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         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 354                desc->hdr.state, desc->hdr.ack,
 355                desc->size, desc->ncookies,
 356                desc->cookies[0].cookie_addr,
 357                desc->cookies[0].cookie_size);
 358
 359         if (desc->hdr.state != VIO_DESC_READY)
 360                 return 1;
 361         err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
 362         if (err == -ECONNRESET)
 363                 return err;
 364         desc->hdr.state = VIO_DESC_DONE;
 365         err = put_rx_desc(port, dr, desc, index);
 366         if (err < 0)
 367                 return err;
 368         *needs_ack = desc->hdr.ack;
 369         return 0;
 370 }
 371
 372 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 373                         u32 start, u32 end)
 374 {
 375         struct vio_driver_state *vio = &port->vio;
 376         int ack_start = -1, ack_end = -1;
 377
 378         end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
 379
 380         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
 381
 382         while (start != end) {
 383                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
 384                 if (err == -ECONNRESET)
 385                         return err;
 386                 if (err != 0)
 387                         break;
 388                 if (ack_start == -1)
 389                         ack_start = start;
 390                 ack_end = start;
 391                 start = next_idx(start, dr);
 392                 if (ack && start != end) {
 393                         err = vnet_send_ack(port, dr, ack_start, ack_end,
 394                                             VIO_DRING_ACTIVE);
 395                         if (err == -ECONNRESET)
 396                                 return err;
 397                         ack_start = -1;
 398                 }
 399         }
 400         if (unlikely(ack_start == -1))
 401                 ack_start = ack_end = prev_idx(start, dr);
 402         return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
 403 }
 404
 405 static int vnet_rx(struct vnet_port *port, void *msgbuf)
 406 {
 407         struct vio_dring_data *pkt = msgbuf;
 408         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
 409         struct vio_driver_state *vio = &port->vio;
 410
 411         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
 412                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
 413
 414         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 415                 return 0;
 416         if (unlikely(pkt->seq != dr->rcv_nxt)) {
 417                 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
 418                        pkt->seq, dr->rcv_nxt);
 419                 return 0;
 420         }
 421
 422         dr->rcv_nxt++;
 423
 424         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
 425
 426         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
 427 }
 428
 429 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
 430 {
 431         u32 idx = dr->cons;
 432         int found = 0;
 433
 434         while (idx != dr->prod) {
 435                 if (idx == end) {
 436                         found = 1;
 437                         break;
 438                 }
 439                 idx = next_idx(idx, dr);
 440         }
 441         return found;
 442 }
 443
 444 static int vnet_ack(struct vnet_port *port, void *msgbuf)
 445 {
 446         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 447         struct vio_dring_data *pkt = msgbuf;
 448         struct net_device *dev;
 449         struct vnet *vp;
 450         u32 end;
 451
 452         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 453                 return 0;
 454
 455         end = pkt->end_idx;
 456         if (unlikely(!idx_is_pending(dr, end)))
 457                 return 0;
 458
 459         dr->cons = next_idx(end, dr);
 460
 461         vp = port->vp;
 462         dev = vp->dev;
 463         if (unlikely(netif_queue_stopped(dev) &&
 464                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
 465                 return 1;
 466
 467         return 0;
 468 }
 469
 470 static int vnet_nack(struct vnet_port *port, void *msgbuf)
 471 {
 472         /* XXX just reset or similar XXX */
 473         return 0;
 474 }
 475
 476 static int handle_mcast(struct vnet_port *port, void *msgbuf)
 477 {
 478         struct vio_net_mcast_info *pkt = msgbuf;
 479
 480         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
 481                 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
 482                        port->vp->dev->name,
 483                        pkt->tag.type,
 484                        pkt->tag.stype,
 485                        pkt->tag.stype_env,
 486                        pkt->tag.sid);
 487
 488         return 0;
 489 }
 490
 491 static void maybe_tx_wakeup(unsigned long param)
 492 {
 493         struct vnet *vp = (struct vnet *)param;
 494         struct net_device *dev = vp->dev;
 495
 496         netif_tx_lock(dev);
 497         if (likely(netif_queue_stopped(dev))) {
 498                 struct vnet_port *port;
 499                 int wake = 1;
 500
 501                 list_for_each_entry(port, &vp->port_list, list) {
 502                         struct vio_dring_state *dr;
 503
 504                         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 505                         if (vnet_tx_dring_avail(dr) <
 506                             VNET_TX_WAKEUP_THRESH(dr)) {
 507                                 wake = 0;
 508                                 break;
 509                         }
 510                 }
 511                 if (wake)
 512                         netif_wake_queue(dev);
 513         }
 514         netif_tx_unlock(dev);
 515 }
 516
 517 static void vnet_event(void *arg, int event)
 518 {
 519         struct vnet_port *port = arg;
 520         struct vio_driver_state *vio = &port->vio;
 521         unsigned long flags;
 522         int tx_wakeup, err;
 523
 524         spin_lock_irqsave(&vio->lock, flags);
 525
 526         if (unlikely(event == LDC_EVENT_RESET ||
 527                      event == LDC_EVENT_UP)) {
 528                 vio_link_state_change(vio, event);
 529                 spin_unlock_irqrestore(&vio->lock, flags);
 530
 531                 if (event == LDC_EVENT_RESET)
 532                         vio_port_up(vio);
 533                 return;
 534         }
 535
 536         if (unlikely(event != LDC_EVENT_DATA_READY)) {
 537                 pr_warning("Unexpected LDC event %d\n", event);
 538                 spin_unlock_irqrestore(&vio->lock, flags);
 539                 return;
 540         }
 541
 542         tx_wakeup = err = 0;
 543         while (1) {
 544                 union {
 545                         struct vio_msg_tag tag;
 546                         u64 raw[8];
 547                 } msgbuf;
 548
 549                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 550                 if (unlikely(err < 0)) {
 551                         if (err == -ECONNRESET)
 552                                 vio_conn_reset(vio);
 553                         break;
 554                 }
 555                 if (err == 0)
 556                         break;
 557                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 558                        msgbuf.tag.type,
 559                        msgbuf.tag.stype,
 560                        msgbuf.tag.stype_env,
 561                        msgbuf.tag.sid);
 562                 err = vio_validate_sid(vio, &msgbuf.tag);
 563                 if (err < 0)
 564                         break;
 565
 566                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 567                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 568                                 err = vnet_rx(port, &msgbuf);
 569                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 570                                 err = vnet_ack(port, &msgbuf);
 571                                 if (err > 0)
 572                                         tx_wakeup |= err;
 573                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 574                                 err = vnet_nack(port, &msgbuf);
 575                         }
 576                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 577                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 578                                 err = handle_mcast(port, &msgbuf);
 579                         else
 580                                 err = vio_control_pkt_engine(vio, &msgbuf);
 581                         if (err)
 582                                 break;
 583                 } else {
 584                         err = vnet_handle_unknown(port, &msgbuf);
 585                 }
 586                 if (err == -ECONNRESET)
 587                         break;
 588         }
 589         spin_unlock(&vio->lock);
 590         /* Kick off a tasklet to wake the queue.  We cannot call
 591          * maybe_tx_wakeup directly here because we could deadlock on
 592          * netif_tx_lock() with dev_watchdog()
 593          */
 594         if (unlikely(tx_wakeup && err != -ECONNRESET))
 595                 tasklet_schedule(&port->vp->vnet_tx_wakeup);
 596
 597         local_irq_restore(flags);
 598 }
 599
 600 static int __vnet_tx_trigger(struct vnet_port *port)
 601 {
 602         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 603         struct vio_dring_data hdr = {
 604                 .tag = {
 605                         .type           = VIO_TYPE_DATA,
 606                         .stype          = VIO_SUBTYPE_INFO,
 607                         .stype_env      = VIO_DRING_DATA,
 608                         .sid            = vio_send_sid(&port->vio),
 609                 },
 610                 .dring_ident            = dr->ident,
 611                 .start_idx              = dr->prod,
 612                 .end_idx                = (u32) -1,
 613         };
 614         int err, delay;
 615         int retries = 0;
 616
 617         hdr.seq = dr->snd_nxt;
 618         delay = 1;
 619         do {
 620                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 621                 if (err > 0) {
 622                         dr->snd_nxt++;
 623                         break;
 624                 }
 625                 udelay(delay);
 626                 if ((delay <<= 1) > 128)
 627                         delay = 128;
 628                 if (retries++ > VNET_MAX_RETRIES)
 629                         break;
 630         } while (err == -EAGAIN);
 631
 632         return err;
 633 }
 634
 635 static inline bool port_is_up(struct vnet_port *vnet)
 636 {
 637         struct vio_driver_state *vio = &vnet->vio;
 638
 639         return !!(vio->hs_state & VIO_HS_COMPLETE);
 640 }
 641
 642 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
 643 {
 644         unsigned int hash = vnet_hashfn(skb->data);
 645         struct hlist_head *hp = &vp->port_hash[hash];
 646         struct vnet_port *port;
 647
 648         hlist_for_each_entry(port, hp, hash) {
 649                 if (!port_is_up(port))
 650                         continue;
 651                 if (ether_addr_equal(port->raddr, skb->data))
 652                         return port;
 653         }
 654         list_for_each_entry(port, &vp->port_list, list) {
 655                 if (!port->switch_port)
 656                         continue;
 657                 if (!port_is_up(port))
 658                         continue;
 659                 return port;
 660         }
 661         return NULL;
 662 }
 663
 664 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
 665 {
 666         struct vnet_port *ret;
 667         unsigned long flags;
 668
 669         spin_lock_irqsave(&vp->lock, flags);
 670         ret = __tx_port_find(vp, skb);
 671         spin_unlock_irqrestore(&vp->lock, flags);
 672
 673         return ret;
 674 }
 675
 676 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 677 {
 678         struct vnet *vp = netdev_priv(dev);
 679         struct vnet_port *port = tx_port_find(vp, skb);
 680         struct vio_dring_state *dr;
 681         struct vio_net_desc *d;
 682         unsigned long flags;
 683         unsigned int len;
 684         void *tx_buf;
 685         int i, err;
 686
 687         if (unlikely(!port))
 688                 goto out_dropped;
 689
 690         spin_lock_irqsave(&port->vio.lock, flags);
 691
 692         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 693         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
 694                 if (!netif_queue_stopped(dev)) {
 695                         netif_stop_queue(dev);
 696
 697                         /* This is a hard error, log it. */
 698                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
 699                         dev->stats.tx_errors++;
 700                 }
 701                 spin_unlock_irqrestore(&port->vio.lock, flags);
 702                 return NETDEV_TX_BUSY;
 703         }
 704
 705         d = vio_dring_cur(dr);
 706
 707         tx_buf = port->tx_bufs[dr->prod].buf;
 708         skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
 709
 710         len = skb->len;
 711         if (len < ETH_ZLEN) {
 712                 len = ETH_ZLEN;
 713                 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
 714         }
 715
 716         /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
 717          * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
 718          * the protocol itself does not require it as long as the peer
 719          * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
 720          *
 721          * An ACK for every packet in the ring is expensive as the
 722          * sending of LDC messages is slow and affects performance.
 723          */
 724         d->hdr.ack = VIO_ACK_DISABLE;
 725         d->size = len;
 726         d->ncookies = port->tx_bufs[dr->prod].ncookies;
 727         for (i = 0; i < d->ncookies; i++)
 728                 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
 729
 730         /* This has to be a non-SMP write barrier because we are writing
 731          * to memory which is shared with the peer LDOM.
 732          */
 733         wmb();
 734
 735         d->hdr.state = VIO_DESC_READY;
 736
 737         err = __vnet_tx_trigger(port);
 738         if (unlikely(err < 0)) {
 739                 netdev_info(dev, "TX trigger error %d\n", err);
 740                 d->hdr.state = VIO_DESC_FREE;
 741                 dev->stats.tx_carrier_errors++;
 742                 goto out_dropped_unlock;
 743         }
 744
 745         dev->stats.tx_packets++;
 746         dev->stats.tx_bytes += skb->len;
 747
 748         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
 749         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
 750                 netif_stop_queue(dev);
 751                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
 752                         netif_wake_queue(dev);
 753         }
 754
 755         spin_unlock_irqrestore(&port->vio.lock, flags);
 756
 757         dev_kfree_skb(skb);
 758
 759         return NETDEV_TX_OK;
 760
 761 out_dropped_unlock:
 762         spin_unlock_irqrestore(&port->vio.lock, flags);
 763
 764 out_dropped:
 765         dev_kfree_skb(skb);
 766         dev->stats.tx_dropped++;
 767         return NETDEV_TX_OK;
 768 }
 769
 770 static void vnet_tx_timeout(struct net_device *dev)
 771 {
 772         /* XXX Implement me XXX */
 773 }
 774
 775 static int vnet_open(struct net_device *dev)
 776 {
 777         netif_carrier_on(dev);
 778         netif_start_queue(dev);
 779
 780         return 0;
 781 }
 782
 783 static int vnet_close(struct net_device *dev)
 784 {
 785         netif_stop_queue(dev);
 786         netif_carrier_off(dev);
 787
 788         return 0;
 789 }
 790
 791 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
 792 {
 793         struct vnet_mcast_entry *m;
 794
 795         for (m = vp->mcast_list; m; m = m->next) {
 796                 if (ether_addr_equal(m->addr, addr))
 797                         return m;
 798         }
 799         return NULL;
 800 }
 801
 802 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
 803 {
 804         struct netdev_hw_addr *ha;
 805
 806         netdev_for_each_mc_addr(ha, dev) {
 807                 struct vnet_mcast_entry *m;
 808
 809                 m = __vnet_mc_find(vp, ha->addr);
 810                 if (m) {
 811                         m->hit = 1;
 812                         continue;
 813                 }
 814
 815                 if (!m) {
 816                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
 817                         if (!m)
 818                                 continue;
 819                         memcpy(m->addr, ha->addr, ETH_ALEN);
 820                         m->hit = 1;
 821
 822                         m->next = vp->mcast_list;
 823                         vp->mcast_list = m;
 824                 }
 825         }
 826 }
 827
 828 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
 829 {
 830         struct vio_net_mcast_info info;
 831         struct vnet_mcast_entry *m, **pp;
 832         int n_addrs;
 833
 834         memset(&info, 0, sizeof(info));
 835
 836         info.tag.type = VIO_TYPE_CTRL;
 837         info.tag.stype = VIO_SUBTYPE_INFO;
 838         info.tag.stype_env = VNET_MCAST_INFO;
 839         info.tag.sid = vio_send_sid(&port->vio);
 840         info.set = 1;
 841
 842         n_addrs = 0;
 843         for (m = vp->mcast_list; m; m = m->next) {
 844                 if (m->sent)
 845                         continue;
 846                 m->sent = 1;
 847                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 848                        m->addr, ETH_ALEN);
 849                 if (++n_addrs == VNET_NUM_MCAST) {
 850                         info.count = n_addrs;
 851
 852                         (void) vio_ldc_send(&port->vio, &info,
 853                                             sizeof(info));
 854                         n_addrs = 0;
 855                 }
 856         }
 857         if (n_addrs) {
 858                 info.count = n_addrs;
 859                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 860         }
 861
 862         info.set = 0;
 863
 864         n_addrs = 0;
 865         pp = &vp->mcast_list;
 866         while ((m = *pp) != NULL) {
 867                 if (m->hit) {
 868                         m->hit = 0;
 869                         pp = &m->next;
 870                         continue;
 871                 }
 872
 873                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
 874                        m->addr, ETH_ALEN);
 875                 if (++n_addrs == VNET_NUM_MCAST) {
 876                         info.count = n_addrs;
 877                         (void) vio_ldc_send(&port->vio, &info,
 878                                             sizeof(info));
 879                         n_addrs = 0;
 880                 }
 881
 882                 *pp = m->next;
 883                 kfree(m);
 884         }
 885         if (n_addrs) {
 886                 info.count = n_addrs;
 887                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
 888         }
 889 }
 890
 891 static void vnet_set_rx_mode(struct net_device *dev)
 892 {
 893         struct vnet *vp = netdev_priv(dev);
 894         struct vnet_port *port;
 895         unsigned long flags;
 896
 897         spin_lock_irqsave(&vp->lock, flags);
 898         if (!list_empty(&vp->port_list)) {
 899                 port = list_entry(vp->port_list.next, struct vnet_port, list);
 900
 901                 if (port->switch_port) {
 902                         __update_mc_list(vp, dev);
 903                         __send_mc_list(vp, port);
 904                 }
 905         }
 906         spin_unlock_irqrestore(&vp->lock, flags);
 907 }
 908
 909 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
 910 {
 911         if (new_mtu != ETH_DATA_LEN)
 912                 return -EINVAL;
 913
 914         dev->mtu = new_mtu;
 915         return 0;
 916 }
 917
 918 static int vnet_set_mac_addr(struct net_device *dev, void *p)
 919 {
 920         return -EINVAL;
 921 }
 922
 923 static void vnet_get_drvinfo(struct net_device *dev,
 924                              struct ethtool_drvinfo *info)
 925 {
 926         strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
 927         strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
 928 }
 929
 930 static u32 vnet_get_msglevel(struct net_device *dev)
 931 {
 932         struct vnet *vp = netdev_priv(dev);
 933         return vp->msg_enable;
 934 }
 935
 936 static void vnet_set_msglevel(struct net_device *dev, u32 value)
 937 {
 938         struct vnet *vp = netdev_priv(dev);
 939         vp->msg_enable = value;
 940 }
 941
 942 static const struct ethtool_ops vnet_ethtool_ops = {
 943         .get_drvinfo            = vnet_get_drvinfo,
 944         .get_msglevel           = vnet_get_msglevel,
 945         .set_msglevel           = vnet_set_msglevel,
 946         .get_link               = ethtool_op_get_link,
 947 };
 948
 949 static void vnet_port_free_tx_bufs(struct vnet_port *port)
 950 {
 951         struct vio_dring_state *dr;
 952         int i;
 953
 954         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 955         if (dr->base) {
 956                 ldc_free_exp_dring(port->vio.lp, dr->base,
 957                                    (dr->entry_size * dr->num_entries),
 958                                    dr->cookies, dr->ncookies);
 959                 dr->base = NULL;
 960                 dr->entry_size = 0;
 961                 dr->num_entries = 0;
 962                 dr->pending = 0;
 963                 dr->ncookies = 0;
 964         }
 965
 966         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 967                 void *buf = port->tx_bufs[i].buf;
 968
 969                 if (!buf)
 970                         continue;
 971
 972                 ldc_unmap(port->vio.lp,
 973                           port->tx_bufs[i].cookies,
 974                           port->tx_bufs[i].ncookies);
 975
 976                 kfree(buf);
 977                 port->tx_bufs[i].buf = NULL;
 978         }
 979 }
 980
 981 static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
 982 {
 983         struct vio_dring_state *dr;
 984         unsigned long len;
 985         int i, err, ncookies;
 986         void *dring;
 987
 988         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
 989                 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
 990                 int map_len = (ETH_FRAME_LEN + 7) & ~7;
 991
 992                 err = -ENOMEM;
 993                 if (!buf)
 994                         goto err_out;
 995
 996                 err = -EFAULT;
 997                 if ((unsigned long)buf & (8UL - 1)) {
 998                         pr_err("TX buffer misaligned\n");
 999                         kfree(buf);
1000                         goto err_out;
1001                 }
1002
1003                 err = ldc_map_single(port->vio.lp, buf, map_len,
1004                                      port->tx_bufs[i].cookies, 2,
1005                                      (LDC_MAP_SHADOW |
1006                                       LDC_MAP_DIRECT |
1007                                       LDC_MAP_RW));
1008                 if (err < 0) {
1009                         kfree(buf);
1010                         goto err_out;
1011                 }
1012                 port->tx_bufs[i].buf = buf;
1013                 port->tx_bufs[i].ncookies = err;
1014         }
1015
1016         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1017
1018         len = (VNET_TX_RING_SIZE *
1019                (sizeof(struct vio_net_desc) +
1020                 (sizeof(struct ldc_trans_cookie) * 2)));
1021
1022         ncookies = VIO_MAX_RING_COOKIES;
1023         dring = ldc_alloc_exp_dring(port->vio.lp, len,
1024                                     dr->cookies, &ncookies,
1025                                     (LDC_MAP_SHADOW |
1026                                      LDC_MAP_DIRECT |
1027                                      LDC_MAP_RW));
1028         if (IS_ERR(dring)) {
1029                 err = PTR_ERR(dring);
1030                 goto err_out;
1031         }
1032
1033         dr->base = dring;
1034         dr->entry_size = (sizeof(struct vio_net_desc) +
1035                           (sizeof(struct ldc_trans_cookie) * 2));
1036         dr->num_entries = VNET_TX_RING_SIZE;
1037         dr->prod = dr->cons = 0;
1038         dr->pending = VNET_TX_RING_SIZE;
1039         dr->ncookies = ncookies;
1040
1041         return 0;
1042
1043 err_out:
1044         vnet_port_free_tx_bufs(port);
1045
1046         return err;
1047 }
1048
1049 static LIST_HEAD(vnet_list);
1050 static DEFINE_MUTEX(vnet_list_mutex);
1051
1052 static const struct net_device_ops vnet_ops = {
1053         .ndo_open               = vnet_open,
1054         .ndo_stop               = vnet_close,
1055         .ndo_set_rx_mode        = vnet_set_rx_mode,
1056         .ndo_set_mac_address    = vnet_set_mac_addr,
1057         .ndo_validate_addr      = eth_validate_addr,
1058         .ndo_tx_timeout         = vnet_tx_timeout,
1059         .ndo_change_mtu         = vnet_change_mtu,
1060         .ndo_start_xmit         = vnet_start_xmit,
1061 };
1062
1063 static struct vnet *vnet_new(const u64 *local_mac)
1064 {
1065         struct net_device *dev;
1066         struct vnet *vp;
1067         int err, i;
1068
1069         dev = alloc_etherdev(sizeof(*vp));
1070         if (!dev)
1071                 return ERR_PTR(-ENOMEM);
1072
1073         for (i = 0; i < ETH_ALEN; i++)
1074                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1075
1076         vp = netdev_priv(dev);
1077
1078         spin_lock_init(&vp->lock);
1079         tasklet_init(&vp->vnet_tx_wakeup, maybe_tx_wakeup, (unsigned long)vp);
1080         vp->dev = dev;
1081
1082         INIT_LIST_HEAD(&vp->port_list);
1083         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1084                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1085         INIT_LIST_HEAD(&vp->list);
1086         vp->local_mac = *local_mac;
1087
1088         dev->netdev_ops = &vnet_ops;
1089         dev->ethtool_ops = &vnet_ethtool_ops;
1090         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1091
1092         err = register_netdev(dev);
1093         if (err) {
1094                 pr_err("Cannot register net device, aborting\n");
1095                 goto err_out_free_dev;
1096         }
1097
1098         netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1099
1100         list_add(&vp->list, &vnet_list);
1101
1102         return vp;
1103
1104 err_out_free_dev:
1105         free_netdev(dev);
1106
1107         return ERR_PTR(err);
1108 }
1109
1110 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1111 {
1112         struct vnet *iter, *vp;
1113
1114         mutex_lock(&vnet_list_mutex);
1115         vp = NULL;
1116         list_for_each_entry(iter, &vnet_list, list) {
1117                 if (iter->local_mac == *local_mac) {
1118                         vp = iter;
1119                         break;
1120                 }
1121         }
1122         if (!vp)
1123                 vp = vnet_new(local_mac);
1124         mutex_unlock(&vnet_list_mutex);
1125
1126         return vp;
1127 }
1128
1129 static void vnet_cleanup(void)
1130 {
1131         struct vnet *vp;
1132         struct net_device *dev;
1133
1134         mutex_lock(&vnet_list_mutex);
1135         while (!list_empty(&vnet_list)) {
1136                 vp = list_first_entry(&vnet_list, struct vnet, list);
1137                 list_del(&vp->list);
1138                 dev = vp->dev;
1139                 tasklet_kill(&vp->vnet_tx_wakeup);
1140                 /* vio_unregister_driver() should have cleaned up port_list */
1141                 BUG_ON(!list_empty(&vp->port_list));
1142                 unregister_netdev(dev);
1143                 free_netdev(dev);
1144         }
1145         mutex_unlock(&vnet_list_mutex);
1146 }
1147
1148 static const char *local_mac_prop = "local-mac-address";
1149
1150 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1151                                                 u64 port_node)
1152 {
1153         const u64 *local_mac = NULL;
1154         u64 a;
1155
1156         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1157                 u64 target = mdesc_arc_target(hp, a);
1158                 const char *name;
1159
1160                 name = mdesc_get_property(hp, target, "name", NULL);
1161                 if (!name || strcmp(name, "network"))
1162                         continue;
1163
1164                 local_mac = mdesc_get_property(hp, target,
1165                                                local_mac_prop, NULL);
1166                 if (local_mac)
1167                         break;
1168         }
1169         if (!local_mac)
1170                 return ERR_PTR(-ENODEV);
1171
1172         return vnet_find_or_create(local_mac);
1173 }
1174
1175 static struct ldc_channel_config vnet_ldc_cfg = {
1176         .event          = vnet_event,
1177         .mtu            = 64,
1178         .mode           = LDC_MODE_UNRELIABLE,
1179 };
1180
1181 static struct vio_driver_ops vnet_vio_ops = {
1182         .send_attr              = vnet_send_attr,
1183         .handle_attr            = vnet_handle_attr,
1184         .handshake_complete     = vnet_handshake_complete,
1185 };
1186
1187 static void print_version(void)
1188 {
1189         printk_once(KERN_INFO "%s", version);
1190 }
1191
1192 const char *remote_macaddr_prop = "remote-mac-address";
1193
1194 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1195 {
1196         struct mdesc_handle *hp;
1197         struct vnet_port *port;
1198         unsigned long flags;
1199         struct vnet *vp;
1200         const u64 *rmac;
1201         int len, i, err, switch_port;
1202
1203         print_version();
1204
1205         hp = mdesc_grab();
1206
1207         vp = vnet_find_parent(hp, vdev->mp);
1208         if (IS_ERR(vp)) {
1209                 pr_err("Cannot find port parent vnet\n");
1210                 err = PTR_ERR(vp);
1211                 goto err_out_put_mdesc;
1212         }
1213
1214         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1215         err = -ENODEV;
1216         if (!rmac) {
1217                 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1218                 goto err_out_put_mdesc;
1219         }
1220
1221         port = kzalloc(sizeof(*port), GFP_KERNEL);
1222         err = -ENOMEM;
1223         if (!port)
1224                 goto err_out_put_mdesc;
1225
1226         for (i = 0; i < ETH_ALEN; i++)
1227                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1228
1229         port->vp = vp;
1230
1231         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1232                               vnet_versions, ARRAY_SIZE(vnet_versions),
1233                               &vnet_vio_ops, vp->dev->name);
1234         if (err)
1235                 goto err_out_free_port;
1236
1237         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1238         if (err)
1239                 goto err_out_free_port;
1240
1241         err = vnet_port_alloc_tx_bufs(port);
1242         if (err)
1243                 goto err_out_free_ldc;
1244
1245         INIT_HLIST_NODE(&port->hash);
1246         INIT_LIST_HEAD(&port->list);
1247
1248         switch_port = 0;
1249         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1250                 switch_port = 1;
1251         port->switch_port = switch_port;
1252
1253         spin_lock_irqsave(&vp->lock, flags);
1254         if (switch_port)
1255                 list_add(&port->list, &vp->port_list);
1256         else
1257                 list_add_tail(&port->list, &vp->port_list);
1258         hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1259         spin_unlock_irqrestore(&vp->lock, flags);
1260
1261         dev_set_drvdata(&vdev->dev, port);
1262
1263         pr_info("%s: PORT ( remote-mac %pM%s )\n",
1264                 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1265
1266         vio_port_up(&port->vio);
1267
1268         mdesc_release(hp);
1269
1270         return 0;
1271
1272 err_out_free_ldc:
1273         vio_ldc_free(&port->vio);
1274
1275 err_out_free_port:
1276         kfree(port);
1277
1278 err_out_put_mdesc:
1279         mdesc_release(hp);
1280         return err;
1281 }
1282
1283 static int vnet_port_remove(struct vio_dev *vdev)
1284 {
1285         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1286
1287         if (port) {
1288                 struct vnet *vp = port->vp;
1289                 unsigned long flags;
1290
1291                 del_timer_sync(&port->vio.timer);
1292
1293                 spin_lock_irqsave(&vp->lock, flags);
1294                 list_del(&port->list);
1295                 hlist_del(&port->hash);
1296                 spin_unlock_irqrestore(&vp->lock, flags);
1297
1298                 vnet_port_free_tx_bufs(port);
1299                 vio_ldc_free(&port->vio);
1300
1301                 dev_set_drvdata(&vdev->dev, NULL);
1302
1303                 kfree(port);
1304
1305         }
1306         return 0;
1307 }
1308
1309 static const struct vio_device_id vnet_port_match[] = {
1310         {
1311                 .type = "vnet-port",
1312         },
1313         {},
1314 };
1315 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1316
1317 static struct vio_driver vnet_port_driver = {
1318         .id_table       = vnet_port_match,
1319         .probe          = vnet_port_probe,
1320         .remove         = vnet_port_remove,
1321         .name           = "vnet_port",
1322 };
1323
1324 static int __init vnet_init(void)
1325 {
1326         return vio_register_driver(&vnet_port_driver);
1327 }
1328
1329 static void __exit vnet_exit(void)
1330 {
1331         vio_unregister_driver(&vnet_port_driver);
1332         vnet_cleanup();
1333 }
1334
1335 module_init(vnet_init);
1336 module_exit(vnet_exit);