d86d8f055a6cd7d12330237baec71550230e4609
[linux-2.6.git] / 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/version.h>
62 #include <linux/types.h>
63 #include <linux/errno.h>
64 #include <linux/ioport.h>
65 #include <linux/kernel.h>
66 #include <linux/netdevice.h>
67 #include <linux/etherdevice.h>
68 #include <linux/skbuff.h>
69 #include <linux/init.h>
70 #include <linux/delay.h>
71 #include <linux/mm.h>
72 #include <linux/ethtool.h>
73
74 #include <asm/abs_addr.h>
75 #include <asm/iseries/mf.h>
76 #include <asm/uaccess.h>
77
78 #include <asm/iseries/hv_lp_config.h>
79 #include <asm/iseries/hv_types.h>
80 #include <asm/iseries/hv_lp_event.h>
81 #include <asm/iommu.h>
82 #include <asm/vio.h>
83
84 #undef DEBUG
85
86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88 MODULE_LICENSE("GPL");
89
90 #define VETH_EVENT_CAP  (0)
91 #define VETH_EVENT_FRAMES       (1)
92 #define VETH_EVENT_MONITOR      (2)
93 #define VETH_EVENT_FRAMES_ACK   (3)
94
95 #define VETH_MAX_ACKS_PER_MSG   (20)
96 #define VETH_MAX_FRAMES_PER_MSG (6)
97
98 struct veth_frames_data {
99         u32 addr[VETH_MAX_FRAMES_PER_MSG];
100         u16 len[VETH_MAX_FRAMES_PER_MSG];
101         u32 eofmask;
102 };
103 #define VETH_EOF_SHIFT          (32-VETH_MAX_FRAMES_PER_MSG)
104
105 struct veth_frames_ack_data {
106         u16 token[VETH_MAX_ACKS_PER_MSG];
107 };
108
109 struct veth_cap_data {
110         u8 caps_version;
111         u8 rsvd1;
112         u16 num_buffers;
113         u16 ack_threshold;
114         u16 rsvd2;
115         u32 ack_timeout;
116         u32 rsvd3;
117         u64 rsvd4[3];
118 };
119
120 struct veth_lpevent {
121         struct HvLpEvent base_event;
122         union {
123                 struct veth_cap_data caps_data;
124                 struct veth_frames_data frames_data;
125                 struct veth_frames_ack_data frames_ack_data;
126         } u;
127
128 };
129
130 #define DRV_NAME        "iseries_veth"
131 #define DRV_VERSION     "2.0"
132
133 #define VETH_NUMBUFFERS         (120)
134 #define VETH_ACKTIMEOUT         (1000000) /* microseconds */
135 #define VETH_MAX_MCAST          (12)
136
137 #define VETH_MAX_MTU            (9000)
138
139 #if VETH_NUMBUFFERS < 10
140 #define ACK_THRESHOLD           (1)
141 #elif VETH_NUMBUFFERS < 20
142 #define ACK_THRESHOLD           (4)
143 #elif VETH_NUMBUFFERS < 40
144 #define ACK_THRESHOLD           (10)
145 #else
146 #define ACK_THRESHOLD           (20)
147 #endif
148
149 #define VETH_STATE_SHUTDOWN     (0x0001)
150 #define VETH_STATE_OPEN         (0x0002)
151 #define VETH_STATE_RESET        (0x0004)
152 #define VETH_STATE_SENTMON      (0x0008)
153 #define VETH_STATE_SENTCAPS     (0x0010)
154 #define VETH_STATE_GOTCAPACK    (0x0020)
155 #define VETH_STATE_GOTCAPS      (0x0040)
156 #define VETH_STATE_SENTCAPACK   (0x0080)
157 #define VETH_STATE_READY        (0x0100)
158
159 struct veth_msg {
160         struct veth_msg *next;
161         struct veth_frames_data data;
162         int token;
163         int in_use;
164         struct sk_buff *skb;
165         struct device *dev;
166 };
167
168 struct veth_lpar_connection {
169         HvLpIndex remote_lp;
170         struct work_struct statemachine_wq;
171         struct veth_msg *msgs;
172         int num_events;
173         struct veth_cap_data local_caps;
174
175         struct kobject kobject;
176         struct timer_list ack_timer;
177
178         struct timer_list reset_timer;
179         unsigned int reset_timeout;
180         unsigned long last_contact;
181         int outstanding_tx;
182
183         spinlock_t lock;
184         unsigned long state;
185         HvLpInstanceId src_inst;
186         HvLpInstanceId dst_inst;
187         struct veth_lpevent cap_event, cap_ack_event;
188         u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189         u32 num_pending_acks;
190
191         int num_ack_events;
192         struct veth_cap_data remote_caps;
193         u32 ack_timeout;
194
195         struct veth_msg *msg_stack_head;
196 };
197
198 struct veth_port {
199         struct device *dev;
200         struct net_device_stats stats;
201         u64 mac_addr;
202         HvLpIndexMap lpar_map;
203
204         /* queue_lock protects the stopped_map and dev's queue. */
205         spinlock_t queue_lock;
206         HvLpIndexMap stopped_map;
207
208         /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
209         rwlock_t mcast_gate;
210         int promiscuous;
211         int num_mcast;
212         u64 mcast_addr[VETH_MAX_MCAST];
213
214         struct kobject kobject;
215 };
216
217 static HvLpIndex this_lp;
218 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
219 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
220
221 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
222 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
223 static void veth_wake_queues(struct veth_lpar_connection *cnx);
224 static void veth_stop_queues(struct veth_lpar_connection *cnx);
225 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
226 static void veth_release_connection(struct kobject *kobject);
227 static void veth_timed_ack(unsigned long ptr);
228 static void veth_timed_reset(unsigned long ptr);
229
230 /*
231  * Utility functions
232  */
233
234 #define veth_info(fmt, args...) \
235         printk(KERN_INFO DRV_NAME ": " fmt, ## args)
236
237 #define veth_error(fmt, args...) \
238         printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
239
240 #ifdef DEBUG
241 #define veth_debug(fmt, args...) \
242         printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
243 #else
244 #define veth_debug(fmt, args...) do {} while (0)
245 #endif
246
247 /* You must hold the connection's lock when you call this function. */
248 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
249                                    struct veth_msg *msg)
250 {
251         msg->next = cnx->msg_stack_head;
252         cnx->msg_stack_head = msg;
253 }
254
255 /* You must hold the connection's lock when you call this function. */
256 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
257 {
258         struct veth_msg *msg;
259
260         msg = cnx->msg_stack_head;
261         if (msg)
262                 cnx->msg_stack_head = cnx->msg_stack_head->next;
263
264         return msg;
265 }
266
267 /* You must hold the connection's lock when you call this function. */
268 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
269 {
270         return cnx->msg_stack_head == NULL;
271 }
272
273 static inline HvLpEvent_Rc
274 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
275                  HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
276                  u64 token,
277                  u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
278 {
279         return HvCallEvent_signalLpEventFast(cnx->remote_lp,
280                                              HvLpEvent_Type_VirtualLan,
281                                              subtype, ackind, acktype,
282                                              cnx->src_inst,
283                                              cnx->dst_inst,
284                                              token, data1, data2, data3,
285                                              data4, data5);
286 }
287
288 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
289                                            u16 subtype, u64 token, void *data)
290 {
291         u64 *p = (u64 *) data;
292
293         return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
294                                 HvLpEvent_AckType_ImmediateAck,
295                                 token, p[0], p[1], p[2], p[3], p[4]);
296 }
297
298 struct veth_allocation {
299         struct completion c;
300         int num;
301 };
302
303 static void veth_complete_allocation(void *parm, int number)
304 {
305         struct veth_allocation *vc = (struct veth_allocation *)parm;
306
307         vc->num = number;
308         complete(&vc->c);
309 }
310
311 static int veth_allocate_events(HvLpIndex rlp, int number)
312 {
313         struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
314
315         mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
316                             sizeof(struct veth_lpevent), number,
317                             &veth_complete_allocation, &vc);
318         wait_for_completion(&vc.c);
319
320         return vc.num;
321 }
322
323 /*
324  * sysfs support
325  */
326
327 struct veth_cnx_attribute {
328         struct attribute attr;
329         ssize_t (*show)(struct veth_lpar_connection *, char *buf);
330         ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
331 };
332
333 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334                 struct attribute *attr, char *buf)
335 {
336         struct veth_cnx_attribute *cnx_attr;
337         struct veth_lpar_connection *cnx;
338
339         cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
341
342         if (!cnx_attr->show)
343                 return -EIO;
344
345         return cnx_attr->show(cnx, buf);
346 }
347
348 #define CUSTOM_CNX_ATTR(_name, _format, _expression)                    \
349 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
350 {                                                                       \
351         return sprintf(buf, _format, _expression);                      \
352 }                                                                       \
353 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
354
355 #define SIMPLE_CNX_ATTR(_name)  \
356         CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
357
358 SIMPLE_CNX_ATTR(outstanding_tx);
359 SIMPLE_CNX_ATTR(remote_lp);
360 SIMPLE_CNX_ATTR(num_events);
361 SIMPLE_CNX_ATTR(src_inst);
362 SIMPLE_CNX_ATTR(dst_inst);
363 SIMPLE_CNX_ATTR(num_pending_acks);
364 SIMPLE_CNX_ATTR(num_ack_events);
365 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369                 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
370
371 #define GET_CNX_ATTR(_name)     (&veth_cnx_attr_##_name.attr)
372
373 static struct attribute *veth_cnx_default_attrs[] = {
374         GET_CNX_ATTR(outstanding_tx),
375         GET_CNX_ATTR(remote_lp),
376         GET_CNX_ATTR(num_events),
377         GET_CNX_ATTR(reset_timeout),
378         GET_CNX_ATTR(last_contact),
379         GET_CNX_ATTR(state),
380         GET_CNX_ATTR(src_inst),
381         GET_CNX_ATTR(dst_inst),
382         GET_CNX_ATTR(num_pending_acks),
383         GET_CNX_ATTR(num_ack_events),
384         GET_CNX_ATTR(ack_timeout),
385         NULL
386 };
387
388 static struct sysfs_ops veth_cnx_sysfs_ops = {
389                 .show = veth_cnx_attribute_show
390 };
391
392 static struct kobj_type veth_lpar_connection_ktype = {
393         .release        = veth_release_connection,
394         .sysfs_ops      = &veth_cnx_sysfs_ops,
395         .default_attrs  = veth_cnx_default_attrs
396 };
397
398 struct veth_port_attribute {
399         struct attribute attr;
400         ssize_t (*show)(struct veth_port *, char *buf);
401         ssize_t (*store)(struct veth_port *, const char *buf);
402 };
403
404 static ssize_t veth_port_attribute_show(struct kobject *kobj,
405                 struct attribute *attr, char *buf)
406 {
407         struct veth_port_attribute *port_attr;
408         struct veth_port *port;
409
410         port_attr = container_of(attr, struct veth_port_attribute, attr);
411         port = container_of(kobj, struct veth_port, kobject);
412
413         if (!port_attr->show)
414                 return -EIO;
415
416         return port_attr->show(port, buf);
417 }
418
419 #define CUSTOM_PORT_ATTR(_name, _format, _expression)                   \
420 static ssize_t _name##_show(struct veth_port *port, char *buf)          \
421 {                                                                       \
422         return sprintf(buf, _format, _expression);                      \
423 }                                                                       \
424 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
425
426 #define SIMPLE_PORT_ATTR(_name) \
427         CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
428
429 SIMPLE_PORT_ATTR(promiscuous);
430 SIMPLE_PORT_ATTR(num_mcast);
431 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
433 CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
434
435 #define GET_PORT_ATTR(_name)    (&veth_port_attr_##_name.attr)
436 static struct attribute *veth_port_default_attrs[] = {
437         GET_PORT_ATTR(mac_addr),
438         GET_PORT_ATTR(lpar_map),
439         GET_PORT_ATTR(stopped_map),
440         GET_PORT_ATTR(promiscuous),
441         GET_PORT_ATTR(num_mcast),
442         NULL
443 };
444
445 static struct sysfs_ops veth_port_sysfs_ops = {
446         .show = veth_port_attribute_show
447 };
448
449 static struct kobj_type veth_port_ktype = {
450         .sysfs_ops      = &veth_port_sysfs_ops,
451         .default_attrs  = veth_port_default_attrs
452 };
453
454 /*
455  * LPAR connection code
456  */
457
458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459 {
460         schedule_work(&cnx->statemachine_wq);
461 }
462
463 static void veth_take_cap(struct veth_lpar_connection *cnx,
464                           struct veth_lpevent *event)
465 {
466         unsigned long flags;
467
468         spin_lock_irqsave(&cnx->lock, flags);
469         /* Receiving caps may mean the other end has just come up, so
470          * we need to reload the instance ID of the far end */
471         cnx->dst_inst =
472                 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473                                                   HvLpEvent_Type_VirtualLan);
474
475         if (cnx->state & VETH_STATE_GOTCAPS) {
476                 veth_error("Received a second capabilities from LPAR %d.\n",
477                            cnx->remote_lp);
478                 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479                 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
480         } else {
481                 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482                 cnx->state |= VETH_STATE_GOTCAPS;
483                 veth_kick_statemachine(cnx);
484         }
485         spin_unlock_irqrestore(&cnx->lock, flags);
486 }
487
488 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
489                               struct veth_lpevent *event)
490 {
491         unsigned long flags;
492
493         spin_lock_irqsave(&cnx->lock, flags);
494         if (cnx->state & VETH_STATE_GOTCAPACK) {
495                 veth_error("Received a second capabilities ack from LPAR %d.\n",
496                            cnx->remote_lp);
497         } else {
498                 memcpy(&cnx->cap_ack_event, event,
499                        sizeof(&cnx->cap_ack_event));
500                 cnx->state |= VETH_STATE_GOTCAPACK;
501                 veth_kick_statemachine(cnx);
502         }
503         spin_unlock_irqrestore(&cnx->lock, flags);
504 }
505
506 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
507                                   struct veth_lpevent *event)
508 {
509         unsigned long flags;
510
511         spin_lock_irqsave(&cnx->lock, flags);
512         veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
513
514         /* Avoid kicking the statemachine once we're shutdown.
515          * It's unnecessary and it could break veth_stop_connection(). */
516
517         if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518                 cnx->state |= VETH_STATE_RESET;
519                 veth_kick_statemachine(cnx);
520         }
521         spin_unlock_irqrestore(&cnx->lock, flags);
522 }
523
524 static void veth_handle_ack(struct veth_lpevent *event)
525 {
526         HvLpIndex rlp = event->base_event.xTargetLp;
527         struct veth_lpar_connection *cnx = veth_cnx[rlp];
528
529         BUG_ON(! cnx);
530
531         switch (event->base_event.xSubtype) {
532         case VETH_EVENT_CAP:
533                 veth_take_cap_ack(cnx, event);
534                 break;
535         case VETH_EVENT_MONITOR:
536                 veth_take_monitor_ack(cnx, event);
537                 break;
538         default:
539                 veth_error("Unknown ack type %d from LPAR %d.\n",
540                                 event->base_event.xSubtype, rlp);
541         };
542 }
543
544 static void veth_handle_int(struct veth_lpevent *event)
545 {
546         HvLpIndex rlp = event->base_event.xSourceLp;
547         struct veth_lpar_connection *cnx = veth_cnx[rlp];
548         unsigned long flags;
549         int i, acked = 0;
550
551         BUG_ON(! cnx);
552
553         switch (event->base_event.xSubtype) {
554         case VETH_EVENT_CAP:
555                 veth_take_cap(cnx, event);
556                 break;
557         case VETH_EVENT_MONITOR:
558                 /* do nothing... this'll hang out here til we're dead,
559                  * and the hypervisor will return it for us. */
560                 break;
561         case VETH_EVENT_FRAMES_ACK:
562                 spin_lock_irqsave(&cnx->lock, flags);
563
564                 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565                         u16 msgnum = event->u.frames_ack_data.token[i];
566
567                         if (msgnum < VETH_NUMBUFFERS) {
568                                 veth_recycle_msg(cnx, cnx->msgs + msgnum);
569                                 cnx->outstanding_tx--;
570                                 acked++;
571                         }
572                 }
573
574                 if (acked > 0) {
575                         cnx->last_contact = jiffies;
576                         veth_wake_queues(cnx);
577                 }
578
579                 spin_unlock_irqrestore(&cnx->lock, flags);
580                 break;
581         case VETH_EVENT_FRAMES:
582                 veth_receive(cnx, event);
583                 break;
584         default:
585                 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586                                 event->base_event.xSubtype, rlp);
587         };
588 }
589
590 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
591 {
592         struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
593
594         if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
595                 veth_handle_ack(veth_event);
596         else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
597                 veth_handle_int(veth_event);
598 }
599
600 static int veth_process_caps(struct veth_lpar_connection *cnx)
601 {
602         struct veth_cap_data *remote_caps = &cnx->remote_caps;
603         int num_acks_needed;
604
605         /* Convert timer to jiffies */
606         cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
608         if ( (remote_caps->num_buffers == 0)
609              || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
610              || (remote_caps->ack_threshold == 0)
611              || (cnx->ack_timeout == 0) ) {
612                 veth_error("Received incompatible capabilities from LPAR %d.\n",
613                                 cnx->remote_lp);
614                 return HvLpEvent_Rc_InvalidSubtypeData;
615         }
616
617         num_acks_needed = (remote_caps->num_buffers
618                            / remote_caps->ack_threshold) + 1;
619
620         /* FIXME: locking on num_ack_events? */
621         if (cnx->num_ack_events < num_acks_needed) {
622                 int num;
623
624                 num = veth_allocate_events(cnx->remote_lp,
625                                            num_acks_needed-cnx->num_ack_events);
626                 if (num > 0)
627                         cnx->num_ack_events += num;
628
629                 if (cnx->num_ack_events < num_acks_needed) {
630                         veth_error("Couldn't allocate enough ack events "
631                                         "for LPAR %d.\n", cnx->remote_lp);
632
633                         return HvLpEvent_Rc_BufferNotAvailable;
634                 }
635         }
636
637
638         return HvLpEvent_Rc_Good;
639 }
640
641 /* FIXME: The gotos here are a bit dubious */
642 static void veth_statemachine(void *p)
643 {
644         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
645         int rlp = cnx->remote_lp;
646         int rc;
647
648         spin_lock_irq(&cnx->lock);
649
650  restart:
651         if (cnx->state & VETH_STATE_RESET) {
652                 if (cnx->state & VETH_STATE_OPEN)
653                         HvCallEvent_closeLpEventPath(cnx->remote_lp,
654                                                      HvLpEvent_Type_VirtualLan);
655
656                 /*
657                  * Reset ack data. This prevents the ack_timer actually
658                  * doing anything, even if it runs one more time when
659                  * we drop the lock below.
660                  */
661                 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
662                 cnx->num_pending_acks = 0;
663
664                 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
665                                 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
666                                 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
667                                 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
668
669                 /* Clean up any leftover messages */
670                 if (cnx->msgs) {
671                         int i;
672                         for (i = 0; i < VETH_NUMBUFFERS; ++i)
673                                 veth_recycle_msg(cnx, cnx->msgs + i);
674                 }
675
676                 cnx->outstanding_tx = 0;
677                 veth_wake_queues(cnx);
678
679                 /* Drop the lock so we can do stuff that might sleep or
680                  * take other locks. */
681                 spin_unlock_irq(&cnx->lock);
682
683                 del_timer_sync(&cnx->ack_timer);
684                 del_timer_sync(&cnx->reset_timer);
685
686                 spin_lock_irq(&cnx->lock);
687
688                 if (cnx->state & VETH_STATE_RESET)
689                         goto restart;
690
691                 /* Hack, wait for the other end to reset itself. */
692                 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
693                         schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
694                         goto out;
695                 }
696         }
697
698         if (cnx->state & VETH_STATE_SHUTDOWN)
699                 /* It's all over, do nothing */
700                 goto out;
701
702         if ( !(cnx->state & VETH_STATE_OPEN) ) {
703                 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
704                         goto cant_cope;
705
706                 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
707                 cnx->src_inst =
708                         HvCallEvent_getSourceLpInstanceId(rlp,
709                                                           HvLpEvent_Type_VirtualLan);
710                 cnx->dst_inst =
711                         HvCallEvent_getTargetLpInstanceId(rlp,
712                                                           HvLpEvent_Type_VirtualLan);
713                 cnx->state |= VETH_STATE_OPEN;
714         }
715
716         if ( (cnx->state & VETH_STATE_OPEN)
717              && !(cnx->state & VETH_STATE_SENTMON) ) {
718                 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
719                                       HvLpEvent_AckInd_DoAck,
720                                       HvLpEvent_AckType_DeferredAck,
721                                       0, 0, 0, 0, 0, 0);
722
723                 if (rc == HvLpEvent_Rc_Good) {
724                         cnx->state |= VETH_STATE_SENTMON;
725                 } else {
726                         if ( (rc != HvLpEvent_Rc_PartitionDead)
727                              && (rc != HvLpEvent_Rc_PathClosed) )
728                                 veth_error("Error sending monitor to LPAR %d, "
729                                                 "rc = %d\n", rlp, rc);
730
731                         /* Oh well, hope we get a cap from the other
732                          * end and do better when that kicks us */
733                         goto out;
734                 }
735         }
736
737         if ( (cnx->state & VETH_STATE_OPEN)
738              && !(cnx->state & VETH_STATE_SENTCAPS)) {
739                 u64 *rawcap = (u64 *)&cnx->local_caps;
740
741                 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
742                                       HvLpEvent_AckInd_DoAck,
743                                       HvLpEvent_AckType_ImmediateAck,
744                                       0, rawcap[0], rawcap[1], rawcap[2],
745                                       rawcap[3], rawcap[4]);
746
747                 if (rc == HvLpEvent_Rc_Good) {
748                         cnx->state |= VETH_STATE_SENTCAPS;
749                 } else {
750                         if ( (rc != HvLpEvent_Rc_PartitionDead)
751                              && (rc != HvLpEvent_Rc_PathClosed) )
752                                 veth_error("Error sending caps to LPAR %d, "
753                                                 "rc = %d\n", rlp, rc);
754
755                         /* Oh well, hope we get a cap from the other
756                          * end and do better when that kicks us */
757                         goto out;
758                 }
759         }
760
761         if ((cnx->state & VETH_STATE_GOTCAPS)
762             && !(cnx->state & VETH_STATE_SENTCAPACK)) {
763                 struct veth_cap_data *remote_caps = &cnx->remote_caps;
764
765                 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
766                        sizeof(*remote_caps));
767
768                 spin_unlock_irq(&cnx->lock);
769                 rc = veth_process_caps(cnx);
770                 spin_lock_irq(&cnx->lock);
771
772                 /* We dropped the lock, so recheck for anything which
773                  * might mess us up */
774                 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
775                         goto restart;
776
777                 cnx->cap_event.base_event.xRc = rc;
778                 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
779                 if (rc == HvLpEvent_Rc_Good)
780                         cnx->state |= VETH_STATE_SENTCAPACK;
781                 else
782                         goto cant_cope;
783         }
784
785         if ((cnx->state & VETH_STATE_GOTCAPACK)
786             && (cnx->state & VETH_STATE_GOTCAPS)
787             && !(cnx->state & VETH_STATE_READY)) {
788                 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
789                         /* Start the ACK timer */
790                         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
791                         add_timer(&cnx->ack_timer);
792                         cnx->state |= VETH_STATE_READY;
793                 } else {
794                         veth_error("Caps rejected by LPAR %d, rc = %d\n",
795                                         rlp, cnx->cap_ack_event.base_event.xRc);
796                         goto cant_cope;
797                 }
798         }
799
800  out:
801         spin_unlock_irq(&cnx->lock);
802         return;
803
804  cant_cope:
805         /* FIXME: we get here if something happens we really can't
806          * cope with.  The link will never work once we get here, and
807          * all we can do is not lock the rest of the system up */
808         veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
809                         " (state = 0x%04lx)\n", rlp, cnx->state);
810         cnx->state |= VETH_STATE_SHUTDOWN;
811         spin_unlock_irq(&cnx->lock);
812 }
813
814 static int veth_init_connection(u8 rlp)
815 {
816         struct veth_lpar_connection *cnx;
817         struct veth_msg *msgs;
818         int i, rc;
819
820         if ( (rlp == this_lp)
821              || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
822                 return 0;
823
824         cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
825         if (! cnx)
826                 return -ENOMEM;
827         memset(cnx, 0, sizeof(*cnx));
828
829         cnx->remote_lp = rlp;
830         spin_lock_init(&cnx->lock);
831         INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
832
833         init_timer(&cnx->ack_timer);
834         cnx->ack_timer.function = veth_timed_ack;
835         cnx->ack_timer.data = (unsigned long) cnx;
836
837         init_timer(&cnx->reset_timer);
838         cnx->reset_timer.function = veth_timed_reset;
839         cnx->reset_timer.data = (unsigned long) cnx;
840         cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
841
842         memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
843
844         veth_cnx[rlp] = cnx;
845
846         /* This gets us 1 reference, which is held on behalf of the driver
847          * infrastructure. It's released at module unload. */
848         kobject_init(&cnx->kobject);
849         cnx->kobject.ktype = &veth_lpar_connection_ktype;
850         rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
851         if (rc != 0)
852                 return rc;
853
854         msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
855         if (! msgs) {
856                 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
857                 return -ENOMEM;
858         }
859
860         cnx->msgs = msgs;
861         memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
862
863         for (i = 0; i < VETH_NUMBUFFERS; i++) {
864                 msgs[i].token = i;
865                 veth_stack_push(cnx, msgs + i);
866         }
867
868         cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
869
870         if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
871                 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
872                 return -ENOMEM;
873         }
874
875         cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
876         cnx->local_caps.ack_threshold = ACK_THRESHOLD;
877         cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
878
879         return 0;
880 }
881
882 static void veth_stop_connection(struct veth_lpar_connection *cnx)
883 {
884         if (!cnx)
885                 return;
886
887         spin_lock_irq(&cnx->lock);
888         cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
889         veth_kick_statemachine(cnx);
890         spin_unlock_irq(&cnx->lock);
891
892         /* There's a slim chance the reset code has just queued the
893          * statemachine to run in five seconds. If so we need to cancel
894          * that and requeue the work to run now. */
895         if (cancel_delayed_work(&cnx->statemachine_wq)) {
896                 spin_lock_irq(&cnx->lock);
897                 veth_kick_statemachine(cnx);
898                 spin_unlock_irq(&cnx->lock);
899         }
900
901         /* Wait for the state machine to run. */
902         flush_scheduled_work();
903 }
904
905 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
906 {
907         if (!cnx)
908                 return;
909
910         if (cnx->num_events > 0)
911                 mf_deallocate_lp_events(cnx->remote_lp,
912                                       HvLpEvent_Type_VirtualLan,
913                                       cnx->num_events,
914                                       NULL, NULL);
915         if (cnx->num_ack_events > 0)
916                 mf_deallocate_lp_events(cnx->remote_lp,
917                                       HvLpEvent_Type_VirtualLan,
918                                       cnx->num_ack_events,
919                                       NULL, NULL);
920
921         kfree(cnx->msgs);
922         veth_cnx[cnx->remote_lp] = NULL;
923         kfree(cnx);
924 }
925
926 static void veth_release_connection(struct kobject *kobj)
927 {
928         struct veth_lpar_connection *cnx;
929         cnx = container_of(kobj, struct veth_lpar_connection, kobject);
930         veth_stop_connection(cnx);
931         veth_destroy_connection(cnx);
932 }
933
934 /*
935  * net_device code
936  */
937
938 static int veth_open(struct net_device *dev)
939 {
940         struct veth_port *port = (struct veth_port *) dev->priv;
941
942         memset(&port->stats, 0, sizeof (port->stats));
943         netif_start_queue(dev);
944         return 0;
945 }
946
947 static int veth_close(struct net_device *dev)
948 {
949         netif_stop_queue(dev);
950         return 0;
951 }
952
953 static struct net_device_stats *veth_get_stats(struct net_device *dev)
954 {
955         struct veth_port *port = (struct veth_port *) dev->priv;
956
957         return &port->stats;
958 }
959
960 static int veth_change_mtu(struct net_device *dev, int new_mtu)
961 {
962         if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
963                 return -EINVAL;
964         dev->mtu = new_mtu;
965         return 0;
966 }
967
968 static void veth_set_multicast_list(struct net_device *dev)
969 {
970         struct veth_port *port = (struct veth_port *) dev->priv;
971         unsigned long flags;
972
973         write_lock_irqsave(&port->mcast_gate, flags);
974
975         if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
976                         (dev->mc_count > VETH_MAX_MCAST)) {
977                 port->promiscuous = 1;
978         } else {
979                 struct dev_mc_list *dmi = dev->mc_list;
980                 int i;
981
982                 port->promiscuous = 0;
983
984                 /* Update table */
985                 port->num_mcast = 0;
986
987                 for (i = 0; i < dev->mc_count; i++) {
988                         u8 *addr = dmi->dmi_addr;
989                         u64 xaddr = 0;
990
991                         if (addr[0] & 0x01) {/* multicast address? */
992                                 memcpy(&xaddr, addr, ETH_ALEN);
993                                 port->mcast_addr[port->num_mcast] = xaddr;
994                                 port->num_mcast++;
995                         }
996                         dmi = dmi->next;
997                 }
998         }
999
1000         write_unlock_irqrestore(&port->mcast_gate, flags);
1001 }
1002
1003 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1004 {
1005         strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
1006         info->driver[sizeof(info->driver) - 1] = '\0';
1007         strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
1008         info->version[sizeof(info->version) - 1] = '\0';
1009 }
1010
1011 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1012 {
1013         ecmd->supported = (SUPPORTED_1000baseT_Full
1014                           | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1015         ecmd->advertising = (SUPPORTED_1000baseT_Full
1016                             | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1017         ecmd->port = PORT_FIBRE;
1018         ecmd->transceiver = XCVR_INTERNAL;
1019         ecmd->phy_address = 0;
1020         ecmd->speed = SPEED_1000;
1021         ecmd->duplex = DUPLEX_FULL;
1022         ecmd->autoneg = AUTONEG_ENABLE;
1023         ecmd->maxtxpkt = 120;
1024         ecmd->maxrxpkt = 120;
1025         return 0;
1026 }
1027
1028 static u32 veth_get_link(struct net_device *dev)
1029 {
1030         return 1;
1031 }
1032
1033 static struct ethtool_ops ops = {
1034         .get_drvinfo = veth_get_drvinfo,
1035         .get_settings = veth_get_settings,
1036         .get_link = veth_get_link,
1037 };
1038
1039 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
1040 {
1041         struct net_device *dev;
1042         struct veth_port *port;
1043         int i, rc;
1044
1045         dev = alloc_etherdev(sizeof (struct veth_port));
1046         if (! dev) {
1047                 veth_error("Unable to allocate net_device structure!\n");
1048                 return NULL;
1049         }
1050
1051         port = (struct veth_port *) dev->priv;
1052
1053         spin_lock_init(&port->queue_lock);
1054         rwlock_init(&port->mcast_gate);
1055         port->stopped_map = 0;
1056
1057         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1058                 HvLpVirtualLanIndexMap map;
1059
1060                 if (i == this_lp)
1061                         continue;
1062                 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1063                 if (map & (0x8000 >> vlan))
1064                         port->lpar_map |= (1 << i);
1065         }
1066         port->dev = vdev;
1067
1068         dev->dev_addr[0] = 0x02;
1069         dev->dev_addr[1] = 0x01;
1070         dev->dev_addr[2] = 0xff;
1071         dev->dev_addr[3] = vlan;
1072         dev->dev_addr[4] = 0xff;
1073         dev->dev_addr[5] = this_lp;
1074
1075         dev->mtu = VETH_MAX_MTU;
1076
1077         memcpy(&port->mac_addr, dev->dev_addr, 6);
1078
1079         dev->open = veth_open;
1080         dev->hard_start_xmit = veth_start_xmit;
1081         dev->stop = veth_close;
1082         dev->get_stats = veth_get_stats;
1083         dev->change_mtu = veth_change_mtu;
1084         dev->set_mac_address = NULL;
1085         dev->set_multicast_list = veth_set_multicast_list;
1086         SET_ETHTOOL_OPS(dev, &ops);
1087
1088         SET_NETDEV_DEV(dev, vdev);
1089
1090         rc = register_netdev(dev);
1091         if (rc != 0) {
1092                 veth_error("Failed registering net device for vlan%d.\n", vlan);
1093                 free_netdev(dev);
1094                 return NULL;
1095         }
1096
1097         kobject_init(&port->kobject);
1098         port->kobject.parent = &dev->class_dev.kobj;
1099         port->kobject.ktype  = &veth_port_ktype;
1100         kobject_set_name(&port->kobject, "veth_port");
1101         if (0 != kobject_add(&port->kobject))
1102                 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1103
1104         veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1105                         dev->name, vlan, port->lpar_map);
1106
1107         return dev;
1108 }
1109
1110 /*
1111  * Tx path
1112  */
1113
1114 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1115                                 struct net_device *dev)
1116 {
1117         struct veth_lpar_connection *cnx = veth_cnx[rlp];
1118         struct veth_port *port = (struct veth_port *) dev->priv;
1119         HvLpEvent_Rc rc;
1120         struct veth_msg *msg = NULL;
1121         unsigned long flags;
1122
1123         if (! cnx)
1124                 return 0;
1125
1126         spin_lock_irqsave(&cnx->lock, flags);
1127
1128         if (! (cnx->state & VETH_STATE_READY))
1129                 goto no_error;
1130
1131         if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1132                 goto drop;
1133
1134         msg = veth_stack_pop(cnx);
1135         if (! msg)
1136                 goto drop;
1137
1138         msg->in_use = 1;
1139         msg->skb = skb_get(skb);
1140
1141         msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1142                                 skb->len, DMA_TO_DEVICE);
1143
1144         if (dma_mapping_error(msg->data.addr[0]))
1145                 goto recycle_and_drop;
1146
1147         msg->dev = port->dev;
1148         msg->data.len[0] = skb->len;
1149         msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1150
1151         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1152
1153         if (rc != HvLpEvent_Rc_Good)
1154                 goto recycle_and_drop;
1155
1156         /* If the timer's not already running, start it now. */
1157         if (0 == cnx->outstanding_tx)
1158                 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1159
1160         cnx->last_contact = jiffies;
1161         cnx->outstanding_tx++;
1162
1163         if (veth_stack_is_empty(cnx))
1164                 veth_stop_queues(cnx);
1165
1166  no_error:
1167         spin_unlock_irqrestore(&cnx->lock, flags);
1168         return 0;
1169
1170  recycle_and_drop:
1171         veth_recycle_msg(cnx, msg);
1172  drop:
1173         spin_unlock_irqrestore(&cnx->lock, flags);
1174         return 1;
1175 }
1176
1177 static void veth_transmit_to_many(struct sk_buff *skb,
1178                                           HvLpIndexMap lpmask,
1179                                           struct net_device *dev)
1180 {
1181         struct veth_port *port = (struct veth_port *) dev->priv;
1182         int i, success, error;
1183
1184         success = error = 0;
1185
1186         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1187                 if ((lpmask & (1 << i)) == 0)
1188                         continue;
1189
1190                 if (veth_transmit_to_one(skb, i, dev))
1191                         error = 1;
1192                 else
1193                         success = 1;
1194         }
1195
1196         if (error)
1197                 port->stats.tx_errors++;
1198
1199         if (success) {
1200                 port->stats.tx_packets++;
1201                 port->stats.tx_bytes += skb->len;
1202         }
1203 }
1204
1205 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1206 {
1207         unsigned char *frame = skb->data;
1208         struct veth_port *port = (struct veth_port *) dev->priv;
1209         HvLpIndexMap lpmask;
1210
1211         if (! (frame[0] & 0x01)) {
1212                 /* unicast packet */
1213                 HvLpIndex rlp = frame[5];
1214
1215                 if ( ! ((1 << rlp) & port->lpar_map) ) {
1216                         dev_kfree_skb(skb);
1217                         return 0;
1218                 }
1219
1220                 lpmask = 1 << rlp;
1221         } else {
1222                 lpmask = port->lpar_map;
1223         }
1224
1225         veth_transmit_to_many(skb, lpmask, dev);
1226
1227         dev_kfree_skb(skb);
1228
1229         return 0;
1230 }
1231
1232 /* You must hold the connection's lock when you call this function. */
1233 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1234                              struct veth_msg *msg)
1235 {
1236         u32 dma_address, dma_length;
1237
1238         if (msg->in_use) {
1239                 msg->in_use = 0;
1240                 dma_address = msg->data.addr[0];
1241                 dma_length = msg->data.len[0];
1242
1243                 if (!dma_mapping_error(dma_address))
1244                         dma_unmap_single(msg->dev, dma_address, dma_length,
1245                                         DMA_TO_DEVICE);
1246
1247                 if (msg->skb) {
1248                         dev_kfree_skb_any(msg->skb);
1249                         msg->skb = NULL;
1250                 }
1251
1252                 memset(&msg->data, 0, sizeof(msg->data));
1253                 veth_stack_push(cnx, msg);
1254         } else if (cnx->state & VETH_STATE_OPEN) {
1255                 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1256                                 cnx->remote_lp, msg->token);
1257         }
1258 }
1259
1260 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1261 {
1262         int i;
1263
1264         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1265                 struct net_device *dev = veth_dev[i];
1266                 struct veth_port *port;
1267                 unsigned long flags;
1268
1269                 if (! dev)
1270                         continue;
1271
1272                 port = (struct veth_port *)dev->priv;
1273
1274                 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1275                         continue;
1276
1277                 spin_lock_irqsave(&port->queue_lock, flags);
1278
1279                 port->stopped_map &= ~(1 << cnx->remote_lp);
1280
1281                 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1282                         veth_debug("cnx %d: woke queue for %s.\n",
1283                                         cnx->remote_lp, dev->name);
1284                         netif_wake_queue(dev);
1285                 }
1286                 spin_unlock_irqrestore(&port->queue_lock, flags);
1287         }
1288 }
1289
1290 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1291 {
1292         int i;
1293
1294         for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1295                 struct net_device *dev = veth_dev[i];
1296                 struct veth_port *port;
1297
1298                 if (! dev)
1299                         continue;
1300
1301                 port = (struct veth_port *)dev->priv;
1302
1303                 /* If this cnx is not on the vlan for this port, continue */
1304                 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1305                         continue;
1306
1307                 spin_lock(&port->queue_lock);
1308
1309                 netif_stop_queue(dev);
1310                 port->stopped_map |= (1 << cnx->remote_lp);
1311
1312                 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1313                                 cnx->remote_lp, dev->name, port->stopped_map);
1314
1315                 spin_unlock(&port->queue_lock);
1316         }
1317 }
1318
1319 static void veth_timed_reset(unsigned long ptr)
1320 {
1321         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1322         unsigned long trigger_time, flags;
1323
1324         /* FIXME is it possible this fires after veth_stop_connection()?
1325          * That would reschedule the statemachine for 5 seconds and probably
1326          * execute it after the module's been unloaded. Hmm. */
1327
1328         spin_lock_irqsave(&cnx->lock, flags);
1329
1330         if (cnx->outstanding_tx > 0) {
1331                 trigger_time = cnx->last_contact + cnx->reset_timeout;
1332
1333                 if (trigger_time < jiffies) {
1334                         cnx->state |= VETH_STATE_RESET;
1335                         veth_kick_statemachine(cnx);
1336                         veth_error("%d packets not acked by LPAR %d within %d "
1337                                         "seconds, resetting.\n",
1338                                         cnx->outstanding_tx, cnx->remote_lp,
1339                                         cnx->reset_timeout / HZ);
1340                 } else {
1341                         /* Reschedule the timer */
1342                         trigger_time = jiffies + cnx->reset_timeout;
1343                         mod_timer(&cnx->reset_timer, trigger_time);
1344                 }
1345         }
1346
1347         spin_unlock_irqrestore(&cnx->lock, flags);
1348 }
1349
1350 /*
1351  * Rx path
1352  */
1353
1354 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1355 {
1356         int wanted = 0;
1357         int i;
1358         unsigned long flags;
1359
1360         if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1361                 return 1;
1362
1363         read_lock_irqsave(&port->mcast_gate, flags);
1364
1365         if (port->promiscuous) {
1366                 wanted = 1;
1367                 goto out;
1368         }
1369
1370         for (i = 0; i < port->num_mcast; ++i) {
1371                 if (port->mcast_addr[i] == mac_addr) {
1372                         wanted = 1;
1373                         break;
1374                 }
1375         }
1376
1377  out:
1378         read_unlock_irqrestore(&port->mcast_gate, flags);
1379
1380         return wanted;
1381 }
1382
1383 struct dma_chunk {
1384         u64 addr;
1385         u64 size;
1386 };
1387
1388 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1389
1390 static inline void veth_build_dma_list(struct dma_chunk *list,
1391                                        unsigned char *p, unsigned long length)
1392 {
1393         unsigned long done;
1394         int i = 1;
1395
1396         /* FIXME: skbs are continguous in real addresses.  Do we
1397          * really need to break it into PAGE_SIZE chunks, or can we do
1398          * it just at the granularity of iSeries real->absolute
1399          * mapping?  Indeed, given the way the allocator works, can we
1400          * count on them being absolutely contiguous? */
1401         list[0].addr = iseries_hv_addr(p);
1402         list[0].size = min(length,
1403                            PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1404
1405         done = list[0].size;
1406         while (done < length) {
1407                 list[i].addr = iseries_hv_addr(p + done);
1408                 list[i].size = min(length-done, PAGE_SIZE);
1409                 done += list[i].size;
1410                 i++;
1411         }
1412 }
1413
1414 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1415 {
1416         HvLpEvent_Rc rc;
1417
1418         rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1419                              0, &cnx->pending_acks);
1420
1421         if (rc != HvLpEvent_Rc_Good)
1422                 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1423                                 cnx->remote_lp, (int)rc);
1424
1425         cnx->num_pending_acks = 0;
1426         memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1427 }
1428
1429 static void veth_receive(struct veth_lpar_connection *cnx,
1430                          struct veth_lpevent *event)
1431 {
1432         struct veth_frames_data *senddata = &event->u.frames_data;
1433         int startchunk = 0;
1434         int nchunks;
1435         unsigned long flags;
1436         HvLpDma_Rc rc;
1437
1438         do {
1439                 u16 length = 0;
1440                 struct sk_buff *skb;
1441                 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1442                 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1443                 u64 dest;
1444                 HvLpVirtualLanIndex vlan;
1445                 struct net_device *dev;
1446                 struct veth_port *port;
1447
1448                 /* FIXME: do we need this? */
1449                 memset(local_list, 0, sizeof(local_list));
1450                 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1451
1452                 /* a 0 address marks the end of the valid entries */
1453                 if (senddata->addr[startchunk] == 0)
1454                         break;
1455
1456                 /* make sure that we have at least 1 EOF entry in the
1457                  * remaining entries */
1458                 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1459                         veth_error("Missing EOF fragment in event "
1460                                         "eofmask = 0x%x startchunk = %d\n",
1461                                         (unsigned)senddata->eofmask,
1462                                         startchunk);
1463                         break;
1464                 }
1465
1466                 /* build list of chunks in this frame */
1467                 nchunks = 0;
1468                 do {
1469                         remote_list[nchunks].addr =
1470                                 (u64) senddata->addr[startchunk+nchunks] << 32;
1471                         remote_list[nchunks].size =
1472                                 senddata->len[startchunk+nchunks];
1473                         length += remote_list[nchunks].size;
1474                 } while (! (senddata->eofmask &
1475                             (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1476
1477                 /* length == total length of all chunks */
1478                 /* nchunks == # of chunks in this frame */
1479
1480                 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1481                         veth_error("Received oversize frame from LPAR %d "
1482                                         "(length = %d)\n",
1483                                         cnx->remote_lp, length);
1484                         continue;
1485                 }
1486
1487                 skb = alloc_skb(length, GFP_ATOMIC);
1488                 if (!skb)
1489                         continue;
1490
1491                 veth_build_dma_list(local_list, skb->data, length);
1492
1493                 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1494                                             event->base_event.xSourceLp,
1495                                             HvLpDma_Direction_RemoteToLocal,
1496                                             cnx->src_inst,
1497                                             cnx->dst_inst,
1498                                             HvLpDma_AddressType_RealAddress,
1499                                             HvLpDma_AddressType_TceIndex,
1500                                             iseries_hv_addr(&local_list),
1501                                             iseries_hv_addr(&remote_list),
1502                                             length);
1503                 if (rc != HvLpDma_Rc_Good) {
1504                         dev_kfree_skb_irq(skb);
1505                         continue;
1506                 }
1507
1508                 vlan = skb->data[9];
1509                 dev = veth_dev[vlan];
1510                 if (! dev) {
1511                         /*
1512                          * Some earlier versions of the driver sent
1513                          * broadcasts down all connections, even to lpars
1514                          * that weren't on the relevant vlan. So ignore
1515                          * packets belonging to a vlan we're not on.
1516                          * We can also be here if we receive packets while
1517                          * the driver is going down, because then dev is NULL.
1518                          */
1519                         dev_kfree_skb_irq(skb);
1520                         continue;
1521                 }
1522
1523                 port = (struct veth_port *)dev->priv;
1524                 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1525
1526                 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1527                         dev_kfree_skb_irq(skb);
1528                         continue;
1529                 }
1530                 if (! veth_frame_wanted(port, dest)) {
1531                         dev_kfree_skb_irq(skb);
1532                         continue;
1533                 }
1534
1535                 skb_put(skb, length);
1536                 skb->dev = dev;
1537                 skb->protocol = eth_type_trans(skb, dev);
1538                 skb->ip_summed = CHECKSUM_NONE;
1539                 netif_rx(skb);  /* send it up */
1540                 port->stats.rx_packets++;
1541                 port->stats.rx_bytes += length;
1542         } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1543
1544         /* Ack it */
1545         spin_lock_irqsave(&cnx->lock, flags);
1546         BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1547
1548         cnx->pending_acks[cnx->num_pending_acks++] =
1549                 event->base_event.xCorrelationToken;
1550
1551         if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1552              || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1553                 veth_flush_acks(cnx);
1554
1555         spin_unlock_irqrestore(&cnx->lock, flags);
1556 }
1557
1558 static void veth_timed_ack(unsigned long ptr)
1559 {
1560         struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1561         unsigned long flags;
1562
1563         /* Ack all the events */
1564         spin_lock_irqsave(&cnx->lock, flags);
1565         if (cnx->num_pending_acks > 0)
1566                 veth_flush_acks(cnx);
1567
1568         /* Reschedule the timer */
1569         cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1570         add_timer(&cnx->ack_timer);
1571         spin_unlock_irqrestore(&cnx->lock, flags);
1572 }
1573
1574 static int veth_remove(struct vio_dev *vdev)
1575 {
1576         struct veth_lpar_connection *cnx;
1577         struct net_device *dev;
1578         struct veth_port *port;
1579         int i;
1580
1581         dev = veth_dev[vdev->unit_address];
1582
1583         if (! dev)
1584                 return 0;
1585
1586         port = netdev_priv(dev);
1587
1588         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1589                 cnx = veth_cnx[i];
1590
1591                 if (cnx && (port->lpar_map & (1 << i))) {
1592                         /* Drop our reference to connections on our VLAN */
1593                         kobject_put(&cnx->kobject);
1594                 }
1595         }
1596
1597         veth_dev[vdev->unit_address] = NULL;
1598         kobject_del(&port->kobject);
1599         kobject_put(&port->kobject);
1600         unregister_netdev(dev);
1601         free_netdev(dev);
1602
1603         return 0;
1604 }
1605
1606 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1607 {
1608         int i = vdev->unit_address;
1609         struct net_device *dev;
1610         struct veth_port *port;
1611
1612         dev = veth_probe_one(i, &vdev->dev);
1613         if (dev == NULL) {
1614                 veth_remove(vdev);
1615                 return 1;
1616         }
1617         veth_dev[i] = dev;
1618
1619         port = (struct veth_port*)netdev_priv(dev);
1620
1621         /* Start the state machine on each connection on this vlan. If we're
1622          * the first dev to do so this will commence link negotiation */
1623         for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1624                 struct veth_lpar_connection *cnx;
1625
1626                 if (! (port->lpar_map & (1 << i)))
1627                         continue;
1628
1629                 cnx = veth_cnx[i];
1630                 if (!cnx)
1631                         continue;
1632
1633                 kobject_get(&cnx->kobject);
1634                 veth_kick_statemachine(cnx);
1635         }
1636
1637         return 0;
1638 }
1639
1640 /**
1641  * veth_device_table: Used by vio.c to match devices that we
1642  * support.
1643  */
1644 static struct vio_device_id veth_device_table[] __devinitdata = {
1645         { "vlan", "" },
1646         { "", "" }
1647 };
1648 MODULE_DEVICE_TABLE(vio, veth_device_table);
1649
1650 static struct vio_driver veth_driver = {
1651         .id_table = veth_device_table,
1652         .probe = veth_probe,
1653         .remove = veth_remove,
1654         .driver = {
1655                 .name = DRV_NAME,
1656                 .owner = THIS_MODULE,
1657         }
1658 };
1659
1660 /*
1661  * Module initialization/cleanup
1662  */
1663
1664 void __exit veth_module_cleanup(void)
1665 {
1666         int i;
1667         struct veth_lpar_connection *cnx;
1668
1669         /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1670         HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1671
1672         /* Make sure any work queued from Hypervisor callbacks is finished. */
1673         flush_scheduled_work();
1674
1675         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1676                 cnx = veth_cnx[i];
1677
1678                 if (!cnx)
1679                         continue;
1680
1681                 /* Remove the connection from sysfs */
1682                 kobject_del(&cnx->kobject);
1683                 /* Drop the driver's reference to the connection */
1684                 kobject_put(&cnx->kobject);
1685         }
1686
1687         /* Unregister the driver, which will close all the netdevs and stop
1688          * the connections when they're no longer referenced. */
1689         vio_unregister_driver(&veth_driver);
1690 }
1691 module_exit(veth_module_cleanup);
1692
1693 int __init veth_module_init(void)
1694 {
1695         int i;
1696         int rc;
1697
1698         this_lp = HvLpConfig_getLpIndex_outline();
1699
1700         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1701                 rc = veth_init_connection(i);
1702                 if (rc != 0)
1703                         goto error;
1704         }
1705
1706         HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1707                                   &veth_handle_event);
1708
1709         rc = vio_register_driver(&veth_driver);
1710         if (rc != 0)
1711                 goto error;
1712
1713         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1714                 struct kobject *kobj;
1715
1716                 if (!veth_cnx[i])
1717                         continue;
1718
1719                 kobj = &veth_cnx[i]->kobject;
1720                 kobj->parent = &veth_driver.driver.kobj;
1721                 /* If the add failes, complain but otherwise continue */
1722                 if (0 != kobject_add(kobj))
1723                         veth_error("cnx %d: Failed adding to sysfs.\n", i);
1724         }
1725
1726         return 0;
1727
1728 error:
1729         for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1730                 veth_destroy_connection(veth_cnx[i]);
1731         }
1732
1733         return rc;
1734 }
1735 module_init(veth_module_init);