drivers/net/ethernet/sun/sunvnet.c

   1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
   2  *
   3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/module.h>
   9 #include <linux/kernel.h>
  10 #include <linux/types.h>
  11 #include <linux/slab.h>
  12 #include <linux/delay.h>
  13 #include <linux/init.h>
  14 #include <linux/netdevice.h>
  15 #include <linux/ethtool.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/mutex.h>
  18
  19 #include <asm/vio.h>
  20 #include <asm/ldc.h>
  21
  22 #include "sunvnet.h"
  23
  24 #define DRV_MODULE_NAME         "sunvnet"
  25 #define DRV_MODULE_VERSION      "1.0"
  26 #define DRV_MODULE_RELDATE      "June 25, 2007"
  27
  28 static char version[] =
  29         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
  30 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
  31 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
  32 MODULE_LICENSE("GPL");
  33 MODULE_VERSION(DRV_MODULE_VERSION);
  34
  35 /* Heuristic for the number of times to exponentially backoff and
  36  * retry sending an LDC trigger when EAGAIN is encountered
  37  */
  38 #define VNET_MAX_RETRIES        10
  39
  40 /* Ordered from largest major to lowest */
  41 static struct vio_version vnet_versions[] = {
  42         { .major = 1, .minor = 0 },
  43 };
  44
  45 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
  46 {
  47         return vio_dring_avail(dr, VNET_TX_RING_SIZE);
  48 }
  49
  50 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
  51 {
  52         struct vio_msg_tag *pkt = arg;
  53
  54         pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
  55                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
  56         pr_err("Resetting connection\n");
  57
  58         ldc_disconnect(port->vio.lp);
  59
  60         return -ECONNRESET;
  61 }
  62
  63 static int vnet_send_attr(struct vio_driver_state *vio)
  64 {
  65         struct vnet_port *port = to_vnet_port(vio);
  66         struct net_device *dev = port->vp->dev;
  67         struct vio_net_attr_info pkt;
  68         int i;
  69
  70         memset(&pkt, 0, sizeof(pkt));
  71         pkt.tag.type = VIO_TYPE_CTRL;
  72         pkt.tag.stype = VIO_SUBTYPE_INFO;
  73         pkt.tag.stype_env = VIO_ATTR_INFO;
  74         pkt.tag.sid = vio_send_sid(vio);
  75         pkt.xfer_mode = VIO_DRING_MODE;
  76         pkt.addr_type = VNET_ADDR_ETHERMAC;
  77         pkt.ack_freq = 0;
  78         for (i = 0; i < 6; i++)
  79                 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
  80         pkt.mtu = ETH_FRAME_LEN;
  81
  82         viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
  83                "ackfreq[%u] mtu[%llu]\n",
  84                pkt.xfer_mode, pkt.addr_type,
  85                (unsigned long long) pkt.addr,
  86                pkt.ack_freq,
  87                (unsigned long long) pkt.mtu);
  88
  89         return vio_ldc_send(vio, &pkt, sizeof(pkt));
  90 }
  91
  92 static int handle_attr_info(struct vio_driver_state *vio,
  93                             struct vio_net_attr_info *pkt)
  94 {
  95         viodbg(HS, "GOT NET ATTR INFO xmode[0x%x] atype[0x%x] addr[%llx] "
  96                "ackfreq[%u] mtu[%llu]\n",
  97                pkt->xfer_mode, pkt->addr_type,
  98                (unsigned long long) pkt->addr,
  99                pkt->ack_freq,
 100                (unsigned long long) pkt->mtu);
 101
 102         pkt->tag.sid = vio_send_sid(vio);
 103
 104         if (pkt->xfer_mode != VIO_DRING_MODE ||
 105             pkt->addr_type != VNET_ADDR_ETHERMAC ||
 106             pkt->mtu != ETH_FRAME_LEN) {
 107                 viodbg(HS, "SEND NET ATTR NACK\n");
 108
 109                 pkt->tag.stype = VIO_SUBTYPE_NACK;
 110
 111                 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
 112
 113                 return -ECONNRESET;
 114         } else {
 115                 viodbg(HS, "SEND NET ATTR ACK\n");
 116
 117                 pkt->tag.stype = VIO_SUBTYPE_ACK;
 118
 119                 return vio_ldc_send(vio, pkt, sizeof(*pkt));
 120         }
 121
 122 }
 123
 124 static int handle_attr_ack(struct vio_driver_state *vio,
 125                            struct vio_net_attr_info *pkt)
 126 {
 127         viodbg(HS, "GOT NET ATTR ACK\n");
 128
 129         return 0;
 130 }
 131
 132 static int handle_attr_nack(struct vio_driver_state *vio,
 133                             struct vio_net_attr_info *pkt)
 134 {
 135         viodbg(HS, "GOT NET ATTR NACK\n");
 136
 137         return -ECONNRESET;
 138 }
 139
 140 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
 141 {
 142         struct vio_net_attr_info *pkt = arg;
 143
 144         switch (pkt->tag.stype) {
 145         case VIO_SUBTYPE_INFO:
 146                 return handle_attr_info(vio, pkt);
 147
 148         case VIO_SUBTYPE_ACK:
 149                 return handle_attr_ack(vio, pkt);
 150
 151         case VIO_SUBTYPE_NACK:
 152                 return handle_attr_nack(vio, pkt);
 153
 154         default:
 155                 return -ECONNRESET;
 156         }
 157 }
 158
 159 static void vnet_handshake_complete(struct vio_driver_state *vio)
 160 {
 161         struct vio_dring_state *dr;
 162
 163         dr = &vio->drings[VIO_DRIVER_RX_RING];
 164         dr->snd_nxt = dr->rcv_nxt = 1;
 165
 166         dr = &vio->drings[VIO_DRIVER_TX_RING];
 167         dr->snd_nxt = dr->rcv_nxt = 1;
 168 }
 169
 170 /* The hypervisor interface that implements copying to/from imported
 171  * memory from another domain requires that copies are done to 8-byte
 172  * aligned buffers, and that the lengths of such copies are also 8-byte
 173  * multiples.
 174  *
 175  * So we align skb->data to an 8-byte multiple and pad-out the data
 176  * area so we can round the copy length up to the next multiple of
 177  * 8 for the copy.
 178  *
 179  * The transmitter puts the actual start of the packet 6 bytes into
 180  * the buffer it sends over, so that the IP headers after the ethernet
 181  * header are aligned properly.  These 6 bytes are not in the descriptor
 182  * length, they are simply implied.  This offset is represented using
 183  * the VNET_PACKET_SKIP macro.
 184  */
 185 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
 186                                            unsigned int len)
 187 {
 188         struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
 189         unsigned long addr, off;
 190
 191         if (unlikely(!skb))
 192                 return NULL;
 193
 194         addr = (unsigned long) skb->data;
 195         off = ((addr + 7UL) & ~7UL) - addr;
 196         if (off)
 197                 skb_reserve(skb, off);
 198
 199         return skb;
 200 }
 201
 202 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
 203                        struct ldc_trans_cookie *cookies, int ncookies)
 204 {
 205         struct net_device *dev = port->vp->dev;
 206         unsigned int copy_len;
 207         struct sk_buff *skb;
 208         int err;
 209
 210         err = -EMSGSIZE;
 211         if (unlikely(len < ETH_ZLEN || len > ETH_FRAME_LEN)) {
 212                 dev->stats.rx_length_errors++;
 213                 goto out_dropped;
 214         }
 215
 216         skb = alloc_and_align_skb(dev, len);
 217         err = -ENOMEM;
 218         if (unlikely(!skb)) {
 219                 dev->stats.rx_missed_errors++;
 220                 goto out_dropped;
 221         }
 222
 223         copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
 224         skb_put(skb, copy_len);
 225         err = ldc_copy(port->vio.lp, LDC_COPY_IN,
 226                        skb->data, copy_len, 0,
 227                        cookies, ncookies);
 228         if (unlikely(err < 0)) {
 229                 dev->stats.rx_frame_errors++;
 230                 goto out_free_skb;
 231         }
 232
 233         skb_pull(skb, VNET_PACKET_SKIP);
 234         skb_trim(skb, len);
 235         skb->protocol = eth_type_trans(skb, dev);
 236
 237         dev->stats.rx_packets++;
 238         dev->stats.rx_bytes += len;
 239
 240         netif_rx(skb);
 241
 242         return 0;
 243
 244 out_free_skb:
 245         kfree_skb(skb);
 246
 247 out_dropped:
 248         dev->stats.rx_dropped++;
 249         return err;
 250 }
 251
 252 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
 253                          u32 start, u32 end, u8 vio_dring_state)
 254 {
 255         struct vio_dring_data hdr = {
 256                 .tag = {
 257                         .type           = VIO_TYPE_DATA,
 258                         .stype          = VIO_SUBTYPE_ACK,
 259                         .stype_env      = VIO_DRING_DATA,
 260                         .sid            = vio_send_sid(&port->vio),
 261                 },
 262                 .dring_ident            = dr->ident,
 263                 .start_idx              = start,
 264                 .end_idx                = end,
 265                 .state                  = vio_dring_state,
 266         };
 267         int err, delay;
 268         int retries = 0;
 269
 270         hdr.seq = dr->snd_nxt;
 271         delay = 1;
 272         do {
 273                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 274                 if (err > 0) {
 275                         dr->snd_nxt++;
 276                         break;
 277                 }
 278                 udelay(delay);
 279                 if ((delay <<= 1) > 128)
 280                         delay = 128;
 281                 if (retries++ > VNET_MAX_RETRIES) {
 282                         pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
 283                                 port->raddr[0], port->raddr[1],
 284                                 port->raddr[2], port->raddr[3],
 285                                 port->raddr[4], port->raddr[5]);
 286                         err = -ECONNRESET;
 287                 }
 288         } while (err == -EAGAIN);
 289
 290         return err;
 291 }
 292
 293 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
 294 {
 295         if (++idx == dr->num_entries)
 296                 idx = 0;
 297         return idx;
 298 }
 299
 300 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
 301 {
 302         if (idx == 0)
 303                 idx = dr->num_entries - 1;
 304         else
 305                 idx--;
 306
 307         return idx;
 308 }
 309
 310 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
 311                                         struct vio_dring_state *dr,
 312                                         u32 index)
 313 {
 314         struct vio_net_desc *desc = port->vio.desc_buf;
 315         int err;
 316
 317         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
 318                                   (index * dr->entry_size),
 319                                   dr->cookies, dr->ncookies);
 320         if (err < 0)
 321                 return ERR_PTR(err);
 322
 323         return desc;
 324 }
 325
 326 static int put_rx_desc(struct vnet_port *port,
 327                        struct vio_dring_state *dr,
 328                        struct vio_net_desc *desc,
 329                        u32 index)
 330 {
 331         int err;
 332
 333         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
 334                                   (index * dr->entry_size),
 335                                   dr->cookies, dr->ncookies);
 336         if (err < 0)
 337                 return err;
 338
 339         return 0;
 340 }
 341
 342 static int vnet_walk_rx_one(struct vnet_port *port,
 343                             struct vio_dring_state *dr,
 344                             u32 index, int *needs_ack)
 345 {
 346         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
 347         struct vio_driver_state *vio = &port->vio;
 348         int err;
 349
 350         if (IS_ERR(desc))
 351                 return PTR_ERR(desc);
 352
 353         if (desc->hdr.state != VIO_DESC_READY)
 354                 return 1;
 355
 356         rmb();
 357
 358         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 359                desc->hdr.state, desc->hdr.ack,
 360                desc->size, desc->ncookies,
 361                desc->cookies[0].cookie_addr,
 362                desc->cookies[0].cookie_size);
 363
 364         err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
 365         if (err == -ECONNRESET)
 366                 return err;
 367         desc->hdr.state = VIO_DESC_DONE;
 368         err = put_rx_desc(port, dr, desc, index);
 369         if (err < 0)
 370                 return err;
 371         *needs_ack = desc->hdr.ack;
 372         return 0;
 373 }
 374
 375 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 376                         u32 start, u32 end)
 377 {
 378         struct vio_driver_state *vio = &port->vio;
 379         int ack_start = -1, ack_end = -1;
 380
 381         end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
 382
 383         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
 384
 385         while (start != end) {
 386                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
 387                 if (err == -ECONNRESET)
 388                         return err;
 389                 if (err != 0)
 390                         break;
 391                 if (ack_start == -1)
 392                         ack_start = start;
 393                 ack_end = start;
 394                 start = next_idx(start, dr);
 395                 if (ack && start != end) {
 396                         err = vnet_send_ack(port, dr, ack_start, ack_end,
 397                                             VIO_DRING_ACTIVE);
 398                         if (err == -ECONNRESET)
 399                                 return err;
 400                         ack_start = -1;
 401                 }
 402         }
 403         if (unlikely(ack_start == -1))
 404                 ack_start = ack_end = prev_idx(start, dr);
 405         return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
 406 }
 407
 408 static int vnet_rx(struct vnet_port *port, void *msgbuf)
 409 {
 410         struct vio_dring_data *pkt = msgbuf;
 411         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
 412         struct vio_driver_state *vio = &port->vio;
 413
 414         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
 415                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
 416
 417         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 418                 return 0;
 419         if (unlikely(pkt->seq != dr->rcv_nxt)) {
 420                 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
 421                        pkt->seq, dr->rcv_nxt);
 422                 return 0;
 423         }
 424
 425         dr->rcv_nxt++;
 426
 427         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
 428
 429         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
 430 }
 431
 432 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
 433 {
 434         u32 idx = dr->cons;
 435         int found = 0;
 436
 437         while (idx != dr->prod) {
 438                 if (idx == end) {
 439                         found = 1;
 440                         break;
 441                 }
 442                 idx = next_idx(idx, dr);
 443         }
 444         return found;
 445 }
 446
 447 static int vnet_ack(struct vnet_port *port, void *msgbuf)
 448 {
 449         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 450         struct vio_dring_data *pkt = msgbuf;
 451         struct net_device *dev;
 452         struct vnet *vp;
 453         u32 end;
 454
 455         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 456                 return 0;
 457
 458         end = pkt->end_idx;
 459         if (unlikely(!idx_is_pending(dr, end)))
 460                 return 0;
 461
 462         dr->cons = next_idx(end, dr);
 463
 464         vp = port->vp;
 465         dev = vp->dev;
 466         if (unlikely(netif_queue_stopped(dev) &&
 467                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
 468                 return 1;
 469
 470         return 0;
 471 }
 472
 473 static int vnet_nack(struct vnet_port *port, void *msgbuf)
 474 {
 475         /* XXX just reset or similar XXX */
 476         return 0;
 477 }
 478
 479 static int handle_mcast(struct vnet_port *port, void *msgbuf)
 480 {
 481         struct vio_net_mcast_info *pkt = msgbuf;
 482
 483         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
 484                 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
 485                        port->vp->dev->name,
 486                        pkt->tag.type,
 487                        pkt->tag.stype,
 488                        pkt->tag.stype_env,
 489                        pkt->tag.sid);
 490
 491         return 0;
 492 }
 493
 494 static void maybe_tx_wakeup(unsigned long param)
 495 {
 496         struct vnet *vp = (struct vnet *)param;
 497         struct net_device *dev = vp->dev;
 498
 499         netif_tx_lock(dev);
 500         if (likely(netif_queue_stopped(dev))) {
 501                 struct vnet_port *port;
 502                 int wake = 1;
 503
 504                 list_for_each_entry(port, &vp->port_list, list) {
 505                         struct vio_dring_state *dr;
 506
 507                         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 508                         if (vnet_tx_dring_avail(dr) <
 509                             VNET_TX_WAKEUP_THRESH(dr)) {
 510                                 wake = 0;
 511                                 break;
 512                         }
 513                 }
 514                 if (wake)
 515                         netif_wake_queue(dev);
 516         }
 517         netif_tx_unlock(dev);
 518 }
 519
 520 static void vnet_event(void *arg, int event)
 521 {
 522         struct vnet_port *port = arg;
 523         struct vio_driver_state *vio = &port->vio;
 524         unsigned long flags;
 525         int tx_wakeup, err;
 526
 527         spin_lock_irqsave(&vio->lock, flags);
 528
 529         if (unlikely(event == LDC_EVENT_RESET ||
 530                      event == LDC_EVENT_UP)) {
 531                 vio_link_state_change(vio, event);
 532                 spin_unlock_irqrestore(&vio->lock, flags);
 533
 534                 if (event == LDC_EVENT_RESET)
 535                         vio_port_up(vio);
 536                 return;
 537         }
 538
 539         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 540                 pr_warning("Unexpected LDC event %d\n", event);
 541                 spin_unlock_irqrestore(&vio->lock, flags);
 542                 return;
 543         }
 544
 545         tx_wakeup = err = 0;
 546         while (1) {
 547                 union {
 548                         struct vio_msg_tag tag;
 549                         u64 raw[8];
 550                 } msgbuf;
 551
 552                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 553                 if (unlikely(err < 0)) {
 554                         if (err == -ECONNRESET)
 555                                 vio_conn_reset(vio);
 556                         break;
 557                 }
 558                 if (err == 0)
 559                         break;
 560                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 561                        msgbuf.tag.type,
 562                        msgbuf.tag.stype,
 563                        msgbuf.tag.stype_env,
 564                        msgbuf.tag.sid);
 565                 err = vio_validate_sid(vio, &msgbuf.tag);
 566                 if (err < 0)
 567                         break;
 568
 569                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 570                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 571                                 err = vnet_rx(port, &msgbuf);
 572                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 573                                 err = vnet_ack(port, &msgbuf);
 574                                 if (err > 0)
 575                                         tx_wakeup |= err;
 576                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 577                                 err = vnet_nack(port, &msgbuf);
 578                         }
 579                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 580                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 581                                 err = handle_mcast(port, &msgbuf);
 582                         else
 583                                 err = vio_control_pkt_engine(vio, &msgbuf);
 584                         if (err)
 585                                 break;
 586                 } else {
 587                         err = vnet_handle_unknown(port, &msgbuf);
 588                 }
 589                 if (err == -ECONNRESET)
 590                         break;
 591         }
 592         spin_unlock(&vio->lock);
 593         /* Kick off a tasklet to wake the queue.  We cannot call
 594          * maybe_tx_wakeup directly here because we could deadlock on
 595          * netif_tx_lock() with dev_watchdog()
 596          */
 597         if (unlikely(tx_wakeup && err != -ECONNRESET))
 598                 tasklet_schedule(&port->vp->vnet_tx_wakeup);
 599
 600         local_irq_restore(flags);
 601 }
 602
 603 static int __vnet_tx_trigger(struct vnet_port *port)
 604 {
 605         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 606         struct vio_dring_data hdr = {
 607                 .tag = {
 608                         .type           = VIO_TYPE_DATA,
 609                         .stype          = VIO_SUBTYPE_INFO,
 610                         .stype_env      = VIO_DRING_DATA,
 611                         .sid            = vio_send_sid(&port->vio),
 612                 },
 613                 .dring_ident            = dr->ident,
 614                 .start_idx              = dr->prod,
 615                 .end_idx                = (u32) -1,
 616         };
 617         int err, delay;
 618         int retries = 0;
 619
 620         hdr.seq = dr->snd_nxt;
 621         delay = 1;
 622         do {
 623                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 624                 if (err > 0) {
 625                         dr->snd_nxt++;
 626                         break;
 627                 }
 628                 udelay(delay);
 629                 if ((delay <<= 1) > 128)
 630                         delay = 128;
 631                 if (retries++ > VNET_MAX_RETRIES)
 632                         break;
 633         } while (err == -EAGAIN);
 634
 635         return err;
 636 }
 637
 638 static inline bool port_is_up(struct vnet_port *vnet)
 639 {
 640         struct vio_driver_state *vio = &vnet->vio;
 641
 642         return !!(vio->hs_state & VIO_HS_COMPLETE);
 643 }
 644
 645 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 646 {
 647         unsigned int hash = vnet_hashfn(skb->data);
 648         struct hlist_head *hp = &vp->port_hash[hash];
 649         struct vnet_port *port;
 650
 651         hlist_for_each_entry(port, hp, hash) {
 652                 if (!port_is_up(port))
 653                         continue;
 654                 if (ether_addr_equal(port->raddr, skb->data))
 655                         return port;
 656         }
 657         list_for_each_entry(port, &vp->port_list, list) {
 658                 if (!port->switch_port)
 659                         continue;
 660                 if (!port_is_up(port))
 661                         continue;
 662                 return port;
 663         }
 664         return NULL;
 665 }
 666
 667 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
 668 {
 669         struct vnet_port *ret;
 670         unsigned long flags;
 671
 672         spin_lock_irqsave(&vp->lock, flags);
 673         ret = __tx_port_find(vp, skb);
 674         spin_unlock_irqrestore(&vp->lock, flags);
 675
 676         return ret;
 677 }
 678
 679 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 680 {
 681         struct vnet *vp = netdev_priv(dev);
 682         struct vnet_port *port = tx_port_find(vp, skb);
 683         struct vio_dring_state *dr;
 684         struct vio_net_desc *d;
 685         unsigned long flags;
 686         unsigned int len;
 687         void *tx_buf;
 688         int i, err;
 689
 690         if (unlikely(!port))
 691                 goto out_dropped;
 692
 693         spin_lock_irqsave(&port->vio.lock, flags);
 694
 695         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 696         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 697                 if (!netif_queue_stopped(dev)) {
 698                         netif_stop_queue(dev);
 699
 700                         /* This is a hard error, log it. */
 701                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
 702                         dev->stats.tx_errors++;
 703                 }
 704                 spin_unlock_irqrestore(&port->vio.lock, flags);
 705                 return NETDEV_TX_BUSY;
 706         }
 707
 708         d = vio_dring_cur(dr);
 709
 710         tx_buf = port->tx_bufs[dr->prod].buf;
 711         skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
 712
 713         len = skb->len;
 714         if (len < ETH_ZLEN) {
 715                 len = ETH_ZLEN;
 716                 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
 717         }
 718
 719         /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
 720          * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
 721          * the protocol itself does not require it as long as the peer
 722          * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
 723          *
 724          * An ACK for every packet in the ring is expensive as the
 725          * sending of LDC messages is slow and affects performance.
 726          */
 727         d->hdr.ack = VIO_ACK_DISABLE;
 728         d->size = len;
 729         d->ncookies = port->tx_bufs[dr->prod].ncookies;
 730         for (i = 0; i < d->ncookies; i++)
 731                 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
 732
 733         /* This has to be a non-SMP write barrier because we are writing
 734          * to memory which is shared with the peer LDOM.
 735          */
 736         wmb();
 737
 738         d->hdr.state = VIO_DESC_READY;
 739
 740         err = __vnet_tx_trigger(port);
 741         if (unlikely(err < 0)) {
 742                 netdev_info(dev, "TX trigger error %d\n", err);
 743                 d->hdr.state = VIO_DESC_FREE;
 744                 dev->stats.tx_carrier_errors++;
 745                 goto out_dropped_unlock;
 746         }
 747
 748         dev->stats.tx_packets++;
 749         dev->stats.tx_bytes += skb->len;
 750
 751         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
 752         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
 753                 netif_stop_queue(dev);
 754                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
 755                         netif_wake_queue(dev);
 756         }
 757
 758         spin_unlock_irqrestore(&port->vio.lock, flags);
 759
 760         dev_kfree_skb(skb);
 761
 762         return NETDEV_TX_OK;
 763
 764 out_dropped_unlock:
 765         spin_unlock_irqrestore(&port->vio.lock, flags);
 766
 767 out_dropped:
 768         dev_kfree_skb(skb);
 769         dev->stats.tx_dropped++;
 770         return NETDEV_TX_OK;
 771 }
 772
 773 static void vnet_tx_timeout(struct net_device *dev)
 774 {
 775         /* XXX Implement me XXX */
 776 }
 777
 778 static int vnet_open(struct net_device *dev)
 779 {
 780         netif_carrier_on(dev);
 781         netif_start_queue(dev);
 782
 783         return 0;
 784 }
 785
 786 static int vnet_close(struct net_device *dev)
 787 {
 788         netif_stop_queue(dev);
 789         netif_carrier_off(dev);
 790
 791         return 0;
 792 }
 793
 794 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
 795 {
 796         struct vnet_mcast_entry *m;
 797
 798         for (m = vp->mcast_list; m; m = m->next) {
 799                 if (ether_addr_equal(m->addr, addr))
 800                         return m;
 801         }
 802         return NULL;
 803 }
 804
 805 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
 806 {
 807         struct netdev_hw_addr *ha;
 808
 809         netdev_for_each_mc_addr(ha, dev) {
 810                 struct vnet_mcast_entry *m;
 811
 812                 m = __vnet_mc_find(vp, ha->addr);
 813                 if (m) {
 814                         m->hit = 1;
 815                         continue;
 816                 }
 817
 818                 if (!m) {
 819                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
 820                         if (!m)
 821                                 continue;
 822                         memcpy(m->addr, ha->addr, ETH_ALEN);
 823                         m->hit = 1;
 824
 825                         m->next = vp->mcast_list;
 826                         vp->mcast_list = m;
 827                 }
 828         }
 829 }
 830
 831 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
 832 {
 833         struct vio_net_mcast_info info;
 834         struct vnet_mcast_entry *m, **pp;
 835         int n_addrs;
 836
 837         memset(&info, 0, sizeof(info));
 838
 839         info.tag.type = VIO_TYPE_CTRL;
 840         info.tag.stype = VIO_SUBTYPE_INFO;
 841         info.tag.stype_env = VNET_MCAST_INFO;
 842         info.tag.sid = vio_send_sid(&port->vio);
 843         info.set = 1;
 844
 845         n_addrs = 0;
 846         for (m = vp->mcast_list; m; m = m->next) {
 847                 if (m->sent)
 848                         continue;
 849                 m->sent = 1;
 850                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 851                        m->addr, ETH_ALEN);
 852                 if (++n_addrs == VNET_NUM_MCAST) {
 853                         info.count = n_addrs;
 854
 855                         (void) vio_ldc_send(&port->vio, &info,
 856                                             sizeof(info));
 857                         n_addrs = 0;
 858                 }
 859         }
 860         if (n_addrs) {
 861                 info.count = n_addrs;
 862                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 863         }
 864
 865         info.set = 0;
 866
 867         n_addrs = 0;
 868         pp = &vp->mcast_list;
 869         while ((m = *pp) != NULL) {
 870                 if (m->hit) {
 871                         m->hit = 0;
 872                         pp = &m->next;
 873                         continue;
 874                 }
 875
 876                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 877                        m->addr, ETH_ALEN);
 878                 if (++n_addrs == VNET_NUM_MCAST) {
 879                         info.count = n_addrs;
 880                         (void) vio_ldc_send(&port->vio, &info,
 881                                             sizeof(info));
 882                         n_addrs = 0;
 883                 }
 884
 885                 *pp = m->next;
 886                 kfree(m);
 887         }
 888         if (n_addrs) {
 889                 info.count = n_addrs;
 890                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 891         }
 892 }
 893
 894 static void vnet_set_rx_mode(struct net_device *dev)
 895 {
 896         struct vnet *vp = netdev_priv(dev);
 897         struct vnet_port *port;
 898         unsigned long flags;
 899
 900         spin_lock_irqsave(&vp->lock, flags);
 901         if (!list_empty(&vp->port_list)) {
 902                 port = list_entry(vp->port_list.next, struct vnet_port, list);
 903
 904                 if (port->switch_port) {
 905                         __update_mc_list(vp, dev);
 906                         __send_mc_list(vp, port);
 907                 }
 908         }
 909         spin_unlock_irqrestore(&vp->lock, flags);
 910 }
 911
 912 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
 913 {
 914         if (new_mtu != ETH_DATA_LEN)
 915                 return -EINVAL;
 916
 917         dev->mtu = new_mtu;
 918         return 0;
 919 }
 920
 921 static int vnet_set_mac_addr(struct net_device *dev, void *p)
 922 {
 923         return -EINVAL;
 924 }
 925
 926 static void vnet_get_drvinfo(struct net_device *dev,
 927                              struct ethtool_drvinfo *info)
 928 {
 929         strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
 930         strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
 931 }
 932
 933 static u32 vnet_get_msglevel(struct net_device *dev)
 934 {
 935         struct vnet *vp = netdev_priv(dev);
 936         return vp->msg_enable;
 937 }
 938
 939 static void vnet_set_msglevel(struct net_device *dev, u32 value)
 940 {
 941         struct vnet *vp = netdev_priv(dev);
 942         vp->msg_enable = value;
 943 }
 944
 945 static const struct ethtool_ops vnet_ethtool_ops = {
 946         .get_drvinfo            = vnet_get_drvinfo,
 947         .get_msglevel           = vnet_get_msglevel,
 948         .set_msglevel           = vnet_set_msglevel,
 949         .get_link               = ethtool_op_get_link,
 950 };
 951
 952 static void vnet_port_free_tx_bufs(struct vnet_port *port)
 953 {
 954         struct vio_dring_state *dr;
 955         int i;
 956
 957         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 958         if (dr->base) {
 959                 ldc_free_exp_dring(port->vio.lp, dr->base,
 960                                    (dr->entry_size * dr->num_entries),
 961                                    dr->cookies, dr->ncookies);
 962                 dr->base = NULL;
 963                 dr->entry_size = 0;
 964                 dr->num_entries = 0;
 965                 dr->pending = 0;
 966                 dr->ncookies = 0;
 967         }
 968
 969         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 970                 void *buf = port->tx_bufs[i].buf;
 971
 972                 if (!buf)
 973                         continue;
 974
 975                 ldc_unmap(port->vio.lp,
 976                           port->tx_bufs[i].cookies,
 977                           port->tx_bufs[i].ncookies);
 978
 979                 kfree(buf);
 980                 port->tx_bufs[i].buf = NULL;
 981         }
 982 }
 983
 984 static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
 985 {
 986         struct vio_dring_state *dr;
 987         unsigned long len;
 988         int i, err, ncookies;
 989         void *dring;
 990
 991         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 992                 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
 993                 int map_len = (ETH_FRAME_LEN + 7) & ~7;
 994
 995                 err = -ENOMEM;
 996                 if (!buf)
 997                         goto err_out;
 998
 999                 err = -EFAULT;
1000                 if ((unsigned long)buf & (8UL - 1)) {
1001                         pr_err("TX buffer misaligned\n");
1002                         kfree(buf);
1003                         goto err_out;
1004                 }
1005
1006                 err = ldc_map_single(port->vio.lp, buf, map_len,
1007                                      port->tx_bufs[i].cookies, 2,
1008                                      (LDC_MAP_SHADOW |
1009                                       LDC_MAP_DIRECT |
1010                                       LDC_MAP_RW));
1011                 if (err < 0) {
1012                         kfree(buf);
1013                         goto err_out;
1014                 }
1015                 port->tx_bufs[i].buf = buf;
1016                 port->tx_bufs[i].ncookies = err;
1017         }
1018
1019         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1020
1021         len = (VNET_TX_RING_SIZE *
1022                (sizeof(struct vio_net_desc) +
1023                 (sizeof(struct ldc_trans_cookie) * 2)));
1024
1025         ncookies = VIO_MAX_RING_COOKIES;
1026         dring = ldc_alloc_exp_dring(port->vio.lp, len,
1027                                     dr->cookies, &ncookies,
1028                                     (LDC_MAP_SHADOW |
1029                                      LDC_MAP_DIRECT |
1030                                      LDC_MAP_RW));
1031         if (IS_ERR(dring)) {
1032                 err = PTR_ERR(dring);
1033                 goto err_out;
1034         }
1035
1036         dr->base = dring;
1037         dr->entry_size = (sizeof(struct vio_net_desc) +
1038                           (sizeof(struct ldc_trans_cookie) * 2));
1039         dr->num_entries = VNET_TX_RING_SIZE;
1040         dr->prod = dr->cons = 0;
1041         dr->pending = VNET_TX_RING_SIZE;
1042         dr->ncookies = ncookies;
1043
1044         return 0;
1045
1046 err_out:
1047         vnet_port_free_tx_bufs(port);
1048
1049         return err;
1050 }
1051
1052 static LIST_HEAD(vnet_list);
1053 static DEFINE_MUTEX(vnet_list_mutex);
1054
1055 static const struct net_device_ops vnet_ops = {
1056         .ndo_open               = vnet_open,
1057         .ndo_stop               = vnet_close,
1058         .ndo_set_rx_mode        = vnet_set_rx_mode,
1059         .ndo_set_mac_address    = vnet_set_mac_addr,
1060         .ndo_validate_addr      = eth_validate_addr,
1061         .ndo_tx_timeout         = vnet_tx_timeout,
1062         .ndo_change_mtu         = vnet_change_mtu,
1063         .ndo_start_xmit         = vnet_start_xmit,
1064 };
1065
1066 static struct vnet *vnet_new(const u64 *local_mac)
1067 {
1068         struct net_device *dev;
1069         struct vnet *vp;
1070         int err, i;
1071
1072         dev = alloc_etherdev(sizeof(*vp));
1073         if (!dev)
1074                 return ERR_PTR(-ENOMEM);
1075
1076         for (i = 0; i < ETH_ALEN; i++)
1077                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1078
1079         vp = netdev_priv(dev);
1080
1081         spin_lock_init(&vp->lock);
1082         tasklet_init(&vp->vnet_tx_wakeup, maybe_tx_wakeup, (unsigned long)vp);
1083         vp->dev = dev;
1084
1085         INIT_LIST_HEAD(&vp->port_list);
1086         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1087                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1088         INIT_LIST_HEAD(&vp->list);
1089         vp->local_mac = *local_mac;
1090
1091         dev->netdev_ops = &vnet_ops;
1092         dev->ethtool_ops = &vnet_ethtool_ops;
1093         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1094
1095         err = register_netdev(dev);
1096         if (err) {
1097                 pr_err("Cannot register net device, aborting\n");
1098                 goto err_out_free_dev;
1099         }
1100
1101         netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1102
1103         list_add(&vp->list, &vnet_list);
1104
1105         return vp;
1106
1107 err_out_free_dev:
1108         free_netdev(dev);
1109
1110         return ERR_PTR(err);
1111 }
1112
1113 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1114 {
1115         struct vnet *iter, *vp;
1116
1117         mutex_lock(&vnet_list_mutex);
1118         vp = NULL;
1119         list_for_each_entry(iter, &vnet_list, list) {
1120                 if (iter->local_mac == *local_mac) {
1121                         vp = iter;
1122                         break;
1123                 }
1124         }
1125         if (!vp)
1126                 vp = vnet_new(local_mac);
1127         mutex_unlock(&vnet_list_mutex);
1128
1129         return vp;
1130 }
1131
1132 static void vnet_cleanup(void)
1133 {
1134         struct vnet *vp;
1135         struct net_device *dev;
1136
1137         mutex_lock(&vnet_list_mutex);
1138         while (!list_empty(&vnet_list)) {
1139                 vp = list_first_entry(&vnet_list, struct vnet, list);
1140                 list_del(&vp->list);
1141                 dev = vp->dev;
1142                 tasklet_kill(&vp->vnet_tx_wakeup);
1143                 /* vio_unregister_driver() should have cleaned up port_list */
1144                 BUG_ON(!list_empty(&vp->port_list));
1145                 unregister_netdev(dev);
1146                 free_netdev(dev);
1147         }
1148         mutex_unlock(&vnet_list_mutex);
1149 }
1150
1151 static const char *local_mac_prop = "local-mac-address";
1152
1153 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1154                                                 u64 port_node)
1155 {
1156         const u64 *local_mac = NULL;
1157         u64 a;
1158
1159         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1160                 u64 target = mdesc_arc_target(hp, a);
1161                 const char *name;
1162
1163                 name = mdesc_get_property(hp, target, "name", NULL);
1164                 if (!name || strcmp(name, "network"))
1165                         continue;
1166
1167                 local_mac = mdesc_get_property(hp, target,
1168                                                local_mac_prop, NULL);
1169                 if (local_mac)
1170                         break;
1171         }
1172         if (!local_mac)
1173                 return ERR_PTR(-ENODEV);
1174
1175         return vnet_find_or_create(local_mac);
1176 }
1177
1178 static struct ldc_channel_config vnet_ldc_cfg = {
1179         .event          = vnet_event,
1180         .mtu            = 64,
1181         .mode           = LDC_MODE_UNRELIABLE,
1182 };
1183
1184 static struct vio_driver_ops vnet_vio_ops = {
1185         .send_attr              = vnet_send_attr,
1186         .handle_attr            = vnet_handle_attr,
1187         .handshake_complete     = vnet_handshake_complete,
1188 };
1189
1190 static void print_version(void)
1191 {
1192         printk_once(KERN_INFO "%s", version);
1193 }
1194
1195 const char *remote_macaddr_prop = "remote-mac-address";
1196
1197 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1198 {
1199         struct mdesc_handle *hp;
1200         struct vnet_port *port;
1201         unsigned long flags;
1202         struct vnet *vp;
1203         const u64 *rmac;
1204         int len, i, err, switch_port;
1205
1206         print_version();
1207
1208         hp = mdesc_grab();
1209
1210         vp = vnet_find_parent(hp, vdev->mp);
1211         if (IS_ERR(vp)) {
1212                 pr_err("Cannot find port parent vnet\n");
1213                 err = PTR_ERR(vp);
1214                 goto err_out_put_mdesc;
1215         }
1216
1217         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1218         err = -ENODEV;
1219         if (!rmac) {
1220                 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1221                 goto err_out_put_mdesc;
1222         }
1223
1224         port = kzalloc(sizeof(*port), GFP_KERNEL);
1225         err = -ENOMEM;
1226         if (!port)
1227                 goto err_out_put_mdesc;
1228
1229         for (i = 0; i < ETH_ALEN; i++)
1230                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1231
1232         port->vp = vp;
1233
1234         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1235                               vnet_versions, ARRAY_SIZE(vnet_versions),
1236                               &vnet_vio_ops, vp->dev->name);
1237         if (err)
1238                 goto err_out_free_port;
1239
1240         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1241         if (err)
1242                 goto err_out_free_port;
1243
1244         err = vnet_port_alloc_tx_bufs(port);
1245         if (err)
1246                 goto err_out_free_ldc;
1247
1248         INIT_HLIST_NODE(&port->hash);
1249         INIT_LIST_HEAD(&port->list);
1250
1251         switch_port = 0;
1252         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1253                 switch_port = 1;
1254         port->switch_port = switch_port;
1255
1256         spin_lock_irqsave(&vp->lock, flags);
1257         if (switch_port)
1258                 list_add(&port->list, &vp->port_list);
1259         else
1260                 list_add_tail(&port->list, &vp->port_list);
1261         hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1262         spin_unlock_irqrestore(&vp->lock, flags);
1263
1264         dev_set_drvdata(&vdev->dev, port);
1265
1266         pr_info("%s: PORT ( remote-mac %pM%s )\n",
1267                 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1268
1269         vio_port_up(&port->vio);
1270
1271         mdesc_release(hp);
1272
1273         return 0;
1274
1275 err_out_free_ldc:
1276         vio_ldc_free(&port->vio);
1277
1278 err_out_free_port:
1279         kfree(port);
1280
1281 err_out_put_mdesc:
1282         mdesc_release(hp);
1283         return err;
1284 }
1285
1286 static int vnet_port_remove(struct vio_dev *vdev)
1287 {
1288         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1289
1290         if (port) {
1291                 struct vnet *vp = port->vp;
1292                 unsigned long flags;
1293
1294                 del_timer_sync(&port->vio.timer);
1295
1296                 spin_lock_irqsave(&vp->lock, flags);
1297                 list_del(&port->list);
1298                 hlist_del(&port->hash);
1299                 spin_unlock_irqrestore(&vp->lock, flags);
1300
1301                 vnet_port_free_tx_bufs(port);
1302                 vio_ldc_free(&port->vio);
1303
1304                 dev_set_drvdata(&vdev->dev, NULL);
1305
1306                 kfree(port);
1307
1308         }
1309         return 0;
1310 }
1311
1312 static const struct vio_device_id vnet_port_match[] = {
1313         {
1314                 .type = "vnet-port",
1315         },
1316         {},
1317 };
1318 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1319
1320 static struct vio_driver vnet_port_driver = {
1321         .id_table       = vnet_port_match,
1322         .probe          = vnet_port_probe,
1323         .remove         = vnet_port_remove,
1324         .name           = "vnet_port",
1325 };
1326
1327 static int __init vnet_init(void)
1328 {
1329         return vio_register_driver(&vnet_port_driver);
1330 }
1331
1332 static void __exit vnet_exit(void)
1333 {
1334         vio_unregister_driver(&vnet_port_driver);
1335         vnet_cleanup();
1336 }
1337
1338 module_init(vnet_init);
1339 module_exit(vnet_exit);