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