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  * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License as
  11  * published by the Free Software Foundation; either version 2 of the
  12  * License, or (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
  22  * USA
  23  *
  24  *
  25  * This module implements the virtual ethernet device for iSeries LPAR
  26  * Linux.  It uses hypervisor message passing to implement an
  27  * ethernet-like network device communicating between partitions on
  28  * the iSeries.
  29  *
  30  * The iSeries LPAR hypervisor currently allows for up to 16 different
  31  * virtual ethernets.  These are all dynamically configurable on
  32  * OS/400 partitions, but dynamic configuration is not supported under
  33  * Linux yet.  An ethXX network device will be created for each
  34  * virtual ethernet this partition is connected to.
  35  *
  36  * - This driver is responsible for routing packets to and from other
  37  *   partitions.  The MAC addresses used by the virtual ethernets
  38  *   contains meaning and must not be modified.
  39  *
  40  * - Having 2 virtual ethernets to the same remote partition DOES NOT
  41  *   double the available bandwidth.  The 2 devices will share the
  42  *   available hypervisor bandwidth.
  43  *
  44  * - If you send a packet to your own mac address, it will just be
  45  *   dropped, you won't get it on the receive side.
  46  *
  47  * - Multicast is implemented by sending the frame frame to every
  48  *   other partition.  It is the responsibility of the receiving
  49  *   partition to filter the addresses desired.
  50  *
  51  * Tunable parameters:
  52  *
  53  * VETH_NUMBUFFERS: This compile time option defaults to 120.  It
  54  * controls how much memory Linux will allocate per remote partition
  55  * it is communicating with.  It can be thought of as the maximum
  56  * number of packets outstanding to a remote partition at a time.
  57  */
  58
  59 #include <linux/module.h>
  60 #include <linux/types.h>
  61 #include <linux/errno.h>
  62 #include <linux/ioport.h>
  63 #include <linux/kernel.h>
  64 #include <linux/netdevice.h>
  65 #include <linux/etherdevice.h>
  66 #include <linux/skbuff.h>
  67 #include <linux/init.h>
  68 #include <linux/delay.h>
  69 #include <linux/mm.h>
  70 #include <linux/ethtool.h>
  71 #include <linux/if_ether.h>
  72
  73 #include <asm/abs_addr.h>
  74 #include <asm/iseries/mf.h>
  75 #include <asm/uaccess.h>
  76 #include <asm/firmware.h>
  77 #include <asm/iseries/hv_lp_config.h>
  78 #include <asm/iseries/hv_types.h>
  79 #include <asm/iseries/hv_lp_event.h>
  80 #include <asm/iommu.h>
  81 #include <asm/vio.h>
  82
  83 #undef DEBUG
  84
  85 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  86 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  87 MODULE_LICENSE("GPL");
  88
  89 #define VETH_EVENT_CAP  (0)
  90 #define VETH_EVENT_FRAMES       (1)
  91 #define VETH_EVENT_MONITOR      (2)
  92 #define VETH_EVENT_FRAMES_ACK   (3)
  93
  94 #define VETH_MAX_ACKS_PER_MSG   (20)
  95 #define VETH_MAX_FRAMES_PER_MSG (6)
  96
  97 struct veth_frames_data {
  98         u32 addr[VETH_MAX_FRAMES_PER_MSG];
  99         u16 len[VETH_MAX_FRAMES_PER_MSG];
 100         u32 eofmask;
 101 };
 102 #define VETH_EOF_SHIFT          (32-VETH_MAX_FRAMES_PER_MSG)
 103
 104 struct veth_frames_ack_data {
 105         u16 token[VETH_MAX_ACKS_PER_MSG];
 106 };
 107
 108 struct veth_cap_data {
 109         u8 caps_version;
 110         u8 rsvd1;
 111         u16 num_buffers;
 112         u16 ack_threshold;
 113         u16 rsvd2;
 114         u32 ack_timeout;
 115         u32 rsvd3;
 116         u64 rsvd4[3];
 117 };
 118
 119 struct veth_lpevent {
 120         struct HvLpEvent base_event;
 121         union {
 122                 struct veth_cap_data caps_data;
 123                 struct veth_frames_data frames_data;
 124                 struct veth_frames_ack_data frames_ack_data;
 125         } u;
 126
 127 };
 128
 129 #define DRV_NAME        "iseries_veth"
 130 #define DRV_VERSION     "2.0"
 131
 132 #define VETH_NUMBUFFERS         (120)
 133 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
 134 #define VETH_MAX_MCAST          (12)
 135
 136 #define VETH_MAX_MTU            (9000)
 137
 138 #if VETH_NUMBUFFERS < 10
 139 #define ACK_THRESHOLD           (1)
 140 #elif VETH_NUMBUFFERS < 20
 141 #define ACK_THRESHOLD           (4)
 142 #elif VETH_NUMBUFFERS < 40
 143 #define ACK_THRESHOLD           (10)
 144 #else
 145 #define ACK_THRESHOLD           (20)
 146 #endif
 147
 148 #define VETH_STATE_SHUTDOWN     (0x0001)
 149 #define VETH_STATE_OPEN         (0x0002)
 150 #define VETH_STATE_RESET        (0x0004)
 151 #define VETH_STATE_SENTMON      (0x0008)
 152 #define VETH_STATE_SENTCAPS     (0x0010)
 153 #define VETH_STATE_GOTCAPACK    (0x0020)
 154 #define VETH_STATE_GOTCAPS      (0x0040)
 155 #define VETH_STATE_SENTCAPACK   (0x0080)
 156 #define VETH_STATE_READY        (0x0100)
 157
 158 struct veth_msg {
 159         struct veth_msg *next;
 160         struct veth_frames_data data;
 161         int token;
 162         int in_use;
 163         struct sk_buff *skb;
 164         struct device *dev;
 165 };
 166
 167 struct veth_lpar_connection {
 168         HvLpIndex remote_lp;
 169         struct delayed_work statemachine_wq;
 170         struct veth_msg *msgs;
 171         int num_events;
 172         struct veth_cap_data local_caps;
 173
 174         struct kobject kobject;
 175         struct timer_list ack_timer;
 176
 177         struct timer_list reset_timer;
 178         unsigned int reset_timeout;
 179         unsigned long last_contact;
 180         int outstanding_tx;
 181
 182         spinlock_t lock;
 183         unsigned long state;
 184         HvLpInstanceId src_inst;
 185         HvLpInstanceId dst_inst;
 186         struct veth_lpevent cap_event, cap_ack_event;
 187         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 188         u32 num_pending_acks;
 189
 190         int num_ack_events;
 191         struct veth_cap_data remote_caps;
 192         u32 ack_timeout;
 193
 194         struct veth_msg *msg_stack_head;
 195 };
 196
 197 struct veth_port {
 198         struct device *dev;
 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 =
 312                 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
 313
 314         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
 315                             sizeof(struct veth_lpevent), number,
 316                             &veth_complete_allocation, &vc);
 317         wait_for_completion(&vc.c);
 318
 319         return vc.num;
 320 }
 321
 322 /*
 323  * sysfs support
 324  */
 325
 326 struct veth_cnx_attribute {
 327         struct attribute attr;
 328         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
 329         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
 330 };
 331
 332 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
 333                 struct attribute *attr, char *buf)
 334 {
 335         struct veth_cnx_attribute *cnx_attr;
 336         struct veth_lpar_connection *cnx;
 337
 338         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
 339         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 340
 341         if (!cnx_attr->show)
 342                 return -EIO;
 343
 344         return cnx_attr->show(cnx, buf);
 345 }
 346
 347 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
 348 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
 349 {                                                                       \
 350         return sprintf(buf, _format, _expression);                      \
 351 }                                                                       \
 352 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
 353
 354 #define SIMPLE_CNX_ATTR(_name)  \
 355         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
 356
 357 SIMPLE_CNX_ATTR(outstanding_tx);
 358 SIMPLE_CNX_ATTR(remote_lp);
 359 SIMPLE_CNX_ATTR(num_events);
 360 SIMPLE_CNX_ATTR(src_inst);
 361 SIMPLE_CNX_ATTR(dst_inst);
 362 SIMPLE_CNX_ATTR(num_pending_acks);
 363 SIMPLE_CNX_ATTR(num_ack_events);
 364 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
 365 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
 366 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
 367 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
 368                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
 369
 370 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
 371
 372 static struct attribute *veth_cnx_default_attrs[] = {
 373         GET_CNX_ATTR(outstanding_tx),
 374         GET_CNX_ATTR(remote_lp),
 375         GET_CNX_ATTR(num_events),
 376         GET_CNX_ATTR(reset_timeout),
 377         GET_CNX_ATTR(last_contact),
 378         GET_CNX_ATTR(state),
 379         GET_CNX_ATTR(src_inst),
 380         GET_CNX_ATTR(dst_inst),
 381         GET_CNX_ATTR(num_pending_acks),
 382         GET_CNX_ATTR(num_ack_events),
 383         GET_CNX_ATTR(ack_timeout),
 384         NULL
 385 };
 386
 387 static struct sysfs_ops veth_cnx_sysfs_ops = {
 388                 .show = veth_cnx_attribute_show
 389 };
 390
 391 static struct kobj_type veth_lpar_connection_ktype = {
 392         .release        = veth_release_connection,
 393         .sysfs_ops      = &veth_cnx_sysfs_ops,
 394         .default_attrs  = veth_cnx_default_attrs
 395 };
 396
 397 struct veth_port_attribute {
 398         struct attribute attr;
 399         ssize_t (*show)(struct veth_port *, char *buf);
 400         ssize_t (*store)(struct veth_port *, const char *buf);
 401 };
 402
 403 static ssize_t veth_port_attribute_show(struct kobject *kobj,
 404                 struct attribute *attr, char *buf)
 405 {
 406         struct veth_port_attribute *port_attr;
 407         struct veth_port *port;
 408
 409         port_attr = container_of(attr, struct veth_port_attribute, attr);
 410         port = container_of(kobj, struct veth_port, kobject);
 411
 412         if (!port_attr->show)
 413                 return -EIO;
 414
 415         return port_attr->show(port, buf);
 416 }
 417
 418 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
 419 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
 420 {                                                                       \
 421         return sprintf(buf, _format, _expression);                      \
 422 }                                                                       \
 423 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
 424
 425 #define SIMPLE_PORT_ATTR(_name) \
 426         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
 427
 428 SIMPLE_PORT_ATTR(promiscuous);
 429 SIMPLE_PORT_ATTR(num_mcast);
 430 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
 431 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
 432 CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
 433
 434 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
 435 static struct attribute *veth_port_default_attrs[] = {
 436         GET_PORT_ATTR(mac_addr),
 437         GET_PORT_ATTR(lpar_map),
 438         GET_PORT_ATTR(stopped_map),
 439         GET_PORT_ATTR(promiscuous),
 440         GET_PORT_ATTR(num_mcast),
 441         NULL
 442 };
 443
 444 static struct sysfs_ops veth_port_sysfs_ops = {
 445         .show = veth_port_attribute_show
 446 };
 447
 448 static struct kobj_type veth_port_ktype = {
 449         .sysfs_ops      = &veth_port_sysfs_ops,
 450         .default_attrs  = veth_port_default_attrs
 451 };
 452
 453 /*
 454  * LPAR connection code
 455  */
 456
 457 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 458 {
 459         schedule_delayed_work(&cnx->statemachine_wq, 0);
 460 }
 461
 462 static void veth_take_cap(struct veth_lpar_connection *cnx,
 463                           struct veth_lpevent *event)
 464 {
 465         unsigned long flags;
 466
 467         spin_lock_irqsave(&cnx->lock, flags);
 468         /* Receiving caps may mean the other end has just come up, so
 469          * we need to reload the instance ID of the far end */
 470         cnx->dst_inst =
 471                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 472                                                   HvLpEvent_Type_VirtualLan);
 473
 474         if (cnx->state & VETH_STATE_GOTCAPS) {
 475                 veth_error("Received a second capabilities from LPAR %d.\n",
 476                            cnx->remote_lp);
 477                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 478                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 479         } else {
 480                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 481                 cnx->state |= VETH_STATE_GOTCAPS;
 482                 veth_kick_statemachine(cnx);
 483         }
 484         spin_unlock_irqrestore(&cnx->lock, flags);
 485 }
 486
 487 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 488                               struct veth_lpevent *event)
 489 {
 490         unsigned long flags;
 491
 492         spin_lock_irqsave(&cnx->lock, flags);
 493         if (cnx->state & VETH_STATE_GOTCAPACK) {
 494                 veth_error("Received a second capabilities ack from LPAR %d.\n",
 495                            cnx->remote_lp);
 496         } else {
 497                 memcpy(&cnx->cap_ack_event, event,
 498                        sizeof(cnx->cap_ack_event));
 499                 cnx->state |= VETH_STATE_GOTCAPACK;
 500                 veth_kick_statemachine(cnx);
 501         }
 502         spin_unlock_irqrestore(&cnx->lock, flags);
 503 }
 504
 505 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 506                                   struct veth_lpevent *event)
 507 {
 508         unsigned long flags;
 509
 510         spin_lock_irqsave(&cnx->lock, flags);
 511         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
 512
 513         /* Avoid kicking the statemachine once we're shutdown.
 514          * It's unnecessary and it could break veth_stop_connection(). */
 515
 516         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 517                 cnx->state |= VETH_STATE_RESET;
 518                 veth_kick_statemachine(cnx);
 519         }
 520         spin_unlock_irqrestore(&cnx->lock, flags);
 521 }
 522
 523 static void veth_handle_ack(struct veth_lpevent *event)
 524 {
 525         HvLpIndex rlp = event->base_event.xTargetLp;
 526         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 527
 528         BUG_ON(! cnx);
 529
 530         switch (event->base_event.xSubtype) {
 531         case VETH_EVENT_CAP:
 532                 veth_take_cap_ack(cnx, event);
 533                 break;
 534         case VETH_EVENT_MONITOR:
 535                 veth_take_monitor_ack(cnx, event);
 536                 break;
 537         default:
 538                 veth_error("Unknown ack type %d from LPAR %d.\n",
 539                                 event->base_event.xSubtype, rlp);
 540         };
 541 }
 542
 543 static void veth_handle_int(struct veth_lpevent *event)
 544 {
 545         HvLpIndex rlp = event->base_event.xSourceLp;
 546         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 547         unsigned long flags;
 548         int i, acked = 0;
 549
 550         BUG_ON(! cnx);
 551
 552         switch (event->base_event.xSubtype) {
 553         case VETH_EVENT_CAP:
 554                 veth_take_cap(cnx, event);
 555                 break;
 556         case VETH_EVENT_MONITOR:
 557                 /* do nothing... this'll hang out here til we're dead,
 558                  * and the hypervisor will return it for us. */
 559                 break;
 560         case VETH_EVENT_FRAMES_ACK:
 561                 spin_lock_irqsave(&cnx->lock, flags);
 562
 563                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 564                         u16 msgnum = event->u.frames_ack_data.token[i];
 565
 566                         if (msgnum < VETH_NUMBUFFERS) {
 567                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 568                                 cnx->outstanding_tx--;
 569                                 acked++;
 570                         }
 571                 }
 572
 573                 if (acked > 0) {
 574                         cnx->last_contact = jiffies;
 575                         veth_wake_queues(cnx);
 576                 }
 577
 578                 spin_unlock_irqrestore(&cnx->lock, flags);
 579                 break;
 580         case VETH_EVENT_FRAMES:
 581                 veth_receive(cnx, event);
 582                 break;
 583         default:
 584                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
 585                                 event->base_event.xSubtype, rlp);
 586         };
 587 }
 588
 589 static void veth_handle_event(struct HvLpEvent *event)
 590 {
 591         struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
 592
 593         if (hvlpevent_is_ack(event))
 594                 veth_handle_ack(veth_event);
 595         else
 596                 veth_handle_int(veth_event);
 597 }
 598
 599 static int veth_process_caps(struct veth_lpar_connection *cnx)
 600 {
 601         struct veth_cap_data *remote_caps = &cnx->remote_caps;
 602         int num_acks_needed;
 603
 604         /* Convert timer to jiffies */
 605         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 606
 607         if ( (remote_caps->num_buffers == 0) ||
 608              (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
 609              (remote_caps->ack_threshold == 0) ||
 610              (cnx->ack_timeout == 0) ) {
 611                 veth_error("Received incompatible capabilities from LPAR %d.\n",
 612                                 cnx->remote_lp);
 613                 return HvLpEvent_Rc_InvalidSubtypeData;
 614         }
 615
 616         num_acks_needed = (remote_caps->num_buffers
 617                            / remote_caps->ack_threshold) + 1;
 618
 619         /* FIXME: locking on num_ack_events? */
 620         if (cnx->num_ack_events < num_acks_needed) {
 621                 int num;
 622
 623                 num = veth_allocate_events(cnx->remote_lp,
 624                                            num_acks_needed-cnx->num_ack_events);
 625                 if (num > 0)
 626                         cnx->num_ack_events += num;
 627
 628                 if (cnx->num_ack_events < num_acks_needed) {
 629                         veth_error("Couldn't allocate enough ack events "
 630                                         "for LPAR %d.\n", cnx->remote_lp);
 631
 632                         return HvLpEvent_Rc_BufferNotAvailable;
 633                 }
 634         }
 635
 636
 637         return HvLpEvent_Rc_Good;
 638 }
 639
 640 /* FIXME: The gotos here are a bit dubious */
 641 static void veth_statemachine(struct work_struct *work)
 642 {
 643         struct veth_lpar_connection *cnx =
 644                 container_of(work, struct veth_lpar_connection,
 645                              statemachine_wq.work);
 646         int rlp = cnx->remote_lp;
 647         int rc;
 648
 649         spin_lock_irq(&cnx->lock);
 650
 651  restart:
 652         if (cnx->state & VETH_STATE_RESET) {
 653                 if (cnx->state & VETH_STATE_OPEN)
 654                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 655                                                      HvLpEvent_Type_VirtualLan);
 656
 657                 /*
 658                  * Reset ack data. This prevents the ack_timer actually
 659                  * doing anything, even if it runs one more time when
 660                  * we drop the lock below.
 661                  */
 662                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 663                 cnx->num_pending_acks = 0;
 664
 665                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 666                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 667                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 668                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 669
 670                 /* Clean up any leftover messages */
 671                 if (cnx->msgs) {
 672                         int i;
 673                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 674                                 veth_recycle_msg(cnx, cnx->msgs + i);
 675                 }
 676
 677                 cnx->outstanding_tx = 0;
 678                 veth_wake_queues(cnx);
 679
 680                 /* Drop the lock so we can do stuff that might sleep or
 681                  * take other locks. */
 682                 spin_unlock_irq(&cnx->lock);
 683
 684                 del_timer_sync(&cnx->ack_timer);
 685                 del_timer_sync(&cnx->reset_timer);
 686
 687                 spin_lock_irq(&cnx->lock);
 688
 689                 if (cnx->state & VETH_STATE_RESET)
 690                         goto restart;
 691
 692                 /* Hack, wait for the other end to reset itself. */
 693                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 694                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
 695                         goto out;
 696                 }
 697         }
 698
 699         if (cnx->state & VETH_STATE_SHUTDOWN)
 700                 /* It's all over, do nothing */
 701                 goto out;
 702
 703         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 704                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 705                         goto cant_cope;
 706
 707                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 708                 cnx->src_inst =
 709                         HvCallEvent_getSourceLpInstanceId(rlp,
 710                                                           HvLpEvent_Type_VirtualLan);
 711                 cnx->dst_inst =
 712                         HvCallEvent_getTargetLpInstanceId(rlp,
 713                                                           HvLpEvent_Type_VirtualLan);
 714                 cnx->state |= VETH_STATE_OPEN;
 715         }
 716
 717         if ( (cnx->state & VETH_STATE_OPEN) &&
 718              !(cnx->state & VETH_STATE_SENTMON) ) {
 719                 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
 720                                       HvLpEvent_AckInd_DoAck,
 721                                       HvLpEvent_AckType_DeferredAck,
 722                                       0, 0, 0, 0, 0, 0);
 723
 724                 if (rc == HvLpEvent_Rc_Good) {
 725                         cnx->state |= VETH_STATE_SENTMON;
 726                 } else {
 727                         if ( (rc != HvLpEvent_Rc_PartitionDead) &&
 728                              (rc != HvLpEvent_Rc_PathClosed) )
 729                                 veth_error("Error sending monitor to LPAR %d, "
 730                                                 "rc = %d\n", rlp, rc);
 731
 732                         /* Oh well, hope we get a cap from the other
 733                          * end and do better when that kicks us */
 734                         goto out;
 735                 }
 736         }
 737
 738         if ( (cnx->state & VETH_STATE_OPEN) &&
 739              !(cnx->state & VETH_STATE_SENTCAPS)) {
 740                 u64 *rawcap = (u64 *)&cnx->local_caps;
 741
 742                 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
 743                                       HvLpEvent_AckInd_DoAck,
 744                                       HvLpEvent_AckType_ImmediateAck,
 745                                       0, rawcap[0], rawcap[1], rawcap[2],
 746                                       rawcap[3], rawcap[4]);
 747
 748                 if (rc == HvLpEvent_Rc_Good) {
 749                         cnx->state |= VETH_STATE_SENTCAPS;
 750                 } else {
 751                         if ( (rc != HvLpEvent_Rc_PartitionDead) &&
 752                              (rc != HvLpEvent_Rc_PathClosed) )
 753                                 veth_error("Error sending caps to LPAR %d, "
 754                                                 "rc = %d\n", rlp, rc);
 755
 756                         /* Oh well, hope we get a cap from the other
 757                          * end and do better when that kicks us */
 758                         goto out;
 759                 }
 760         }
 761
 762         if ((cnx->state & VETH_STATE_GOTCAPS) &&
 763             !(cnx->state & VETH_STATE_SENTCAPACK)) {
 764                 struct veth_cap_data *remote_caps = &cnx->remote_caps;
 765
 766                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 767                        sizeof(*remote_caps));
 768
 769                 spin_unlock_irq(&cnx->lock);
 770                 rc = veth_process_caps(cnx);
 771                 spin_lock_irq(&cnx->lock);
 772
 773                 /* We dropped the lock, so recheck for anything which
 774                  * might mess us up */
 775                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 776                         goto restart;
 777
 778                 cnx->cap_event.base_event.xRc = rc;
 779                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 780                 if (rc == HvLpEvent_Rc_Good)
 781                         cnx->state |= VETH_STATE_SENTCAPACK;
 782                 else
 783                         goto cant_cope;
 784         }
 785
 786         if ((cnx->state & VETH_STATE_GOTCAPACK) &&
 787             (cnx->state & VETH_STATE_GOTCAPS) &&
 788             !(cnx->state & VETH_STATE_READY)) {
 789                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 790                         /* Start the ACK timer */
 791                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 792                         add_timer(&cnx->ack_timer);
 793                         cnx->state |= VETH_STATE_READY;
 794                 } else {
 795                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
 796                                         rlp, cnx->cap_ack_event.base_event.xRc);
 797                         goto cant_cope;
 798                 }
 799         }
 800
 801  out:
 802         spin_unlock_irq(&cnx->lock);
 803         return;
 804
 805  cant_cope:
 806         /* FIXME: we get here if something happens we really can't
 807          * cope with.  The link will never work once we get here, and
 808          * all we can do is not lock the rest of the system up */
 809         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
 810                         " (state = 0x%04lx)\n", rlp, cnx->state);
 811         cnx->state |= VETH_STATE_SHUTDOWN;
 812         spin_unlock_irq(&cnx->lock);
 813 }
 814
 815 static int veth_init_connection(u8 rlp)
 816 {
 817         struct veth_lpar_connection *cnx;
 818         struct veth_msg *msgs;
 819         int i;
 820
 821         if ( (rlp == this_lp) ||
 822              ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 823                 return 0;
 824
 825         cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
 826         if (! cnx)
 827                 return -ENOMEM;
 828
 829         cnx->remote_lp = rlp;
 830         spin_lock_init(&cnx->lock);
 831         INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
 832
 833         init_timer(&cnx->ack_timer);
 834         cnx->ack_timer.function = veth_timed_ack;
 835         cnx->ack_timer.data = (unsigned long) cnx;
 836
 837         init_timer(&cnx->reset_timer);
 838         cnx->reset_timer.function = veth_timed_reset;
 839         cnx->reset_timer.data = (unsigned long) cnx;
 840         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
 841
 842         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 843
 844         veth_cnx[rlp] = cnx;
 845
 846         /* This gets us 1 reference, which is held on behalf of the driver
 847          * infrastructure. It's released at module unload. */
 848         kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
 849
 850         msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
 851         if (! msgs) {
 852                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
 853                 return -ENOMEM;
 854         }
 855
 856         cnx->msgs = msgs;
 857
 858         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 859                 msgs[i].token = i;
 860                 veth_stack_push(cnx, msgs + i);
 861         }
 862
 863         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 864
 865         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 866                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
 867                 return -ENOMEM;
 868         }
 869
 870         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 871         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 872         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 873
 874         return 0;
 875 }
 876
 877 static void veth_stop_connection(struct veth_lpar_connection *cnx)
 878 {
 879         if (!cnx)
 880                 return;
 881
 882         spin_lock_irq(&cnx->lock);
 883         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 884         veth_kick_statemachine(cnx);
 885         spin_unlock_irq(&cnx->lock);
 886
 887         /* There's a slim chance the reset code has just queued the
 888          * statemachine to run in five seconds. If so we need to cancel
 889          * that and requeue the work to run now. */
 890         if (cancel_delayed_work(&cnx->statemachine_wq)) {
 891                 spin_lock_irq(&cnx->lock);
 892                 veth_kick_statemachine(cnx);
 893                 spin_unlock_irq(&cnx->lock);
 894         }
 895
 896         /* Wait for the state machine to run. */
 897         flush_scheduled_work();
 898 }
 899
 900 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
 901 {
 902         if (!cnx)
 903                 return;
 904
 905         if (cnx->num_events > 0)
 906                 mf_deallocate_lp_events(cnx->remote_lp,
 907                                       HvLpEvent_Type_VirtualLan,
 908                                       cnx->num_events,
 909                                       NULL, NULL);
 910         if (cnx->num_ack_events > 0)
 911                 mf_deallocate_lp_events(cnx->remote_lp,
 912                                       HvLpEvent_Type_VirtualLan,
 913                                       cnx->num_ack_events,
 914                                       NULL, NULL);
 915
 916         kfree(cnx->msgs);
 917         veth_cnx[cnx->remote_lp] = NULL;
 918         kfree(cnx);
 919 }
 920
 921 static void veth_release_connection(struct kobject *kobj)
 922 {
 923         struct veth_lpar_connection *cnx;
 924         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 925         veth_stop_connection(cnx);
 926         veth_destroy_connection(cnx);
 927 }
 928
 929 /*
 930  * net_device code
 931  */
 932
 933 static int veth_open(struct net_device *dev)
 934 {
 935         netif_start_queue(dev);
 936         return 0;
 937 }
 938
 939 static int veth_close(struct net_device *dev)
 940 {
 941         netif_stop_queue(dev);
 942         return 0;
 943 }
 944
 945 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 946 {
 947         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 948                 return -EINVAL;
 949         dev->mtu = new_mtu;
 950         return 0;
 951 }
 952
 953 static void veth_set_multicast_list(struct net_device *dev)
 954 {
 955         struct veth_port *port = netdev_priv(dev);
 956         unsigned long flags;
 957
 958         write_lock_irqsave(&port->mcast_gate, flags);
 959
 960         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
 961                         (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
 962                 port->promiscuous = 1;
 963         } else {
 964                 struct dev_mc_list *dmi;
 965
 966                 port->promiscuous = 0;
 967
 968                 /* Update table */
 969                 port->num_mcast = 0;
 970
 971                 netdev_for_each_mc_addr(dmi, dev) {
 972                         u8 *addr = dmi->dmi_addr;
 973                         u64 xaddr = 0;
 974
 975                         if (addr[0] & 0x01) {/* multicast address? */
 976                                 memcpy(&xaddr, addr, ETH_ALEN);
 977                                 port->mcast_addr[port->num_mcast] = xaddr;
 978                                 port->num_mcast++;
 979                         }
 980                 }
 981         }
 982
 983         write_unlock_irqrestore(&port->mcast_gate, flags);
 984 }
 985
 986 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 987 {
 988         strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
 989         info->driver[sizeof(info->driver) - 1] = '\0';
 990         strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
 991         info->version[sizeof(info->version) - 1] = '\0';
 992 }
 993
 994 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
 995 {
 996         ecmd->supported = (SUPPORTED_1000baseT_Full
 997                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 998         ecmd->advertising = (SUPPORTED_1000baseT_Full
 999                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1000         ecmd->port = PORT_FIBRE;
1001         ecmd->transceiver = XCVR_INTERNAL;
1002         ecmd->phy_address = 0;
1003         ecmd->speed = SPEED_1000;
1004         ecmd->duplex = DUPLEX_FULL;
1005         ecmd->autoneg = AUTONEG_ENABLE;
1006         ecmd->maxtxpkt = 120;
1007         ecmd->maxrxpkt = 120;
1008         return 0;
1009 }
1010
1011 static u32 veth_get_link(struct net_device *dev)
1012 {
1013         return 1;
1014 }
1015
1016 static const struct ethtool_ops ops = {
1017         .get_drvinfo = veth_get_drvinfo,
1018         .get_settings = veth_get_settings,
1019         .get_link = veth_get_link,
1020 };
1021
1022 static const struct net_device_ops veth_netdev_ops = {
1023         .ndo_open               = veth_open,
1024         .ndo_stop               = veth_close,
1025         .ndo_start_xmit         = veth_start_xmit,
1026         .ndo_change_mtu         = veth_change_mtu,
1027         .ndo_set_multicast_list = veth_set_multicast_list,
1028         .ndo_set_mac_address    = NULL,
1029         .ndo_validate_addr      = eth_validate_addr,
1030 };
1031
1032 static struct net_device *veth_probe_one(int vlan,
1033                 struct vio_dev *vio_dev)
1034 {
1035         struct net_device *dev;
1036         struct veth_port *port;
1037         struct device *vdev = &vio_dev->dev;
1038         int i, rc;
1039         const unsigned char *mac_addr;
1040
1041         mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1042         if (mac_addr == NULL)
1043                 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1044         if (mac_addr == NULL) {
1045                 veth_error("Unable to fetch MAC address from device tree.\n");
1046                 return NULL;
1047         }
1048
1049         dev = alloc_etherdev(sizeof (struct veth_port));
1050         if (! dev) {
1051                 veth_error("Unable to allocate net_device structure!\n");
1052                 return NULL;
1053         }
1054
1055         port = netdev_priv(dev);
1056
1057         spin_lock_init(&port->queue_lock);
1058         rwlock_init(&port->mcast_gate);
1059         port->stopped_map = 0;
1060
1061         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1062                 HvLpVirtualLanIndexMap map;
1063
1064                 if (i == this_lp)
1065                         continue;
1066                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1067                 if (map & (0x8000 >> vlan))
1068                         port->lpar_map |= (1 << i);
1069         }
1070         port->dev = vdev;
1071
1072         memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1073
1074         dev->mtu = VETH_MAX_MTU;
1075
1076         memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1077
1078         dev->netdev_ops = &veth_netdev_ops;
1079         SET_ETHTOOL_OPS(dev, &ops);
1080
1081         SET_NETDEV_DEV(dev, vdev);
1082
1083         rc = register_netdev(dev);
1084         if (rc != 0) {
1085                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1086                 free_netdev(dev);
1087                 return NULL;
1088         }
1089
1090         kobject_init(&port->kobject, &veth_port_ktype);
1091         if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
1092                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1093
1094         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1095                         dev->name, vlan, port->lpar_map);
1096
1097         return dev;
1098 }
1099
1100 /*
1101  * Tx path
1102  */
1103
1104 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1105                                 struct net_device *dev)
1106 {
1107         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1108         struct veth_port *port = netdev_priv(dev);
1109         HvLpEvent_Rc rc;
1110         struct veth_msg *msg = NULL;
1111         unsigned long flags;
1112
1113         if (! cnx)
1114                 return 0;
1115
1116         spin_lock_irqsave(&cnx->lock, flags);
1117
1118         if (! (cnx->state & VETH_STATE_READY))
1119                 goto no_error;
1120
1121         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1122                 goto drop;
1123
1124         msg = veth_stack_pop(cnx);
1125         if (! msg)
1126                 goto drop;
1127
1128         msg->in_use = 1;
1129         msg->skb = skb_get(skb);
1130
1131         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1132                                 skb->len, DMA_TO_DEVICE);
1133
1134         if (dma_mapping_error(port->dev, msg->data.addr[0]))
1135                 goto recycle_and_drop;
1136
1137         msg->dev = port->dev;
1138         msg->data.len[0] = skb->len;
1139         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1140
1141         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1142
1143         if (rc != HvLpEvent_Rc_Good)
1144                 goto recycle_and_drop;
1145
1146         /* If the timer's not already running, start it now. */
1147         if (0 == cnx->outstanding_tx)
1148                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1149
1150         cnx->last_contact = jiffies;
1151         cnx->outstanding_tx++;
1152
1153         if (veth_stack_is_empty(cnx))
1154                 veth_stop_queues(cnx);
1155
1156  no_error:
1157         spin_unlock_irqrestore(&cnx->lock, flags);
1158         return 0;
1159
1160  recycle_and_drop:
1161         veth_recycle_msg(cnx, msg);
1162  drop:
1163         spin_unlock_irqrestore(&cnx->lock, flags);
1164         return 1;
1165 }
1166
1167 static void veth_transmit_to_many(struct sk_buff *skb,
1168                                           HvLpIndexMap lpmask,
1169                                           struct net_device *dev)
1170 {
1171         int i, success, error;
1172
1173         success = error = 0;
1174
1175         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1176                 if ((lpmask & (1 << i)) == 0)
1177                         continue;
1178
1179                 if (veth_transmit_to_one(skb, i, dev))
1180                         error = 1;
1181                 else
1182                         success = 1;
1183         }
1184
1185         if (error)
1186                 dev->stats.tx_errors++;
1187
1188         if (success) {
1189                 dev->stats.tx_packets++;
1190                 dev->stats.tx_bytes += skb->len;
1191         }
1192 }
1193
1194 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1195 {
1196         unsigned char *frame = skb->data;
1197         struct veth_port *port = netdev_priv(dev);
1198         HvLpIndexMap lpmask;
1199
1200         if (! (frame[0] & 0x01)) {
1201                 /* unicast packet */
1202                 HvLpIndex rlp = frame[5];
1203
1204                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1205                         dev_kfree_skb(skb);
1206                         return NETDEV_TX_OK;
1207                 }
1208
1209                 lpmask = 1 << rlp;
1210         } else {
1211                 lpmask = port->lpar_map;
1212         }
1213
1214         veth_transmit_to_many(skb, lpmask, dev);
1215
1216         dev_kfree_skb(skb);
1217
1218         return NETDEV_TX_OK;
1219 }
1220
1221 /* You must hold the connection's lock when you call this function. */
1222 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1223                              struct veth_msg *msg)
1224 {
1225         u32 dma_address, dma_length;
1226
1227         if (msg->in_use) {
1228                 msg->in_use = 0;
1229                 dma_address = msg->data.addr[0];
1230                 dma_length = msg->data.len[0];
1231
1232                 if (!dma_mapping_error(msg->dev, dma_address))
1233                         dma_unmap_single(msg->dev, dma_address, dma_length,
1234                                         DMA_TO_DEVICE);
1235
1236                 if (msg->skb) {
1237                         dev_kfree_skb_any(msg->skb);
1238                         msg->skb = NULL;
1239                 }
1240
1241                 memset(&msg->data, 0, sizeof(msg->data));
1242                 veth_stack_push(cnx, msg);
1243         } else if (cnx->state & VETH_STATE_OPEN) {
1244                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1245                                 cnx->remote_lp, msg->token);
1246         }
1247 }
1248
1249 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1250 {
1251         int i;
1252
1253         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1254                 struct net_device *dev = veth_dev[i];
1255                 struct veth_port *port;
1256                 unsigned long flags;
1257
1258                 if (! dev)
1259                         continue;
1260
1261                 port = netdev_priv(dev);
1262
1263                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1264                         continue;
1265
1266                 spin_lock_irqsave(&port->queue_lock, flags);
1267
1268                 port->stopped_map &= ~(1 << cnx->remote_lp);
1269
1270                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1271                         veth_debug("cnx %d: woke queue for %s.\n",
1272                                         cnx->remote_lp, dev->name);
1273                         netif_wake_queue(dev);
1274                 }
1275                 spin_unlock_irqrestore(&port->queue_lock, flags);
1276         }
1277 }
1278
1279 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1280 {
1281         int i;
1282
1283         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1284                 struct net_device *dev = veth_dev[i];
1285                 struct veth_port *port;
1286
1287                 if (! dev)
1288                         continue;
1289
1290                 port = netdev_priv(dev);
1291
1292                 /* If this cnx is not on the vlan for this port, continue */
1293                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1294                         continue;
1295
1296                 spin_lock(&port->queue_lock);
1297
1298                 netif_stop_queue(dev);
1299                 port->stopped_map |= (1 << cnx->remote_lp);
1300
1301                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1302                                 cnx->remote_lp, dev->name, port->stopped_map);
1303
1304                 spin_unlock(&port->queue_lock);
1305         }
1306 }
1307
1308 static void veth_timed_reset(unsigned long ptr)
1309 {
1310         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1311         unsigned long trigger_time, flags;
1312
1313         /* FIXME is it possible this fires after veth_stop_connection()?
1314          * That would reschedule the statemachine for 5 seconds and probably
1315          * execute it after the module's been unloaded. Hmm. */
1316
1317         spin_lock_irqsave(&cnx->lock, flags);
1318
1319         if (cnx->outstanding_tx > 0) {
1320                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1321
1322                 if (trigger_time < jiffies) {
1323                         cnx->state |= VETH_STATE_RESET;
1324                         veth_kick_statemachine(cnx);
1325                         veth_error("%d packets not acked by LPAR %d within %d "
1326                                         "seconds, resetting.\n",
1327                                         cnx->outstanding_tx, cnx->remote_lp,
1328                                         cnx->reset_timeout / HZ);
1329                 } else {
1330                         /* Reschedule the timer */
1331                         trigger_time = jiffies + cnx->reset_timeout;
1332                         mod_timer(&cnx->reset_timer, trigger_time);
1333                 }
1334         }
1335
1336         spin_unlock_irqrestore(&cnx->lock, flags);
1337 }
1338
1339 /*
1340  * Rx path
1341  */
1342
1343 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1344 {
1345         int wanted = 0;
1346         int i;
1347         unsigned long flags;
1348
1349         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1350                 return 1;
1351
1352         read_lock_irqsave(&port->mcast_gate, flags);
1353
1354         if (port->promiscuous) {
1355                 wanted = 1;
1356                 goto out;
1357         }
1358
1359         for (i = 0; i < port->num_mcast; ++i) {
1360                 if (port->mcast_addr[i] == mac_addr) {
1361                         wanted = 1;
1362                         break;
1363                 }
1364         }
1365
1366  out:
1367         read_unlock_irqrestore(&port->mcast_gate, flags);
1368
1369         return wanted;
1370 }
1371
1372 struct dma_chunk {
1373         u64 addr;
1374         u64 size;
1375 };
1376
1377 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1378
1379 static inline void veth_build_dma_list(struct dma_chunk *list,
1380                                        unsigned char *p, unsigned long length)
1381 {
1382         unsigned long done;
1383         int i = 1;
1384
1385         /* FIXME: skbs are contiguous in real addresses.  Do we
1386          * really need to break it into PAGE_SIZE chunks, or can we do
1387          * it just at the granularity of iSeries real->absolute
1388          * mapping?  Indeed, given the way the allocator works, can we
1389          * count on them being absolutely contiguous? */
1390         list[0].addr = iseries_hv_addr(p);
1391         list[0].size = min(length,
1392                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1393
1394         done = list[0].size;
1395         while (done < length) {
1396                 list[i].addr = iseries_hv_addr(p + done);
1397                 list[i].size = min(length-done, PAGE_SIZE);
1398                 done += list[i].size;
1399                 i++;
1400         }
1401 }
1402
1403 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1404 {
1405         HvLpEvent_Rc rc;
1406
1407         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1408                              0, &cnx->pending_acks);
1409
1410         if (rc != HvLpEvent_Rc_Good)
1411                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1412                                 cnx->remote_lp, (int)rc);
1413
1414         cnx->num_pending_acks = 0;
1415         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1416 }
1417
1418 static void veth_receive(struct veth_lpar_connection *cnx,
1419                          struct veth_lpevent *event)
1420 {
1421         struct veth_frames_data *senddata = &event->u.frames_data;
1422         int startchunk = 0;
1423         int nchunks;
1424         unsigned long flags;
1425         HvLpDma_Rc rc;
1426
1427         do {
1428                 u16 length = 0;
1429                 struct sk_buff *skb;
1430                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1431                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1432                 u64 dest;
1433                 HvLpVirtualLanIndex vlan;
1434                 struct net_device *dev;
1435                 struct veth_port *port;
1436
1437                 /* FIXME: do we need this? */
1438                 memset(local_list, 0, sizeof(local_list));
1439                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1440
1441                 /* a 0 address marks the end of the valid entries */
1442                 if (senddata->addr[startchunk] == 0)
1443                         break;
1444
1445                 /* make sure that we have at least 1 EOF entry in the
1446                  * remaining entries */
1447                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1448                         veth_error("Missing EOF fragment in event "
1449                                         "eofmask = 0x%x startchunk = %d\n",
1450                                         (unsigned)senddata->eofmask,
1451                                         startchunk);
1452                         break;
1453                 }
1454
1455                 /* build list of chunks in this frame */
1456                 nchunks = 0;
1457                 do {
1458                         remote_list[nchunks].addr =
1459                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1460                         remote_list[nchunks].size =
1461                                 senddata->len[startchunk+nchunks];
1462                         length += remote_list[nchunks].size;
1463                 } while (! (senddata->eofmask &
1464                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1465
1466                 /* length == total length of all chunks */
1467                 /* nchunks == # of chunks in this frame */
1468
1469                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1470                         veth_error("Received oversize frame from LPAR %d "
1471                                         "(length = %d)\n",
1472                                         cnx->remote_lp, length);
1473                         continue;
1474                 }
1475
1476                 skb = alloc_skb(length, GFP_ATOMIC);
1477                 if (!skb)
1478                         continue;
1479
1480                 veth_build_dma_list(local_list, skb->data, length);
1481
1482                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1483                                             event->base_event.xSourceLp,
1484                                             HvLpDma_Direction_RemoteToLocal,
1485                                             cnx->src_inst,
1486                                             cnx->dst_inst,
1487                                             HvLpDma_AddressType_RealAddress,
1488                                             HvLpDma_AddressType_TceIndex,
1489                                             iseries_hv_addr(&local_list),
1490                                             iseries_hv_addr(&remote_list),
1491                                             length);
1492                 if (rc != HvLpDma_Rc_Good) {
1493                         dev_kfree_skb_irq(skb);
1494                         continue;
1495                 }
1496
1497                 vlan = skb->data[9];
1498                 dev = veth_dev[vlan];
1499                 if (! dev) {
1500                         /*
1501                          * Some earlier versions of the driver sent
1502                          * broadcasts down all connections, even to lpars
1503                          * that weren't on the relevant vlan. So ignore
1504                          * packets belonging to a vlan we're not on.
1505                          * We can also be here if we receive packets while
1506                          * the driver is going down, because then dev is NULL.
1507                          */
1508                         dev_kfree_skb_irq(skb);
1509                         continue;
1510                 }
1511
1512                 port = netdev_priv(dev);
1513                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1514
1515                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1516                         dev_kfree_skb_irq(skb);
1517                         continue;
1518                 }
1519                 if (! veth_frame_wanted(port, dest)) {
1520                         dev_kfree_skb_irq(skb);
1521                         continue;
1522                 }
1523
1524                 skb_put(skb, length);
1525                 skb->protocol = eth_type_trans(skb, dev);
1526                 skb->ip_summed = CHECKSUM_NONE;
1527                 netif_rx(skb);  /* send it up */
1528                 dev->stats.rx_packets++;
1529                 dev->stats.rx_bytes += length;
1530         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1531
1532         /* Ack it */
1533         spin_lock_irqsave(&cnx->lock, flags);
1534         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1535
1536         cnx->pending_acks[cnx->num_pending_acks++] =
1537                 event->base_event.xCorrelationToken;
1538
1539         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1540              (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1541                 veth_flush_acks(cnx);
1542
1543         spin_unlock_irqrestore(&cnx->lock, flags);
1544 }
1545
1546 static void veth_timed_ack(unsigned long ptr)
1547 {
1548         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1549         unsigned long flags;
1550
1551         /* Ack all the events */
1552         spin_lock_irqsave(&cnx->lock, flags);
1553         if (cnx->num_pending_acks > 0)
1554                 veth_flush_acks(cnx);
1555
1556         /* Reschedule the timer */
1557         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1558         add_timer(&cnx->ack_timer);
1559         spin_unlock_irqrestore(&cnx->lock, flags);
1560 }
1561
1562 static int veth_remove(struct vio_dev *vdev)
1563 {
1564         struct veth_lpar_connection *cnx;
1565         struct net_device *dev;
1566         struct veth_port *port;
1567         int i;
1568
1569         dev = veth_dev[vdev->unit_address];
1570
1571         if (! dev)
1572                 return 0;
1573
1574         port = netdev_priv(dev);
1575
1576         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1577                 cnx = veth_cnx[i];
1578
1579                 if (cnx && (port->lpar_map & (1 << i))) {
1580                         /* Drop our reference to connections on our VLAN */
1581                         kobject_put(&cnx->kobject);
1582                 }
1583         }
1584
1585         veth_dev[vdev->unit_address] = NULL;
1586         kobject_del(&port->kobject);
1587         kobject_put(&port->kobject);
1588         unregister_netdev(dev);
1589         free_netdev(dev);
1590
1591         return 0;
1592 }
1593
1594 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1595 {
1596         int i = vdev->unit_address;
1597         struct net_device *dev;
1598         struct veth_port *port;
1599
1600         dev = veth_probe_one(i, vdev);
1601         if (dev == NULL) {
1602                 veth_remove(vdev);
1603                 return 1;
1604         }
1605         veth_dev[i] = dev;
1606
1607         port = (struct veth_port*)netdev_priv(dev);
1608
1609         /* Start the state machine on each connection on this vlan. If we're
1610          * the first dev to do so this will commence link negotiation */
1611         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1612                 struct veth_lpar_connection *cnx;
1613
1614                 if (! (port->lpar_map & (1 << i)))
1615                         continue;
1616
1617                 cnx = veth_cnx[i];
1618                 if (!cnx)
1619                         continue;
1620
1621                 kobject_get(&cnx->kobject);
1622                 veth_kick_statemachine(cnx);
1623         }
1624
1625         return 0;
1626 }
1627
1628 /**
1629  * veth_device_table: Used by vio.c to match devices that we
1630  * support.
1631  */
1632 static struct vio_device_id veth_device_table[] __devinitdata = {
1633         { "network", "IBM,iSeries-l-lan" },
1634         { "", "" }
1635 };
1636 MODULE_DEVICE_TABLE(vio, veth_device_table);
1637
1638 static struct vio_driver veth_driver = {
1639         .id_table = veth_device_table,
1640         .probe = veth_probe,
1641         .remove = veth_remove,
1642         .driver = {
1643                 .name = DRV_NAME,
1644                 .owner = THIS_MODULE,
1645         }
1646 };
1647
1648 /*
1649  * Module initialization/cleanup
1650  */
1651
1652 static void __exit veth_module_cleanup(void)
1653 {
1654         int i;
1655         struct veth_lpar_connection *cnx;
1656
1657         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1658         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1659
1660         /* Make sure any work queued from Hypervisor callbacks is finished. */
1661         flush_scheduled_work();
1662
1663         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1664                 cnx = veth_cnx[i];
1665
1666                 if (!cnx)
1667                         continue;
1668
1669                 /* Remove the connection from sysfs */
1670                 kobject_del(&cnx->kobject);
1671                 /* Drop the driver's reference to the connection */
1672                 kobject_put(&cnx->kobject);
1673         }
1674
1675         /* Unregister the driver, which will close all the netdevs and stop
1676          * the connections when they're no longer referenced. */
1677         vio_unregister_driver(&veth_driver);
1678 }
1679 module_exit(veth_module_cleanup);
1680
1681 static int __init veth_module_init(void)
1682 {
1683         int i;
1684         int rc;
1685
1686         if (!firmware_has_feature(FW_FEATURE_ISERIES))
1687                 return -ENODEV;
1688
1689         this_lp = HvLpConfig_getLpIndex_outline();
1690
1691         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1692                 rc = veth_init_connection(i);
1693                 if (rc != 0)
1694                         goto error;
1695         }
1696
1697         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1698                                   &veth_handle_event);
1699
1700         rc = vio_register_driver(&veth_driver);
1701         if (rc != 0)
1702                 goto error;
1703
1704         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1705                 struct kobject *kobj;
1706
1707                 if (!veth_cnx[i])
1708                         continue;
1709
1710                 kobj = &veth_cnx[i]->kobject;
1711                 /* If the add failes, complain but otherwise continue */
1712                 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1713                                         "cnx%.2d", veth_cnx[i]->remote_lp))
1714                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1715         }
1716
1717         return 0;
1718
1719 error:
1720         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1721                 veth_destroy_connection(veth_cnx[i]);
1722         }
1723
1724         return rc;
1725 }
1726 module_init(veth_module_init);