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