Merge git://git.skbuff.net/gitroot/yoshfuji/linux-2.6.14+advapi-fix/
[linux-2.6] / 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/config.h>
60 #include <linux/module.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
73 #include <asm/abs_addr.h>
74 #include <asm/iseries/mf.h>
75 #include <asm/uaccess.h>
76
77 #include <asm/iseries/hv_lp_config.h>
78 #include <asm/iseries/hv_types.h>
79 #include <asm/iseries/hv_lp_event.h>
80 #include <asm/iommu.h>
81 #include <asm/vio.h>
82
83 #undef DEBUG
84
85 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
86 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
87 MODULE_LICENSE("GPL");
88
89 #define VETH_EVENT_CAP  (0)
90 #define VETH_EVENT_FRAMES       (1)
91 #define VETH_EVENT_MONITOR      (2)
92 #define VETH_EVENT_FRAMES_ACK   (3)
93
94 #define VETH_MAX_ACKS_PER_MSG   (20)
95 #define VETH_MAX_FRAMES_PER_MSG (6)
96
97 struct veth_frames_data {
98         u32 addr[VETH_MAX_FRAMES_PER_MSG];
99         u16 len[VETH_MAX_FRAMES_PER_MSG];
100         u32 eofmask;
101 };
102 #define VETH_EOF_SHIFT          (32-VETH_MAX_FRAMES_PER_MSG)
103
104 struct veth_frames_ack_data {
105         u16 token[VETH_MAX_ACKS_PER_MSG];
106 };
107
108 struct veth_cap_data {
109         u8 caps_version;
110         u8 rsvd1;
111         u16 num_buffers;
112         u16 ack_threshold;
113         u16 rsvd2;
114         u32 ack_timeout;
115         u32 rsvd3;
116         u64 rsvd4[3];
117 };
118
119 struct veth_lpevent {
120         struct HvLpEvent base_event;
121         union {
122                 struct veth_cap_data caps_data;
123                 struct veth_frames_data frames_data;
124                 struct veth_frames_ack_data frames_ack_data;
125         } u;
126
127 };
128
129 #define DRV_NAME        "iseries_veth"
130 #define DRV_VERSION     "2.0"
131
132 #define VETH_NUMBUFFERS         (120)
133 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
134 #define VETH_MAX_MCAST          (12)
135
136 #define VETH_MAX_MTU            (9000)
137
138 #if VETH_NUMBUFFERS < 10
139 #define ACK_THRESHOLD           (1)
140 #elif VETH_NUMBUFFERS < 20
141 #define ACK_THRESHOLD           (4)
142 #elif VETH_NUMBUFFERS < 40
143 #define ACK_THRESHOLD           (10)
144 #else
145 #define ACK_THRESHOLD           (20)
146 #endif
147
148 #define VETH_STATE_SHUTDOWN     (0x0001)
149 #define VETH_STATE_OPEN         (0x0002)
150 #define VETH_STATE_RESET        (0x0004)
151 #define VETH_STATE_SENTMON      (0x0008)
152 #define VETH_STATE_SENTCAPS     (0x0010)
153 #define VETH_STATE_GOTCAPACK    (0x0020)
154 #define VETH_STATE_GOTCAPS      (0x0040)
155 #define VETH_STATE_SENTCAPACK   (0x0080)
156 #define VETH_STATE_READY        (0x0100)
157
158 struct veth_msg {
159         struct veth_msg *next;
160         struct veth_frames_data data;
161         int token;
162         int in_use;
163         struct sk_buff *skb;
164         struct device *dev;
165 };
166
167 struct veth_lpar_connection {
168         HvLpIndex remote_lp;
169         struct work_struct statemachine_wq;
170         struct veth_msg *msgs;
171         int num_events;
172         struct veth_cap_data local_caps;
173
174         struct kobject kobject;
175         struct timer_list ack_timer;
176
177         struct timer_list reset_timer;
178         unsigned int reset_timeout;
179         unsigned long last_contact;
180         int outstanding_tx;
181
182         spinlock_t lock;
183         unsigned long state;
184         HvLpInstanceId src_inst;
185         HvLpInstanceId dst_inst;
186         struct veth_lpevent cap_event, cap_ack_event;
187         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
188         u32 num_pending_acks;
189
190         int num_ack_events;
191         struct veth_cap_data remote_caps;
192         u32 ack_timeout;
193
194         struct veth_msg *msg_stack_head;
195 };
196
197 struct veth_port {
198         struct device *dev;
199         struct net_device_stats stats;
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 = { COMPLETION_INITIALIZER(vc.c), 0 };
313
314         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
315                             sizeof(struct veth_lpevent), number,
316                             &veth_complete_allocation, &vc);
317         wait_for_completion(&vc.c);
318
319         return vc.num;
320 }
321
322 /*
323  * sysfs support
324  */
325
326 struct veth_cnx_attribute {
327         struct attribute attr;
328         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
329         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
330 };
331
332 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
333                 struct attribute *attr, char *buf)
334 {
335         struct veth_cnx_attribute *cnx_attr;
336         struct veth_lpar_connection *cnx;
337
338         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
339         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
340
341         if (!cnx_attr->show)
342                 return -EIO;
343
344         return cnx_attr->show(cnx, buf);
345 }
346
347 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
348 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
349 {                                                                       \
350         return sprintf(buf, _format, _expression);                      \
351 }                                                                       \
352 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
353
354 #define SIMPLE_CNX_ATTR(_name)  \
355         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
356
357 SIMPLE_CNX_ATTR(outstanding_tx);
358 SIMPLE_CNX_ATTR(remote_lp);
359 SIMPLE_CNX_ATTR(num_events);
360 SIMPLE_CNX_ATTR(src_inst);
361 SIMPLE_CNX_ATTR(dst_inst);
362 SIMPLE_CNX_ATTR(num_pending_acks);
363 SIMPLE_CNX_ATTR(num_ack_events);
364 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
365 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
366 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
367 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
368                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
369
370 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
371
372 static struct attribute *veth_cnx_default_attrs[] = {
373         GET_CNX_ATTR(outstanding_tx),
374         GET_CNX_ATTR(remote_lp),
375         GET_CNX_ATTR(num_events),
376         GET_CNX_ATTR(reset_timeout),
377         GET_CNX_ATTR(last_contact),
378         GET_CNX_ATTR(state),
379         GET_CNX_ATTR(src_inst),
380         GET_CNX_ATTR(dst_inst),
381         GET_CNX_ATTR(num_pending_acks),
382         GET_CNX_ATTR(num_ack_events),
383         GET_CNX_ATTR(ack_timeout),
384         NULL
385 };
386
387 static struct sysfs_ops veth_cnx_sysfs_ops = {
388                 .show = veth_cnx_attribute_show
389 };
390
391 static struct kobj_type veth_lpar_connection_ktype = {
392         .release        = veth_release_connection,
393         .sysfs_ops      = &veth_cnx_sysfs_ops,
394         .default_attrs  = veth_cnx_default_attrs
395 };
396
397 struct veth_port_attribute {
398         struct attribute attr;
399         ssize_t (*show)(struct veth_port *, char *buf);
400         ssize_t (*store)(struct veth_port *, const char *buf);
401 };
402
403 static ssize_t veth_port_attribute_show(struct kobject *kobj,
404                 struct attribute *attr, char *buf)
405 {
406         struct veth_port_attribute *port_attr;
407         struct veth_port *port;
408
409         port_attr = container_of(attr, struct veth_port_attribute, attr);
410         port = container_of(kobj, struct veth_port, kobject);
411
412         if (!port_attr->show)
413                 return -EIO;
414
415         return port_attr->show(port, buf);
416 }
417
418 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
419 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
420 {                                                                       \
421         return sprintf(buf, _format, _expression);                      \
422 }                                                                       \
423 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
424
425 #define SIMPLE_PORT_ATTR(_name) \
426         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
427
428 SIMPLE_PORT_ATTR(promiscuous);
429 SIMPLE_PORT_ATTR(num_mcast);
430 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
431 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
432 CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
433
434 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
435 static struct attribute *veth_port_default_attrs[] = {
436         GET_PORT_ATTR(mac_addr),
437         GET_PORT_ATTR(lpar_map),
438         GET_PORT_ATTR(stopped_map),
439         GET_PORT_ATTR(promiscuous),
440         GET_PORT_ATTR(num_mcast),
441         NULL
442 };
443
444 static struct sysfs_ops veth_port_sysfs_ops = {
445         .show = veth_port_attribute_show
446 };
447
448 static struct kobj_type veth_port_ktype = {
449         .sysfs_ops      = &veth_port_sysfs_ops,
450         .default_attrs  = veth_port_default_attrs
451 };
452
453 /*
454  * LPAR connection code
455  */
456
457 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
458 {
459         schedule_work(&cnx->statemachine_wq);
460 }
461
462 static void veth_take_cap(struct veth_lpar_connection *cnx,
463                           struct veth_lpevent *event)
464 {
465         unsigned long flags;
466
467         spin_lock_irqsave(&cnx->lock, flags);
468         /* Receiving caps may mean the other end has just come up, so
469          * we need to reload the instance ID of the far end */
470         cnx->dst_inst =
471                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
472                                                   HvLpEvent_Type_VirtualLan);
473
474         if (cnx->state & VETH_STATE_GOTCAPS) {
475                 veth_error("Received a second capabilities from LPAR %d.\n",
476                            cnx->remote_lp);
477                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
478                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
479         } else {
480                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
481                 cnx->state |= VETH_STATE_GOTCAPS;
482                 veth_kick_statemachine(cnx);
483         }
484         spin_unlock_irqrestore(&cnx->lock, flags);
485 }
486
487 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
488                               struct veth_lpevent *event)
489 {
490         unsigned long flags;
491
492         spin_lock_irqsave(&cnx->lock, flags);
493         if (cnx->state & VETH_STATE_GOTCAPACK) {
494                 veth_error("Received a second capabilities ack from LPAR %d.\n",
495                            cnx->remote_lp);
496         } else {
497                 memcpy(&cnx->cap_ack_event, event,
498                        sizeof(&cnx->cap_ack_event));
499                 cnx->state |= VETH_STATE_GOTCAPACK;
500                 veth_kick_statemachine(cnx);
501         }
502         spin_unlock_irqrestore(&cnx->lock, flags);
503 }
504
505 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
506                                   struct veth_lpevent *event)
507 {
508         unsigned long flags;
509
510         spin_lock_irqsave(&cnx->lock, flags);
511         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
512
513         /* Avoid kicking the statemachine once we're shutdown.
514          * It's unnecessary and it could break veth_stop_connection(). */
515
516         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
517                 cnx->state |= VETH_STATE_RESET;
518                 veth_kick_statemachine(cnx);
519         }
520         spin_unlock_irqrestore(&cnx->lock, flags);
521 }
522
523 static void veth_handle_ack(struct veth_lpevent *event)
524 {
525         HvLpIndex rlp = event->base_event.xTargetLp;
526         struct veth_lpar_connection *cnx = veth_cnx[rlp];
527
528         BUG_ON(! cnx);
529
530         switch (event->base_event.xSubtype) {
531         case VETH_EVENT_CAP:
532                 veth_take_cap_ack(cnx, event);
533                 break;
534         case VETH_EVENT_MONITOR:
535                 veth_take_monitor_ack(cnx, event);
536                 break;
537         default:
538                 veth_error("Unknown ack type %d from LPAR %d.\n",
539                                 event->base_event.xSubtype, rlp);
540         };
541 }
542
543 static void veth_handle_int(struct veth_lpevent *event)
544 {
545         HvLpIndex rlp = event->base_event.xSourceLp;
546         struct veth_lpar_connection *cnx = veth_cnx[rlp];
547         unsigned long flags;
548         int i, acked = 0;
549
550         BUG_ON(! cnx);
551
552         switch (event->base_event.xSubtype) {
553         case VETH_EVENT_CAP:
554                 veth_take_cap(cnx, event);
555                 break;
556         case VETH_EVENT_MONITOR:
557                 /* do nothing... this'll hang out here til we're dead,
558                  * and the hypervisor will return it for us. */
559                 break;
560         case VETH_EVENT_FRAMES_ACK:
561                 spin_lock_irqsave(&cnx->lock, flags);
562
563                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
564                         u16 msgnum = event->u.frames_ack_data.token[i];
565
566                         if (msgnum < VETH_NUMBUFFERS) {
567                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
568                                 cnx->outstanding_tx--;
569                                 acked++;
570                         }
571                 }
572
573                 if (acked > 0) {
574                         cnx->last_contact = jiffies;
575                         veth_wake_queues(cnx);
576                 }
577
578                 spin_unlock_irqrestore(&cnx->lock, flags);
579                 break;
580         case VETH_EVENT_FRAMES:
581                 veth_receive(cnx, event);
582                 break;
583         default:
584                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
585                                 event->base_event.xSubtype, rlp);
586         };
587 }
588
589 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
590 {
591         struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
592
593         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
594                 veth_handle_ack(veth_event);
595         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
596                 veth_handle_int(veth_event);
597 }
598
599 static int veth_process_caps(struct veth_lpar_connection *cnx)
600 {
601         struct veth_cap_data *remote_caps = &cnx->remote_caps;
602         int num_acks_needed;
603
604         /* Convert timer to jiffies */
605         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
606
607         if ( (remote_caps->num_buffers == 0)
608              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
609              || (remote_caps->ack_threshold == 0)
610              || (cnx->ack_timeout == 0) ) {
611                 veth_error("Received incompatible capabilities from LPAR %d.\n",
612                                 cnx->remote_lp);
613                 return HvLpEvent_Rc_InvalidSubtypeData;
614         }
615
616         num_acks_needed = (remote_caps->num_buffers
617                            / remote_caps->ack_threshold) + 1;
618
619         /* FIXME: locking on num_ack_events? */
620         if (cnx->num_ack_events < num_acks_needed) {
621                 int num;
622
623                 num = veth_allocate_events(cnx->remote_lp,
624                                            num_acks_needed-cnx->num_ack_events);
625                 if (num > 0)
626                         cnx->num_ack_events += num;
627
628                 if (cnx->num_ack_events < num_acks_needed) {
629                         veth_error("Couldn't allocate enough ack events "
630                                         "for LPAR %d.\n", cnx->remote_lp);
631
632                         return HvLpEvent_Rc_BufferNotAvailable;
633                 }
634         }
635
636
637         return HvLpEvent_Rc_Good;
638 }
639
640 /* FIXME: The gotos here are a bit dubious */
641 static void veth_statemachine(void *p)
642 {
643         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
644         int rlp = cnx->remote_lp;
645         int rc;
646
647         spin_lock_irq(&cnx->lock);
648
649  restart:
650         if (cnx->state & VETH_STATE_RESET) {
651                 if (cnx->state & VETH_STATE_OPEN)
652                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
653                                                      HvLpEvent_Type_VirtualLan);
654
655                 /*
656                  * Reset ack data. This prevents the ack_timer actually
657                  * doing anything, even if it runs one more time when
658                  * we drop the lock below.
659                  */
660                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
661                 cnx->num_pending_acks = 0;
662
663                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
664                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
665                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
666                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
667
668                 /* Clean up any leftover messages */
669                 if (cnx->msgs) {
670                         int i;
671                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
672                                 veth_recycle_msg(cnx, cnx->msgs + i);
673                 }
674
675                 cnx->outstanding_tx = 0;
676                 veth_wake_queues(cnx);
677
678                 /* Drop the lock so we can do stuff that might sleep or
679                  * take other locks. */
680                 spin_unlock_irq(&cnx->lock);
681
682                 del_timer_sync(&cnx->ack_timer);
683                 del_timer_sync(&cnx->reset_timer);
684
685                 spin_lock_irq(&cnx->lock);
686
687                 if (cnx->state & VETH_STATE_RESET)
688                         goto restart;
689
690                 /* Hack, wait for the other end to reset itself. */
691                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
692                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
693                         goto out;
694                 }
695         }
696
697         if (cnx->state & VETH_STATE_SHUTDOWN)
698                 /* It's all over, do nothing */
699                 goto out;
700
701         if ( !(cnx->state & VETH_STATE_OPEN) ) {
702                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
703                         goto cant_cope;
704
705                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
706                 cnx->src_inst =
707                         HvCallEvent_getSourceLpInstanceId(rlp,
708                                                           HvLpEvent_Type_VirtualLan);
709                 cnx->dst_inst =
710                         HvCallEvent_getTargetLpInstanceId(rlp,
711                                                           HvLpEvent_Type_VirtualLan);
712                 cnx->state |= VETH_STATE_OPEN;
713         }
714
715         if ( (cnx->state & VETH_STATE_OPEN)
716              && !(cnx->state & VETH_STATE_SENTMON) ) {
717                 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
718                                       HvLpEvent_AckInd_DoAck,
719                                       HvLpEvent_AckType_DeferredAck,
720                                       0, 0, 0, 0, 0, 0);
721
722                 if (rc == HvLpEvent_Rc_Good) {
723                         cnx->state |= VETH_STATE_SENTMON;
724                 } else {
725                         if ( (rc != HvLpEvent_Rc_PartitionDead)
726                              && (rc != HvLpEvent_Rc_PathClosed) )
727                                 veth_error("Error sending monitor to LPAR %d, "
728                                                 "rc = %d\n", rlp, rc);
729
730                         /* Oh well, hope we get a cap from the other
731                          * end and do better when that kicks us */
732                         goto out;
733                 }
734         }
735
736         if ( (cnx->state & VETH_STATE_OPEN)
737              && !(cnx->state & VETH_STATE_SENTCAPS)) {
738                 u64 *rawcap = (u64 *)&cnx->local_caps;
739
740                 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
741                                       HvLpEvent_AckInd_DoAck,
742                                       HvLpEvent_AckType_ImmediateAck,
743                                       0, rawcap[0], rawcap[1], rawcap[2],
744                                       rawcap[3], rawcap[4]);
745
746                 if (rc == HvLpEvent_Rc_Good) {
747                         cnx->state |= VETH_STATE_SENTCAPS;
748                 } else {
749                         if ( (rc != HvLpEvent_Rc_PartitionDead)
750                              && (rc != HvLpEvent_Rc_PathClosed) )
751                                 veth_error("Error sending caps to LPAR %d, "
752                                                 "rc = %d\n", rlp, rc);
753
754                         /* Oh well, hope we get a cap from the other
755                          * end and do better when that kicks us */
756                         goto out;
757                 }
758         }
759
760         if ((cnx->state & VETH_STATE_GOTCAPS)
761             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
762                 struct veth_cap_data *remote_caps = &cnx->remote_caps;
763
764                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
765                        sizeof(*remote_caps));
766
767                 spin_unlock_irq(&cnx->lock);
768                 rc = veth_process_caps(cnx);
769                 spin_lock_irq(&cnx->lock);
770
771                 /* We dropped the lock, so recheck for anything which
772                  * might mess us up */
773                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
774                         goto restart;
775
776                 cnx->cap_event.base_event.xRc = rc;
777                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
778                 if (rc == HvLpEvent_Rc_Good)
779                         cnx->state |= VETH_STATE_SENTCAPACK;
780                 else
781                         goto cant_cope;
782         }
783
784         if ((cnx->state & VETH_STATE_GOTCAPACK)
785             && (cnx->state & VETH_STATE_GOTCAPS)
786             && !(cnx->state & VETH_STATE_READY)) {
787                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
788                         /* Start the ACK timer */
789                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
790                         add_timer(&cnx->ack_timer);
791                         cnx->state |= VETH_STATE_READY;
792                 } else {
793                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
794                                         rlp, cnx->cap_ack_event.base_event.xRc);
795                         goto cant_cope;
796                 }
797         }
798
799  out:
800         spin_unlock_irq(&cnx->lock);
801         return;
802
803  cant_cope:
804         /* FIXME: we get here if something happens we really can't
805          * cope with.  The link will never work once we get here, and
806          * all we can do is not lock the rest of the system up */
807         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
808                         " (state = 0x%04lx)\n", rlp, cnx->state);
809         cnx->state |= VETH_STATE_SHUTDOWN;
810         spin_unlock_irq(&cnx->lock);
811 }
812
813 static int veth_init_connection(u8 rlp)
814 {
815         struct veth_lpar_connection *cnx;
816         struct veth_msg *msgs;
817         int i, rc;
818
819         if ( (rlp == this_lp)
820              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
821                 return 0;
822
823         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
824         if (! cnx)
825                 return -ENOMEM;
826         memset(cnx, 0, sizeof(*cnx));
827
828         cnx->remote_lp = rlp;
829         spin_lock_init(&cnx->lock);
830         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
831
832         init_timer(&cnx->ack_timer);
833         cnx->ack_timer.function = veth_timed_ack;
834         cnx->ack_timer.data = (unsigned long) cnx;
835
836         init_timer(&cnx->reset_timer);
837         cnx->reset_timer.function = veth_timed_reset;
838         cnx->reset_timer.data = (unsigned long) cnx;
839         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
840
841         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
842
843         veth_cnx[rlp] = cnx;
844
845         /* This gets us 1 reference, which is held on behalf of the driver
846          * infrastructure. It's released at module unload. */
847         kobject_init(&cnx->kobject);
848         cnx->kobject.ktype = &veth_lpar_connection_ktype;
849         rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
850         if (rc != 0)
851                 return rc;
852
853         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
854         if (! msgs) {
855                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
856                 return -ENOMEM;
857         }
858
859         cnx->msgs = msgs;
860         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
861
862         for (i = 0; i < VETH_NUMBUFFERS; i++) {
863                 msgs[i].token = i;
864                 veth_stack_push(cnx, msgs + i);
865         }
866
867         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
868
869         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
870                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
871                 return -ENOMEM;
872         }
873
874         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
875         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
876         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
877
878         return 0;
879 }
880
881 static void veth_stop_connection(struct veth_lpar_connection *cnx)
882 {
883         if (!cnx)
884                 return;
885
886         spin_lock_irq(&cnx->lock);
887         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
888         veth_kick_statemachine(cnx);
889         spin_unlock_irq(&cnx->lock);
890
891         /* There's a slim chance the reset code has just queued the
892          * statemachine to run in five seconds. If so we need to cancel
893          * that and requeue the work to run now. */
894         if (cancel_delayed_work(&cnx->statemachine_wq)) {
895                 spin_lock_irq(&cnx->lock);
896                 veth_kick_statemachine(cnx);
897                 spin_unlock_irq(&cnx->lock);
898         }
899
900         /* Wait for the state machine to run. */
901         flush_scheduled_work();
902 }
903
904 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
905 {
906         if (!cnx)
907                 return;
908
909         if (cnx->num_events > 0)
910                 mf_deallocate_lp_events(cnx->remote_lp,
911                                       HvLpEvent_Type_VirtualLan,
912                                       cnx->num_events,
913                                       NULL, NULL);
914         if (cnx->num_ack_events > 0)
915                 mf_deallocate_lp_events(cnx->remote_lp,
916                                       HvLpEvent_Type_VirtualLan,
917                                       cnx->num_ack_events,
918                                       NULL, NULL);
919
920         kfree(cnx->msgs);
921         veth_cnx[cnx->remote_lp] = NULL;
922         kfree(cnx);
923 }
924
925 static void veth_release_connection(struct kobject *kobj)
926 {
927         struct veth_lpar_connection *cnx;
928         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
929         veth_stop_connection(cnx);
930         veth_destroy_connection(cnx);
931 }
932
933 /*
934  * net_device code
935  */
936
937 static int veth_open(struct net_device *dev)
938 {
939         struct veth_port *port = (struct veth_port *) dev->priv;
940
941         memset(&port->stats, 0, sizeof (port->stats));
942         netif_start_queue(dev);
943         return 0;
944 }
945
946 static int veth_close(struct net_device *dev)
947 {
948         netif_stop_queue(dev);
949         return 0;
950 }
951
952 static struct net_device_stats *veth_get_stats(struct net_device *dev)
953 {
954         struct veth_port *port = (struct veth_port *) dev->priv;
955
956         return &port->stats;
957 }
958
959 static int veth_change_mtu(struct net_device *dev, int new_mtu)
960 {
961         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
962                 return -EINVAL;
963         dev->mtu = new_mtu;
964         return 0;
965 }
966
967 static void veth_set_multicast_list(struct net_device *dev)
968 {
969         struct veth_port *port = (struct veth_port *) dev->priv;
970         unsigned long flags;
971
972         write_lock_irqsave(&port->mcast_gate, flags);
973
974         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
975                         (dev->mc_count > VETH_MAX_MCAST)) {
976                 port->promiscuous = 1;
977         } else {
978                 struct dev_mc_list *dmi = dev->mc_list;
979                 int i;
980
981                 port->promiscuous = 0;
982
983                 /* Update table */
984                 port->num_mcast = 0;
985
986                 for (i = 0; i < dev->mc_count; i++) {
987                         u8 *addr = dmi->dmi_addr;
988                         u64 xaddr = 0;
989
990                         if (addr[0] & 0x01) {/* multicast address? */
991                                 memcpy(&xaddr, addr, ETH_ALEN);
992                                 port->mcast_addr[port->num_mcast] = xaddr;
993                                 port->num_mcast++;
994                         }
995                         dmi = dmi->next;
996                 }
997         }
998
999         write_unlock_irqrestore(&port->mcast_gate, flags);
1000 }
1001
1002 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1003 {
1004         strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
1005         info->driver[sizeof(info->driver) - 1] = '\0';
1006         strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
1007         info->version[sizeof(info->version) - 1] = '\0';
1008 }
1009
1010 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1011 {
1012         ecmd->supported = (SUPPORTED_1000baseT_Full
1013                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1014         ecmd->advertising = (SUPPORTED_1000baseT_Full
1015                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1016         ecmd->port = PORT_FIBRE;
1017         ecmd->transceiver = XCVR_INTERNAL;
1018         ecmd->phy_address = 0;
1019         ecmd->speed = SPEED_1000;
1020         ecmd->duplex = DUPLEX_FULL;
1021         ecmd->autoneg = AUTONEG_ENABLE;
1022         ecmd->maxtxpkt = 120;
1023         ecmd->maxrxpkt = 120;
1024         return 0;
1025 }
1026
1027 static u32 veth_get_link(struct net_device *dev)
1028 {
1029         return 1;
1030 }
1031
1032 static struct ethtool_ops ops = {
1033         .get_drvinfo = veth_get_drvinfo,
1034         .get_settings = veth_get_settings,
1035         .get_link = veth_get_link,
1036 };
1037
1038 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
1039 {
1040         struct net_device *dev;
1041         struct veth_port *port;
1042         int i, rc;
1043
1044         dev = alloc_etherdev(sizeof (struct veth_port));
1045         if (! dev) {
1046                 veth_error("Unable to allocate net_device structure!\n");
1047                 return NULL;
1048         }
1049
1050         port = (struct veth_port *) dev->priv;
1051
1052         spin_lock_init(&port->queue_lock);
1053         rwlock_init(&port->mcast_gate);
1054         port->stopped_map = 0;
1055
1056         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1057                 HvLpVirtualLanIndexMap map;
1058
1059                 if (i == this_lp)
1060                         continue;
1061                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1062                 if (map & (0x8000 >> vlan))
1063                         port->lpar_map |= (1 << i);
1064         }
1065         port->dev = vdev;
1066
1067         dev->dev_addr[0] = 0x02;
1068         dev->dev_addr[1] = 0x01;
1069         dev->dev_addr[2] = 0xff;
1070         dev->dev_addr[3] = vlan;
1071         dev->dev_addr[4] = 0xff;
1072         dev->dev_addr[5] = this_lp;
1073
1074         dev->mtu = VETH_MAX_MTU;
1075
1076         memcpy(&port->mac_addr, dev->dev_addr, 6);
1077
1078         dev->open = veth_open;
1079         dev->hard_start_xmit = veth_start_xmit;
1080         dev->stop = veth_close;
1081         dev->get_stats = veth_get_stats;
1082         dev->change_mtu = veth_change_mtu;
1083         dev->set_mac_address = NULL;
1084         dev->set_multicast_list = veth_set_multicast_list;
1085         SET_ETHTOOL_OPS(dev, &ops);
1086
1087         SET_NETDEV_DEV(dev, vdev);
1088
1089         rc = register_netdev(dev);
1090         if (rc != 0) {
1091                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1092                 free_netdev(dev);
1093                 return NULL;
1094         }
1095
1096         kobject_init(&port->kobject);
1097         port->kobject.parent = &dev->class_dev.kobj;
1098         port->kobject.ktype  = &veth_port_ktype;
1099         kobject_set_name(&port->kobject, "veth_port");
1100         if (0 != kobject_add(&port->kobject))
1101                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1102
1103         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1104                         dev->name, vlan, port->lpar_map);
1105
1106         return dev;
1107 }
1108
1109 /*
1110  * Tx path
1111  */
1112
1113 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1114                                 struct net_device *dev)
1115 {
1116         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1117         struct veth_port *port = (struct veth_port *) dev->priv;
1118         HvLpEvent_Rc rc;
1119         struct veth_msg *msg = NULL;
1120         unsigned long flags;
1121
1122         if (! cnx)
1123                 return 0;
1124
1125         spin_lock_irqsave(&cnx->lock, flags);
1126
1127         if (! (cnx->state & VETH_STATE_READY))
1128                 goto no_error;
1129
1130         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1131                 goto drop;
1132
1133         msg = veth_stack_pop(cnx);
1134         if (! msg)
1135                 goto drop;
1136
1137         msg->in_use = 1;
1138         msg->skb = skb_get(skb);
1139
1140         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1141                                 skb->len, DMA_TO_DEVICE);
1142
1143         if (dma_mapping_error(msg->data.addr[0]))
1144                 goto recycle_and_drop;
1145
1146         msg->dev = port->dev;
1147         msg->data.len[0] = skb->len;
1148         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1149
1150         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1151
1152         if (rc != HvLpEvent_Rc_Good)
1153                 goto recycle_and_drop;
1154
1155         /* If the timer's not already running, start it now. */
1156         if (0 == cnx->outstanding_tx)
1157                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1158
1159         cnx->last_contact = jiffies;
1160         cnx->outstanding_tx++;
1161
1162         if (veth_stack_is_empty(cnx))
1163                 veth_stop_queues(cnx);
1164
1165  no_error:
1166         spin_unlock_irqrestore(&cnx->lock, flags);
1167         return 0;
1168
1169  recycle_and_drop:
1170         veth_recycle_msg(cnx, msg);
1171  drop:
1172         spin_unlock_irqrestore(&cnx->lock, flags);
1173         return 1;
1174 }
1175
1176 static void veth_transmit_to_many(struct sk_buff *skb,
1177                                           HvLpIndexMap lpmask,
1178                                           struct net_device *dev)
1179 {
1180         struct veth_port *port = (struct veth_port *) dev->priv;
1181         int i, success, error;
1182
1183         success = error = 0;
1184
1185         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1186                 if ((lpmask & (1 << i)) == 0)
1187                         continue;
1188
1189                 if (veth_transmit_to_one(skb, i, dev))
1190                         error = 1;
1191                 else
1192                         success = 1;
1193         }
1194
1195         if (error)
1196                 port->stats.tx_errors++;
1197
1198         if (success) {
1199                 port->stats.tx_packets++;
1200                 port->stats.tx_bytes += skb->len;
1201         }
1202 }
1203
1204 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1205 {
1206         unsigned char *frame = skb->data;
1207         struct veth_port *port = (struct veth_port *) dev->priv;
1208         HvLpIndexMap lpmask;
1209
1210         if (! (frame[0] & 0x01)) {
1211                 /* unicast packet */
1212                 HvLpIndex rlp = frame[5];
1213
1214                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1215                         dev_kfree_skb(skb);
1216                         return 0;
1217                 }
1218
1219                 lpmask = 1 << rlp;
1220         } else {
1221                 lpmask = port->lpar_map;
1222         }
1223
1224         veth_transmit_to_many(skb, lpmask, dev);
1225
1226         dev_kfree_skb(skb);
1227
1228         return 0;
1229 }
1230
1231 /* You must hold the connection's lock when you call this function. */
1232 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1233                              struct veth_msg *msg)
1234 {
1235         u32 dma_address, dma_length;
1236
1237         if (msg->in_use) {
1238                 msg->in_use = 0;
1239                 dma_address = msg->data.addr[0];
1240                 dma_length = msg->data.len[0];
1241
1242                 if (!dma_mapping_error(dma_address))
1243                         dma_unmap_single(msg->dev, dma_address, dma_length,
1244                                         DMA_TO_DEVICE);
1245
1246                 if (msg->skb) {
1247                         dev_kfree_skb_any(msg->skb);
1248                         msg->skb = NULL;
1249                 }
1250
1251                 memset(&msg->data, 0, sizeof(msg->data));
1252                 veth_stack_push(cnx, msg);
1253         } else if (cnx->state & VETH_STATE_OPEN) {
1254                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1255                                 cnx->remote_lp, msg->token);
1256         }
1257 }
1258
1259 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1260 {
1261         int i;
1262
1263         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1264                 struct net_device *dev = veth_dev[i];
1265                 struct veth_port *port;
1266                 unsigned long flags;
1267
1268                 if (! dev)
1269                         continue;
1270
1271                 port = (struct veth_port *)dev->priv;
1272
1273                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1274                         continue;
1275
1276                 spin_lock_irqsave(&port->queue_lock, flags);
1277
1278                 port->stopped_map &= ~(1 << cnx->remote_lp);
1279
1280                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1281                         veth_debug("cnx %d: woke queue for %s.\n",
1282                                         cnx->remote_lp, dev->name);
1283                         netif_wake_queue(dev);
1284                 }
1285                 spin_unlock_irqrestore(&port->queue_lock, flags);
1286         }
1287 }
1288
1289 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1290 {
1291         int i;
1292
1293         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1294                 struct net_device *dev = veth_dev[i];
1295                 struct veth_port *port;
1296
1297                 if (! dev)
1298                         continue;
1299
1300                 port = (struct veth_port *)dev->priv;
1301
1302                 /* If this cnx is not on the vlan for this port, continue */
1303                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1304                         continue;
1305
1306                 spin_lock(&port->queue_lock);
1307
1308                 netif_stop_queue(dev);
1309                 port->stopped_map |= (1 << cnx->remote_lp);
1310
1311                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1312                                 cnx->remote_lp, dev->name, port->stopped_map);
1313
1314                 spin_unlock(&port->queue_lock);
1315         }
1316 }
1317
1318 static void veth_timed_reset(unsigned long ptr)
1319 {
1320         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1321         unsigned long trigger_time, flags;
1322
1323         /* FIXME is it possible this fires after veth_stop_connection()?
1324          * That would reschedule the statemachine for 5 seconds and probably
1325          * execute it after the module's been unloaded. Hmm. */
1326
1327         spin_lock_irqsave(&cnx->lock, flags);
1328
1329         if (cnx->outstanding_tx > 0) {
1330                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1331
1332                 if (trigger_time < jiffies) {
1333                         cnx->state |= VETH_STATE_RESET;
1334                         veth_kick_statemachine(cnx);
1335                         veth_error("%d packets not acked by LPAR %d within %d "
1336                                         "seconds, resetting.\n",
1337                                         cnx->outstanding_tx, cnx->remote_lp,
1338                                         cnx->reset_timeout / HZ);
1339                 } else {
1340                         /* Reschedule the timer */
1341                         trigger_time = jiffies + cnx->reset_timeout;
1342                         mod_timer(&cnx->reset_timer, trigger_time);
1343                 }
1344         }
1345
1346         spin_unlock_irqrestore(&cnx->lock, flags);
1347 }
1348
1349 /*
1350  * Rx path
1351  */
1352
1353 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1354 {
1355         int wanted = 0;
1356         int i;
1357         unsigned long flags;
1358
1359         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1360                 return 1;
1361
1362         read_lock_irqsave(&port->mcast_gate, flags);
1363
1364         if (port->promiscuous) {
1365                 wanted = 1;
1366                 goto out;
1367         }
1368
1369         for (i = 0; i < port->num_mcast; ++i) {
1370                 if (port->mcast_addr[i] == mac_addr) {
1371                         wanted = 1;
1372                         break;
1373                 }
1374         }
1375
1376  out:
1377         read_unlock_irqrestore(&port->mcast_gate, flags);
1378
1379         return wanted;
1380 }
1381
1382 struct dma_chunk {
1383         u64 addr;
1384         u64 size;
1385 };
1386
1387 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1388
1389 static inline void veth_build_dma_list(struct dma_chunk *list,
1390                                        unsigned char *p, unsigned long length)
1391 {
1392         unsigned long done;
1393         int i = 1;
1394
1395         /* FIXME: skbs are continguous in real addresses.  Do we
1396          * really need to break it into PAGE_SIZE chunks, or can we do
1397          * it just at the granularity of iSeries real->absolute
1398          * mapping?  Indeed, given the way the allocator works, can we
1399          * count on them being absolutely contiguous? */
1400         list[0].addr = iseries_hv_addr(p);
1401         list[0].size = min(length,
1402                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1403
1404         done = list[0].size;
1405         while (done < length) {
1406                 list[i].addr = iseries_hv_addr(p + done);
1407                 list[i].size = min(length-done, PAGE_SIZE);
1408                 done += list[i].size;
1409                 i++;
1410         }
1411 }
1412
1413 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1414 {
1415         HvLpEvent_Rc rc;
1416
1417         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1418                              0, &cnx->pending_acks);
1419
1420         if (rc != HvLpEvent_Rc_Good)
1421                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1422                                 cnx->remote_lp, (int)rc);
1423
1424         cnx->num_pending_acks = 0;
1425         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1426 }
1427
1428 static void veth_receive(struct veth_lpar_connection *cnx,
1429                          struct veth_lpevent *event)
1430 {
1431         struct veth_frames_data *senddata = &event->u.frames_data;
1432         int startchunk = 0;
1433         int nchunks;
1434         unsigned long flags;
1435         HvLpDma_Rc rc;
1436
1437         do {
1438                 u16 length = 0;
1439                 struct sk_buff *skb;
1440                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1441                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1442                 u64 dest;
1443                 HvLpVirtualLanIndex vlan;
1444                 struct net_device *dev;
1445                 struct veth_port *port;
1446
1447                 /* FIXME: do we need this? */
1448                 memset(local_list, 0, sizeof(local_list));
1449                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1450
1451                 /* a 0 address marks the end of the valid entries */
1452                 if (senddata->addr[startchunk] == 0)
1453                         break;
1454
1455                 /* make sure that we have at least 1 EOF entry in the
1456                  * remaining entries */
1457                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1458                         veth_error("Missing EOF fragment in event "
1459                                         "eofmask = 0x%x startchunk = %d\n",
1460                                         (unsigned)senddata->eofmask,
1461                                         startchunk);
1462                         break;
1463                 }
1464
1465                 /* build list of chunks in this frame */
1466                 nchunks = 0;
1467                 do {
1468                         remote_list[nchunks].addr =
1469                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1470                         remote_list[nchunks].size =
1471                                 senddata->len[startchunk+nchunks];
1472                         length += remote_list[nchunks].size;
1473                 } while (! (senddata->eofmask &
1474                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1475
1476                 /* length == total length of all chunks */
1477                 /* nchunks == # of chunks in this frame */
1478
1479                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1480                         veth_error("Received oversize frame from LPAR %d "
1481                                         "(length = %d)\n",
1482                                         cnx->remote_lp, length);
1483                         continue;
1484                 }
1485
1486                 skb = alloc_skb(length, GFP_ATOMIC);
1487                 if (!skb)
1488                         continue;
1489
1490                 veth_build_dma_list(local_list, skb->data, length);
1491
1492                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1493                                             event->base_event.xSourceLp,
1494                                             HvLpDma_Direction_RemoteToLocal,
1495                                             cnx->src_inst,
1496                                             cnx->dst_inst,
1497                                             HvLpDma_AddressType_RealAddress,
1498                                             HvLpDma_AddressType_TceIndex,
1499                                             iseries_hv_addr(&local_list),
1500                                             iseries_hv_addr(&remote_list),
1501                                             length);
1502                 if (rc != HvLpDma_Rc_Good) {
1503                         dev_kfree_skb_irq(skb);
1504                         continue;
1505                 }
1506
1507                 vlan = skb->data[9];
1508                 dev = veth_dev[vlan];
1509                 if (! dev) {
1510                         /*
1511                          * Some earlier versions of the driver sent
1512                          * broadcasts down all connections, even to lpars
1513                          * that weren't on the relevant vlan. So ignore
1514                          * packets belonging to a vlan we're not on.
1515                          * We can also be here if we receive packets while
1516                          * the driver is going down, because then dev is NULL.
1517                          */
1518                         dev_kfree_skb_irq(skb);
1519                         continue;
1520                 }
1521
1522                 port = (struct veth_port *)dev->priv;
1523                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1524
1525                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1526                         dev_kfree_skb_irq(skb);
1527                         continue;
1528                 }
1529                 if (! veth_frame_wanted(port, dest)) {
1530                         dev_kfree_skb_irq(skb);
1531                         continue;
1532                 }
1533
1534                 skb_put(skb, length);
1535                 skb->dev = dev;
1536                 skb->protocol = eth_type_trans(skb, dev);
1537                 skb->ip_summed = CHECKSUM_NONE;
1538                 netif_rx(skb);  /* send it up */
1539                 port->stats.rx_packets++;
1540                 port->stats.rx_bytes += length;
1541         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1542
1543         /* Ack it */
1544         spin_lock_irqsave(&cnx->lock, flags);
1545         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1546
1547         cnx->pending_acks[cnx->num_pending_acks++] =
1548                 event->base_event.xCorrelationToken;
1549
1550         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1551              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1552                 veth_flush_acks(cnx);
1553
1554         spin_unlock_irqrestore(&cnx->lock, flags);
1555 }
1556
1557 static void veth_timed_ack(unsigned long ptr)
1558 {
1559         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1560         unsigned long flags;
1561
1562         /* Ack all the events */
1563         spin_lock_irqsave(&cnx->lock, flags);
1564         if (cnx->num_pending_acks > 0)
1565                 veth_flush_acks(cnx);
1566
1567         /* Reschedule the timer */
1568         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1569         add_timer(&cnx->ack_timer);
1570         spin_unlock_irqrestore(&cnx->lock, flags);
1571 }
1572
1573 static int veth_remove(struct vio_dev *vdev)
1574 {
1575         struct veth_lpar_connection *cnx;
1576         struct net_device *dev;
1577         struct veth_port *port;
1578         int i;
1579
1580         dev = veth_dev[vdev->unit_address];
1581
1582         if (! dev)
1583                 return 0;
1584
1585         port = netdev_priv(dev);
1586
1587         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1588                 cnx = veth_cnx[i];
1589
1590                 if (cnx && (port->lpar_map & (1 << i))) {
1591                         /* Drop our reference to connections on our VLAN */
1592                         kobject_put(&cnx->kobject);
1593                 }
1594         }
1595
1596         veth_dev[vdev->unit_address] = NULL;
1597         kobject_del(&port->kobject);
1598         kobject_put(&port->kobject);
1599         unregister_netdev(dev);
1600         free_netdev(dev);
1601
1602         return 0;
1603 }
1604
1605 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1606 {
1607         int i = vdev->unit_address;
1608         struct net_device *dev;
1609         struct veth_port *port;
1610
1611         dev = veth_probe_one(i, &vdev->dev);
1612         if (dev == NULL) {
1613                 veth_remove(vdev);
1614                 return 1;
1615         }
1616         veth_dev[i] = dev;
1617
1618         port = (struct veth_port*)netdev_priv(dev);
1619
1620         /* Start the state machine on each connection on this vlan. If we're
1621          * the first dev to do so this will commence link negotiation */
1622         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1623                 struct veth_lpar_connection *cnx;
1624
1625                 if (! (port->lpar_map & (1 << i)))
1626                         continue;
1627
1628                 cnx = veth_cnx[i];
1629                 if (!cnx)
1630                         continue;
1631
1632                 kobject_get(&cnx->kobject);
1633                 veth_kick_statemachine(cnx);
1634         }
1635
1636         return 0;
1637 }
1638
1639 /**
1640  * veth_device_table: Used by vio.c to match devices that we
1641  * support.
1642  */
1643 static struct vio_device_id veth_device_table[] __devinitdata = {
1644         { "vlan", "" },
1645         { "", "" }
1646 };
1647 MODULE_DEVICE_TABLE(vio, veth_device_table);
1648
1649 static struct vio_driver veth_driver = {
1650         .id_table = veth_device_table,
1651         .probe = veth_probe,
1652         .remove = veth_remove,
1653         .driver = {
1654                 .name = DRV_NAME,
1655                 .owner = THIS_MODULE,
1656         }
1657 };
1658
1659 /*
1660  * Module initialization/cleanup
1661  */
1662
1663 void __exit veth_module_cleanup(void)
1664 {
1665         int i;
1666         struct veth_lpar_connection *cnx;
1667
1668         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1669         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1670
1671         /* Make sure any work queued from Hypervisor callbacks is finished. */
1672         flush_scheduled_work();
1673
1674         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1675                 cnx = veth_cnx[i];
1676
1677                 if (!cnx)
1678                         continue;
1679
1680                 /* Remove the connection from sysfs */
1681                 kobject_del(&cnx->kobject);
1682                 /* Drop the driver's reference to the connection */
1683                 kobject_put(&cnx->kobject);
1684         }
1685
1686         /* Unregister the driver, which will close all the netdevs and stop
1687          * the connections when they're no longer referenced. */
1688         vio_unregister_driver(&veth_driver);
1689 }
1690 module_exit(veth_module_cleanup);
1691
1692 int __init veth_module_init(void)
1693 {
1694         int i;
1695         int rc;
1696
1697         this_lp = HvLpConfig_getLpIndex_outline();
1698
1699         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1700                 rc = veth_init_connection(i);
1701                 if (rc != 0)
1702                         goto error;
1703         }
1704
1705         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1706                                   &veth_handle_event);
1707
1708         rc = vio_register_driver(&veth_driver);
1709         if (rc != 0)
1710                 goto error;
1711
1712         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1713                 struct kobject *kobj;
1714
1715                 if (!veth_cnx[i])
1716                         continue;
1717
1718                 kobj = &veth_cnx[i]->kobject;
1719                 kobj->parent = &veth_driver.driver.kobj;
1720                 /* If the add failes, complain but otherwise continue */
1721                 if (0 != kobject_add(kobj))
1722                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1723         }
1724
1725         return 0;
1726
1727 error:
1728         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1729                 veth_destroy_connection(veth_cnx[i]);
1730         }
1731
1732         return rc;
1733 }
1734 module_init(veth_module_init);