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 #include <linux/slab.h>
  73
  74 #include <asm/abs_addr.h>
  75 #include <asm/iseries/mf.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/firmware.h>
  78 #include <asm/iseries/hv_lp_config.h>
  79 #include <asm/iseries/hv_types.h>
  80 #include <asm/iseries/hv_lp_event.h>
  81 #include <asm/iommu.h>
  82 #include <asm/vio.h>
  83
  84 #undef DEBUG
  85
  86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  88 MODULE_LICENSE("GPL");
  89
  90 #define VETH_EVENT_CAP  (0)
  91 #define VETH_EVENT_FRAMES       (1)
  92 #define VETH_EVENT_MONITOR      (2)
  93 #define VETH_EVENT_FRAMES_ACK   (3)
  94
  95 #define VETH_MAX_ACKS_PER_MSG   (20)
  96 #define VETH_MAX_FRAMES_PER_MSG (6)
  97
  98 struct veth_frames_data {
  99         u32 addr[VETH_MAX_FRAMES_PER_MSG];
 100         u16 len[VETH_MAX_FRAMES_PER_MSG];
 101         u32 eofmask;
 102 };
 103 #define VETH_EOF_SHIFT          (32-VETH_MAX_FRAMES_PER_MSG)
 104
 105 struct veth_frames_ack_data {
 106         u16 token[VETH_MAX_ACKS_PER_MSG];
 107 };
 108
 109 struct veth_cap_data {
 110         u8 caps_version;
 111         u8 rsvd1;
 112         u16 num_buffers;
 113         u16 ack_threshold;
 114         u16 rsvd2;
 115         u32 ack_timeout;
 116         u32 rsvd3;
 117         u64 rsvd4[3];
 118 };
 119
 120 struct veth_lpevent {
 121         struct HvLpEvent base_event;
 122         union {
 123                 struct veth_cap_data caps_data;
 124                 struct veth_frames_data frames_data;
 125                 struct veth_frames_ack_data frames_ack_data;
 126         } u;
 127
 128 };
 129
 130 #define DRV_NAME        "iseries_veth"
 131 #define DRV_VERSION     "2.0"
 132
 133 #define VETH_NUMBUFFERS         (120)
 134 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
 135 #define VETH_MAX_MCAST          (12)
 136
 137 #define VETH_MAX_MTU            (9000)
 138
 139 #if VETH_NUMBUFFERS < 10
 140 #define ACK_THRESHOLD           (1)
 141 #elif VETH_NUMBUFFERS < 20
 142 #define ACK_THRESHOLD           (4)
 143 #elif VETH_NUMBUFFERS < 40
 144 #define ACK_THRESHOLD           (10)
 145 #else
 146 #define ACK_THRESHOLD           (20)
 147 #endif
 148
 149 #define VETH_STATE_SHUTDOWN     (0x0001)
 150 #define VETH_STATE_OPEN         (0x0002)
 151 #define VETH_STATE_RESET        (0x0004)
 152 #define VETH_STATE_SENTMON      (0x0008)
 153 #define VETH_STATE_SENTCAPS     (0x0010)
 154 #define VETH_STATE_GOTCAPACK    (0x0020)
 155 #define VETH_STATE_GOTCAPS      (0x0040)
 156 #define VETH_STATE_SENTCAPACK   (0x0080)
 157 #define VETH_STATE_READY        (0x0100)
 158
 159 struct veth_msg {
 160         struct veth_msg *next;
 161         struct veth_frames_data data;
 162         int token;
 163         int in_use;
 164         struct sk_buff *skb;
 165         struct device *dev;
 166 };
 167
 168 struct veth_lpar_connection {
 169         HvLpIndex remote_lp;
 170         struct delayed_work statemachine_wq;
 171         struct veth_msg *msgs;
 172         int num_events;
 173         struct veth_cap_data local_caps;
 174
 175         struct kobject kobject;
 176         struct timer_list ack_timer;
 177
 178         struct timer_list reset_timer;
 179         unsigned int reset_timeout;
 180         unsigned long last_contact;
 181         int outstanding_tx;
 182
 183         spinlock_t lock;
 184         unsigned long state;
 185         HvLpInstanceId src_inst;
 186         HvLpInstanceId dst_inst;
 187         struct veth_lpevent cap_event, cap_ack_event;
 188         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 189         u32 num_pending_acks;
 190
 191         int num_ack_events;
 192         struct veth_cap_data remote_caps;
 193         u32 ack_timeout;
 194
 195         struct veth_msg *msg_stack_head;
 196 };
 197
 198 struct veth_port {
 199         struct device *dev;
 200         u64 mac_addr;
 201         HvLpIndexMap lpar_map;
 202
 203         /* queue_lock protects the stopped_map and dev's queue. */
 204         spinlock_t queue_lock;
 205         HvLpIndexMap stopped_map;
 206
 207         /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
 208         rwlock_t mcast_gate;
 209         int promiscuous;
 210         int num_mcast;
 211         u64 mcast_addr[VETH_MAX_MCAST];
 212
 213         struct kobject kobject;
 214 };
 215
 216 static HvLpIndex this_lp;
 217 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
 218 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
 219
 220 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
 221 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
 222 static void veth_wake_queues(struct veth_lpar_connection *cnx);
 223 static void veth_stop_queues(struct veth_lpar_connection *cnx);
 224 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
 225 static void veth_release_connection(struct kobject *kobject);
 226 static void veth_timed_ack(unsigned long ptr);
 227 static void veth_timed_reset(unsigned long ptr);
 228
 229 /*
 230  * Utility functions
 231  */
 232
 233 #define veth_info(fmt, args...) \
 234         printk(KERN_INFO DRV_NAME ": " fmt, ## args)
 235
 236 #define veth_error(fmt, args...) \
 237         printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
 238
 239 #ifdef DEBUG
 240 #define veth_debug(fmt, args...) \
 241         printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
 242 #else
 243 #define veth_debug(fmt, args...) do {} while (0)
 244 #endif
 245
 246 /* You must hold the connection's lock when you call this function. */
 247 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
 248                                    struct veth_msg *msg)
 249 {
 250         msg->next = cnx->msg_stack_head;
 251         cnx->msg_stack_head = msg;
 252 }
 253
 254 /* You must hold the connection's lock when you call this function. */
 255 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
 256 {
 257         struct veth_msg *msg;
 258
 259         msg = cnx->msg_stack_head;
 260         if (msg)
 261                 cnx->msg_stack_head = cnx->msg_stack_head->next;
 262
 263         return msg;
 264 }
 265
 266 /* You must hold the connection's lock when you call this function. */
 267 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
 268 {
 269         return cnx->msg_stack_head == NULL;
 270 }
 271
 272 static inline HvLpEvent_Rc
 273 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
 274                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
 275                  u64 token,
 276                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
 277 {
 278         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
 279                                              HvLpEvent_Type_VirtualLan,
 280                                              subtype, ackind, acktype,
 281                                              cnx->src_inst,
 282                                              cnx->dst_inst,
 283                                              token, data1, data2, data3,
 284                                              data4, data5);
 285 }
 286
 287 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
 288                                            u16 subtype, u64 token, void *data)
 289 {
 290         u64 *p = (u64 *) data;
 291
 292         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
 293                                 HvLpEvent_AckType_ImmediateAck,
 294                                 token, p[0], p[1], p[2], p[3], p[4]);
 295 }
 296
 297 struct veth_allocation {
 298         struct completion c;
 299         int num;
 300 };
 301
 302 static void veth_complete_allocation(void *parm, int number)
 303 {
 304         struct veth_allocation *vc = (struct veth_allocation *)parm;
 305
 306         vc->num = number;
 307         complete(&vc->c);
 308 }
 309
 310 static int veth_allocate_events(HvLpIndex rlp, int number)
 311 {
 312         struct veth_allocation vc =
 313                 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
 314
 315         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
 316                             sizeof(struct veth_lpevent), number,
 317                             &veth_complete_allocation, &vc);
 318         wait_for_completion(&vc.c);
 319
 320         return vc.num;
 321 }
 322
 323 /*
 324  * sysfs support
 325  */
 326
 327 struct veth_cnx_attribute {
 328         struct attribute attr;
 329         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
 330         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
 331 };
 332
 333 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
 334                 struct attribute *attr, char *buf)
 335 {
 336         struct veth_cnx_attribute *cnx_attr;
 337         struct veth_lpar_connection *cnx;
 338
 339         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
 340         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 341
 342         if (!cnx_attr->show)
 343                 return -EIO;
 344
 345         return cnx_attr->show(cnx, buf);
 346 }
 347
 348 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
 349 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
 350 {                                                                       \
 351         return sprintf(buf, _format, _expression);                      \
 352 }                                                                       \
 353 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
 354
 355 #define SIMPLE_CNX_ATTR(_name)  \
 356         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
 357
 358 SIMPLE_CNX_ATTR(outstanding_tx);
 359 SIMPLE_CNX_ATTR(remote_lp);
 360 SIMPLE_CNX_ATTR(num_events);
 361 SIMPLE_CNX_ATTR(src_inst);
 362 SIMPLE_CNX_ATTR(dst_inst);
 363 SIMPLE_CNX_ATTR(num_pending_acks);
 364 SIMPLE_CNX_ATTR(num_ack_events);
 365 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
 366 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
 367 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
 368 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
 369                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
 370
 371 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
 372
 373 static struct attribute *veth_cnx_default_attrs[] = {
 374         GET_CNX_ATTR(outstanding_tx),
 375         GET_CNX_ATTR(remote_lp),
 376         GET_CNX_ATTR(num_events),
 377         GET_CNX_ATTR(reset_timeout),
 378         GET_CNX_ATTR(last_contact),
 379         GET_CNX_ATTR(state),
 380         GET_CNX_ATTR(src_inst),
 381         GET_CNX_ATTR(dst_inst),
 382         GET_CNX_ATTR(num_pending_acks),
 383         GET_CNX_ATTR(num_ack_events),
 384         GET_CNX_ATTR(ack_timeout),
 385         NULL
 386 };
 387
 388 static const struct sysfs_ops veth_cnx_sysfs_ops = {
 389                 .show = veth_cnx_attribute_show
 390 };
 391
 392 static struct kobj_type veth_lpar_connection_ktype = {
 393         .release        = veth_release_connection,
 394         .sysfs_ops      = &veth_cnx_sysfs_ops,
 395         .default_attrs  = veth_cnx_default_attrs
 396 };
 397
 398 struct veth_port_attribute {
 399         struct attribute attr;
 400         ssize_t (*show)(struct veth_port *, char *buf);
 401         ssize_t (*store)(struct veth_port *, const char *buf);
 402 };
 403
 404 static ssize_t veth_port_attribute_show(struct kobject *kobj,
 405                 struct attribute *attr, char *buf)
 406 {
 407         struct veth_port_attribute *port_attr;
 408         struct veth_port *port;
 409
 410         port_attr = container_of(attr, struct veth_port_attribute, attr);
 411         port = container_of(kobj, struct veth_port, kobject);
 412
 413         if (!port_attr->show)
 414                 return -EIO;
 415
 416         return port_attr->show(port, buf);
 417 }
 418
 419 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
 420 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
 421 {                                                                       \
 422         return sprintf(buf, _format, _expression);                      \
 423 }                                                                       \
 424 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
 425
 426 #define SIMPLE_PORT_ATTR(_name) \
 427         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
 428
 429 SIMPLE_PORT_ATTR(promiscuous);
 430 SIMPLE_PORT_ATTR(num_mcast);
 431 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
 432 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
 433 CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
 434
 435 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
 436 static struct attribute *veth_port_default_attrs[] = {
 437         GET_PORT_ATTR(mac_addr),
 438         GET_PORT_ATTR(lpar_map),
 439         GET_PORT_ATTR(stopped_map),
 440         GET_PORT_ATTR(promiscuous),
 441         GET_PORT_ATTR(num_mcast),
 442         NULL
 443 };
 444
 445 static const struct sysfs_ops veth_port_sysfs_ops = {
 446         .show = veth_port_attribute_show
 447 };
 448
 449 static struct kobj_type veth_port_ktype = {
 450         .sysfs_ops      = &veth_port_sysfs_ops,
 451         .default_attrs  = veth_port_default_attrs
 452 };
 453
 454 /*
 455  * LPAR connection code
 456  */
 457
 458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 459 {
 460         schedule_delayed_work(&cnx->statemachine_wq, 0);
 461 }
 462
 463 static void veth_take_cap(struct veth_lpar_connection *cnx,
 464                           struct veth_lpevent *event)
 465 {
 466         unsigned long flags;
 467
 468         spin_lock_irqsave(&cnx->lock, flags);
 469         /* Receiving caps may mean the other end has just come up, so
 470          * we need to reload the instance ID of the far end */
 471         cnx->dst_inst =
 472                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 473                                                   HvLpEvent_Type_VirtualLan);
 474
 475         if (cnx->state & VETH_STATE_GOTCAPS) {
 476                 veth_error("Received a second capabilities from LPAR %d.\n",
 477                            cnx->remote_lp);
 478                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 479                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 480         } else {
 481                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 482                 cnx->state |= VETH_STATE_GOTCAPS;
 483                 veth_kick_statemachine(cnx);
 484         }
 485         spin_unlock_irqrestore(&cnx->lock, flags);
 486 }
 487
 488 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 489                               struct veth_lpevent *event)
 490 {
 491         unsigned long flags;
 492
 493         spin_lock_irqsave(&cnx->lock, flags);
 494         if (cnx->state & VETH_STATE_GOTCAPACK) {
 495                 veth_error("Received a second capabilities ack from LPAR %d.\n",
 496                            cnx->remote_lp);
 497         } else {
 498                 memcpy(&cnx->cap_ack_event, event,
 499                        sizeof(cnx->cap_ack_event));
 500                 cnx->state |= VETH_STATE_GOTCAPACK;
 501                 veth_kick_statemachine(cnx);
 502         }
 503         spin_unlock_irqrestore(&cnx->lock, flags);
 504 }
 505
 506 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 507                                   struct veth_lpevent *event)
 508 {
 509         unsigned long flags;
 510
 511         spin_lock_irqsave(&cnx->lock, flags);
 512         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
 513
 514         /* Avoid kicking the statemachine once we're shutdown.
 515          * It's unnecessary and it could break veth_stop_connection(). */
 516
 517         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 518                 cnx->state |= VETH_STATE_RESET;
 519                 veth_kick_statemachine(cnx);
 520         }
 521         spin_unlock_irqrestore(&cnx->lock, flags);
 522 }
 523
 524 static void veth_handle_ack(struct veth_lpevent *event)
 525 {
 526         HvLpIndex rlp = event->base_event.xTargetLp;
 527         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 528
 529         BUG_ON(! cnx);
 530
 531         switch (event->base_event.xSubtype) {
 532         case VETH_EVENT_CAP:
 533                 veth_take_cap_ack(cnx, event);
 534                 break;
 535         case VETH_EVENT_MONITOR:
 536                 veth_take_monitor_ack(cnx, event);
 537                 break;
 538         default:
 539                 veth_error("Unknown ack type %d from LPAR %d.\n",
 540                                 event->base_event.xSubtype, rlp);
 541         };
 542 }
 543
 544 static void veth_handle_int(struct veth_lpevent *event)
 545 {
 546         HvLpIndex rlp = event->base_event.xSourceLp;
 547         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 548         unsigned long flags;
 549         int i, acked = 0;
 550
 551         BUG_ON(! cnx);
 552
 553         switch (event->base_event.xSubtype) {
 554         case VETH_EVENT_CAP:
 555                 veth_take_cap(cnx, event);
 556                 break;
 557         case VETH_EVENT_MONITOR:
 558                 /* do nothing... this'll hang out here til we're dead,
 559                  * and the hypervisor will return it for us. */
 560                 break;
 561         case VETH_EVENT_FRAMES_ACK:
 562                 spin_lock_irqsave(&cnx->lock, flags);
 563
 564                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 565                         u16 msgnum = event->u.frames_ack_data.token[i];
 566
 567                         if (msgnum < VETH_NUMBUFFERS) {
 568                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 569                                 cnx->outstanding_tx--;
 570                                 acked++;
 571                         }
 572                 }
 573
 574                 if (acked > 0) {
 575                         cnx->last_contact = jiffies;
 576                         veth_wake_queues(cnx);
 577                 }
 578
 579                 spin_unlock_irqrestore(&cnx->lock, flags);
 580                 break;
 581         case VETH_EVENT_FRAMES:
 582                 veth_receive(cnx, event);
 583                 break;
 584         default:
 585                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
 586                                 event->base_event.xSubtype, rlp);
 587         };
 588 }
 589
 590 static void veth_handle_event(struct HvLpEvent *event)
 591 {
 592         struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
 593
 594         if (hvlpevent_is_ack(event))
 595                 veth_handle_ack(veth_event);
 596         else
 597                 veth_handle_int(veth_event);
 598 }
 599
 600 static int veth_process_caps(struct veth_lpar_connection *cnx)
 601 {
 602         struct veth_cap_data *remote_caps = &cnx->remote_caps;
 603         int num_acks_needed;
 604
 605         /* Convert timer to jiffies */
 606         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 607
 608         if ( (remote_caps->num_buffers == 0) ||
 609              (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
 610              (remote_caps->ack_threshold == 0) ||
 611              (cnx->ack_timeout == 0) ) {
 612                 veth_error("Received incompatible capabilities from LPAR %d.\n",
 613                                 cnx->remote_lp);
 614                 return HvLpEvent_Rc_InvalidSubtypeData;
 615         }
 616
 617         num_acks_needed = (remote_caps->num_buffers
 618                            / remote_caps->ack_threshold) + 1;
 619
 620         /* FIXME: locking on num_ack_events? */
 621         if (cnx->num_ack_events < num_acks_needed) {
 622                 int num;
 623
 624                 num = veth_allocate_events(cnx->remote_lp,
 625                                            num_acks_needed-cnx->num_ack_events);
 626                 if (num > 0)
 627                         cnx->num_ack_events += num;
 628
 629                 if (cnx->num_ack_events < num_acks_needed) {
 630                         veth_error("Couldn't allocate enough ack events "
 631                                         "for LPAR %d.\n", cnx->remote_lp);
 632
 633                         return HvLpEvent_Rc_BufferNotAvailable;
 634                 }
 635         }
 636
 637
 638         return HvLpEvent_Rc_Good;
 639 }
 640
 641 /* FIXME: The gotos here are a bit dubious */
 642 static void veth_statemachine(struct work_struct *work)
 643 {
 644         struct veth_lpar_connection *cnx =
 645                 container_of(work, struct veth_lpar_connection,
 646                              statemachine_wq.work);
 647         int rlp = cnx->remote_lp;
 648         int rc;
 649
 650         spin_lock_irq(&cnx->lock);
 651
 652  restart:
 653         if (cnx->state & VETH_STATE_RESET) {
 654                 if (cnx->state & VETH_STATE_OPEN)
 655                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 656                                                      HvLpEvent_Type_VirtualLan);
 657
 658                 /*
 659                  * Reset ack data. This prevents the ack_timer actually
 660                  * doing anything, even if it runs one more time when
 661                  * we drop the lock below.
 662                  */
 663                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 664                 cnx->num_pending_acks = 0;
 665
 666                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 667                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 668                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 669                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 670
 671                 /* Clean up any leftover messages */
 672                 if (cnx->msgs) {
 673                         int i;
 674                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 675                                 veth_recycle_msg(cnx, cnx->msgs + i);
 676                 }
 677
 678                 cnx->outstanding_tx = 0;
 679                 veth_wake_queues(cnx);
 680
 681                 /* Drop the lock so we can do stuff that might sleep or
 682                  * take other locks. */
 683                 spin_unlock_irq(&cnx->lock);
 684
 685                 del_timer_sync(&cnx->ack_timer);
 686                 del_timer_sync(&cnx->reset_timer);
 687
 688                 spin_lock_irq(&cnx->lock);
 689
 690                 if (cnx->state & VETH_STATE_RESET)
 691                         goto restart;
 692
 693                 /* Hack, wait for the other end to reset itself. */
 694                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 695                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
 696                         goto out;
 697                 }
 698         }
 699
 700         if (cnx->state & VETH_STATE_SHUTDOWN)
 701                 /* It's all over, do nothing */
 702                 goto out;
 703
 704         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 705                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 706                         goto cant_cope;
 707
 708                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 709                 cnx->src_inst =
 710                         HvCallEvent_getSourceLpInstanceId(rlp,
 711                                                           HvLpEvent_Type_VirtualLan);
 712                 cnx->dst_inst =
 713                         HvCallEvent_getTargetLpInstanceId(rlp,
 714                                                           HvLpEvent_Type_VirtualLan);
 715                 cnx->state |= VETH_STATE_OPEN;
 716         }
 717
 718         if ( (cnx->state & VETH_STATE_OPEN) &&
 719              !(cnx->state & VETH_STATE_SENTMON) ) {
 720                 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
 721                                       HvLpEvent_AckInd_DoAck,
 722                                       HvLpEvent_AckType_DeferredAck,
 723                                       0, 0, 0, 0, 0, 0);
 724
 725                 if (rc == HvLpEvent_Rc_Good) {
 726                         cnx->state |= VETH_STATE_SENTMON;
 727                 } else {
 728                         if ( (rc != HvLpEvent_Rc_PartitionDead) &&
 729                              (rc != HvLpEvent_Rc_PathClosed) )
 730                                 veth_error("Error sending monitor to LPAR %d, "
 731                                                 "rc = %d\n", rlp, rc);
 732
 733                         /* Oh well, hope we get a cap from the other
 734                          * end and do better when that kicks us */
 735                         goto out;
 736                 }
 737         }
 738
 739         if ( (cnx->state & VETH_STATE_OPEN) &&
 740              !(cnx->state & VETH_STATE_SENTCAPS)) {
 741                 u64 *rawcap = (u64 *)&cnx->local_caps;
 742
 743                 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
 744                                       HvLpEvent_AckInd_DoAck,
 745                                       HvLpEvent_AckType_ImmediateAck,
 746                                       0, rawcap[0], rawcap[1], rawcap[2],
 747                                       rawcap[3], rawcap[4]);
 748
 749                 if (rc == HvLpEvent_Rc_Good) {
 750                         cnx->state |= VETH_STATE_SENTCAPS;
 751                 } else {
 752                         if ( (rc != HvLpEvent_Rc_PartitionDead) &&
 753                              (rc != HvLpEvent_Rc_PathClosed) )
 754                                 veth_error("Error sending caps to LPAR %d, "
 755                                                 "rc = %d\n", rlp, rc);
 756
 757                         /* Oh well, hope we get a cap from the other
 758                          * end and do better when that kicks us */
 759                         goto out;
 760                 }
 761         }
 762
 763         if ((cnx->state & VETH_STATE_GOTCAPS) &&
 764             !(cnx->state & VETH_STATE_SENTCAPACK)) {
 765                 struct veth_cap_data *remote_caps = &cnx->remote_caps;
 766
 767                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 768                        sizeof(*remote_caps));
 769
 770                 spin_unlock_irq(&cnx->lock);
 771                 rc = veth_process_caps(cnx);
 772                 spin_lock_irq(&cnx->lock);
 773
 774                 /* We dropped the lock, so recheck for anything which
 775                  * might mess us up */
 776                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 777                         goto restart;
 778
 779                 cnx->cap_event.base_event.xRc = rc;
 780                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 781                 if (rc == HvLpEvent_Rc_Good)
 782                         cnx->state |= VETH_STATE_SENTCAPACK;
 783                 else
 784                         goto cant_cope;
 785         }
 786
 787         if ((cnx->state & VETH_STATE_GOTCAPACK) &&
 788             (cnx->state & VETH_STATE_GOTCAPS) &&
 789             !(cnx->state & VETH_STATE_READY)) {
 790                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 791                         /* Start the ACK timer */
 792                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 793                         add_timer(&cnx->ack_timer);
 794                         cnx->state |= VETH_STATE_READY;
 795                 } else {
 796                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
 797                                         rlp, cnx->cap_ack_event.base_event.xRc);
 798                         goto cant_cope;
 799                 }
 800         }
 801
 802  out:
 803         spin_unlock_irq(&cnx->lock);
 804         return;
 805
 806  cant_cope:
 807         /* FIXME: we get here if something happens we really can't
 808          * cope with.  The link will never work once we get here, and
 809          * all we can do is not lock the rest of the system up */
 810         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
 811                         " (state = 0x%04lx)\n", rlp, cnx->state);
 812         cnx->state |= VETH_STATE_SHUTDOWN;
 813         spin_unlock_irq(&cnx->lock);
 814 }
 815
 816 static int veth_init_connection(u8 rlp)
 817 {
 818         struct veth_lpar_connection *cnx;
 819         struct veth_msg *msgs;
 820         int i;
 821
 822         if ( (rlp == this_lp) ||
 823              ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 824                 return 0;
 825
 826         cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
 827         if (! cnx)
 828                 return -ENOMEM;
 829
 830         cnx->remote_lp = rlp;
 831         spin_lock_init(&cnx->lock);
 832         INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
 833
 834         init_timer(&cnx->ack_timer);
 835         cnx->ack_timer.function = veth_timed_ack;
 836         cnx->ack_timer.data = (unsigned long) cnx;
 837
 838         init_timer(&cnx->reset_timer);
 839         cnx->reset_timer.function = veth_timed_reset;
 840         cnx->reset_timer.data = (unsigned long) cnx;
 841         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
 842
 843         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 844
 845         veth_cnx[rlp] = cnx;
 846
 847         /* This gets us 1 reference, which is held on behalf of the driver
 848          * infrastructure. It's released at module unload. */
 849         kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
 850
 851         msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
 852         if (! msgs) {
 853                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
 854                 return -ENOMEM;
 855         }
 856
 857         cnx->msgs = msgs;
 858
 859         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 860                 msgs[i].token = i;
 861                 veth_stack_push(cnx, msgs + i);
 862         }
 863
 864         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 865
 866         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 867                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
 868                 return -ENOMEM;
 869         }
 870
 871         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 872         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 873         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 874
 875         return 0;
 876 }
 877
 878 static void veth_stop_connection(struct veth_lpar_connection *cnx)
 879 {
 880         if (!cnx)
 881                 return;
 882
 883         spin_lock_irq(&cnx->lock);
 884         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 885         veth_kick_statemachine(cnx);
 886         spin_unlock_irq(&cnx->lock);
 887
 888         /* There's a slim chance the reset code has just queued the
 889          * statemachine to run in five seconds. If so we need to cancel
 890          * that and requeue the work to run now. */
 891         if (cancel_delayed_work(&cnx->statemachine_wq)) {
 892                 spin_lock_irq(&cnx->lock);
 893                 veth_kick_statemachine(cnx);
 894                 spin_unlock_irq(&cnx->lock);
 895         }
 896
 897         /* Wait for the state machine to run. */
 898         flush_scheduled_work();
 899 }
 900
 901 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
 902 {
 903         if (!cnx)
 904                 return;
 905
 906         if (cnx->num_events > 0)
 907                 mf_deallocate_lp_events(cnx->remote_lp,
 908                                       HvLpEvent_Type_VirtualLan,
 909                                       cnx->num_events,
 910                                       NULL, NULL);
 911         if (cnx->num_ack_events > 0)
 912                 mf_deallocate_lp_events(cnx->remote_lp,
 913                                       HvLpEvent_Type_VirtualLan,
 914                                       cnx->num_ack_events,
 915                                       NULL, NULL);
 916
 917         kfree(cnx->msgs);
 918         veth_cnx[cnx->remote_lp] = NULL;
 919         kfree(cnx);
 920 }
 921
 922 static void veth_release_connection(struct kobject *kobj)
 923 {
 924         struct veth_lpar_connection *cnx;
 925         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 926         veth_stop_connection(cnx);
 927         veth_destroy_connection(cnx);
 928 }
 929
 930 /*
 931  * net_device code
 932  */
 933
 934 static int veth_open(struct net_device *dev)
 935 {
 936         netif_start_queue(dev);
 937         return 0;
 938 }
 939
 940 static int veth_close(struct net_device *dev)
 941 {
 942         netif_stop_queue(dev);
 943         return 0;
 944 }
 945
 946 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 947 {
 948         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 949                 return -EINVAL;
 950         dev->mtu = new_mtu;
 951         return 0;
 952 }
 953
 954 static void veth_set_multicast_list(struct net_device *dev)
 955 {
 956         struct veth_port *port = netdev_priv(dev);
 957         unsigned long flags;
 958
 959         write_lock_irqsave(&port->mcast_gate, flags);
 960
 961         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
 962                         (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
 963                 port->promiscuous = 1;
 964         } else {
 965                 struct netdev_hw_addr *ha;
 966
 967                 port->promiscuous = 0;
 968
 969                 /* Update table */
 970                 port->num_mcast = 0;
 971
 972                 netdev_for_each_mc_addr(ha, dev) {
 973                         u8 *addr = ha->addr;
 974                         u64 xaddr = 0;
 975
 976                         if (addr[0] & 0x01) {/* multicast address? */
 977                                 memcpy(&xaddr, addr, ETH_ALEN);
 978                                 port->mcast_addr[port->num_mcast] = xaddr;
 979                                 port->num_mcast++;
 980                         }
 981                 }
 982         }
 983
 984         write_unlock_irqrestore(&port->mcast_gate, flags);
 985 }
 986
 987 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 988 {
 989         strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
 990         info->driver[sizeof(info->driver) - 1] = '\0';
 991         strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
 992         info->version[sizeof(info->version) - 1] = '\0';
 993 }
 994
 995 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
 996 {
 997         ecmd->supported = (SUPPORTED_1000baseT_Full
 998                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 999         ecmd->advertising = (SUPPORTED_1000baseT_Full
1000                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1001         ecmd->port = PORT_FIBRE;
1002         ecmd->transceiver = XCVR_INTERNAL;
1003         ecmd->phy_address = 0;
1004         ecmd->speed = SPEED_1000;
1005         ecmd->duplex = DUPLEX_FULL;
1006         ecmd->autoneg = AUTONEG_ENABLE;
1007         ecmd->maxtxpkt = 120;
1008         ecmd->maxrxpkt = 120;
1009         return 0;
1010 }
1011
1012 static u32 veth_get_link(struct net_device *dev)
1013 {
1014         return 1;
1015 }
1016
1017 static const struct ethtool_ops ops = {
1018         .get_drvinfo = veth_get_drvinfo,
1019         .get_settings = veth_get_settings,
1020         .get_link = veth_get_link,
1021 };
1022
1023 static const struct net_device_ops veth_netdev_ops = {
1024         .ndo_open               = veth_open,
1025         .ndo_stop               = veth_close,
1026         .ndo_start_xmit         = veth_start_xmit,
1027         .ndo_change_mtu         = veth_change_mtu,
1028         .ndo_set_multicast_list = veth_set_multicast_list,
1029         .ndo_set_mac_address    = NULL,
1030         .ndo_validate_addr      = eth_validate_addr,
1031 };
1032
1033 static struct net_device *veth_probe_one(int vlan,
1034                 struct vio_dev *vio_dev)
1035 {
1036         struct net_device *dev;
1037         struct veth_port *port;
1038         struct device *vdev = &vio_dev->dev;
1039         int i, rc;
1040         const unsigned char *mac_addr;
1041
1042         mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1043         if (mac_addr == NULL)
1044                 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1045         if (mac_addr == NULL) {
1046                 veth_error("Unable to fetch MAC address from device tree.\n");
1047                 return NULL;
1048         }
1049
1050         dev = alloc_etherdev(sizeof (struct veth_port));
1051         if (! dev) {
1052                 veth_error("Unable to allocate net_device structure!\n");
1053                 return NULL;
1054         }
1055
1056         port = netdev_priv(dev);
1057
1058         spin_lock_init(&port->queue_lock);
1059         rwlock_init(&port->mcast_gate);
1060         port->stopped_map = 0;
1061
1062         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1063                 HvLpVirtualLanIndexMap map;
1064
1065                 if (i == this_lp)
1066                         continue;
1067                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1068                 if (map & (0x8000 >> vlan))
1069                         port->lpar_map |= (1 << i);
1070         }
1071         port->dev = vdev;
1072
1073         memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1074
1075         dev->mtu = VETH_MAX_MTU;
1076
1077         memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1078
1079         dev->netdev_ops = &veth_netdev_ops;
1080         SET_ETHTOOL_OPS(dev, &ops);
1081
1082         SET_NETDEV_DEV(dev, vdev);
1083
1084         rc = register_netdev(dev);
1085         if (rc != 0) {
1086                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1087                 free_netdev(dev);
1088                 return NULL;
1089         }
1090
1091         kobject_init(&port->kobject, &veth_port_ktype);
1092         if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
1093                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1094
1095         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1096                         dev->name, vlan, port->lpar_map);
1097
1098         return dev;
1099 }
1100
1101 /*
1102  * Tx path
1103  */
1104
1105 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1106                                 struct net_device *dev)
1107 {
1108         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1109         struct veth_port *port = netdev_priv(dev);
1110         HvLpEvent_Rc rc;
1111         struct veth_msg *msg = NULL;
1112         unsigned long flags;
1113
1114         if (! cnx)
1115                 return 0;
1116
1117         spin_lock_irqsave(&cnx->lock, flags);
1118
1119         if (! (cnx->state & VETH_STATE_READY))
1120                 goto no_error;
1121
1122         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1123                 goto drop;
1124
1125         msg = veth_stack_pop(cnx);
1126         if (! msg)
1127                 goto drop;
1128
1129         msg->in_use = 1;
1130         msg->skb = skb_get(skb);
1131
1132         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1133                                 skb->len, DMA_TO_DEVICE);
1134
1135         if (dma_mapping_error(port->dev, msg->data.addr[0]))
1136                 goto recycle_and_drop;
1137
1138         msg->dev = port->dev;
1139         msg->data.len[0] = skb->len;
1140         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1141
1142         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1143
1144         if (rc != HvLpEvent_Rc_Good)
1145                 goto recycle_and_drop;
1146
1147         /* If the timer's not already running, start it now. */
1148         if (0 == cnx->outstanding_tx)
1149                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1150
1151         cnx->last_contact = jiffies;
1152         cnx->outstanding_tx++;
1153
1154         if (veth_stack_is_empty(cnx))
1155                 veth_stop_queues(cnx);
1156
1157  no_error:
1158         spin_unlock_irqrestore(&cnx->lock, flags);
1159         return 0;
1160
1161  recycle_and_drop:
1162         veth_recycle_msg(cnx, msg);
1163  drop:
1164         spin_unlock_irqrestore(&cnx->lock, flags);
1165         return 1;
1166 }
1167
1168 static void veth_transmit_to_many(struct sk_buff *skb,
1169                                           HvLpIndexMap lpmask,
1170                                           struct net_device *dev)
1171 {
1172         int i, success, error;
1173
1174         success = error = 0;
1175
1176         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1177                 if ((lpmask & (1 << i)) == 0)
1178                         continue;
1179
1180                 if (veth_transmit_to_one(skb, i, dev))
1181                         error = 1;
1182                 else
1183                         success = 1;
1184         }
1185
1186         if (error)
1187                 dev->stats.tx_errors++;
1188
1189         if (success) {
1190                 dev->stats.tx_packets++;
1191                 dev->stats.tx_bytes += skb->len;
1192         }
1193 }
1194
1195 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1196 {
1197         unsigned char *frame = skb->data;
1198         struct veth_port *port = netdev_priv(dev);
1199         HvLpIndexMap lpmask;
1200
1201         if (! (frame[0] & 0x01)) {
1202                 /* unicast packet */
1203                 HvLpIndex rlp = frame[5];
1204
1205                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1206                         dev_kfree_skb(skb);
1207                         return NETDEV_TX_OK;
1208                 }
1209
1210                 lpmask = 1 << rlp;
1211         } else {
1212                 lpmask = port->lpar_map;
1213         }
1214
1215         veth_transmit_to_many(skb, lpmask, dev);
1216
1217         dev_kfree_skb(skb);
1218
1219         return NETDEV_TX_OK;
1220 }
1221
1222 /* You must hold the connection's lock when you call this function. */
1223 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1224                              struct veth_msg *msg)
1225 {
1226         u32 dma_address, dma_length;
1227
1228         if (msg->in_use) {
1229                 msg->in_use = 0;
1230                 dma_address = msg->data.addr[0];
1231                 dma_length = msg->data.len[0];
1232
1233                 if (!dma_mapping_error(msg->dev, dma_address))
1234                         dma_unmap_single(msg->dev, dma_address, dma_length,
1235                                         DMA_TO_DEVICE);
1236
1237                 if (msg->skb) {
1238                         dev_kfree_skb_any(msg->skb);
1239                         msg->skb = NULL;
1240                 }
1241
1242                 memset(&msg->data, 0, sizeof(msg->data));
1243                 veth_stack_push(cnx, msg);
1244         } else if (cnx->state & VETH_STATE_OPEN) {
1245                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1246                                 cnx->remote_lp, msg->token);
1247         }
1248 }
1249
1250 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1251 {
1252         int i;
1253
1254         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1255                 struct net_device *dev = veth_dev[i];
1256                 struct veth_port *port;
1257                 unsigned long flags;
1258
1259                 if (! dev)
1260                         continue;
1261
1262                 port = netdev_priv(dev);
1263
1264                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1265                         continue;
1266
1267                 spin_lock_irqsave(&port->queue_lock, flags);
1268
1269                 port->stopped_map &= ~(1 << cnx->remote_lp);
1270
1271                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1272                         veth_debug("cnx %d: woke queue for %s.\n",
1273                                         cnx->remote_lp, dev->name);
1274                         netif_wake_queue(dev);
1275                 }
1276                 spin_unlock_irqrestore(&port->queue_lock, flags);
1277         }
1278 }
1279
1280 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1281 {
1282         int i;
1283
1284         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1285                 struct net_device *dev = veth_dev[i];
1286                 struct veth_port *port;
1287
1288                 if (! dev)
1289                         continue;
1290
1291                 port = netdev_priv(dev);
1292
1293                 /* If this cnx is not on the vlan for this port, continue */
1294                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1295                         continue;
1296
1297                 spin_lock(&port->queue_lock);
1298
1299                 netif_stop_queue(dev);
1300                 port->stopped_map |= (1 << cnx->remote_lp);
1301
1302                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1303                                 cnx->remote_lp, dev->name, port->stopped_map);
1304
1305                 spin_unlock(&port->queue_lock);
1306         }
1307 }
1308
1309 static void veth_timed_reset(unsigned long ptr)
1310 {
1311         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1312         unsigned long trigger_time, flags;
1313
1314         /* FIXME is it possible this fires after veth_stop_connection()?
1315          * That would reschedule the statemachine for 5 seconds and probably
1316          * execute it after the module's been unloaded. Hmm. */
1317
1318         spin_lock_irqsave(&cnx->lock, flags);
1319
1320         if (cnx->outstanding_tx > 0) {
1321                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1322
1323                 if (trigger_time < jiffies) {
1324                         cnx->state |= VETH_STATE_RESET;
1325                         veth_kick_statemachine(cnx);
1326                         veth_error("%d packets not acked by LPAR %d within %d "
1327                                         "seconds, resetting.\n",
1328                                         cnx->outstanding_tx, cnx->remote_lp,
1329                                         cnx->reset_timeout / HZ);
1330                 } else {
1331                         /* Reschedule the timer */
1332                         trigger_time = jiffies + cnx->reset_timeout;
1333                         mod_timer(&cnx->reset_timer, trigger_time);
1334                 }
1335         }
1336
1337         spin_unlock_irqrestore(&cnx->lock, flags);
1338 }
1339
1340 /*
1341  * Rx path
1342  */
1343
1344 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1345 {
1346         int wanted = 0;
1347         int i;
1348         unsigned long flags;
1349
1350         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1351                 return 1;
1352
1353         read_lock_irqsave(&port->mcast_gate, flags);
1354
1355         if (port->promiscuous) {
1356                 wanted = 1;
1357                 goto out;
1358         }
1359
1360         for (i = 0; i < port->num_mcast; ++i) {
1361                 if (port->mcast_addr[i] == mac_addr) {
1362                         wanted = 1;
1363                         break;
1364                 }
1365         }
1366
1367  out:
1368         read_unlock_irqrestore(&port->mcast_gate, flags);
1369
1370         return wanted;
1371 }
1372
1373 struct dma_chunk {
1374         u64 addr;
1375         u64 size;
1376 };
1377
1378 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1379
1380 static inline void veth_build_dma_list(struct dma_chunk *list,
1381                                        unsigned char *p, unsigned long length)
1382 {
1383         unsigned long done;
1384         int i = 1;
1385
1386         /* FIXME: skbs are contiguous in real addresses.  Do we
1387          * really need to break it into PAGE_SIZE chunks, or can we do
1388          * it just at the granularity of iSeries real->absolute
1389          * mapping?  Indeed, given the way the allocator works, can we
1390          * count on them being absolutely contiguous? */
1391         list[0].addr = iseries_hv_addr(p);
1392         list[0].size = min(length,
1393                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1394
1395         done = list[0].size;
1396         while (done < length) {
1397                 list[i].addr = iseries_hv_addr(p + done);
1398                 list[i].size = min(length-done, PAGE_SIZE);
1399                 done += list[i].size;
1400                 i++;
1401         }
1402 }
1403
1404 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1405 {
1406         HvLpEvent_Rc rc;
1407
1408         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1409                              0, &cnx->pending_acks);
1410
1411         if (rc != HvLpEvent_Rc_Good)
1412                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1413                                 cnx->remote_lp, (int)rc);
1414
1415         cnx->num_pending_acks = 0;
1416         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1417 }
1418
1419 static void veth_receive(struct veth_lpar_connection *cnx,
1420                          struct veth_lpevent *event)
1421 {
1422         struct veth_frames_data *senddata = &event->u.frames_data;
1423         int startchunk = 0;
1424         int nchunks;
1425         unsigned long flags;
1426         HvLpDma_Rc rc;
1427
1428         do {
1429                 u16 length = 0;
1430                 struct sk_buff *skb;
1431                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1432                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1433                 u64 dest;
1434                 HvLpVirtualLanIndex vlan;
1435                 struct net_device *dev;
1436                 struct veth_port *port;
1437
1438                 /* FIXME: do we need this? */
1439                 memset(local_list, 0, sizeof(local_list));
1440                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1441
1442                 /* a 0 address marks the end of the valid entries */
1443                 if (senddata->addr[startchunk] == 0)
1444                         break;
1445
1446                 /* make sure that we have at least 1 EOF entry in the
1447                  * remaining entries */
1448                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1449                         veth_error("Missing EOF fragment in event "
1450                                         "eofmask = 0x%x startchunk = %d\n",
1451                                         (unsigned)senddata->eofmask,
1452                                         startchunk);
1453                         break;
1454                 }
1455
1456                 /* build list of chunks in this frame */
1457                 nchunks = 0;
1458                 do {
1459                         remote_list[nchunks].addr =
1460                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1461                         remote_list[nchunks].size =
1462                                 senddata->len[startchunk+nchunks];
1463                         length += remote_list[nchunks].size;
1464                 } while (! (senddata->eofmask &
1465                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1466
1467                 /* length == total length of all chunks */
1468                 /* nchunks == # of chunks in this frame */
1469
1470                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1471                         veth_error("Received oversize frame from LPAR %d "
1472                                         "(length = %d)\n",
1473                                         cnx->remote_lp, length);
1474                         continue;
1475                 }
1476
1477                 skb = alloc_skb(length, GFP_ATOMIC);
1478                 if (!skb)
1479                         continue;
1480
1481                 veth_build_dma_list(local_list, skb->data, length);
1482
1483                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1484                                             event->base_event.xSourceLp,
1485                                             HvLpDma_Direction_RemoteToLocal,
1486                                             cnx->src_inst,
1487                                             cnx->dst_inst,
1488                                             HvLpDma_AddressType_RealAddress,
1489                                             HvLpDma_AddressType_TceIndex,
1490                                             iseries_hv_addr(&local_list),
1491                                             iseries_hv_addr(&remote_list),
1492                                             length);
1493                 if (rc != HvLpDma_Rc_Good) {
1494                         dev_kfree_skb_irq(skb);
1495                         continue;
1496                 }
1497
1498                 vlan = skb->data[9];
1499                 dev = veth_dev[vlan];
1500                 if (! dev) {
1501                         /*
1502                          * Some earlier versions of the driver sent
1503                          * broadcasts down all connections, even to lpars
1504                          * that weren't on the relevant vlan. So ignore
1505                          * packets belonging to a vlan we're not on.
1506                          * We can also be here if we receive packets while
1507                          * the driver is going down, because then dev is NULL.
1508                          */
1509                         dev_kfree_skb_irq(skb);
1510                         continue;
1511                 }
1512
1513                 port = netdev_priv(dev);
1514                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1515
1516                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1517                         dev_kfree_skb_irq(skb);
1518                         continue;
1519                 }
1520                 if (! veth_frame_wanted(port, dest)) {
1521                         dev_kfree_skb_irq(skb);
1522                         continue;
1523                 }
1524
1525                 skb_put(skb, length);
1526                 skb->protocol = eth_type_trans(skb, dev);
1527                 skb->ip_summed = CHECKSUM_NONE;
1528                 netif_rx(skb);  /* send it up */
1529                 dev->stats.rx_packets++;
1530                 dev->stats.rx_bytes += length;
1531         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1532
1533         /* Ack it */
1534         spin_lock_irqsave(&cnx->lock, flags);
1535         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1536
1537         cnx->pending_acks[cnx->num_pending_acks++] =
1538                 event->base_event.xCorrelationToken;
1539
1540         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1541              (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1542                 veth_flush_acks(cnx);
1543
1544         spin_unlock_irqrestore(&cnx->lock, flags);
1545 }
1546
1547 static void veth_timed_ack(unsigned long ptr)
1548 {
1549         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1550         unsigned long flags;
1551
1552         /* Ack all the events */
1553         spin_lock_irqsave(&cnx->lock, flags);
1554         if (cnx->num_pending_acks > 0)
1555                 veth_flush_acks(cnx);
1556
1557         /* Reschedule the timer */
1558         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1559         add_timer(&cnx->ack_timer);
1560         spin_unlock_irqrestore(&cnx->lock, flags);
1561 }
1562
1563 static int veth_remove(struct vio_dev *vdev)
1564 {
1565         struct veth_lpar_connection *cnx;
1566         struct net_device *dev;
1567         struct veth_port *port;
1568         int i;
1569
1570         dev = veth_dev[vdev->unit_address];
1571
1572         if (! dev)
1573                 return 0;
1574
1575         port = netdev_priv(dev);
1576
1577         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1578                 cnx = veth_cnx[i];
1579
1580                 if (cnx && (port->lpar_map & (1 << i))) {
1581                         /* Drop our reference to connections on our VLAN */
1582                         kobject_put(&cnx->kobject);
1583                 }
1584         }
1585
1586         veth_dev[vdev->unit_address] = NULL;
1587         kobject_del(&port->kobject);
1588         kobject_put(&port->kobject);
1589         unregister_netdev(dev);
1590         free_netdev(dev);
1591
1592         return 0;
1593 }
1594
1595 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1596 {
1597         int i = vdev->unit_address;
1598         struct net_device *dev;
1599         struct veth_port *port;
1600
1601         dev = veth_probe_one(i, vdev);
1602         if (dev == NULL) {
1603                 veth_remove(vdev);
1604                 return 1;
1605         }
1606         veth_dev[i] = dev;
1607
1608         port = (struct veth_port*)netdev_priv(dev);
1609
1610         /* Start the state machine on each connection on this vlan. If we're
1611          * the first dev to do so this will commence link negotiation */
1612         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1613                 struct veth_lpar_connection *cnx;
1614
1615                 if (! (port->lpar_map & (1 << i)))
1616                         continue;
1617
1618                 cnx = veth_cnx[i];
1619                 if (!cnx)
1620                         continue;
1621
1622                 kobject_get(&cnx->kobject);
1623                 veth_kick_statemachine(cnx);
1624         }
1625
1626         return 0;
1627 }
1628
1629 /**
1630  * veth_device_table: Used by vio.c to match devices that we
1631  * support.
1632  */
1633 static struct vio_device_id veth_device_table[] __devinitdata = {
1634         { "network", "IBM,iSeries-l-lan" },
1635         { "", "" }
1636 };
1637 MODULE_DEVICE_TABLE(vio, veth_device_table);
1638
1639 static struct vio_driver veth_driver = {
1640         .id_table = veth_device_table,
1641         .probe = veth_probe,
1642         .remove = veth_remove,
1643         .driver = {
1644                 .name = DRV_NAME,
1645                 .owner = THIS_MODULE,
1646         }
1647 };
1648
1649 /*
1650  * Module initialization/cleanup
1651  */
1652
1653 static void __exit veth_module_cleanup(void)
1654 {
1655         int i;
1656         struct veth_lpar_connection *cnx;
1657
1658         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1659         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1660
1661         /* Make sure any work queued from Hypervisor callbacks is finished. */
1662         flush_scheduled_work();
1663
1664         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1665                 cnx = veth_cnx[i];
1666
1667                 if (!cnx)
1668                         continue;
1669
1670                 /* Remove the connection from sysfs */
1671                 kobject_del(&cnx->kobject);
1672                 /* Drop the driver's reference to the connection */
1673                 kobject_put(&cnx->kobject);
1674         }
1675
1676         /* Unregister the driver, which will close all the netdevs and stop
1677          * the connections when they're no longer referenced. */
1678         vio_unregister_driver(&veth_driver);
1679 }
1680 module_exit(veth_module_cleanup);
1681
1682 static int __init veth_module_init(void)
1683 {
1684         int i;
1685         int rc;
1686
1687         if (!firmware_has_feature(FW_FEATURE_ISERIES))
1688                 return -ENODEV;
1689
1690         this_lp = HvLpConfig_getLpIndex_outline();
1691
1692         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1693                 rc = veth_init_connection(i);
1694                 if (rc != 0)
1695                         goto error;
1696         }
1697
1698         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1699                                   &veth_handle_event);
1700
1701         rc = vio_register_driver(&veth_driver);
1702         if (rc != 0)
1703                 goto error;
1704
1705         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1706                 struct kobject *kobj;
1707
1708                 if (!veth_cnx[i])
1709                         continue;
1710
1711                 kobj = &veth_cnx[i]->kobject;
1712                 /* If the add failes, complain but otherwise continue */
1713                 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1714                                         "cnx%.2d", veth_cnx[i]->remote_lp))
1715                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1716         }
1717
1718         return 0;
1719
1720 error:
1721         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1722                 veth_destroy_connection(veth_cnx[i]);
1723         }
1724
1725         return rc;
1726 }
1727 module_init(veth_module_init);