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 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  85 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  86 MODULE_LICENSE("GPL");
  87
  88 #define VETH_EVENT_CAP  (0)
  89 #define VETH_EVENT_FRAMES       (1)
  90 #define VETH_EVENT_MONITOR      (2)
  91 #define VETH_EVENT_FRAMES_ACK   (3)
  92
  93 #define VETH_MAX_ACKS_PER_MSG   (20)
  94 #define VETH_MAX_FRAMES_PER_MSG (6)
  95
  96 struct veth_frames_data {
  97         u32 addr[VETH_MAX_FRAMES_PER_MSG];
  98         u16 len[VETH_MAX_FRAMES_PER_MSG];
  99         u32 eofmask;
 100 };
 101 #define VETH_EOF_SHIFT          (32-VETH_MAX_FRAMES_PER_MSG)
 102
 103 struct veth_frames_ack_data {
 104         u16 token[VETH_MAX_ACKS_PER_MSG];
 105 };
 106
 107 struct veth_cap_data {
 108         u8 caps_version;
 109         u8 rsvd1;
 110         u16 num_buffers;
 111         u16 ack_threshold;
 112         u16 rsvd2;
 113         u32 ack_timeout;
 114         u32 rsvd3;
 115         u64 rsvd4[3];
 116 };
 117
 118 struct veth_lpevent {
 119         struct HvLpEvent base_event;
 120         union {
 121                 struct veth_cap_data caps_data;
 122                 struct veth_frames_data frames_data;
 123                 struct veth_frames_ack_data frames_ack_data;
 124         } u;
 125
 126 };
 127
 128 #define DRV_NAME        "iseries_veth"
 129 #define DRV_VERSION     "2.0"
 130
 131 #define VETH_NUMBUFFERS         (120)
 132 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
 133 #define VETH_MAX_MCAST          (12)
 134
 135 #define VETH_MAX_MTU            (9000)
 136
 137 #if VETH_NUMBUFFERS < 10
 138 #define ACK_THRESHOLD           (1)
 139 #elif VETH_NUMBUFFERS < 20
 140 #define ACK_THRESHOLD           (4)
 141 #elif VETH_NUMBUFFERS < 40
 142 #define ACK_THRESHOLD           (10)
 143 #else
 144 #define ACK_THRESHOLD           (20)
 145 #endif
 146
 147 #define VETH_STATE_SHUTDOWN     (0x0001)
 148 #define VETH_STATE_OPEN         (0x0002)
 149 #define VETH_STATE_RESET        (0x0004)
 150 #define VETH_STATE_SENTMON      (0x0008)
 151 #define VETH_STATE_SENTCAPS     (0x0010)
 152 #define VETH_STATE_GOTCAPACK    (0x0020)
 153 #define VETH_STATE_GOTCAPS      (0x0040)
 154 #define VETH_STATE_SENTCAPACK   (0x0080)
 155 #define VETH_STATE_READY        (0x0100)
 156
 157 struct veth_msg {
 158         struct veth_msg *next;
 159         struct veth_frames_data data;
 160         int token;
 161         int in_use;
 162         struct sk_buff *skb;
 163         struct device *dev;
 164 };
 165
 166 struct veth_lpar_connection {
 167         HvLpIndex remote_lp;
 168         struct work_struct statemachine_wq;
 169         struct veth_msg *msgs;
 170         int num_events;
 171         struct veth_cap_data local_caps;
 172
 173         struct kobject kobject;
 174         struct timer_list ack_timer;
 175
 176         struct timer_list reset_timer;
 177         unsigned int reset_timeout;
 178         unsigned long last_contact;
 179         int outstanding_tx;
 180
 181         spinlock_t lock;
 182         unsigned long state;
 183         HvLpInstanceId src_inst;
 184         HvLpInstanceId dst_inst;
 185         struct veth_lpevent cap_event, cap_ack_event;
 186         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 187         u32 num_pending_acks;
 188
 189         int num_ack_events;
 190         struct veth_cap_data remote_caps;
 191         u32 ack_timeout;
 192
 193         struct veth_msg *msg_stack_head;
 194 };
 195
 196 struct veth_port {
 197         struct device *dev;
 198         struct net_device_stats stats;
 199         u64 mac_addr;
 200         HvLpIndexMap lpar_map;
 201
 202         /* queue_lock protects the stopped_map and dev's queue. */
 203         spinlock_t queue_lock;
 204         HvLpIndexMap stopped_map;
 205
 206         /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
 207         rwlock_t mcast_gate;
 208         int promiscuous;
 209         int num_mcast;
 210         u64 mcast_addr[VETH_MAX_MCAST];
 211
 212         struct kobject kobject;
 213 };
 214
 215 static HvLpIndex this_lp;
 216 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
 217 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
 218
 219 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
 220 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
 221 static void veth_wake_queues(struct veth_lpar_connection *cnx);
 222 static void veth_stop_queues(struct veth_lpar_connection *cnx);
 223 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
 224 static void veth_release_connection(struct kobject *kobject);
 225 static void veth_timed_ack(unsigned long ptr);
 226 static void veth_timed_reset(unsigned long ptr);
 227
 228 /*
 229  * Utility functions
 230  */
 231
 232 #define veth_info(fmt, args...) \
 233         printk(KERN_INFO DRV_NAME ": " fmt, ## args)
 234
 235 #define veth_error(fmt, args...) \
 236         printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
 237
 238 #ifdef DEBUG
 239 #define veth_debug(fmt, args...) \
 240         printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
 241 #else
 242 #define veth_debug(fmt, args...) do {} while (0)
 243 #endif
 244
 245 /* You must hold the connection's lock when you call this function. */
 246 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
 247                                    struct veth_msg *msg)
 248 {
 249         msg->next = cnx->msg_stack_head;
 250         cnx->msg_stack_head = msg;
 251 }
 252
 253 /* You must hold the connection's lock when you call this function. */
 254 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
 255 {
 256         struct veth_msg *msg;
 257
 258         msg = cnx->msg_stack_head;
 259         if (msg)
 260                 cnx->msg_stack_head = cnx->msg_stack_head->next;
 261
 262         return msg;
 263 }
 264
 265 /* You must hold the connection's lock when you call this function. */
 266 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
 267 {
 268         return cnx->msg_stack_head == NULL;
 269 }
 270
 271 static inline HvLpEvent_Rc
 272 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
 273                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
 274                  u64 token,
 275                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
 276 {
 277         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
 278                                              HvLpEvent_Type_VirtualLan,
 279                                              subtype, ackind, acktype,
 280                                              cnx->src_inst,
 281                                              cnx->dst_inst,
 282                                              token, data1, data2, data3,
 283                                              data4, data5);
 284 }
 285
 286 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
 287                                            u16 subtype, u64 token, void *data)
 288 {
 289         u64 *p = (u64 *) data;
 290
 291         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
 292                                 HvLpEvent_AckType_ImmediateAck,
 293                                 token, p[0], p[1], p[2], p[3], p[4]);
 294 }
 295
 296 struct veth_allocation {
 297         struct completion c;
 298         int num;
 299 };
 300
 301 static void veth_complete_allocation(void *parm, int number)
 302 {
 303         struct veth_allocation *vc = (struct veth_allocation *)parm;
 304
 305         vc->num = number;
 306         complete(&vc->c);
 307 }
 308
 309 static int veth_allocate_events(HvLpIndex rlp, int number)
 310 {
 311         struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
 312
 313         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
 314                             sizeof(struct veth_lpevent), number,
 315                             &veth_complete_allocation, &vc);
 316         wait_for_completion(&vc.c);
 317
 318         return vc.num;
 319 }
 320
 321 /*
 322  * sysfs support
 323  */
 324
 325 struct veth_cnx_attribute {
 326         struct attribute attr;
 327         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
 328         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
 329 };
 330
 331 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
 332                 struct attribute *attr, char *buf)
 333 {
 334         struct veth_cnx_attribute *cnx_attr;
 335         struct veth_lpar_connection *cnx;
 336
 337         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
 338         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 339
 340         if (!cnx_attr->show)
 341                 return -EIO;
 342
 343         return cnx_attr->show(cnx, buf);
 344 }
 345
 346 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
 347 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
 348 {                                                                       \
 349         return sprintf(buf, _format, _expression);                      \
 350 }                                                                       \
 351 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
 352
 353 #define SIMPLE_CNX_ATTR(_name)  \
 354         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
 355
 356 SIMPLE_CNX_ATTR(outstanding_tx);
 357 SIMPLE_CNX_ATTR(remote_lp);
 358 SIMPLE_CNX_ATTR(num_events);
 359 SIMPLE_CNX_ATTR(src_inst);
 360 SIMPLE_CNX_ATTR(dst_inst);
 361 SIMPLE_CNX_ATTR(num_pending_acks);
 362 SIMPLE_CNX_ATTR(num_ack_events);
 363 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
 364 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
 365 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
 366 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
 367                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
 368
 369 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
 370
 371 static struct attribute *veth_cnx_default_attrs[] = {
 372         GET_CNX_ATTR(outstanding_tx),
 373         GET_CNX_ATTR(remote_lp),
 374         GET_CNX_ATTR(num_events),
 375         GET_CNX_ATTR(reset_timeout),
 376         GET_CNX_ATTR(last_contact),
 377         GET_CNX_ATTR(state),
 378         GET_CNX_ATTR(src_inst),
 379         GET_CNX_ATTR(dst_inst),
 380         GET_CNX_ATTR(num_pending_acks),
 381         GET_CNX_ATTR(num_ack_events),
 382         GET_CNX_ATTR(ack_timeout),
 383         NULL
 384 };
 385
 386 static struct sysfs_ops veth_cnx_sysfs_ops = {
 387                 .show = veth_cnx_attribute_show
 388 };
 389
 390 static struct kobj_type veth_lpar_connection_ktype = {
 391         .release        = veth_release_connection,
 392         .sysfs_ops      = &veth_cnx_sysfs_ops,
 393         .default_attrs  = veth_cnx_default_attrs
 394 };
 395
 396 struct veth_port_attribute {
 397         struct attribute attr;
 398         ssize_t (*show)(struct veth_port *, char *buf);
 399         ssize_t (*store)(struct veth_port *, const char *buf);
 400 };
 401
 402 static ssize_t veth_port_attribute_show(struct kobject *kobj,
 403                 struct attribute *attr, char *buf)
 404 {
 405         struct veth_port_attribute *port_attr;
 406         struct veth_port *port;
 407
 408         port_attr = container_of(attr, struct veth_port_attribute, attr);
 409         port = container_of(kobj, struct veth_port, kobject);
 410
 411         if (!port_attr->show)
 412                 return -EIO;
 413
 414         return port_attr->show(port, buf);
 415 }
 416
 417 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
 418 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
 419 {                                                                       \
 420         return sprintf(buf, _format, _expression);                      \
 421 }                                                                       \
 422 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
 423
 424 #define SIMPLE_PORT_ATTR(_name) \
 425         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
 426
 427 SIMPLE_PORT_ATTR(promiscuous);
 428 SIMPLE_PORT_ATTR(num_mcast);
 429 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
 430 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
 431 CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
 432
 433 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
 434 static struct attribute *veth_port_default_attrs[] = {
 435         GET_PORT_ATTR(mac_addr),
 436         GET_PORT_ATTR(lpar_map),
 437         GET_PORT_ATTR(stopped_map),
 438         GET_PORT_ATTR(promiscuous),
 439         GET_PORT_ATTR(num_mcast),
 440         NULL
 441 };
 442
 443 static struct sysfs_ops veth_port_sysfs_ops = {
 444         .show = veth_port_attribute_show
 445 };
 446
 447 static struct kobj_type veth_port_ktype = {
 448         .sysfs_ops      = &veth_port_sysfs_ops,
 449         .default_attrs  = veth_port_default_attrs
 450 };
 451
 452 /*
 453  * LPAR connection code
 454  */
 455
 456 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 457 {
 458         schedule_work(&cnx->statemachine_wq);
 459 }
 460
 461 static void veth_take_cap(struct veth_lpar_connection *cnx,
 462                           struct veth_lpevent *event)
 463 {
 464         unsigned long flags;
 465
 466         spin_lock_irqsave(&cnx->lock, flags);
 467         /* Receiving caps may mean the other end has just come up, so
 468          * we need to reload the instance ID of the far end */
 469         cnx->dst_inst =
 470                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 471                                                   HvLpEvent_Type_VirtualLan);
 472
 473         if (cnx->state & VETH_STATE_GOTCAPS) {
 474                 veth_error("Received a second capabilities from LPAR %d.\n",
 475                            cnx->remote_lp);
 476                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 477                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 478         } else {
 479                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 480                 cnx->state |= VETH_STATE_GOTCAPS;
 481                 veth_kick_statemachine(cnx);
 482         }
 483         spin_unlock_irqrestore(&cnx->lock, flags);
 484 }
 485
 486 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 487                               struct veth_lpevent *event)
 488 {
 489         unsigned long flags;
 490
 491         spin_lock_irqsave(&cnx->lock, flags);
 492         if (cnx->state & VETH_STATE_GOTCAPACK) {
 493                 veth_error("Received a second capabilities ack from LPAR %d.\n",
 494                            cnx->remote_lp);
 495         } else {
 496                 memcpy(&cnx->cap_ack_event, event,
 497                        sizeof(&cnx->cap_ack_event));
 498                 cnx->state |= VETH_STATE_GOTCAPACK;
 499                 veth_kick_statemachine(cnx);
 500         }
 501         spin_unlock_irqrestore(&cnx->lock, flags);
 502 }
 503
 504 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 505                                   struct veth_lpevent *event)
 506 {
 507         unsigned long flags;
 508
 509         spin_lock_irqsave(&cnx->lock, flags);
 510         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
 511
 512         /* Avoid kicking the statemachine once we're shutdown.
 513          * It's unnecessary and it could break veth_stop_connection(). */
 514
 515         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 516                 cnx->state |= VETH_STATE_RESET;
 517                 veth_kick_statemachine(cnx);
 518         }
 519         spin_unlock_irqrestore(&cnx->lock, flags);
 520 }
 521
 522 static void veth_handle_ack(struct veth_lpevent *event)
 523 {
 524         HvLpIndex rlp = event->base_event.xTargetLp;
 525         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 526
 527         BUG_ON(! cnx);
 528
 529         switch (event->base_event.xSubtype) {
 530         case VETH_EVENT_CAP:
 531                 veth_take_cap_ack(cnx, event);
 532                 break;
 533         case VETH_EVENT_MONITOR:
 534                 veth_take_monitor_ack(cnx, event);
 535                 break;
 536         default:
 537                 veth_error("Unknown ack type %d from LPAR %d.\n",
 538                                 event->base_event.xSubtype, rlp);
 539         };
 540 }
 541
 542 static void veth_handle_int(struct veth_lpevent *event)
 543 {
 544         HvLpIndex rlp = event->base_event.xSourceLp;
 545         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 546         unsigned long flags;
 547         int i, acked = 0;
 548
 549         BUG_ON(! cnx);
 550
 551         switch (event->base_event.xSubtype) {
 552         case VETH_EVENT_CAP:
 553                 veth_take_cap(cnx, event);
 554                 break;
 555         case VETH_EVENT_MONITOR:
 556                 /* do nothing... this'll hang out here til we're dead,
 557                  * and the hypervisor will return it for us. */
 558                 break;
 559         case VETH_EVENT_FRAMES_ACK:
 560                 spin_lock_irqsave(&cnx->lock, flags);
 561
 562                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 563                         u16 msgnum = event->u.frames_ack_data.token[i];
 564
 565                         if (msgnum < VETH_NUMBUFFERS) {
 566                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 567                                 cnx->outstanding_tx--;
 568                                 acked++;
 569                         }
 570                 }
 571
 572                 if (acked > 0) {
 573                         cnx->last_contact = jiffies;
 574                         veth_wake_queues(cnx);
 575                 }
 576
 577                 spin_unlock_irqrestore(&cnx->lock, flags);
 578                 break;
 579         case VETH_EVENT_FRAMES:
 580                 veth_receive(cnx, event);
 581                 break;
 582         default:
 583                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
 584                                 event->base_event.xSubtype, rlp);
 585         };
 586 }
 587
 588 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
 589 {
 590         struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
 591
 592         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
 593                 veth_handle_ack(veth_event);
 594         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
 595                 veth_handle_int(veth_event);
 596 }
 597
 598 static int veth_process_caps(struct veth_lpar_connection *cnx)
 599 {
 600         struct veth_cap_data *remote_caps = &cnx->remote_caps;
 601         int num_acks_needed;
 602
 603         /* Convert timer to jiffies */
 604         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 605
 606         if ( (remote_caps->num_buffers == 0)
 607              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
 608              || (remote_caps->ack_threshold == 0)
 609              || (cnx->ack_timeout == 0) ) {
 610                 veth_error("Received incompatible capabilities from LPAR %d.\n",
 611                                 cnx->remote_lp);
 612                 return HvLpEvent_Rc_InvalidSubtypeData;
 613         }
 614
 615         num_acks_needed = (remote_caps->num_buffers
 616                            / remote_caps->ack_threshold) + 1;
 617
 618         /* FIXME: locking on num_ack_events? */
 619         if (cnx->num_ack_events < num_acks_needed) {
 620                 int num;
 621
 622                 num = veth_allocate_events(cnx->remote_lp,
 623                                            num_acks_needed-cnx->num_ack_events);
 624                 if (num > 0)
 625                         cnx->num_ack_events += num;
 626
 627                 if (cnx->num_ack_events < num_acks_needed) {
 628                         veth_error("Couldn't allocate enough ack events "
 629                                         "for LPAR %d.\n", cnx->remote_lp);
 630
 631                         return HvLpEvent_Rc_BufferNotAvailable;
 632                 }
 633         }
 634
 635
 636         return HvLpEvent_Rc_Good;
 637 }
 638
 639 /* FIXME: The gotos here are a bit dubious */
 640 static void veth_statemachine(void *p)
 641 {
 642         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
 643         int rlp = cnx->remote_lp;
 644         int rc;
 645
 646         spin_lock_irq(&cnx->lock);
 647
 648  restart:
 649         if (cnx->state & VETH_STATE_RESET) {
 650                 if (cnx->state & VETH_STATE_OPEN)
 651                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 652                                                      HvLpEvent_Type_VirtualLan);
 653
 654                 /*
 655                  * Reset ack data. This prevents the ack_timer actually
 656                  * doing anything, even if it runs one more time when
 657                  * we drop the lock below.
 658                  */
 659                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 660                 cnx->num_pending_acks = 0;
 661
 662                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 663                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 664                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 665                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 666
 667                 /* Clean up any leftover messages */
 668                 if (cnx->msgs) {
 669                         int i;
 670                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 671                                 veth_recycle_msg(cnx, cnx->msgs + i);
 672                 }
 673
 674                 cnx->outstanding_tx = 0;
 675                 veth_wake_queues(cnx);
 676
 677                 /* Drop the lock so we can do stuff that might sleep or
 678                  * take other locks. */
 679                 spin_unlock_irq(&cnx->lock);
 680
 681                 del_timer_sync(&cnx->ack_timer);
 682                 del_timer_sync(&cnx->reset_timer);
 683
 684                 spin_lock_irq(&cnx->lock);
 685
 686                 if (cnx->state & VETH_STATE_RESET)
 687                         goto restart;
 688
 689                 /* Hack, wait for the other end to reset itself. */
 690                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 691                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
 692                         goto out;
 693                 }
 694         }
 695
 696         if (cnx->state & VETH_STATE_SHUTDOWN)
 697                 /* It's all over, do nothing */
 698                 goto out;
 699
 700         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 701                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 702                         goto cant_cope;
 703
 704                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 705                 cnx->src_inst =
 706                         HvCallEvent_getSourceLpInstanceId(rlp,
 707                                                           HvLpEvent_Type_VirtualLan);
 708                 cnx->dst_inst =
 709                         HvCallEvent_getTargetLpInstanceId(rlp,
 710                                                           HvLpEvent_Type_VirtualLan);
 711                 cnx->state |= VETH_STATE_OPEN;
 712         }
 713
 714         if ( (cnx->state & VETH_STATE_OPEN)
 715              && !(cnx->state & VETH_STATE_SENTMON) ) {
 716                 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
 717                                       HvLpEvent_AckInd_DoAck,
 718                                       HvLpEvent_AckType_DeferredAck,
 719                                       0, 0, 0, 0, 0, 0);
 720
 721                 if (rc == HvLpEvent_Rc_Good) {
 722                         cnx->state |= VETH_STATE_SENTMON;
 723                 } else {
 724                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 725                              && (rc != HvLpEvent_Rc_PathClosed) )
 726                                 veth_error("Error sending monitor to LPAR %d, "
 727                                                 "rc = %d\n", rlp, rc);
 728
 729                         /* Oh well, hope we get a cap from the other
 730                          * end and do better when that kicks us */
 731                         goto out;
 732                 }
 733         }
 734
 735         if ( (cnx->state & VETH_STATE_OPEN)
 736              && !(cnx->state & VETH_STATE_SENTCAPS)) {
 737                 u64 *rawcap = (u64 *)&cnx->local_caps;
 738
 739                 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
 740                                       HvLpEvent_AckInd_DoAck,
 741                                       HvLpEvent_AckType_ImmediateAck,
 742                                       0, rawcap[0], rawcap[1], rawcap[2],
 743                                       rawcap[3], rawcap[4]);
 744
 745                 if (rc == HvLpEvent_Rc_Good) {
 746                         cnx->state |= VETH_STATE_SENTCAPS;
 747                 } else {
 748                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 749                              && (rc != HvLpEvent_Rc_PathClosed) )
 750                                 veth_error("Error sending caps to LPAR %d, "
 751                                                 "rc = %d\n", rlp, rc);
 752
 753                         /* Oh well, hope we get a cap from the other
 754                          * end and do better when that kicks us */
 755                         goto out;
 756                 }
 757         }
 758
 759         if ((cnx->state & VETH_STATE_GOTCAPS)
 760             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
 761                 struct veth_cap_data *remote_caps = &cnx->remote_caps;
 762
 763                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 764                        sizeof(*remote_caps));
 765
 766                 spin_unlock_irq(&cnx->lock);
 767                 rc = veth_process_caps(cnx);
 768                 spin_lock_irq(&cnx->lock);
 769
 770                 /* We dropped the lock, so recheck for anything which
 771                  * might mess us up */
 772                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 773                         goto restart;
 774
 775                 cnx->cap_event.base_event.xRc = rc;
 776                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 777                 if (rc == HvLpEvent_Rc_Good)
 778                         cnx->state |= VETH_STATE_SENTCAPACK;
 779                 else
 780                         goto cant_cope;
 781         }
 782
 783         if ((cnx->state & VETH_STATE_GOTCAPACK)
 784             && (cnx->state & VETH_STATE_GOTCAPS)
 785             && !(cnx->state & VETH_STATE_READY)) {
 786                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 787                         /* Start the ACK timer */
 788                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 789                         add_timer(&cnx->ack_timer);
 790                         cnx->state |= VETH_STATE_READY;
 791                 } else {
 792                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
 793                                         rlp, cnx->cap_ack_event.base_event.xRc);
 794                         goto cant_cope;
 795                 }
 796         }
 797
 798  out:
 799         spin_unlock_irq(&cnx->lock);
 800         return;
 801
 802  cant_cope:
 803         /* FIXME: we get here if something happens we really can't
 804          * cope with.  The link will never work once we get here, and
 805          * all we can do is not lock the rest of the system up */
 806         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
 807                         " (state = 0x%04lx)\n", rlp, cnx->state);
 808         cnx->state |= VETH_STATE_SHUTDOWN;
 809         spin_unlock_irq(&cnx->lock);
 810 }
 811
 812 static int veth_init_connection(u8 rlp)
 813 {
 814         struct veth_lpar_connection *cnx;
 815         struct veth_msg *msgs;
 816         int i, rc;
 817
 818         if ( (rlp == this_lp)
 819              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 820                 return 0;
 821
 822         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
 823         if (! cnx)
 824                 return -ENOMEM;
 825         memset(cnx, 0, sizeof(*cnx));
 826
 827         cnx->remote_lp = rlp;
 828         spin_lock_init(&cnx->lock);
 829         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
 830
 831         init_timer(&cnx->ack_timer);
 832         cnx->ack_timer.function = veth_timed_ack;
 833         cnx->ack_timer.data = (unsigned long) cnx;
 834
 835         init_timer(&cnx->reset_timer);
 836         cnx->reset_timer.function = veth_timed_reset;
 837         cnx->reset_timer.data = (unsigned long) cnx;
 838         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
 839
 840         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 841
 842         veth_cnx[rlp] = cnx;
 843
 844         /* This gets us 1 reference, which is held on behalf of the driver
 845          * infrastructure. It's released at module unload. */
 846         kobject_init(&cnx->kobject);
 847         cnx->kobject.ktype = &veth_lpar_connection_ktype;
 848         rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
 849         if (rc != 0)
 850                 return rc;
 851
 852         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
 853         if (! msgs) {
 854                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
 855                 return -ENOMEM;
 856         }
 857
 858         cnx->msgs = msgs;
 859         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
 860
 861         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 862                 msgs[i].token = i;
 863                 veth_stack_push(cnx, msgs + i);
 864         }
 865
 866         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 867
 868         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 869                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
 870                 return -ENOMEM;
 871         }
 872
 873         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 874         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 875         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 876
 877         return 0;
 878 }
 879
 880 static void veth_stop_connection(struct veth_lpar_connection *cnx)
 881 {
 882         if (!cnx)
 883                 return;
 884
 885         spin_lock_irq(&cnx->lock);
 886         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 887         veth_kick_statemachine(cnx);
 888         spin_unlock_irq(&cnx->lock);
 889
 890         /* There's a slim chance the reset code has just queued the
 891          * statemachine to run in five seconds. If so we need to cancel
 892          * that and requeue the work to run now. */
 893         if (cancel_delayed_work(&cnx->statemachine_wq)) {
 894                 spin_lock_irq(&cnx->lock);
 895                 veth_kick_statemachine(cnx);
 896                 spin_unlock_irq(&cnx->lock);
 897         }
 898
 899         /* Wait for the state machine to run. */
 900         flush_scheduled_work();
 901 }
 902
 903 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
 904 {
 905         if (!cnx)
 906                 return;
 907
 908         if (cnx->num_events > 0)
 909                 mf_deallocate_lp_events(cnx->remote_lp,
 910                                       HvLpEvent_Type_VirtualLan,
 911                                       cnx->num_events,
 912                                       NULL, NULL);
 913         if (cnx->num_ack_events > 0)
 914                 mf_deallocate_lp_events(cnx->remote_lp,
 915                                       HvLpEvent_Type_VirtualLan,
 916                                       cnx->num_ack_events,
 917                                       NULL, NULL);
 918
 919         kfree(cnx->msgs);
 920         veth_cnx[cnx->remote_lp] = NULL;
 921         kfree(cnx);
 922 }
 923
 924 static void veth_release_connection(struct kobject *kobj)
 925 {
 926         struct veth_lpar_connection *cnx;
 927         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 928         veth_stop_connection(cnx);
 929         veth_destroy_connection(cnx);
 930 }
 931
 932 /*
 933  * net_device code
 934  */
 935
 936 static int veth_open(struct net_device *dev)
 937 {
 938         struct veth_port *port = (struct veth_port *) dev->priv;
 939
 940         memset(&port->stats, 0, sizeof (port->stats));
 941         netif_start_queue(dev);
 942         return 0;
 943 }
 944
 945 static int veth_close(struct net_device *dev)
 946 {
 947         netif_stop_queue(dev);
 948         return 0;
 949 }
 950
 951 static struct net_device_stats *veth_get_stats(struct net_device *dev)
 952 {
 953         struct veth_port *port = (struct veth_port *) dev->priv;
 954
 955         return &port->stats;
 956 }
 957
 958 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 959 {
 960         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 961                 return -EINVAL;
 962         dev->mtu = new_mtu;
 963         return 0;
 964 }
 965
 966 static void veth_set_multicast_list(struct net_device *dev)
 967 {
 968         struct veth_port *port = (struct veth_port *) dev->priv;
 969         unsigned long flags;
 970
 971         write_lock_irqsave(&port->mcast_gate, flags);
 972
 973         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
 974                         (dev->mc_count > VETH_MAX_MCAST)) {
 975                 port->promiscuous = 1;
 976         } else {
 977                 struct dev_mc_list *dmi = dev->mc_list;
 978                 int i;
 979
 980                 port->promiscuous = 0;
 981
 982                 /* Update table */
 983                 port->num_mcast = 0;
 984
 985                 for (i = 0; i < dev->mc_count; i++) {
 986                         u8 *addr = dmi->dmi_addr;
 987                         u64 xaddr = 0;
 988
 989                         if (addr[0] & 0x01) {/* multicast address? */
 990                                 memcpy(&xaddr, addr, ETH_ALEN);
 991                                 port->mcast_addr[port->num_mcast] = xaddr;
 992                                 port->num_mcast++;
 993                         }
 994                         dmi = dmi->next;
 995                 }
 996         }
 997
 998         write_unlock_irqrestore(&port->mcast_gate, flags);
 999 }
1000
1001 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1002 {
1003         strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
1004         info->driver[sizeof(info->driver) - 1] = '\0';
1005         strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
1006         info->version[sizeof(info->version) - 1] = '\0';
1007 }
1008
1009 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1010 {
1011         ecmd->supported = (SUPPORTED_1000baseT_Full
1012                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1013         ecmd->advertising = (SUPPORTED_1000baseT_Full
1014                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1015         ecmd->port = PORT_FIBRE;
1016         ecmd->transceiver = XCVR_INTERNAL;
1017         ecmd->phy_address = 0;
1018         ecmd->speed = SPEED_1000;
1019         ecmd->duplex = DUPLEX_FULL;
1020         ecmd->autoneg = AUTONEG_ENABLE;
1021         ecmd->maxtxpkt = 120;
1022         ecmd->maxrxpkt = 120;
1023         return 0;
1024 }
1025
1026 static u32 veth_get_link(struct net_device *dev)
1027 {
1028         return 1;
1029 }
1030
1031 static struct ethtool_ops ops = {
1032         .get_drvinfo = veth_get_drvinfo,
1033         .get_settings = veth_get_settings,
1034         .get_link = veth_get_link,
1035 };
1036
1037 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
1038 {
1039         struct net_device *dev;
1040         struct veth_port *port;
1041         int i, rc;
1042
1043         dev = alloc_etherdev(sizeof (struct veth_port));
1044         if (! dev) {
1045                 veth_error("Unable to allocate net_device structure!\n");
1046                 return NULL;
1047         }
1048
1049         port = (struct veth_port *) dev->priv;
1050
1051         spin_lock_init(&port->queue_lock);
1052         rwlock_init(&port->mcast_gate);
1053         port->stopped_map = 0;
1054
1055         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1056                 HvLpVirtualLanIndexMap map;
1057
1058                 if (i == this_lp)
1059                         continue;
1060                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1061                 if (map & (0x8000 >> vlan))
1062                         port->lpar_map |= (1 << i);
1063         }
1064         port->dev = vdev;
1065
1066         dev->dev_addr[0] = 0x02;
1067         dev->dev_addr[1] = 0x01;
1068         dev->dev_addr[2] = 0xff;
1069         dev->dev_addr[3] = vlan;
1070         dev->dev_addr[4] = 0xff;
1071         dev->dev_addr[5] = this_lp;
1072
1073         dev->mtu = VETH_MAX_MTU;
1074
1075         memcpy(&port->mac_addr, dev->dev_addr, 6);
1076
1077         dev->open = veth_open;
1078         dev->hard_start_xmit = veth_start_xmit;
1079         dev->stop = veth_close;
1080         dev->get_stats = veth_get_stats;
1081         dev->change_mtu = veth_change_mtu;
1082         dev->set_mac_address = NULL;
1083         dev->set_multicast_list = veth_set_multicast_list;
1084         SET_ETHTOOL_OPS(dev, &ops);
1085
1086         SET_NETDEV_DEV(dev, vdev);
1087
1088         rc = register_netdev(dev);
1089         if (rc != 0) {
1090                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1091                 free_netdev(dev);
1092                 return NULL;
1093         }
1094
1095         kobject_init(&port->kobject);
1096         port->kobject.parent = &dev->class_dev.kobj;
1097         port->kobject.ktype  = &veth_port_ktype;
1098         kobject_set_name(&port->kobject, "veth_port");
1099         if (0 != kobject_add(&port->kobject))
1100                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1101
1102         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1103                         dev->name, vlan, port->lpar_map);
1104
1105         return dev;
1106 }
1107
1108 /*
1109  * Tx path
1110  */
1111
1112 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1113                                 struct net_device *dev)
1114 {
1115         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1116         struct veth_port *port = (struct veth_port *) dev->priv;
1117         HvLpEvent_Rc rc;
1118         struct veth_msg *msg = NULL;
1119         unsigned long flags;
1120
1121         if (! cnx)
1122                 return 0;
1123
1124         spin_lock_irqsave(&cnx->lock, flags);
1125
1126         if (! (cnx->state & VETH_STATE_READY))
1127                 goto no_error;
1128
1129         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1130                 goto drop;
1131
1132         msg = veth_stack_pop(cnx);
1133         if (! msg)
1134                 goto drop;
1135
1136         msg->in_use = 1;
1137         msg->skb = skb_get(skb);
1138
1139         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1140                                 skb->len, DMA_TO_DEVICE);
1141
1142         if (dma_mapping_error(msg->data.addr[0]))
1143                 goto recycle_and_drop;
1144
1145         msg->dev = port->dev;
1146         msg->data.len[0] = skb->len;
1147         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1148
1149         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1150
1151         if (rc != HvLpEvent_Rc_Good)
1152                 goto recycle_and_drop;
1153
1154         /* If the timer's not already running, start it now. */
1155         if (0 == cnx->outstanding_tx)
1156                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1157
1158         cnx->last_contact = jiffies;
1159         cnx->outstanding_tx++;
1160
1161         if (veth_stack_is_empty(cnx))
1162                 veth_stop_queues(cnx);
1163
1164  no_error:
1165         spin_unlock_irqrestore(&cnx->lock, flags);
1166         return 0;
1167
1168  recycle_and_drop:
1169         veth_recycle_msg(cnx, msg);
1170  drop:
1171         spin_unlock_irqrestore(&cnx->lock, flags);
1172         return 1;
1173 }
1174
1175 static void veth_transmit_to_many(struct sk_buff *skb,
1176                                           HvLpIndexMap lpmask,
1177                                           struct net_device *dev)
1178 {
1179         struct veth_port *port = (struct veth_port *) dev->priv;
1180         int i, success, error;
1181
1182         success = error = 0;
1183
1184         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1185                 if ((lpmask & (1 << i)) == 0)
1186                         continue;
1187
1188                 if (veth_transmit_to_one(skb, i, dev))
1189                         error = 1;
1190                 else
1191                         success = 1;
1192         }
1193
1194         if (error)
1195                 port->stats.tx_errors++;
1196
1197         if (success) {
1198                 port->stats.tx_packets++;
1199                 port->stats.tx_bytes += skb->len;
1200         }
1201 }
1202
1203 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1204 {
1205         unsigned char *frame = skb->data;
1206         struct veth_port *port = (struct veth_port *) dev->priv;
1207         HvLpIndexMap lpmask;
1208
1209         if (! (frame[0] & 0x01)) {
1210                 /* unicast packet */
1211                 HvLpIndex rlp = frame[5];
1212
1213                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1214                         dev_kfree_skb(skb);
1215                         return 0;
1216                 }
1217
1218                 lpmask = 1 << rlp;
1219         } else {
1220                 lpmask = port->lpar_map;
1221         }
1222
1223         veth_transmit_to_many(skb, lpmask, dev);
1224
1225         dev_kfree_skb(skb);
1226
1227         return 0;
1228 }
1229
1230 /* You must hold the connection's lock when you call this function. */
1231 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1232                              struct veth_msg *msg)
1233 {
1234         u32 dma_address, dma_length;
1235
1236         if (msg->in_use) {
1237                 msg->in_use = 0;
1238                 dma_address = msg->data.addr[0];
1239                 dma_length = msg->data.len[0];
1240
1241                 if (!dma_mapping_error(dma_address))
1242                         dma_unmap_single(msg->dev, dma_address, dma_length,
1243                                         DMA_TO_DEVICE);
1244
1245                 if (msg->skb) {
1246                         dev_kfree_skb_any(msg->skb);
1247                         msg->skb = NULL;
1248                 }
1249
1250                 memset(&msg->data, 0, sizeof(msg->data));
1251                 veth_stack_push(cnx, msg);
1252         } else if (cnx->state & VETH_STATE_OPEN) {
1253                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1254                                 cnx->remote_lp, msg->token);
1255         }
1256 }
1257
1258 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1259 {
1260         int i;
1261
1262         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1263                 struct net_device *dev = veth_dev[i];
1264                 struct veth_port *port;
1265                 unsigned long flags;
1266
1267                 if (! dev)
1268                         continue;
1269
1270                 port = (struct veth_port *)dev->priv;
1271
1272                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1273                         continue;
1274
1275                 spin_lock_irqsave(&port->queue_lock, flags);
1276
1277                 port->stopped_map &= ~(1 << cnx->remote_lp);
1278
1279                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1280                         veth_debug("cnx %d: woke queue for %s.\n",
1281                                         cnx->remote_lp, dev->name);
1282                         netif_wake_queue(dev);
1283                 }
1284                 spin_unlock_irqrestore(&port->queue_lock, flags);
1285         }
1286 }
1287
1288 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1289 {
1290         int i;
1291
1292         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1293                 struct net_device *dev = veth_dev[i];
1294                 struct veth_port *port;
1295
1296                 if (! dev)
1297                         continue;
1298
1299                 port = (struct veth_port *)dev->priv;
1300
1301                 /* If this cnx is not on the vlan for this port, continue */
1302                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1303                         continue;
1304
1305                 spin_lock(&port->queue_lock);
1306
1307                 netif_stop_queue(dev);
1308                 port->stopped_map |= (1 << cnx->remote_lp);
1309
1310                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1311                                 cnx->remote_lp, dev->name, port->stopped_map);
1312
1313                 spin_unlock(&port->queue_lock);
1314         }
1315 }
1316
1317 static void veth_timed_reset(unsigned long ptr)
1318 {
1319         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1320         unsigned long trigger_time, flags;
1321
1322         /* FIXME is it possible this fires after veth_stop_connection()?
1323          * That would reschedule the statemachine for 5 seconds and probably
1324          * execute it after the module's been unloaded. Hmm. */
1325
1326         spin_lock_irqsave(&cnx->lock, flags);
1327
1328         if (cnx->outstanding_tx > 0) {
1329                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1330
1331                 if (trigger_time < jiffies) {
1332                         cnx->state |= VETH_STATE_RESET;
1333                         veth_kick_statemachine(cnx);
1334                         veth_error("%d packets not acked by LPAR %d within %d "
1335                                         "seconds, resetting.\n",
1336                                         cnx->outstanding_tx, cnx->remote_lp,
1337                                         cnx->reset_timeout / HZ);
1338                 } else {
1339                         /* Reschedule the timer */
1340                         trigger_time = jiffies + cnx->reset_timeout;
1341                         mod_timer(&cnx->reset_timer, trigger_time);
1342                 }
1343         }
1344
1345         spin_unlock_irqrestore(&cnx->lock, flags);
1346 }
1347
1348 /*
1349  * Rx path
1350  */
1351
1352 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1353 {
1354         int wanted = 0;
1355         int i;
1356         unsigned long flags;
1357
1358         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1359                 return 1;
1360
1361         read_lock_irqsave(&port->mcast_gate, flags);
1362
1363         if (port->promiscuous) {
1364                 wanted = 1;
1365                 goto out;
1366         }
1367
1368         for (i = 0; i < port->num_mcast; ++i) {
1369                 if (port->mcast_addr[i] == mac_addr) {
1370                         wanted = 1;
1371                         break;
1372                 }
1373         }
1374
1375  out:
1376         read_unlock_irqrestore(&port->mcast_gate, flags);
1377
1378         return wanted;
1379 }
1380
1381 struct dma_chunk {
1382         u64 addr;
1383         u64 size;
1384 };
1385
1386 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1387
1388 static inline void veth_build_dma_list(struct dma_chunk *list,
1389                                        unsigned char *p, unsigned long length)
1390 {
1391         unsigned long done;
1392         int i = 1;
1393
1394         /* FIXME: skbs are continguous in real addresses.  Do we
1395          * really need to break it into PAGE_SIZE chunks, or can we do
1396          * it just at the granularity of iSeries real->absolute
1397          * mapping?  Indeed, given the way the allocator works, can we
1398          * count on them being absolutely contiguous? */
1399         list[0].addr = ISERIES_HV_ADDR(p);
1400         list[0].size = min(length,
1401                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1402
1403         done = list[0].size;
1404         while (done < length) {
1405                 list[i].addr = ISERIES_HV_ADDR(p + done);
1406                 list[i].size = min(length-done, PAGE_SIZE);
1407                 done += list[i].size;
1408                 i++;
1409         }
1410 }
1411
1412 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1413 {
1414         HvLpEvent_Rc rc;
1415
1416         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1417                              0, &cnx->pending_acks);
1418
1419         if (rc != HvLpEvent_Rc_Good)
1420                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1421                                 cnx->remote_lp, (int)rc);
1422
1423         cnx->num_pending_acks = 0;
1424         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1425 }
1426
1427 static void veth_receive(struct veth_lpar_connection *cnx,
1428                          struct veth_lpevent *event)
1429 {
1430         struct veth_frames_data *senddata = &event->u.frames_data;
1431         int startchunk = 0;
1432         int nchunks;
1433         unsigned long flags;
1434         HvLpDma_Rc rc;
1435
1436         do {
1437                 u16 length = 0;
1438                 struct sk_buff *skb;
1439                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1440                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1441                 u64 dest;
1442                 HvLpVirtualLanIndex vlan;
1443                 struct net_device *dev;
1444                 struct veth_port *port;
1445
1446                 /* FIXME: do we need this? */
1447                 memset(local_list, 0, sizeof(local_list));
1448                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1449
1450                 /* a 0 address marks the end of the valid entries */
1451                 if (senddata->addr[startchunk] == 0)
1452                         break;
1453
1454                 /* make sure that we have at least 1 EOF entry in the
1455                  * remaining entries */
1456                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1457                         veth_error("Missing EOF fragment in event "
1458                                         "eofmask = 0x%x startchunk = %d\n",
1459                                         (unsigned)senddata->eofmask,
1460                                         startchunk);
1461                         break;
1462                 }
1463
1464                 /* build list of chunks in this frame */
1465                 nchunks = 0;
1466                 do {
1467                         remote_list[nchunks].addr =
1468                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1469                         remote_list[nchunks].size =
1470                                 senddata->len[startchunk+nchunks];
1471                         length += remote_list[nchunks].size;
1472                 } while (! (senddata->eofmask &
1473                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1474
1475                 /* length == total length of all chunks */
1476                 /* nchunks == # of chunks in this frame */
1477
1478                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1479                         veth_error("Received oversize frame from LPAR %d "
1480                                         "(length = %d)\n",
1481                                         cnx->remote_lp, length);
1482                         continue;
1483                 }
1484
1485                 skb = alloc_skb(length, GFP_ATOMIC);
1486                 if (!skb)
1487                         continue;
1488
1489                 veth_build_dma_list(local_list, skb->data, length);
1490
1491                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1492                                             event->base_event.xSourceLp,
1493                                             HvLpDma_Direction_RemoteToLocal,
1494                                             cnx->src_inst,
1495                                             cnx->dst_inst,
1496                                             HvLpDma_AddressType_RealAddress,
1497                                             HvLpDma_AddressType_TceIndex,
1498                                             ISERIES_HV_ADDR(&local_list),
1499                                             ISERIES_HV_ADDR(&remote_list),
1500                                             length);
1501                 if (rc != HvLpDma_Rc_Good) {
1502                         dev_kfree_skb_irq(skb);
1503                         continue;
1504                 }
1505
1506                 vlan = skb->data[9];
1507                 dev = veth_dev[vlan];
1508                 if (! dev) {
1509                         /*
1510                          * Some earlier versions of the driver sent
1511                          * broadcasts down all connections, even to lpars
1512                          * that weren't on the relevant vlan. So ignore
1513                          * packets belonging to a vlan we're not on.
1514                          * We can also be here if we receive packets while
1515                          * the driver is going down, because then dev is NULL.
1516                          */
1517                         dev_kfree_skb_irq(skb);
1518                         continue;
1519                 }
1520
1521                 port = (struct veth_port *)dev->priv;
1522                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1523
1524                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1525                         dev_kfree_skb_irq(skb);
1526                         continue;
1527                 }
1528                 if (! veth_frame_wanted(port, dest)) {
1529                         dev_kfree_skb_irq(skb);
1530                         continue;
1531                 }
1532
1533                 skb_put(skb, length);
1534                 skb->dev = dev;
1535                 skb->protocol = eth_type_trans(skb, dev);
1536                 skb->ip_summed = CHECKSUM_NONE;
1537                 netif_rx(skb);  /* send it up */
1538                 port->stats.rx_packets++;
1539                 port->stats.rx_bytes += length;
1540         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1541
1542         /* Ack it */
1543         spin_lock_irqsave(&cnx->lock, flags);
1544         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1545
1546         cnx->pending_acks[cnx->num_pending_acks++] =
1547                 event->base_event.xCorrelationToken;
1548
1549         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1550              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1551                 veth_flush_acks(cnx);
1552
1553         spin_unlock_irqrestore(&cnx->lock, flags);
1554 }
1555
1556 static void veth_timed_ack(unsigned long ptr)
1557 {
1558         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1559         unsigned long flags;
1560
1561         /* Ack all the events */
1562         spin_lock_irqsave(&cnx->lock, flags);
1563         if (cnx->num_pending_acks > 0)
1564                 veth_flush_acks(cnx);
1565
1566         /* Reschedule the timer */
1567         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1568         add_timer(&cnx->ack_timer);
1569         spin_unlock_irqrestore(&cnx->lock, flags);
1570 }
1571
1572 static int veth_remove(struct vio_dev *vdev)
1573 {
1574         struct veth_lpar_connection *cnx;
1575         struct net_device *dev;
1576         struct veth_port *port;
1577         int i;
1578
1579         dev = veth_dev[vdev->unit_address];
1580
1581         if (! dev)
1582                 return 0;
1583
1584         port = netdev_priv(dev);
1585
1586         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1587                 cnx = veth_cnx[i];
1588
1589                 if (cnx && (port->lpar_map & (1 << i))) {
1590                         /* Drop our reference to connections on our VLAN */
1591                         kobject_put(&cnx->kobject);
1592                 }
1593         }
1594
1595         veth_dev[vdev->unit_address] = NULL;
1596         kobject_del(&port->kobject);
1597         kobject_put(&port->kobject);
1598         unregister_netdev(dev);
1599         free_netdev(dev);
1600
1601         return 0;
1602 }
1603
1604 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1605 {
1606         int i = vdev->unit_address;
1607         struct net_device *dev;
1608         struct veth_port *port;
1609
1610         dev = veth_probe_one(i, &vdev->dev);
1611         if (dev == NULL) {
1612                 veth_remove(vdev);
1613                 return 1;
1614         }
1615         veth_dev[i] = dev;
1616
1617         port = (struct veth_port*)netdev_priv(dev);
1618
1619         /* Start the state machine on each connection on this vlan. If we're
1620          * the first dev to do so this will commence link negotiation */
1621         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1622                 struct veth_lpar_connection *cnx;
1623
1624                 if (! (port->lpar_map & (1 << i)))
1625                         continue;
1626
1627                 cnx = veth_cnx[i];
1628                 if (!cnx)
1629                         continue;
1630
1631                 kobject_get(&cnx->kobject);
1632                 veth_kick_statemachine(cnx);
1633         }
1634
1635         return 0;
1636 }
1637
1638 /**
1639  * veth_device_table: Used by vio.c to match devices that we
1640  * support.
1641  */
1642 static struct vio_device_id veth_device_table[] __devinitdata = {
1643         { "vlan", "" },
1644         { "", "" }
1645 };
1646 MODULE_DEVICE_TABLE(vio, veth_device_table);
1647
1648 static struct vio_driver veth_driver = {
1649         .name = DRV_NAME,
1650         .id_table = veth_device_table,
1651         .probe = veth_probe,
1652         .remove = veth_remove
1653 };
1654
1655 /*
1656  * Module initialization/cleanup
1657  */
1658
1659 void __exit veth_module_cleanup(void)
1660 {
1661         int i;
1662         struct veth_lpar_connection *cnx;
1663
1664         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1665         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1666
1667         /* Make sure any work queued from Hypervisor callbacks is finished. */
1668         flush_scheduled_work();
1669
1670         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1671                 cnx = veth_cnx[i];
1672
1673                 if (!cnx)
1674                         continue;
1675
1676                 /* Remove the connection from sysfs */
1677                 kobject_del(&cnx->kobject);
1678                 /* Drop the driver's reference to the connection */
1679                 kobject_put(&cnx->kobject);
1680         }
1681
1682         /* Unregister the driver, which will close all the netdevs and stop
1683          * the connections when they're no longer referenced. */
1684         vio_unregister_driver(&veth_driver);
1685 }
1686 module_exit(veth_module_cleanup);
1687
1688 int __init veth_module_init(void)
1689 {
1690         int i;
1691         int rc;
1692
1693         this_lp = HvLpConfig_getLpIndex_outline();
1694
1695         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1696                 rc = veth_init_connection(i);
1697                 if (rc != 0)
1698                         goto error;
1699         }
1700
1701         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1702                                   &veth_handle_event);
1703
1704         rc = vio_register_driver(&veth_driver);
1705         if (rc != 0)
1706                 goto error;
1707
1708         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1709                 struct kobject *kobj;
1710
1711                 if (!veth_cnx[i])
1712                         continue;
1713
1714                 kobj = &veth_cnx[i]->kobject;
1715                 kobj->parent = &veth_driver.driver.kobj;
1716                 /* If the add failes, complain but otherwise continue */
1717                 if (0 != kobject_add(kobj))
1718                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1719         }
1720
1721         return 0;
1722
1723 error:
1724         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1725                 veth_destroy_connection(veth_cnx[i]);
1726         }
1727
1728         return rc;
1729 }
1730 module_init(veth_module_init);