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 u32 next_idx(u32 idx, struct vio_dring_state *dr)
 470 {
 471         if (++idx == dr->num_entries)
 472                 idx = 0;
 473         return idx;
 474 }
 475
 476 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
 477 {
 478         if (idx == 0)
 479                 idx = dr->num_entries - 1;
 480         else
 481                 idx--;
 482
 483         return idx;
 484 }
 485
 486 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
 487                                         struct vio_dring_state *dr,
 488                                         u32 index)
 489 {
 490         struct vio_net_desc *desc = port->vio.desc_buf;
 491         int err;
 492
 493         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
 494                                   (index * dr->entry_size),
 495                                   dr->cookies, dr->ncookies);
 496         if (err < 0)
 497                 return ERR_PTR(err);
 498
 499         return desc;
 500 }
 501
 502 static int put_rx_desc(struct vnet_port *port,
 503                        struct vio_dring_state *dr,
 504                        struct vio_net_desc *desc,
 505                        u32 index)
 506 {
 507         int err;
 508
 509         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
 510                                   (index * dr->entry_size),
 511                                   dr->cookies, dr->ncookies);
 512         if (err < 0)
 513                 return err;
 514
 515         return 0;
 516 }
 517
 518 static int vnet_walk_rx_one(struct vnet_port *port,
 519                             struct vio_dring_state *dr,
 520                             u32 index, int *needs_ack)
 521 {
 522         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
 523         struct vio_driver_state *vio = &port->vio;
 524         int err;
 525
 526         BUG_ON(desc == NULL);
 527         if (IS_ERR(desc))
 528                 return PTR_ERR(desc);
 529
 530         if (desc->hdr.state != VIO_DESC_READY)
 531                 return 1;
 532
 533         rmb();
 534
 535         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 536                desc->hdr.state, desc->hdr.ack,
 537                desc->size, desc->ncookies,
 538                desc->cookies[0].cookie_addr,
 539                desc->cookies[0].cookie_size);
 540
 541         err = vnet_rx_one(port, desc);
 542         if (err == -ECONNRESET)
 543                 return err;
 544         desc->hdr.state = VIO_DESC_DONE;
 545         err = put_rx_desc(port, dr, desc, index);
 546         if (err < 0)
 547                 return err;
 548         *needs_ack = desc->hdr.ack;
 549         return 0;
 550 }
 551
 552 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 553                         u32 start, u32 end, int *npkts, int budget)
 554 {
 555         struct vio_driver_state *vio = &port->vio;
 556         int ack_start = -1, ack_end = -1;
 557         bool send_ack = true;
 558
 559         end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
 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 = next_idx(start, dr);
 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 = prev_idx(start, dr);
 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 = next_idx(idx, dr);
 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 = next_idx(end, dr);
 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                         port->rmtu = 0;
 751                         port->tso = true;
 752                         port->tsolen = 0;
 753                         vio_port_up(vio);
 754                 }
 755                 port->rx_event = 0;
 756                 return 0;
 757         }
 758         /* We may have multiple LDC events in rx_event. Unroll send_events() */
 759         event = (port->rx_event & LDC_EVENT_UP);
 760         port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
 761         if (event == LDC_EVENT_UP)
 762                 goto ldc_ctrl;
 763         event = port->rx_event;
 764         if (!(event & LDC_EVENT_DATA_READY))
 765                 return 0;
 766
 767         /* we dont expect any other bits than RESET, UP, DATA_READY */
 768         BUG_ON(event != LDC_EVENT_DATA_READY);
 769
 770         tx_wakeup = err = 0;
 771         while (1) {
 772                 union {
 773                         struct vio_msg_tag tag;
 774                         u64 raw[8];
 775                 } msgbuf;
 776
 777                 if (port->napi_resume) {
 778                         struct vio_dring_data *pkt =
 779                                 (struct vio_dring_data *)&msgbuf;
 780                         struct vio_dring_state *dr =
 781                                 &port->vio.drings[VIO_DRIVER_RX_RING];
 782
 783                         pkt->tag.type = VIO_TYPE_DATA;
 784                         pkt->tag.stype = VIO_SUBTYPE_INFO;
 785                         pkt->tag.stype_env = VIO_DRING_DATA;
 786                         pkt->seq = dr->rcv_nxt;
 787                         pkt->start_idx = next_idx(port->napi_stop_idx, dr);
 788                         pkt->end_idx = -1;
 789                         goto napi_resume;
 790                 }
 791                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 792                 if (unlikely(err < 0)) {
 793                         if (err == -ECONNRESET)
 794                                 vio_conn_reset(vio);
 795                         break;
 796                 }
 797                 if (err == 0)
 798                         break;
 799                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 800                        msgbuf.tag.type,
 801                        msgbuf.tag.stype,
 802                        msgbuf.tag.stype_env,
 803                        msgbuf.tag.sid);
 804                 err = vio_validate_sid(vio, &msgbuf.tag);
 805                 if (err < 0)
 806                         break;
 807 napi_resume:
 808                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 809                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 810                                 if (!port_is_up(port)) {
 811                                         /* failures like handshake_failure()
 812                                          * may have cleaned up dring, but
 813                                          * NAPI polling may bring us here.
 814                                          */
 815                                         err = -ECONNRESET;
 816                                         break;
 817                                 }
 818                                 err = vnet_rx(port, &msgbuf, &npkts, budget);
 819                                 if (npkts >= budget)
 820                                         break;
 821                                 if (npkts == 0)
 822                                         break;
 823                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 824                                 err = vnet_ack(port, &msgbuf);
 825                                 if (err > 0)
 826                                         tx_wakeup |= err;
 827                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 828                                 err = vnet_nack(port, &msgbuf);
 829                         }
 830                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 831                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 832                                 err = handle_mcast(port, &msgbuf);
 833                         else
 834                                 err = vio_control_pkt_engine(vio, &msgbuf);
 835                         if (err)
 836                                 break;
 837                 } else {
 838                         err = vnet_handle_unknown(port, &msgbuf);
 839                 }
 840                 if (err == -ECONNRESET)
 841                         break;
 842         }
 843         if (unlikely(tx_wakeup && err != -ECONNRESET))
 844                 maybe_tx_wakeup(port);
 845         return npkts;
 846 }
 847
 848 static int vnet_poll(struct napi_struct *napi, int budget)
 849 {
 850         struct vnet_port *port = container_of(napi, struct vnet_port, napi);
 851         struct vio_driver_state *vio = &port->vio;
 852         int processed = vnet_event_napi(port, budget);
 853
 854         if (processed < budget) {
 855                 napi_complete(napi);
 856                 port->rx_event &= ~LDC_EVENT_DATA_READY;
 857                 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
 858         }
 859         return processed;
 860 }
 861
 862 static void vnet_event(void *arg, int event)
 863 {
 864         struct vnet_port *port = arg;
 865         struct vio_driver_state *vio = &port->vio;
 866
 867         port->rx_event |= event;
 868         vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
 869         napi_schedule(&port->napi);
 870
 871 }
 872
 873 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
 874 {
 875         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 876         struct vio_dring_data hdr = {
 877                 .tag = {
 878                         .type           = VIO_TYPE_DATA,
 879                         .stype          = VIO_SUBTYPE_INFO,
 880                         .stype_env      = VIO_DRING_DATA,
 881                         .sid            = vio_send_sid(&port->vio),
 882                 },
 883                 .dring_ident            = dr->ident,
 884                 .start_idx              = start,
 885                 .end_idx                = (u32) -1,
 886         };
 887         int err, delay;
 888         int retries = 0;
 889
 890         if (port->stop_rx) {
 891                 err = vnet_send_ack(port,
 892                                     &port->vio.drings[VIO_DRIVER_RX_RING],
 893                                     port->stop_rx_idx, -1,
 894                                     VIO_DRING_STOPPED);
 895                 if (err <= 0)
 896                         return err;
 897         }
 898
 899         hdr.seq = dr->snd_nxt;
 900         delay = 1;
 901         do {
 902                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 903                 if (err > 0) {
 904                         dr->snd_nxt++;
 905                         break;
 906                 }
 907                 udelay(delay);
 908                 if ((delay <<= 1) > 128)
 909                         delay = 128;
 910                 if (retries++ > VNET_MAX_RETRIES)
 911                         break;
 912         } while (err == -EAGAIN);
 913
 914         return err;
 915 }
 916
 917 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 918 {
 919         unsigned int hash = vnet_hashfn(skb->data);
 920         struct hlist_head *hp = &vp->port_hash[hash];
 921         struct vnet_port *port;
 922
 923         hlist_for_each_entry_rcu(port, hp, hash) {
 924                 if (!port_is_up(port))
 925                         continue;
 926                 if (ether_addr_equal(port->raddr, skb->data))
 927                         return port;
 928         }
 929         list_for_each_entry_rcu(port, &vp->port_list, list) {
 930                 if (!port->switch_port)
 931                         continue;
 932                 if (!port_is_up(port))
 933                         continue;
 934                 return port;
 935         }
 936         return NULL;
 937 }
 938
 939 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
 940                                           unsigned *pending)
 941 {
 942         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 943         struct sk_buff *skb = NULL;
 944         int i, txi;
 945
 946         *pending = 0;
 947
 948         txi = dr->prod-1;
 949         if (txi < 0)
 950                 txi = VNET_TX_RING_SIZE-1;
 951
 952         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
 953                 struct vio_net_desc *d;
 954
 955                 d = vio_dring_entry(dr, txi);
 956
 957                 if (d->hdr.state == VIO_DESC_DONE) {
 958                         if (port->tx_bufs[txi].skb) {
 959                                 BUG_ON(port->tx_bufs[txi].skb->next);
 960
 961                                 port->tx_bufs[txi].skb->next = skb;
 962                                 skb = port->tx_bufs[txi].skb;
 963                                 port->tx_bufs[txi].skb = NULL;
 964
 965                                 ldc_unmap(port->vio.lp,
 966                                           port->tx_bufs[txi].cookies,
 967                                           port->tx_bufs[txi].ncookies);
 968                         }
 969                         d->hdr.state = VIO_DESC_FREE;
 970                 } else if (d->hdr.state == VIO_DESC_READY) {
 971                         (*pending)++;
 972                 } else if (d->hdr.state == VIO_DESC_FREE) {
 973                         break;
 974                 }
 975                 --txi;
 976                 if (txi < 0)
 977                         txi = VNET_TX_RING_SIZE-1;
 978         }
 979         return skb;
 980 }
 981
 982 static inline void vnet_free_skbs(struct sk_buff *skb)
 983 {
 984         struct sk_buff *next;
 985
 986         while (skb) {
 987                 next = skb->next;
 988                 skb->next = NULL;
 989                 dev_kfree_skb(skb);
 990                 skb = next;
 991         }
 992 }
 993
 994 static void vnet_clean_timer_expire(unsigned long port0)
 995 {
 996         struct vnet_port *port = (struct vnet_port *)port0;
 997         struct sk_buff *freeskbs;
 998         unsigned pending;
 999
1000         netif_tx_lock(port->vp->dev);
1001         freeskbs = vnet_clean_tx_ring(port, &pending);
1002         netif_tx_unlock(port->vp->dev);
1003
1004         vnet_free_skbs(freeskbs);
1005
1006         if (pending)
1007                 (void)mod_timer(&port->clean_timer,
1008                                 jiffies + VNET_CLEAN_TIMEOUT);
1009          else
1010                 del_timer(&port->clean_timer);
1011 }
1012
1013 static inline int vnet_skb_map(struct ldc_channel *lp, struct sk_buff *skb,
1014                                struct ldc_trans_cookie *cookies, int ncookies,
1015                                unsigned int map_perm)
1016 {
1017         int i, nc, err, blen;
1018
1019         /* header */
1020         blen = skb_headlen(skb);
1021         if (blen < ETH_ZLEN)
1022                 blen = ETH_ZLEN;
1023         blen += VNET_PACKET_SKIP;
1024         blen += 8 - (blen & 7);
1025
1026         err = ldc_map_single(lp, skb->data-VNET_PACKET_SKIP, blen, cookies,
1027                              ncookies, map_perm);
1028         if (err < 0)
1029                 return err;
1030         nc = err;
1031
1032         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1033                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1034                 u8 *vaddr;
1035
1036                 if (nc < ncookies) {
1037                         vaddr = kmap_atomic(skb_frag_page(f));
1038                         blen = skb_frag_size(f);
1039                         blen += 8 - (blen & 7);
1040                         err = ldc_map_single(lp, vaddr + f->page_offset,
1041                                              blen, cookies + nc, ncookies - nc,
1042                                              map_perm);
1043                         kunmap_atomic(vaddr);
1044                 } else {
1045                         err = -EMSGSIZE;
1046                 }
1047
1048                 if (err < 0) {
1049                         ldc_unmap(lp, cookies, nc);
1050                         return err;
1051                 }
1052                 nc += err;
1053         }
1054         return nc;
1055 }
1056
1057 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, int ncookies)
1058 {
1059         struct sk_buff *nskb;
1060         int i, len, pad, docopy;
1061
1062         len = skb->len;
1063         pad = 0;
1064         if (len < ETH_ZLEN) {
1065                 pad += ETH_ZLEN - skb->len;
1066                 len += pad;
1067         }
1068         len += VNET_PACKET_SKIP;
1069         pad += 8 - (len & 7);
1070
1071         /* make sure we have enough cookies and alignment in every frag */
1072         docopy = skb_shinfo(skb)->nr_frags >= ncookies;
1073         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1074                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1075
1076                 docopy |= f->page_offset & 7;
1077         }
1078         if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
1079             skb_tailroom(skb) < pad ||
1080             skb_headroom(skb) < VNET_PACKET_SKIP || docopy) {
1081                 int start = 0, offset;
1082                 __wsum csum;
1083
1084                 len = skb->len > ETH_ZLEN ? skb->len : ETH_ZLEN;
1085                 nskb = alloc_and_align_skb(skb->dev, len);
1086                 if (nskb == NULL) {
1087                         dev_kfree_skb(skb);
1088                         return NULL;
1089                 }
1090                 skb_reserve(nskb, VNET_PACKET_SKIP);
1091
1092                 nskb->protocol = skb->protocol;
1093                 offset = skb_mac_header(skb) - skb->data;
1094                 skb_set_mac_header(nskb, offset);
1095                 offset = skb_network_header(skb) - skb->data;
1096                 skb_set_network_header(nskb, offset);
1097                 offset = skb_transport_header(skb) - skb->data;
1098                 skb_set_transport_header(nskb, offset);
1099
1100                 offset = 0;
1101                 nskb->csum_offset = skb->csum_offset;
1102                 nskb->ip_summed = skb->ip_summed;
1103
1104                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1105                         start = skb_checksum_start_offset(skb);
1106                 if (start) {
1107                         struct iphdr *iph = ip_hdr(nskb);
1108                         int offset = start + nskb->csum_offset;
1109
1110                         if (skb_copy_bits(skb, 0, nskb->data, start)) {
1111                                 dev_kfree_skb(nskb);
1112                                 dev_kfree_skb(skb);
1113                                 return NULL;
1114                         }
1115                         *(__sum16 *)(skb->data + offset) = 0;
1116                         csum = skb_copy_and_csum_bits(skb, start,
1117                                                       nskb->data + start,
1118                                                       skb->len - start, 0);
1119                         if (iph->protocol == IPPROTO_TCP ||
1120                             iph->protocol == IPPROTO_UDP) {
1121                                 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
1122                                                          skb->len - start,
1123                                                          iph->protocol, csum);
1124                         }
1125                         *(__sum16 *)(nskb->data + offset) = csum;
1126
1127                         nskb->ip_summed = CHECKSUM_NONE;
1128                 } else if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1129                         dev_kfree_skb(nskb);
1130                         dev_kfree_skb(skb);
1131                         return NULL;
1132                 }
1133                 (void)skb_put(nskb, skb->len);
1134                 if (skb_is_gso(skb)) {
1135                         skb_shinfo(nskb)->gso_size = skb_shinfo(skb)->gso_size;
1136                         skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
1137                 }
1138                 dev_kfree_skb(skb);
1139                 skb = nskb;
1140         }
1141         return skb;
1142 }
1143
1144 static u16
1145 vnet_select_queue(struct net_device *dev, struct sk_buff *skb,
1146                   void *accel_priv, select_queue_fallback_t fallback)
1147 {
1148         struct vnet *vp = netdev_priv(dev);
1149         struct vnet_port *port = __tx_port_find(vp, skb);
1150
1151         if (port == NULL)
1152                 return 0;
1153         return port->q_index;
1154 }
1155
1156 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev);
1157
1158 static int vnet_handle_offloads(struct vnet_port *port, struct sk_buff *skb)
1159 {
1160         struct net_device *dev = port->vp->dev;
1161         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1162         struct sk_buff *segs;
1163         int maclen, datalen;
1164         int status;
1165         int gso_size, gso_type, gso_segs;
1166         int hlen = skb_transport_header(skb) - skb_mac_header(skb);
1167         int proto = IPPROTO_IP;
1168
1169         if (skb->protocol == htons(ETH_P_IP))
1170                 proto = ip_hdr(skb)->protocol;
1171         else if (skb->protocol == htons(ETH_P_IPV6))
1172                 proto = ipv6_hdr(skb)->nexthdr;
1173
1174         if (proto == IPPROTO_TCP)
1175                 hlen += tcp_hdr(skb)->doff * 4;
1176         else if (proto == IPPROTO_UDP)
1177                 hlen += sizeof(struct udphdr);
1178         else {
1179                 pr_err("vnet_handle_offloads GSO with unknown transport "
1180                        "protocol %d tproto %d\n", skb->protocol, proto);
1181                 hlen = 128; /* XXX */
1182         }
1183         datalen = port->tsolen - hlen;
1184
1185         gso_size = skb_shinfo(skb)->gso_size;
1186         gso_type = skb_shinfo(skb)->gso_type;
1187         gso_segs = skb_shinfo(skb)->gso_segs;
1188
1189         if (port->tso && gso_size < datalen)
1190                 gso_segs = DIV_ROUND_UP(skb->len - hlen, datalen);
1191
1192         if (unlikely(vnet_tx_dring_avail(dr) < gso_segs)) {
1193                 struct netdev_queue *txq;
1194
1195                 txq  = netdev_get_tx_queue(dev, port->q_index);
1196                 netif_tx_stop_queue(txq);
1197                 if (vnet_tx_dring_avail(dr) < skb_shinfo(skb)->gso_segs)
1198                         return NETDEV_TX_BUSY;
1199                 netif_tx_wake_queue(txq);
1200         }
1201
1202         maclen = skb_network_header(skb) - skb_mac_header(skb);
1203         skb_pull(skb, maclen);
1204
1205         if (port->tso && gso_size < datalen) {
1206                 /* segment to TSO size */
1207                 skb_shinfo(skb)->gso_size = datalen;
1208                 skb_shinfo(skb)->gso_segs = gso_segs;
1209
1210                 segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1211
1212                 /* restore gso_size & gso_segs */
1213                 skb_shinfo(skb)->gso_size = gso_size;
1214                 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len - hlen,
1215                                                          gso_size);
1216         } else
1217                 segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1218         if (IS_ERR(segs)) {
1219                 dev->stats.tx_dropped++;
1220                 return NETDEV_TX_OK;
1221         }
1222
1223         skb_push(skb, maclen);
1224         skb_reset_mac_header(skb);
1225
1226         status = 0;
1227         while (segs) {
1228                 struct sk_buff *curr = segs;
1229
1230                 segs = segs->next;
1231                 curr->next = NULL;
1232                 if (port->tso && curr->len > dev->mtu) {
1233                         skb_shinfo(curr)->gso_size = gso_size;
1234                         skb_shinfo(curr)->gso_type = gso_type;
1235                         skb_shinfo(curr)->gso_segs =
1236                                 DIV_ROUND_UP(curr->len - hlen, gso_size);
1237                 } else
1238                         skb_shinfo(curr)->gso_size = 0;
1239
1240                 skb_push(curr, maclen);
1241                 skb_reset_mac_header(curr);
1242                 memcpy(skb_mac_header(curr), skb_mac_header(skb),
1243                        maclen);
1244                 curr->csum_start = skb_transport_header(curr) - curr->head;
1245                 if (ip_hdr(curr)->protocol == IPPROTO_TCP)
1246                         curr->csum_offset = offsetof(struct tcphdr, check);
1247                 else if (ip_hdr(curr)->protocol == IPPROTO_UDP)
1248                         curr->csum_offset = offsetof(struct udphdr, check);
1249
1250                 if (!(status & NETDEV_TX_MASK))
1251                         status = vnet_start_xmit(curr, dev);
1252                 if (status & NETDEV_TX_MASK)
1253                         dev_kfree_skb_any(curr);
1254         }
1255
1256         if (!(status & NETDEV_TX_MASK))
1257                 dev_kfree_skb_any(skb);
1258         return status;
1259 }
1260
1261 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
1262 {
1263         struct vnet *vp = netdev_priv(dev);
1264         struct vnet_port *port = NULL;
1265         struct vio_dring_state *dr;
1266         struct vio_net_desc *d;
1267         unsigned int len;
1268         struct sk_buff *freeskbs = NULL;
1269         int i, err, txi;
1270         unsigned pending = 0;
1271         struct netdev_queue *txq;
1272
1273         rcu_read_lock();
1274         port = __tx_port_find(vp, skb);
1275         if (unlikely(!port)) {
1276                 rcu_read_unlock();
1277                 goto out_dropped;
1278         }
1279
1280         if (skb_is_gso(skb) && skb->len > port->tsolen) {
1281                 err = vnet_handle_offloads(port, skb);
1282                 rcu_read_unlock();
1283                 return err;
1284         }
1285
1286         if (!skb_is_gso(skb) && skb->len > port->rmtu) {
1287                 unsigned long localmtu = port->rmtu - ETH_HLEN;
1288
1289                 if (vio_version_after_eq(&port->vio, 1, 3))
1290                         localmtu -= VLAN_HLEN;
1291
1292                 if (skb->protocol == htons(ETH_P_IP)) {
1293                         struct flowi4 fl4;
1294                         struct rtable *rt = NULL;
1295
1296                         memset(&fl4, 0, sizeof(fl4));
1297                         fl4.flowi4_oif = dev->ifindex;
1298                         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
1299                         fl4.daddr = ip_hdr(skb)->daddr;
1300                         fl4.saddr = ip_hdr(skb)->saddr;
1301
1302                         rt = ip_route_output_key(dev_net(dev), &fl4);
1303                         rcu_read_unlock();
1304                         if (!IS_ERR(rt)) {
1305                                 skb_dst_set(skb, &rt->dst);
1306                                 icmp_send(skb, ICMP_DEST_UNREACH,
1307                                           ICMP_FRAG_NEEDED,
1308                                           htonl(localmtu));
1309                         }
1310                 }
1311 #if IS_ENABLED(CONFIG_IPV6)
1312                 else if (skb->protocol == htons(ETH_P_IPV6))
1313                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1314 #endif
1315                 goto out_dropped;
1316         }
1317
1318         skb = vnet_skb_shape(skb, 2);
1319
1320         if (unlikely(!skb))
1321                 goto out_dropped;
1322
1323         if (skb->ip_summed == CHECKSUM_PARTIAL)
1324                 vnet_fullcsum(skb);
1325
1326         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1327         i = skb_get_queue_mapping(skb);
1328         txq = netdev_get_tx_queue(dev, i);
1329         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1330                 if (!netif_tx_queue_stopped(txq)) {
1331                         netif_tx_stop_queue(txq);
1332
1333                         /* This is a hard error, log it. */
1334                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1335                         dev->stats.tx_errors++;
1336                 }
1337                 rcu_read_unlock();
1338                 return NETDEV_TX_BUSY;
1339         }
1340
1341         d = vio_dring_cur(dr);
1342
1343         txi = dr->prod;
1344
1345         freeskbs = vnet_clean_tx_ring(port, &pending);
1346
1347         BUG_ON(port->tx_bufs[txi].skb);
1348
1349         len = skb->len;
1350         if (len < ETH_ZLEN)
1351                 len = ETH_ZLEN;
1352
1353         err = vnet_skb_map(port->vio.lp, skb, port->tx_bufs[txi].cookies, 2,
1354                            (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1355         if (err < 0) {
1356                 netdev_info(dev, "tx buffer map error %d\n", err);
1357                 goto out_dropped;
1358         }
1359
1360         port->tx_bufs[txi].skb = skb;
1361         skb = NULL;
1362         port->tx_bufs[txi].ncookies = err;
1363
1364         /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1365          * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1366          * the protocol itself does not require it as long as the peer
1367          * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1368          *
1369          * An ACK for every packet in the ring is expensive as the
1370          * sending of LDC messages is slow and affects performance.
1371          */
1372         d->hdr.ack = VIO_ACK_DISABLE;
1373         d->size = len;
1374         d->ncookies = port->tx_bufs[txi].ncookies;
1375         for (i = 0; i < d->ncookies; i++)
1376                 d->cookies[i] = port->tx_bufs[txi].cookies[i];
1377         if (vio_version_after_eq(&port->vio, 1, 7)) {
1378                 struct vio_net_dext *dext = vio_net_ext(d);
1379
1380                 memset(dext, 0, sizeof(*dext));
1381                 if (skb_is_gso(port->tx_bufs[txi].skb)) {
1382                         dext->ipv4_lso_mss = skb_shinfo(port->tx_bufs[txi].skb)
1383                                              ->gso_size;
1384                         dext->flags |= VNET_PKT_IPV4_LSO;
1385                 }
1386                 if (vio_version_after_eq(&port->vio, 1, 8) &&
1387                     !port->switch_port) {
1388                         dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1389                         dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1390                 }
1391         }
1392
1393         /* This has to be a non-SMP write barrier because we are writing
1394          * to memory which is shared with the peer LDOM.
1395          */
1396         wmb();
1397
1398         d->hdr.state = VIO_DESC_READY;
1399
1400         /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1401          * to notify the consumer that some descriptors are READY.
1402          * After that "start" trigger, no additional triggers are needed until
1403          * a DRING_STOPPED is received from the consumer. The dr->cons field
1404          * (set up by vnet_ack()) has the value of the next dring index
1405          * that has not yet been ack-ed. We send a "start" trigger here
1406          * if, and only if, start_cons is true (reset it afterward). Conversely,
1407          * vnet_ack() should check if the dring corresponding to cons
1408          * is marked READY, but start_cons was false.
1409          * If so, vnet_ack() should send out the missed "start" trigger.
1410          *
1411          * Note that the wmb() above makes sure the cookies et al. are
1412          * not globally visible before the VIO_DESC_READY, and that the
1413          * stores are ordered correctly by the compiler. The consumer will
1414          * not proceed until the VIO_DESC_READY is visible assuring that
1415          * the consumer does not observe anything related to descriptors
1416          * out of order. The HV trap from the LDC start trigger is the
1417          * producer to consumer announcement that work is available to the
1418          * consumer
1419          */
1420         if (!port->start_cons)
1421                 goto ldc_start_done; /* previous trigger suffices */
1422
1423         err = __vnet_tx_trigger(port, dr->cons);
1424         if (unlikely(err < 0)) {
1425                 netdev_info(dev, "TX trigger error %d\n", err);
1426                 d->hdr.state = VIO_DESC_FREE;
1427                 dev->stats.tx_carrier_errors++;
1428                 goto out_dropped;
1429         }
1430
1431 ldc_start_done:
1432         port->start_cons = false;
1433
1434         dev->stats.tx_packets++;
1435         dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1436
1437         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1438         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1439                 netif_tx_stop_queue(txq);
1440                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1441                         netif_tx_wake_queue(txq);
1442         }
1443
1444         (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1445         rcu_read_unlock();
1446
1447         vnet_free_skbs(freeskbs);
1448
1449         return NETDEV_TX_OK;
1450
1451 out_dropped:
1452         if (pending)
1453                 (void)mod_timer(&port->clean_timer,
1454                                 jiffies + VNET_CLEAN_TIMEOUT);
1455         else if (port)
1456                 del_timer(&port->clean_timer);
1457         if (port)
1458                 rcu_read_unlock();
1459         if (skb)
1460                 dev_kfree_skb(skb);
1461         vnet_free_skbs(freeskbs);
1462         dev->stats.tx_dropped++;
1463         return NETDEV_TX_OK;
1464 }
1465
1466 static void vnet_tx_timeout(struct net_device *dev)
1467 {
1468         /* XXX Implement me XXX */
1469 }
1470
1471 static int vnet_open(struct net_device *dev)
1472 {
1473         netif_carrier_on(dev);
1474         netif_tx_start_all_queues(dev);
1475
1476         return 0;
1477 }
1478
1479 static int vnet_close(struct net_device *dev)
1480 {
1481         netif_tx_stop_all_queues(dev);
1482         netif_carrier_off(dev);
1483
1484         return 0;
1485 }
1486
1487 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1488 {
1489         struct vnet_mcast_entry *m;
1490
1491         for (m = vp->mcast_list; m; m = m->next) {
1492                 if (ether_addr_equal(m->addr, addr))
1493                         return m;
1494         }
1495         return NULL;
1496 }
1497
1498 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1499 {
1500         struct netdev_hw_addr *ha;
1501
1502         netdev_for_each_mc_addr(ha, dev) {
1503                 struct vnet_mcast_entry *m;
1504
1505                 m = __vnet_mc_find(vp, ha->addr);
1506                 if (m) {
1507                         m->hit = 1;
1508                         continue;
1509                 }
1510
1511                 if (!m) {
1512                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
1513                         if (!m)
1514                                 continue;
1515                         memcpy(m->addr, ha->addr, ETH_ALEN);
1516                         m->hit = 1;
1517
1518                         m->next = vp->mcast_list;
1519                         vp->mcast_list = m;
1520                 }
1521         }
1522 }
1523
1524 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1525 {
1526         struct vio_net_mcast_info info;
1527         struct vnet_mcast_entry *m, **pp;
1528         int n_addrs;
1529
1530         memset(&info, 0, sizeof(info));
1531
1532         info.tag.type = VIO_TYPE_CTRL;
1533         info.tag.stype = VIO_SUBTYPE_INFO;
1534         info.tag.stype_env = VNET_MCAST_INFO;
1535         info.tag.sid = vio_send_sid(&port->vio);
1536         info.set = 1;
1537
1538         n_addrs = 0;
1539         for (m = vp->mcast_list; m; m = m->next) {
1540                 if (m->sent)
1541                         continue;
1542                 m->sent = 1;
1543                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1544                        m->addr, ETH_ALEN);
1545                 if (++n_addrs == VNET_NUM_MCAST) {
1546                         info.count = n_addrs;
1547
1548                         (void) vio_ldc_send(&port->vio, &info,
1549                                             sizeof(info));
1550                         n_addrs = 0;
1551                 }
1552         }
1553         if (n_addrs) {
1554                 info.count = n_addrs;
1555                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1556         }
1557
1558         info.set = 0;
1559
1560         n_addrs = 0;
1561         pp = &vp->mcast_list;
1562         while ((m = *pp) != NULL) {
1563                 if (m->hit) {
1564                         m->hit = 0;
1565                         pp = &m->next;
1566                         continue;
1567                 }
1568
1569                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1570                        m->addr, ETH_ALEN);
1571                 if (++n_addrs == VNET_NUM_MCAST) {
1572                         info.count = n_addrs;
1573                         (void) vio_ldc_send(&port->vio, &info,
1574                                             sizeof(info));
1575                         n_addrs = 0;
1576                 }
1577
1578                 *pp = m->next;
1579                 kfree(m);
1580         }
1581         if (n_addrs) {
1582                 info.count = n_addrs;
1583                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1584         }
1585 }
1586
1587 static void vnet_set_rx_mode(struct net_device *dev)
1588 {
1589         struct vnet *vp = netdev_priv(dev);
1590         struct vnet_port *port;
1591
1592         rcu_read_lock();
1593         list_for_each_entry_rcu(port, &vp->port_list, list) {
1594
1595                 if (port->switch_port) {
1596                         __update_mc_list(vp, dev);
1597                         __send_mc_list(vp, port);
1598                         break;
1599                 }
1600         }
1601         rcu_read_unlock();
1602 }
1603
1604 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1605 {
1606         if (new_mtu < 68 || new_mtu > 65535)
1607                 return -EINVAL;
1608
1609         dev->mtu = new_mtu;
1610         return 0;
1611 }
1612
1613 static int vnet_set_mac_addr(struct net_device *dev, void *p)
1614 {
1615         return -EINVAL;
1616 }
1617
1618 static void vnet_get_drvinfo(struct net_device *dev,
1619                              struct ethtool_drvinfo *info)
1620 {
1621         strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1622         strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
1623 }
1624
1625 static u32 vnet_get_msglevel(struct net_device *dev)
1626 {
1627         struct vnet *vp = netdev_priv(dev);
1628         return vp->msg_enable;
1629 }
1630
1631 static void vnet_set_msglevel(struct net_device *dev, u32 value)
1632 {
1633         struct vnet *vp = netdev_priv(dev);
1634         vp->msg_enable = value;
1635 }
1636
1637 static const struct ethtool_ops vnet_ethtool_ops = {
1638         .get_drvinfo            = vnet_get_drvinfo,
1639         .get_msglevel           = vnet_get_msglevel,
1640         .set_msglevel           = vnet_set_msglevel,
1641         .get_link               = ethtool_op_get_link,
1642 };
1643
1644 static void vnet_port_free_tx_bufs(struct vnet_port *port)
1645 {
1646         struct vio_dring_state *dr;
1647         int i;
1648
1649         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1650         if (dr->base) {
1651                 ldc_free_exp_dring(port->vio.lp, dr->base,
1652                                    (dr->entry_size * dr->num_entries),
1653                                    dr->cookies, dr->ncookies);
1654                 dr->base = NULL;
1655                 dr->entry_size = 0;
1656                 dr->num_entries = 0;
1657                 dr->pending = 0;
1658                 dr->ncookies = 0;
1659         }
1660
1661         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1662                 struct vio_net_desc *d;
1663                 void *skb = port->tx_bufs[i].skb;
1664
1665                 if (!skb)
1666                         continue;
1667
1668                 d = vio_dring_entry(dr, i);
1669                 if (d->hdr.state == VIO_DESC_READY)
1670                         pr_warn("active transmit buffers freed\n");
1671
1672                 ldc_unmap(port->vio.lp,
1673                           port->tx_bufs[i].cookies,
1674                           port->tx_bufs[i].ncookies);
1675                 dev_kfree_skb(skb);
1676                 port->tx_bufs[i].skb = NULL;
1677                 d->hdr.state = VIO_DESC_FREE;
1678         }
1679 }
1680
1681 static int vnet_port_alloc_tx_ring(struct vnet_port *port)
1682 {
1683         struct vio_dring_state *dr;
1684         unsigned long len, elen;
1685         int i, err, ncookies;
1686         void *dring;
1687
1688         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1689
1690         elen = sizeof(struct vio_net_desc) +
1691                sizeof(struct ldc_trans_cookie) * 2;
1692         if (vio_version_after_eq(&port->vio, 1, 7))
1693                 elen += sizeof(struct vio_net_dext);
1694         len = VNET_TX_RING_SIZE * elen;
1695
1696         ncookies = VIO_MAX_RING_COOKIES;
1697         dring = ldc_alloc_exp_dring(port->vio.lp, len,
1698                                     dr->cookies, &ncookies,
1699                                     (LDC_MAP_SHADOW |
1700                                      LDC_MAP_DIRECT |
1701                                      LDC_MAP_RW));
1702         if (IS_ERR(dring)) {
1703                 err = PTR_ERR(dring);
1704                 goto err_out;
1705         }
1706
1707         dr->base = dring;
1708         dr->entry_size = elen;
1709         dr->num_entries = VNET_TX_RING_SIZE;
1710         dr->prod = dr->cons = 0;
1711         port->start_cons  = true; /* need an initial trigger */
1712         dr->pending = VNET_TX_RING_SIZE;
1713         dr->ncookies = ncookies;
1714
1715         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1716                 struct vio_net_desc *d;
1717
1718                 d = vio_dring_entry(dr, i);
1719                 d->hdr.state = VIO_DESC_FREE;
1720         }
1721         return 0;
1722
1723 err_out:
1724         vnet_port_free_tx_bufs(port);
1725
1726         return err;
1727 }
1728
1729 #ifdef CONFIG_NET_POLL_CONTROLLER
1730 static void vnet_poll_controller(struct net_device *dev)
1731 {
1732         struct vnet *vp = netdev_priv(dev);
1733         struct vnet_port *port;
1734         unsigned long flags;
1735
1736         spin_lock_irqsave(&vp->lock, flags);
1737         if (!list_empty(&vp->port_list)) {
1738                 port = list_entry(vp->port_list.next, struct vnet_port, list);
1739                 napi_schedule(&port->napi);
1740         }
1741         spin_unlock_irqrestore(&vp->lock, flags);
1742 }
1743 #endif
1744 static LIST_HEAD(vnet_list);
1745 static DEFINE_MUTEX(vnet_list_mutex);
1746
1747 static const struct net_device_ops vnet_ops = {
1748         .ndo_open               = vnet_open,
1749         .ndo_stop               = vnet_close,
1750         .ndo_set_rx_mode        = vnet_set_rx_mode,
1751         .ndo_set_mac_address    = vnet_set_mac_addr,
1752         .ndo_validate_addr      = eth_validate_addr,
1753         .ndo_tx_timeout         = vnet_tx_timeout,
1754         .ndo_change_mtu         = vnet_change_mtu,
1755         .ndo_start_xmit         = vnet_start_xmit,
1756         .ndo_select_queue       = vnet_select_queue,
1757 #ifdef CONFIG_NET_POLL_CONTROLLER
1758         .ndo_poll_controller    = vnet_poll_controller,
1759 #endif
1760 };
1761
1762 static struct vnet *vnet_new(const u64 *local_mac)
1763 {
1764         struct net_device *dev;
1765         struct vnet *vp;
1766         int err, i;
1767
1768         dev = alloc_etherdev_mqs(sizeof(*vp), VNET_MAX_TXQS, 1);
1769         if (!dev)
1770                 return ERR_PTR(-ENOMEM);
1771         dev->needed_headroom = VNET_PACKET_SKIP + 8;
1772         dev->needed_tailroom = 8;
1773
1774         for (i = 0; i < ETH_ALEN; i++)
1775                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1776
1777         vp = netdev_priv(dev);
1778
1779         spin_lock_init(&vp->lock);
1780         vp->dev = dev;
1781
1782         INIT_LIST_HEAD(&vp->port_list);
1783         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1784                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1785         INIT_LIST_HEAD(&vp->list);
1786         vp->local_mac = *local_mac;
1787
1788         dev->netdev_ops = &vnet_ops;
1789         dev->ethtool_ops = &vnet_ethtool_ops;
1790         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1791
1792         dev->hw_features = NETIF_F_TSO | NETIF_F_GSO | NETIF_F_GSO_SOFTWARE |
1793                            NETIF_F_HW_CSUM | NETIF_F_SG;
1794         dev->features = dev->hw_features;
1795
1796         err = register_netdev(dev);
1797         if (err) {
1798                 pr_err("Cannot register net device, aborting\n");
1799                 goto err_out_free_dev;
1800         }
1801
1802         netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1803
1804         list_add(&vp->list, &vnet_list);
1805
1806         return vp;
1807
1808 err_out_free_dev:
1809         free_netdev(dev);
1810
1811         return ERR_PTR(err);
1812 }
1813
1814 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1815 {
1816         struct vnet *iter, *vp;
1817
1818         mutex_lock(&vnet_list_mutex);
1819         vp = NULL;
1820         list_for_each_entry(iter, &vnet_list, list) {
1821                 if (iter->local_mac == *local_mac) {
1822                         vp = iter;
1823                         break;
1824                 }
1825         }
1826         if (!vp)
1827                 vp = vnet_new(local_mac);
1828         mutex_unlock(&vnet_list_mutex);
1829
1830         return vp;
1831 }
1832
1833 static void vnet_cleanup(void)
1834 {
1835         struct vnet *vp;
1836         struct net_device *dev;
1837
1838         mutex_lock(&vnet_list_mutex);
1839         while (!list_empty(&vnet_list)) {
1840                 vp = list_first_entry(&vnet_list, struct vnet, list);
1841                 list_del(&vp->list);
1842                 dev = vp->dev;
1843                 /* vio_unregister_driver() should have cleaned up port_list */
1844                 BUG_ON(!list_empty(&vp->port_list));
1845                 unregister_netdev(dev);
1846                 free_netdev(dev);
1847         }
1848         mutex_unlock(&vnet_list_mutex);
1849 }
1850
1851 static const char *local_mac_prop = "local-mac-address";
1852
1853 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1854                                                 u64 port_node)
1855 {
1856         const u64 *local_mac = NULL;
1857         u64 a;
1858
1859         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1860                 u64 target = mdesc_arc_target(hp, a);
1861                 const char *name;
1862
1863                 name = mdesc_get_property(hp, target, "name", NULL);
1864                 if (!name || strcmp(name, "network"))
1865                         continue;
1866
1867                 local_mac = mdesc_get_property(hp, target,
1868                                                local_mac_prop, NULL);
1869                 if (local_mac)
1870                         break;
1871         }
1872         if (!local_mac)
1873                 return ERR_PTR(-ENODEV);
1874
1875         return vnet_find_or_create(local_mac);
1876 }
1877
1878 static struct ldc_channel_config vnet_ldc_cfg = {
1879         .event          = vnet_event,
1880         .mtu            = 64,
1881         .mode           = LDC_MODE_UNRELIABLE,
1882 };
1883
1884 static struct vio_driver_ops vnet_vio_ops = {
1885         .send_attr              = vnet_send_attr,
1886         .handle_attr            = vnet_handle_attr,
1887         .handshake_complete     = vnet_handshake_complete,
1888 };
1889
1890 static void print_version(void)
1891 {
1892         printk_once(KERN_INFO "%s", version);
1893 }
1894
1895 const char *remote_macaddr_prop = "remote-mac-address";
1896
1897 static void
1898 vnet_port_add_txq(struct vnet_port *port)
1899 {
1900         struct vnet *vp = port->vp;
1901         int n;
1902
1903         n = vp->nports++;
1904         n = n & (VNET_MAX_TXQS - 1);
1905         port->q_index = n;
1906         netif_tx_wake_queue(netdev_get_tx_queue(vp->dev, port->q_index));
1907 }
1908
1909 static void
1910 vnet_port_rm_txq(struct vnet_port *port)
1911 {
1912         port->vp->nports--;
1913         netif_tx_stop_queue(netdev_get_tx_queue(port->vp->dev, port->q_index));
1914 }
1915
1916 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1917 {
1918         struct mdesc_handle *hp;
1919         struct vnet_port *port;
1920         unsigned long flags;
1921         struct vnet *vp;
1922         const u64 *rmac;
1923         int len, i, err, switch_port;
1924
1925         print_version();
1926
1927         hp = mdesc_grab();
1928
1929         vp = vnet_find_parent(hp, vdev->mp);
1930         if (IS_ERR(vp)) {
1931                 pr_err("Cannot find port parent vnet\n");
1932                 err = PTR_ERR(vp);
1933                 goto err_out_put_mdesc;
1934         }
1935
1936         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1937         err = -ENODEV;
1938         if (!rmac) {
1939                 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1940                 goto err_out_put_mdesc;
1941         }
1942
1943         port = kzalloc(sizeof(*port), GFP_KERNEL);
1944         err = -ENOMEM;
1945         if (!port)
1946                 goto err_out_put_mdesc;
1947
1948         for (i = 0; i < ETH_ALEN; i++)
1949                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1950
1951         port->vp = vp;
1952
1953         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1954                               vnet_versions, ARRAY_SIZE(vnet_versions),
1955                               &vnet_vio_ops, vp->dev->name);
1956         if (err)
1957                 goto err_out_free_port;
1958
1959         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1960         if (err)
1961                 goto err_out_free_port;
1962
1963         netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1964
1965         INIT_HLIST_NODE(&port->hash);
1966         INIT_LIST_HEAD(&port->list);
1967
1968         switch_port = 0;
1969         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1970                 switch_port = 1;
1971         port->switch_port = switch_port;
1972         port->tso = true;
1973         port->tsolen = 0;
1974
1975         spin_lock_irqsave(&vp->lock, flags);
1976         if (switch_port)
1977                 list_add_rcu(&port->list, &vp->port_list);
1978         else
1979                 list_add_tail_rcu(&port->list, &vp->port_list);
1980         hlist_add_head_rcu(&port->hash,
1981                            &vp->port_hash[vnet_hashfn(port->raddr)]);
1982         vnet_port_add_txq(port);
1983         spin_unlock_irqrestore(&vp->lock, flags);
1984
1985         dev_set_drvdata(&vdev->dev, port);
1986
1987         pr_info("%s: PORT ( remote-mac %pM%s )\n",
1988                 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1989
1990         setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1991                     (unsigned long)port);
1992
1993         napi_enable(&port->napi);
1994         vio_port_up(&port->vio);
1995
1996         mdesc_release(hp);
1997
1998         return 0;
1999
2000 err_out_free_port:
2001         kfree(port);
2002
2003 err_out_put_mdesc:
2004         mdesc_release(hp);
2005         return err;
2006 }
2007
2008 static int vnet_port_remove(struct vio_dev *vdev)
2009 {
2010         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
2011
2012         if (port) {
2013
2014                 del_timer_sync(&port->vio.timer);
2015
2016                 napi_disable(&port->napi);
2017
2018                 list_del_rcu(&port->list);
2019                 hlist_del_rcu(&port->hash);
2020
2021                 synchronize_rcu();
2022                 del_timer_sync(&port->clean_timer);
2023                 vnet_port_rm_txq(port);
2024                 netif_napi_del(&port->napi);
2025                 vnet_port_free_tx_bufs(port);
2026                 vio_ldc_free(&port->vio);
2027
2028                 dev_set_drvdata(&vdev->dev, NULL);
2029
2030                 kfree(port);
2031
2032         }
2033         return 0;
2034 }
2035
2036 static const struct vio_device_id vnet_port_match[] = {
2037         {
2038                 .type = "vnet-port",
2039         },
2040         {},
2041 };
2042 MODULE_DEVICE_TABLE(vio, vnet_port_match);
2043
2044 static struct vio_driver vnet_port_driver = {
2045         .id_table       = vnet_port_match,
2046         .probe          = vnet_port_probe,
2047         .remove         = vnet_port_remove,
2048         .name           = "vnet_port",
2049 };
2050
2051 static int __init vnet_init(void)
2052 {
2053         return vio_register_driver(&vnet_port_driver);
2054 }
2055
2056 static void __exit vnet_exit(void)
2057 {
2058         vio_unregister_driver(&vnet_port_driver);
2059         vnet_cleanup();
2060 }
2061
2062 module_init(vnet_init);
2063 module_exit(vnet_exit);