git.oblomov.eu Git - linux-2.6/blob - drivers/net/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 #include <linux/module.h>
   7 #include <linux/kernel.h>
   8 #include <linux/types.h>
   9 #include <linux/slab.h>
  10 #include <linux/delay.h>
  11 #include <linux/init.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/ethtool.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/mutex.h>
  16
  17 #include <asm/vio.h>
  18 #include <asm/ldc.h>
  19
  20 #include "sunvnet.h"
  21
  22 #define DRV_MODULE_NAME         "sunvnet"
  23 #define PFX DRV_MODULE_NAME     ": "
  24 #define DRV_MODULE_VERSION      "1.0"
  25 #define DRV_MODULE_RELDATE      "June 25, 2007"
  26
  27 static char version[] __devinitdata =
  28         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
  29 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
  30 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
  31 MODULE_LICENSE("GPL");
  32 MODULE_VERSION(DRV_MODULE_VERSION);
  33
  34 /* Ordered from largest major to lowest */
  35 static struct vio_version vnet_versions[] = {
  36         { .major = 1, .minor = 0 },
  37 };
  38
  39 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
  40 {
  41         return vio_dring_avail(dr, VNET_TX_RING_SIZE);
  42 }
  43
  44 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
  45 {
  46         struct vio_msg_tag *pkt = arg;
  47
  48         printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
  49                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
  50         printk(KERN_ERR PFX "Resetting connection.\n");
  51
  52         ldc_disconnect(port->vio.lp);
  53
  54         return -ECONNRESET;
  55 }
  56
  57 static int vnet_send_attr(struct vio_driver_state *vio)
  58 {
  59         struct vnet_port *port = to_vnet_port(vio);
  60         struct net_device *dev = port->vp->dev;
  61         struct vio_net_attr_info pkt;
  62         int i;
  63
  64         memset(&pkt, 0, sizeof(pkt));
  65         pkt.tag.type = VIO_TYPE_CTRL;
  66         pkt.tag.stype = VIO_SUBTYPE_INFO;
  67         pkt.tag.stype_env = VIO_ATTR_INFO;
  68         pkt.tag.sid = vio_send_sid(vio);
  69         pkt.xfer_mode = VIO_DRING_MODE;
  70         pkt.addr_type = VNET_ADDR_ETHERMAC;
  71         pkt.ack_freq = 0;
  72         for (i = 0; i < 6; i++)
  73                 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
  74         pkt.mtu = ETH_FRAME_LEN;
  75
  76         viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
  77                "ackfreq[%u] mtu[%llu]\n",
  78                pkt.xfer_mode, pkt.addr_type,
  79                (unsigned long long) pkt.addr,
  80                pkt.ack_freq,
  81                (unsigned long long) pkt.mtu);
  82
  83         return vio_ldc_send(vio, &pkt, sizeof(pkt));
  84 }
  85
  86 static int handle_attr_info(struct vio_driver_state *vio,
  87                             struct vio_net_attr_info *pkt)
  88 {
  89         viodbg(HS, "GOT NET ATTR INFO xmode[0x%x] atype[0x%x] addr[%llx] "
  90                "ackfreq[%u] mtu[%llu]\n",
  91                pkt->xfer_mode, pkt->addr_type,
  92                (unsigned long long) pkt->addr,
  93                pkt->ack_freq,
  94                (unsigned long long) pkt->mtu);
  95
  96         pkt->tag.sid = vio_send_sid(vio);
  97
  98         if (pkt->xfer_mode != VIO_DRING_MODE ||
  99             pkt->addr_type != VNET_ADDR_ETHERMAC ||
 100             pkt->mtu != ETH_FRAME_LEN) {
 101                 viodbg(HS, "SEND NET ATTR NACK\n");
 102
 103                 pkt->tag.stype = VIO_SUBTYPE_NACK;
 104
 105                 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
 106
 107                 return -ECONNRESET;
 108         } else {
 109                 viodbg(HS, "SEND NET ATTR ACK\n");
 110
 111                 pkt->tag.stype = VIO_SUBTYPE_ACK;
 112
 113                 return vio_ldc_send(vio, pkt, sizeof(*pkt));
 114         }
 115
 116 }
 117
 118 static int handle_attr_ack(struct vio_driver_state *vio,
 119                            struct vio_net_attr_info *pkt)
 120 {
 121         viodbg(HS, "GOT NET ATTR ACK\n");
 122
 123         return 0;
 124 }
 125
 126 static int handle_attr_nack(struct vio_driver_state *vio,
 127                             struct vio_net_attr_info *pkt)
 128 {
 129         viodbg(HS, "GOT NET ATTR NACK\n");
 130
 131         return -ECONNRESET;
 132 }
 133
 134 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
 135 {
 136         struct vio_net_attr_info *pkt = arg;
 137
 138         switch (pkt->tag.stype) {
 139         case VIO_SUBTYPE_INFO:
 140                 return handle_attr_info(vio, pkt);
 141
 142         case VIO_SUBTYPE_ACK:
 143                 return handle_attr_ack(vio, pkt);
 144
 145         case VIO_SUBTYPE_NACK:
 146                 return handle_attr_nack(vio, pkt);
 147
 148         default:
 149                 return -ECONNRESET;
 150         }
 151 }
 152
 153 static void vnet_handshake_complete(struct vio_driver_state *vio)
 154 {
 155         struct vio_dring_state *dr;
 156
 157         dr = &vio->drings[VIO_DRIVER_RX_RING];
 158         dr->snd_nxt = dr->rcv_nxt = 1;
 159
 160         dr = &vio->drings[VIO_DRIVER_TX_RING];
 161         dr->snd_nxt = dr->rcv_nxt = 1;
 162 }
 163
 164 /* The hypervisor interface that implements copying to/from imported
 165  * memory from another domain requires that copies are done to 8-byte
 166  * aligned buffers, and that the lengths of such copies are also 8-byte
 167  * multiples.
 168  *
 169  * So we align skb->data to an 8-byte multiple and pad-out the data
 170  * area so we can round the copy length up to the next multiple of
 171  * 8 for the copy.
 172  *
 173  * The transmitter puts the actual start of the packet 6 bytes into
 174  * the buffer it sends over, so that the IP headers after the ethernet
 175  * header are aligned properly.  These 6 bytes are not in the descriptor
 176  * length, they are simply implied.  This offset is represented using
 177  * the VNET_PACKET_SKIP macro.
 178  */
 179 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
 180                                            unsigned int len)
 181 {
 182         struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
 183         unsigned long addr, off;
 184
 185         if (unlikely(!skb))
 186                 return NULL;
 187
 188         addr = (unsigned long) skb->data;
 189         off = ((addr + 7UL) & ~7UL) - addr;
 190         if (off)
 191                 skb_reserve(skb, off);
 192
 193         return skb;
 194 }
 195
 196 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
 197                        struct ldc_trans_cookie *cookies, int ncookies)
 198 {
 199         struct net_device *dev = port->vp->dev;
 200         unsigned int copy_len;
 201         struct sk_buff *skb;
 202         int err;
 203
 204         err = -EMSGSIZE;
 205         if (unlikely(len < ETH_ZLEN || len > ETH_FRAME_LEN)) {
 206                 dev->stats.rx_length_errors++;
 207                 goto out_dropped;
 208         }
 209
 210         skb = alloc_and_align_skb(dev, len);
 211         err = -ENOMEM;
 212         if (unlikely(!skb)) {
 213                 dev->stats.rx_missed_errors++;
 214                 goto out_dropped;
 215         }
 216
 217         copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
 218         skb_put(skb, copy_len);
 219         err = ldc_copy(port->vio.lp, LDC_COPY_IN,
 220                        skb->data, copy_len, 0,
 221                        cookies, ncookies);
 222         if (unlikely(err < 0)) {
 223                 dev->stats.rx_frame_errors++;
 224                 goto out_free_skb;
 225         }
 226
 227         skb_pull(skb, VNET_PACKET_SKIP);
 228         skb_trim(skb, len);
 229         skb->protocol = eth_type_trans(skb, dev);
 230
 231         dev->stats.rx_packets++;
 232         dev->stats.rx_bytes += len;
 233
 234         netif_rx(skb);
 235
 236         return 0;
 237
 238 out_free_skb:
 239         kfree_skb(skb);
 240
 241 out_dropped:
 242         dev->stats.rx_dropped++;
 243         return err;
 244 }
 245
 246 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
 247                          u32 start, u32 end, u8 vio_dring_state)
 248 {
 249         struct vio_dring_data hdr = {
 250                 .tag = {
 251                         .type           = VIO_TYPE_DATA,
 252                         .stype          = VIO_SUBTYPE_ACK,
 253                         .stype_env      = VIO_DRING_DATA,
 254                         .sid            = vio_send_sid(&port->vio),
 255                 },
 256                 .dring_ident            = dr->ident,
 257                 .start_idx              = start,
 258                 .end_idx                = end,
 259                 .state                  = vio_dring_state,
 260         };
 261         int err, delay;
 262
 263         hdr.seq = dr->snd_nxt;
 264         delay = 1;
 265         do {
 266                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 267                 if (err > 0) {
 268                         dr->snd_nxt++;
 269                         break;
 270                 }
 271                 udelay(delay);
 272                 if ((delay <<= 1) > 128)
 273                         delay = 128;
 274         } while (err == -EAGAIN);
 275
 276         return err;
 277 }
 278
 279 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
 280 {
 281         if (++idx == dr->num_entries)
 282                 idx = 0;
 283         return idx;
 284 }
 285
 286 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
 287 {
 288         if (idx == 0)
 289                 idx = dr->num_entries - 1;
 290         else
 291                 idx--;
 292
 293         return idx;
 294 }
 295
 296 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
 297                                         struct vio_dring_state *dr,
 298                                         u32 index)
 299 {
 300         struct vio_net_desc *desc = port->vio.desc_buf;
 301         int err;
 302
 303         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
 304                                   (index * dr->entry_size),
 305                                   dr->cookies, dr->ncookies);
 306         if (err < 0)
 307                 return ERR_PTR(err);
 308
 309         return desc;
 310 }
 311
 312 static int put_rx_desc(struct vnet_port *port,
 313                        struct vio_dring_state *dr,
 314                        struct vio_net_desc *desc,
 315                        u32 index)
 316 {
 317         int err;
 318
 319         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
 320                                   (index * dr->entry_size),
 321                                   dr->cookies, dr->ncookies);
 322         if (err < 0)
 323                 return err;
 324
 325         return 0;
 326 }
 327
 328 static int vnet_walk_rx_one(struct vnet_port *port,
 329                             struct vio_dring_state *dr,
 330                             u32 index, int *needs_ack)
 331 {
 332         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
 333         struct vio_driver_state *vio = &port->vio;
 334         int err;
 335
 336         if (IS_ERR(desc))
 337                 return PTR_ERR(desc);
 338
 339         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 340                desc->hdr.state, desc->hdr.ack,
 341                desc->size, desc->ncookies,
 342                desc->cookies[0].cookie_addr,
 343                desc->cookies[0].cookie_size);
 344
 345         if (desc->hdr.state != VIO_DESC_READY)
 346                 return 1;
 347         err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
 348         if (err == -ECONNRESET)
 349                 return err;
 350         desc->hdr.state = VIO_DESC_DONE;
 351         err = put_rx_desc(port, dr, desc, index);
 352         if (err < 0)
 353                 return err;
 354         *needs_ack = desc->hdr.ack;
 355         return 0;
 356 }
 357
 358 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 359                         u32 start, u32 end)
 360 {
 361         struct vio_driver_state *vio = &port->vio;
 362         int ack_start = -1, ack_end = -1;
 363
 364         end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
 365
 366         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
 367
 368         while (start != end) {
 369                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
 370                 if (err == -ECONNRESET)
 371                         return err;
 372                 if (err != 0)
 373                         break;
 374                 if (ack_start == -1)
 375                         ack_start = start;
 376                 ack_end = start;
 377                 start = next_idx(start, dr);
 378                 if (ack && start != end) {
 379                         err = vnet_send_ack(port, dr, ack_start, ack_end,
 380                                             VIO_DRING_ACTIVE);
 381                         if (err == -ECONNRESET)
 382                                 return err;
 383                         ack_start = -1;
 384                 }
 385         }
 386         if (unlikely(ack_start == -1))
 387                 ack_start = ack_end = prev_idx(start, dr);
 388         return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
 389 }
 390
 391 static int vnet_rx(struct vnet_port *port, void *msgbuf)
 392 {
 393         struct vio_dring_data *pkt = msgbuf;
 394         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
 395         struct vio_driver_state *vio = &port->vio;
 396
 397         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
 398                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
 399
 400         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 401                 return 0;
 402         if (unlikely(pkt->seq != dr->rcv_nxt)) {
 403                 printk(KERN_ERR PFX "RX out of sequence seq[0x%llx] "
 404                        "rcv_nxt[0x%llx]\n", pkt->seq, dr->rcv_nxt);
 405                 return 0;
 406         }
 407
 408         dr->rcv_nxt++;
 409
 410         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
 411
 412         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
 413 }
 414
 415 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
 416 {
 417         u32 idx = dr->cons;
 418         int found = 0;
 419
 420         while (idx != dr->prod) {
 421                 if (idx == end) {
 422                         found = 1;
 423                         break;
 424                 }
 425                 idx = next_idx(idx, dr);
 426         }
 427         return found;
 428 }
 429
 430 static int vnet_ack(struct vnet_port *port, void *msgbuf)
 431 {
 432         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 433         struct vio_dring_data *pkt = msgbuf;
 434         struct net_device *dev;
 435         struct vnet *vp;
 436         u32 end;
 437
 438         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 439                 return 0;
 440
 441         end = pkt->end_idx;
 442         if (unlikely(!idx_is_pending(dr, end)))
 443                 return 0;
 444
 445         dr->cons = next_idx(end, dr);
 446
 447         vp = port->vp;
 448         dev = vp->dev;
 449         if (unlikely(netif_queue_stopped(dev) &&
 450                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
 451                 return 1;
 452
 453         return 0;
 454 }
 455
 456 static int vnet_nack(struct vnet_port *port, void *msgbuf)
 457 {
 458         /* XXX just reset or similar XXX */
 459         return 0;
 460 }
 461
 462 static int handle_mcast(struct vnet_port *port, void *msgbuf)
 463 {
 464         struct vio_net_mcast_info *pkt = msgbuf;
 465
 466         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
 467                 printk(KERN_ERR PFX "%s: Got unexpected MCAST reply "
 468                        "[%02x:%02x:%04x:%08x]\n",
 469                        port->vp->dev->name,
 470                        pkt->tag.type,
 471                        pkt->tag.stype,
 472                        pkt->tag.stype_env,
 473                        pkt->tag.sid);
 474
 475         return 0;
 476 }
 477
 478 static void maybe_tx_wakeup(struct vnet *vp)
 479 {
 480         struct net_device *dev = vp->dev;
 481
 482         netif_tx_lock(dev);
 483         if (likely(netif_queue_stopped(dev))) {
 484                 struct vnet_port *port;
 485                 int wake = 1;
 486
 487                 list_for_each_entry(port, &vp->port_list, list) {
 488                         struct vio_dring_state *dr;
 489
 490                         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 491                         if (vnet_tx_dring_avail(dr) <
 492                             VNET_TX_WAKEUP_THRESH(dr)) {
 493                                 wake = 0;
 494                                 break;
 495                         }
 496                 }
 497                 if (wake)
 498                         netif_wake_queue(dev);
 499         }
 500         netif_tx_unlock(dev);
 501 }
 502
 503 static void vnet_event(void *arg, int event)
 504 {
 505         struct vnet_port *port = arg;
 506         struct vio_driver_state *vio = &port->vio;
 507         unsigned long flags;
 508         int tx_wakeup, err;
 509
 510         spin_lock_irqsave(&vio->lock, flags);
 511
 512         if (unlikely(event == LDC_EVENT_RESET ||
 513                      event == LDC_EVENT_UP)) {
 514                 vio_link_state_change(vio, event);
 515                 spin_unlock_irqrestore(&vio->lock, flags);
 516
 517                 if (event == LDC_EVENT_RESET)
 518                         vio_port_up(vio);
 519                 return;
 520         }
 521
 522         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 523                 printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
 524                 spin_unlock_irqrestore(&vio->lock, flags);
 525                 return;
 526         }
 527
 528         tx_wakeup = err = 0;
 529         while (1) {
 530                 union {
 531                         struct vio_msg_tag tag;
 532                         u64 raw[8];
 533                 } msgbuf;
 534
 535                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 536                 if (unlikely(err < 0)) {
 537                         if (err == -ECONNRESET)
 538                                 vio_conn_reset(vio);
 539                         break;
 540                 }
 541                 if (err == 0)
 542                         break;
 543                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 544                        msgbuf.tag.type,
 545                        msgbuf.tag.stype,
 546                        msgbuf.tag.stype_env,
 547                        msgbuf.tag.sid);
 548                 err = vio_validate_sid(vio, &msgbuf.tag);
 549                 if (err < 0)
 550                         break;
 551
 552                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 553                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 554                                 err = vnet_rx(port, &msgbuf);
 555                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 556                                 err = vnet_ack(port, &msgbuf);
 557                                 if (err > 0)
 558                                         tx_wakeup |= err;
 559                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 560                                 err = vnet_nack(port, &msgbuf);
 561                         }
 562                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 563                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 564                                 err = handle_mcast(port, &msgbuf);
 565                         else
 566                                 err = vio_control_pkt_engine(vio, &msgbuf);
 567                         if (err)
 568                                 break;
 569                 } else {
 570                         err = vnet_handle_unknown(port, &msgbuf);
 571                 }
 572                 if (err == -ECONNRESET)
 573                         break;
 574         }
 575         spin_unlock(&vio->lock);
 576         if (unlikely(tx_wakeup && err != -ECONNRESET))
 577                 maybe_tx_wakeup(port->vp);
 578         local_irq_restore(flags);
 579 }
 580
 581 static int __vnet_tx_trigger(struct vnet_port *port)
 582 {
 583         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 584         struct vio_dring_data hdr = {
 585                 .tag = {
 586                         .type           = VIO_TYPE_DATA,
 587                         .stype          = VIO_SUBTYPE_INFO,
 588                         .stype_env      = VIO_DRING_DATA,
 589                         .sid            = vio_send_sid(&port->vio),
 590                 },
 591                 .dring_ident            = dr->ident,
 592                 .start_idx              = dr->prod,
 593                 .end_idx                = (u32) -1,
 594         };
 595         int err, delay;
 596
 597         hdr.seq = dr->snd_nxt;
 598         delay = 1;
 599         do {
 600                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 601                 if (err > 0) {
 602                         dr->snd_nxt++;
 603                         break;
 604                 }
 605                 udelay(delay);
 606                 if ((delay <<= 1) > 128)
 607                         delay = 128;
 608         } while (err == -EAGAIN);
 609
 610         return err;
 611 }
 612
 613 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 614 {
 615         unsigned int hash = vnet_hashfn(skb->data);
 616         struct hlist_head *hp = &vp->port_hash[hash];
 617         struct hlist_node *n;
 618         struct vnet_port *port;
 619
 620         hlist_for_each_entry(port, n, hp, hash) {
 621                 if (!compare_ether_addr(port->raddr, skb->data))
 622                         return port;
 623         }
 624         port = NULL;
 625         if (!list_empty(&vp->port_list))
 626                 port = list_entry(vp->port_list.next, struct vnet_port, list);
 627
 628         return port;
 629 }
 630
 631 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
 632 {
 633         struct vnet_port *ret;
 634         unsigned long flags;
 635
 636         spin_lock_irqsave(&vp->lock, flags);
 637         ret = __tx_port_find(vp, skb);
 638         spin_unlock_irqrestore(&vp->lock, flags);
 639
 640         return ret;
 641 }
 642
 643 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 644 {
 645         struct vnet *vp = netdev_priv(dev);
 646         struct vnet_port *port = tx_port_find(vp, skb);
 647         struct vio_dring_state *dr;
 648         struct vio_net_desc *d;
 649         unsigned long flags;
 650         unsigned int len;
 651         void *tx_buf;
 652         int i, err;
 653
 654         if (unlikely(!port))
 655                 goto out_dropped;
 656
 657         spin_lock_irqsave(&port->vio.lock, flags);
 658
 659         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 660         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 661                 if (!netif_queue_stopped(dev)) {
 662                         netif_stop_queue(dev);
 663
 664                         /* This is a hard error, log it. */
 665                         printk(KERN_ERR PFX "%s: BUG! Tx Ring full when "
 666                                "queue awake!\n", dev->name);
 667                         dev->stats.tx_errors++;
 668                 }
 669                 spin_unlock_irqrestore(&port->vio.lock, flags);
 670                 return NETDEV_TX_BUSY;
 671         }
 672
 673         d = vio_dring_cur(dr);
 674
 675         tx_buf = port->tx_bufs[dr->prod].buf;
 676         skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
 677
 678         len = skb->len;
 679         if (len < ETH_ZLEN) {
 680                 len = ETH_ZLEN;
 681                 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
 682         }
 683
 684         d->hdr.ack = VIO_ACK_ENABLE;
 685         d->size = len;
 686         d->ncookies = port->tx_bufs[dr->prod].ncookies;
 687         for (i = 0; i < d->ncookies; i++)
 688                 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
 689
 690         /* This has to be a non-SMP write barrier because we are writing
 691          * to memory which is shared with the peer LDOM.
 692          */
 693         wmb();
 694
 695         d->hdr.state = VIO_DESC_READY;
 696
 697         err = __vnet_tx_trigger(port);
 698         if (unlikely(err < 0)) {
 699                 printk(KERN_INFO PFX "%s: TX trigger error %d\n",
 700                        dev->name, err);
 701                 d->hdr.state = VIO_DESC_FREE;
 702                 dev->stats.tx_carrier_errors++;
 703                 goto out_dropped_unlock;
 704         }
 705
 706         dev->stats.tx_packets++;
 707         dev->stats.tx_bytes += skb->len;
 708
 709         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
 710         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 711                 netif_stop_queue(dev);
 712                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
 713                         netif_wake_queue(dev);
 714         }
 715
 716         spin_unlock_irqrestore(&port->vio.lock, flags);
 717
 718         dev_kfree_skb(skb);
 719
 720         dev->trans_start = jiffies;
 721         return NETDEV_TX_OK;
 722
 723 out_dropped_unlock:
 724         spin_unlock_irqrestore(&port->vio.lock, flags);
 725
 726 out_dropped:
 727         dev_kfree_skb(skb);
 728         dev->stats.tx_dropped++;
 729         return NETDEV_TX_OK;
 730 }
 731
 732 static void vnet_tx_timeout(struct net_device *dev)
 733 {
 734         /* XXX Implement me XXX */
 735 }
 736
 737 static int vnet_open(struct net_device *dev)
 738 {
 739         netif_carrier_on(dev);
 740         netif_start_queue(dev);
 741
 742         return 0;
 743 }
 744
 745 static int vnet_close(struct net_device *dev)
 746 {
 747         netif_stop_queue(dev);
 748         netif_carrier_off(dev);
 749
 750         return 0;
 751 }
 752
 753 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
 754 {
 755         struct vnet_mcast_entry *m;
 756
 757         for (m = vp->mcast_list; m; m = m->next) {
 758                 if (!memcmp(m->addr, addr, ETH_ALEN))
 759                         return m;
 760         }
 761         return NULL;
 762 }
 763
 764 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
 765 {
 766         struct dev_addr_list *p;
 767
 768         for (p = dev->mc_list; p; p = p->next) {
 769                 struct vnet_mcast_entry *m;
 770
 771                 m = __vnet_mc_find(vp, p->dmi_addr);
 772                 if (m) {
 773                         m->hit = 1;
 774                         continue;
 775                 }
 776
 777                 if (!m) {
 778                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
 779                         if (!m)
 780                                 continue;
 781                         memcpy(m->addr, p->dmi_addr, ETH_ALEN);
 782                         m->hit = 1;
 783
 784                         m->next = vp->mcast_list;
 785                         vp->mcast_list = m;
 786                 }
 787         }
 788 }
 789
 790 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
 791 {
 792         struct vio_net_mcast_info info;
 793         struct vnet_mcast_entry *m, **pp;
 794         int n_addrs;
 795
 796         memset(&info, 0, sizeof(info));
 797
 798         info.tag.type = VIO_TYPE_CTRL;
 799         info.tag.stype = VIO_SUBTYPE_INFO;
 800         info.tag.stype_env = VNET_MCAST_INFO;
 801         info.tag.sid = vio_send_sid(&port->vio);
 802         info.set = 1;
 803
 804         n_addrs = 0;
 805         for (m = vp->mcast_list; m; m = m->next) {
 806                 if (m->sent)
 807                         continue;
 808                 m->sent = 1;
 809                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 810                        m->addr, ETH_ALEN);
 811                 if (++n_addrs == VNET_NUM_MCAST) {
 812                         info.count = n_addrs;
 813
 814                         (void) vio_ldc_send(&port->vio, &info,
 815                                             sizeof(info));
 816                         n_addrs = 0;
 817                 }
 818         }
 819         if (n_addrs) {
 820                 info.count = n_addrs;
 821                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 822         }
 823
 824         info.set = 0;
 825
 826         n_addrs = 0;
 827         pp = &vp->mcast_list;
 828         while ((m = *pp) != NULL) {
 829                 if (m->hit) {
 830                         m->hit = 0;
 831                         pp = &m->next;
 832                         continue;
 833                 }
 834
 835                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 836                        m->addr, ETH_ALEN);
 837                 if (++n_addrs == VNET_NUM_MCAST) {
 838                         info.count = n_addrs;
 839                         (void) vio_ldc_send(&port->vio, &info,
 840                                             sizeof(info));
 841                         n_addrs = 0;
 842                 }
 843
 844                 *pp = m->next;
 845                 kfree(m);
 846         }
 847         if (n_addrs) {
 848                 info.count = n_addrs;
 849                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 850         }
 851 }
 852
 853 static void vnet_set_rx_mode(struct net_device *dev)
 854 {
 855         struct vnet *vp = netdev_priv(dev);
 856         struct vnet_port *port;
 857         unsigned long flags;
 858
 859         spin_lock_irqsave(&vp->lock, flags);
 860         if (!list_empty(&vp->port_list)) {
 861                 port = list_entry(vp->port_list.next, struct vnet_port, list);
 862
 863                 if (port->switch_port) {
 864                         __update_mc_list(vp, dev);
 865                         __send_mc_list(vp, port);
 866                 }
 867         }
 868         spin_unlock_irqrestore(&vp->lock, flags);
 869 }
 870
 871 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
 872 {
 873         if (new_mtu != ETH_DATA_LEN)
 874                 return -EINVAL;
 875
 876         dev->mtu = new_mtu;
 877         return 0;
 878 }
 879
 880 static int vnet_set_mac_addr(struct net_device *dev, void *p)
 881 {
 882         return -EINVAL;
 883 }
 884
 885 static void vnet_get_drvinfo(struct net_device *dev,
 886                              struct ethtool_drvinfo *info)
 887 {
 888         strcpy(info->driver, DRV_MODULE_NAME);
 889         strcpy(info->version, DRV_MODULE_VERSION);
 890 }
 891
 892 static u32 vnet_get_msglevel(struct net_device *dev)
 893 {
 894         struct vnet *vp = netdev_priv(dev);
 895         return vp->msg_enable;
 896 }
 897
 898 static void vnet_set_msglevel(struct net_device *dev, u32 value)
 899 {
 900         struct vnet *vp = netdev_priv(dev);
 901         vp->msg_enable = value;
 902 }
 903
 904 static const struct ethtool_ops vnet_ethtool_ops = {
 905         .get_drvinfo            = vnet_get_drvinfo,
 906         .get_msglevel           = vnet_get_msglevel,
 907         .set_msglevel           = vnet_set_msglevel,
 908         .get_link               = ethtool_op_get_link,
 909 };
 910
 911 static void vnet_port_free_tx_bufs(struct vnet_port *port)
 912 {
 913         struct vio_dring_state *dr;
 914         int i;
 915
 916         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 917         if (dr->base) {
 918                 ldc_free_exp_dring(port->vio.lp, dr->base,
 919                                    (dr->entry_size * dr->num_entries),
 920                                    dr->cookies, dr->ncookies);
 921                 dr->base = NULL;
 922                 dr->entry_size = 0;
 923                 dr->num_entries = 0;
 924                 dr->pending = 0;
 925                 dr->ncookies = 0;
 926         }
 927
 928         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 929                 void *buf = port->tx_bufs[i].buf;
 930
 931                 if (!buf)
 932                         continue;
 933
 934                 ldc_unmap(port->vio.lp,
 935                           port->tx_bufs[i].cookies,
 936                           port->tx_bufs[i].ncookies);
 937
 938                 kfree(buf);
 939                 port->tx_bufs[i].buf = NULL;
 940         }
 941 }
 942
 943 static int __devinit vnet_port_alloc_tx_bufs(struct vnet_port *port)
 944 {
 945         struct vio_dring_state *dr;
 946         unsigned long len;
 947         int i, err, ncookies;
 948         void *dring;
 949
 950         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 951                 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
 952                 int map_len = (ETH_FRAME_LEN + 7) & ~7;
 953
 954                 err = -ENOMEM;
 955                 if (!buf) {
 956                         printk(KERN_ERR "TX buffer allocation failure\n");
 957                         goto err_out;
 958                 }
 959                 err = -EFAULT;
 960                 if ((unsigned long)buf & (8UL - 1)) {
 961                         printk(KERN_ERR "TX buffer misaligned\n");
 962                         kfree(buf);
 963                         goto err_out;
 964                 }
 965
 966                 err = ldc_map_single(port->vio.lp, buf, map_len,
 967                                      port->tx_bufs[i].cookies, 2,
 968                                      (LDC_MAP_SHADOW |
 969                                       LDC_MAP_DIRECT |
 970                                       LDC_MAP_RW));
 971                 if (err < 0) {
 972                         kfree(buf);
 973                         goto err_out;
 974                 }
 975                 port->tx_bufs[i].buf = buf;
 976                 port->tx_bufs[i].ncookies = err;
 977         }
 978
 979         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 980
 981         len = (VNET_TX_RING_SIZE *
 982                (sizeof(struct vio_net_desc) +
 983                 (sizeof(struct ldc_trans_cookie) * 2)));
 984
 985         ncookies = VIO_MAX_RING_COOKIES;
 986         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 987                                     dr->cookies, &ncookies,
 988                                     (LDC_MAP_SHADOW |
 989                                      LDC_MAP_DIRECT |
 990                                      LDC_MAP_RW));
 991         if (IS_ERR(dring)) {
 992                 err = PTR_ERR(dring);
 993                 goto err_out;
 994         }
 995
 996         dr->base = dring;
 997         dr->entry_size = (sizeof(struct vio_net_desc) +
 998                           (sizeof(struct ldc_trans_cookie) * 2));
 999         dr->num_entries = VNET_TX_RING_SIZE;
1000         dr->prod = dr->cons = 0;
1001         dr->pending = VNET_TX_RING_SIZE;
1002         dr->ncookies = ncookies;
1003
1004         return 0;
1005
1006 err_out:
1007         vnet_port_free_tx_bufs(port);
1008
1009         return err;
1010 }
1011
1012 static LIST_HEAD(vnet_list);
1013 static DEFINE_MUTEX(vnet_list_mutex);
1014
1015 static const struct net_device_ops vnet_ops = {
1016         .ndo_open               = vnet_open,
1017         .ndo_stop               = vnet_close,
1018         .ndo_set_multicast_list = vnet_set_rx_mode,
1019         .ndo_set_mac_address    = vnet_set_mac_addr,
1020         .ndo_tx_timeout         = vnet_tx_timeout,
1021         .ndo_change_mtu         = vnet_change_mtu,
1022         .ndo_start_xmit         = vnet_start_xmit,
1023 };
1024
1025 static struct vnet * __devinit vnet_new(const u64 *local_mac)
1026 {
1027         struct net_device *dev;
1028         struct vnet *vp;
1029         int err, i;
1030
1031         dev = alloc_etherdev(sizeof(*vp));
1032         if (!dev) {
1033                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1034                 return ERR_PTR(-ENOMEM);
1035         }
1036
1037         for (i = 0; i < ETH_ALEN; i++)
1038                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1039
1040         memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
1041
1042         vp = netdev_priv(dev);
1043
1044         spin_lock_init(&vp->lock);
1045         vp->dev = dev;
1046
1047         INIT_LIST_HEAD(&vp->port_list);
1048         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1049                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1050         INIT_LIST_HEAD(&vp->list);
1051         vp->local_mac = *local_mac;
1052
1053         dev->netdev_ops = &vnet_ops;
1054         dev->ethtool_ops = &vnet_ethtool_ops;
1055         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1056
1057         err = register_netdev(dev);
1058         if (err) {
1059                 printk(KERN_ERR PFX "Cannot register net device, "
1060                        "aborting.\n");
1061                 goto err_out_free_dev;
1062         }
1063
1064         printk(KERN_INFO "%s: Sun LDOM vnet ", dev->name);
1065
1066         for (i = 0; i < 6; i++)
1067                 printk("%2.2x%c", dev->dev_addr[i], i == 5 ? '\n' : ':');
1068
1069         list_add(&vp->list, &vnet_list);
1070
1071         return vp;
1072
1073 err_out_free_dev:
1074         free_netdev(dev);
1075
1076         return ERR_PTR(err);
1077 }
1078
1079 static struct vnet * __devinit vnet_find_or_create(const u64 *local_mac)
1080 {
1081         struct vnet *iter, *vp;
1082
1083         mutex_lock(&vnet_list_mutex);
1084         vp = NULL;
1085         list_for_each_entry(iter, &vnet_list, list) {
1086                 if (iter->local_mac == *local_mac) {
1087                         vp = iter;
1088                         break;
1089                 }
1090         }
1091         if (!vp)
1092                 vp = vnet_new(local_mac);
1093         mutex_unlock(&vnet_list_mutex);
1094
1095         return vp;
1096 }
1097
1098 static const char *local_mac_prop = "local-mac-address";
1099
1100 static struct vnet * __devinit vnet_find_parent(struct mdesc_handle *hp,
1101                                                 u64 port_node)
1102 {
1103         const u64 *local_mac = NULL;
1104         u64 a;
1105
1106         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1107                 u64 target = mdesc_arc_target(hp, a);
1108                 const char *name;
1109
1110                 name = mdesc_get_property(hp, target, "name", NULL);
1111                 if (!name || strcmp(name, "network"))
1112                         continue;
1113
1114                 local_mac = mdesc_get_property(hp, target,
1115                                                local_mac_prop, NULL);
1116                 if (local_mac)
1117                         break;
1118         }
1119         if (!local_mac)
1120                 return ERR_PTR(-ENODEV);
1121
1122         return vnet_find_or_create(local_mac);
1123 }
1124
1125 static struct ldc_channel_config vnet_ldc_cfg = {
1126         .event          = vnet_event,
1127         .mtu            = 64,
1128         .mode           = LDC_MODE_UNRELIABLE,
1129 };
1130
1131 static struct vio_driver_ops vnet_vio_ops = {
1132         .send_attr              = vnet_send_attr,
1133         .handle_attr            = vnet_handle_attr,
1134         .handshake_complete     = vnet_handshake_complete,
1135 };
1136
1137 static void __devinit print_version(void)
1138 {
1139         static int version_printed;
1140
1141         if (version_printed++ == 0)
1142                 printk(KERN_INFO "%s", version);
1143 }
1144
1145 const char *remote_macaddr_prop = "remote-mac-address";
1146
1147 static int __devinit vnet_port_probe(struct vio_dev *vdev,
1148                                      const struct vio_device_id *id)
1149 {
1150         struct mdesc_handle *hp;
1151         struct vnet_port *port;
1152         unsigned long flags;
1153         struct vnet *vp;
1154         const u64 *rmac;
1155         int len, i, err, switch_port;
1156
1157         print_version();
1158
1159         hp = mdesc_grab();
1160
1161         vp = vnet_find_parent(hp, vdev->mp);
1162         if (IS_ERR(vp)) {
1163                 printk(KERN_ERR PFX "Cannot find port parent vnet.\n");
1164                 err = PTR_ERR(vp);
1165                 goto err_out_put_mdesc;
1166         }
1167
1168         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1169         err = -ENODEV;
1170         if (!rmac) {
1171                 printk(KERN_ERR PFX "Port lacks %s property.\n",
1172                        remote_macaddr_prop);
1173                 goto err_out_put_mdesc;
1174         }
1175
1176         port = kzalloc(sizeof(*port), GFP_KERNEL);
1177         err = -ENOMEM;
1178         if (!port) {
1179                 printk(KERN_ERR PFX "Cannot allocate vnet_port.\n");
1180                 goto err_out_put_mdesc;
1181         }
1182
1183         for (i = 0; i < ETH_ALEN; i++)
1184                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1185
1186         port->vp = vp;
1187
1188         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1189                               vnet_versions, ARRAY_SIZE(vnet_versions),
1190                               &vnet_vio_ops, vp->dev->name);
1191         if (err)
1192                 goto err_out_free_port;
1193
1194         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1195         if (err)
1196                 goto err_out_free_port;
1197
1198         err = vnet_port_alloc_tx_bufs(port);
1199         if (err)
1200                 goto err_out_free_ldc;
1201
1202         INIT_HLIST_NODE(&port->hash);
1203         INIT_LIST_HEAD(&port->list);
1204
1205         switch_port = 0;
1206         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1207                 switch_port = 1;
1208         port->switch_port = switch_port;
1209
1210         spin_lock_irqsave(&vp->lock, flags);
1211         if (switch_port)
1212                 list_add(&port->list, &vp->port_list);
1213         else
1214                 list_add_tail(&port->list, &vp->port_list);
1215         hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1216         spin_unlock_irqrestore(&vp->lock, flags);
1217
1218         dev_set_drvdata(&vdev->dev, port);
1219
1220         printk(KERN_INFO "%s: PORT ( remote-mac %pM%s )\n",
1221                vp->dev->name, port->raddr,
1222                switch_port ? " switch-port" : "");
1223
1224         vio_port_up(&port->vio);
1225
1226         mdesc_release(hp);
1227
1228         return 0;
1229
1230 err_out_free_ldc:
1231         vio_ldc_free(&port->vio);
1232
1233 err_out_free_port:
1234         kfree(port);
1235
1236 err_out_put_mdesc:
1237         mdesc_release(hp);
1238         return err;
1239 }
1240
1241 static int vnet_port_remove(struct vio_dev *vdev)
1242 {
1243         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1244
1245         if (port) {
1246                 struct vnet *vp = port->vp;
1247                 unsigned long flags;
1248
1249                 del_timer_sync(&port->vio.timer);
1250
1251                 spin_lock_irqsave(&vp->lock, flags);
1252                 list_del(&port->list);
1253                 hlist_del(&port->hash);
1254                 spin_unlock_irqrestore(&vp->lock, flags);
1255
1256                 vnet_port_free_tx_bufs(port);
1257                 vio_ldc_free(&port->vio);
1258
1259                 dev_set_drvdata(&vdev->dev, NULL);
1260
1261                 kfree(port);
1262         }
1263         return 0;
1264 }
1265
1266 static const struct vio_device_id vnet_port_match[] = {
1267         {
1268                 .type = "vnet-port",
1269         },
1270         {},
1271 };
1272 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1273
1274 static struct vio_driver vnet_port_driver = {
1275         .id_table       = vnet_port_match,
1276         .probe          = vnet_port_probe,
1277         .remove         = vnet_port_remove,
1278         .driver         = {
1279                 .name   = "vnet_port",
1280                 .owner  = THIS_MODULE,
1281         }
1282 };
1283
1284 static int __init vnet_init(void)
1285 {
1286         return vio_register_driver(&vnet_port_driver);
1287 }
1288
1289 static void __exit vnet_exit(void)
1290 {
1291         vio_unregister_driver(&vnet_port_driver);
1292 }
1293
1294 module_init(vnet_init);
1295 module_exit(vnet_exit);