drivers/net/iseries_veth.c

   1 /* File veth.c created by Kyle A. Lucke on Mon Aug  7 2000. */
   2 /*
   3  * IBM eServer iSeries Virtual Ethernet Device Driver
   4  * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
   5  * Substantially cleaned up by:
   6  * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License as
  10  * published by the Free Software Foundation; either version 2 of the
  11  * License, or (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
  21  * USA
  22  *
  23  *
  24  * This module implements the virtual ethernet device for iSeries LPAR
  25  * Linux.  It uses hypervisor message passing to implement an
  26  * ethernet-like network device communicating between partitions on
  27  * the iSeries.
  28  *
  29  * The iSeries LPAR hypervisor currently allows for up to 16 different
  30  * virtual ethernets.  These are all dynamically configurable on
  31  * OS/400 partitions, but dynamic configuration is not supported under
  32  * Linux yet.  An ethXX network device will be created for each
  33  * virtual ethernet this partition is connected to.
  34  *
  35  * - This driver is responsible for routing packets to and from other
  36  *   partitions.  The MAC addresses used by the virtual ethernets
  37  *   contains meaning and must not be modified.
  38  *
  39  * - Having 2 virtual ethernets to the same remote partition DOES NOT
  40  *   double the available bandwidth.  The 2 devices will share the
  41  *   available hypervisor bandwidth.
  42  *
  43  * - If you send a packet to your own mac address, it will just be
  44  *   dropped, you won't get it on the receive side.
  45  *
  46  * - Multicast is implemented by sending the frame frame to every
  47  *   other partition.  It is the responsibility of the receiving
  48  *   partition to filter the addresses desired.
  49  *
  50  * Tunable parameters:
  51  *
  52  * VETH_NUMBUFFERS: This compile time option defaults to 120.  It
  53  * controls how much memory Linux will allocate per remote partition
  54  * it is communicating with.  It can be thought of as the maximum
  55  * number of packets outstanding to a remote partition at a time.
  56  */
  57
  58 #include <linux/config.h>
  59 #include <linux/module.h>
  60 #include <linux/version.h>
  61 #include <linux/types.h>
  62 #include <linux/errno.h>
  63 #include <linux/ioport.h>
  64 #include <linux/kernel.h>
  65 #include <linux/netdevice.h>
  66 #include <linux/etherdevice.h>
  67 #include <linux/skbuff.h>
  68 #include <linux/init.h>
  69 #include <linux/delay.h>
  70 #include <linux/mm.h>
  71 #include <linux/ethtool.h>
  72 #include <asm/iSeries/mf.h>
  73 #include <asm/iSeries/iSeries_pci.h>
  74 #include <asm/uaccess.h>
  75
  76 #include <asm/iSeries/HvLpConfig.h>
  77 #include <asm/iSeries/HvTypes.h>
  78 #include <asm/iSeries/HvLpEvent.h>
  79 #include <asm/iommu.h>
  80 #include <asm/vio.h>
  81
  82 #include "iseries_veth.h"
  83
  84 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  85 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  86 MODULE_LICENSE("GPL");
  87
  88 #define VETH_NUMBUFFERS         (120)
  89 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
  90 #define VETH_MAX_MCAST          (12)
  91
  92 #define VETH_MAX_MTU            (9000)
  93
  94 #if VETH_NUMBUFFERS < 10
  95 #define ACK_THRESHOLD           (1)
  96 #elif VETH_NUMBUFFERS < 20
  97 #define ACK_THRESHOLD           (4)
  98 #elif VETH_NUMBUFFERS < 40
  99 #define ACK_THRESHOLD           (10)
 100 #else
 101 #define ACK_THRESHOLD           (20)
 102 #endif
 103
 104 #define VETH_STATE_SHUTDOWN     (0x0001)
 105 #define VETH_STATE_OPEN         (0x0002)
 106 #define VETH_STATE_RESET        (0x0004)
 107 #define VETH_STATE_SENTMON      (0x0008)
 108 #define VETH_STATE_SENTCAPS     (0x0010)
 109 #define VETH_STATE_GOTCAPACK    (0x0020)
 110 #define VETH_STATE_GOTCAPS      (0x0040)
 111 #define VETH_STATE_SENTCAPACK   (0x0080)
 112 #define VETH_STATE_READY        (0x0100)
 113
 114 struct veth_msg {
 115         struct veth_msg *next;
 116         struct VethFramesData data;
 117         int token;
 118         unsigned long in_use;
 119         struct sk_buff *skb;
 120         struct device *dev;
 121 };
 122
 123 struct veth_lpar_connection {
 124         HvLpIndex remote_lp;
 125         struct work_struct statemachine_wq;
 126         struct veth_msg *msgs;
 127         int num_events;
 128         struct VethCapData local_caps;
 129
 130         struct timer_list ack_timer;
 131
 132         spinlock_t lock;
 133         unsigned long state;
 134         HvLpInstanceId src_inst;
 135         HvLpInstanceId dst_inst;
 136         struct VethLpEvent cap_event, cap_ack_event;
 137         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 138         u32 num_pending_acks;
 139
 140         int num_ack_events;
 141         struct VethCapData remote_caps;
 142         u32 ack_timeout;
 143
 144         spinlock_t msg_stack_lock;
 145         struct veth_msg *msg_stack_head;
 146 };
 147
 148 struct veth_port {
 149         struct device *dev;
 150         struct net_device_stats stats;
 151         u64 mac_addr;
 152         HvLpIndexMap lpar_map;
 153
 154         spinlock_t pending_gate;
 155         struct sk_buff *pending_skb;
 156         HvLpIndexMap pending_lpmask;
 157
 158         rwlock_t mcast_gate;
 159         int promiscuous;
 160         int all_mcast;
 161         int num_mcast;
 162         u64 mcast_addr[VETH_MAX_MCAST];
 163 };
 164
 165 static HvLpIndex this_lp;
 166 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
 167 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
 168
 169 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
 170 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
 171 static void veth_flush_pending(struct veth_lpar_connection *cnx);
 172 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
 173 static void veth_timed_ack(unsigned long connectionPtr);
 174
 175 /*
 176  * Utility functions
 177  */
 178
 179 #define veth_printk(prio, fmt, args...) \
 180         printk(prio "%s: " fmt, __FILE__, ## args)
 181
 182 #define veth_error(fmt, args...) \
 183         printk(KERN_ERR "(%s:%3.3d) ERROR: " fmt, __FILE__, __LINE__ , ## args)
 184
 185 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
 186                                    struct veth_msg *msg)
 187 {
 188         unsigned long flags;
 189
 190         spin_lock_irqsave(&cnx->msg_stack_lock, flags);
 191         msg->next = cnx->msg_stack_head;
 192         cnx->msg_stack_head = msg;
 193         spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
 194 }
 195
 196 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
 197 {
 198         unsigned long flags;
 199         struct veth_msg *msg;
 200
 201         spin_lock_irqsave(&cnx->msg_stack_lock, flags);
 202         msg = cnx->msg_stack_head;
 203         if (msg)
 204                 cnx->msg_stack_head = cnx->msg_stack_head->next;
 205         spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
 206         return msg;
 207 }
 208
 209 static inline HvLpEvent_Rc
 210 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
 211                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
 212                  u64 token,
 213                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
 214 {
 215         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
 216                                              HvLpEvent_Type_VirtualLan,
 217                                              subtype, ackind, acktype,
 218                                              cnx->src_inst,
 219                                              cnx->dst_inst,
 220                                              token, data1, data2, data3,
 221                                              data4, data5);
 222 }
 223
 224 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
 225                                            u16 subtype, u64 token, void *data)
 226 {
 227         u64 *p = (u64 *) data;
 228
 229         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
 230                                 HvLpEvent_AckType_ImmediateAck,
 231                                 token, p[0], p[1], p[2], p[3], p[4]);
 232 }
 233
 234 struct veth_allocation {
 235         struct completion c;
 236         int num;
 237 };
 238
 239 static void veth_complete_allocation(void *parm, int number)
 240 {
 241         struct veth_allocation *vc = (struct veth_allocation *)parm;
 242
 243         vc->num = number;
 244         complete(&vc->c);
 245 }
 246
 247 static int veth_allocate_events(HvLpIndex rlp, int number)
 248 {
 249         struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
 250
 251         mf_allocateLpEvents(rlp, HvLpEvent_Type_VirtualLan,
 252                             sizeof(struct VethLpEvent), number,
 253                             &veth_complete_allocation, &vc);
 254         wait_for_completion(&vc.c);
 255
 256         return vc.num;
 257 }
 258
 259 /*
 260  * LPAR connection code
 261  */
 262
 263 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 264 {
 265         schedule_work(&cnx->statemachine_wq);
 266 }
 267
 268 static void veth_take_cap(struct veth_lpar_connection *cnx,
 269                           struct VethLpEvent *event)
 270 {
 271         unsigned long flags;
 272
 273         spin_lock_irqsave(&cnx->lock, flags);
 274         /* Receiving caps may mean the other end has just come up, so
 275          * we need to reload the instance ID of the far end */
 276         cnx->dst_inst =
 277                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 278                                                   HvLpEvent_Type_VirtualLan);
 279
 280         if (cnx->state & VETH_STATE_GOTCAPS) {
 281                 veth_error("Received a second capabilities from lpar %d\n",
 282                            cnx->remote_lp);
 283                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 284                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 285         } else {
 286                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 287                 cnx->state |= VETH_STATE_GOTCAPS;
 288                 veth_kick_statemachine(cnx);
 289         }
 290         spin_unlock_irqrestore(&cnx->lock, flags);
 291 }
 292
 293 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 294                               struct VethLpEvent *event)
 295 {
 296         unsigned long flags;
 297
 298         spin_lock_irqsave(&cnx->lock, flags);
 299         if (cnx->state & VETH_STATE_GOTCAPACK) {
 300                 veth_error("Received a second capabilities ack from lpar %d\n",
 301                            cnx->remote_lp);
 302         } else {
 303                 memcpy(&cnx->cap_ack_event, event,
 304                        sizeof(&cnx->cap_ack_event));
 305                 cnx->state |= VETH_STATE_GOTCAPACK;
 306                 veth_kick_statemachine(cnx);
 307         }
 308         spin_unlock_irqrestore(&cnx->lock, flags);
 309 }
 310
 311 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 312                                   struct VethLpEvent *event)
 313 {
 314         unsigned long flags;
 315
 316         spin_lock_irqsave(&cnx->lock, flags);
 317         veth_printk(KERN_DEBUG, "Monitor ack returned for lpar %d\n",
 318                     cnx->remote_lp);
 319         cnx->state |= VETH_STATE_RESET;
 320         veth_kick_statemachine(cnx);
 321         spin_unlock_irqrestore(&cnx->lock, flags);
 322 }
 323
 324 static void veth_handle_ack(struct VethLpEvent *event)
 325 {
 326         HvLpIndex rlp = event->base_event.xTargetLp;
 327         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 328
 329         BUG_ON(! cnx);
 330
 331         switch (event->base_event.xSubtype) {
 332         case VethEventTypeCap:
 333                 veth_take_cap_ack(cnx, event);
 334                 break;
 335         case VethEventTypeMonitor:
 336                 veth_take_monitor_ack(cnx, event);
 337                 break;
 338         default:
 339                 veth_error("Unknown ack type %d from lpar %d\n",
 340                            event->base_event.xSubtype, rlp);
 341         };
 342 }
 343
 344 static void veth_handle_int(struct VethLpEvent *event)
 345 {
 346         HvLpIndex rlp = event->base_event.xSourceLp;
 347         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 348         unsigned long flags;
 349         int i;
 350
 351         BUG_ON(! cnx);
 352
 353         switch (event->base_event.xSubtype) {
 354         case VethEventTypeCap:
 355                 veth_take_cap(cnx, event);
 356                 break;
 357         case VethEventTypeMonitor:
 358                 /* do nothing... this'll hang out here til we're dead,
 359                  * and the hypervisor will return it for us. */
 360                 break;
 361         case VethEventTypeFramesAck:
 362                 spin_lock_irqsave(&cnx->lock, flags);
 363                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 364                         u16 msgnum = event->u.frames_ack_data.token[i];
 365
 366                         if (msgnum < VETH_NUMBUFFERS)
 367                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 368                 }
 369                 spin_unlock_irqrestore(&cnx->lock, flags);
 370                 veth_flush_pending(cnx);
 371                 break;
 372         case VethEventTypeFrames:
 373                 veth_receive(cnx, event);
 374                 break;
 375         default:
 376                 veth_error("Unknown interrupt type %d from lpar %d\n",
 377                            event->base_event.xSubtype, rlp);
 378         };
 379 }
 380
 381 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
 382 {
 383         struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
 384
 385         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
 386                 veth_handle_ack(veth_event);
 387         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
 388                 veth_handle_int(veth_event);
 389 }
 390
 391 static int veth_process_caps(struct veth_lpar_connection *cnx)
 392 {
 393         struct VethCapData *remote_caps = &cnx->remote_caps;
 394         int num_acks_needed;
 395
 396         /* Convert timer to jiffies */
 397         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 398
 399         if ( (remote_caps->num_buffers == 0)
 400              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
 401              || (remote_caps->ack_threshold == 0)
 402              || (cnx->ack_timeout == 0) ) {
 403                 veth_error("Received incompatible capabilities from lpar %d\n",
 404                            cnx->remote_lp);
 405                 return HvLpEvent_Rc_InvalidSubtypeData;
 406         }
 407
 408         num_acks_needed = (remote_caps->num_buffers
 409                            / remote_caps->ack_threshold) + 1;
 410
 411         /* FIXME: locking on num_ack_events? */
 412         if (cnx->num_ack_events < num_acks_needed) {
 413                 int num;
 414
 415                 num = veth_allocate_events(cnx->remote_lp,
 416                                            num_acks_needed-cnx->num_ack_events);
 417                 if (num > 0)
 418                         cnx->num_ack_events += num;
 419
 420                 if (cnx->num_ack_events < num_acks_needed) {
 421                         veth_error("Couldn't allocate enough ack events for lpar %d\n",
 422                                    cnx->remote_lp);
 423
 424                         return HvLpEvent_Rc_BufferNotAvailable;
 425                 }
 426         }
 427
 428
 429         return HvLpEvent_Rc_Good;
 430 }
 431
 432 /* FIXME: The gotos here are a bit dubious */
 433 static void veth_statemachine(void *p)
 434 {
 435         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
 436         int rlp = cnx->remote_lp;
 437         int rc;
 438
 439         spin_lock_irq(&cnx->lock);
 440
 441  restart:
 442         if (cnx->state & VETH_STATE_RESET) {
 443                 int i;
 444
 445                 del_timer(&cnx->ack_timer);
 446
 447                 if (cnx->state & VETH_STATE_OPEN)
 448                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 449                                                      HvLpEvent_Type_VirtualLan);
 450
 451                 /* reset ack data */
 452                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 453                 cnx->num_pending_acks = 0;
 454
 455                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 456                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 457                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 458                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 459
 460                 /* Clean up any leftover messages */
 461                 if (cnx->msgs)
 462                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 463                                 veth_recycle_msg(cnx, cnx->msgs + i);
 464                 spin_unlock_irq(&cnx->lock);
 465                 veth_flush_pending(cnx);
 466                 spin_lock_irq(&cnx->lock);
 467                 if (cnx->state & VETH_STATE_RESET)
 468                         goto restart;
 469         }
 470
 471         if (cnx->state & VETH_STATE_SHUTDOWN)
 472                 /* It's all over, do nothing */
 473                 goto out;
 474
 475         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 476                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 477                         goto cant_cope;
 478
 479                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 480                 cnx->src_inst =
 481                         HvCallEvent_getSourceLpInstanceId(rlp,
 482                                                           HvLpEvent_Type_VirtualLan);
 483                 cnx->dst_inst =
 484                         HvCallEvent_getTargetLpInstanceId(rlp,
 485                                                           HvLpEvent_Type_VirtualLan);
 486                 cnx->state |= VETH_STATE_OPEN;
 487         }
 488
 489         if ( (cnx->state & VETH_STATE_OPEN)
 490              && !(cnx->state & VETH_STATE_SENTMON) ) {
 491                 rc = veth_signalevent(cnx, VethEventTypeMonitor,
 492                                       HvLpEvent_AckInd_DoAck,
 493                                       HvLpEvent_AckType_DeferredAck,
 494                                       0, 0, 0, 0, 0, 0);
 495
 496                 if (rc == HvLpEvent_Rc_Good) {
 497                         cnx->state |= VETH_STATE_SENTMON;
 498                 } else {
 499                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 500                              && (rc != HvLpEvent_Rc_PathClosed) )
 501                                 veth_error("Error sending monitor to "
 502                                            "lpar %d, rc=%x\n",
 503                                            rlp, (int) rc);
 504
 505                         /* Oh well, hope we get a cap from the other
 506                          * end and do better when that kicks us */
 507                         goto out;
 508                 }
 509         }
 510
 511         if ( (cnx->state & VETH_STATE_OPEN)
 512              && !(cnx->state & VETH_STATE_SENTCAPS)) {
 513                 u64 *rawcap = (u64 *)&cnx->local_caps;
 514
 515                 rc = veth_signalevent(cnx, VethEventTypeCap,
 516                                       HvLpEvent_AckInd_DoAck,
 517                                       HvLpEvent_AckType_ImmediateAck,
 518                                       0, rawcap[0], rawcap[1], rawcap[2],
 519                                       rawcap[3], rawcap[4]);
 520
 521                 if (rc == HvLpEvent_Rc_Good) {
 522                         cnx->state |= VETH_STATE_SENTCAPS;
 523                 } else {
 524                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 525                              && (rc != HvLpEvent_Rc_PathClosed) )
 526                                 veth_error("Error sending caps to "
 527                                            "lpar %d, rc=%x\n",
 528                                            rlp, (int) rc);
 529                         /* Oh well, hope we get a cap from the other
 530                          * end and do better when that kicks us */
 531                         goto out;
 532                 }
 533         }
 534
 535         if ((cnx->state & VETH_STATE_GOTCAPS)
 536             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
 537                 struct VethCapData *remote_caps = &cnx->remote_caps;
 538
 539                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 540                        sizeof(*remote_caps));
 541
 542                 spin_unlock_irq(&cnx->lock);
 543                 rc = veth_process_caps(cnx);
 544                 spin_lock_irq(&cnx->lock);
 545
 546                 /* We dropped the lock, so recheck for anything which
 547                  * might mess us up */
 548                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 549                         goto restart;
 550
 551                 cnx->cap_event.base_event.xRc = rc;
 552                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 553                 if (rc == HvLpEvent_Rc_Good)
 554                         cnx->state |= VETH_STATE_SENTCAPACK;
 555                 else
 556                         goto cant_cope;
 557         }
 558
 559         if ((cnx->state & VETH_STATE_GOTCAPACK)
 560             && (cnx->state & VETH_STATE_GOTCAPS)
 561             && !(cnx->state & VETH_STATE_READY)) {
 562                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 563                         /* Start the ACK timer */
 564                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 565                         add_timer(&cnx->ack_timer);
 566                         cnx->state |= VETH_STATE_READY;
 567                 } else {
 568                         veth_printk(KERN_ERR, "Caps rejected (rc=%d) by "
 569                                     "lpar %d\n",
 570                                     cnx->cap_ack_event.base_event.xRc,
 571                                     rlp);
 572                         goto cant_cope;
 573                 }
 574         }
 575
 576  out:
 577         spin_unlock_irq(&cnx->lock);
 578         return;
 579
 580  cant_cope:
 581         /* FIXME: we get here if something happens we really can't
 582          * cope with.  The link will never work once we get here, and
 583          * all we can do is not lock the rest of the system up */
 584         veth_error("Badness on connection to lpar %d (state=%04lx) "
 585                    " - shutting down\n", rlp, cnx->state);
 586         cnx->state |= VETH_STATE_SHUTDOWN;
 587         spin_unlock_irq(&cnx->lock);
 588 }
 589
 590 static int veth_init_connection(u8 rlp)
 591 {
 592         struct veth_lpar_connection *cnx;
 593         struct veth_msg *msgs;
 594         int i;
 595
 596         if ( (rlp == this_lp)
 597              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 598                 return 0;
 599
 600         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
 601         if (! cnx)
 602                 return -ENOMEM;
 603         memset(cnx, 0, sizeof(*cnx));
 604
 605         cnx->remote_lp = rlp;
 606         spin_lock_init(&cnx->lock);
 607         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
 608         init_timer(&cnx->ack_timer);
 609         cnx->ack_timer.function = veth_timed_ack;
 610         cnx->ack_timer.data = (unsigned long) cnx;
 611         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 612
 613         veth_cnx[rlp] = cnx;
 614
 615         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
 616         if (! msgs) {
 617                 veth_error("Can't allocate buffers for lpar %d\n", rlp);
 618                 return -ENOMEM;
 619         }
 620
 621         cnx->msgs = msgs;
 622         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
 623         spin_lock_init(&cnx->msg_stack_lock);
 624
 625         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 626                 msgs[i].token = i;
 627                 veth_stack_push(cnx, msgs + i);
 628         }
 629
 630         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 631
 632         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 633                 veth_error("Can't allocate events for lpar %d, only got %d\n",
 634                            rlp, cnx->num_events);
 635                 return -ENOMEM;
 636         }
 637
 638         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 639         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 640         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 641
 642         return 0;
 643 }
 644
 645 static void veth_destroy_connection(u8 rlp)
 646 {
 647         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 648
 649         if (! cnx)
 650                 return;
 651
 652         spin_lock_irq(&cnx->lock);
 653         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 654         veth_kick_statemachine(cnx);
 655         spin_unlock_irq(&cnx->lock);
 656
 657         flush_scheduled_work();
 658
 659         /* FIXME: not sure if this is necessary - will already have
 660          * been deleted by the state machine, just want to make sure
 661          * its not running any more */
 662         del_timer_sync(&cnx->ack_timer);
 663
 664         if (cnx->num_events > 0)
 665                 mf_deallocateLpEvents(cnx->remote_lp,
 666                                       HvLpEvent_Type_VirtualLan,
 667                                       cnx->num_events,
 668                                       NULL, NULL);
 669         if (cnx->num_ack_events > 0)
 670                 mf_deallocateLpEvents(cnx->remote_lp,
 671                                       HvLpEvent_Type_VirtualLan,
 672                                       cnx->num_ack_events,
 673                                       NULL, NULL);
 674
 675         if (cnx->msgs)
 676                 kfree(cnx->msgs);
 677 }
 678
 679 /*
 680  * net_device code
 681  */
 682
 683 static int veth_open(struct net_device *dev)
 684 {
 685         struct veth_port *port = (struct veth_port *) dev->priv;
 686
 687         memset(&port->stats, 0, sizeof (port->stats));
 688         netif_start_queue(dev);
 689         return 0;
 690 }
 691
 692 static int veth_close(struct net_device *dev)
 693 {
 694         netif_stop_queue(dev);
 695         return 0;
 696 }
 697
 698 static struct net_device_stats *veth_get_stats(struct net_device *dev)
 699 {
 700         struct veth_port *port = (struct veth_port *) dev->priv;
 701
 702         return &port->stats;
 703 }
 704
 705 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 706 {
 707         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 708                 return -EINVAL;
 709         dev->mtu = new_mtu;
 710         return 0;
 711 }
 712
 713 static void veth_set_multicast_list(struct net_device *dev)
 714 {
 715         struct veth_port *port = (struct veth_port *) dev->priv;
 716         unsigned long flags;
 717
 718         write_lock_irqsave(&port->mcast_gate, flags);
 719
 720         if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */
 721                 printk(KERN_INFO "%s: Promiscuous mode enabled.\n",
 722                        dev->name);
 723                 port->promiscuous = 1;
 724         } else if ( (dev->flags & IFF_ALLMULTI)
 725                     || (dev->mc_count > VETH_MAX_MCAST) ) {
 726                 port->all_mcast = 1;
 727         } else {
 728                 struct dev_mc_list *dmi = dev->mc_list;
 729                 int i;
 730
 731                 /* Update table */
 732                 port->num_mcast = 0;
 733
 734                 for (i = 0; i < dev->mc_count; i++) {
 735                         u8 *addr = dmi->dmi_addr;
 736                         u64 xaddr = 0;
 737
 738                         if (addr[0] & 0x01) {/* multicast address? */
 739                                 memcpy(&xaddr, addr, ETH_ALEN);
 740                                 port->mcast_addr[port->num_mcast] = xaddr;
 741                                 port->num_mcast++;
 742                         }
 743                         dmi = dmi->next;
 744                 }
 745         }
 746
 747         write_unlock_irqrestore(&port->mcast_gate, flags);
 748 }
 749
 750 static int veth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 751 {
 752 #ifdef SIOCETHTOOL
 753         struct ethtool_cmd ecmd;
 754
 755         if (cmd != SIOCETHTOOL)
 756                 return -EOPNOTSUPP;
 757         if (copy_from_user(&ecmd, ifr->ifr_data, sizeof (ecmd)))
 758                 return -EFAULT;
 759         switch (ecmd.cmd) {
 760         case ETHTOOL_GSET:
 761                 ecmd.supported = (SUPPORTED_1000baseT_Full
 762                                   | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 763                 ecmd.advertising = (SUPPORTED_1000baseT_Full
 764                                     | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 765
 766                 ecmd.port = PORT_FIBRE;
 767                 ecmd.transceiver = XCVR_INTERNAL;
 768                 ecmd.phy_address = 0;
 769                 ecmd.speed = SPEED_1000;
 770                 ecmd.duplex = DUPLEX_FULL;
 771                 ecmd.autoneg = AUTONEG_ENABLE;
 772                 ecmd.maxtxpkt = 120;
 773                 ecmd.maxrxpkt = 120;
 774                 if (copy_to_user(ifr->ifr_data, &ecmd, sizeof(ecmd)))
 775                         return -EFAULT;
 776                 return 0;
 777
 778         case ETHTOOL_GDRVINFO:{
 779                         struct ethtool_drvinfo info = { ETHTOOL_GDRVINFO };
 780                         strncpy(info.driver, "veth", sizeof(info.driver) - 1);
 781                         info.driver[sizeof(info.driver) - 1] = '\0';
 782                         strncpy(info.version, "1.0", sizeof(info.version) - 1);
 783                         if (copy_to_user(ifr->ifr_data, &info, sizeof(info)))
 784                                 return -EFAULT;
 785                         return 0;
 786                 }
 787                 /* get link status */
 788         case ETHTOOL_GLINK:{
 789                         struct ethtool_value edata = { ETHTOOL_GLINK };
 790                         edata.data = 1;
 791                         if (copy_to_user(ifr->ifr_data, &edata, sizeof(edata)))
 792                                 return -EFAULT;
 793                         return 0;
 794                 }
 795
 796         default:
 797                 break;
 798         }
 799
 800 #endif
 801         return -EOPNOTSUPP;
 802 }
 803
 804 static void veth_tx_timeout(struct net_device *dev)
 805 {
 806         struct veth_port *port = (struct veth_port *)dev->priv;
 807         struct net_device_stats *stats = &port->stats;
 808         unsigned long flags;
 809         int i;
 810
 811         stats->tx_errors++;
 812
 813         spin_lock_irqsave(&port->pending_gate, flags);
 814
 815         printk(KERN_WARNING "%s: Tx timeout!  Resetting lp connections: %08x\n",
 816                dev->name, port->pending_lpmask);
 817
 818         /* If we've timed out the queue must be stopped, which should
 819          * only ever happen when there is a pending packet. */
 820         WARN_ON(! port->pending_lpmask);
 821
 822         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 823                 struct veth_lpar_connection *cnx = veth_cnx[i];
 824
 825                 if (! (port->pending_lpmask & (1<<i)))
 826                         continue;
 827
 828                 /* If we're pending on it, we must be connected to it,
 829                  * so we should certainly have a structure for it. */
 830                 BUG_ON(! cnx);
 831
 832                 /* Theoretically we could be kicking a connection
 833                  * which doesn't deserve it, but in practice if we've
 834                  * had a Tx timeout, the pending_lpmask will have
 835                  * exactly one bit set - the connection causing the
 836                  * problem. */
 837                 spin_lock(&cnx->lock);
 838                 cnx->state |= VETH_STATE_RESET;
 839                 veth_kick_statemachine(cnx);
 840                 spin_unlock(&cnx->lock);
 841         }
 842
 843         spin_unlock_irqrestore(&port->pending_gate, flags);
 844 }
 845
 846 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
 847 {
 848         struct net_device *dev;
 849         struct veth_port *port;
 850         int i, rc;
 851
 852         dev = alloc_etherdev(sizeof (struct veth_port));
 853         if (! dev) {
 854                 veth_error("Unable to allocate net_device structure!\n");
 855                 return NULL;
 856         }
 857
 858         port = (struct veth_port *) dev->priv;
 859
 860         spin_lock_init(&port->pending_gate);
 861         rwlock_init(&port->mcast_gate);
 862
 863         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 864                 HvLpVirtualLanIndexMap map;
 865
 866                 if (i == this_lp)
 867                         continue;
 868                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
 869                 if (map & (0x8000 >> vlan))
 870                         port->lpar_map |= (1 << i);
 871         }
 872         port->dev = vdev;
 873
 874         dev->dev_addr[0] = 0x02;
 875         dev->dev_addr[1] = 0x01;
 876         dev->dev_addr[2] = 0xff;
 877         dev->dev_addr[3] = vlan;
 878         dev->dev_addr[4] = 0xff;
 879         dev->dev_addr[5] = this_lp;
 880
 881         dev->mtu = VETH_MAX_MTU;
 882
 883         memcpy(&port->mac_addr, dev->dev_addr, 6);
 884
 885         dev->open = veth_open;
 886         dev->hard_start_xmit = veth_start_xmit;
 887         dev->stop = veth_close;
 888         dev->get_stats = veth_get_stats;
 889         dev->change_mtu = veth_change_mtu;
 890         dev->set_mac_address = NULL;
 891         dev->set_multicast_list = veth_set_multicast_list;
 892         dev->do_ioctl = veth_ioctl;
 893
 894         dev->watchdog_timeo = 2 * (VETH_ACKTIMEOUT * HZ / 1000000);
 895         dev->tx_timeout = veth_tx_timeout;
 896
 897         SET_NETDEV_DEV(dev, vdev);
 898
 899         rc = register_netdev(dev);
 900         if (rc != 0) {
 901                 veth_printk(KERN_ERR,
 902                             "Failed to register ethernet device for vlan %d\n",
 903                             vlan);
 904                 free_netdev(dev);
 905                 return NULL;
 906         }
 907
 908         veth_printk(KERN_DEBUG, "%s attached to iSeries vlan %d (lpar_map=0x%04x)\n",
 909                     dev->name, vlan, port->lpar_map);
 910
 911         return dev;
 912 }
 913
 914 /*
 915  * Tx path
 916  */
 917
 918 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
 919                                 struct net_device *dev)
 920 {
 921         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 922         struct veth_port *port = (struct veth_port *) dev->priv;
 923         HvLpEvent_Rc rc;
 924         u32 dma_address, dma_length;
 925         struct veth_msg *msg = NULL;
 926         int err = 0;
 927         unsigned long flags;
 928
 929         if (! cnx) {
 930                 port->stats.tx_errors++;
 931                 dev_kfree_skb(skb);
 932                 return 0;
 933         }
 934
 935         spin_lock_irqsave(&cnx->lock, flags);
 936
 937         if (! cnx->state & VETH_STATE_READY)
 938                 goto drop;
 939
 940         if ((skb->len - 14) > VETH_MAX_MTU)
 941                 goto drop;
 942
 943         msg = veth_stack_pop(cnx);
 944
 945         if (! msg) {
 946                 err = 1;
 947                 goto drop;
 948         }
 949
 950         dma_length = skb->len;
 951         dma_address = dma_map_single(port->dev, skb->data,
 952                                      dma_length, DMA_TO_DEVICE);
 953
 954         if (dma_mapping_error(dma_address))
 955                 goto recycle_and_drop;
 956
 957         /* Is it really necessary to check the length and address
 958          * fields of the first entry here? */
 959         msg->skb = skb;
 960         msg->dev = port->dev;
 961         msg->data.addr[0] = dma_address;
 962         msg->data.len[0] = dma_length;
 963         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
 964         set_bit(0, &(msg->in_use));
 965         rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
 966
 967         if (rc != HvLpEvent_Rc_Good)
 968                 goto recycle_and_drop;
 969
 970         spin_unlock_irqrestore(&cnx->lock, flags);
 971         return 0;
 972
 973  recycle_and_drop:
 974         msg->skb = NULL;
 975         /* need to set in use to make veth_recycle_msg in case this
 976          * was a mapping failure */
 977         set_bit(0, &msg->in_use);
 978         veth_recycle_msg(cnx, msg);
 979  drop:
 980         port->stats.tx_errors++;
 981         dev_kfree_skb(skb);
 982         spin_unlock_irqrestore(&cnx->lock, flags);
 983         return err;
 984 }
 985
 986 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
 987                                           HvLpIndexMap lpmask,
 988                                           struct net_device *dev)
 989 {
 990         struct veth_port *port = (struct veth_port *) dev->priv;
 991         int i;
 992         int rc;
 993
 994         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 995                 if ((lpmask & (1 << i)) == 0)
 996                         continue;
 997
 998                 rc = veth_transmit_to_one(skb_get(skb), i, dev);
 999                 if (! rc)
1000                         lpmask &= ~(1<<i);
1001         }
1002
1003         if (! lpmask) {
1004                 port->stats.tx_packets++;
1005                 port->stats.tx_bytes += skb->len;
1006         }
1007
1008         return lpmask;
1009 }
1010
1011 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1012 {
1013         unsigned char *frame = skb->data;
1014         struct veth_port *port = (struct veth_port *) dev->priv;
1015         unsigned long flags;
1016         HvLpIndexMap lpmask;
1017
1018         if (! (frame[0] & 0x01)) {
1019                 /* unicast packet */
1020                 HvLpIndex rlp = frame[5];
1021
1022                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1023                         dev_kfree_skb(skb);
1024                         return 0;
1025                 }
1026
1027                 lpmask = 1 << rlp;
1028         } else {
1029                 lpmask = port->lpar_map;
1030         }
1031
1032         spin_lock_irqsave(&port->pending_gate, flags);
1033
1034         lpmask = veth_transmit_to_many(skb, lpmask, dev);
1035
1036         if (! lpmask) {
1037                 dev_kfree_skb(skb);
1038         } else {
1039                 if (port->pending_skb) {
1040                         veth_error("%s: Tx while skb was pending!\n",
1041                                    dev->name);
1042                         dev_kfree_skb(skb);
1043                         spin_unlock_irqrestore(&port->pending_gate, flags);
1044                         return 1;
1045                 }
1046
1047                 port->pending_skb = skb;
1048                 port->pending_lpmask = lpmask;
1049                 netif_stop_queue(dev);
1050         }
1051
1052         spin_unlock_irqrestore(&port->pending_gate, flags);
1053
1054         return 0;
1055 }
1056
1057 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1058                              struct veth_msg *msg)
1059 {
1060         u32 dma_address, dma_length;
1061
1062         if (test_and_clear_bit(0, &msg->in_use)) {
1063                 dma_address = msg->data.addr[0];
1064                 dma_length = msg->data.len[0];
1065
1066                 dma_unmap_single(msg->dev, dma_address, dma_length,
1067                                  DMA_TO_DEVICE);
1068
1069                 if (msg->skb) {
1070                         dev_kfree_skb_any(msg->skb);
1071                         msg->skb = NULL;
1072                 }
1073
1074                 memset(&msg->data, 0, sizeof(msg->data));
1075                 veth_stack_push(cnx, msg);
1076         } else
1077                 if (cnx->state & VETH_STATE_OPEN)
1078                         veth_error("Bogus frames ack from lpar %d (#%d)\n",
1079                                    cnx->remote_lp, msg->token);
1080 }
1081
1082 static void veth_flush_pending(struct veth_lpar_connection *cnx)
1083 {
1084         int i;
1085         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1086                 struct net_device *dev = veth_dev[i];
1087                 struct veth_port *port;
1088                 unsigned long flags;
1089
1090                 if (! dev)
1091                         continue;
1092
1093                 port = (struct veth_port *)dev->priv;
1094
1095                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1096                         continue;
1097
1098                 spin_lock_irqsave(&port->pending_gate, flags);
1099                 if (port->pending_skb) {
1100                         port->pending_lpmask =
1101                                 veth_transmit_to_many(port->pending_skb,
1102                                                       port->pending_lpmask,
1103                                                       dev);
1104                         if (! port->pending_lpmask) {
1105                                 dev_kfree_skb_any(port->pending_skb);
1106                                 port->pending_skb = NULL;
1107                                 netif_wake_queue(dev);
1108                         }
1109                 }
1110                 spin_unlock_irqrestore(&port->pending_gate, flags);
1111         }
1112 }
1113
1114 /*
1115  * Rx path
1116  */
1117
1118 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1119 {
1120         int wanted = 0;
1121         int i;
1122         unsigned long flags;
1123
1124         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1125                 return 1;
1126
1127         if (! (((char *) &mac_addr)[0] & 0x01))
1128                 return 0;
1129
1130         read_lock_irqsave(&port->mcast_gate, flags);
1131
1132         if (port->promiscuous || port->all_mcast) {
1133                 wanted = 1;
1134                 goto out;
1135         }
1136
1137         for (i = 0; i < port->num_mcast; ++i) {
1138                 if (port->mcast_addr[i] == mac_addr) {
1139                         wanted = 1;
1140                         break;
1141                 }
1142         }
1143
1144  out:
1145         read_unlock_irqrestore(&port->mcast_gate, flags);
1146
1147         return wanted;
1148 }
1149
1150 struct dma_chunk {
1151         u64 addr;
1152         u64 size;
1153 };
1154
1155 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1156
1157 static inline void veth_build_dma_list(struct dma_chunk *list,
1158                                        unsigned char *p, unsigned long length)
1159 {
1160         unsigned long done;
1161         int i = 1;
1162
1163         /* FIXME: skbs are continguous in real addresses.  Do we
1164          * really need to break it into PAGE_SIZE chunks, or can we do
1165          * it just at the granularity of iSeries real->absolute
1166          * mapping?  Indeed, given the way the allocator works, can we
1167          * count on them being absolutely contiguous? */
1168         list[0].addr = ISERIES_HV_ADDR(p);
1169         list[0].size = min(length,
1170                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1171
1172         done = list[0].size;
1173         while (done < length) {
1174                 list[i].addr = ISERIES_HV_ADDR(p + done);
1175                 list[i].size = min(length-done, PAGE_SIZE);
1176                 done += list[i].size;
1177                 i++;
1178         }
1179 }
1180
1181 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1182 {
1183         HvLpEvent_Rc rc;
1184
1185         rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1186                              0, &cnx->pending_acks);
1187
1188         if (rc != HvLpEvent_Rc_Good)
1189                 veth_error("Error 0x%x acking frames from lpar %d!\n",
1190                            (unsigned)rc, cnx->remote_lp);
1191
1192         cnx->num_pending_acks = 0;
1193         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1194 }
1195
1196 static void veth_receive(struct veth_lpar_connection *cnx,
1197                          struct VethLpEvent *event)
1198 {
1199         struct VethFramesData *senddata = &event->u.frames_data;
1200         int startchunk = 0;
1201         int nchunks;
1202         unsigned long flags;
1203         HvLpDma_Rc rc;
1204
1205         do {
1206                 u16 length = 0;
1207                 struct sk_buff *skb;
1208                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1209                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1210                 u64 dest;
1211                 HvLpVirtualLanIndex vlan;
1212                 struct net_device *dev;
1213                 struct veth_port *port;
1214
1215                 /* FIXME: do we need this? */
1216                 memset(local_list, 0, sizeof(local_list));
1217                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1218
1219                 /* a 0 address marks the end of the valid entries */
1220                 if (senddata->addr[startchunk] == 0)
1221                         break;
1222
1223                 /* make sure that we have at least 1 EOF entry in the
1224                  * remaining entries */
1225                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1226                         veth_error("missing EOF frag in event "
1227                                    "eofmask=0x%x startchunk=%d\n",
1228                                    (unsigned) senddata->eofmask, startchunk);
1229                         break;
1230                 }
1231
1232                 /* build list of chunks in this frame */
1233                 nchunks = 0;
1234                 do {
1235                         remote_list[nchunks].addr =
1236                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1237                         remote_list[nchunks].size =
1238                                 senddata->len[startchunk+nchunks];
1239                         length += remote_list[nchunks].size;
1240                 } while (! (senddata->eofmask &
1241                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1242
1243                 /* length == total length of all chunks */
1244                 /* nchunks == # of chunks in this frame */
1245
1246                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1247                         veth_error("Received oversize frame from lpar %d "
1248                                    "(length=%d)\n", cnx->remote_lp, length);
1249                         continue;
1250                 }
1251
1252                 skb = alloc_skb(length, GFP_ATOMIC);
1253                 if (!skb)
1254                         continue;
1255
1256                 veth_build_dma_list(local_list, skb->data, length);
1257
1258                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1259                                             event->base_event.xSourceLp,
1260                                             HvLpDma_Direction_RemoteToLocal,
1261                                             cnx->src_inst,
1262                                             cnx->dst_inst,
1263                                             HvLpDma_AddressType_RealAddress,
1264                                             HvLpDma_AddressType_TceIndex,
1265                                             ISERIES_HV_ADDR(&local_list),
1266                                             ISERIES_HV_ADDR(&remote_list),
1267                                             length);
1268                 if (rc != HvLpDma_Rc_Good) {
1269                         dev_kfree_skb_irq(skb);
1270                         continue;
1271                 }
1272
1273                 vlan = skb->data[9];
1274                 dev = veth_dev[vlan];
1275                 if (! dev)
1276                         /* Some earlier versions of the driver sent
1277                            broadcasts down all connections, even to
1278                            lpars that weren't on the relevant vlan.
1279                            So ignore packets belonging to a vlan we're
1280                            not on. */
1281                         continue;
1282
1283                 port = (struct veth_port *)dev->priv;
1284                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1285
1286                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1287                         dev_kfree_skb_irq(skb);
1288                         continue;
1289                 }
1290                 if (! veth_frame_wanted(port, dest)) {
1291                         dev_kfree_skb_irq(skb);
1292                         continue;
1293                 }
1294
1295                 skb_put(skb, length);
1296                 skb->dev = dev;
1297                 skb->protocol = eth_type_trans(skb, dev);
1298                 skb->ip_summed = CHECKSUM_NONE;
1299                 netif_rx(skb);  /* send it up */
1300                 port->stats.rx_packets++;
1301                 port->stats.rx_bytes += length;
1302         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1303
1304         /* Ack it */
1305         spin_lock_irqsave(&cnx->lock, flags);
1306         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1307
1308         cnx->pending_acks[cnx->num_pending_acks++] =
1309                 event->base_event.xCorrelationToken;
1310
1311         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1312              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1313                 veth_flush_acks(cnx);
1314
1315         spin_unlock_irqrestore(&cnx->lock, flags);
1316 }
1317
1318 static void veth_timed_ack(unsigned long ptr)
1319 {
1320         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1321         unsigned long flags;
1322
1323         /* Ack all the events */
1324         spin_lock_irqsave(&cnx->lock, flags);
1325         if (cnx->num_pending_acks > 0)
1326                 veth_flush_acks(cnx);
1327
1328         /* Reschedule the timer */
1329         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1330         add_timer(&cnx->ack_timer);
1331         spin_unlock_irqrestore(&cnx->lock, flags);
1332 }
1333
1334 static int veth_remove(struct vio_dev *vdev)
1335 {
1336         int i = vdev->unit_address;
1337         struct net_device *dev;
1338
1339         dev = veth_dev[i];
1340         if (dev != NULL) {
1341                 veth_dev[i] = NULL;
1342                 unregister_netdev(dev);
1343                 free_netdev(dev);
1344         }
1345         return 0;
1346 }
1347
1348 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1349 {
1350         int i = vdev->unit_address;
1351         struct net_device *dev;
1352
1353         dev = veth_probe_one(i, &vdev->dev);
1354         if (dev == NULL) {
1355                 veth_remove(vdev);
1356                 return 1;
1357         }
1358         veth_dev[i] = dev;
1359
1360         /* Start the state machine on each connection, to commence
1361          * link negotiation */
1362         for (i = 0; i < HVMAXARCHITECTEDLPS; i++)
1363                 if (veth_cnx[i])
1364                         veth_kick_statemachine(veth_cnx[i]);
1365
1366         return 0;
1367 }
1368
1369 /**
1370  * veth_device_table: Used by vio.c to match devices that we
1371  * support.
1372  */
1373 static struct vio_device_id veth_device_table[] __devinitdata = {
1374         { "vlan", "" },
1375         { NULL, NULL }
1376 };
1377 MODULE_DEVICE_TABLE(vio, veth_device_table);
1378
1379 static struct vio_driver veth_driver = {
1380         .name = "iseries_veth",
1381         .id_table = veth_device_table,
1382         .probe = veth_probe,
1383         .remove = veth_remove
1384 };
1385
1386 /*
1387  * Module initialization/cleanup
1388  */
1389
1390 void __exit veth_module_cleanup(void)
1391 {
1392         int i;
1393
1394         vio_unregister_driver(&veth_driver);
1395
1396         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
1397                 veth_destroy_connection(i);
1398
1399         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1400 }
1401 module_exit(veth_module_cleanup);
1402
1403 int __init veth_module_init(void)
1404 {
1405         int i;
1406         int rc;
1407
1408         this_lp = HvLpConfig_getLpIndex_outline();
1409
1410         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1411                 rc = veth_init_connection(i);
1412                 if (rc != 0) {
1413                         veth_module_cleanup();
1414                         return rc;
1415                 }
1416         }
1417
1418         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1419                                   &veth_handle_event);
1420
1421         return vio_register_driver(&veth_driver);
1422 }
1423 module_init(veth_module_init);