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 extern struct vio_dev *iSeries_veth_dev;
  85
  86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  88 MODULE_LICENSE("GPL");
  89
  90 #define VETH_NUMBUFFERS         (120)
  91 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
  92 #define VETH_MAX_MCAST          (12)
  93
  94 #define VETH_MAX_MTU            (9000)
  95
  96 #if VETH_NUMBUFFERS < 10
  97 #define ACK_THRESHOLD           (1)
  98 #elif VETH_NUMBUFFERS < 20
  99 #define ACK_THRESHOLD           (4)
 100 #elif VETH_NUMBUFFERS < 40
 101 #define ACK_THRESHOLD           (10)
 102 #else
 103 #define ACK_THRESHOLD           (20)
 104 #endif
 105
 106 #define VETH_STATE_SHUTDOWN     (0x0001)
 107 #define VETH_STATE_OPEN         (0x0002)
 108 #define VETH_STATE_RESET        (0x0004)
 109 #define VETH_STATE_SENTMON      (0x0008)
 110 #define VETH_STATE_SENTCAPS     (0x0010)
 111 #define VETH_STATE_GOTCAPACK    (0x0020)
 112 #define VETH_STATE_GOTCAPS      (0x0040)
 113 #define VETH_STATE_SENTCAPACK   (0x0080)
 114 #define VETH_STATE_READY        (0x0100)
 115
 116 struct veth_msg {
 117         struct veth_msg *next;
 118         struct VethFramesData data;
 119         int token;
 120         unsigned long in_use;
 121         struct sk_buff *skb;
 122 };
 123
 124 struct veth_lpar_connection {
 125         HvLpIndex remote_lp;
 126         struct work_struct statemachine_wq;
 127         struct veth_msg *msgs;
 128         int num_events;
 129         struct VethCapData local_caps;
 130
 131         struct timer_list ack_timer;
 132
 133         spinlock_t lock;
 134         unsigned long state;
 135         HvLpInstanceId src_inst;
 136         HvLpInstanceId dst_inst;
 137         struct VethLpEvent cap_event, cap_ack_event;
 138         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 139         u32 num_pending_acks;
 140
 141         int num_ack_events;
 142         struct VethCapData remote_caps;
 143         u32 ack_timeout;
 144
 145         spinlock_t msg_stack_lock;
 146         struct veth_msg *msg_stack_head;
 147 };
 148
 149 struct veth_port {
 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         }
 465
 466         if (cnx->state & VETH_STATE_SHUTDOWN)
 467                 /* It's all over, do nothing */
 468                 goto out;
 469
 470         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 471                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 472                         goto cant_cope;
 473
 474                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 475                 cnx->src_inst =
 476                         HvCallEvent_getSourceLpInstanceId(rlp,
 477                                                           HvLpEvent_Type_VirtualLan);
 478                 cnx->dst_inst =
 479                         HvCallEvent_getTargetLpInstanceId(rlp,
 480                                                           HvLpEvent_Type_VirtualLan);
 481                 cnx->state |= VETH_STATE_OPEN;
 482         }
 483
 484         if ( (cnx->state & VETH_STATE_OPEN)
 485              && !(cnx->state & VETH_STATE_SENTMON) ) {
 486                 rc = veth_signalevent(cnx, VethEventTypeMonitor,
 487                                       HvLpEvent_AckInd_DoAck,
 488                                       HvLpEvent_AckType_DeferredAck,
 489                                       0, 0, 0, 0, 0, 0);
 490
 491                 if (rc == HvLpEvent_Rc_Good) {
 492                         cnx->state |= VETH_STATE_SENTMON;
 493                 } else {
 494                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 495                              && (rc != HvLpEvent_Rc_PathClosed) )
 496                                 veth_error("Error sending monitor to "
 497                                            "lpar %d, rc=%x\n",
 498                                            rlp, (int) rc);
 499
 500                         /* Oh well, hope we get a cap from the other
 501                          * end and do better when that kicks us */
 502                         goto out;
 503                 }
 504         }
 505
 506         if ( (cnx->state & VETH_STATE_OPEN)
 507              && !(cnx->state & VETH_STATE_SENTCAPS)) {
 508                 u64 *rawcap = (u64 *)&cnx->local_caps;
 509
 510                 rc = veth_signalevent(cnx, VethEventTypeCap,
 511                                       HvLpEvent_AckInd_DoAck,
 512                                       HvLpEvent_AckType_ImmediateAck,
 513                                       0, rawcap[0], rawcap[1], rawcap[2],
 514                                       rawcap[3], rawcap[4]);
 515
 516                 if (rc == HvLpEvent_Rc_Good) {
 517                         cnx->state |= VETH_STATE_SENTCAPS;
 518                 } else {
 519                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 520                              && (rc != HvLpEvent_Rc_PathClosed) )
 521                                 veth_error("Error sending caps to "
 522                                            "lpar %d, rc=%x\n",
 523                                            rlp, (int) rc);
 524                         /* Oh well, hope we get a cap from the other
 525                          * end and do better when that kicks us */
 526                         goto out;
 527                 }
 528         }
 529
 530         if ((cnx->state & VETH_STATE_GOTCAPS)
 531             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
 532                 struct VethCapData *remote_caps = &cnx->remote_caps;
 533
 534                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 535                        sizeof(*remote_caps));
 536
 537                 spin_unlock_irq(&cnx->lock);
 538                 rc = veth_process_caps(cnx);
 539                 spin_lock_irq(&cnx->lock);
 540
 541                 /* We dropped the lock, so recheck for anything which
 542                  * might mess us up */
 543                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 544                         goto restart;
 545
 546                 cnx->cap_event.base_event.xRc = rc;
 547                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 548                 if (rc == HvLpEvent_Rc_Good)
 549                         cnx->state |= VETH_STATE_SENTCAPACK;
 550                 else
 551                         goto cant_cope;
 552         }
 553
 554         if ((cnx->state & VETH_STATE_GOTCAPACK)
 555             && (cnx->state & VETH_STATE_GOTCAPS)
 556             && !(cnx->state & VETH_STATE_READY)) {
 557                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 558                         /* Start the ACK timer */
 559                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 560                         add_timer(&cnx->ack_timer);
 561                         cnx->state |= VETH_STATE_READY;
 562                 } else {
 563                         veth_printk(KERN_ERR, "Caps rejected (rc=%d) by "
 564                                     "lpar %d\n",
 565                                     cnx->cap_ack_event.base_event.xRc,
 566                                     rlp);
 567                         goto cant_cope;
 568                 }
 569         }
 570
 571  out:
 572         spin_unlock_irq(&cnx->lock);
 573         return;
 574
 575  cant_cope:
 576         /* FIXME: we get here if something happens we really can't
 577          * cope with.  The link will never work once we get here, and
 578          * all we can do is not lock the rest of the system up */
 579         veth_error("Badness on connection to lpar %d (state=%04lx) "
 580                    " - shutting down\n", rlp, cnx->state);
 581         cnx->state |= VETH_STATE_SHUTDOWN;
 582         spin_unlock_irq(&cnx->lock);
 583 }
 584
 585 static int veth_init_connection(u8 rlp)
 586 {
 587         struct veth_lpar_connection *cnx;
 588         struct veth_msg *msgs;
 589         int i;
 590
 591         if ( (rlp == this_lp)
 592              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 593                 return 0;
 594
 595         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
 596         if (! cnx)
 597                 return -ENOMEM;
 598         memset(cnx, 0, sizeof(*cnx));
 599
 600         cnx->remote_lp = rlp;
 601         spin_lock_init(&cnx->lock);
 602         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
 603         init_timer(&cnx->ack_timer);
 604         cnx->ack_timer.function = veth_timed_ack;
 605         cnx->ack_timer.data = (unsigned long) cnx;
 606         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 607
 608         veth_cnx[rlp] = cnx;
 609
 610         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
 611         if (! msgs) {
 612                 veth_error("Can't allocate buffers for lpar %d\n", rlp);
 613                 return -ENOMEM;
 614         }
 615
 616         cnx->msgs = msgs;
 617         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
 618         spin_lock_init(&cnx->msg_stack_lock);
 619
 620         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 621                 msgs[i].token = i;
 622                 veth_stack_push(cnx, msgs + i);
 623         }
 624
 625         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 626
 627         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 628                 veth_error("Can't allocate events for lpar %d, only got %d\n",
 629                            rlp, cnx->num_events);
 630                 return -ENOMEM;
 631         }
 632
 633         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 634         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 635         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 636
 637         return 0;
 638 }
 639
 640 static void veth_destroy_connection(u8 rlp)
 641 {
 642         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 643
 644         if (! cnx)
 645                 return;
 646
 647         spin_lock_irq(&cnx->lock);
 648         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 649         veth_kick_statemachine(cnx);
 650         spin_unlock_irq(&cnx->lock);
 651
 652         flush_scheduled_work();
 653
 654         /* FIXME: not sure if this is necessary - will already have
 655          * been deleted by the state machine, just want to make sure
 656          * its not running any more */
 657         del_timer_sync(&cnx->ack_timer);
 658
 659         if (cnx->num_events > 0)
 660                 mf_deallocateLpEvents(cnx->remote_lp,
 661                                       HvLpEvent_Type_VirtualLan,
 662                                       cnx->num_events,
 663                                       NULL, NULL);
 664         if (cnx->num_ack_events > 0)
 665                 mf_deallocateLpEvents(cnx->remote_lp,
 666                                       HvLpEvent_Type_VirtualLan,
 667                                       cnx->num_ack_events,
 668                                       NULL, NULL);
 669
 670         if (cnx->msgs)
 671                 kfree(cnx->msgs);
 672 }
 673
 674 /*
 675  * net_device code
 676  */
 677
 678 static int veth_open(struct net_device *dev)
 679 {
 680         struct veth_port *port = (struct veth_port *) dev->priv;
 681
 682         memset(&port->stats, 0, sizeof (port->stats));
 683         netif_start_queue(dev);
 684         return 0;
 685 }
 686
 687 static int veth_close(struct net_device *dev)
 688 {
 689         netif_stop_queue(dev);
 690         return 0;
 691 }
 692
 693 static struct net_device_stats *veth_get_stats(struct net_device *dev)
 694 {
 695         struct veth_port *port = (struct veth_port *) dev->priv;
 696
 697         return &port->stats;
 698 }
 699
 700 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 701 {
 702         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 703                 return -EINVAL;
 704         dev->mtu = new_mtu;
 705         return 0;
 706 }
 707
 708 static void veth_set_multicast_list(struct net_device *dev)
 709 {
 710         struct veth_port *port = (struct veth_port *) dev->priv;
 711         unsigned long flags;
 712
 713         write_lock_irqsave(&port->mcast_gate, flags);
 714
 715         if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */
 716                 printk(KERN_INFO "%s: Promiscuous mode enabled.\n",
 717                        dev->name);
 718                 port->promiscuous = 1;
 719         } else if ( (dev->flags & IFF_ALLMULTI)
 720                     || (dev->mc_count > VETH_MAX_MCAST) ) {
 721                 port->all_mcast = 1;
 722         } else {
 723                 struct dev_mc_list *dmi = dev->mc_list;
 724                 int i;
 725
 726                 /* Update table */
 727                 port->num_mcast = 0;
 728
 729                 for (i = 0; i < dev->mc_count; i++) {
 730                         u8 *addr = dmi->dmi_addr;
 731                         u64 xaddr = 0;
 732
 733                         if (addr[0] & 0x01) {/* multicast address? */
 734                                 memcpy(&xaddr, addr, ETH_ALEN);
 735                                 port->mcast_addr[port->num_mcast] = xaddr;
 736                                 port->num_mcast++;
 737                         }
 738                         dmi = dmi->next;
 739                 }
 740         }
 741
 742         write_unlock_irqrestore(&port->mcast_gate, flags);
 743 }
 744
 745 static int veth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 746 {
 747 #ifdef SIOCETHTOOL
 748         struct ethtool_cmd ecmd;
 749
 750         if (cmd != SIOCETHTOOL)
 751                 return -EOPNOTSUPP;
 752         if (copy_from_user(&ecmd, ifr->ifr_data, sizeof (ecmd)))
 753                 return -EFAULT;
 754         switch (ecmd.cmd) {
 755         case ETHTOOL_GSET:
 756                 ecmd.supported = (SUPPORTED_1000baseT_Full
 757                                   | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 758                 ecmd.advertising = (SUPPORTED_1000baseT_Full
 759                                     | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 760
 761                 ecmd.port = PORT_FIBRE;
 762                 ecmd.transceiver = XCVR_INTERNAL;
 763                 ecmd.phy_address = 0;
 764                 ecmd.speed = SPEED_1000;
 765                 ecmd.duplex = DUPLEX_FULL;
 766                 ecmd.autoneg = AUTONEG_ENABLE;
 767                 ecmd.maxtxpkt = 120;
 768                 ecmd.maxrxpkt = 120;
 769                 if (copy_to_user(ifr->ifr_data, &ecmd, sizeof(ecmd)))
 770                         return -EFAULT;
 771                 return 0;
 772
 773         case ETHTOOL_GDRVINFO:{
 774                         struct ethtool_drvinfo info = { ETHTOOL_GDRVINFO };
 775                         strncpy(info.driver, "veth", sizeof(info.driver) - 1);
 776                         info.driver[sizeof(info.driver) - 1] = '\0';
 777                         strncpy(info.version, "1.0", sizeof(info.version) - 1);
 778                         if (copy_to_user(ifr->ifr_data, &info, sizeof(info)))
 779                                 return -EFAULT;
 780                         return 0;
 781                 }
 782                 /* get link status */
 783         case ETHTOOL_GLINK:{
 784                         struct ethtool_value edata = { ETHTOOL_GLINK };
 785                         edata.data = 1;
 786                         if (copy_to_user(ifr->ifr_data, &edata, sizeof(edata)))
 787                                 return -EFAULT;
 788                         return 0;
 789                 }
 790
 791         default:
 792                 break;
 793         }
 794
 795 #endif
 796         return -EOPNOTSUPP;
 797 }
 798
 799 struct net_device * __init veth_probe_one(int vlan)
 800 {
 801         struct net_device *dev;
 802         struct veth_port *port;
 803         int i, rc;
 804
 805         dev = alloc_etherdev(sizeof (struct veth_port));
 806         if (! dev) {
 807                 veth_error("Unable to allocate net_device structure!\n");
 808                 return NULL;
 809         }
 810
 811         port = (struct veth_port *) dev->priv;
 812
 813         spin_lock_init(&port->pending_gate);
 814         rwlock_init(&port->mcast_gate);
 815
 816         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 817                 HvLpVirtualLanIndexMap map;
 818
 819                 if (i == this_lp)
 820                         continue;
 821                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
 822                 if (map & (0x8000 >> vlan))
 823                         port->lpar_map |= (1 << i);
 824         }
 825
 826         dev->dev_addr[0] = 0x02;
 827         dev->dev_addr[1] = 0x01;
 828         dev->dev_addr[2] = 0xff;
 829         dev->dev_addr[3] = vlan;
 830         dev->dev_addr[4] = 0xff;
 831         dev->dev_addr[5] = this_lp;
 832
 833         dev->mtu = VETH_MAX_MTU;
 834
 835         memcpy(&port->mac_addr, dev->dev_addr, 6);
 836
 837         dev->open = veth_open;
 838         dev->hard_start_xmit = veth_start_xmit;
 839         dev->stop = veth_close;
 840         dev->get_stats = veth_get_stats;
 841         dev->change_mtu = veth_change_mtu;
 842         dev->set_mac_address = NULL;
 843         dev->set_multicast_list = veth_set_multicast_list;
 844         dev->do_ioctl = veth_ioctl;
 845
 846         rc = register_netdev(dev);
 847         if (rc != 0) {
 848                 veth_printk(KERN_ERR,
 849                             "Failed to register ethernet device for vlan %d\n",
 850                             vlan);
 851                 free_netdev(dev);
 852                 return NULL;
 853         }
 854
 855         veth_printk(KERN_DEBUG, "%s attached to iSeries vlan %d (lpar_map=0x%04x)\n",
 856                     dev->name, vlan, port->lpar_map);
 857
 858         return dev;
 859 }
 860
 861 /*
 862  * Tx path
 863  */
 864
 865 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
 866                                 struct net_device *dev)
 867 {
 868         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 869         struct veth_port *port = (struct veth_port *) dev->priv;
 870         HvLpEvent_Rc rc;
 871         u32 dma_address, dma_length;
 872         struct veth_msg *msg = NULL;
 873         int err = 0;
 874         unsigned long flags;
 875
 876         if (! cnx) {
 877                 port->stats.tx_errors++;
 878                 dev_kfree_skb(skb);
 879                 return 0;
 880         }
 881
 882         spin_lock_irqsave(&cnx->lock, flags);
 883
 884         if (! cnx->state & VETH_STATE_READY)
 885                 goto drop;
 886
 887         if ((skb->len - 14) > VETH_MAX_MTU)
 888                 goto drop;
 889
 890         msg = veth_stack_pop(cnx);
 891
 892         if (! msg) {
 893                 err = 1;
 894                 goto drop;
 895         }
 896
 897         dma_length = skb->len;
 898         dma_address = vio_map_single(iSeries_veth_dev, skb->data,
 899                                      dma_length, DMA_TO_DEVICE);
 900
 901         if (dma_mapping_error(dma_address))
 902                 goto recycle_and_drop;
 903
 904         /* Is it really necessary to check the length and address
 905          * fields of the first entry here? */
 906         msg->skb = skb;
 907         msg->data.addr[0] = dma_address;
 908         msg->data.len[0] = dma_length;
 909         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
 910         set_bit(0, &(msg->in_use));
 911         rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
 912
 913         if (rc != HvLpEvent_Rc_Good)
 914                 goto recycle_and_drop;
 915
 916         spin_unlock_irqrestore(&cnx->lock, flags);
 917         return 0;
 918
 919  recycle_and_drop:
 920         msg->skb = NULL;
 921         /* need to set in use to make veth_recycle_msg in case this
 922          * was a mapping failure */
 923         set_bit(0, &msg->in_use);
 924         veth_recycle_msg(cnx, msg);
 925  drop:
 926         port->stats.tx_errors++;
 927         dev_kfree_skb(skb);
 928         spin_unlock_irqrestore(&cnx->lock, flags);
 929         return err;
 930 }
 931
 932 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
 933                                           HvLpIndexMap lpmask,
 934                                           struct net_device *dev)
 935 {
 936         struct veth_port *port = (struct veth_port *) dev->priv;
 937         int i;
 938         int rc;
 939
 940         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 941                 struct sk_buff *clone;
 942
 943                 if (! lpmask & (1<<i))
 944                         continue;
 945
 946                 clone = skb_clone(skb, GFP_ATOMIC);
 947                 if (! clone) {
 948                         veth_error("%s: skb_clone failed %p\n",
 949                                    dev->name, skb);
 950                         continue;
 951                 }
 952
 953                 rc = veth_transmit_to_one(clone, i, dev);
 954                 if (! rc)
 955                         lpmask &= ~(1<<i);
 956         }
 957
 958         if (! lpmask) {
 959                 port->stats.tx_packets++;
 960                 port->stats.tx_bytes += skb->len;
 961         }
 962
 963         return lpmask;
 964 }
 965
 966 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
 967 {
 968         unsigned char *frame = skb->data;
 969         struct veth_port *port = (struct veth_port *) dev->priv;
 970         unsigned long flags;
 971         HvLpIndexMap lpmask;
 972
 973         if (! (frame[0] & 0x01)) {
 974                 /* unicast packet */
 975                 HvLpIndex rlp = frame[5];
 976
 977                 if ( ! ((1 << rlp) & port->lpar_map) ) {
 978                         dev_kfree_skb(skb);
 979                         return 0;
 980                 }
 981
 982                 lpmask = 1 << rlp;
 983         } else {
 984                 lpmask = port->lpar_map;
 985         }
 986
 987         lpmask = veth_transmit_to_many(skb, lpmask, dev);
 988
 989         if (! lpmask) {
 990                 dev_kfree_skb(skb);
 991         } else {
 992                 spin_lock_irqsave(&port->pending_gate, flags);
 993                 if (port->pending_skb) {
 994                         veth_error("%s: Tx while skb was pending!\n",
 995                                    dev->name);
 996                         dev_kfree_skb(skb);
 997                         return 1;
 998                 }
 999
1000                 port->pending_skb = skb;
1001                 port->pending_lpmask = lpmask;
1002                 netif_stop_queue(dev);
1003
1004                 spin_unlock_irqrestore(&port->pending_gate, flags);
1005         }
1006
1007         return 0;
1008 }
1009
1010 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1011                              struct veth_msg *msg)
1012 {
1013         u32 dma_address, dma_length;
1014
1015         if (test_and_clear_bit(0, &msg->in_use)) {
1016                 dma_address = msg->data.addr[0];
1017                 dma_length = msg->data.len[0];
1018
1019                 vio_unmap_single(iSeries_veth_dev, dma_address, dma_length,
1020                                  DMA_TO_DEVICE);
1021
1022                 if (msg->skb) {
1023                         dev_kfree_skb_any(msg->skb);
1024                         msg->skb = NULL;
1025                 }
1026
1027                 memset(&msg->data, 0, sizeof(msg->data));
1028                 veth_stack_push(cnx, msg);
1029         } else
1030                 if (cnx->state & VETH_STATE_OPEN)
1031                         veth_error("Bogus frames ack from lpar %d (#%d)\n",
1032                                    cnx->remote_lp, msg->token);
1033 }
1034
1035 static void veth_flush_pending(struct veth_lpar_connection *cnx)
1036 {
1037         int i;
1038         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1039                 struct net_device *dev = veth_dev[i];
1040                 struct veth_port *port;
1041                 unsigned long flags;
1042
1043                 if (! dev)
1044                         continue;
1045
1046                 port = (struct veth_port *)dev->priv;
1047
1048                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1049                         continue;
1050
1051                 spin_lock_irqsave(&port->pending_gate, flags);
1052                 if (port->pending_skb) {
1053                         port->pending_lpmask =
1054                                 veth_transmit_to_many(port->pending_skb,
1055                                                       port->pending_lpmask,
1056                                                       dev);
1057                         if (! port->pending_lpmask) {
1058                                 dev_kfree_skb_any(port->pending_skb);
1059                                 port->pending_skb = NULL;
1060                                 netif_start_queue(dev);
1061                         }
1062                 }
1063                 spin_unlock_irqrestore(&port->pending_gate, flags);
1064         }
1065 }
1066
1067 /*
1068  * Rx path
1069  */
1070
1071 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1072 {
1073         int wanted = 0;
1074         int i;
1075         unsigned long flags;
1076
1077         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1078                 return 1;
1079
1080         if (! (((char *) &mac_addr)[0] & 0x01))
1081                 return 0;
1082
1083         read_lock_irqsave(&port->mcast_gate, flags);
1084
1085         if (port->promiscuous || port->all_mcast) {
1086                 wanted = 1;
1087                 goto out;
1088         }
1089
1090         for (i = 0; i < port->num_mcast; ++i) {
1091                 if (port->mcast_addr[i] == mac_addr) {
1092                         wanted = 1;
1093                         break;
1094                 }
1095         }
1096
1097  out:
1098         read_unlock_irqrestore(&port->mcast_gate, flags);
1099
1100         return wanted;
1101 }
1102
1103 struct dma_chunk {
1104         u64 addr;
1105         u64 size;
1106 };
1107
1108 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1109
1110 static inline void veth_build_dma_list(struct dma_chunk *list,
1111                                        unsigned char *p, unsigned long length)
1112 {
1113         unsigned long done;
1114         int i = 1;
1115
1116         /* FIXME: skbs are continguous in real addresses.  Do we
1117          * really need to break it into PAGE_SIZE chunks, or can we do
1118          * it just at the granularity of iSeries real->absolute
1119          * mapping?  Indeed, given the way the allocator works, can we
1120          * count on them being absolutely contiguous? */
1121         list[0].addr = ISERIES_HV_ADDR(p);
1122         list[0].size = min(length,
1123                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1124
1125         done = list[0].size;
1126         while (done < length) {
1127                 list[i].addr = ISERIES_HV_ADDR(p + done);
1128                 list[i].size = min(length-done, PAGE_SIZE);
1129                 done += list[i].size;
1130                 i++;
1131         }
1132 }
1133
1134 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1135 {
1136         HvLpEvent_Rc rc;
1137
1138         rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1139                              0, &cnx->pending_acks);
1140
1141         if (rc != HvLpEvent_Rc_Good)
1142                 veth_error("Error 0x%x acking frames from lpar %d!\n",
1143                            (unsigned)rc, cnx->remote_lp);
1144
1145         cnx->num_pending_acks = 0;
1146         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1147 }
1148
1149 static void veth_receive(struct veth_lpar_connection *cnx,
1150                          struct VethLpEvent *event)
1151 {
1152         struct VethFramesData *senddata = &event->u.frames_data;
1153         int startchunk = 0;
1154         int nchunks;
1155         unsigned long flags;
1156         HvLpDma_Rc rc;
1157
1158         do {
1159                 u16 length = 0;
1160                 struct sk_buff *skb;
1161                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1162                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1163                 u64 dest;
1164                 HvLpVirtualLanIndex vlan;
1165                 struct net_device *dev;
1166                 struct veth_port *port;
1167
1168                 /* FIXME: do we need this? */
1169                 memset(local_list, 0, sizeof(local_list));
1170                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1171
1172                 /* a 0 address marks the end of the valid entries */
1173                 if (senddata->addr[startchunk] == 0)
1174                         break;
1175
1176                 /* make sure that we have at least 1 EOF entry in the
1177                  * remaining entries */
1178                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1179                         veth_error("missing EOF frag in event "
1180                                    "eofmask=0x%x startchunk=%d\n",
1181                                    (unsigned) senddata->eofmask, startchunk);
1182                         break;
1183                 }
1184
1185                 /* build list of chunks in this frame */
1186                 nchunks = 0;
1187                 do {
1188                         remote_list[nchunks].addr =
1189                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1190                         remote_list[nchunks].size =
1191                                 senddata->len[startchunk+nchunks];
1192                         length += remote_list[nchunks].size;
1193                 } while (! (senddata->eofmask &
1194                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1195
1196                 /* length == total length of all chunks */
1197                 /* nchunks == # of chunks in this frame */
1198
1199                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1200                         veth_error("Received oversize frame from lpar %d "
1201                                    "(length=%d)\n", cnx->remote_lp, length);
1202                         continue;
1203                 }
1204
1205                 skb = alloc_skb(length, GFP_ATOMIC);
1206                 if (!skb)
1207                         continue;
1208
1209                 veth_build_dma_list(local_list, skb->data, length);
1210
1211                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1212                                             event->base_event.xSourceLp,
1213                                             HvLpDma_Direction_RemoteToLocal,
1214                                             cnx->src_inst,
1215                                             cnx->dst_inst,
1216                                             HvLpDma_AddressType_RealAddress,
1217                                             HvLpDma_AddressType_TceIndex,
1218                                             ISERIES_HV_ADDR(&local_list),
1219                                             ISERIES_HV_ADDR(&remote_list),
1220                                             length);
1221                 if (rc != HvLpDma_Rc_Good) {
1222                         dev_kfree_skb_irq(skb);
1223                         continue;
1224                 }
1225
1226                 vlan = skb->data[9];
1227                 dev = veth_dev[vlan];
1228                 if (! dev)
1229                         /* Some earlier versions of the driver sent
1230                            broadcasts down all connections, even to
1231                            lpars that weren't on the relevant vlan.
1232                            So ignore packets belonging to a vlan we're
1233                            not on. */
1234                         continue;
1235
1236                 port = (struct veth_port *)dev->priv;
1237                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1238
1239                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1240                         dev_kfree_skb_irq(skb);
1241                         continue;
1242                 }
1243                 if (! veth_frame_wanted(port, dest)) {
1244                         dev_kfree_skb_irq(skb);
1245                         continue;
1246                 }
1247
1248                 skb_put(skb, length);
1249                 skb->dev = dev;
1250                 skb->protocol = eth_type_trans(skb, dev);
1251                 skb->ip_summed = CHECKSUM_NONE;
1252                 netif_rx(skb);  /* send it up */
1253                 port->stats.rx_packets++;
1254                 port->stats.rx_bytes += length;
1255         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1256
1257         /* Ack it */
1258         spin_lock_irqsave(&cnx->lock, flags);
1259         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1260
1261         cnx->pending_acks[cnx->num_pending_acks++] =
1262                 event->base_event.xCorrelationToken;
1263
1264         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1265              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1266                 veth_flush_acks(cnx);
1267
1268         spin_unlock_irqrestore(&cnx->lock, flags);
1269 }
1270
1271 static void veth_timed_ack(unsigned long ptr)
1272 {
1273         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1274         unsigned long flags;
1275
1276         /* Ack all the events */
1277         spin_lock_irqsave(&cnx->lock, flags);
1278         if (cnx->num_pending_acks > 0)
1279                 veth_flush_acks(cnx);
1280
1281         /* Reschedule the timer */
1282         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1283         add_timer(&cnx->ack_timer);
1284         spin_unlock_irqrestore(&cnx->lock, flags);
1285 }
1286
1287 /*
1288  * Module initialization/cleanup
1289  */
1290
1291 void __exit veth_module_cleanup(void)
1292 {
1293         int i;
1294
1295         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
1296                 veth_destroy_connection(i);
1297
1298         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1299
1300         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; ++i) {
1301                 struct net_device *dev = veth_dev[i];
1302
1303                 if (! dev)
1304                         continue;
1305
1306                 veth_dev[i] = NULL;
1307                 unregister_netdev(dev);
1308                 free_netdev(dev);
1309         }
1310 }
1311 module_exit(veth_module_cleanup);
1312
1313 int __init veth_module_init(void)
1314 {
1315         HvLpIndexMap vlan_map = HvLpConfig_getVirtualLanIndexMap();
1316         int i;
1317         int rc;
1318
1319         this_lp = HvLpConfig_getLpIndex_outline();
1320
1321         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1322                 rc = veth_init_connection(i);
1323                 if (rc != 0) {
1324                         veth_module_cleanup();
1325                         return rc;
1326                 }
1327         }
1328
1329         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; ++i) {
1330                 struct net_device *dev;
1331
1332                 if (! (vlan_map & (0x8000 >> i)))
1333                         continue;
1334
1335                 dev = veth_probe_one(i);
1336
1337                 if (! dev) {
1338                         veth_module_cleanup();
1339                         return rc;
1340                 }
1341
1342                 veth_dev[i] = dev;
1343         }
1344
1345         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1346                                   &veth_handle_event);
1347
1348         /* Start the state machine on each connection, to commence
1349          * link negotiation */
1350         for (i = 0; i < HVMAXARCHITECTEDLPS; i++)
1351                 if (veth_cnx[i])
1352                         veth_kick_statemachine(veth_cnx[i]);
1353
1354         return 0;
1355 }
1356 module_init(veth_module_init);