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