491                 if (ether_addr_equal(m->addr, addr))
1492                         return m;
1493         }
1494         return NULL;
1495 }
1496
1497 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1498 {
1499         struct netdev_hw_addr *ha;
1500
1501         netdev_for_each_mc_addr(ha, dev) {
1502                 struct vnet_mcast_entry *m;
1503
1504                 m = __vnet_mc_find(vp, ha->addr);
1505                 if (m) {
1506                         m->hit = 1;
1507                         continue;
1508                 }
1509
1510                 if (!m) {
1511                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
1512                         if (!m)
1513                                 continue;
1514                         memcpy(m->addr, ha->addr, ETH_ALEN);
1515                         m->hit = 1;
1516
1517                         m->next = vp->mcast_list;
1518                         vp->mcast_list = m;
1519                 }
1520         }
1521 }
1522
1523 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1524 {
1525         struct vio_net_mcast_info info;
1526         struct vnet_mcast_entry *m, **pp;
1527         int n_addrs;
1528
1529         memset(&info, 0, sizeof(info));
1530
1531         info.tag.type = VIO_TYPE_CTRL;
1532         info.tag.stype = VIO_SUBTYPE_INFO;
1533         info.tag.stype_env = VNET_MCAST_INFO;
1534         info.tag.sid = vio_send_sid(&port->vio);
1535         info.set = 1;
1536
1537         n_addrs = 0;
1538         for (m = vp->mcast_list; m; m = m->next) {
1539                 if (m->sent)
1540                         continue;
1541                 m->sent = 1;
1542                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1543                        m->addr, ETH_ALEN);
1544                 if (++n_addrs == VNET_NUM_MCAST) {
1545                         info.count = n_addrs;
1546
1547                         (void) vio_ldc_send(&port->vio, &info,
1548                                             sizeof(info));
1549                         n_addrs = 0;
1550                 }
1551         }
1552         if (n_addrs) {
1553                 info.count = n_addrs;
1554                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1555         }
1556
1557         info.set = 0;
1558
1559         n_addrs = 0;
1560         pp = &vp->mcast_list;
1561         while ((m = *pp) != NULL) {
1562                 if (m->hit) {
1563                         m->hit = 0;
1564                         pp = &m->next;
1565                         continue;
1566                 }
1567
1568                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1569                        m->addr, ETH_ALEN);
1570                 if (++n_addrs == VNET_NUM_MCAST) {
1571                         info.count = n_addrs;
1572                         (void) vio_ldc_send(&port->vio, &info,
1573                                             sizeof(info));
1574                         n_addrs = 0;
1575                 }
1576
1577                 *pp = m->next;
1578                 kfree(m);
1579         }
1580         if (n_addrs) {
1581                 info.count = n_addrs;
1582                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1583         }
1584 }
1585
1586 static void vnet_set_rx_mode(struct net_device *dev)
1587 {
1588         struct vnet *vp = netdev_priv(dev);
1589         struct vnet_port *port;
1590
1591         rcu_read_lock();
1592         list_for_each_entry_rcu(port, &vp->port_list, list) {
1593
1594                 if (port->switch_port) {
1595                         __update_mc_list(vp, dev);
1596                         __send_mc_list(vp, port);
1597                         break;
1598                 }
1599         }
1600         rcu_read_unlock();
1601 }
1602
1603 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1604 {
1605         if (new_mtu < 68 || new_mtu > 65535)
1606                 return -EINVAL;
1607
1608         dev->mtu = new_mtu;
1609         return 0;
1610 }
1611
1612 static int vnet_set_mac_addr(struct net_device *dev, void *p)
1613 {
1614         return -EINVAL;
1615 }
1616
1617 static void vnet_get_drvinfo(struct net_device *dev,
1618                              struct ethtool_drvinfo *info)
1619 {
1620         strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1621         strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
1622 }
1623
1624 static u32 vnet_get_msglevel(struct net_device *dev)
1625 {
1626         struct vnet *vp = netdev_priv(dev);
1627         return vp->msg_enable;
1628 }
1629
1630 static void vnet_set_msglevel(struct net_device *dev, u32 value)
1631 {
1632         struct vnet *vp = netdev_priv(dev);
1633         vp->msg_enable = value;
1634 }
1635
1636 static const struct ethtool_ops vnet_ethtool_ops = {
1637         .get_drvinfo            = vnet_get_drvinfo,
1638         .get_msglevel           = vnet_get_msglevel,
1639         .set_msglevel           = vnet_set_msglevel,
1640         .get_link               = ethtool_op_get_link,
1641 };
1642
1643 static void vnet_port_free_tx_bufs(struct vnet_port *port)
1644 {
1645         struct vio_dring_state *dr;
1646         int i;
1647
1648         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1649
1650         if (dr->base == NULL)
1651                 return;
1652
1653         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1654                 struct vio_net_desc *d;
1655                 void *skb = port->tx_bufs[i].skb;
1656
1657                 if (!skb)
1658                         continue;
1659
1660                 d = vio_dring_entry(dr, i);
1661
1662                 ldc_unmap(port->vio.lp,
1663                           port->tx_bufs[i].cookies,
1664                           port->tx_bufs[i].ncookies);
1665                 dev_kfree_skb(skb);
1666                 port->tx_bufs[i].skb = NULL;
1667                 d->hdr.state = VIO_DESC_FREE;
1668         }
1669         ldc_free_exp_dring(port->vio.lp, dr->base,
1670                            (dr->entry_size * dr->num_entries),
1671                            dr->cookies, dr->ncookies);
1672         dr->base = NULL;
1673         dr->entry_size = 0;
1674         dr->num_entries = 0;
1675         dr->pending = 0;
1676         dr->ncookies = 0;
1677 }
1678
1679 static void vnet_port_reset(struct vnet_port *port)
1680 {
1681         del_timer(&port->clean_timer);
1682         vnet_port_free_tx_bufs(port);
1683         port->rmtu = 0;
1684         port->tso = true;
1685         port->tsolen = 0;
1686 }
1687
1688 static int vnet_port_alloc_tx_ring(struct vnet_port *port)
1689 {
1690         struct vio_dring_state *dr;
1691         unsigned long len, elen;
1692         int i, err, ncookies;
1693         void *dring;
1694
1695         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1696
1697         elen = sizeof(struct vio_net_desc) +
1698                sizeof(struct ldc_trans_cookie) * 2;
1699         if (vio_version_after_eq(&port->vio, 1, 7))
1700                 elen += sizeof(struct vio_net_dext);
1701         len = VNET_TX_RING_SIZE * elen;
1702
1703         ncookies = VIO_MAX_RING_COOKIES;
1704         dring = ldc_alloc_exp_dring(port->vio.lp, len,
1705                                     dr->cookies, &ncookies,
1706                                     (LDC_MAP_SHADOW |
1707                                      LDC_MAP_DIRECT |
1708                                      LDC_MAP_RW));
1709         if (IS_ERR(dring)) {
1710                 err = PTR_ERR(dring);
1711                 goto err_out;
1712         }
1713
1714         dr->base = dring;
1715         dr->entry_size = elen;
1716         dr->num_entries = VNET_TX_RING_SIZE;
1717         dr->prod = dr->cons = 0;
1718         port->start_cons  = true; /* need an initial trigger */
1719         dr->pending = VNET_TX_RING_SIZE;
1720         dr->ncookies = ncookies;
1721
1722         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1723                 struct vio_net_desc *d;
1724
1725                 d = vio_dring_entry(dr, i);
1726                 d->hdr.state = VIO_DESC_FREE;
1727         }
1728         return 0;
1729
1730 err_out:
1731         vnet_port_free_tx_bufs(port);
1732
1733         return err;
1734 }
1735
1736 #ifdef CONFIG_NET_POLL_CONTROLLER
1737 static void vnet_poll_controller(struct net_device *dev)
1738 {
1739         struct vnet *vp = netdev_priv(dev);
1740         struct vnet_port *port;
1741         unsigned long flags;
1742
1743         spin_lock_irqsave(&vp->lock, flags);
1744         if (!list_empty(&vp->port_list)) {
1745                 port = list_entry(vp->port_list.next, struct vnet_port, list);
1746                 napi_schedule(&port->napi);
1747         }
1748         spin_unlock_irqrestore(&vp->lock, flags);
1749 }
1750 #endif
1751 static LIST_HEAD(vnet_list);
1752 static DEFINE_MUTEX(vnet_list_mutex);
1753
1754 static const struct net_device_ops vnet_ops = {
1755         .ndo_open               = vnet_open,
1756         .ndo_stop               = vnet_close,
1757         .ndo_set_rx_mode        = vnet_set_rx_mode,
1758         .ndo_set_mac_address    = vnet_set_mac_addr,
1759         .ndo_validate_addr      = eth_validate_addr,
1760         .ndo_tx_timeout         = vnet_tx_timeout,
1761         .ndo_change_mtu         = vnet_change_mtu,
1762         .ndo_start_xmit         = vnet_start_xmit,
1763         .ndo_select_queue       = vnet_select_queue,
1764 #ifdef CONFIG_NET_POLL_CONTROLLER
1765         .ndo_poll_controller    = vnet_poll_controller,
1766 #endif
1767 };
1768
1769 static struct vnet *vnet_new(const u64 *local_mac,
1770                              struct vio_dev *vdev)
1771 {
1772         struct net_device *dev;
1773         struct vnet *vp;
1774         int err, i;
1775
1776         dev = alloc_etherdev_mqs(sizeof(*vp), VNET_MAX_TXQS, 1);
1777         if (!dev)
1778                 return ERR_PTR(-ENOMEM);
1779         dev->needed_headroom = VNET_PACKET_SKIP + 8;
1780         dev->needed_tailroom = 8;
1781
1782         for (i = 0; i < ETH_ALEN; i++)
1783                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1784
1785         vp = netdev_priv(dev);
1786
1787         spin_lock_init(&vp->lock);
1788         vp->dev = dev;
1789
1790         INIT_LIST_HEAD(&vp->port_list);
1791         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1792                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1793         INIT_LIST_HEAD(&vp->list);
1794         vp->local_mac = *local_mac;
1795
1796         dev->netdev_ops = &vnet_ops;
1797         dev->ethtool_ops = &vnet_ethtool_ops;
1798         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1799
1800         dev->hw_features = NETIF_F_TSO | NETIF_F_GSO | NETIF_F_GSO_SOFTWARE |
1801                            NETIF_F_HW_CSUM | NETIF_F_SG;
1802         dev->features = dev->hw_features;
1803
1804         SET_NETDEV_DEV(dev, &vdev->dev);
1805
1806         err = register_netdev(dev);
1807         if (err) {
1808                 pr_err("Cannot register net device, aborting\n");
1809                 goto err_out_free_dev;
1810         }
1811
1812         netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1813
1814         list_add(&vp->list, &vnet_list);
1815
1816         return vp;
1817
1818 err_out_free_dev:
1819         free_netdev(dev);
1820
1821         return ERR_PTR(err);
1822 }
1823
1824 static struct vnet *vnet_find_or_create(const u64 *local_mac,
1825                                         struct vio_dev *vdev)
1826 {
1827         struct vnet *iter, *vp;
1828
1829         mutex_lock(&vnet_list_mutex);
1830         vp = NULL;
1831         list_for_each_entry(iter, &vnet_list, list) {
1832                 if (iter->local_mac == *local_mac) {
1833                         vp = iter;
1834                         break;
1835                 }
1836         }
1837         if (!vp)
1838                 vp = vnet_new(local_mac, vdev);
1839         mutex_unlock(&vnet_list_mutex);
1840
1841         return vp;
1842 }
1843
1844 static void vnet_cleanup(void)
1845 {
1846         struct vnet *vp;
1847         struct net_device *dev;
1848
1849         mutex_lock(&vnet_list_mutex);
1850         while (!list_empty(&vnet_list)) {
1851                 vp = list_first_entry(&vnet_list, struct vnet, list);
1852                 list_del(&vp->list);
1853                 dev = vp->dev;
1854                 /* vio_unregister_driver() should have cleaned up port_list */
1855                 BUG_ON(!list_empty(&vp->port_list));
1856                 unregister_netdev(dev);
1857                 free_netdev(dev);
1858         }
1859         mutex_unlock(&vnet_list_mutex);
1860 }
1861
1862 static const char *local_mac_prop = "local-mac-address";
1863
1864 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1865                                      u64 port_node,
1866                                      struct vio_dev *vdev)
1867 {
1868         const u64 *local_mac = NULL;
1869         u64 a;
1870
1871         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1872                 u64 target = mdesc_arc_target(hp, a);
1873                 const char *name;
1874
1875                 name = mdesc_get_property(hp, target, "name", NULL);
1876                 if (!name || strcmp(name, "network"))
1877                         continue;
1878
1879                 local_mac = mdesc_get_property(hp, target,
1880                                                local_mac_prop, NULL);
1881                 if (local_mac)
1882                         break;
1883         }
1884         if (!local_mac)
1885                 return ERR_PTR(-ENODEV);
1886
1887         return vnet_find_or_create(local_mac, vdev);
1888 }
1889
1890 static struct ldc_channel_config vnet_ldc_cfg = {
1891         .event          = vnet_event,
1892         .mtu            = 64,
1893         .mode           = LDC_MODE_UNRELIABLE,
1894 };
1895
1896 static struct vio_driver_ops vnet_vio_ops = {
1897         .send_attr              = vnet_send_attr,
1898         .handle_attr            = vnet_handle_attr,
1899         .handshake_complete     = vnet_handshake_complete,
1900 };
1901
1902 static void print_version(void)
1903 {
1904         printk_once(KERN_INFO "%s", version);
1905 }
1906
1907 const char *remote_macaddr_prop = "remote-mac-address";
1908
1909 static void
1910 vnet_port_add_txq(struct vnet_port *port)
1911 {
1912         struct vnet *vp = port->vp;
1913         int n;
1914
1915         n = vp->nports++;
1916         n = n & (VNET_MAX_TXQS - 1);
1917         port->q_index = n;
1918         netif_tx_wake_queue(netdev_get_tx_queue(vp->dev, port->q_index));
1919 }
1920
1921 static void
1922 vnet_port_rm_txq(struct vnet_port *port)
1923 {
1924         port->vp->nports--;
1925         netif_tx_stop_queue(netdev_get_tx_queue(port->vp->dev, port->q_index));
1926 }
1927
1928 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1929 {
1930         struct mdesc_handle *hp;
1931         struct vnet_port *port;
1932         unsigned long flags;
1933         struct vnet *vp;
1934         const u64 *rmac;
1935         int len, i, err, switch_port;
1936
1937         print_version();
1938
1939         hp = mdesc_grab();
1940
1941         vp = vnet_find_parent(hp, vdev->mp, vdev);
1942         if (IS_ERR(vp)) {
1943                 pr_err("Cannot find port parent vnet\n");
1944                 err = PTR_ERR(vp);
1945                 goto err_out_put_mdesc;
1946         }
1947
1948         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1949         err = -ENODEV;
1950         if (!rmac) {
1951                 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1952                 goto err_out_put_mdesc;
1953         }
1954
1955         port = kzalloc(sizeof(*port), GFP_KERNEL);
1956         err = -ENOMEM;
1957         if (!port)
1958                 goto err_out_put_mdesc;
1959
1960         for (i = 0; i < ETH_ALEN; i++)
1961                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1962
1963         port->vp = vp;
1964
1965         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1966                               vnet_versions, ARRAY_SIZE(vnet_versions),
1967                               &vnet_vio_ops, vp->dev->name);
1968         if (err)
1969                 goto err_out_free_port;
1970
1971         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1972         if (err)
1973                 goto err_out_free_port;
1974
1975         netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1976
1977         INIT_HLIST_NODE(&port->hash);
1978         INIT_LIST_HEAD(&port->list);
1979
1980         switch_port = 0;
1981         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1982                 switch_port = 1;
1983         port->switch_port = switch_port;
1984         port->tso = true;
1985         port->tsolen = 0;
1986
1987         spin_lock_irqsave(&vp->lock, flags);
1988         if (switch_port)
1989                 list_add_rcu(&port->list, &vp->port_list);
1990         else
1991                 list_add_tail_rcu(&port->list, &vp->port_list);
1992         hlist_add_head_rcu(&port->hash,
1993                            &vp->port_hash[vnet_hashfn(port->raddr)]);
1994         vnet_port_add_txq(port);
1995         spin_unlock_irqrestore(&vp->lock, flags);
1996
1997         dev_set_drvdata(&vdev->dev, port);
1998
1999         pr_info("%s: PORT ( remote-mac %pM%s )\n",
2000                 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
2001
2002         setup_timer(&port->clean_timer, vnet_clean_timer_expire,
2003                     (unsigned long)port);
2004
2005         napi_enable(&port->napi);
2006         vio_port_up(&port->vio);
2007
2008         mdesc_release(hp);
2009
2010         return 0;
2011
2012 err_out_free_port:
2013         kfree(port);
2014
2015 err_out_put_mdesc:
2016         mdesc_release(hp);
2017         return err;
2018 }
2019
2020 static int vnet_port_remove(struct vio_dev *vdev)
2021 {
2022         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
2023
2024         if (port) {
2025
2026                 del_timer_sync(&port->vio.timer);
2027
2028                 napi_disable(&port->napi);
2029
2030                 list_del_rcu(&port->list);
2031                 hlist_del_rcu(&port->hash);
2032
2033                 synchronize_rcu();
2034                 del_timer_sync(&port->clean_timer);
2035                 vnet_port_rm_txq(port);
2036                 netif_napi_del(&port->napi);
2037                 vnet_port_free_tx_bufs(port);
2038                 vio_ldc_free(&port->vio);
2039
2040                 dev_set_drvdata(&vdev->dev, NULL);
2041
2042                 kfree(port);
2043
2044         }
2045         return 0;
2046 }
2047
2048 static const struct vio_device_id vnet_port_match[] = {
2049         {
2050                 .type = "vnet-port",
2051         },
2052         {},
2053 };
2054 MODULE_DEVICE_TABLE(vio, vnet_port_match);
2055
2056 static struct vio_driver vnet_port_driver = {
2057         .id_table       = vnet_port_match,
2058         .probe          = vnet_port_probe,
2059         .remove         = vnet_port_remove,
2060         .name           = "vnet_port",
2061 };
2062
2063 static int __init vnet_init(void)
2064 {
2065         return vio_register_driver(&vnet_port_driver);
2066 }
2067
2068 static void __exit vnet_exit(void)
2069 {
2070         vio_unregister_driver(&vnet_port_driver);
2071         vnet_cleanup();
2072 }
2073
2074 module_init(vnet_init);
2075 module_exit(vnet_exit);