drivers/net/iseries_veth.c

   1 /* File veth.c created by Kyle A. Lucke on Mon Aug  7 2000. */
   2 /*
   3  * IBM eServer iSeries Virtual Ethernet Device Driver
   4  * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
   5  * Substantially cleaned up by:
   6  * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License as
  10  * published by the Free Software Foundation; either version 2 of the
  11  * License, or (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
  21  * USA
  22  *
  23  *
  24  * This module implements the virtual ethernet device for iSeries LPAR
  25  * Linux.  It uses hypervisor message passing to implement an
  26  * ethernet-like network device communicating between partitions on
  27  * the iSeries.
  28  *
  29  * The iSeries LPAR hypervisor currently allows for up to 16 different
  30  * virtual ethernets.  These are all dynamically configurable on
  31  * OS/400 partitions, but dynamic configuration is not supported under
  32  * Linux yet.  An ethXX network device will be created for each
  33  * virtual ethernet this partition is connected to.
  34  *
  35  * - This driver is responsible for routing packets to and from other
  36  *   partitions.  The MAC addresses used by the virtual ethernets
  37  *   contains meaning and must not be modified.
  38  *
  39  * - Having 2 virtual ethernets to the same remote partition DOES NOT
  40  *   double the available bandwidth.  The 2 devices will share the
  41  *   available hypervisor bandwidth.
  42  *
  43  * - If you send a packet to your own mac address, it will just be
  44  *   dropped, you won't get it on the receive side.
  45  *
  46  * - Multicast is implemented by sending the frame frame to every
  47  *   other partition.  It is the responsibility of the receiving
  48  *   partition to filter the addresses desired.
  49  *
  50  * Tunable parameters:
  51  *
  52  * VETH_NUMBUFFERS: This compile time option defaults to 120.  It
  53  * controls how much memory Linux will allocate per remote partition
  54  * it is communicating with.  It can be thought of as the maximum
  55  * number of packets outstanding to a remote partition at a time.
  56  */
  57
  58 #include <linux/config.h>
  59 #include <linux/module.h>
  60 #include <linux/version.h>
  61 #include <linux/types.h>
  62 #include <linux/errno.h>
  63 #include <linux/ioport.h>
  64 #include <linux/kernel.h>
  65 #include <linux/netdevice.h>
  66 #include <linux/etherdevice.h>
  67 #include <linux/skbuff.h>
  68 #include <linux/init.h>
  69 #include <linux/delay.h>
  70 #include <linux/mm.h>
  71 #include <linux/ethtool.h>
  72 #include <asm/iSeries/mf.h>
  73 #include <asm/iSeries/iSeries_pci.h>
  74 #include <asm/uaccess.h>
  75
  76 #include <asm/iSeries/HvLpConfig.h>
  77 #include <asm/iSeries/HvTypes.h>
  78 #include <asm/iSeries/HvLpEvent.h>
  79 #include <asm/iommu.h>
  80 #include <asm/vio.h>
  81
  82 #undef DEBUG
  83
  84 #include "iseries_veth.h"
  85
  86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
  87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
  88 MODULE_LICENSE("GPL");
  89
  90 #define VETH_NUMBUFFERS         (120)
  91 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
  92 #define VETH_MAX_MCAST          (12)
  93
  94 #define VETH_MAX_MTU            (9000)
  95
  96 #if VETH_NUMBUFFERS < 10
  97 #define ACK_THRESHOLD           (1)
  98 #elif VETH_NUMBUFFERS < 20
  99 #define ACK_THRESHOLD           (4)
 100 #elif VETH_NUMBUFFERS < 40
 101 #define ACK_THRESHOLD           (10)
 102 #else
 103 #define ACK_THRESHOLD           (20)
 104 #endif
 105
 106 #define VETH_STATE_SHUTDOWN     (0x0001)
 107 #define VETH_STATE_OPEN         (0x0002)
 108 #define VETH_STATE_RESET        (0x0004)
 109 #define VETH_STATE_SENTMON      (0x0008)
 110 #define VETH_STATE_SENTCAPS     (0x0010)
 111 #define VETH_STATE_GOTCAPACK    (0x0020)
 112 #define VETH_STATE_GOTCAPS      (0x0040)
 113 #define VETH_STATE_SENTCAPACK   (0x0080)
 114 #define VETH_STATE_READY        (0x0100)
 115
 116 struct veth_msg {
 117         struct veth_msg *next;
 118         struct VethFramesData data;
 119         int token;
 120         int in_use;
 121         struct sk_buff *skb;
 122         struct device *dev;
 123 };
 124
 125 struct veth_lpar_connection {
 126         HvLpIndex remote_lp;
 127         struct work_struct statemachine_wq;
 128         struct veth_msg *msgs;
 129         int num_events;
 130         struct VethCapData local_caps;
 131
 132         struct kobject kobject;
 133         struct timer_list ack_timer;
 134
 135         struct timer_list reset_timer;
 136         unsigned int reset_timeout;
 137         unsigned long last_contact;
 138         int outstanding_tx;
 139
 140         spinlock_t lock;
 141         unsigned long state;
 142         HvLpInstanceId src_inst;
 143         HvLpInstanceId dst_inst;
 144         struct VethLpEvent cap_event, cap_ack_event;
 145         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
 146         u32 num_pending_acks;
 147
 148         int num_ack_events;
 149         struct VethCapData remote_caps;
 150         u32 ack_timeout;
 151
 152         struct veth_msg *msg_stack_head;
 153 };
 154
 155 struct veth_port {
 156         struct device *dev;
 157         struct net_device_stats stats;
 158         u64 mac_addr;
 159         HvLpIndexMap lpar_map;
 160
 161         /* queue_lock protects the stopped_map and dev's queue. */
 162         spinlock_t queue_lock;
 163         HvLpIndexMap stopped_map;
 164
 165         /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
 166         rwlock_t mcast_gate;
 167         int promiscuous;
 168         int num_mcast;
 169         u64 mcast_addr[VETH_MAX_MCAST];
 170
 171         struct kobject kobject;
 172 };
 173
 174 static HvLpIndex this_lp;
 175 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
 176 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
 177
 178 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
 179 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
 180 static void veth_wake_queues(struct veth_lpar_connection *cnx);
 181 static void veth_stop_queues(struct veth_lpar_connection *cnx);
 182 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
 183 static void veth_release_connection(struct kobject *kobject);
 184 static void veth_timed_ack(unsigned long ptr);
 185 static void veth_timed_reset(unsigned long ptr);
 186
 187 /*
 188  * Utility functions
 189  */
 190
 191 #define veth_info(fmt, args...) \
 192         printk(KERN_INFO "iseries_veth: " fmt, ## args)
 193
 194 #define veth_error(fmt, args...) \
 195         printk(KERN_ERR "iseries_veth: Error: " fmt, ## args)
 196
 197 #ifdef DEBUG
 198 #define veth_debug(fmt, args...) \
 199         printk(KERN_DEBUG "iseries_veth: " fmt, ## args)
 200 #else
 201 #define veth_debug(fmt, args...) do {} while (0)
 202 #endif
 203
 204 /* You must hold the connection's lock when you call this function. */
 205 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
 206                                    struct veth_msg *msg)
 207 {
 208         msg->next = cnx->msg_stack_head;
 209         cnx->msg_stack_head = msg;
 210 }
 211
 212 /* You must hold the connection's lock when you call this function. */
 213 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
 214 {
 215         struct veth_msg *msg;
 216
 217         msg = cnx->msg_stack_head;
 218         if (msg)
 219                 cnx->msg_stack_head = cnx->msg_stack_head->next;
 220
 221         return msg;
 222 }
 223
 224 /* You must hold the connection's lock when you call this function. */
 225 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
 226 {
 227         return cnx->msg_stack_head == NULL;
 228 }
 229
 230 static inline HvLpEvent_Rc
 231 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
 232                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
 233                  u64 token,
 234                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
 235 {
 236         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
 237                                              HvLpEvent_Type_VirtualLan,
 238                                              subtype, ackind, acktype,
 239                                              cnx->src_inst,
 240                                              cnx->dst_inst,
 241                                              token, data1, data2, data3,
 242                                              data4, data5);
 243 }
 244
 245 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
 246                                            u16 subtype, u64 token, void *data)
 247 {
 248         u64 *p = (u64 *) data;
 249
 250         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
 251                                 HvLpEvent_AckType_ImmediateAck,
 252                                 token, p[0], p[1], p[2], p[3], p[4]);
 253 }
 254
 255 struct veth_allocation {
 256         struct completion c;
 257         int num;
 258 };
 259
 260 static void veth_complete_allocation(void *parm, int number)
 261 {
 262         struct veth_allocation *vc = (struct veth_allocation *)parm;
 263
 264         vc->num = number;
 265         complete(&vc->c);
 266 }
 267
 268 static int veth_allocate_events(HvLpIndex rlp, int number)
 269 {
 270         struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
 271
 272         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
 273                             sizeof(struct VethLpEvent), number,
 274                             &veth_complete_allocation, &vc);
 275         wait_for_completion(&vc.c);
 276
 277         return vc.num;
 278 }
 279
 280 /*
 281  * sysfs support
 282  */
 283
 284 struct veth_cnx_attribute {
 285         struct attribute attr;
 286         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
 287         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
 288 };
 289
 290 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
 291                 struct attribute *attr, char *buf)
 292 {
 293         struct veth_cnx_attribute *cnx_attr;
 294         struct veth_lpar_connection *cnx;
 295
 296         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
 297         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 298
 299         if (!cnx_attr->show)
 300                 return -EIO;
 301
 302         return cnx_attr->show(cnx, buf);
 303 }
 304
 305 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
 306 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
 307 {                                                                       \
 308         return sprintf(buf, _format, _expression);                      \
 309 }                                                                       \
 310 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
 311
 312 #define SIMPLE_CNX_ATTR(_name)  \
 313         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
 314
 315 SIMPLE_CNX_ATTR(outstanding_tx);
 316 SIMPLE_CNX_ATTR(remote_lp);
 317 SIMPLE_CNX_ATTR(num_events);
 318 SIMPLE_CNX_ATTR(src_inst);
 319 SIMPLE_CNX_ATTR(dst_inst);
 320 SIMPLE_CNX_ATTR(num_pending_acks);
 321 SIMPLE_CNX_ATTR(num_ack_events);
 322 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
 323 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
 324 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
 325 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
 326                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
 327
 328 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
 329
 330 static struct attribute *veth_cnx_default_attrs[] = {
 331         GET_CNX_ATTR(outstanding_tx),
 332         GET_CNX_ATTR(remote_lp),
 333         GET_CNX_ATTR(num_events),
 334         GET_CNX_ATTR(reset_timeout),
 335         GET_CNX_ATTR(last_contact),
 336         GET_CNX_ATTR(state),
 337         GET_CNX_ATTR(src_inst),
 338         GET_CNX_ATTR(dst_inst),
 339         GET_CNX_ATTR(num_pending_acks),
 340         GET_CNX_ATTR(num_ack_events),
 341         GET_CNX_ATTR(ack_timeout),
 342         NULL
 343 };
 344
 345 static struct sysfs_ops veth_cnx_sysfs_ops = {
 346                 .show = veth_cnx_attribute_show
 347 };
 348
 349 static struct kobj_type veth_lpar_connection_ktype = {
 350         .release        = veth_release_connection,
 351         .sysfs_ops      = &veth_cnx_sysfs_ops,
 352         .default_attrs  = veth_cnx_default_attrs
 353 };
 354
 355 struct veth_port_attribute {
 356         struct attribute attr;
 357         ssize_t (*show)(struct veth_port *, char *buf);
 358         ssize_t (*store)(struct veth_port *, const char *buf);
 359 };
 360
 361 static ssize_t veth_port_attribute_show(struct kobject *kobj,
 362                 struct attribute *attr, char *buf)
 363 {
 364         struct veth_port_attribute *port_attr;
 365         struct veth_port *port;
 366
 367         port_attr = container_of(attr, struct veth_port_attribute, attr);
 368         port = container_of(kobj, struct veth_port, kobject);
 369
 370         if (!port_attr->show)
 371                 return -EIO;
 372
 373         return port_attr->show(port, buf);
 374 }
 375
 376 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
 377 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
 378 {                                                                       \
 379         return sprintf(buf, _format, _expression);                      \
 380 }                                                                       \
 381 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
 382
 383 #define SIMPLE_PORT_ATTR(_name) \
 384         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
 385
 386 SIMPLE_PORT_ATTR(promiscuous);
 387 SIMPLE_PORT_ATTR(num_mcast);
 388 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
 389 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
 390 CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
 391
 392 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
 393 static struct attribute *veth_port_default_attrs[] = {
 394         GET_PORT_ATTR(mac_addr),
 395         GET_PORT_ATTR(lpar_map),
 396         GET_PORT_ATTR(stopped_map),
 397         GET_PORT_ATTR(promiscuous),
 398         GET_PORT_ATTR(num_mcast),
 399         NULL
 400 };
 401
 402 static struct sysfs_ops veth_port_sysfs_ops = {
 403         .show = veth_port_attribute_show
 404 };
 405
 406 static struct kobj_type veth_port_ktype = {
 407         .sysfs_ops      = &veth_port_sysfs_ops,
 408         .default_attrs  = veth_port_default_attrs
 409 };
 410
 411 /*
 412  * LPAR connection code
 413  */
 414
 415 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
 416 {
 417         schedule_work(&cnx->statemachine_wq);
 418 }
 419
 420 static void veth_take_cap(struct veth_lpar_connection *cnx,
 421                           struct VethLpEvent *event)
 422 {
 423         unsigned long flags;
 424
 425         spin_lock_irqsave(&cnx->lock, flags);
 426         /* Receiving caps may mean the other end has just come up, so
 427          * we need to reload the instance ID of the far end */
 428         cnx->dst_inst =
 429                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
 430                                                   HvLpEvent_Type_VirtualLan);
 431
 432         if (cnx->state & VETH_STATE_GOTCAPS) {
 433                 veth_error("Received a second capabilities from LPAR %d.\n",
 434                            cnx->remote_lp);
 435                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
 436                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
 437         } else {
 438                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
 439                 cnx->state |= VETH_STATE_GOTCAPS;
 440                 veth_kick_statemachine(cnx);
 441         }
 442         spin_unlock_irqrestore(&cnx->lock, flags);
 443 }
 444
 445 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
 446                               struct VethLpEvent *event)
 447 {
 448         unsigned long flags;
 449
 450         spin_lock_irqsave(&cnx->lock, flags);
 451         if (cnx->state & VETH_STATE_GOTCAPACK) {
 452                 veth_error("Received a second capabilities ack from LPAR %d.\n",
 453                            cnx->remote_lp);
 454         } else {
 455                 memcpy(&cnx->cap_ack_event, event,
 456                        sizeof(&cnx->cap_ack_event));
 457                 cnx->state |= VETH_STATE_GOTCAPACK;
 458                 veth_kick_statemachine(cnx);
 459         }
 460         spin_unlock_irqrestore(&cnx->lock, flags);
 461 }
 462
 463 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
 464                                   struct VethLpEvent *event)
 465 {
 466         unsigned long flags;
 467
 468         spin_lock_irqsave(&cnx->lock, flags);
 469         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
 470
 471         /* Avoid kicking the statemachine once we're shutdown.
 472          * It's unnecessary and it could break veth_stop_connection(). */
 473
 474         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 475                 cnx->state |= VETH_STATE_RESET;
 476                 veth_kick_statemachine(cnx);
 477         }
 478         spin_unlock_irqrestore(&cnx->lock, flags);
 479 }
 480
 481 static void veth_handle_ack(struct VethLpEvent *event)
 482 {
 483         HvLpIndex rlp = event->base_event.xTargetLp;
 484         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 485
 486         BUG_ON(! cnx);
 487
 488         switch (event->base_event.xSubtype) {
 489         case VethEventTypeCap:
 490                 veth_take_cap_ack(cnx, event);
 491                 break;
 492         case VethEventTypeMonitor:
 493                 veth_take_monitor_ack(cnx, event);
 494                 break;
 495         default:
 496                 veth_error("Unknown ack type %d from LPAR %d.\n",
 497                                 event->base_event.xSubtype, rlp);
 498         };
 499 }
 500
 501 static void veth_handle_int(struct VethLpEvent *event)
 502 {
 503         HvLpIndex rlp = event->base_event.xSourceLp;
 504         struct veth_lpar_connection *cnx = veth_cnx[rlp];
 505         unsigned long flags;
 506         int i, acked = 0;
 507
 508         BUG_ON(! cnx);
 509
 510         switch (event->base_event.xSubtype) {
 511         case VethEventTypeCap:
 512                 veth_take_cap(cnx, event);
 513                 break;
 514         case VethEventTypeMonitor:
 515                 /* do nothing... this'll hang out here til we're dead,
 516                  * and the hypervisor will return it for us. */
 517                 break;
 518         case VethEventTypeFramesAck:
 519                 spin_lock_irqsave(&cnx->lock, flags);
 520
 521                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
 522                         u16 msgnum = event->u.frames_ack_data.token[i];
 523
 524                         if (msgnum < VETH_NUMBUFFERS) {
 525                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
 526                                 cnx->outstanding_tx--;
 527                                 acked++;
 528                         }
 529                 }
 530
 531                 if (acked > 0) {
 532                         cnx->last_contact = jiffies;
 533                         veth_wake_queues(cnx);
 534                 }
 535
 536                 spin_unlock_irqrestore(&cnx->lock, flags);
 537                 break;
 538         case VethEventTypeFrames:
 539                 veth_receive(cnx, event);
 540                 break;
 541         default:
 542                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
 543                                 event->base_event.xSubtype, rlp);
 544         };
 545 }
 546
 547 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
 548 {
 549         struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
 550
 551         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
 552                 veth_handle_ack(veth_event);
 553         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
 554                 veth_handle_int(veth_event);
 555 }
 556
 557 static int veth_process_caps(struct veth_lpar_connection *cnx)
 558 {
 559         struct VethCapData *remote_caps = &cnx->remote_caps;
 560         int num_acks_needed;
 561
 562         /* Convert timer to jiffies */
 563         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
 564
 565         if ( (remote_caps->num_buffers == 0)
 566              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
 567              || (remote_caps->ack_threshold == 0)
 568              || (cnx->ack_timeout == 0) ) {
 569                 veth_error("Received incompatible capabilities from LPAR %d.\n",
 570                                 cnx->remote_lp);
 571                 return HvLpEvent_Rc_InvalidSubtypeData;
 572         }
 573
 574         num_acks_needed = (remote_caps->num_buffers
 575                            / remote_caps->ack_threshold) + 1;
 576
 577         /* FIXME: locking on num_ack_events? */
 578         if (cnx->num_ack_events < num_acks_needed) {
 579                 int num;
 580
 581                 num = veth_allocate_events(cnx->remote_lp,
 582                                            num_acks_needed-cnx->num_ack_events);
 583                 if (num > 0)
 584                         cnx->num_ack_events += num;
 585
 586                 if (cnx->num_ack_events < num_acks_needed) {
 587                         veth_error("Couldn't allocate enough ack events "
 588                                         "for LPAR %d.\n", cnx->remote_lp);
 589
 590                         return HvLpEvent_Rc_BufferNotAvailable;
 591                 }
 592         }
 593
 594
 595         return HvLpEvent_Rc_Good;
 596 }
 597
 598 /* FIXME: The gotos here are a bit dubious */
 599 static void veth_statemachine(void *p)
 600 {
 601         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
 602         int rlp = cnx->remote_lp;
 603         int rc;
 604
 605         spin_lock_irq(&cnx->lock);
 606
 607  restart:
 608         if (cnx->state & VETH_STATE_RESET) {
 609                 if (cnx->state & VETH_STATE_OPEN)
 610                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
 611                                                      HvLpEvent_Type_VirtualLan);
 612
 613                 /*
 614                  * Reset ack data. This prevents the ack_timer actually
 615                  * doing anything, even if it runs one more time when
 616                  * we drop the lock below.
 617                  */
 618                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 619                 cnx->num_pending_acks = 0;
 620
 621                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
 622                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
 623                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
 624                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
 625
 626                 /* Clean up any leftover messages */
 627                 if (cnx->msgs) {
 628                         int i;
 629                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
 630                                 veth_recycle_msg(cnx, cnx->msgs + i);
 631                 }
 632
 633                 cnx->outstanding_tx = 0;
 634                 veth_wake_queues(cnx);
 635
 636                 /* Drop the lock so we can do stuff that might sleep or
 637                  * take other locks. */
 638                 spin_unlock_irq(&cnx->lock);
 639
 640                 del_timer_sync(&cnx->ack_timer);
 641                 del_timer_sync(&cnx->reset_timer);
 642
 643                 spin_lock_irq(&cnx->lock);
 644
 645                 if (cnx->state & VETH_STATE_RESET)
 646                         goto restart;
 647
 648                 /* Hack, wait for the other end to reset itself. */
 649                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
 650                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
 651                         goto out;
 652                 }
 653         }
 654
 655         if (cnx->state & VETH_STATE_SHUTDOWN)
 656                 /* It's all over, do nothing */
 657                 goto out;
 658
 659         if ( !(cnx->state & VETH_STATE_OPEN) ) {
 660                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
 661                         goto cant_cope;
 662
 663                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
 664                 cnx->src_inst =
 665                         HvCallEvent_getSourceLpInstanceId(rlp,
 666                                                           HvLpEvent_Type_VirtualLan);
 667                 cnx->dst_inst =
 668                         HvCallEvent_getTargetLpInstanceId(rlp,
 669                                                           HvLpEvent_Type_VirtualLan);
 670                 cnx->state |= VETH_STATE_OPEN;
 671         }
 672
 673         if ( (cnx->state & VETH_STATE_OPEN)
 674              && !(cnx->state & VETH_STATE_SENTMON) ) {
 675                 rc = veth_signalevent(cnx, VethEventTypeMonitor,
 676                                       HvLpEvent_AckInd_DoAck,
 677                                       HvLpEvent_AckType_DeferredAck,
 678                                       0, 0, 0, 0, 0, 0);
 679
 680                 if (rc == HvLpEvent_Rc_Good) {
 681                         cnx->state |= VETH_STATE_SENTMON;
 682                 } else {
 683                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 684                              && (rc != HvLpEvent_Rc_PathClosed) )
 685                                 veth_error("Error sending monitor to LPAR %d, "
 686                                                 "rc = %d\n", rlp, rc);
 687
 688                         /* Oh well, hope we get a cap from the other
 689                          * end and do better when that kicks us */
 690                         goto out;
 691                 }
 692         }
 693
 694         if ( (cnx->state & VETH_STATE_OPEN)
 695              && !(cnx->state & VETH_STATE_SENTCAPS)) {
 696                 u64 *rawcap = (u64 *)&cnx->local_caps;
 697
 698                 rc = veth_signalevent(cnx, VethEventTypeCap,
 699                                       HvLpEvent_AckInd_DoAck,
 700                                       HvLpEvent_AckType_ImmediateAck,
 701                                       0, rawcap[0], rawcap[1], rawcap[2],
 702                                       rawcap[3], rawcap[4]);
 703
 704                 if (rc == HvLpEvent_Rc_Good) {
 705                         cnx->state |= VETH_STATE_SENTCAPS;
 706                 } else {
 707                         if ( (rc != HvLpEvent_Rc_PartitionDead)
 708                              && (rc != HvLpEvent_Rc_PathClosed) )
 709                                 veth_error("Error sending caps to LPAR %d, "
 710                                                 "rc = %d\n", rlp, rc);
 711
 712                         /* Oh well, hope we get a cap from the other
 713                          * end and do better when that kicks us */
 714                         goto out;
 715                 }
 716         }
 717
 718         if ((cnx->state & VETH_STATE_GOTCAPS)
 719             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
 720                 struct VethCapData *remote_caps = &cnx->remote_caps;
 721
 722                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
 723                        sizeof(*remote_caps));
 724
 725                 spin_unlock_irq(&cnx->lock);
 726                 rc = veth_process_caps(cnx);
 727                 spin_lock_irq(&cnx->lock);
 728
 729                 /* We dropped the lock, so recheck for anything which
 730                  * might mess us up */
 731                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
 732                         goto restart;
 733
 734                 cnx->cap_event.base_event.xRc = rc;
 735                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
 736                 if (rc == HvLpEvent_Rc_Good)
 737                         cnx->state |= VETH_STATE_SENTCAPACK;
 738                 else
 739                         goto cant_cope;
 740         }
 741
 742         if ((cnx->state & VETH_STATE_GOTCAPACK)
 743             && (cnx->state & VETH_STATE_GOTCAPS)
 744             && !(cnx->state & VETH_STATE_READY)) {
 745                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
 746                         /* Start the ACK timer */
 747                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
 748                         add_timer(&cnx->ack_timer);
 749                         cnx->state |= VETH_STATE_READY;
 750                 } else {
 751                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
 752                                         rlp, cnx->cap_ack_event.base_event.xRc);
 753                         goto cant_cope;
 754                 }
 755         }
 756
 757  out:
 758         spin_unlock_irq(&cnx->lock);
 759         return;
 760
 761  cant_cope:
 762         /* FIXME: we get here if something happens we really can't
 763          * cope with.  The link will never work once we get here, and
 764          * all we can do is not lock the rest of the system up */
 765         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
 766                         " (state = 0x%04lx)\n", rlp, cnx->state);
 767         cnx->state |= VETH_STATE_SHUTDOWN;
 768         spin_unlock_irq(&cnx->lock);
 769 }
 770
 771 static int veth_init_connection(u8 rlp)
 772 {
 773         struct veth_lpar_connection *cnx;
 774         struct veth_msg *msgs;
 775         int i, rc;
 776
 777         if ( (rlp == this_lp)
 778              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
 779                 return 0;
 780
 781         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
 782         if (! cnx)
 783                 return -ENOMEM;
 784         memset(cnx, 0, sizeof(*cnx));
 785
 786         cnx->remote_lp = rlp;
 787         spin_lock_init(&cnx->lock);
 788         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
 789
 790         init_timer(&cnx->ack_timer);
 791         cnx->ack_timer.function = veth_timed_ack;
 792         cnx->ack_timer.data = (unsigned long) cnx;
 793
 794         init_timer(&cnx->reset_timer);
 795         cnx->reset_timer.function = veth_timed_reset;
 796         cnx->reset_timer.data = (unsigned long) cnx;
 797         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
 798
 799         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
 800
 801         veth_cnx[rlp] = cnx;
 802
 803         /* This gets us 1 reference, which is held on behalf of the driver
 804          * infrastructure. It's released at module unload. */
 805         kobject_init(&cnx->kobject);
 806         cnx->kobject.ktype = &veth_lpar_connection_ktype;
 807         rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
 808         if (rc != 0)
 809                 return rc;
 810
 811         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
 812         if (! msgs) {
 813                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
 814                 return -ENOMEM;
 815         }
 816
 817         cnx->msgs = msgs;
 818         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
 819
 820         for (i = 0; i < VETH_NUMBUFFERS; i++) {
 821                 msgs[i].token = i;
 822                 veth_stack_push(cnx, msgs + i);
 823         }
 824
 825         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
 826
 827         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
 828                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
 829                 return -ENOMEM;
 830         }
 831
 832         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
 833         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
 834         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
 835
 836         return 0;
 837 }
 838
 839 static void veth_stop_connection(struct veth_lpar_connection *cnx)
 840 {
 841         if (!cnx)
 842                 return;
 843
 844         spin_lock_irq(&cnx->lock);
 845         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
 846         veth_kick_statemachine(cnx);
 847         spin_unlock_irq(&cnx->lock);
 848
 849         /* There's a slim chance the reset code has just queued the
 850          * statemachine to run in five seconds. If so we need to cancel
 851          * that and requeue the work to run now. */
 852         if (cancel_delayed_work(&cnx->statemachine_wq)) {
 853                 spin_lock_irq(&cnx->lock);
 854                 veth_kick_statemachine(cnx);
 855                 spin_unlock_irq(&cnx->lock);
 856         }
 857
 858         /* Wait for the state machine to run. */
 859         flush_scheduled_work();
 860 }
 861
 862 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
 863 {
 864         if (!cnx)
 865                 return;
 866
 867         if (cnx->num_events > 0)
 868                 mf_deallocate_lp_events(cnx->remote_lp,
 869                                       HvLpEvent_Type_VirtualLan,
 870                                       cnx->num_events,
 871                                       NULL, NULL);
 872         if (cnx->num_ack_events > 0)
 873                 mf_deallocate_lp_events(cnx->remote_lp,
 874                                       HvLpEvent_Type_VirtualLan,
 875                                       cnx->num_ack_events,
 876                                       NULL, NULL);
 877
 878         kfree(cnx->msgs);
 879         veth_cnx[cnx->remote_lp] = NULL;
 880         kfree(cnx);
 881 }
 882
 883 static void veth_release_connection(struct kobject *kobj)
 884 {
 885         struct veth_lpar_connection *cnx;
 886         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
 887         veth_stop_connection(cnx);
 888         veth_destroy_connection(cnx);
 889 }
 890
 891 /*
 892  * net_device code
 893  */
 894
 895 static int veth_open(struct net_device *dev)
 896 {
 897         struct veth_port *port = (struct veth_port *) dev->priv;
 898
 899         memset(&port->stats, 0, sizeof (port->stats));
 900         netif_start_queue(dev);
 901         return 0;
 902 }
 903
 904 static int veth_close(struct net_device *dev)
 905 {
 906         netif_stop_queue(dev);
 907         return 0;
 908 }
 909
 910 static struct net_device_stats *veth_get_stats(struct net_device *dev)
 911 {
 912         struct veth_port *port = (struct veth_port *) dev->priv;
 913
 914         return &port->stats;
 915 }
 916
 917 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 918 {
 919         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
 920                 return -EINVAL;
 921         dev->mtu = new_mtu;
 922         return 0;
 923 }
 924
 925 static void veth_set_multicast_list(struct net_device *dev)
 926 {
 927         struct veth_port *port = (struct veth_port *) dev->priv;
 928         unsigned long flags;
 929
 930         write_lock_irqsave(&port->mcast_gate, flags);
 931
 932         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
 933                         (dev->mc_count > VETH_MAX_MCAST)) {
 934                 port->promiscuous = 1;
 935         } else {
 936                 struct dev_mc_list *dmi = dev->mc_list;
 937                 int i;
 938
 939                 port->promiscuous = 0;
 940
 941                 /* Update table */
 942                 port->num_mcast = 0;
 943
 944                 for (i = 0; i < dev->mc_count; i++) {
 945                         u8 *addr = dmi->dmi_addr;
 946                         u64 xaddr = 0;
 947
 948                         if (addr[0] & 0x01) {/* multicast address? */
 949                                 memcpy(&xaddr, addr, ETH_ALEN);
 950                                 port->mcast_addr[port->num_mcast] = xaddr;
 951                                 port->num_mcast++;
 952                         }
 953                         dmi = dmi->next;
 954                 }
 955         }
 956
 957         write_unlock_irqrestore(&port->mcast_gate, flags);
 958 }
 959
 960 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
 961 {
 962         strncpy(info->driver, "veth", sizeof(info->driver) - 1);
 963         info->driver[sizeof(info->driver) - 1] = '\0';
 964         strncpy(info->version, "1.0", sizeof(info->version) - 1);
 965 }
 966
 967 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
 968 {
 969         ecmd->supported = (SUPPORTED_1000baseT_Full
 970                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 971         ecmd->advertising = (SUPPORTED_1000baseT_Full
 972                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 973         ecmd->port = PORT_FIBRE;
 974         ecmd->transceiver = XCVR_INTERNAL;
 975         ecmd->phy_address = 0;
 976         ecmd->speed = SPEED_1000;
 977         ecmd->duplex = DUPLEX_FULL;
 978         ecmd->autoneg = AUTONEG_ENABLE;
 979         ecmd->maxtxpkt = 120;
 980         ecmd->maxrxpkt = 120;
 981         return 0;
 982 }
 983
 984 static u32 veth_get_link(struct net_device *dev)
 985 {
 986         return 1;
 987 }
 988
 989 static struct ethtool_ops ops = {
 990         .get_drvinfo = veth_get_drvinfo,
 991         .get_settings = veth_get_settings,
 992         .get_link = veth_get_link,
 993 };
 994
 995 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
 996 {
 997         struct net_device *dev;
 998         struct veth_port *port;
 999         int i, rc;
1000
1001         dev = alloc_etherdev(sizeof (struct veth_port));
1002         if (! dev) {
1003                 veth_error("Unable to allocate net_device structure!\n");
1004                 return NULL;
1005         }
1006
1007         port = (struct veth_port *) dev->priv;
1008
1009         spin_lock_init(&port->queue_lock);
1010         rwlock_init(&port->mcast_gate);
1011         port->stopped_map = 0;
1012
1013         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1014                 HvLpVirtualLanIndexMap map;
1015
1016                 if (i == this_lp)
1017                         continue;
1018                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1019                 if (map & (0x8000 >> vlan))
1020                         port->lpar_map |= (1 << i);
1021         }
1022         port->dev = vdev;
1023
1024         dev->dev_addr[0] = 0x02;
1025         dev->dev_addr[1] = 0x01;
1026         dev->dev_addr[2] = 0xff;
1027         dev->dev_addr[3] = vlan;
1028         dev->dev_addr[4] = 0xff;
1029         dev->dev_addr[5] = this_lp;
1030
1031         dev->mtu = VETH_MAX_MTU;
1032
1033         memcpy(&port->mac_addr, dev->dev_addr, 6);
1034
1035         dev->open = veth_open;
1036         dev->hard_start_xmit = veth_start_xmit;
1037         dev->stop = veth_close;
1038         dev->get_stats = veth_get_stats;
1039         dev->change_mtu = veth_change_mtu;
1040         dev->set_mac_address = NULL;
1041         dev->set_multicast_list = veth_set_multicast_list;
1042         SET_ETHTOOL_OPS(dev, &ops);
1043
1044         SET_NETDEV_DEV(dev, vdev);
1045
1046         rc = register_netdev(dev);
1047         if (rc != 0) {
1048                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1049                 free_netdev(dev);
1050                 return NULL;
1051         }
1052
1053         kobject_init(&port->kobject);
1054         port->kobject.parent = &dev->class_dev.kobj;
1055         port->kobject.ktype  = &veth_port_ktype;
1056         kobject_set_name(&port->kobject, "veth_port");
1057         if (0 != kobject_add(&port->kobject))
1058                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1059
1060         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1061                         dev->name, vlan, port->lpar_map);
1062
1063         return dev;
1064 }
1065
1066 /*
1067  * Tx path
1068  */
1069
1070 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1071                                 struct net_device *dev)
1072 {
1073         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1074         struct veth_port *port = (struct veth_port *) dev->priv;
1075         HvLpEvent_Rc rc;
1076         struct veth_msg *msg = NULL;
1077         unsigned long flags;
1078
1079         if (! cnx)
1080                 return 0;
1081
1082         spin_lock_irqsave(&cnx->lock, flags);
1083
1084         if (! (cnx->state & VETH_STATE_READY))
1085                 goto no_error;
1086
1087         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1088                 goto drop;
1089
1090         msg = veth_stack_pop(cnx);
1091         if (! msg)
1092                 goto drop;
1093
1094         msg->in_use = 1;
1095         msg->skb = skb_get(skb);
1096
1097         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1098                                 skb->len, DMA_TO_DEVICE);
1099
1100         if (dma_mapping_error(msg->data.addr[0]))
1101                 goto recycle_and_drop;
1102
1103         msg->dev = port->dev;
1104         msg->data.len[0] = skb->len;
1105         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1106
1107         rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
1108
1109         if (rc != HvLpEvent_Rc_Good)
1110                 goto recycle_and_drop;
1111
1112         /* If the timer's not already running, start it now. */
1113         if (0 == cnx->outstanding_tx)
1114                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1115
1116         cnx->last_contact = jiffies;
1117         cnx->outstanding_tx++;
1118
1119         if (veth_stack_is_empty(cnx))
1120                 veth_stop_queues(cnx);
1121
1122  no_error:
1123         spin_unlock_irqrestore(&cnx->lock, flags);
1124         return 0;
1125
1126  recycle_and_drop:
1127         veth_recycle_msg(cnx, msg);
1128  drop:
1129         spin_unlock_irqrestore(&cnx->lock, flags);
1130         return 1;
1131 }
1132
1133 static void veth_transmit_to_many(struct sk_buff *skb,
1134                                           HvLpIndexMap lpmask,
1135                                           struct net_device *dev)
1136 {
1137         struct veth_port *port = (struct veth_port *) dev->priv;
1138         int i, success, error;
1139
1140         success = error = 0;
1141
1142         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1143                 if ((lpmask & (1 << i)) == 0)
1144                         continue;
1145
1146                 if (veth_transmit_to_one(skb, i, dev))
1147                         error = 1;
1148                 else
1149                         success = 1;
1150         }
1151
1152         if (error)
1153                 port->stats.tx_errors++;
1154
1155         if (success) {
1156                 port->stats.tx_packets++;
1157                 port->stats.tx_bytes += skb->len;
1158         }
1159 }
1160
1161 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1162 {
1163         unsigned char *frame = skb->data;
1164         struct veth_port *port = (struct veth_port *) dev->priv;
1165         HvLpIndexMap lpmask;
1166
1167         if (! (frame[0] & 0x01)) {
1168                 /* unicast packet */
1169                 HvLpIndex rlp = frame[5];
1170
1171                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1172                         dev_kfree_skb(skb);
1173                         return 0;
1174                 }
1175
1176                 lpmask = 1 << rlp;
1177         } else {
1178                 lpmask = port->lpar_map;
1179         }
1180
1181         veth_transmit_to_many(skb, lpmask, dev);
1182
1183         dev_kfree_skb(skb);
1184
1185         return 0;
1186 }
1187
1188 /* You must hold the connection's lock when you call this function. */
1189 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1190                              struct veth_msg *msg)
1191 {
1192         u32 dma_address, dma_length;
1193
1194         if (msg->in_use) {
1195                 msg->in_use = 0;
1196                 dma_address = msg->data.addr[0];
1197                 dma_length = msg->data.len[0];
1198
1199                 if (!dma_mapping_error(dma_address))
1200                         dma_unmap_single(msg->dev, dma_address, dma_length,
1201                                         DMA_TO_DEVICE);
1202
1203                 if (msg->skb) {
1204                         dev_kfree_skb_any(msg->skb);
1205                         msg->skb = NULL;
1206                 }
1207
1208                 memset(&msg->data, 0, sizeof(msg->data));
1209                 veth_stack_push(cnx, msg);
1210         } else if (cnx->state & VETH_STATE_OPEN) {
1211                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1212                                 cnx->remote_lp, msg->token);
1213         }
1214 }
1215
1216 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1217 {
1218         int i;
1219
1220         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1221                 struct net_device *dev = veth_dev[i];
1222                 struct veth_port *port;
1223                 unsigned long flags;
1224
1225                 if (! dev)
1226                         continue;
1227
1228                 port = (struct veth_port *)dev->priv;
1229
1230                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1231                         continue;
1232
1233                 spin_lock_irqsave(&port->queue_lock, flags);
1234
1235                 port->stopped_map &= ~(1 << cnx->remote_lp);
1236
1237                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1238                         veth_debug("cnx %d: woke queue for %s.\n",
1239                                         cnx->remote_lp, dev->name);
1240                         netif_wake_queue(dev);
1241                 }
1242                 spin_unlock_irqrestore(&port->queue_lock, flags);
1243         }
1244 }
1245
1246 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1247 {
1248         int i;
1249
1250         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1251                 struct net_device *dev = veth_dev[i];
1252                 struct veth_port *port;
1253
1254                 if (! dev)
1255                         continue;
1256
1257                 port = (struct veth_port *)dev->priv;
1258
1259                 /* If this cnx is not on the vlan for this port, continue */
1260                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1261                         continue;
1262
1263                 spin_lock(&port->queue_lock);
1264
1265                 netif_stop_queue(dev);
1266                 port->stopped_map |= (1 << cnx->remote_lp);
1267
1268                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1269                                 cnx->remote_lp, dev->name, port->stopped_map);
1270
1271                 spin_unlock(&port->queue_lock);
1272         }
1273 }
1274
1275 static void veth_timed_reset(unsigned long ptr)
1276 {
1277         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1278         unsigned long trigger_time, flags;
1279
1280         /* FIXME is it possible this fires after veth_stop_connection()?
1281          * That would reschedule the statemachine for 5 seconds and probably
1282          * execute it after the module's been unloaded. Hmm. */
1283
1284         spin_lock_irqsave(&cnx->lock, flags);
1285
1286         if (cnx->outstanding_tx > 0) {
1287                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1288
1289                 if (trigger_time < jiffies) {
1290                         cnx->state |= VETH_STATE_RESET;
1291                         veth_kick_statemachine(cnx);
1292                         veth_error("%d packets not acked by LPAR %d within %d "
1293                                         "seconds, resetting.\n",
1294                                         cnx->outstanding_tx, cnx->remote_lp,
1295                                         cnx->reset_timeout / HZ);
1296                 } else {
1297                         /* Reschedule the timer */
1298                         trigger_time = jiffies + cnx->reset_timeout;
1299                         mod_timer(&cnx->reset_timer, trigger_time);
1300                 }
1301         }
1302
1303         spin_unlock_irqrestore(&cnx->lock, flags);
1304 }
1305
1306 /*
1307  * Rx path
1308  */
1309
1310 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1311 {
1312         int wanted = 0;
1313         int i;
1314         unsigned long flags;
1315
1316         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1317                 return 1;
1318
1319         read_lock_irqsave(&port->mcast_gate, flags);
1320
1321         if (port->promiscuous) {
1322                 wanted = 1;
1323                 goto out;
1324         }
1325
1326         for (i = 0; i < port->num_mcast; ++i) {
1327                 if (port->mcast_addr[i] == mac_addr) {
1328                         wanted = 1;
1329                         break;
1330                 }
1331         }
1332
1333  out:
1334         read_unlock_irqrestore(&port->mcast_gate, flags);
1335
1336         return wanted;
1337 }
1338
1339 struct dma_chunk {
1340         u64 addr;
1341         u64 size;
1342 };
1343
1344 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1345
1346 static inline void veth_build_dma_list(struct dma_chunk *list,
1347                                        unsigned char *p, unsigned long length)
1348 {
1349         unsigned long done;
1350         int i = 1;
1351
1352         /* FIXME: skbs are continguous in real addresses.  Do we
1353          * really need to break it into PAGE_SIZE chunks, or can we do
1354          * it just at the granularity of iSeries real->absolute
1355          * mapping?  Indeed, given the way the allocator works, can we
1356          * count on them being absolutely contiguous? */
1357         list[0].addr = ISERIES_HV_ADDR(p);
1358         list[0].size = min(length,
1359                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1360
1361         done = list[0].size;
1362         while (done < length) {
1363                 list[i].addr = ISERIES_HV_ADDR(p + done);
1364                 list[i].size = min(length-done, PAGE_SIZE);
1365                 done += list[i].size;
1366                 i++;
1367         }
1368 }
1369
1370 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1371 {
1372         HvLpEvent_Rc rc;
1373
1374         rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1375                              0, &cnx->pending_acks);
1376
1377         if (rc != HvLpEvent_Rc_Good)
1378                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1379                                 cnx->remote_lp, (int)rc);
1380
1381         cnx->num_pending_acks = 0;
1382         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1383 }
1384
1385 static void veth_receive(struct veth_lpar_connection *cnx,
1386                          struct VethLpEvent *event)
1387 {
1388         struct VethFramesData *senddata = &event->u.frames_data;
1389         int startchunk = 0;
1390         int nchunks;
1391         unsigned long flags;
1392         HvLpDma_Rc rc;
1393
1394         do {
1395                 u16 length = 0;
1396                 struct sk_buff *skb;
1397                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1398                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1399                 u64 dest;
1400                 HvLpVirtualLanIndex vlan;
1401                 struct net_device *dev;
1402                 struct veth_port *port;
1403
1404                 /* FIXME: do we need this? */
1405                 memset(local_list, 0, sizeof(local_list));
1406                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1407
1408                 /* a 0 address marks the end of the valid entries */
1409                 if (senddata->addr[startchunk] == 0)
1410                         break;
1411
1412                 /* make sure that we have at least 1 EOF entry in the
1413                  * remaining entries */
1414                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1415                         veth_error("Missing EOF fragment in event "
1416                                         "eofmask = 0x%x startchunk = %d\n",
1417                                         (unsigned)senddata->eofmask,
1418                                         startchunk);
1419                         break;
1420                 }
1421
1422                 /* build list of chunks in this frame */
1423                 nchunks = 0;
1424                 do {
1425                         remote_list[nchunks].addr =
1426                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1427                         remote_list[nchunks].size =
1428                                 senddata->len[startchunk+nchunks];
1429                         length += remote_list[nchunks].size;
1430                 } while (! (senddata->eofmask &
1431                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1432
1433                 /* length == total length of all chunks */
1434                 /* nchunks == # of chunks in this frame */
1435
1436                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1437                         veth_error("Received oversize frame from LPAR %d "
1438                                         "(length = %d)\n",
1439                                         cnx->remote_lp, length);
1440                         continue;
1441                 }
1442
1443                 skb = alloc_skb(length, GFP_ATOMIC);
1444                 if (!skb)
1445                         continue;
1446
1447                 veth_build_dma_list(local_list, skb->data, length);
1448
1449                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1450                                             event->base_event.xSourceLp,
1451                                             HvLpDma_Direction_RemoteToLocal,
1452                                             cnx->src_inst,
1453                                             cnx->dst_inst,
1454                                             HvLpDma_AddressType_RealAddress,
1455                                             HvLpDma_AddressType_TceIndex,
1456                                             ISERIES_HV_ADDR(&local_list),
1457                                             ISERIES_HV_ADDR(&remote_list),
1458                                             length);
1459                 if (rc != HvLpDma_Rc_Good) {
1460                         dev_kfree_skb_irq(skb);
1461                         continue;
1462                 }
1463
1464                 vlan = skb->data[9];
1465                 dev = veth_dev[vlan];
1466                 if (! dev) {
1467                         /*
1468                          * Some earlier versions of the driver sent
1469                          * broadcasts down all connections, even to lpars
1470                          * that weren't on the relevant vlan. So ignore
1471                          * packets belonging to a vlan we're not on.
1472                          * We can also be here if we receive packets while
1473                          * the driver is going down, because then dev is NULL.
1474                          */
1475                         dev_kfree_skb_irq(skb);
1476                         continue;
1477                 }
1478
1479                 port = (struct veth_port *)dev->priv;
1480                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1481
1482                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1483                         dev_kfree_skb_irq(skb);
1484                         continue;
1485                 }
1486                 if (! veth_frame_wanted(port, dest)) {
1487                         dev_kfree_skb_irq(skb);
1488                         continue;
1489                 }
1490
1491                 skb_put(skb, length);
1492                 skb->dev = dev;
1493                 skb->protocol = eth_type_trans(skb, dev);
1494                 skb->ip_summed = CHECKSUM_NONE;
1495                 netif_rx(skb);  /* send it up */
1496                 port->stats.rx_packets++;
1497                 port->stats.rx_bytes += length;
1498         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1499
1500         /* Ack it */
1501         spin_lock_irqsave(&cnx->lock, flags);
1502         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1503
1504         cnx->pending_acks[cnx->num_pending_acks++] =
1505                 event->base_event.xCorrelationToken;
1506
1507         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1508              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1509                 veth_flush_acks(cnx);
1510
1511         spin_unlock_irqrestore(&cnx->lock, flags);
1512 }
1513
1514 static void veth_timed_ack(unsigned long ptr)
1515 {
1516         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1517         unsigned long flags;
1518
1519         /* Ack all the events */
1520         spin_lock_irqsave(&cnx->lock, flags);
1521         if (cnx->num_pending_acks > 0)
1522                 veth_flush_acks(cnx);
1523
1524         /* Reschedule the timer */
1525         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1526         add_timer(&cnx->ack_timer);
1527         spin_unlock_irqrestore(&cnx->lock, flags);
1528 }
1529
1530 static int veth_remove(struct vio_dev *vdev)
1531 {
1532         struct veth_lpar_connection *cnx;
1533         struct net_device *dev;
1534         struct veth_port *port;
1535         int i;
1536
1537         dev = veth_dev[vdev->unit_address];
1538
1539         if (! dev)
1540                 return 0;
1541
1542         port = netdev_priv(dev);
1543
1544         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1545                 cnx = veth_cnx[i];
1546
1547                 if (cnx && (port->lpar_map & (1 << i))) {
1548                         /* Drop our reference to connections on our VLAN */
1549                         kobject_put(&cnx->kobject);
1550                 }
1551         }
1552
1553         veth_dev[vdev->unit_address] = NULL;
1554         kobject_del(&port->kobject);
1555         kobject_put(&port->kobject);
1556         unregister_netdev(dev);
1557         free_netdev(dev);
1558
1559         return 0;
1560 }
1561
1562 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1563 {
1564         int i = vdev->unit_address;
1565         struct net_device *dev;
1566         struct veth_port *port;
1567
1568         dev = veth_probe_one(i, &vdev->dev);
1569         if (dev == NULL) {
1570                 veth_remove(vdev);
1571                 return 1;
1572         }
1573         veth_dev[i] = dev;
1574
1575         port = (struct veth_port*)netdev_priv(dev);
1576
1577         /* Start the state machine on each connection on this vlan. If we're
1578          * the first dev to do so this will commence link negotiation */
1579         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1580                 struct veth_lpar_connection *cnx;
1581
1582                 if (! (port->lpar_map & (1 << i)))
1583                         continue;
1584
1585                 cnx = veth_cnx[i];
1586                 if (!cnx)
1587                         continue;
1588
1589                 kobject_get(&cnx->kobject);
1590                 veth_kick_statemachine(cnx);
1591         }
1592
1593         return 0;
1594 }
1595
1596 /**
1597  * veth_device_table: Used by vio.c to match devices that we
1598  * support.
1599  */
1600 static struct vio_device_id veth_device_table[] __devinitdata = {
1601         { "vlan", "" },
1602         { "", "" }
1603 };
1604 MODULE_DEVICE_TABLE(vio, veth_device_table);
1605
1606 static struct vio_driver veth_driver = {
1607         .name = "iseries_veth",
1608         .id_table = veth_device_table,
1609         .probe = veth_probe,
1610         .remove = veth_remove
1611 };
1612
1613 /*
1614  * Module initialization/cleanup
1615  */
1616
1617 void __exit veth_module_cleanup(void)
1618 {
1619         int i;
1620         struct veth_lpar_connection *cnx;
1621
1622         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1623         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1624
1625         /* Make sure any work queued from Hypervisor callbacks is finished. */
1626         flush_scheduled_work();
1627
1628         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1629                 cnx = veth_cnx[i];
1630
1631                 if (!cnx)
1632                         continue;
1633
1634                 /* Remove the connection from sysfs */
1635                 kobject_del(&cnx->kobject);
1636                 /* Drop the driver's reference to the connection */
1637                 kobject_put(&cnx->kobject);
1638         }
1639
1640         /* Unregister the driver, which will close all the netdevs and stop
1641          * the connections when they're no longer referenced. */
1642         vio_unregister_driver(&veth_driver);
1643 }
1644 module_exit(veth_module_cleanup);
1645
1646 int __init veth_module_init(void)
1647 {
1648         int i;
1649         int rc;
1650
1651         this_lp = HvLpConfig_getLpIndex_outline();
1652
1653         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1654                 rc = veth_init_connection(i);
1655                 if (rc != 0)
1656                         goto error;
1657         }
1658
1659         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1660                                   &veth_handle_event);
1661
1662         rc = vio_register_driver(&veth_driver);
1663         if (rc != 0)
1664                 goto error;
1665
1666         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1667                 struct kobject *kobj;
1668
1669                 if (!veth_cnx[i])
1670                         continue;
1671
1672                 kobj = &veth_cnx[i]->kobject;
1673                 kobj->parent = &veth_driver.driver.kobj;
1674                 /* If the add failes, complain but otherwise continue */
1675                 if (0 != kobject_add(kobj))
1676                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1677         }
1678
1679         return 0;
1680
1681 error:
1682         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1683                 veth_destroy_connection(veth_cnx[i]);
1684         }
1685
1686         return rc;
1687 }
1688 module_init(veth_module_init);