drivers/net/sunvnet.c

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