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 #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:%lx:%lx]\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[%016lx] rcv_nxt[%016lx]\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%lx] "
 404                        "rcv_nxt[0x%lx]\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 void maybe_tx_wakeup(struct vnet *vp)
 463 {
 464         struct net_device *dev = vp->dev;
 465
 466         netif_tx_lock(dev);
 467         if (likely(netif_queue_stopped(dev))) {
 468                 struct vnet_port *port;
 469                 int wake = 1;
 470
 471                 list_for_each_entry(port, &vp->port_list, list) {
 472                         struct vio_dring_state *dr;
 473
 474                         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 475                         if (vnet_tx_dring_avail(dr) <
 476                             VNET_TX_WAKEUP_THRESH(dr)) {
 477                                 wake = 0;
 478                                 break;
 479                         }
 480                 }
 481                 if (wake)
 482                         netif_wake_queue(dev);
 483         }
 484         netif_tx_unlock(dev);
 485 }
 486
 487 static void vnet_event(void *arg, int event)
 488 {
 489         struct vnet_port *port = arg;
 490         struct vio_driver_state *vio = &port->vio;
 491         unsigned long flags;
 492         int tx_wakeup, err;
 493
 494         spin_lock_irqsave(&vio->lock, flags);
 495
 496         if (unlikely(event == LDC_EVENT_RESET ||
 497                      event == LDC_EVENT_UP)) {
 498                 vio_link_state_change(vio, event);
 499                 spin_unlock_irqrestore(&vio->lock, flags);
 500
 501                 if (event == LDC_EVENT_RESET)
 502                         vio_port_up(vio);
 503                 return;
 504         }
 505
 506         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 507                 printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
 508                 spin_unlock_irqrestore(&vio->lock, flags);
 509                 return;
 510         }
 511
 512         tx_wakeup = err = 0;
 513         while (1) {
 514                 union {
 515                         struct vio_msg_tag tag;
 516                         u64 raw[8];
 517                 } msgbuf;
 518
 519                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 520                 if (unlikely(err < 0)) {
 521                         if (err == -ECONNRESET)
 522                                 vio_conn_reset(vio);
 523                         break;
 524                 }
 525                 if (err == 0)
 526                         break;
 527                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 528                        msgbuf.tag.type,
 529                        msgbuf.tag.stype,
 530                        msgbuf.tag.stype_env,
 531                        msgbuf.tag.sid);
 532                 err = vio_validate_sid(vio, &msgbuf.tag);
 533                 if (err < 0)
 534                         break;
 535
 536                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 537                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 538                                 err = vnet_rx(port, &msgbuf);
 539                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 540                                 err = vnet_ack(port, &msgbuf);
 541                                 if (err > 0)
 542                                         tx_wakeup |= err;
 543                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 544                                 err = vnet_nack(port, &msgbuf);
 545                         }
 546                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 547                         err = vio_control_pkt_engine(vio, &msgbuf);
 548                         if (err)
 549                                 break;
 550                 } else {
 551                         err = vnet_handle_unknown(port, &msgbuf);
 552                 }
 553                 if (err == -ECONNRESET)
 554                         break;
 555         }
 556         spin_unlock(&vio->lock);
 557         if (unlikely(tx_wakeup && err != -ECONNRESET))
 558                 maybe_tx_wakeup(port->vp);
 559         local_irq_restore(flags);
 560 }
 561
 562 static int __vnet_tx_trigger(struct vnet_port *port)
 563 {
 564         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 565         struct vio_dring_data hdr = {
 566                 .tag = {
 567                         .type           = VIO_TYPE_DATA,
 568                         .stype          = VIO_SUBTYPE_INFO,
 569                         .stype_env      = VIO_DRING_DATA,
 570                         .sid            = vio_send_sid(&port->vio),
 571                 },
 572                 .dring_ident            = dr->ident,
 573                 .start_idx              = dr->prod,
 574                 .end_idx                = (u32) -1,
 575         };
 576         int err, delay;
 577
 578         hdr.seq = dr->snd_nxt;
 579         delay = 1;
 580         do {
 581                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 582                 if (err > 0) {
 583                         dr->snd_nxt++;
 584                         break;
 585                 }
 586                 udelay(delay);
 587                 if ((delay <<= 1) > 128)
 588                         delay = 128;
 589         } while (err == -EAGAIN);
 590
 591         return err;
 592 }
 593
 594 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 595 {
 596         unsigned int hash = vnet_hashfn(skb->data);
 597         struct hlist_head *hp = &vp->port_hash[hash];
 598         struct hlist_node *n;
 599         struct vnet_port *port;
 600
 601         hlist_for_each_entry(port, n, hp, hash) {
 602                 if (!compare_ether_addr(port->raddr, skb->data))
 603                         return port;
 604         }
 605         port = NULL;
 606         if (!list_empty(&vp->port_list))
 607                 port = list_entry(vp->port_list.next, struct vnet_port, list);
 608
 609         return port;
 610 }
 611
 612 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
 613 {
 614         struct vnet_port *ret;
 615         unsigned long flags;
 616
 617         spin_lock_irqsave(&vp->lock, flags);
 618         ret = __tx_port_find(vp, skb);
 619         spin_unlock_irqrestore(&vp->lock, flags);
 620
 621         return ret;
 622 }
 623
 624 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 625 {
 626         struct vnet *vp = netdev_priv(dev);
 627         struct vnet_port *port = tx_port_find(vp, skb);
 628         struct vio_dring_state *dr;
 629         struct vio_net_desc *d;
 630         unsigned long flags;
 631         unsigned int len;
 632         void *tx_buf;
 633         int i, err;
 634
 635         if (unlikely(!port))
 636                 goto out_dropped;
 637
 638         spin_lock_irqsave(&port->vio.lock, flags);
 639
 640         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 641         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 642                 if (!netif_queue_stopped(dev)) {
 643                         netif_stop_queue(dev);
 644
 645                         /* This is a hard error, log it. */
 646                         printk(KERN_ERR PFX "%s: BUG! Tx Ring full when "
 647                                "queue awake!\n", dev->name);
 648                         dev->stats.tx_errors++;
 649                 }
 650                 spin_unlock_irqrestore(&port->vio.lock, flags);
 651                 return NETDEV_TX_BUSY;
 652         }
 653
 654         d = vio_dring_cur(dr);
 655
 656         tx_buf = port->tx_bufs[dr->prod].buf;
 657         skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
 658
 659         len = skb->len;
 660         if (len < ETH_ZLEN) {
 661                 len = ETH_ZLEN;
 662                 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
 663         }
 664
 665         d->hdr.ack = VIO_ACK_ENABLE;
 666         d->size = len;
 667         d->ncookies = port->tx_bufs[dr->prod].ncookies;
 668         for (i = 0; i < d->ncookies; i++)
 669                 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
 670
 671         /* This has to be a non-SMP write barrier because we are writing
 672          * to memory which is shared with the peer LDOM.
 673          */
 674         wmb();
 675
 676         d->hdr.state = VIO_DESC_READY;
 677
 678         err = __vnet_tx_trigger(port);
 679         if (unlikely(err < 0)) {
 680                 printk(KERN_INFO PFX "%s: TX trigger error %d\n",
 681                        dev->name, err);
 682                 d->hdr.state = VIO_DESC_FREE;
 683                 dev->stats.tx_carrier_errors++;
 684                 goto out_dropped_unlock;
 685         }
 686
 687         dev->stats.tx_packets++;
 688         dev->stats.tx_bytes += skb->len;
 689
 690         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
 691         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 692                 netif_stop_queue(dev);
 693                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
 694                         netif_wake_queue(dev);
 695         }
 696
 697         spin_unlock_irqrestore(&port->vio.lock, flags);
 698
 699         dev_kfree_skb(skb);
 700
 701         dev->trans_start = jiffies;
 702         return NETDEV_TX_OK;
 703
 704 out_dropped_unlock:
 705         spin_unlock_irqrestore(&port->vio.lock, flags);
 706
 707 out_dropped:
 708         dev_kfree_skb(skb);
 709         dev->stats.tx_dropped++;
 710         return NETDEV_TX_OK;
 711 }
 712
 713 static void vnet_tx_timeout(struct net_device *dev)
 714 {
 715         /* XXX Implement me XXX */
 716 }
 717
 718 static int vnet_open(struct net_device *dev)
 719 {
 720         netif_carrier_on(dev);
 721         netif_start_queue(dev);
 722
 723         return 0;
 724 }
 725
 726 static int vnet_close(struct net_device *dev)
 727 {
 728         netif_stop_queue(dev);
 729         netif_carrier_off(dev);
 730
 731         return 0;
 732 }
 733
 734 static void vnet_set_rx_mode(struct net_device *dev)
 735 {
 736         /* XXX Implement multicast support XXX */
 737 }
 738
 739 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
 740 {
 741         if (new_mtu != ETH_DATA_LEN)
 742                 return -EINVAL;
 743
 744         dev->mtu = new_mtu;
 745         return 0;
 746 }
 747
 748 static int vnet_set_mac_addr(struct net_device *dev, void *p)
 749 {
 750         return -EINVAL;
 751 }
 752
 753 static void vnet_get_drvinfo(struct net_device *dev,
 754                              struct ethtool_drvinfo *info)
 755 {
 756         strcpy(info->driver, DRV_MODULE_NAME);
 757         strcpy(info->version, DRV_MODULE_VERSION);
 758 }
 759
 760 static u32 vnet_get_msglevel(struct net_device *dev)
 761 {
 762         struct vnet *vp = netdev_priv(dev);
 763         return vp->msg_enable;
 764 }
 765
 766 static void vnet_set_msglevel(struct net_device *dev, u32 value)
 767 {
 768         struct vnet *vp = netdev_priv(dev);
 769         vp->msg_enable = value;
 770 }
 771
 772 static const struct ethtool_ops vnet_ethtool_ops = {
 773         .get_drvinfo            = vnet_get_drvinfo,
 774         .get_msglevel           = vnet_get_msglevel,
 775         .set_msglevel           = vnet_set_msglevel,
 776         .get_link               = ethtool_op_get_link,
 777         .get_perm_addr          = ethtool_op_get_perm_addr,
 778 };
 779
 780 static void vnet_port_free_tx_bufs(struct vnet_port *port)
 781 {
 782         struct vio_dring_state *dr;
 783         int i;
 784
 785         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 786         if (dr->base) {
 787                 ldc_free_exp_dring(port->vio.lp, dr->base,
 788                                    (dr->entry_size * dr->num_entries),
 789                                    dr->cookies, dr->ncookies);
 790                 dr->base = NULL;
 791                 dr->entry_size = 0;
 792                 dr->num_entries = 0;
 793                 dr->pending = 0;
 794                 dr->ncookies = 0;
 795         }
 796
 797         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 798                 void *buf = port->tx_bufs[i].buf;
 799
 800                 if (!buf)
 801                         continue;
 802
 803                 ldc_unmap(port->vio.lp,
 804                           port->tx_bufs[i].cookies,
 805                           port->tx_bufs[i].ncookies);
 806
 807                 kfree(buf);
 808                 port->tx_bufs[i].buf = NULL;
 809         }
 810 }
 811
 812 static int __devinit vnet_port_alloc_tx_bufs(struct vnet_port *port)
 813 {
 814         struct vio_dring_state *dr;
 815         unsigned long len;
 816         int i, err, ncookies;
 817         void *dring;
 818
 819         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 820                 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
 821                 int map_len = (ETH_FRAME_LEN + 7) & ~7;
 822
 823                 err = -ENOMEM;
 824                 if (!buf) {
 825                         printk(KERN_ERR "TX buffer allocation failure\n");
 826                         goto err_out;
 827                 }
 828                 err = -EFAULT;
 829                 if ((unsigned long)buf & (8UL - 1)) {
 830                         printk(KERN_ERR "TX buffer misaligned\n");
 831                         kfree(buf);
 832                         goto err_out;
 833                 }
 834
 835                 err = ldc_map_single(port->vio.lp, buf, map_len,
 836                                      port->tx_bufs[i].cookies, 2,
 837                                      (LDC_MAP_SHADOW |
 838                                       LDC_MAP_DIRECT |
 839                                       LDC_MAP_RW));
 840                 if (err < 0) {
 841                         kfree(buf);
 842                         goto err_out;
 843                 }
 844                 port->tx_bufs[i].buf = buf;
 845                 port->tx_bufs[i].ncookies = err;
 846         }
 847
 848         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 849
 850         len = (VNET_TX_RING_SIZE *
 851                (sizeof(struct vio_net_desc) +
 852                 (sizeof(struct ldc_trans_cookie) * 2)));
 853
 854         ncookies = VIO_MAX_RING_COOKIES;
 855         dring = ldc_alloc_exp_dring(port->vio.lp, len,
 856                                     dr->cookies, &ncookies,
 857                                     (LDC_MAP_SHADOW |
 858                                      LDC_MAP_DIRECT |
 859                                      LDC_MAP_RW));
 860         if (IS_ERR(dring)) {
 861                 err = PTR_ERR(dring);
 862                 goto err_out;
 863         }
 864
 865         dr->base = dring;
 866         dr->entry_size = (sizeof(struct vio_net_desc) +
 867                           (sizeof(struct ldc_trans_cookie) * 2));
 868         dr->num_entries = VNET_TX_RING_SIZE;
 869         dr->prod = dr->cons = 0;
 870         dr->pending = VNET_TX_RING_SIZE;
 871         dr->ncookies = ncookies;
 872
 873         return 0;
 874
 875 err_out:
 876         vnet_port_free_tx_bufs(port);
 877
 878         return err;
 879 }
 880
 881 static LIST_HEAD(vnet_list);
 882 static DEFINE_MUTEX(vnet_list_mutex);
 883
 884 static struct vnet * __devinit vnet_new(const u64 *local_mac)
 885 {
 886         struct net_device *dev;
 887         struct vnet *vp;
 888         int err, i;
 889
 890         dev = alloc_etherdev(sizeof(*vp));
 891         if (!dev) {
 892                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
 893                 return ERR_PTR(-ENOMEM);
 894         }
 895
 896         for (i = 0; i < ETH_ALEN; i++)
 897                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
 898
 899         memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
 900
 901         vp = netdev_priv(dev);
 902
 903         spin_lock_init(&vp->lock);
 904         vp->dev = dev;
 905
 906         INIT_LIST_HEAD(&vp->port_list);
 907         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
 908                 INIT_HLIST_HEAD(&vp->port_hash[i]);
 909         INIT_LIST_HEAD(&vp->list);
 910         vp->local_mac = *local_mac;
 911
 912         dev->open = vnet_open;
 913         dev->stop = vnet_close;
 914         dev->set_multicast_list = vnet_set_rx_mode;
 915         dev->set_mac_address = vnet_set_mac_addr;
 916         dev->tx_timeout = vnet_tx_timeout;
 917         dev->ethtool_ops = &vnet_ethtool_ops;
 918         dev->watchdog_timeo = VNET_TX_TIMEOUT;
 919         dev->change_mtu = vnet_change_mtu;
 920         dev->hard_start_xmit = vnet_start_xmit;
 921
 922         err = register_netdev(dev);
 923         if (err) {
 924                 printk(KERN_ERR PFX "Cannot register net device, "
 925                        "aborting.\n");
 926                 goto err_out_free_dev;
 927         }
 928
 929         printk(KERN_INFO "%s: Sun LDOM vnet ", dev->name);
 930
 931         for (i = 0; i < 6; i++)
 932                 printk("%2.2x%c", dev->dev_addr[i], i == 5 ? '\n' : ':');
 933
 934         list_add(&vp->list, &vnet_list);
 935
 936         return vp;
 937
 938 err_out_free_dev:
 939         free_netdev(dev);
 940
 941         return ERR_PTR(err);
 942 }
 943
 944 static struct vnet * __devinit vnet_find_or_create(const u64 *local_mac)
 945 {
 946         struct vnet *iter, *vp;
 947
 948         mutex_lock(&vnet_list_mutex);
 949         vp = NULL;
 950         list_for_each_entry(iter, &vnet_list, list) {
 951                 if (iter->local_mac == *local_mac) {
 952                         vp = iter;
 953                         break;
 954                 }
 955         }
 956         if (!vp)
 957                 vp = vnet_new(local_mac);
 958         mutex_unlock(&vnet_list_mutex);
 959
 960         return vp;
 961 }
 962
 963 static const char *local_mac_prop = "local-mac-address";
 964
 965 static struct vnet * __devinit vnet_find_parent(struct mdesc_handle *hp,
 966                                                 u64 port_node)
 967 {
 968         const u64 *local_mac = NULL;
 969         u64 a;
 970
 971         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
 972                 u64 target = mdesc_arc_target(hp, a);
 973                 const char *name;
 974
 975                 name = mdesc_get_property(hp, target, "name", NULL);
 976                 if (!name || strcmp(name, "network"))
 977                         continue;
 978
 979                 local_mac = mdesc_get_property(hp, target,
 980                                                local_mac_prop, NULL);
 981                 if (local_mac)
 982                         break;
 983         }
 984         if (!local_mac)
 985                 return ERR_PTR(-ENODEV);
 986
 987         return vnet_find_or_create(local_mac);
 988 }
 989
 990 static struct ldc_channel_config vnet_ldc_cfg = {
 991         .event          = vnet_event,
 992         .mtu            = 64,
 993         .mode           = LDC_MODE_UNRELIABLE,
 994 };
 995
 996 static struct vio_driver_ops vnet_vio_ops = {
 997         .send_attr              = vnet_send_attr,
 998         .handle_attr            = vnet_handle_attr,
 999         .handshake_complete     = vnet_handshake_complete,
1000 };
1001
1002 static void print_version(void)
1003 {
1004         static int version_printed;
1005
1006         if (version_printed++ == 0)
1007                 printk(KERN_INFO "%s", version);
1008 }
1009
1010 const char *remote_macaddr_prop = "remote-mac-address";
1011
1012 static int __devinit vnet_port_probe(struct vio_dev *vdev,
1013                                      const struct vio_device_id *id)
1014 {
1015         struct mdesc_handle *hp;
1016         struct vnet_port *port;
1017         unsigned long flags;
1018         struct vnet *vp;
1019         const u64 *rmac;
1020         int len, i, err, switch_port;
1021
1022         print_version();
1023
1024         hp = mdesc_grab();
1025
1026         vp = vnet_find_parent(hp, vdev->mp);
1027         if (IS_ERR(vp)) {
1028                 printk(KERN_ERR PFX "Cannot find port parent vnet.\n");
1029                 err = PTR_ERR(vp);
1030                 goto err_out_put_mdesc;
1031         }
1032
1033         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1034         err = -ENODEV;
1035         if (!rmac) {
1036                 printk(KERN_ERR PFX "Port lacks %s property.\n",
1037                        remote_macaddr_prop);
1038                 goto err_out_put_mdesc;
1039         }
1040
1041         port = kzalloc(sizeof(*port), GFP_KERNEL);
1042         err = -ENOMEM;
1043         if (!port) {
1044                 printk(KERN_ERR PFX "Cannot allocate vnet_port.\n");
1045                 goto err_out_put_mdesc;
1046         }
1047
1048         for (i = 0; i < ETH_ALEN; i++)
1049                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1050
1051         port->vp = vp;
1052
1053         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1054                               vnet_versions, ARRAY_SIZE(vnet_versions),
1055                               &vnet_vio_ops, vp->dev->name);
1056         if (err)
1057                 goto err_out_free_port;
1058
1059         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1060         if (err)
1061                 goto err_out_free_port;
1062
1063         err = vnet_port_alloc_tx_bufs(port);
1064         if (err)
1065                 goto err_out_free_ldc;
1066
1067         INIT_HLIST_NODE(&port->hash);
1068         INIT_LIST_HEAD(&port->list);
1069
1070         switch_port = 0;
1071         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1072                 switch_port = 1;
1073
1074         spin_lock_irqsave(&vp->lock, flags);
1075         if (switch_port)
1076                 list_add(&port->list, &vp->port_list);
1077         else
1078                 list_add_tail(&port->list, &vp->port_list);
1079         hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1080         spin_unlock_irqrestore(&vp->lock, flags);
1081
1082         dev_set_drvdata(&vdev->dev, port);
1083
1084         printk(KERN_INFO "%s: PORT ( remote-mac ", vp->dev->name);
1085         for (i = 0; i < 6; i++)
1086                 printk("%2.2x%c", port->raddr[i], i == 5 ? ' ' : ':');
1087         if (switch_port)
1088                 printk("switch-port ");
1089         printk(")\n");
1090
1091         vio_port_up(&port->vio);
1092
1093         mdesc_release(hp);
1094
1095         return 0;
1096
1097 err_out_free_ldc:
1098         vio_ldc_free(&port->vio);
1099
1100 err_out_free_port:
1101         kfree(port);
1102
1103 err_out_put_mdesc:
1104         mdesc_release(hp);
1105         return err;
1106 }
1107
1108 static int vnet_port_remove(struct vio_dev *vdev)
1109 {
1110         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1111
1112         if (port) {
1113                 struct vnet *vp = port->vp;
1114                 unsigned long flags;
1115
1116                 del_timer_sync(&port->vio.timer);
1117
1118                 spin_lock_irqsave(&vp->lock, flags);
1119                 list_del(&port->list);
1120                 hlist_del(&port->hash);
1121                 spin_unlock_irqrestore(&vp->lock, flags);
1122
1123                 vnet_port_free_tx_bufs(port);
1124                 vio_ldc_free(&port->vio);
1125
1126                 dev_set_drvdata(&vdev->dev, NULL);
1127
1128                 kfree(port);
1129         }
1130         return 0;
1131 }
1132
1133 static struct vio_device_id vnet_port_match[] = {
1134         {
1135                 .type = "vnet-port",
1136         },
1137         {},
1138 };
1139 MODULE_DEVICE_TABLE(vio, vnet_match);
1140
1141 static struct vio_driver vnet_port_driver = {
1142         .id_table       = vnet_port_match,
1143         .probe          = vnet_port_probe,
1144         .remove         = vnet_port_remove,
1145         .driver         = {
1146                 .name   = "vnet_port",
1147                 .owner  = THIS_MODULE,
1148         }
1149 };
1150
1151 static int __init vnet_init(void)
1152 {
1153         return vio_register_driver(&vnet_port_driver);
1154 }
1155
1156 static void __exit vnet_exit(void)
1157 {
1158         vio_unregister_driver(&vnet_port_driver);
1159 }
1160
1161 module_init(vnet_init);
1162 module_exit(vnet_exit);