git.oblomov.eu Git - linux-2.6/blob - 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 #undef DEBUG
  83
  84 #include "iseries_veth.h"
  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         int in_use;
 121         struct sk_buff *skb;
 122         struct device *dev;
 123 };
 124
 125 struct veth_lpar_connection {
 126         HvLpIndex remote_lp;
 127         struct work_struct statemachine_wq;
 128         struct veth_msg *msgs;
 129         int num_events;
 130         struct VethCapData local_caps;
 131
 132         struct kobject kobject;
 133         struct timer_list ack_timer;
 134
 135         struct timer_list reset_timer;
 136         unsigned int reset_timeout;
 137         unsigned long last_contact;
 138         int outstanding_tx;
 139
 140         spinlock_t lock;
 141         unsigned long state;
 142         HvLpInstanceId src_inst;
 143         HvLpInstanceId dst_inst;
 144         struct VethLpEvent cap_event, cap_ack_event;
 145         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 146         u32 num_pending_acks;
 147
 148         int num_ack_events;
 149         struct VethCapData remote_caps;
 150         u32 ack_timeout;
 151
 152         struct veth_msg *msg_stack_head;
 153 };
 154
 155 struct veth_port {
 156         struct device *dev;
 157         struct net_device_stats stats;
 158         u64 mac_addr;
 159         HvLpIndexMap lpar_map;
 160
 161         spinlock_t pending_gate;
 162         struct sk_buff *pending_skb;
 163         HvLpIndexMap pending_lpmask;
 164
 165         rwlock_t mcast_gate;
 166         int promiscuous;
 167         int num_mcast;
 168         u64 mcast_addr[VETH_MAX_MCAST];
 169 };
 170
 171 static HvLpIndex this_lp;
 172 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
 173 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
 174
 175 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
 176 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
 177 static void veth_flush_pending(struct veth_lpar_connection *cnx);
 178 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
 179 static void veth_release_connection(struct kobject *kobject);
 180 static void veth_timed_ack(unsigned long ptr);
 181 static void veth_timed_reset(unsigned long ptr);
 182
 183 static struct kobj_type veth_lpar_connection_ktype = {
 184         .release        = veth_release_connection
 185 };
 186
 187 /*
 188  * Utility functions
 189  */
 190
 191 #define veth_info(fmt, args...) \
 192         printk(KERN_INFO "iseries_veth: " fmt, ## args)
 193
 194 #define veth_error(fmt, args...) \
 195         printk(KERN_ERR "iseries_veth: Error: " fmt, ## args)
 196
 197 #ifdef DEBUG
 198 #define veth_debug(fmt, args...) \
 199         printk(KERN_DEBUG "iseries_veth: " fmt, ## args)
 200 #else
 201 #define veth_debug(fmt, args...) do {} while (0)
 202 #endif
 203
 204 /* You must hold the connection's lock when you call this function. */
 205 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
 206                                    struct veth_msg *msg)
 207 {
 208         msg->next = cnx->msg_stack_head;
 209         cnx->msg_stack_head = msg;
 210 }
 211
 212 /* You must hold the connection's lock when you call this function. */
 213 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
 214 {
 215         struct veth_msg *msg;
 216
 217         msg = cnx->msg_stack_head;
 218         if (msg)
 219                 cnx->msg_stack_head = cnx->msg_stack_head->next;
 220
 221         return msg;
 222 }
 223
 224 static inline HvLpEvent_Rc
 225 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
 226                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
 227                  u64 token,
 228                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
 229 {
 230         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
 231                                              HvLpEvent_Type_VirtualLan,
 232                                              subtype, ackind, acktype,
 233                                              cnx->src_inst,
 234                                              cnx->dst_inst,
 235                                              token, data1, data2, data3,
 236                                              data4, data5);
 237 }
 238
 239 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
 240                                            u16 subtype, u64 token, void *data)
 241 {
 242         u64 *p = (u64 *) data;
 243
 244         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
 245                                 HvLpEvent_AckType_ImmediateAck,
 246                                 token, p[0], p[1], p[2], p[3], p[4]);
 247 }
 248
 249 struct veth_allocation {
 250         struct completion c;
 251         int num;
 252 };
 253
 254 static void veth_complete_allocation(void *parm, int number)
 255 {
 256         struct veth_allocation *vc = (struct veth_allocation *)parm;
 257
 258         vc->num = number;
 259         complete(&vc->c);
 260 }
 261
 262 static int veth_allocate_events(HvLpIndex rlp, int number)
 263 {
 264         struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
 265
 266         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
 267                             sizeof(struct VethLpEvent), number,
 268                             &veth_complete_allocation, &vc);
 269         wait_for_completion(&vc.c);
 270
 271         return vc.num;
 272 }
 273
 274 /*
 275  * LPAR connection code
 276  */
 277
 278 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 279 {
 280         schedule_work(&cnx->statemachine_wq);
 281 }
 282
 283 static void veth_take_cap(struct veth_lpar_connection *cnx,
 284                           struct VethLpEvent *event)
 285 {
 286         unsigned long flags;
 287
 288         spin_lock_irqsave(&cnx->lock, flags);
 289         /* Receiving caps may mean the other end has just come up, so
 290          * we need to reload the instance ID of the far end */
 291         cnx->dst_inst =
 292                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 293                                                   HvLpEvent_Type_VirtualLan);
 294
 295         if (cnx->state & VETH_STATE_GOTCAPS) {
 296                 veth_error("Received a second capabilities from LPAR %d.\n",
 297                            cnx->remote_lp);
 298                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 299                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 300         } else {
 301                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 302                 cnx->state |= VETH_STATE_GOTCAPS;
 303                 veth_kick_statemachine(cnx);
 304         }
 305         spin_unlock_irqrestore(&cnx->lock, flags);
 306 }
 307
 308 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 309                               struct VethLpEvent *event)
 310 {
 311         unsigned long flags;
 312
 313         spin_lock_irqsave(&cnx->lock, flags);
 314         if (cnx->state & VETH_STATE_GOTCAPACK) {
 315                 veth_error("Received a second capabilities ack from LPAR %d.\n",
 316                            cnx->remote_lp);
 317         } else {
 318                 memcpy(&cnx->cap_ack_event, event,
 319                        sizeof(&cnx->cap_ack_event));
 320                 cnx->state |= VETH_STATE_GOTCAPACK;
 321                 veth_kick_statemachine(cnx);
 322         }
 323         spin_unlock_irqrestore(&cnx->lock, flags);
 324 }
 325
 326 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 327                                   struct VethLpEvent *event)
 328 {
 329         unsigned long flags;
 330
 331         spin_lock_irqsave(&cnx->lock, flags);
 332         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
 333
 334         /* Avoid kicking the statemachine once we're shutdown.
 335          * It's unnecessary and it could break veth_stop_connection(). */
 336
 337         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 338                 cnx->state |= VETH_STATE_RESET;
 339                 veth_kick_statemachine(cnx);
 340         }
 341         spin_unlock_irqrestore(&cnx->lock, flags);
 342 }
 343
 344 static void veth_handle_ack(struct VethLpEvent *event)
 345 {
 346         HvLpIndex rlp = event->base_event.xTargetLp;
 347         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 348
 349         BUG_ON(! cnx);
 350
 351         switch (event->base_event.xSubtype) {
 352         case VethEventTypeCap:
 353                 veth_take_cap_ack(cnx, event);
 354                 break;
 355         case VethEventTypeMonitor:
 356                 veth_take_monitor_ack(cnx, event);
 357                 break;
 358         default:
 359                 veth_error("Unknown ack type %d from LPAR %d.\n",
 360                                 event->base_event.xSubtype, rlp);
 361         };
 362 }
 363
 364 static void veth_handle_int(struct VethLpEvent *event)
 365 {
 366         HvLpIndex rlp = event->base_event.xSourceLp;
 367         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 368         unsigned long flags;
 369         int i, acked = 0;
 370
 371         BUG_ON(! cnx);
 372
 373         switch (event->base_event.xSubtype) {
 374         case VethEventTypeCap:
 375                 veth_take_cap(cnx, event);
 376                 break;
 377         case VethEventTypeMonitor:
 378                 /* do nothing... this'll hang out here til we're dead,
 379                  * and the hypervisor will return it for us. */
 380                 break;
 381         case VethEventTypeFramesAck:
 382                 spin_lock_irqsave(&cnx->lock, flags);
 383
 384                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 385                         u16 msgnum = event->u.frames_ack_data.token[i];
 386
 387                         if (msgnum < VETH_NUMBUFFERS) {
 388                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 389                                 cnx->outstanding_tx--;
 390                                 acked++;
 391                         }
 392                 }
 393
 394                 if (acked > 0)
 395                         cnx->last_contact = jiffies;
 396
 397                 spin_unlock_irqrestore(&cnx->lock, flags);
 398
 399                 veth_flush_pending(cnx);
 400                 break;
 401         case VethEventTypeFrames:
 402                 veth_receive(cnx, event);
 403                 break;
 404         default:
 405                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
 406                                 event->base_event.xSubtype, rlp);
 407         };
 408 }
 409
 410 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
 411 {
 412         struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
 413
 414         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
 415                 veth_handle_ack(veth_event);
 416         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
 417                 veth_handle_int(veth_event);
 418 }
 419
 420 static int veth_process_caps(struct veth_lpar_connection *cnx)
 421 {
 422         struct VethCapData *remote_caps = &cnx->remote_caps;
 423         int num_acks_needed;
 424
 425         /* Convert timer to jiffies */
 426         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 427
 428         if ( (remote_caps->num_buffers == 0)
 429              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
 430              || (remote_caps->ack_threshold == 0)
 431              || (cnx->ack_timeout == 0) ) {
 432                 veth_error("Received incompatible capabilities from LPAR %d.\n",
 433                                 cnx->remote_lp);
 434                 return HvLpEvent_Rc_InvalidSubtypeData;
 435         }
 436
 437         num_acks_needed = (remote_caps->num_buffers
 438                            / remote_caps->ack_threshold) + 1;
 439
 440         /* FIXME: locking on num_ack_events? */
 441         if (cnx->num_ack_events < num_acks_needed) {
 442                 int num;
 443
 444                 num = veth_allocate_events(cnx->remote_lp,
 445                                            num_acks_needed-cnx->num_ack_events);
 446                 if (num > 0)
 447                         cnx->num_ack_events += num;
 448
 449                 if (cnx->num_ack_events < num_acks_needed) {
 450                         veth_error("Couldn't allocate enough ack events "
 451                                         "for LPAR %d.\n", cnx->remote_lp);
 452
 453                         return HvLpEvent_Rc_BufferNotAvailable;
 454                 }
 455         }
 456
 457
 458         return HvLpEvent_Rc_Good;
 459 }
 460
 461 /* FIXME: The gotos here are a bit dubious */
 462 static void veth_statemachine(void *p)
 463 {
 464         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
 465         int rlp = cnx->remote_lp;
 466         int rc;
 467
 468         spin_lock_irq(&cnx->lock);
 469
 470  restart:
 471         if (cnx->state & VETH_STATE_RESET) {
 472                 if (cnx->state & VETH_STATE_OPEN)
 473                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 474                                                      HvLpEvent_Type_VirtualLan);
 475
 476                 /*
 477                  * Reset ack data. This prevents the ack_timer actually
 478                  * doing anything, even if it runs one more time when
 479                  * we drop the lock below.
 480                  */
 481                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 482                 cnx->num_pending_acks = 0;
 483
 484                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 485                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 486                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 487                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 488
 489                 /* Clean up any leftover messages */
 490                 if (cnx->msgs) {
 491                         int i;
 492                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 493                                 veth_recycle_msg(cnx, cnx->msgs + i);
 494                 }
 495                 cnx->outstanding_tx = 0;
 496
 497                 /* Drop the lock so we can do stuff that might sleep or
 498                  * take other locks. */
 499                 spin_unlock_irq(&cnx->lock);
 500
 501                 del_timer_sync(&cnx->ack_timer);
 502                 del_timer_sync(&cnx->reset_timer);
 503
 504                 veth_flush_pending(cnx);
 505
 506                 spin_lock_irq(&cnx->lock);
 507
 508                 if (cnx->state & VETH_STATE_RESET)
 509                         goto restart;
 510
 511                 /* Hack, wait for the other end to reset itself. */
 512                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 513                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
 514                         goto out;
 515                 }
 516         }
 517
 518         if (cnx->state & VETH_STATE_SHUTDOWN)
 519                 /* It's all over, do nothing */
 520                 goto out;
 521
 522         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 523                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 524                         goto cant_cope;
 525
 526                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 527                 cnx->src_inst =
 528                         HvCallEvent_getSourceLpInstanceId(rlp,
 529                                                           HvLpEvent_Type_VirtualLan);
 530                 cnx->dst_inst =
 531                         HvCallEvent_getTargetLpInstanceId(rlp,
 532                                                           HvLpEvent_Type_VirtualLan);
 533                 cnx->state |= VETH_STATE_OPEN;
 534         }
 535
 536         if ( (cnx->state & VETH_STATE_OPEN)
 537              && !(cnx->state & VETH_STATE_SENTMON) ) {
 538                 rc = veth_signalevent(cnx, VethEventTypeMonitor,
 539                                       HvLpEvent_AckInd_DoAck,
 540                                       HvLpEvent_AckType_DeferredAck,
 541                                       0, 0, 0, 0, 0, 0);
 542
 543                 if (rc == HvLpEvent_Rc_Good) {
 544                         cnx->state |= VETH_STATE_SENTMON;
 545                 } else {
 546                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 547                              && (rc != HvLpEvent_Rc_PathClosed) )
 548                                 veth_error("Error sending monitor to LPAR %d, "
 549                                                 "rc = %d\n", rlp, rc);
 550
 551                         /* Oh well, hope we get a cap from the other
 552                          * end and do better when that kicks us */
 553                         goto out;
 554                 }
 555         }
 556
 557         if ( (cnx->state & VETH_STATE_OPEN)
 558              && !(cnx->state & VETH_STATE_SENTCAPS)) {
 559                 u64 *rawcap = (u64 *)&cnx->local_caps;
 560
 561                 rc = veth_signalevent(cnx, VethEventTypeCap,
 562                                       HvLpEvent_AckInd_DoAck,
 563                                       HvLpEvent_AckType_ImmediateAck,
 564                                       0, rawcap[0], rawcap[1], rawcap[2],
 565                                       rawcap[3], rawcap[4]);
 566
 567                 if (rc == HvLpEvent_Rc_Good) {
 568                         cnx->state |= VETH_STATE_SENTCAPS;
 569                 } else {
 570                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 571                              && (rc != HvLpEvent_Rc_PathClosed) )
 572                                 veth_error("Error sending caps to LPAR %d, "
 573                                                 "rc = %d\n", rlp, rc);
 574
 575                         /* Oh well, hope we get a cap from the other
 576                          * end and do better when that kicks us */
 577                         goto out;
 578                 }
 579         }
 580
 581         if ((cnx->state & VETH_STATE_GOTCAPS)
 582             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
 583                 struct VethCapData *remote_caps = &cnx->remote_caps;
 584
 585                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 586                        sizeof(*remote_caps));
 587
 588                 spin_unlock_irq(&cnx->lock);
 589                 rc = veth_process_caps(cnx);
 590                 spin_lock_irq(&cnx->lock);
 591
 592                 /* We dropped the lock, so recheck for anything which
 593                  * might mess us up */
 594                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 595                         goto restart;
 596
 597                 cnx->cap_event.base_event.xRc = rc;
 598                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 599                 if (rc == HvLpEvent_Rc_Good)
 600                         cnx->state |= VETH_STATE_SENTCAPACK;
 601                 else
 602                         goto cant_cope;
 603         }
 604
 605         if ((cnx->state & VETH_STATE_GOTCAPACK)
 606             && (cnx->state & VETH_STATE_GOTCAPS)
 607             && !(cnx->state & VETH_STATE_READY)) {
 608                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 609                         /* Start the ACK timer */
 610                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 611                         add_timer(&cnx->ack_timer);
 612                         cnx->state |= VETH_STATE_READY;
 613                 } else {
 614                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
 615                                         rlp, cnx->cap_ack_event.base_event.xRc);
 616                         goto cant_cope;
 617                 }
 618         }
 619
 620  out:
 621         spin_unlock_irq(&cnx->lock);
 622         return;
 623
 624  cant_cope:
 625         /* FIXME: we get here if something happens we really can't
 626          * cope with.  The link will never work once we get here, and
 627          * all we can do is not lock the rest of the system up */
 628         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
 629                         " (state = 0x%04lx)\n", rlp, cnx->state);
 630         cnx->state |= VETH_STATE_SHUTDOWN;
 631         spin_unlock_irq(&cnx->lock);
 632 }
 633
 634 static int veth_init_connection(u8 rlp)
 635 {
 636         struct veth_lpar_connection *cnx;
 637         struct veth_msg *msgs;
 638         int i, rc;
 639
 640         if ( (rlp == this_lp)
 641              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 642                 return 0;
 643
 644         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
 645         if (! cnx)
 646                 return -ENOMEM;
 647         memset(cnx, 0, sizeof(*cnx));
 648
 649         cnx->remote_lp = rlp;
 650         spin_lock_init(&cnx->lock);
 651         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
 652
 653         init_timer(&cnx->ack_timer);
 654         cnx->ack_timer.function = veth_timed_ack;
 655         cnx->ack_timer.data = (unsigned long) cnx;
 656
 657         init_timer(&cnx->reset_timer);
 658         cnx->reset_timer.function = veth_timed_reset;
 659         cnx->reset_timer.data = (unsigned long) cnx;
 660         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
 661
 662         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 663
 664         veth_cnx[rlp] = cnx;
 665
 666         /* This gets us 1 reference, which is held on behalf of the driver
 667          * infrastructure. It's released at module unload. */
 668         kobject_init(&cnx->kobject);
 669         cnx->kobject.ktype = &veth_lpar_connection_ktype;
 670         rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
 671         if (rc != 0)
 672                 return rc;
 673
 674         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
 675         if (! msgs) {
 676                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
 677                 return -ENOMEM;
 678         }
 679
 680         cnx->msgs = msgs;
 681         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
 682
 683         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 684                 msgs[i].token = i;
 685                 veth_stack_push(cnx, msgs + i);
 686         }
 687
 688         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 689
 690         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 691                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
 692                 return -ENOMEM;
 693         }
 694
 695         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 696         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 697         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 698
 699         return 0;
 700 }
 701
 702 static void veth_stop_connection(struct veth_lpar_connection *cnx)
 703 {
 704         if (!cnx)
 705                 return;
 706
 707         spin_lock_irq(&cnx->lock);
 708         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 709         veth_kick_statemachine(cnx);
 710         spin_unlock_irq(&cnx->lock);
 711
 712         /* There's a slim chance the reset code has just queued the
 713          * statemachine to run in five seconds. If so we need to cancel
 714          * that and requeue the work to run now. */
 715         if (cancel_delayed_work(&cnx->statemachine_wq)) {
 716                 spin_lock_irq(&cnx->lock);
 717                 veth_kick_statemachine(cnx);
 718                 spin_unlock_irq(&cnx->lock);
 719         }
 720
 721         /* Wait for the state machine to run. */
 722         flush_scheduled_work();
 723 }
 724
 725 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
 726 {
 727         if (!cnx)
 728                 return;
 729
 730         if (cnx->num_events > 0)
 731                 mf_deallocate_lp_events(cnx->remote_lp,
 732                                       HvLpEvent_Type_VirtualLan,
 733                                       cnx->num_events,
 734                                       NULL, NULL);
 735         if (cnx->num_ack_events > 0)
 736                 mf_deallocate_lp_events(cnx->remote_lp,
 737                                       HvLpEvent_Type_VirtualLan,
 738                                       cnx->num_ack_events,
 739                                       NULL, NULL);
 740
 741         kfree(cnx->msgs);
 742         veth_cnx[cnx->remote_lp] = NULL;
 743         kfree(cnx);
 744 }
 745
 746 static void veth_release_connection(struct kobject *kobj)
 747 {
 748         struct veth_lpar_connection *cnx;
 749         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 750         veth_stop_connection(cnx);
 751         veth_destroy_connection(cnx);
 752 }
 753
 754 /*
 755  * net_device code
 756  */
 757
 758 static int veth_open(struct net_device *dev)
 759 {
 760         struct veth_port *port = (struct veth_port *) dev->priv;
 761
 762         memset(&port->stats, 0, sizeof (port->stats));
 763         netif_start_queue(dev);
 764         return 0;
 765 }
 766
 767 static int veth_close(struct net_device *dev)
 768 {
 769         netif_stop_queue(dev);
 770         return 0;
 771 }
 772
 773 static struct net_device_stats *veth_get_stats(struct net_device *dev)
 774 {
 775         struct veth_port *port = (struct veth_port *) dev->priv;
 776
 777         return &port->stats;
 778 }
 779
 780 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 781 {
 782         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 783                 return -EINVAL;
 784         dev->mtu = new_mtu;
 785         return 0;
 786 }
 787
 788 static void veth_set_multicast_list(struct net_device *dev)
 789 {
 790         struct veth_port *port = (struct veth_port *) dev->priv;
 791         unsigned long flags;
 792
 793         write_lock_irqsave(&port->mcast_gate, flags);
 794
 795         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
 796                         (dev->mc_count > VETH_MAX_MCAST)) {
 797                 port->promiscuous = 1;
 798         } else {
 799                 struct dev_mc_list *dmi = dev->mc_list;
 800                 int i;
 801
 802                 port->promiscuous = 0;
 803
 804                 /* Update table */
 805                 port->num_mcast = 0;
 806
 807                 for (i = 0; i < dev->mc_count; i++) {
 808                         u8 *addr = dmi->dmi_addr;
 809                         u64 xaddr = 0;
 810
 811                         if (addr[0] & 0x01) {/* multicast address? */
 812                                 memcpy(&xaddr, addr, ETH_ALEN);
 813                                 port->mcast_addr[port->num_mcast] = xaddr;
 814                                 port->num_mcast++;
 815                         }
 816                         dmi = dmi->next;
 817                 }
 818         }
 819
 820         write_unlock_irqrestore(&port->mcast_gate, flags);
 821 }
 822
 823 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 824 {
 825         strncpy(info->driver, "veth", sizeof(info->driver) - 1);
 826         info->driver[sizeof(info->driver) - 1] = '\0';
 827         strncpy(info->version, "1.0", sizeof(info->version) - 1);
 828 }
 829
 830 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
 831 {
 832         ecmd->supported = (SUPPORTED_1000baseT_Full
 833                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 834         ecmd->advertising = (SUPPORTED_1000baseT_Full
 835                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 836         ecmd->port = PORT_FIBRE;
 837         ecmd->transceiver = XCVR_INTERNAL;
 838         ecmd->phy_address = 0;
 839         ecmd->speed = SPEED_1000;
 840         ecmd->duplex = DUPLEX_FULL;
 841         ecmd->autoneg = AUTONEG_ENABLE;
 842         ecmd->maxtxpkt = 120;
 843         ecmd->maxrxpkt = 120;
 844         return 0;
 845 }
 846
 847 static u32 veth_get_link(struct net_device *dev)
 848 {
 849         return 1;
 850 }
 851
 852 static struct ethtool_ops ops = {
 853         .get_drvinfo = veth_get_drvinfo,
 854         .get_settings = veth_get_settings,
 855         .get_link = veth_get_link,
 856 };
 857
 858 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
 859 {
 860         struct net_device *dev;
 861         struct veth_port *port;
 862         int i, rc;
 863
 864         dev = alloc_etherdev(sizeof (struct veth_port));
 865         if (! dev) {
 866                 veth_error("Unable to allocate net_device structure!\n");
 867                 return NULL;
 868         }
 869
 870         port = (struct veth_port *) dev->priv;
 871
 872         spin_lock_init(&port->pending_gate);
 873         rwlock_init(&port->mcast_gate);
 874
 875         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
 876                 HvLpVirtualLanIndexMap map;
 877
 878                 if (i == this_lp)
 879                         continue;
 880                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
 881                 if (map & (0x8000 >> vlan))
 882                         port->lpar_map |= (1 << i);
 883         }
 884         port->dev = vdev;
 885
 886         dev->dev_addr[0] = 0x02;
 887         dev->dev_addr[1] = 0x01;
 888         dev->dev_addr[2] = 0xff;
 889         dev->dev_addr[3] = vlan;
 890         dev->dev_addr[4] = 0xff;
 891         dev->dev_addr[5] = this_lp;
 892
 893         dev->mtu = VETH_MAX_MTU;
 894
 895         memcpy(&port->mac_addr, dev->dev_addr, 6);
 896
 897         dev->open = veth_open;
 898         dev->hard_start_xmit = veth_start_xmit;
 899         dev->stop = veth_close;
 900         dev->get_stats = veth_get_stats;
 901         dev->change_mtu = veth_change_mtu;
 902         dev->set_mac_address = NULL;
 903         dev->set_multicast_list = veth_set_multicast_list;
 904         SET_ETHTOOL_OPS(dev, &ops);
 905
 906         SET_NETDEV_DEV(dev, vdev);
 907
 908         rc = register_netdev(dev);
 909         if (rc != 0) {
 910                 veth_error("Failed registering net device for vlan%d.\n", vlan);
 911                 free_netdev(dev);
 912                 return NULL;
 913         }
 914
 915         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
 916                         dev->name, vlan, port->lpar_map);
 917
 918         return dev;
 919 }
 920
 921 /*
 922  * Tx path
 923  */
 924
 925 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
 926                                 struct net_device *dev)
 927 {
 928         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 929         struct veth_port *port = (struct veth_port *) dev->priv;
 930         HvLpEvent_Rc rc;
 931         struct veth_msg *msg = NULL;
 932         int err = 0;
 933         unsigned long flags;
 934
 935         if (! cnx) {
 936                 port->stats.tx_errors++;
 937                 dev_kfree_skb(skb);
 938                 return 0;
 939         }
 940
 941         spin_lock_irqsave(&cnx->lock, flags);
 942
 943         if (! (cnx->state & VETH_STATE_READY))
 944                 goto drop;
 945
 946         if ((skb->len - 14) > VETH_MAX_MTU)
 947                 goto drop;
 948
 949         msg = veth_stack_pop(cnx);
 950
 951         if (! msg) {
 952                 err = 1;
 953                 goto drop;
 954         }
 955
 956         msg->in_use = 1;
 957
 958         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
 959                                 skb->len, DMA_TO_DEVICE);
 960
 961         if (dma_mapping_error(msg->data.addr[0]))
 962                 goto recycle_and_drop;
 963
 964         /* Is it really necessary to check the length and address
 965          * fields of the first entry here? */
 966         msg->skb = skb;
 967         msg->dev = port->dev;
 968         msg->data.len[0] = skb->len;
 969         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
 970
 971         rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
 972
 973         if (rc != HvLpEvent_Rc_Good)
 974                 goto recycle_and_drop;
 975
 976         /* If the timer's not already running, start it now. */
 977         if (0 == cnx->outstanding_tx)
 978                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
 979
 980         cnx->last_contact = jiffies;
 981         cnx->outstanding_tx++;
 982
 983         spin_unlock_irqrestore(&cnx->lock, flags);
 984         return 0;
 985
 986  recycle_and_drop:
 987         /* we free the skb below, so tell veth_recycle_msg() not to. */
 988         msg->skb = NULL;
 989         veth_recycle_msg(cnx, msg);
 990  drop:
 991         port->stats.tx_errors++;
 992         dev_kfree_skb(skb);
 993         spin_unlock_irqrestore(&cnx->lock, flags);
 994         return err;
 995 }
 996
 997 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
 998                                           HvLpIndexMap lpmask,
 999                                           struct net_device *dev)
1000 {
1001         struct veth_port *port = (struct veth_port *) dev->priv;
1002         int i;
1003         int rc;
1004
1005         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1006                 if ((lpmask & (1 << i)) == 0)
1007                         continue;
1008
1009                 rc = veth_transmit_to_one(skb_get(skb), i, dev);
1010                 if (! rc)
1011                         lpmask &= ~(1<<i);
1012         }
1013
1014         if (! lpmask) {
1015                 port->stats.tx_packets++;
1016                 port->stats.tx_bytes += skb->len;
1017         }
1018
1019         return lpmask;
1020 }
1021
1022 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1023 {
1024         unsigned char *frame = skb->data;
1025         struct veth_port *port = (struct veth_port *) dev->priv;
1026         unsigned long flags;
1027         HvLpIndexMap lpmask;
1028
1029         if (! (frame[0] & 0x01)) {
1030                 /* unicast packet */
1031                 HvLpIndex rlp = frame[5];
1032
1033                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1034                         dev_kfree_skb(skb);
1035                         return 0;
1036                 }
1037
1038                 lpmask = 1 << rlp;
1039         } else {
1040                 lpmask = port->lpar_map;
1041         }
1042
1043         spin_lock_irqsave(&port->pending_gate, flags);
1044
1045         lpmask = veth_transmit_to_many(skb, lpmask, dev);
1046
1047         if (! lpmask) {
1048                 dev_kfree_skb(skb);
1049         } else {
1050                 if (port->pending_skb) {
1051                         veth_error("%s: TX while skb was pending!\n",
1052                                    dev->name);
1053                         dev_kfree_skb(skb);
1054                         spin_unlock_irqrestore(&port->pending_gate, flags);
1055                         return 1;
1056                 }
1057
1058                 port->pending_skb = skb;
1059                 port->pending_lpmask = lpmask;
1060                 netif_stop_queue(dev);
1061         }
1062
1063         spin_unlock_irqrestore(&port->pending_gate, flags);
1064
1065         return 0;
1066 }
1067
1068 /* You must hold the connection's lock when you call this function. */
1069 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1070                              struct veth_msg *msg)
1071 {
1072         u32 dma_address, dma_length;
1073
1074         if (msg->in_use) {
1075                 msg->in_use = 0;
1076                 dma_address = msg->data.addr[0];
1077                 dma_length = msg->data.len[0];
1078
1079                 if (!dma_mapping_error(dma_address))
1080                         dma_unmap_single(msg->dev, dma_address, dma_length,
1081                                         DMA_TO_DEVICE);
1082
1083                 if (msg->skb) {
1084                         dev_kfree_skb_any(msg->skb);
1085                         msg->skb = NULL;
1086                 }
1087
1088                 memset(&msg->data, 0, sizeof(msg->data));
1089                 veth_stack_push(cnx, msg);
1090         } else if (cnx->state & VETH_STATE_OPEN) {
1091                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1092                                 cnx->remote_lp, msg->token);
1093         }
1094 }
1095
1096 static void veth_flush_pending(struct veth_lpar_connection *cnx)
1097 {
1098         int i;
1099         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1100                 struct net_device *dev = veth_dev[i];
1101                 struct veth_port *port;
1102                 unsigned long flags;
1103
1104                 if (! dev)
1105                         continue;
1106
1107                 port = (struct veth_port *)dev->priv;
1108
1109                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1110                         continue;
1111
1112                 spin_lock_irqsave(&port->pending_gate, flags);
1113                 if (port->pending_skb) {
1114                         port->pending_lpmask =
1115                                 veth_transmit_to_many(port->pending_skb,
1116                                                       port->pending_lpmask,
1117                                                       dev);
1118                         if (! port->pending_lpmask) {
1119                                 dev_kfree_skb_any(port->pending_skb);
1120                                 port->pending_skb = NULL;
1121                                 netif_wake_queue(dev);
1122                         }
1123                 }
1124                 spin_unlock_irqrestore(&port->pending_gate, flags);
1125         }
1126 }
1127
1128 static void veth_timed_reset(unsigned long ptr)
1129 {
1130         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1131         unsigned long trigger_time, flags;
1132
1133         /* FIXME is it possible this fires after veth_stop_connection()?
1134          * That would reschedule the statemachine for 5 seconds and probably
1135          * execute it after the module's been unloaded. Hmm. */
1136
1137         spin_lock_irqsave(&cnx->lock, flags);
1138
1139         if (cnx->outstanding_tx > 0) {
1140                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1141
1142                 if (trigger_time < jiffies) {
1143                         cnx->state |= VETH_STATE_RESET;
1144                         veth_kick_statemachine(cnx);
1145                         veth_error("%d packets not acked by LPAR %d within %d "
1146                                         "seconds, resetting.\n",
1147                                         cnx->outstanding_tx, cnx->remote_lp,
1148                                         cnx->reset_timeout / HZ);
1149                 } else {
1150                         /* Reschedule the timer */
1151                         trigger_time = jiffies + cnx->reset_timeout;
1152                         mod_timer(&cnx->reset_timer, trigger_time);
1153                 }
1154         }
1155
1156         spin_unlock_irqrestore(&cnx->lock, flags);
1157 }
1158
1159 /*
1160  * Rx path
1161  */
1162
1163 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1164 {
1165         int wanted = 0;
1166         int i;
1167         unsigned long flags;
1168
1169         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1170                 return 1;
1171
1172         read_lock_irqsave(&port->mcast_gate, flags);
1173
1174         if (port->promiscuous) {
1175                 wanted = 1;
1176                 goto out;
1177         }
1178
1179         for (i = 0; i < port->num_mcast; ++i) {
1180                 if (port->mcast_addr[i] == mac_addr) {
1181                         wanted = 1;
1182                         break;
1183                 }
1184         }
1185
1186  out:
1187         read_unlock_irqrestore(&port->mcast_gate, flags);
1188
1189         return wanted;
1190 }
1191
1192 struct dma_chunk {
1193         u64 addr;
1194         u64 size;
1195 };
1196
1197 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1198
1199 static inline void veth_build_dma_list(struct dma_chunk *list,
1200                                        unsigned char *p, unsigned long length)
1201 {
1202         unsigned long done;
1203         int i = 1;
1204
1205         /* FIXME: skbs are continguous in real addresses.  Do we
1206          * really need to break it into PAGE_SIZE chunks, or can we do
1207          * it just at the granularity of iSeries real->absolute
1208          * mapping?  Indeed, given the way the allocator works, can we
1209          * count on them being absolutely contiguous? */
1210         list[0].addr = ISERIES_HV_ADDR(p);
1211         list[0].size = min(length,
1212                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1213
1214         done = list[0].size;
1215         while (done < length) {
1216                 list[i].addr = ISERIES_HV_ADDR(p + done);
1217                 list[i].size = min(length-done, PAGE_SIZE);
1218                 done += list[i].size;
1219                 i++;
1220         }
1221 }
1222
1223 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1224 {
1225         HvLpEvent_Rc rc;
1226
1227         rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1228                              0, &cnx->pending_acks);
1229
1230         if (rc != HvLpEvent_Rc_Good)
1231                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1232                                 cnx->remote_lp, (int)rc);
1233
1234         cnx->num_pending_acks = 0;
1235         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1236 }
1237
1238 static void veth_receive(struct veth_lpar_connection *cnx,
1239                          struct VethLpEvent *event)
1240 {
1241         struct VethFramesData *senddata = &event->u.frames_data;
1242         int startchunk = 0;
1243         int nchunks;
1244         unsigned long flags;
1245         HvLpDma_Rc rc;
1246
1247         do {
1248                 u16 length = 0;
1249                 struct sk_buff *skb;
1250                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1251                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1252                 u64 dest;
1253                 HvLpVirtualLanIndex vlan;
1254                 struct net_device *dev;
1255                 struct veth_port *port;
1256
1257                 /* FIXME: do we need this? */
1258                 memset(local_list, 0, sizeof(local_list));
1259                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1260
1261                 /* a 0 address marks the end of the valid entries */
1262                 if (senddata->addr[startchunk] == 0)
1263                         break;
1264
1265                 /* make sure that we have at least 1 EOF entry in the
1266                  * remaining entries */
1267                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1268                         veth_error("Missing EOF fragment in event "
1269                                         "eofmask = 0x%x startchunk = %d\n",
1270                                         (unsigned)senddata->eofmask,
1271                                         startchunk);
1272                         break;
1273                 }
1274
1275                 /* build list of chunks in this frame */
1276                 nchunks = 0;
1277                 do {
1278                         remote_list[nchunks].addr =
1279                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1280                         remote_list[nchunks].size =
1281                                 senddata->len[startchunk+nchunks];
1282                         length += remote_list[nchunks].size;
1283                 } while (! (senddata->eofmask &
1284                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1285
1286                 /* length == total length of all chunks */
1287                 /* nchunks == # of chunks in this frame */
1288
1289                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1290                         veth_error("Received oversize frame from LPAR %d "
1291                                         "(length = %d)\n",
1292                                         cnx->remote_lp, length);
1293                         continue;
1294                 }
1295
1296                 skb = alloc_skb(length, GFP_ATOMIC);
1297                 if (!skb)
1298                         continue;
1299
1300                 veth_build_dma_list(local_list, skb->data, length);
1301
1302                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1303                                             event->base_event.xSourceLp,
1304                                             HvLpDma_Direction_RemoteToLocal,
1305                                             cnx->src_inst,
1306                                             cnx->dst_inst,
1307                                             HvLpDma_AddressType_RealAddress,
1308                                             HvLpDma_AddressType_TceIndex,
1309                                             ISERIES_HV_ADDR(&local_list),
1310                                             ISERIES_HV_ADDR(&remote_list),
1311                                             length);
1312                 if (rc != HvLpDma_Rc_Good) {
1313                         dev_kfree_skb_irq(skb);
1314                         continue;
1315                 }
1316
1317                 vlan = skb->data[9];
1318                 dev = veth_dev[vlan];
1319                 if (! dev) {
1320                         /*
1321                          * Some earlier versions of the driver sent
1322                          * broadcasts down all connections, even to lpars
1323                          * that weren't on the relevant vlan. So ignore
1324                          * packets belonging to a vlan we're not on.
1325                          * We can also be here if we receive packets while
1326                          * the driver is going down, because then dev is NULL.
1327                          */
1328                         dev_kfree_skb_irq(skb);
1329                         continue;
1330                 }
1331
1332                 port = (struct veth_port *)dev->priv;
1333                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1334
1335                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1336                         dev_kfree_skb_irq(skb);
1337                         continue;
1338                 }
1339                 if (! veth_frame_wanted(port, dest)) {
1340                         dev_kfree_skb_irq(skb);
1341                         continue;
1342                 }
1343
1344                 skb_put(skb, length);
1345                 skb->dev = dev;
1346                 skb->protocol = eth_type_trans(skb, dev);
1347                 skb->ip_summed = CHECKSUM_NONE;
1348                 netif_rx(skb);  /* send it up */
1349                 port->stats.rx_packets++;
1350                 port->stats.rx_bytes += length;
1351         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1352
1353         /* Ack it */
1354         spin_lock_irqsave(&cnx->lock, flags);
1355         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1356
1357         cnx->pending_acks[cnx->num_pending_acks++] =
1358                 event->base_event.xCorrelationToken;
1359
1360         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1361              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1362                 veth_flush_acks(cnx);
1363
1364         spin_unlock_irqrestore(&cnx->lock, flags);
1365 }
1366
1367 static void veth_timed_ack(unsigned long ptr)
1368 {
1369         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1370         unsigned long flags;
1371
1372         /* Ack all the events */
1373         spin_lock_irqsave(&cnx->lock, flags);
1374         if (cnx->num_pending_acks > 0)
1375                 veth_flush_acks(cnx);
1376
1377         /* Reschedule the timer */
1378         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1379         add_timer(&cnx->ack_timer);
1380         spin_unlock_irqrestore(&cnx->lock, flags);
1381 }
1382
1383 static int veth_remove(struct vio_dev *vdev)
1384 {
1385         struct veth_lpar_connection *cnx;
1386         struct net_device *dev;
1387         struct veth_port *port;
1388         int i;
1389
1390         dev = veth_dev[vdev->unit_address];
1391
1392         if (! dev)
1393                 return 0;
1394
1395         port = netdev_priv(dev);
1396
1397         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1398                 cnx = veth_cnx[i];
1399
1400                 if (cnx && (port->lpar_map & (1 << i))) {
1401                         /* Drop our reference to connections on our VLAN */
1402                         kobject_put(&cnx->kobject);
1403                 }
1404         }
1405
1406         veth_dev[vdev->unit_address] = NULL;
1407         unregister_netdev(dev);
1408         free_netdev(dev);
1409
1410         return 0;
1411 }
1412
1413 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1414 {
1415         int i = vdev->unit_address;
1416         struct net_device *dev;
1417         struct veth_port *port;
1418
1419         dev = veth_probe_one(i, &vdev->dev);
1420         if (dev == NULL) {
1421                 veth_remove(vdev);
1422                 return 1;
1423         }
1424         veth_dev[i] = dev;
1425
1426         port = (struct veth_port*)netdev_priv(dev);
1427
1428         /* Start the state machine on each connection on this vlan. If we're
1429          * the first dev to do so this will commence link negotiation */
1430         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1431                 struct veth_lpar_connection *cnx;
1432
1433                 if (! (port->lpar_map & (1 << i)))
1434                         continue;
1435
1436                 cnx = veth_cnx[i];
1437                 if (!cnx)
1438                         continue;
1439
1440                 kobject_get(&cnx->kobject);
1441                 veth_kick_statemachine(cnx);
1442         }
1443
1444         return 0;
1445 }
1446
1447 /**
1448  * veth_device_table: Used by vio.c to match devices that we
1449  * support.
1450  */
1451 static struct vio_device_id veth_device_table[] __devinitdata = {
1452         { "vlan", "" },
1453         { "", "" }
1454 };
1455 MODULE_DEVICE_TABLE(vio, veth_device_table);
1456
1457 static struct vio_driver veth_driver = {
1458         .name = "iseries_veth",
1459         .id_table = veth_device_table,
1460         .probe = veth_probe,
1461         .remove = veth_remove
1462 };
1463
1464 /*
1465  * Module initialization/cleanup
1466  */
1467
1468 void __exit veth_module_cleanup(void)
1469 {
1470         int i;
1471         struct veth_lpar_connection *cnx;
1472
1473         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1474         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1475
1476         /* Make sure any work queued from Hypervisor callbacks is finished. */
1477         flush_scheduled_work();
1478
1479         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1480                 cnx = veth_cnx[i];
1481
1482                 if (!cnx)
1483                         continue;
1484
1485                 /* Drop the driver's reference to the connection */
1486                 kobject_put(&cnx->kobject);
1487         }
1488
1489         /* Unregister the driver, which will close all the netdevs and stop
1490          * the connections when they're no longer referenced. */
1491         vio_unregister_driver(&veth_driver);
1492 }
1493 module_exit(veth_module_cleanup);
1494
1495 int __init veth_module_init(void)
1496 {
1497         int i;
1498         int rc;
1499
1500         this_lp = HvLpConfig_getLpIndex_outline();
1501
1502         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1503                 rc = veth_init_connection(i);
1504                 if (rc != 0)
1505                         goto error;
1506         }
1507
1508         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1509                                   &veth_handle_event);
1510
1511         rc = vio_register_driver(&veth_driver);
1512         if (rc != 0)
1513                 goto error;
1514
1515         return 0;
1516
1517 error:
1518         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1519                 veth_destroy_connection(veth_cnx[i]);
1520         }
1521
1522         return rc;
1523 }
1524 module_init(veth_module_init);