?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / drivers / net / iseries_veth.c
blobb69c11306ad0c1e89428d79a775b514714e9561d
1 /* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
2 /*
3 * IBM eServer iSeries Virtual Ethernet Device Driver
4 * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
5 * Substantially cleaned up by:
6 * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
7 * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
8 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License as
11 * published by the Free Software Foundation; either version 2 of the
12 * License, or (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
22 * USA
23 *
24 *
25 * This module implements the virtual ethernet device for iSeries LPAR
26 * Linux. It uses hypervisor message passing to implement an
27 * ethernet-like network device communicating between partitions on
28 * the iSeries.
29 *
30 * The iSeries LPAR hypervisor currently allows for up to 16 different
31 * virtual ethernets. These are all dynamically configurable on
32 * OS/400 partitions, but dynamic configuration is not supported under
33 * Linux yet. An ethXX network device will be created for each
34 * virtual ethernet this partition is connected to.
35 *
36 * - This driver is responsible for routing packets to and from other
37 * partitions. The MAC addresses used by the virtual ethernets
38 * contains meaning and must not be modified.
39 *
40 * - Having 2 virtual ethernets to the same remote partition DOES NOT
41 * double the available bandwidth. The 2 devices will share the
42 * available hypervisor bandwidth.
43 *
44 * - If you send a packet to your own mac address, it will just be
45 * dropped, you won't get it on the receive side.
46 *
47 * - Multicast is implemented by sending the frame frame to every
48 * other partition. It is the responsibility of the receiving
49 * partition to filter the addresses desired.
50 *
51 * Tunable parameters:
52 *
53 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
54 * controls how much memory Linux will allocate per remote partition
55 * it is communicating with. It can be thought of as the maximum
56 * number of packets outstanding to a remote partition at a time.
57 */
58
59 #include <linux/module.h>
60 #include <linux/types.h>
61 #include <linux/errno.h>
62 #include <linux/ioport.h>
63 #include <linux/kernel.h>
64 #include <linux/netdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <linux/init.h>
68 #include <linux/delay.h>
69 #include <linux/mm.h>
70 #include <linux/ethtool.h>
71 #include <linux/if_ether.h>
72 #include <linux/slab.h>
73
74 #include <asm/abs_addr.h>
75 #include <asm/iseries/mf.h>
76 #include <asm/uaccess.h>
77 #include <asm/firmware.h>
78 #include <asm/iseries/hv_lp_config.h>
79 #include <asm/iseries/hv_types.h>
80 #include <asm/iseries/hv_lp_event.h>
81 #include <asm/iommu.h>
82 #include <asm/vio.h>
83
84 #undef DEBUG
85
86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88 MODULE_LICENSE("GPL");
89
90 #define VETH_EVENT_CAP (0)
91 #define VETH_EVENT_FRAMES (1)
92 #define VETH_EVENT_MONITOR (2)
93 #define VETH_EVENT_FRAMES_ACK (3)
94
95 #define VETH_MAX_ACKS_PER_MSG (20)
96 #define VETH_MAX_FRAMES_PER_MSG (6)
97
98 struct veth_frames_data {
99 u32 addr[VETH_MAX_FRAMES_PER_MSG];
100 u16 len[VETH_MAX_FRAMES_PER_MSG];
101 u32 eofmask;
102 };
103 #define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
104
105 struct veth_frames_ack_data {
106 u16 token[VETH_MAX_ACKS_PER_MSG];
107 };
108
109 struct veth_cap_data {
110 u8 caps_version;
111 u8 rsvd1;
112 u16 num_buffers;
113 u16 ack_threshold;
114 u16 rsvd2;
115 u32 ack_timeout;
116 u32 rsvd3;
117 u64 rsvd4[3];
118 };
119
120 struct veth_lpevent {
121 struct HvLpEvent base_event;
122 union {
123 struct veth_cap_data caps_data;
124 struct veth_frames_data frames_data;
125 struct veth_frames_ack_data frames_ack_data;
126 } u;
127
128 };
129
130 #define DRV_NAME "iseries_veth"
131 #define DRV_VERSION "2.0"
132
133 #define VETH_NUMBUFFERS (120)
134 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
135 #define VETH_MAX_MCAST (12)
136
137 #define VETH_MAX_MTU (9000)
138
139 #if VETH_NUMBUFFERS < 10
140 #define ACK_THRESHOLD (1)
141 #elif VETH_NUMBUFFERS < 20
142 #define ACK_THRESHOLD (4)
143 #elif VETH_NUMBUFFERS < 40
144 #define ACK_THRESHOLD (10)
145 #else
146 #define ACK_THRESHOLD (20)
147 #endif
148
149 #define VETH_STATE_SHUTDOWN (0x0001)
150 #define VETH_STATE_OPEN (0x0002)
151 #define VETH_STATE_RESET (0x0004)
152 #define VETH_STATE_SENTMON (0x0008)
153 #define VETH_STATE_SENTCAPS (0x0010)
154 #define VETH_STATE_GOTCAPACK (0x0020)
155 #define VETH_STATE_GOTCAPS (0x0040)
156 #define VETH_STATE_SENTCAPACK (0x0080)
157 #define VETH_STATE_READY (0x0100)
158
159 struct veth_msg {
160 struct veth_msg *next;
161 struct veth_frames_data data;
162 int token;
163 int in_use;
164 struct sk_buff *skb;
165 struct device *dev;
166 };
167
168 struct veth_lpar_connection {
169 HvLpIndex remote_lp;
170 struct delayed_work statemachine_wq;
171 struct veth_msg *msgs;
172 int num_events;
173 struct veth_cap_data local_caps;
174
175 struct kobject kobject;
176 struct timer_list ack_timer;
177
178 struct timer_list reset_timer;
179 unsigned int reset_timeout;
180 unsigned long last_contact;
181 int outstanding_tx;
182
183 spinlock_t lock;
184 unsigned long state;
185 HvLpInstanceId src_inst;
186 HvLpInstanceId dst_inst;
187 struct veth_lpevent cap_event, cap_ack_event;
188 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189 u32 num_pending_acks;
190
191 int num_ack_events;
192 struct veth_cap_data remote_caps;
193 u32 ack_timeout;
194
195 struct veth_msg *msg_stack_head;
196 };
197
198 struct veth_port {
199 struct device *dev;
200 u64 mac_addr;
201 HvLpIndexMap lpar_map;
202
203 /* queue_lock protects the stopped_map and dev's queue. */
204 spinlock_t queue_lock;
205 HvLpIndexMap stopped_map;
206
207 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
208 rwlock_t mcast_gate;
209 int promiscuous;
210 int num_mcast;
211 u64 mcast_addr[VETH_MAX_MCAST];
212
213 struct kobject kobject;
214 };
215
216 static HvLpIndex this_lp;
217 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
218 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
219
220 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
221 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
222 static void veth_wake_queues(struct veth_lpar_connection *cnx);
223 static void veth_stop_queues(struct veth_lpar_connection *cnx);
224 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
225 static void veth_release_connection(struct kobject *kobject);
226 static void veth_timed_ack(unsigned long ptr);
227 static void veth_timed_reset(unsigned long ptr);
228
229 /*
230 * Utility functions
231 */
232
233 #define veth_info(fmt, args...) \
234 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
235
236 #define veth_error(fmt, args...) \
237 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
238
239 #ifdef DEBUG
240 #define veth_debug(fmt, args...) \
241 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
242 #else
243 #define veth_debug(fmt, args...) do {} while (0)
244 #endif
245
246 /* You must hold the connection's lock when you call this function. */
247 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
248 struct veth_msg *msg)
249 {
250 msg->next = cnx->msg_stack_head;
251 cnx->msg_stack_head = msg;
252 }
253
254 /* You must hold the connection's lock when you call this function. */
255 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
256 {
257 struct veth_msg *msg;
258
259 msg = cnx->msg_stack_head;
260 if (msg)
261 cnx->msg_stack_head = cnx->msg_stack_head->next;
262
263 return msg;
264 }
265
266 /* You must hold the connection's lock when you call this function. */
267 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
268 {
269 return cnx->msg_stack_head == NULL;
270 }
271
272 static inline HvLpEvent_Rc
273 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
274 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
275 u64 token,
276 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
277 {
278 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
279 HvLpEvent_Type_VirtualLan,
280 subtype, ackind, acktype,
281 cnx->src_inst,
282 cnx->dst_inst,
283 token, data1, data2, data3,
284 data4, data5);
285 }
286
287 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
288 u16 subtype, u64 token, void *data)
289 {
290 u64 *p = (u64 *) data;
291
292 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
293 HvLpEvent_AckType_ImmediateAck,
294 token, p[0], p[1], p[2], p[3], p[4]);
295 }
296
297 struct veth_allocation {
298 struct completion c;
299 int num;
300 };
301
302 static void veth_complete_allocation(void *parm, int number)
303 {
304 struct veth_allocation *vc = (struct veth_allocation *)parm;
305
306 vc->num = number;
307 complete(&vc->c);
308 }
309
310 static int veth_allocate_events(HvLpIndex rlp, int number)
311 {
312 struct veth_allocation vc =
313 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
314
315 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
316 sizeof(struct veth_lpevent), number,
317 &veth_complete_allocation, &vc);
318 wait_for_completion(&vc.c);
319
320 return vc.num;
321 }
322
323 /*
324 * sysfs support
325 */
326
327 struct veth_cnx_attribute {
328 struct attribute attr;
329 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
330 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
331 };
332
333 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334 struct attribute *attr, char *buf)
335 {
336 struct veth_cnx_attribute *cnx_attr;
337 struct veth_lpar_connection *cnx;
338
339 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
341
342 if (!cnx_attr->show)
343 return -EIO;
344
345 return cnx_attr->show(cnx, buf);
346 }
347
348 #define CUSTOM_CNX_ATTR(_name, _format, _expression) \
349 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
350 { \
351 return sprintf(buf, _format, _expression); \
352 } \
353 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
354
355 #define SIMPLE_CNX_ATTR(_name) \
356 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
357
358 SIMPLE_CNX_ATTR(outstanding_tx);
359 SIMPLE_CNX_ATTR(remote_lp);
360 SIMPLE_CNX_ATTR(num_events);
361 SIMPLE_CNX_ATTR(src_inst);
362 SIMPLE_CNX_ATTR(dst_inst);
363 SIMPLE_CNX_ATTR(num_pending_acks);
364 SIMPLE_CNX_ATTR(num_ack_events);
365 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
370
371 #define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
372
373 static struct attribute *veth_cnx_default_attrs[] = {
374 GET_CNX_ATTR(outstanding_tx),
375 GET_CNX_ATTR(remote_lp),
376 GET_CNX_ATTR(num_events),
377 GET_CNX_ATTR(reset_timeout),
378 GET_CNX_ATTR(last_contact),
379 GET_CNX_ATTR(state),
380 GET_CNX_ATTR(src_inst),
381 GET_CNX_ATTR(dst_inst),
382 GET_CNX_ATTR(num_pending_acks),
383 GET_CNX_ATTR(num_ack_events),
384 GET_CNX_ATTR(ack_timeout),
385 NULL
386 };
387
388 static const struct sysfs_ops veth_cnx_sysfs_ops = {
389 .show = veth_cnx_attribute_show
390 };
391
392 static struct kobj_type veth_lpar_connection_ktype = {
393 .release = veth_release_connection,
394 .sysfs_ops = &veth_cnx_sysfs_ops,
395 .default_attrs = veth_cnx_default_attrs
396 };
397
398 struct veth_port_attribute {
399 struct attribute attr;
400 ssize_t (*show)(struct veth_port *, char *buf);
401 ssize_t (*store)(struct veth_port *, const char *buf);
402 };
403
404 static ssize_t veth_port_attribute_show(struct kobject *kobj,
405 struct attribute *attr, char *buf)
406 {
407 struct veth_port_attribute *port_attr;
408 struct veth_port *port;
409
410 port_attr = container_of(attr, struct veth_port_attribute, attr);
411 port = container_of(kobj, struct veth_port, kobject);
412
413 if (!port_attr->show)
414 return -EIO;
415
416 return port_attr->show(port, buf);
417 }
418
419 #define CUSTOM_PORT_ATTR(_name, _format, _expression) \
420 static ssize_t _name##_show(struct veth_port *port, char *buf) \
421 { \
422 return sprintf(buf, _format, _expression); \
423 } \
424 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
425
426 #define SIMPLE_PORT_ATTR(_name) \
427 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
428
429 SIMPLE_PORT_ATTR(promiscuous);
430 SIMPLE_PORT_ATTR(num_mcast);
431 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
433 CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
434
435 #define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
436 static struct attribute *veth_port_default_attrs[] = {
437 GET_PORT_ATTR(mac_addr),
438 GET_PORT_ATTR(lpar_map),
439 GET_PORT_ATTR(stopped_map),
440 GET_PORT_ATTR(promiscuous),
441 GET_PORT_ATTR(num_mcast),
442 NULL
443 };
444
445 static const struct sysfs_ops veth_port_sysfs_ops = {
446 .show = veth_port_attribute_show
447 };
448
449 static struct kobj_type veth_port_ktype = {
450 .sysfs_ops = &veth_port_sysfs_ops,
451 .default_attrs = veth_port_default_attrs
452 };
453
454 /*
455 * LPAR connection code
456 */
457
458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459 {
460 schedule_delayed_work(&cnx->statemachine_wq, 0);
461 }
462
463 static void veth_take_cap(struct veth_lpar_connection *cnx,
464 struct veth_lpevent *event)
465 {
466 unsigned long flags;
467
468 spin_lock_irqsave(&cnx->lock, flags);
469 /* Receiving caps may mean the other end has just come up, so
470 * we need to reload the instance ID of the far end */
471 cnx->dst_inst =
472 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473 HvLpEvent_Type_VirtualLan);
474
475 if (cnx->state & VETH_STATE_GOTCAPS) {
476 veth_error("Received a second capabilities from LPAR %d.\n",
477 cnx->remote_lp);
478 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
480 } else {
481 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482 cnx->state |= VETH_STATE_GOTCAPS;
483 veth_kick_statemachine(cnx);
484 }
485 spin_unlock_irqrestore(&cnx->lock, flags);
486 }
487
488 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
489 struct veth_lpevent *event)
490 {
491 unsigned long flags;
492
493 spin_lock_irqsave(&cnx->lock, flags);
494 if (cnx->state & VETH_STATE_GOTCAPACK) {
495 veth_error("Received a second capabilities ack from LPAR %d.\n",
496 cnx->remote_lp);
497 } else {
498 memcpy(&cnx->cap_ack_event, event,
499 sizeof(cnx->cap_ack_event));
500 cnx->state |= VETH_STATE_GOTCAPACK;
501 veth_kick_statemachine(cnx);
502 }
503 spin_unlock_irqrestore(&cnx->lock, flags);
504 }
505
506 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
507 struct veth_lpevent *event)
508 {
509 unsigned long flags;
510
511 spin_lock_irqsave(&cnx->lock, flags);
512 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
513
514 /* Avoid kicking the statemachine once we're shutdown.
515 * It's unnecessary and it could break veth_stop_connection(). */
516
517 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518 cnx->state |= VETH_STATE_RESET;
519 veth_kick_statemachine(cnx);
520 }
521 spin_unlock_irqrestore(&cnx->lock, flags);
522 }
523
524 static void veth_handle_ack(struct veth_lpevent *event)
525 {
526 HvLpIndex rlp = event->base_event.xTargetLp;
527 struct veth_lpar_connection *cnx = veth_cnx[rlp];
528
529 BUG_ON(! cnx);
530
531 switch (event->base_event.xSubtype) {
532 case VETH_EVENT_CAP:
533 veth_take_cap_ack(cnx, event);
534 break;
535 case VETH_EVENT_MONITOR:
536 veth_take_monitor_ack(cnx, event);
537 break;
538 default:
539 veth_error("Unknown ack type %d from LPAR %d.\n",
540 event->base_event.xSubtype, rlp);
541 };
542 }
543
544 static void veth_handle_int(struct veth_lpevent *event)
545 {
546 HvLpIndex rlp = event->base_event.xSourceLp;
547 struct veth_lpar_connection *cnx = veth_cnx[rlp];
548 unsigned long flags;
549 int i, acked = 0;
550
551 BUG_ON(! cnx);
552
553 switch (event->base_event.xSubtype) {
554 case VETH_EVENT_CAP:
555 veth_take_cap(cnx, event);
556 break;
557 case VETH_EVENT_MONITOR:
558 /* do nothing... this'll hang out here til we're dead,
559 * and the hypervisor will return it for us. */
560 break;
561 case VETH_EVENT_FRAMES_ACK:
562 spin_lock_irqsave(&cnx->lock, flags);
563
564 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565 u16 msgnum = event->u.frames_ack_data.token[i];
566
567 if (msgnum < VETH_NUMBUFFERS) {
568 veth_recycle_msg(cnx, cnx->msgs + msgnum);
569 cnx->outstanding_tx--;
570 acked++;
571 }
572 }
573
574 if (acked > 0) {
575 cnx->last_contact = jiffies;
576 veth_wake_queues(cnx);
577 }
578
579 spin_unlock_irqrestore(&cnx->lock, flags);
580 break;
581 case VETH_EVENT_FRAMES:
582 veth_receive(cnx, event);
583 break;
584 default:
585 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586 event->base_event.xSubtype, rlp);
587 };
588 }
589
590 static void veth_handle_event(struct HvLpEvent *event)
591 {
592 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
593
594 if (hvlpevent_is_ack(event))
595 veth_handle_ack(veth_event);
596 else
597 veth_handle_int(veth_event);
598 }
599
600 static int veth_process_caps(struct veth_lpar_connection *cnx)
601 {
602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
603 int num_acks_needed;
604
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
608 if ( (remote_caps->num_buffers == 0) ||
609 (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
610 (remote_caps->ack_threshold == 0) ||
611 (cnx->ack_timeout == 0) ) {
612 veth_error("Received incompatible capabilities from LPAR %d.\n",
613 cnx->remote_lp);
614 return HvLpEvent_Rc_InvalidSubtypeData;
615 }
616
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
619
620 if (cnx->num_ack_events < num_acks_needed) {
621 int num;
622
623 num = veth_allocate_events(cnx->remote_lp,
624 num_acks_needed-cnx->num_ack_events);
625 if (num > 0)
626 cnx->num_ack_events += num;
627
628 if (cnx->num_ack_events < num_acks_needed) {
629 veth_error("Couldn't allocate enough ack events "
630 "for LPAR %d.\n", cnx->remote_lp);
631
632 return HvLpEvent_Rc_BufferNotAvailable;
633 }
634 }
635
636
637 return HvLpEvent_Rc_Good;
638 }
639
640 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 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
806 " (state = 0x%04lx)\n", rlp, cnx->state);
807 cnx->state |= VETH_STATE_SHUTDOWN;
808 spin_unlock_irq(&cnx->lock);
809 }
810
811 static int veth_init_connection(u8 rlp)
812 {
813 struct veth_lpar_connection *cnx;
814 struct veth_msg *msgs;
815 int i;
816
817 if ( (rlp == this_lp) ||
818 ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
819 return 0;
820
821 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
822 if (! cnx)
823 return -ENOMEM;
824
825 cnx->remote_lp = rlp;
826 spin_lock_init(&cnx->lock);
827 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
828
829 init_timer(&cnx->ack_timer);
830 cnx->ack_timer.function = veth_timed_ack;
831 cnx->ack_timer.data = (unsigned long) cnx;
832
833 init_timer(&cnx->reset_timer);
834 cnx->reset_timer.function = veth_timed_reset;
835 cnx->reset_timer.data = (unsigned long) cnx;
836 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
837
838 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
839
840 veth_cnx[rlp] = cnx;
841
842 /* This gets us 1 reference, which is held on behalf of the driver
843 * infrastructure. It's released at module unload. */
844 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
845
846 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
847 if (! msgs) {
848 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
849 return -ENOMEM;
850 }
851
852 cnx->msgs = msgs;
853
854 for (i = 0; i < VETH_NUMBUFFERS; i++) {
855 msgs[i].token = i;
856 veth_stack_push(cnx, msgs + i);
857 }
858
859 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
860
861 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
862 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
863 return -ENOMEM;
864 }
865
866 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
867 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
868 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
869
870 return 0;
871 }
872
873 static void veth_stop_connection(struct veth_lpar_connection *cnx)
874 {
875 if (!cnx)
876 return;
877
878 spin_lock_irq(&cnx->lock);
879 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
880 veth_kick_statemachine(cnx);
881 spin_unlock_irq(&cnx->lock);
882
883 /* There's a slim chance the reset code has just queued the
884 * statemachine to run in five seconds. If so we need to cancel
885 * that and requeue the work to run now. */
886 if (cancel_delayed_work(&cnx->statemachine_wq)) {
887 spin_lock_irq(&cnx->lock);
888 veth_kick_statemachine(cnx);
889 spin_unlock_irq(&cnx->lock);
890 }
891
892 /* Wait for the state machine to run. */
893 flush_scheduled_work();
894 }
895
896 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
897 {
898 if (!cnx)
899 return;
900
901 if (cnx->num_events > 0)
902 mf_deallocate_lp_events(cnx->remote_lp,
903 HvLpEvent_Type_VirtualLan,
904 cnx->num_events,
905 NULL, NULL);
906 if (cnx->num_ack_events > 0)
907 mf_deallocate_lp_events(cnx->remote_lp,
908 HvLpEvent_Type_VirtualLan,
909 cnx->num_ack_events,
910 NULL, NULL);
911
912 kfree(cnx->msgs);
913 veth_cnx[cnx->remote_lp] = NULL;
914 kfree(cnx);
915 }
916
917 static void veth_release_connection(struct kobject *kobj)
918 {
919 struct veth_lpar_connection *cnx;
920 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
921 veth_stop_connection(cnx);
922 veth_destroy_connection(cnx);
923 }
924
925 /*
926 * net_device code
927 */
928
929 static int veth_open(struct net_device *dev)
930 {
931 netif_start_queue(dev);
932 return 0;
933 }
934
935 static int veth_close(struct net_device *dev)
936 {
937 netif_stop_queue(dev);
938 return 0;
939 }
940
941 static int veth_change_mtu(struct net_device *dev, int new_mtu)
942 {
943 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
944 return -EINVAL;
945 dev->mtu = new_mtu;
946 return 0;
947 }
948
949 static void veth_set_multicast_list(struct net_device *dev)
950 {
951 struct veth_port *port = netdev_priv(dev);
952 unsigned long flags;
953
954 write_lock_irqsave(&port->mcast_gate, flags);
955
956 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
957 (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
958 port->promiscuous = 1;
959 } else {
960 struct netdev_hw_addr *ha;
961
962 port->promiscuous = 0;
963
964 /* Update table */
965 port->num_mcast = 0;
966
967 netdev_for_each_mc_addr(ha, dev) {
968 u8 *addr = ha->addr;
969 u64 xaddr = 0;
970
971 if (addr[0] & 0x01) {/* multicast address? */
972 memcpy(&xaddr, addr, ETH_ALEN);
973 port->mcast_addr[port->num_mcast] = xaddr;
974 port->num_mcast++;
975 }
976 }
977 }
978
979 write_unlock_irqrestore(&port->mcast_gate, flags);
980 }
981
982 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
983 {
984 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
985 info->driver[sizeof(info->driver) - 1] = '\0';
986 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
987 info->version[sizeof(info->version) - 1] = '\0';
988 }
989
990 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
991 {
992 ecmd->supported = (SUPPORTED_1000baseT_Full
993 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
994 ecmd->advertising = (SUPPORTED_1000baseT_Full
995 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
996 ecmd->port = PORT_FIBRE;
997 ecmd->transceiver = XCVR_INTERNAL;
998 ecmd->phy_address = 0;
999 ecmd->speed = SPEED_1000;
1000 ecmd->duplex = DUPLEX_FULL;
1001 ecmd->autoneg = AUTONEG_ENABLE;
1002 ecmd->maxtxpkt = 120;
1003 ecmd->maxrxpkt = 120;
1004 return 0;
1005 }
1006
1007 static u32 veth_get_link(struct net_device *dev)
1008 {
1009 return 1;
1010 }
1011
1012 static const struct ethtool_ops ops = {
1013 .get_drvinfo = veth_get_drvinfo,
1014 .get_settings = veth_get_settings,
1015 .get_link = veth_get_link,
1016 };
1017
1018 static const struct net_device_ops veth_netdev_ops = {
1019 .ndo_open = veth_open,
1020 .ndo_stop = veth_close,
1021 .ndo_start_xmit = veth_start_xmit,
1022 .ndo_change_mtu = veth_change_mtu,
1023 .ndo_set_multicast_list = veth_set_multicast_list,
1024 .ndo_set_mac_address = NULL,
1025 .ndo_validate_addr = eth_validate_addr,
1026 };
1027
1028 static struct net_device *veth_probe_one(int vlan,
1029 struct vio_dev *vio_dev)
1030 {
1031 struct net_device *dev;
1032 struct veth_port *port;
1033 struct device *vdev = &vio_dev->dev;
1034 int i, rc;
1035 const unsigned char *mac_addr;
1036
1037 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1038 if (mac_addr == NULL)
1039 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1040 if (mac_addr == NULL) {
1041 veth_error("Unable to fetch MAC address from device tree.\n");
1042 return NULL;
1043 }
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 = netdev_priv(dev);
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 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1069
1070 dev->mtu = VETH_MAX_MTU;
1071
1072 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1073
1074 dev->netdev_ops = &veth_netdev_ops;
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 = netdev_priv(dev);
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(port->dev, 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 = netdev_priv(dev);
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 NETDEV_TX_OK;
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 NETDEV_TX_OK;
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(msg->dev, 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 = netdev_priv(dev);
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 = netdev_priv(dev);
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
1310 spin_lock_irqsave(&cnx->lock, flags);
1311
1312 if (cnx->outstanding_tx > 0) {
1313 trigger_time = cnx->last_contact + cnx->reset_timeout;
1314
1315 if (trigger_time < jiffies) {
1316 cnx->state |= VETH_STATE_RESET;
1317 veth_kick_statemachine(cnx);
1318 veth_error("%d packets not acked by LPAR %d within %d "
1319 "seconds, resetting.\n",
1320 cnx->outstanding_tx, cnx->remote_lp,
1321 cnx->reset_timeout / HZ);
1322 } else {
1323 /* Reschedule the timer */
1324 trigger_time = jiffies + cnx->reset_timeout;
1325 mod_timer(&cnx->reset_timer, trigger_time);
1326 }
1327 }
1328
1329 spin_unlock_irqrestore(&cnx->lock, flags);
1330 }
1331
1332 /*
1333 * Rx path
1334 */
1335
1336 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1337 {
1338 int wanted = 0;
1339 int i;
1340 unsigned long flags;
1341
1342 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1343 return 1;
1344
1345 read_lock_irqsave(&port->mcast_gate, flags);
1346
1347 if (port->promiscuous) {
1348 wanted = 1;
1349 goto out;
1350 }
1351
1352 for (i = 0; i < port->num_mcast; ++i) {
1353 if (port->mcast_addr[i] == mac_addr) {
1354 wanted = 1;
1355 break;
1356 }
1357 }
1358
1359 out:
1360 read_unlock_irqrestore(&port->mcast_gate, flags);
1361
1362 return wanted;
1363 }
1364
1365 struct dma_chunk {
1366 u64 addr;
1367 u64 size;
1368 };
1369
1370 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1371
1372 static inline void veth_build_dma_list(struct dma_chunk *list,
1373 unsigned char *p, unsigned long length)
1374 {
1375 unsigned long done;
1376 int i = 1;
1377
1378 list[0].addr = iseries_hv_addr(p);
1379 list[0].size = min(length,
1380 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1381
1382 done = list[0].size;
1383 while (done < length) {
1384 list[i].addr = iseries_hv_addr(p + done);
1385 list[i].size = min(length-done, PAGE_SIZE);
1386 done += list[i].size;
1387 i++;
1388 }
1389 }
1390
1391 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1392 {
1393 HvLpEvent_Rc rc;
1394
1395 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1396 0, &cnx->pending_acks);
1397
1398 if (rc != HvLpEvent_Rc_Good)
1399 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1400 cnx->remote_lp, (int)rc);
1401
1402 cnx->num_pending_acks = 0;
1403 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1404 }
1405
1406 static void veth_receive(struct veth_lpar_connection *cnx,
1407 struct veth_lpevent *event)
1408 {
1409 struct veth_frames_data *senddata = &event->u.frames_data;
1410 int startchunk = 0;
1411 int nchunks;
1412 unsigned long flags;
1413 HvLpDma_Rc rc;
1414
1415 do {
1416 u16 length = 0;
1417 struct sk_buff *skb;
1418 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1419 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1420 u64 dest;
1421 HvLpVirtualLanIndex vlan;
1422 struct net_device *dev;
1423 struct veth_port *port;
1424
1425 memset(local_list, 0, sizeof(local_list));
1426 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1427
1428 /* a 0 address marks the end of the valid entries */
1429 if (senddata->addr[startchunk] == 0)
1430 break;
1431
1432 /* make sure that we have at least 1 EOF entry in the
1433 * remaining entries */
1434 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1435 veth_error("Missing EOF fragment in event "
1436 "eofmask = 0x%x startchunk = %d\n",
1437 (unsigned)senddata->eofmask,
1438 startchunk);
1439 break;
1440 }
1441
1442 /* build list of chunks in this frame */
1443 nchunks = 0;
1444 do {
1445 remote_list[nchunks].addr =
1446 (u64) senddata->addr[startchunk+nchunks] << 32;
1447 remote_list[nchunks].size =
1448 senddata->len[startchunk+nchunks];
1449 length += remote_list[nchunks].size;
1450 } while (! (senddata->eofmask &
1451 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1452
1453 /* length == total length of all chunks */
1454 /* nchunks == # of chunks in this frame */
1455
1456 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1457 veth_error("Received oversize frame from LPAR %d "
1458 "(length = %d)\n",
1459 cnx->remote_lp, length);
1460 continue;
1461 }
1462
1463 skb = alloc_skb(length, GFP_ATOMIC);
1464 if (!skb)
1465 continue;
1466
1467 veth_build_dma_list(local_list, skb->data, length);
1468
1469 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1470 event->base_event.xSourceLp,
1471 HvLpDma_Direction_RemoteToLocal,
1472 cnx->src_inst,
1473 cnx->dst_inst,
1474 HvLpDma_AddressType_RealAddress,
1475 HvLpDma_AddressType_TceIndex,
1476 iseries_hv_addr(&local_list),
1477 iseries_hv_addr(&remote_list),
1478 length);
1479 if (rc != HvLpDma_Rc_Good) {
1480 dev_kfree_skb_irq(skb);
1481 continue;
1482 }
1483
1484 vlan = skb->data[9];
1485 dev = veth_dev[vlan];
1486 if (! dev) {
1487 /*
1488 * Some earlier versions of the driver sent
1489 * broadcasts down all connections, even to lpars
1490 * that weren't on the relevant vlan. So ignore
1491 * packets belonging to a vlan we're not on.
1492 * We can also be here if we receive packets while
1493 * the driver is going down, because then dev is NULL.
1494 */
1495 dev_kfree_skb_irq(skb);
1496 continue;
1497 }
1498
1499 port = netdev_priv(dev);
1500 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1501
1502 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1503 dev_kfree_skb_irq(skb);
1504 continue;
1505 }
1506 if (! veth_frame_wanted(port, dest)) {
1507 dev_kfree_skb_irq(skb);
1508 continue;
1509 }
1510
1511 skb_put(skb, length);
1512 skb->protocol = eth_type_trans(skb, dev);
1513 skb->ip_summed = CHECKSUM_NONE;
1514 netif_rx(skb); /* send it up */
1515 dev->stats.rx_packets++;
1516 dev->stats.rx_bytes += length;
1517 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1518
1519 /* Ack it */
1520 spin_lock_irqsave(&cnx->lock, flags);
1521 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1522
1523 cnx->pending_acks[cnx->num_pending_acks++] =
1524 event->base_event.xCorrelationToken;
1525
1526 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1527 (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1528 veth_flush_acks(cnx);
1529
1530 spin_unlock_irqrestore(&cnx->lock, flags);
1531 }
1532
1533 static void veth_timed_ack(unsigned long ptr)
1534 {
1535 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1536 unsigned long flags;
1537
1538 /* Ack all the events */
1539 spin_lock_irqsave(&cnx->lock, flags);
1540 if (cnx->num_pending_acks > 0)
1541 veth_flush_acks(cnx);
1542
1543 /* Reschedule the timer */
1544 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1545 add_timer(&cnx->ack_timer);
1546 spin_unlock_irqrestore(&cnx->lock, flags);
1547 }
1548
1549 static int veth_remove(struct vio_dev *vdev)
1550 {
1551 struct veth_lpar_connection *cnx;
1552 struct net_device *dev;
1553 struct veth_port *port;
1554 int i;
1555
1556 dev = veth_dev[vdev->unit_address];
1557
1558 if (! dev)
1559 return 0;
1560
1561 port = netdev_priv(dev);
1562
1563 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1564 cnx = veth_cnx[i];
1565
1566 if (cnx && (port->lpar_map & (1 << i))) {
1567 /* Drop our reference to connections on our VLAN */
1568 kobject_put(&cnx->kobject);
1569 }
1570 }
1571
1572 veth_dev[vdev->unit_address] = NULL;
1573 kobject_del(&port->kobject);
1574 kobject_put(&port->kobject);
1575 unregister_netdev(dev);
1576 free_netdev(dev);
1577
1578 return 0;
1579 }
1580
1581 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1582 {
1583 int i = vdev->unit_address;
1584 struct net_device *dev;
1585 struct veth_port *port;
1586
1587 dev = veth_probe_one(i, vdev);
1588 if (dev == NULL) {
1589 veth_remove(vdev);
1590 return 1;
1591 }
1592 veth_dev[i] = dev;
1593
1594 port = (struct veth_port*)netdev_priv(dev);
1595
1596 /* Start the state machine on each connection on this vlan. If we're
1597 * the first dev to do so this will commence link negotiation */
1598 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1599 struct veth_lpar_connection *cnx;
1600
1601 if (! (port->lpar_map & (1 << i)))
1602 continue;
1603
1604 cnx = veth_cnx[i];
1605 if (!cnx)
1606 continue;
1607
1608 kobject_get(&cnx->kobject);
1609 veth_kick_statemachine(cnx);
1610 }
1611
1612 return 0;
1613 }
1614
1615 /**
1616 * veth_device_table: Used by vio.c to match devices that we
1617 * support.
1618 */
1619 static struct vio_device_id veth_device_table[] __devinitdata = {
1620 { "network", "IBM,iSeries-l-lan" },
1621 { "", "" }
1622 };
1623 MODULE_DEVICE_TABLE(vio, veth_device_table);
1624
1625 static struct vio_driver veth_driver = {
1626 .id_table = veth_device_table,
1627 .probe = veth_probe,
1628 .remove = veth_remove,
1629 .driver = {
1630 .name = DRV_NAME,
1631 .owner = THIS_MODULE,
1632 }
1633 };
1634
1635 /*
1636 * Module initialization/cleanup
1637 */
1638
1639 static void __exit veth_module_cleanup(void)
1640 {
1641 int i;
1642 struct veth_lpar_connection *cnx;
1643
1644 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1645 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1646
1647 /* Make sure any work queued from Hypervisor callbacks is finished. */
1648 flush_scheduled_work();
1649
1650 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1651 cnx = veth_cnx[i];
1652
1653 if (!cnx)
1654 continue;
1655
1656 /* Remove the connection from sysfs */
1657 kobject_del(&cnx->kobject);
1658 /* Drop the driver's reference to the connection */
1659 kobject_put(&cnx->kobject);
1660 }
1661
1662 /* Unregister the driver, which will close all the netdevs and stop
1663 * the connections when they're no longer referenced. */
1664 vio_unregister_driver(&veth_driver);
1665 }
1666 module_exit(veth_module_cleanup);
1667
1668 static int __init veth_module_init(void)
1669 {
1670 int i;
1671 int rc;
1672
1673 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1674 return -ENODEV;
1675
1676 this_lp = HvLpConfig_getLpIndex_outline();
1677
1678 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1679 rc = veth_init_connection(i);
1680 if (rc != 0)
1681 goto error;
1682 }
1683
1684 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1685 &veth_handle_event);
1686
1687 rc = vio_register_driver(&veth_driver);
1688 if (rc != 0)
1689 goto error;
1690
1691 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1692 struct kobject *kobj;
1693
1694 if (!veth_cnx[i])
1695 continue;
1696
1697 kobj = &veth_cnx[i]->kobject;
1698 /* If the add failes, complain but otherwise continue */
1699 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1700 "cnx%.2d", veth_cnx[i]->remote_lp))
1701 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1702 }
1703
1704 return 0;
1705
1706 error:
1707 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1708 veth_destroy_connection(veth_cnx[i]);
1709 }
1710
1711 return rc;
1712 }
1713 module_init(veth_module_init);
Tomato firmware and modifications.