drivers/net/ethernet/sun/sunvnet_common.c

   1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
   2  *
   3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
   4  * Copyright (C) 2016 Oracle. All rights reserved.
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/kernel.h>
   9 #include <linux/types.h>
  10 #include <linux/slab.h>
  11 #include <linux/delay.h>
  12 #include <linux/init.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/ethtool.h>
  15 #include <linux/etherdevice.h>
  16 #include <linux/mutex.h>
  17 #include <linux/highmem.h>
  18 #include <linux/if_vlan.h>
  19 #define CREATE_TRACE_POINTS
  20 #include <trace/events/sunvnet.h>
  21
  22 #if IS_ENABLED(CONFIG_IPV6)
  23 #include <linux/icmpv6.h>
  24 #endif
  25
  26 #include <net/ip.h>
  27 #include <net/icmp.h>
  28 #include <net/route.h>
  29
  30 #include <asm/vio.h>
  31 #include <asm/ldc.h>
  32
  33 #include "sunvnet_common.h"
  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 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
  41 static void vnet_port_reset(struct vnet_port *port);
  42
  43 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
  44 {
  45         return vio_dring_avail(dr, VNET_TX_RING_SIZE);
  46 }
  47
  48 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
  49 {
  50         struct vio_msg_tag *pkt = arg;
  51
  52         pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
  53                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
  54         pr_err("Resetting connection\n");
  55
  56         ldc_disconnect(port->vio.lp);
  57
  58         return -ECONNRESET;
  59 }
  60
  61 static int vnet_port_alloc_tx_ring(struct vnet_port *port);
  62
  63 int sunvnet_send_attr_common(struct vio_driver_state *vio)
  64 {
  65         struct vnet_port *port = to_vnet_port(vio);
  66         struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
  67         struct vio_net_attr_info pkt;
  68         int framelen = ETH_FRAME_LEN;
  69         int i, err;
  70
  71         err = vnet_port_alloc_tx_ring(to_vnet_port(vio));
  72         if (err)
  73                 return err;
  74
  75         memset(&pkt, 0, sizeof(pkt));
  76         pkt.tag.type = VIO_TYPE_CTRL;
  77         pkt.tag.stype = VIO_SUBTYPE_INFO;
  78         pkt.tag.stype_env = VIO_ATTR_INFO;
  79         pkt.tag.sid = vio_send_sid(vio);
  80         if (vio_version_before(vio, 1, 2))
  81                 pkt.xfer_mode = VIO_DRING_MODE;
  82         else
  83                 pkt.xfer_mode = VIO_NEW_DRING_MODE;
  84         pkt.addr_type = VNET_ADDR_ETHERMAC;
  85         pkt.ack_freq = 0;
  86         for (i = 0; i < 6; i++)
  87                 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
  88         if (vio_version_after(vio, 1, 3)) {
  89                 if (port->rmtu) {
  90                         port->rmtu = min(VNET_MAXPACKET, port->rmtu);
  91                         pkt.mtu = port->rmtu;
  92                 } else {
  93                         port->rmtu = VNET_MAXPACKET;
  94                         pkt.mtu = port->rmtu;
  95                 }
  96                 if (vio_version_after_eq(vio, 1, 6))
  97                         pkt.options = VIO_TX_DRING;
  98         } else if (vio_version_before(vio, 1, 3)) {
  99                 pkt.mtu = framelen;
 100         } else { /* v1.3 */
 101                 pkt.mtu = framelen + VLAN_HLEN;
 102         }
 103
 104         pkt.cflags = 0;
 105         if (vio_version_after_eq(vio, 1, 7) && port->tso) {
 106                 pkt.cflags |= VNET_LSO_IPV4_CAPAB;
 107                 if (!port->tsolen)
 108                         port->tsolen = VNET_MAXTSO;
 109                 pkt.ipv4_lso_maxlen = port->tsolen;
 110         }
 111
 112         pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
 113
 114         viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
 115                "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
 116                "cflags[0x%04x] lso_max[%u]\n",
 117                pkt.xfer_mode, pkt.addr_type,
 118                (unsigned long long)pkt.addr,
 119                pkt.ack_freq, pkt.plnk_updt, pkt.options,
 120                (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
 121
 122         return vio_ldc_send(vio, &pkt, sizeof(pkt));
 123 }
 124 EXPORT_SYMBOL_GPL(sunvnet_send_attr_common);
 125
 126 static int handle_attr_info(struct vio_driver_state *vio,
 127                             struct vio_net_attr_info *pkt)
 128 {
 129         struct vnet_port *port = to_vnet_port(vio);
 130         u64     localmtu;
 131         u8      xfer_mode;
 132
 133         viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
 134                "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
 135                " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
 136                pkt->xfer_mode, pkt->addr_type,
 137                (unsigned long long)pkt->addr,
 138                pkt->ack_freq, pkt->plnk_updt, pkt->options,
 139                (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
 140                pkt->ipv4_lso_maxlen);
 141
 142         pkt->tag.sid = vio_send_sid(vio);
 143
 144         xfer_mode = pkt->xfer_mode;
 145         /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
 146         if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
 147                 xfer_mode = VIO_NEW_DRING_MODE;
 148
 149         /* MTU negotiation:
 150          *      < v1.3 - ETH_FRAME_LEN exactly
 151          *      > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
 152          *                      pkt->mtu for ACK
 153          *      = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
 154          */
 155         if (vio_version_before(vio, 1, 3)) {
 156                 localmtu = ETH_FRAME_LEN;
 157         } else if (vio_version_after(vio, 1, 3)) {
 158                 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
 159                 localmtu = min(pkt->mtu, localmtu);
 160                 pkt->mtu = localmtu;
 161         } else { /* v1.3 */
 162                 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
 163         }
 164         port->rmtu = localmtu;
 165
 166         /* LSO negotiation */
 167         if (vio_version_after_eq(vio, 1, 7))
 168                 port->tso &= !!(pkt->cflags & VNET_LSO_IPV4_CAPAB);
 169         else
 170                 port->tso = false;
 171         if (port->tso) {
 172                 if (!port->tsolen)
 173                         port->tsolen = VNET_MAXTSO;
 174                 port->tsolen = min(port->tsolen, pkt->ipv4_lso_maxlen);
 175                 if (port->tsolen < VNET_MINTSO) {
 176                         port->tso = false;
 177                         port->tsolen = 0;
 178                         pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
 179                 }
 180                 pkt->ipv4_lso_maxlen = port->tsolen;
 181         } else {
 182                 pkt->cflags &= ~VNET_LSO_IPV4_CAPAB;
 183                 pkt->ipv4_lso_maxlen = 0;
 184         }
 185
 186         /* for version >= 1.6, ACK packet mode we support */
 187         if (vio_version_after_eq(vio, 1, 6)) {
 188                 pkt->xfer_mode = VIO_NEW_DRING_MODE;
 189                 pkt->options = VIO_TX_DRING;
 190         }
 191
 192         if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
 193             pkt->addr_type != VNET_ADDR_ETHERMAC ||
 194             pkt->mtu != localmtu) {
 195                 viodbg(HS, "SEND NET ATTR NACK\n");
 196
 197                 pkt->tag.stype = VIO_SUBTYPE_NACK;
 198
 199                 (void)vio_ldc_send(vio, pkt, sizeof(*pkt));
 200
 201                 return -ECONNRESET;
 202         }
 203
 204         viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
 205                "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
 206                "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
 207                pkt->xfer_mode, pkt->addr_type,
 208                (unsigned long long)pkt->addr,
 209                pkt->ack_freq, pkt->plnk_updt, pkt->options,
 210                (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
 211                pkt->ipv4_lso_maxlen);
 212
 213         pkt->tag.stype = VIO_SUBTYPE_ACK;
 214
 215         return vio_ldc_send(vio, pkt, sizeof(*pkt));
 216 }
 217
 218 static int handle_attr_ack(struct vio_driver_state *vio,
 219                            struct vio_net_attr_info *pkt)
 220 {
 221         viodbg(HS, "GOT NET ATTR ACK\n");
 222
 223         return 0;
 224 }
 225
 226 static int handle_attr_nack(struct vio_driver_state *vio,
 227                             struct vio_net_attr_info *pkt)
 228 {
 229         viodbg(HS, "GOT NET ATTR NACK\n");
 230
 231         return -ECONNRESET;
 232 }
 233
 234 int sunvnet_handle_attr_common(struct vio_driver_state *vio, void *arg)
 235 {
 236         struct vio_net_attr_info *pkt = arg;
 237
 238         switch (pkt->tag.stype) {
 239         case VIO_SUBTYPE_INFO:
 240                 return handle_attr_info(vio, pkt);
 241
 242         case VIO_SUBTYPE_ACK:
 243                 return handle_attr_ack(vio, pkt);
 244
 245         case VIO_SUBTYPE_NACK:
 246                 return handle_attr_nack(vio, pkt);
 247
 248         default:
 249                 return -ECONNRESET;
 250         }
 251 }
 252 EXPORT_SYMBOL_GPL(sunvnet_handle_attr_common);
 253
 254 void sunvnet_handshake_complete_common(struct vio_driver_state *vio)
 255 {
 256         struct vio_dring_state *dr;
 257
 258         dr = &vio->drings[VIO_DRIVER_RX_RING];
 259         dr->rcv_nxt = 1;
 260         dr->snd_nxt = 1;
 261
 262         dr = &vio->drings[VIO_DRIVER_TX_RING];
 263         dr->rcv_nxt = 1;
 264         dr->snd_nxt = 1;
 265 }
 266 EXPORT_SYMBOL_GPL(sunvnet_handshake_complete_common);
 267
 268 /* The hypervisor interface that implements copying to/from imported
 269  * memory from another domain requires that copies are done to 8-byte
 270  * aligned buffers, and that the lengths of such copies are also 8-byte
 271  * multiples.
 272  *
 273  * So we align skb->data to an 8-byte multiple and pad-out the data
 274  * area so we can round the copy length up to the next multiple of
 275  * 8 for the copy.
 276  *
 277  * The transmitter puts the actual start of the packet 6 bytes into
 278  * the buffer it sends over, so that the IP headers after the ethernet
 279  * header are aligned properly.  These 6 bytes are not in the descriptor
 280  * length, they are simply implied.  This offset is represented using
 281  * the VNET_PACKET_SKIP macro.
 282  */
 283 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
 284                                            unsigned int len)
 285 {
 286         struct sk_buff *skb;
 287         unsigned long addr, off;
 288
 289         skb = netdev_alloc_skb(dev, len + VNET_PACKET_SKIP + 8 + 8);
 290         if (unlikely(!skb))
 291                 return NULL;
 292
 293         addr = (unsigned long)skb->data;
 294         off = ((addr + 7UL) & ~7UL) - addr;
 295         if (off)
 296                 skb_reserve(skb, off);
 297
 298         return skb;
 299 }
 300
 301 static inline void vnet_fullcsum(struct sk_buff *skb)
 302 {
 303         struct iphdr *iph = ip_hdr(skb);
 304         int offset = skb_transport_offset(skb);
 305
 306         if (skb->protocol != htons(ETH_P_IP))
 307                 return;
 308         if (iph->protocol != IPPROTO_TCP &&
 309             iph->protocol != IPPROTO_UDP)
 310                 return;
 311         skb->ip_summed = CHECKSUM_NONE;
 312         skb->csum_level = 1;
 313         skb->csum = 0;
 314         if (iph->protocol == IPPROTO_TCP) {
 315                 struct tcphdr *ptcp = tcp_hdr(skb);
 316
 317                 ptcp->check = 0;
 318                 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
 319                 ptcp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 320                                                 skb->len - offset, IPPROTO_TCP,
 321                                                 skb->csum);
 322         } else if (iph->protocol == IPPROTO_UDP) {
 323                 struct udphdr *pudp = udp_hdr(skb);
 324
 325                 pudp->check = 0;
 326                 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
 327                 pudp->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
 328                                                 skb->len - offset, IPPROTO_UDP,
 329                                                 skb->csum);
 330         }
 331 }
 332
 333 static int vnet_rx_one(struct vnet_port *port, struct vio_net_desc *desc)
 334 {
 335         struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
 336         unsigned int len = desc->size;
 337         unsigned int copy_len;
 338         struct sk_buff *skb;
 339         int maxlen;
 340         int err;
 341
 342         err = -EMSGSIZE;
 343         if (port->tso && port->tsolen > port->rmtu)
 344                 maxlen = port->tsolen;
 345         else
 346                 maxlen = port->rmtu;
 347         if (unlikely(len < ETH_ZLEN || len > maxlen)) {
 348                 dev->stats.rx_length_errors++;
 349                 goto out_dropped;
 350         }
 351
 352         skb = alloc_and_align_skb(dev, len);
 353         err = -ENOMEM;
 354         if (unlikely(!skb)) {
 355                 dev->stats.rx_missed_errors++;
 356                 goto out_dropped;
 357         }
 358
 359         copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
 360         skb_put(skb, copy_len);
 361         err = ldc_copy(port->vio.lp, LDC_COPY_IN,
 362                        skb->data, copy_len, 0,
 363                        desc->cookies, desc->ncookies);
 364         if (unlikely(err < 0)) {
 365                 dev->stats.rx_frame_errors++;
 366                 goto out_free_skb;
 367         }
 368
 369         skb_pull(skb, VNET_PACKET_SKIP);
 370         skb_trim(skb, len);
 371         skb->protocol = eth_type_trans(skb, dev);
 372
 373         if (vio_version_after_eq(&port->vio, 1, 8)) {
 374                 struct vio_net_dext *dext = vio_net_ext(desc);
 375
 376                 skb_reset_network_header(skb);
 377
 378                 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM) {
 379                         if (skb->protocol == ETH_P_IP) {
 380                                 struct iphdr *iph = ip_hdr(skb);
 381
 382                                 iph->check = 0;
 383                                 ip_send_check(iph);
 384                         }
 385                 }
 386                 if ((dext->flags & VNET_PKT_HCK_FULLCKSUM) &&
 387                     skb->ip_summed == CHECKSUM_NONE) {
 388                         if (skb->protocol == htons(ETH_P_IP)) {
 389                                 struct iphdr *iph = ip_hdr(skb);
 390                                 int ihl = iph->ihl * 4;
 391
 392                                 skb_reset_transport_header(skb);
 393                                 skb_set_transport_header(skb, ihl);
 394                                 vnet_fullcsum(skb);
 395                         }
 396                 }
 397                 if (dext->flags & VNET_PKT_HCK_IPV4_HDRCKSUM_OK) {
 398                         skb->ip_summed = CHECKSUM_PARTIAL;
 399                         skb->csum_level = 0;
 400                         if (dext->flags & VNET_PKT_HCK_FULLCKSUM_OK)
 401                                 skb->csum_level = 1;
 402                 }
 403         }
 404
 405         skb->ip_summed = port->switch_port ? CHECKSUM_NONE : CHECKSUM_PARTIAL;
 406
 407         dev->stats.rx_packets++;
 408         dev->stats.rx_bytes += len;
 409         napi_gro_receive(&port->napi, skb);
 410         return 0;
 411
 412 out_free_skb:
 413         kfree_skb(skb);
 414
 415 out_dropped:
 416         dev->stats.rx_dropped++;
 417         return err;
 418 }
 419
 420 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
 421                          u32 start, u32 end, u8 vio_dring_state)
 422 {
 423         struct vio_dring_data hdr = {
 424                 .tag = {
 425                         .type           = VIO_TYPE_DATA,
 426                         .stype          = VIO_SUBTYPE_ACK,
 427                         .stype_env      = VIO_DRING_DATA,
 428                         .sid            = vio_send_sid(&port->vio),
 429                 },
 430                 .dring_ident            = dr->ident,
 431                 .start_idx              = start,
 432                 .end_idx                = end,
 433                 .state                  = vio_dring_state,
 434         };
 435         int err, delay;
 436         int retries = 0;
 437
 438         hdr.seq = dr->snd_nxt;
 439         delay = 1;
 440         do {
 441                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 442                 if (err > 0) {
 443                         dr->snd_nxt++;
 444                         break;
 445                 }
 446                 udelay(delay);
 447                 if ((delay <<= 1) > 128)
 448                         delay = 128;
 449                 if (retries++ > VNET_MAX_RETRIES) {
 450                         pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
 451                                 port->raddr[0], port->raddr[1],
 452                                 port->raddr[2], port->raddr[3],
 453                                 port->raddr[4], port->raddr[5]);
 454                         break;
 455                 }
 456         } while (err == -EAGAIN);
 457
 458         if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
 459                 port->stop_rx_idx = end;
 460                 port->stop_rx = true;
 461         } else {
 462                 port->stop_rx_idx = 0;
 463                 port->stop_rx = false;
 464         }
 465
 466         return err;
 467 }
 468
 469 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
 470                                         struct vio_dring_state *dr,
 471                                         u32 index)
 472 {
 473         struct vio_net_desc *desc = port->vio.desc_buf;
 474         int err;
 475
 476         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
 477                                   (index * dr->entry_size),
 478                                   dr->cookies, dr->ncookies);
 479         if (err < 0)
 480                 return ERR_PTR(err);
 481
 482         return desc;
 483 }
 484
 485 static int put_rx_desc(struct vnet_port *port,
 486                        struct vio_dring_state *dr,
 487                        struct vio_net_desc *desc,
 488                        u32 index)
 489 {
 490         int err;
 491
 492         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
 493                                   (index * dr->entry_size),
 494                                   dr->cookies, dr->ncookies);
 495         if (err < 0)
 496                 return err;
 497
 498         return 0;
 499 }
 500
 501 static int vnet_walk_rx_one(struct vnet_port *port,
 502                             struct vio_dring_state *dr,
 503                             u32 index, int *needs_ack)
 504 {
 505         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
 506         struct vio_driver_state *vio = &port->vio;
 507         int err;
 508
 509         BUG_ON(!desc);
 510         if (IS_ERR(desc))
 511                 return PTR_ERR(desc);
 512
 513         if (desc->hdr.state != VIO_DESC_READY)
 514                 return 1;
 515
 516         dma_rmb();
 517
 518         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
 519                desc->hdr.state, desc->hdr.ack,
 520                desc->size, desc->ncookies,
 521                desc->cookies[0].cookie_addr,
 522                desc->cookies[0].cookie_size);
 523
 524         err = vnet_rx_one(port, desc);
 525         if (err == -ECONNRESET)
 526                 return err;
 527         trace_vnet_rx_one(port->vio._local_sid, port->vio._peer_sid,
 528                           index, desc->hdr.ack);
 529         desc->hdr.state = VIO_DESC_DONE;
 530         err = put_rx_desc(port, dr, desc, index);
 531         if (err < 0)
 532                 return err;
 533         *needs_ack = desc->hdr.ack;
 534         return 0;
 535 }
 536
 537 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
 538                         u32 start, u32 end, int *npkts, int budget)
 539 {
 540         struct vio_driver_state *vio = &port->vio;
 541         int ack_start = -1, ack_end = -1;
 542         bool send_ack = true;
 543
 544         end = (end == (u32)-1) ? vio_dring_prev(dr, start)
 545                                : vio_dring_next(dr, end);
 546
 547         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
 548
 549         while (start != end) {
 550                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
 551
 552                 if (err == -ECONNRESET)
 553                         return err;
 554                 if (err != 0)
 555                         break;
 556                 (*npkts)++;
 557                 if (ack_start == -1)
 558                         ack_start = start;
 559                 ack_end = start;
 560                 start = vio_dring_next(dr, start);
 561                 if (ack && start != end) {
 562                         err = vnet_send_ack(port, dr, ack_start, ack_end,
 563                                             VIO_DRING_ACTIVE);
 564                         if (err == -ECONNRESET)
 565                                 return err;
 566                         ack_start = -1;
 567                 }
 568                 if ((*npkts) >= budget) {
 569                         send_ack = false;
 570                         break;
 571                 }
 572         }
 573         if (unlikely(ack_start == -1)) {
 574                 ack_end = vio_dring_prev(dr, start);
 575                 ack_start = ack_end;
 576         }
 577         if (send_ack) {
 578                 port->napi_resume = false;
 579                 trace_vnet_tx_send_stopped_ack(port->vio._local_sid,
 580                                                port->vio._peer_sid,
 581                                                ack_end, *npkts);
 582                 return vnet_send_ack(port, dr, ack_start, ack_end,
 583                                      VIO_DRING_STOPPED);
 584         } else  {
 585                 trace_vnet_tx_defer_stopped_ack(port->vio._local_sid,
 586                                                 port->vio._peer_sid,
 587                                                 ack_end, *npkts);
 588                 port->napi_resume = true;
 589                 port->napi_stop_idx = ack_end;
 590                 return 1;
 591         }
 592 }
 593
 594 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
 595                    int budget)
 596 {
 597         struct vio_dring_data *pkt = msgbuf;
 598         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
 599         struct vio_driver_state *vio = &port->vio;
 600
 601         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
 602                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
 603
 604         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 605                 return 0;
 606         if (unlikely(pkt->seq != dr->rcv_nxt)) {
 607                 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
 608                        pkt->seq, dr->rcv_nxt);
 609                 return 0;
 610         }
 611
 612         if (!port->napi_resume)
 613                 dr->rcv_nxt++;
 614
 615         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
 616
 617         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
 618                             npkts, budget);
 619 }
 620
 621 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
 622 {
 623         u32 idx = dr->cons;
 624         int found = 0;
 625
 626         while (idx != dr->prod) {
 627                 if (idx == end) {
 628                         found = 1;
 629                         break;
 630                 }
 631                 idx = vio_dring_next(dr, idx);
 632         }
 633         return found;
 634 }
 635
 636 static int vnet_ack(struct vnet_port *port, void *msgbuf)
 637 {
 638         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 639         struct vio_dring_data *pkt = msgbuf;
 640         struct net_device *dev;
 641         u32 end;
 642         struct vio_net_desc *desc;
 643         struct netdev_queue *txq;
 644
 645         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
 646                 return 0;
 647
 648         end = pkt->end_idx;
 649         dev = VNET_PORT_TO_NET_DEVICE(port);
 650         netif_tx_lock(dev);
 651         if (unlikely(!idx_is_pending(dr, end))) {
 652                 netif_tx_unlock(dev);
 653                 return 0;
 654         }
 655
 656         /* sync for race conditions with vnet_start_xmit() and tell xmit it
 657          * is time to send a trigger.
 658          */
 659         trace_vnet_rx_stopped_ack(port->vio._local_sid,
 660                                   port->vio._peer_sid, end);
 661         dr->cons = vio_dring_next(dr, end);
 662         desc = vio_dring_entry(dr, dr->cons);
 663         if (desc->hdr.state == VIO_DESC_READY && !port->start_cons) {
 664                 /* vnet_start_xmit() just populated this dring but missed
 665                  * sending the "start" LDC message to the consumer.
 666                  * Send a "start" trigger on its behalf.
 667                  */
 668                 if (__vnet_tx_trigger(port, dr->cons) > 0)
 669                         port->start_cons = false;
 670                 else
 671                         port->start_cons = true;
 672         } else {
 673                 port->start_cons = true;
 674         }
 675         netif_tx_unlock(dev);
 676
 677         txq = netdev_get_tx_queue(dev, port->q_index);
 678         if (unlikely(netif_tx_queue_stopped(txq) &&
 679                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
 680                 return 1;
 681
 682         return 0;
 683 }
 684
 685 static int vnet_nack(struct vnet_port *port, void *msgbuf)
 686 {
 687         /* XXX just reset or similar XXX */
 688         return 0;
 689 }
 690
 691 static int handle_mcast(struct vnet_port *port, void *msgbuf)
 692 {
 693         struct vio_net_mcast_info *pkt = msgbuf;
 694         struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
 695
 696         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
 697                 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
 698                        dev->name,
 699                        pkt->tag.type,
 700                        pkt->tag.stype,
 701                        pkt->tag.stype_env,
 702                        pkt->tag.sid);
 703
 704         return 0;
 705 }
 706
 707 /* Got back a STOPPED LDC message on port. If the queue is stopped,
 708  * wake it up so that we'll send out another START message at the
 709  * next TX.
 710  */
 711 static void maybe_tx_wakeup(struct vnet_port *port)
 712 {
 713         struct netdev_queue *txq;
 714
 715         txq = netdev_get_tx_queue(VNET_PORT_TO_NET_DEVICE(port),
 716                                   port->q_index);
 717         __netif_tx_lock(txq, smp_processor_id());
 718         if (likely(netif_tx_queue_stopped(txq))) {
 719                 struct vio_dring_state *dr;
 720
 721                 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 722                 netif_tx_wake_queue(txq);
 723         }
 724         __netif_tx_unlock(txq);
 725 }
 726
 727 bool sunvnet_port_is_up_common(struct vnet_port *vnet)
 728 {
 729         struct vio_driver_state *vio = &vnet->vio;
 730
 731         return !!(vio->hs_state & VIO_HS_COMPLETE);
 732 }
 733 EXPORT_SYMBOL_GPL(sunvnet_port_is_up_common);
 734
 735 static int vnet_event_napi(struct vnet_port *port, int budget)
 736 {
 737         struct vio_driver_state *vio = &port->vio;
 738         int tx_wakeup, err;
 739         int npkts = 0;
 740         int event = (port->rx_event & LDC_EVENT_RESET);
 741
 742 ldc_ctrl:
 743         if (unlikely(event == LDC_EVENT_RESET ||
 744                      event == LDC_EVENT_UP)) {
 745                 vio_link_state_change(vio, event);
 746
 747                 if (event == LDC_EVENT_RESET) {
 748                         vnet_port_reset(port);
 749                         vio_port_up(vio);
 750                 }
 751                 port->rx_event = 0;
 752                 return 0;
 753         }
 754         /* We may have multiple LDC events in rx_event. Unroll send_events() */
 755         event = (port->rx_event & LDC_EVENT_UP);
 756         port->rx_event &= ~(LDC_EVENT_RESET | LDC_EVENT_UP);
 757         if (event == LDC_EVENT_UP)
 758                 goto ldc_ctrl;
 759         event = port->rx_event;
 760         if (!(event & LDC_EVENT_DATA_READY))
 761                 return 0;
 762
 763         /* we dont expect any other bits than RESET, UP, DATA_READY */
 764         BUG_ON(event != LDC_EVENT_DATA_READY);
 765
 766         err = 0;
 767         tx_wakeup = 0;
 768         while (1) {
 769                 union {
 770                         struct vio_msg_tag tag;
 771                         u64 raw[8];
 772                 } msgbuf;
 773
 774                 if (port->napi_resume) {
 775                         struct vio_dring_data *pkt =
 776                                 (struct vio_dring_data *)&msgbuf;
 777                         struct vio_dring_state *dr =
 778                                 &port->vio.drings[VIO_DRIVER_RX_RING];
 779
 780                         pkt->tag.type = VIO_TYPE_DATA;
 781                         pkt->tag.stype = VIO_SUBTYPE_INFO;
 782                         pkt->tag.stype_env = VIO_DRING_DATA;
 783                         pkt->seq = dr->rcv_nxt;
 784                         pkt->start_idx = vio_dring_next(dr,
 785                                                         port->napi_stop_idx);
 786                         pkt->end_idx = -1;
 787                         goto napi_resume;
 788                 }
 789                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
 790                 if (unlikely(err < 0)) {
 791                         if (err == -ECONNRESET)
 792                                 vio_conn_reset(vio);
 793                         break;
 794                 }
 795                 if (err == 0)
 796                         break;
 797                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
 798                        msgbuf.tag.type,
 799                        msgbuf.tag.stype,
 800                        msgbuf.tag.stype_env,
 801                        msgbuf.tag.sid);
 802                 err = vio_validate_sid(vio, &msgbuf.tag);
 803                 if (err < 0)
 804                         break;
 805 napi_resume:
 806                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
 807                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
 808                                 if (!sunvnet_port_is_up_common(port)) {
 809                                         /* failures like handshake_failure()
 810                                          * may have cleaned up dring, but
 811                                          * NAPI polling may bring us here.
 812                                          */
 813                                         err = -ECONNRESET;
 814                                         break;
 815                                 }
 816                                 err = vnet_rx(port, &msgbuf, &npkts, budget);
 817                                 if (npkts >= budget)
 818                                         break;
 819                                 if (npkts == 0)
 820                                         break;
 821                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
 822                                 err = vnet_ack(port, &msgbuf);
 823                                 if (err > 0)
 824                                         tx_wakeup |= err;
 825                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
 826                                 err = vnet_nack(port, &msgbuf);
 827                         }
 828                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
 829                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
 830                                 err = handle_mcast(port, &msgbuf);
 831                         else
 832                                 err = vio_control_pkt_engine(vio, &msgbuf);
 833                         if (err)
 834                                 break;
 835                 } else {
 836                         err = vnet_handle_unknown(port, &msgbuf);
 837                 }
 838                 if (err == -ECONNRESET)
 839                         break;
 840         }
 841         if (unlikely(tx_wakeup && err != -ECONNRESET))
 842                 maybe_tx_wakeup(port);
 843         return npkts;
 844 }
 845
 846 int sunvnet_poll_common(struct napi_struct *napi, int budget)
 847 {
 848         struct vnet_port *port = container_of(napi, struct vnet_port, napi);
 849         struct vio_driver_state *vio = &port->vio;
 850         int processed = vnet_event_napi(port, budget);
 851
 852         if (processed < budget) {
 853                 napi_complete(napi);
 854                 port->rx_event &= ~LDC_EVENT_DATA_READY;
 855                 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
 856         }
 857         return processed;
 858 }
 859 EXPORT_SYMBOL_GPL(sunvnet_poll_common);
 860
 861 void sunvnet_event_common(void *arg, int event)
 862 {
 863         struct vnet_port *port = arg;
 864         struct vio_driver_state *vio = &port->vio;
 865
 866         port->rx_event |= event;
 867         vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
 868         napi_schedule(&port->napi);
 869 }
 870 EXPORT_SYMBOL_GPL(sunvnet_event_common);
 871
 872 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
 873 {
 874         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 875         struct vio_dring_data hdr = {
 876                 .tag = {
 877                         .type           = VIO_TYPE_DATA,
 878                         .stype          = VIO_SUBTYPE_INFO,
 879                         .stype_env      = VIO_DRING_DATA,
 880                         .sid            = vio_send_sid(&port->vio),
 881                 },
 882                 .dring_ident            = dr->ident,
 883                 .start_idx              = start,
 884                 .end_idx                = (u32)-1,
 885         };
 886         int err, delay;
 887         int retries = 0;
 888
 889         if (port->stop_rx) {
 890                 trace_vnet_tx_pending_stopped_ack(port->vio._local_sid,
 891                                                   port->vio._peer_sid,
 892                                                   port->stop_rx_idx, -1);
 893                 err = vnet_send_ack(port,
 894                                     &port->vio.drings[VIO_DRIVER_RX_RING],
 895                                     port->stop_rx_idx, -1,
 896                                     VIO_DRING_STOPPED);
 897                 if (err <= 0)
 898                         return err;
 899         }
 900
 901         hdr.seq = dr->snd_nxt;
 902         delay = 1;
 903         do {
 904                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
 905                 if (err > 0) {
 906                         dr->snd_nxt++;
 907                         break;
 908                 }
 909                 udelay(delay);
 910                 if ((delay <<= 1) > 128)
 911                         delay = 128;
 912                 if (retries++ > VNET_MAX_RETRIES)
 913                         break;
 914         } while (err == -EAGAIN);
 915         trace_vnet_tx_trigger(port->vio._local_sid,
 916                               port->vio._peer_sid, start, err);
 917
 918         return err;
 919 }
 920
 921 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
 922                                           unsigned *pending)
 923 {
 924         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
 925         struct sk_buff *skb = NULL;
 926         int i, txi;
 927
 928         *pending = 0;
 929
 930         txi = dr->prod;
 931         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
 932                 struct vio_net_desc *d;
 933
 934                 --txi;
 935                 if (txi < 0)
 936                         txi = VNET_TX_RING_SIZE - 1;
 937
 938                 d = vio_dring_entry(dr, txi);
 939
 940                 if (d->hdr.state == VIO_DESC_READY) {
 941                         (*pending)++;
 942                         continue;
 943                 }
 944                 if (port->tx_bufs[txi].skb) {
 945                         if (d->hdr.state != VIO_DESC_DONE)
 946                                 pr_notice("invalid ring buffer state %d\n",
 947                                           d->hdr.state);
 948                         BUG_ON(port->tx_bufs[txi].skb->next);
 949
 950                         port->tx_bufs[txi].skb->next = skb;
 951                         skb = port->tx_bufs[txi].skb;
 952                         port->tx_bufs[txi].skb = NULL;
 953
 954                         ldc_unmap(port->vio.lp,
 955                                   port->tx_bufs[txi].cookies,
 956                                   port->tx_bufs[txi].ncookies);
 957                 } else if (d->hdr.state == VIO_DESC_FREE) {
 958                         break;
 959                 }
 960                 d->hdr.state = VIO_DESC_FREE;
 961         }
 962         return skb;
 963 }
 964
 965 static inline void vnet_free_skbs(struct sk_buff *skb)
 966 {
 967         struct sk_buff *next;
 968
 969         while (skb) {
 970                 next = skb->next;
 971                 skb->next = NULL;
 972                 dev_kfree_skb(skb);
 973                 skb = next;
 974         }
 975 }
 976
 977 void sunvnet_clean_timer_expire_common(unsigned long port0)
 978 {
 979         struct vnet_port *port = (struct vnet_port *)port0;
 980         struct sk_buff *freeskbs;
 981         unsigned pending;
 982
 983         netif_tx_lock(VNET_PORT_TO_NET_DEVICE(port));
 984         freeskbs = vnet_clean_tx_ring(port, &pending);
 985         netif_tx_unlock(VNET_PORT_TO_NET_DEVICE(port));
 986
 987         vnet_free_skbs(freeskbs);
 988
 989         if (pending)
 990                 (void)mod_timer(&port->clean_timer,
 991                                 jiffies + VNET_CLEAN_TIMEOUT);
 992          else
 993                 del_timer(&port->clean_timer);
 994 }
 995 EXPORT_SYMBOL_GPL(sunvnet_clean_timer_expire_common);
 996
 997 static inline int vnet_skb_map(struct ldc_channel *lp, struct sk_buff *skb,
 998                                struct ldc_trans_cookie *cookies, int ncookies,
 999                                unsigned int map_perm)
1000 {
1001         int i, nc, err, blen;
1002
1003         /* header */
1004         blen = skb_headlen(skb);
1005         if (blen < ETH_ZLEN)
1006                 blen = ETH_ZLEN;
1007         blen += VNET_PACKET_SKIP;
1008         blen += 8 - (blen & 7);
1009
1010         err = ldc_map_single(lp, skb->data - VNET_PACKET_SKIP, blen, cookies,
1011                              ncookies, map_perm);
1012         if (err < 0)
1013                 return err;
1014         nc = err;
1015
1016         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1017                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1018                 u8 *vaddr;
1019
1020                 if (nc < ncookies) {
1021                         vaddr = kmap_atomic(skb_frag_page(f));
1022                         blen = skb_frag_size(f);
1023                         blen += 8 - (blen & 7);
1024                         err = ldc_map_single(lp, vaddr + f->page_offset,
1025                                              blen, cookies + nc, ncookies - nc,
1026                                              map_perm);
1027                         kunmap_atomic(vaddr);
1028                 } else {
1029                         err = -EMSGSIZE;
1030                 }
1031
1032                 if (err < 0) {
1033                         ldc_unmap(lp, cookies, nc);
1034                         return err;
1035                 }
1036                 nc += err;
1037         }
1038         return nc;
1039 }
1040
1041 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, int ncookies)
1042 {
1043         struct sk_buff *nskb;
1044         int i, len, pad, docopy;
1045
1046         len = skb->len;
1047         pad = 0;
1048         if (len < ETH_ZLEN) {
1049                 pad += ETH_ZLEN - skb->len;
1050                 len += pad;
1051         }
1052         len += VNET_PACKET_SKIP;
1053         pad += 8 - (len & 7);
1054
1055         /* make sure we have enough cookies and alignment in every frag */
1056         docopy = skb_shinfo(skb)->nr_frags >= ncookies;
1057         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1058                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1059
1060                 docopy |= f->page_offset & 7;
1061         }
1062         if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
1063             skb_tailroom(skb) < pad ||
1064             skb_headroom(skb) < VNET_PACKET_SKIP || docopy) {
1065                 int start = 0, offset;
1066                 __wsum csum;
1067
1068                 len = skb->len > ETH_ZLEN ? skb->len : ETH_ZLEN;
1069                 nskb = alloc_and_align_skb(skb->dev, len);
1070                 if (!nskb) {
1071                         dev_kfree_skb(skb);
1072                         return NULL;
1073                 }
1074                 skb_reserve(nskb, VNET_PACKET_SKIP);
1075
1076                 nskb->protocol = skb->protocol;
1077                 offset = skb_mac_header(skb) - skb->data;
1078                 skb_set_mac_header(nskb, offset);
1079                 offset = skb_network_header(skb) - skb->data;
1080                 skb_set_network_header(nskb, offset);
1081                 offset = skb_transport_header(skb) - skb->data;
1082                 skb_set_transport_header(nskb, offset);
1083
1084                 offset = 0;
1085                 nskb->csum_offset = skb->csum_offset;
1086                 nskb->ip_summed = skb->ip_summed;
1087
1088                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1089                         start = skb_checksum_start_offset(skb);
1090                 if (start) {
1091                         struct iphdr *iph = ip_hdr(nskb);
1092                         int offset = start + nskb->csum_offset;
1093
1094                         if (skb_copy_bits(skb, 0, nskb->data, start)) {
1095                                 dev_kfree_skb(nskb);
1096                                 dev_kfree_skb(skb);
1097                                 return NULL;
1098                         }
1099                         *(__sum16 *)(skb->data + offset) = 0;
1100                         csum = skb_copy_and_csum_bits(skb, start,
1101                                                       nskb->data + start,
1102                                                       skb->len - start, 0);
1103                         if (iph->protocol == IPPROTO_TCP ||
1104                             iph->protocol == IPPROTO_UDP) {
1105                                 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
1106                                                          skb->len - start,
1107                                                          iph->protocol, csum);
1108                         }
1109                         *(__sum16 *)(nskb->data + offset) = csum;
1110
1111                         nskb->ip_summed = CHECKSUM_NONE;
1112                 } else if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
1113                         dev_kfree_skb(nskb);
1114                         dev_kfree_skb(skb);
1115                         return NULL;
1116                 }
1117                 (void)skb_put(nskb, skb->len);
1118                 if (skb_is_gso(skb)) {
1119                         skb_shinfo(nskb)->gso_size = skb_shinfo(skb)->gso_size;
1120                         skb_shinfo(nskb)->gso_type = skb_shinfo(skb)->gso_type;
1121                 }
1122                 nskb->queue_mapping = skb->queue_mapping;
1123                 dev_kfree_skb(skb);
1124                 skb = nskb;
1125         }
1126         return skb;
1127 }
1128
1129 static int vnet_handle_offloads(struct vnet_port *port, struct sk_buff *skb,
1130                                 struct vnet_port *(*vnet_tx_port)
1131                                 (struct sk_buff *, struct net_device *))
1132 {
1133         struct net_device *dev = VNET_PORT_TO_NET_DEVICE(port);
1134         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1135         struct sk_buff *segs;
1136         int maclen, datalen;
1137         int status;
1138         int gso_size, gso_type, gso_segs;
1139         int hlen = skb_transport_header(skb) - skb_mac_header(skb);
1140         int proto = IPPROTO_IP;
1141
1142         if (skb->protocol == htons(ETH_P_IP))
1143                 proto = ip_hdr(skb)->protocol;
1144         else if (skb->protocol == htons(ETH_P_IPV6))
1145                 proto = ipv6_hdr(skb)->nexthdr;
1146
1147         if (proto == IPPROTO_TCP) {
1148                 hlen += tcp_hdr(skb)->doff * 4;
1149         } else if (proto == IPPROTO_UDP) {
1150                 hlen += sizeof(struct udphdr);
1151         } else {
1152                 pr_err("vnet_handle_offloads GSO with unknown transport "
1153                        "protocol %d tproto %d\n", skb->protocol, proto);
1154                 hlen = 128; /* XXX */
1155         }
1156         datalen = port->tsolen - hlen;
1157
1158         gso_size = skb_shinfo(skb)->gso_size;
1159         gso_type = skb_shinfo(skb)->gso_type;
1160         gso_segs = skb_shinfo(skb)->gso_segs;
1161
1162         if (port->tso && gso_size < datalen)
1163                 gso_segs = DIV_ROUND_UP(skb->len - hlen, datalen);
1164
1165         if (unlikely(vnet_tx_dring_avail(dr) < gso_segs)) {
1166                 struct netdev_queue *txq;
1167
1168                 txq  = netdev_get_tx_queue(dev, port->q_index);
1169                 netif_tx_stop_queue(txq);
1170                 if (vnet_tx_dring_avail(dr) < skb_shinfo(skb)->gso_segs)
1171                         return NETDEV_TX_BUSY;
1172                 netif_tx_wake_queue(txq);
1173         }
1174
1175         maclen = skb_network_header(skb) - skb_mac_header(skb);
1176         skb_pull(skb, maclen);
1177
1178         if (port->tso && gso_size < datalen) {
1179                 if (skb_unclone(skb, GFP_ATOMIC))
1180                         goto out_dropped;
1181
1182                 /* segment to TSO size */
1183                 skb_shinfo(skb)->gso_size = datalen;
1184                 skb_shinfo(skb)->gso_segs = gso_segs;
1185         }
1186         segs = skb_gso_segment(skb, dev->features & ~NETIF_F_TSO);
1187         if (IS_ERR(segs))
1188                 goto out_dropped;
1189
1190         skb_push(skb, maclen);
1191         skb_reset_mac_header(skb);
1192
1193         status = 0;
1194         while (segs) {
1195                 struct sk_buff *curr = segs;
1196
1197                 segs = segs->next;
1198                 curr->next = NULL;
1199                 if (port->tso && curr->len > dev->mtu) {
1200                         skb_shinfo(curr)->gso_size = gso_size;
1201                         skb_shinfo(curr)->gso_type = gso_type;
1202                         skb_shinfo(curr)->gso_segs =
1203                                 DIV_ROUND_UP(curr->len - hlen, gso_size);
1204                 } else {
1205                         skb_shinfo(curr)->gso_size = 0;
1206                 }
1207
1208                 skb_push(curr, maclen);
1209                 skb_reset_mac_header(curr);
1210                 memcpy(skb_mac_header(curr), skb_mac_header(skb),
1211                        maclen);
1212                 curr->csum_start = skb_transport_header(curr) - curr->head;
1213                 if (ip_hdr(curr)->protocol == IPPROTO_TCP)
1214                         curr->csum_offset = offsetof(struct tcphdr, check);
1215                 else if (ip_hdr(curr)->protocol == IPPROTO_UDP)
1216                         curr->csum_offset = offsetof(struct udphdr, check);
1217
1218                 if (!(status & NETDEV_TX_MASK))
1219                         status = sunvnet_start_xmit_common(curr, dev,
1220                                                            vnet_tx_port);
1221                 if (status & NETDEV_TX_MASK)
1222                         dev_kfree_skb_any(curr);
1223         }
1224
1225         if (!(status & NETDEV_TX_MASK))
1226                 dev_kfree_skb_any(skb);
1227         return status;
1228 out_dropped:
1229         dev->stats.tx_dropped++;
1230         dev_kfree_skb_any(skb);
1231         return NETDEV_TX_OK;
1232 }
1233
1234 int sunvnet_start_xmit_common(struct sk_buff *skb, struct net_device *dev,
1235                               struct vnet_port *(*vnet_tx_port)
1236                               (struct sk_buff *, struct net_device *))
1237 {
1238         struct vnet_port *port = NULL;
1239         struct vio_dring_state *dr;
1240         struct vio_net_desc *d;
1241         unsigned int len;
1242         struct sk_buff *freeskbs = NULL;
1243         int i, err, txi;
1244         unsigned pending = 0;
1245         struct netdev_queue *txq;
1246
1247         rcu_read_lock();
1248         port = vnet_tx_port(skb, dev);
1249         if (unlikely(!port)) {
1250                 rcu_read_unlock();
1251                 goto out_dropped;
1252         }
1253
1254         if (skb_is_gso(skb) && skb->len > port->tsolen) {
1255                 err = vnet_handle_offloads(port, skb, vnet_tx_port);
1256                 rcu_read_unlock();
1257                 return err;
1258         }
1259
1260         if (!skb_is_gso(skb) && skb->len > port->rmtu) {
1261                 unsigned long localmtu = port->rmtu - ETH_HLEN;
1262
1263                 if (vio_version_after_eq(&port->vio, 1, 3))
1264                         localmtu -= VLAN_HLEN;
1265
1266                 if (skb->protocol == htons(ETH_P_IP))
1267                         icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
1268                                       htonl(localmtu));
1269 #if IS_ENABLED(CONFIG_IPV6)
1270                 else if (skb->protocol == htons(ETH_P_IPV6))
1271                         icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1272 #endif
1273                 goto out_dropped;
1274         }
1275
1276         skb = vnet_skb_shape(skb, 2);
1277
1278         if (unlikely(!skb))
1279                 goto out_dropped;
1280
1281         if (skb->ip_summed == CHECKSUM_PARTIAL)
1282                 vnet_fullcsum(skb);
1283
1284         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1285         i = skb_get_queue_mapping(skb);
1286         txq = netdev_get_tx_queue(dev, i);
1287         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1288                 if (!netif_tx_queue_stopped(txq)) {
1289                         netif_tx_stop_queue(txq);
1290
1291                         /* This is a hard error, log it. */
1292                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1293                         dev->stats.tx_errors++;
1294                 }
1295                 rcu_read_unlock();
1296                 return NETDEV_TX_BUSY;
1297         }
1298
1299         d = vio_dring_cur(dr);
1300
1301         txi = dr->prod;
1302
1303         freeskbs = vnet_clean_tx_ring(port, &pending);
1304
1305         BUG_ON(port->tx_bufs[txi].skb);
1306
1307         len = skb->len;
1308         if (len < ETH_ZLEN)
1309                 len = ETH_ZLEN;
1310
1311         err = vnet_skb_map(port->vio.lp, skb, port->tx_bufs[txi].cookies, 2,
1312                            (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1313         if (err < 0) {
1314                 netdev_info(dev, "tx buffer map error %d\n", err);
1315                 goto out_dropped;
1316         }
1317
1318         port->tx_bufs[txi].skb = skb;
1319         skb = NULL;
1320         port->tx_bufs[txi].ncookies = err;
1321
1322         /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1323          * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1324          * the protocol itself does not require it as long as the peer
1325          * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1326          *
1327          * An ACK for every packet in the ring is expensive as the
1328          * sending of LDC messages is slow and affects performance.
1329          */
1330         d->hdr.ack = VIO_ACK_DISABLE;
1331         d->size = len;
1332         d->ncookies = port->tx_bufs[txi].ncookies;
1333         for (i = 0; i < d->ncookies; i++)
1334                 d->cookies[i] = port->tx_bufs[txi].cookies[i];
1335         if (vio_version_after_eq(&port->vio, 1, 7)) {
1336                 struct vio_net_dext *dext = vio_net_ext(d);
1337
1338                 memset(dext, 0, sizeof(*dext));
1339                 if (skb_is_gso(port->tx_bufs[txi].skb)) {
1340                         dext->ipv4_lso_mss = skb_shinfo(port->tx_bufs[txi].skb)
1341                                              ->gso_size;
1342                         dext->flags |= VNET_PKT_IPV4_LSO;
1343                 }
1344                 if (vio_version_after_eq(&port->vio, 1, 8) &&
1345                     !port->switch_port) {
1346                         dext->flags |= VNET_PKT_HCK_IPV4_HDRCKSUM_OK;
1347                         dext->flags |= VNET_PKT_HCK_FULLCKSUM_OK;
1348                 }
1349         }
1350
1351         /* This has to be a non-SMP write barrier because we are writing
1352          * to memory which is shared with the peer LDOM.
1353          */
1354         dma_wmb();
1355
1356         d->hdr.state = VIO_DESC_READY;
1357
1358         /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1359          * to notify the consumer that some descriptors are READY.
1360          * After that "start" trigger, no additional triggers are needed until
1361          * a DRING_STOPPED is received from the consumer. The dr->cons field
1362          * (set up by vnet_ack()) has the value of the next dring index
1363          * that has not yet been ack-ed. We send a "start" trigger here
1364          * if, and only if, start_cons is true (reset it afterward). Conversely,
1365          * vnet_ack() should check if the dring corresponding to cons
1366          * is marked READY, but start_cons was false.
1367          * If so, vnet_ack() should send out the missed "start" trigger.
1368          *
1369          * Note that the dma_wmb() above makes sure the cookies et al. are
1370          * not globally visible before the VIO_DESC_READY, and that the
1371          * stores are ordered correctly by the compiler. The consumer will
1372          * not proceed until the VIO_DESC_READY is visible assuring that
1373          * the consumer does not observe anything related to descriptors
1374          * out of order. The HV trap from the LDC start trigger is the
1375          * producer to consumer announcement that work is available to the
1376          * consumer
1377          */
1378         if (!port->start_cons) { /* previous trigger suffices */
1379                 trace_vnet_skip_tx_trigger(port->vio._local_sid,
1380                                            port->vio._peer_sid, dr->cons);
1381                 goto ldc_start_done;
1382         }
1383
1384         err = __vnet_tx_trigger(port, dr->cons);
1385         if (unlikely(err < 0)) {
1386                 netdev_info(dev, "TX trigger error %d\n", err);
1387                 d->hdr.state = VIO_DESC_FREE;
1388                 skb = port->tx_bufs[txi].skb;
1389                 port->tx_bufs[txi].skb = NULL;
1390                 dev->stats.tx_carrier_errors++;
1391                 goto out_dropped;
1392         }
1393
1394 ldc_start_done:
1395         port->start_cons = false;
1396
1397         dev->stats.tx_packets++;
1398         dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1399
1400         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1401         if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1402                 netif_tx_stop_queue(txq);
1403                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1404                         netif_tx_wake_queue(txq);
1405         }
1406
1407         (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1408         rcu_read_unlock();
1409
1410         vnet_free_skbs(freeskbs);
1411
1412         return NETDEV_TX_OK;
1413
1414 out_dropped:
1415         if (pending)
1416                 (void)mod_timer(&port->clean_timer,
1417                                 jiffies + VNET_CLEAN_TIMEOUT);
1418         else if (port)
1419                 del_timer(&port->clean_timer);
1420         if (port)
1421                 rcu_read_unlock();
1422         if (skb)
1423                 dev_kfree_skb(skb);
1424         vnet_free_skbs(freeskbs);
1425         dev->stats.tx_dropped++;
1426         return NETDEV_TX_OK;
1427 }
1428 EXPORT_SYMBOL_GPL(sunvnet_start_xmit_common);
1429
1430 void sunvnet_tx_timeout_common(struct net_device *dev)
1431 {
1432         /* XXX Implement me XXX */
1433 }
1434 EXPORT_SYMBOL_GPL(sunvnet_tx_timeout_common);
1435
1436 int sunvnet_open_common(struct net_device *dev)
1437 {
1438         netif_carrier_on(dev);
1439         netif_tx_start_all_queues(dev);
1440
1441         return 0;
1442 }
1443 EXPORT_SYMBOL_GPL(sunvnet_open_common);
1444
1445 int sunvnet_close_common(struct net_device *dev)
1446 {
1447         netif_tx_stop_all_queues(dev);
1448         netif_carrier_off(dev);
1449
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(sunvnet_close_common);
1453
1454 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1455 {
1456         struct vnet_mcast_entry *m;
1457
1458         for (m = vp->mcast_list; m; m = m->next) {
1459                 if (ether_addr_equal(m->addr, addr))
1460                         return m;
1461         }
1462         return NULL;
1463 }
1464
1465 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1466 {
1467         struct netdev_hw_addr *ha;
1468
1469         netdev_for_each_mc_addr(ha, dev) {
1470                 struct vnet_mcast_entry *m;
1471
1472                 m = __vnet_mc_find(vp, ha->addr);
1473                 if (m) {
1474                         m->hit = 1;
1475                         continue;
1476                 }
1477
1478                 if (!m) {
1479                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
1480                         if (!m)
1481                                 continue;
1482                         memcpy(m->addr, ha->addr, ETH_ALEN);
1483                         m->hit = 1;
1484
1485                         m->next = vp->mcast_list;
1486                         vp->mcast_list = m;
1487                 }
1488         }
1489 }
1490
1491 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1492 {
1493         struct vio_net_mcast_info info;
1494         struct vnet_mcast_entry *m, **pp;
1495         int n_addrs;
1496
1497         memset(&info, 0, sizeof(info));
1498
1499         info.tag.type = VIO_TYPE_CTRL;
1500         info.tag.stype = VIO_SUBTYPE_INFO;
1501         info.tag.stype_env = VNET_MCAST_INFO;
1502         info.tag.sid = vio_send_sid(&port->vio);
1503         info.set = 1;
1504
1505         n_addrs = 0;
1506         for (m = vp->mcast_list; m; m = m->next) {
1507                 if (m->sent)
1508                         continue;
1509                 m->sent = 1;
1510                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1511                        m->addr, ETH_ALEN);
1512                 if (++n_addrs == VNET_NUM_MCAST) {
1513                         info.count = n_addrs;
1514
1515                         (void)vio_ldc_send(&port->vio, &info,
1516                                            sizeof(info));
1517                         n_addrs = 0;
1518                 }
1519         }
1520         if (n_addrs) {
1521                 info.count = n_addrs;
1522                 (void)vio_ldc_send(&port->vio, &info, sizeof(info));
1523         }
1524
1525         info.set = 0;
1526
1527         n_addrs = 0;
1528         pp = &vp->mcast_list;
1529         while ((m = *pp) != NULL) {
1530                 if (m->hit) {
1531                         m->hit = 0;
1532                         pp = &m->next;
1533                         continue;
1534                 }
1535
1536                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1537                        m->addr, ETH_ALEN);
1538                 if (++n_addrs == VNET_NUM_MCAST) {
1539                         info.count = n_addrs;
1540                         (void)vio_ldc_send(&port->vio, &info,
1541                                            sizeof(info));
1542                         n_addrs = 0;
1543                 }
1544
1545                 *pp = m->next;
1546                 kfree(m);
1547         }
1548         if (n_addrs) {
1549                 info.count = n_addrs;
1550                 (void)vio_ldc_send(&port->vio, &info, sizeof(info));
1551         }
1552 }
1553
1554 void sunvnet_set_rx_mode_common(struct net_device *dev, struct vnet *vp)
1555 {
1556         struct vnet_port *port;
1557
1558         rcu_read_lock();
1559         list_for_each_entry_rcu(port, &vp->port_list, list) {
1560                 if (port->switch_port) {
1561                         __update_mc_list(vp, dev);
1562                         __send_mc_list(vp, port);
1563                         break;
1564                 }
1565         }
1566         rcu_read_unlock();
1567 }
1568 EXPORT_SYMBOL_GPL(sunvnet_set_rx_mode_common);
1569
1570 int sunvnet_change_mtu_common(struct net_device *dev, int new_mtu)
1571 {
1572         if (new_mtu < 68 || new_mtu > 65535)
1573                 return -EINVAL;
1574
1575         dev->mtu = new_mtu;
1576         return 0;
1577 }
1578 EXPORT_SYMBOL_GPL(sunvnet_change_mtu_common);
1579
1580 int sunvnet_set_mac_addr_common(struct net_device *dev, void *p)
1581 {
1582         return -EINVAL;
1583 }
1584 EXPORT_SYMBOL_GPL(sunvnet_set_mac_addr_common);
1585
1586 void sunvnet_port_free_tx_bufs_common(struct vnet_port *port)
1587 {
1588         struct vio_dring_state *dr;
1589         int i;
1590
1591         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1592
1593         if (!dr->base)
1594                 return;
1595
1596         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1597                 struct vio_net_desc *d;
1598                 void *skb = port->tx_bufs[i].skb;
1599
1600                 if (!skb)
1601                         continue;
1602
1603                 d = vio_dring_entry(dr, i);
1604
1605                 ldc_unmap(port->vio.lp,
1606                           port->tx_bufs[i].cookies,
1607                           port->tx_bufs[i].ncookies);
1608                 dev_kfree_skb(skb);
1609                 port->tx_bufs[i].skb = NULL;
1610                 d->hdr.state = VIO_DESC_FREE;
1611         }
1612         ldc_free_exp_dring(port->vio.lp, dr->base,
1613                            (dr->entry_size * dr->num_entries),
1614                            dr->cookies, dr->ncookies);
1615         dr->base = NULL;
1616         dr->entry_size = 0;
1617         dr->num_entries = 0;
1618         dr->pending = 0;
1619         dr->ncookies = 0;
1620 }
1621 EXPORT_SYMBOL_GPL(sunvnet_port_free_tx_bufs_common);
1622
1623 static void vnet_port_reset(struct vnet_port *port)
1624 {
1625         del_timer(&port->clean_timer);
1626         sunvnet_port_free_tx_bufs_common(port);
1627         port->rmtu = 0;
1628         port->tso = true;
1629         port->tsolen = 0;
1630 }
1631
1632 static int vnet_port_alloc_tx_ring(struct vnet_port *port)
1633 {
1634         struct vio_dring_state *dr;
1635         unsigned long len, elen;
1636         int i, err, ncookies;
1637         void *dring;
1638
1639         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1640
1641         elen = sizeof(struct vio_net_desc) +
1642                sizeof(struct ldc_trans_cookie) * 2;
1643         if (vio_version_after_eq(&port->vio, 1, 7))
1644                 elen += sizeof(struct vio_net_dext);
1645         len = VNET_TX_RING_SIZE * elen;
1646
1647         ncookies = VIO_MAX_RING_COOKIES;
1648         dring = ldc_alloc_exp_dring(port->vio.lp, len,
1649                                     dr->cookies, &ncookies,
1650                                     (LDC_MAP_SHADOW |
1651                                      LDC_MAP_DIRECT |
1652                                      LDC_MAP_RW));
1653         if (IS_ERR(dring)) {
1654                 err = PTR_ERR(dring);
1655                 goto err_out;
1656         }
1657
1658         dr->base = dring;
1659         dr->entry_size = elen;
1660         dr->num_entries = VNET_TX_RING_SIZE;
1661         dr->prod = 0;
1662         dr->cons = 0;
1663         port->start_cons  = true; /* need an initial trigger */
1664         dr->pending = VNET_TX_RING_SIZE;
1665         dr->ncookies = ncookies;
1666
1667         for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1668                 struct vio_net_desc *d;
1669
1670                 d = vio_dring_entry(dr, i);
1671                 d->hdr.state = VIO_DESC_FREE;
1672         }
1673         return 0;
1674
1675 err_out:
1676         sunvnet_port_free_tx_bufs_common(port);
1677
1678         return err;
1679 }
1680
1681 #ifdef CONFIG_NET_POLL_CONTROLLER
1682 void sunvnet_poll_controller_common(struct net_device *dev, struct vnet *vp)
1683 {
1684         struct vnet_port *port;
1685         unsigned long flags;
1686
1687         spin_lock_irqsave(&vp->lock, flags);
1688         if (!list_empty(&vp->port_list)) {
1689                 port = list_entry(vp->port_list.next, struct vnet_port, list);
1690                 napi_schedule(&port->napi);
1691         }
1692         spin_unlock_irqrestore(&vp->lock, flags);
1693 }
1694 EXPORT_SYMBOL_GPL(sunvnet_poll_controller_common);
1695 #endif
1696
1697 void sunvnet_port_add_txq_common(struct vnet_port *port)
1698 {
1699         struct vnet *vp = port->vp;
1700         int n;
1701
1702         n = vp->nports++;
1703         n = n & (VNET_MAX_TXQS - 1);
1704         port->q_index = n;
1705         netif_tx_wake_queue(netdev_get_tx_queue(VNET_PORT_TO_NET_DEVICE(port),
1706                                                 port->q_index));
1707 }
1708 EXPORT_SYMBOL_GPL(sunvnet_port_add_txq_common);
1709
1710 void sunvnet_port_rm_txq_common(struct vnet_port *port)
1711 {
1712         port->vp->nports--;
1713         netif_tx_stop_queue(netdev_get_tx_queue(VNET_PORT_TO_NET_DEVICE(port),
1714                                                 port->q_index));
1715 }
1716 EXPORT_SYMBOL_GPL(sunvnet_port_rm_txq_common);