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. */
veth_stack_push(struct veth_lpar_connection * cnx,struct veth_msg * msg)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. */
veth_stack_pop(struct veth_lpar_connection * cnx)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. */
veth_stack_is_empty(struct veth_lpar_connection * cnx)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
veth_signalevent(struct veth_lpar_connection * cnx,u16 subtype,HvLpEvent_AckInd ackind,HvLpEvent_AckType acktype,u64 token,u64 data1,u64 data2,u64 data3,u64 data4,u64 data5)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
veth_signaldata(struct veth_lpar_connection * cnx,u16 subtype,u64 token,void * data)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
veth_complete_allocation(void * parm,int number)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
veth_allocate_events(HvLpIndex rlp,int number)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
veth_cnx_attribute_show(struct kobject * kobj,struct attribute * attr,char * buf)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
veth_port_attribute_show(struct kobject * kobj,struct attribute * attr,char * buf)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
veth_kick_statemachine(struct veth_lpar_connection * cnx)458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459 {
460 schedule_delayed_work(&cnx->statemachine_wq, 0);
461 }
462
veth_take_cap(struct veth_lpar_connection * cnx,struct veth_lpevent * event)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
veth_take_cap_ack(struct veth_lpar_connection * cnx,struct veth_lpevent * event)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
veth_take_monitor_ack(struct veth_lpar_connection * cnx,struct veth_lpevent * event)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
veth_handle_ack(struct veth_lpevent * event)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
veth_handle_int(struct veth_lpevent * event)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
veth_handle_event(struct HvLpEvent * event)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
veth_process_caps(struct veth_lpar_connection * cnx)600 static int veth_process_caps(struct veth_lpar_connection *cnx)
601 {
602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
603 int num_acks_needed;
604
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
608 if ( (remote_caps->num_buffers == 0) ||
609 (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
610 (remote_caps->ack_threshold == 0) ||
611 (cnx->ack_timeout == 0) ) {
612 veth_error("Received incompatible capabilities from LPAR %d.\n",
613 cnx->remote_lp);
614 return HvLpEvent_Rc_InvalidSubtypeData;
615 }
616
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
619
620 /* FIXME: locking on num_ack_events? */
621 if (cnx->num_ack_events < num_acks_needed) {
622 int num;
623
624 num = veth_allocate_events(cnx->remote_lp,
625 num_acks_needed-cnx->num_ack_events);
626 if (num > 0)
627 cnx->num_ack_events += num;
628
629 if (cnx->num_ack_events < num_acks_needed) {
630 veth_error("Couldn't allocate enough ack events "
631 "for LPAR %d.\n", cnx->remote_lp);
632
633 return HvLpEvent_Rc_BufferNotAvailable;
634 }
635 }
636
637
638 return HvLpEvent_Rc_Good;
639 }
640
641 /* FIXME: The gotos here are a bit dubious */
veth_statemachine(struct work_struct * work)642 static void veth_statemachine(struct work_struct *work)
643 {
644 struct veth_lpar_connection *cnx =
645 container_of(work, struct veth_lpar_connection,
646 statemachine_wq.work);
647 int rlp = cnx->remote_lp;
648 int rc;
649
650 spin_lock_irq(&cnx->lock);
651
652 restart:
653 if (cnx->state & VETH_STATE_RESET) {
654 if (cnx->state & VETH_STATE_OPEN)
655 HvCallEvent_closeLpEventPath(cnx->remote_lp,
656 HvLpEvent_Type_VirtualLan);
657
658 /*
659 * Reset ack data. This prevents the ack_timer actually
660 * doing anything, even if it runs one more time when
661 * we drop the lock below.
662 */
663 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
664 cnx->num_pending_acks = 0;
665
666 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
667 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
668 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
669 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
670
671 /* Clean up any leftover messages */
672 if (cnx->msgs) {
673 int i;
674 for (i = 0; i < VETH_NUMBUFFERS; ++i)
675 veth_recycle_msg(cnx, cnx->msgs + i);
676 }
677
678 cnx->outstanding_tx = 0;
679 veth_wake_queues(cnx);
680
681 /* Drop the lock so we can do stuff that might sleep or
682 * take other locks. */
683 spin_unlock_irq(&cnx->lock);
684
685 del_timer_sync(&cnx->ack_timer);
686 del_timer_sync(&cnx->reset_timer);
687
688 spin_lock_irq(&cnx->lock);
689
690 if (cnx->state & VETH_STATE_RESET)
691 goto restart;
692
693 /* Hack, wait for the other end to reset itself. */
694 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
695 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
696 goto out;
697 }
698 }
699
700 if (cnx->state & VETH_STATE_SHUTDOWN)
701 /* It's all over, do nothing */
702 goto out;
703
704 if ( !(cnx->state & VETH_STATE_OPEN) ) {
705 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
706 goto cant_cope;
707
708 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
709 cnx->src_inst =
710 HvCallEvent_getSourceLpInstanceId(rlp,
711 HvLpEvent_Type_VirtualLan);
712 cnx->dst_inst =
713 HvCallEvent_getTargetLpInstanceId(rlp,
714 HvLpEvent_Type_VirtualLan);
715 cnx->state |= VETH_STATE_OPEN;
716 }
717
718 if ( (cnx->state & VETH_STATE_OPEN) &&
719 !(cnx->state & VETH_STATE_SENTMON) ) {
720 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
721 HvLpEvent_AckInd_DoAck,
722 HvLpEvent_AckType_DeferredAck,
723 0, 0, 0, 0, 0, 0);
724
725 if (rc == HvLpEvent_Rc_Good) {
726 cnx->state |= VETH_STATE_SENTMON;
727 } else {
728 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
729 (rc != HvLpEvent_Rc_PathClosed) )
730 veth_error("Error sending monitor to LPAR %d, "
731 "rc = %d\n", rlp, rc);
732
733 /* Oh well, hope we get a cap from the other
734 * end and do better when that kicks us */
735 goto out;
736 }
737 }
738
739 if ( (cnx->state & VETH_STATE_OPEN) &&
740 !(cnx->state & VETH_STATE_SENTCAPS)) {
741 u64 *rawcap = (u64 *)&cnx->local_caps;
742
743 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
744 HvLpEvent_AckInd_DoAck,
745 HvLpEvent_AckType_ImmediateAck,
746 0, rawcap[0], rawcap[1], rawcap[2],
747 rawcap[3], rawcap[4]);
748
749 if (rc == HvLpEvent_Rc_Good) {
750 cnx->state |= VETH_STATE_SENTCAPS;
751 } else {
752 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
753 (rc != HvLpEvent_Rc_PathClosed) )
754 veth_error("Error sending caps to LPAR %d, "
755 "rc = %d\n", rlp, rc);
756
757 /* Oh well, hope we get a cap from the other
758 * end and do better when that kicks us */
759 goto out;
760 }
761 }
762
763 if ((cnx->state & VETH_STATE_GOTCAPS) &&
764 !(cnx->state & VETH_STATE_SENTCAPACK)) {
765 struct veth_cap_data *remote_caps = &cnx->remote_caps;
766
767 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
768 sizeof(*remote_caps));
769
770 spin_unlock_irq(&cnx->lock);
771 rc = veth_process_caps(cnx);
772 spin_lock_irq(&cnx->lock);
773
774 /* We dropped the lock, so recheck for anything which
775 * might mess us up */
776 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
777 goto restart;
778
779 cnx->cap_event.base_event.xRc = rc;
780 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
781 if (rc == HvLpEvent_Rc_Good)
782 cnx->state |= VETH_STATE_SENTCAPACK;
783 else
784 goto cant_cope;
785 }
786
787 if ((cnx->state & VETH_STATE_GOTCAPACK) &&
788 (cnx->state & VETH_STATE_GOTCAPS) &&
789 !(cnx->state & VETH_STATE_READY)) {
790 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
791 /* Start the ACK timer */
792 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
793 add_timer(&cnx->ack_timer);
794 cnx->state |= VETH_STATE_READY;
795 } else {
796 veth_error("Caps rejected by LPAR %d, rc = %d\n",
797 rlp, cnx->cap_ack_event.base_event.xRc);
798 goto cant_cope;
799 }
800 }
801
802 out:
803 spin_unlock_irq(&cnx->lock);
804 return;
805
806 cant_cope:
807 /* FIXME: we get here if something happens we really can't
808 * cope with. The link will never work once we get here, and
809 * all we can do is not lock the rest of the system up */
810 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
811 " (state = 0x%04lx)\n", rlp, cnx->state);
812 cnx->state |= VETH_STATE_SHUTDOWN;
813 spin_unlock_irq(&cnx->lock);
814 }
815
veth_init_connection(u8 rlp)816 static int veth_init_connection(u8 rlp)
817 {
818 struct veth_lpar_connection *cnx;
819 struct veth_msg *msgs;
820 int i;
821
822 if ( (rlp == this_lp) ||
823 ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
824 return 0;
825
826 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
827 if (! cnx)
828 return -ENOMEM;
829
830 cnx->remote_lp = rlp;
831 spin_lock_init(&cnx->lock);
832 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
833
834 init_timer(&cnx->ack_timer);
835 cnx->ack_timer.function = veth_timed_ack;
836 cnx->ack_timer.data = (unsigned long) cnx;
837
838 init_timer(&cnx->reset_timer);
839 cnx->reset_timer.function = veth_timed_reset;
840 cnx->reset_timer.data = (unsigned long) cnx;
841 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
842
843 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
844
845 veth_cnx[rlp] = cnx;
846
847 /* This gets us 1 reference, which is held on behalf of the driver
848 * infrastructure. It's released at module unload. */
849 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
850
851 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
852 if (! msgs) {
853 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
854 return -ENOMEM;
855 }
856
857 cnx->msgs = msgs;
858
859 for (i = 0; i < VETH_NUMBUFFERS; i++) {
860 msgs[i].token = i;
861 veth_stack_push(cnx, msgs + i);
862 }
863
864 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
865
866 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
867 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
868 return -ENOMEM;
869 }
870
871 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
872 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
873 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
874
875 return 0;
876 }
877
veth_stop_connection(struct veth_lpar_connection * cnx)878 static void veth_stop_connection(struct veth_lpar_connection *cnx)
879 {
880 if (!cnx)
881 return;
882
883 spin_lock_irq(&cnx->lock);
884 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
885 veth_kick_statemachine(cnx);
886 spin_unlock_irq(&cnx->lock);
887
888 /* ensure the statemachine runs now and waits for its completion */
889 flush_delayed_work_sync(&cnx->statemachine_wq);
890 }
891
veth_destroy_connection(struct veth_lpar_connection * cnx)892 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
893 {
894 if (!cnx)
895 return;
896
897 if (cnx->num_events > 0)
898 mf_deallocate_lp_events(cnx->remote_lp,
899 HvLpEvent_Type_VirtualLan,
900 cnx->num_events,
901 NULL, NULL);
902 if (cnx->num_ack_events > 0)
903 mf_deallocate_lp_events(cnx->remote_lp,
904 HvLpEvent_Type_VirtualLan,
905 cnx->num_ack_events,
906 NULL, NULL);
907
908 kfree(cnx->msgs);
909 veth_cnx[cnx->remote_lp] = NULL;
910 kfree(cnx);
911 }
912
veth_release_connection(struct kobject * kobj)913 static void veth_release_connection(struct kobject *kobj)
914 {
915 struct veth_lpar_connection *cnx;
916 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
917 veth_stop_connection(cnx);
918 veth_destroy_connection(cnx);
919 }
920
921 /*
922 * net_device code
923 */
924
veth_open(struct net_device * dev)925 static int veth_open(struct net_device *dev)
926 {
927 netif_start_queue(dev);
928 return 0;
929 }
930
veth_close(struct net_device * dev)931 static int veth_close(struct net_device *dev)
932 {
933 netif_stop_queue(dev);
934 return 0;
935 }
936
veth_change_mtu(struct net_device * dev,int new_mtu)937 static int veth_change_mtu(struct net_device *dev, int new_mtu)
938 {
939 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
940 return -EINVAL;
941 dev->mtu = new_mtu;
942 return 0;
943 }
944
veth_set_multicast_list(struct net_device * dev)945 static void veth_set_multicast_list(struct net_device *dev)
946 {
947 struct veth_port *port = netdev_priv(dev);
948 unsigned long flags;
949
950 write_lock_irqsave(&port->mcast_gate, flags);
951
952 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
953 (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
954 port->promiscuous = 1;
955 } else {
956 struct netdev_hw_addr *ha;
957
958 port->promiscuous = 0;
959
960 /* Update table */
961 port->num_mcast = 0;
962
963 netdev_for_each_mc_addr(ha, dev) {
964 u8 *addr = ha->addr;
965 u64 xaddr = 0;
966
967 memcpy(&xaddr, addr, ETH_ALEN);
968 port->mcast_addr[port->num_mcast] = xaddr;
969 port->num_mcast++;
970 }
971 }
972
973 write_unlock_irqrestore(&port->mcast_gate, flags);
974 }
975
veth_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * info)976 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
977 {
978 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
979 info->driver[sizeof(info->driver) - 1] = '\0';
980 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
981 info->version[sizeof(info->version) - 1] = '\0';
982 }
983
veth_get_settings(struct net_device * dev,struct ethtool_cmd * ecmd)984 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
985 {
986 ecmd->supported = (SUPPORTED_1000baseT_Full
987 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
988 ecmd->advertising = (SUPPORTED_1000baseT_Full
989 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
990 ecmd->port = PORT_FIBRE;
991 ecmd->transceiver = XCVR_INTERNAL;
992 ecmd->phy_address = 0;
993 ecmd->speed = SPEED_1000;
994 ecmd->duplex = DUPLEX_FULL;
995 ecmd->autoneg = AUTONEG_ENABLE;
996 ecmd->maxtxpkt = 120;
997 ecmd->maxrxpkt = 120;
998 return 0;
999 }
1000
1001 static const struct ethtool_ops ops = {
1002 .get_drvinfo = veth_get_drvinfo,
1003 .get_settings = veth_get_settings,
1004 .get_link = ethtool_op_get_link,
1005 };
1006
1007 static const struct net_device_ops veth_netdev_ops = {
1008 .ndo_open = veth_open,
1009 .ndo_stop = veth_close,
1010 .ndo_start_xmit = veth_start_xmit,
1011 .ndo_change_mtu = veth_change_mtu,
1012 .ndo_set_rx_mode = veth_set_multicast_list,
1013 .ndo_set_mac_address = NULL,
1014 .ndo_validate_addr = eth_validate_addr,
1015 };
1016
veth_probe_one(int vlan,struct vio_dev * vio_dev)1017 static struct net_device *veth_probe_one(int vlan,
1018 struct vio_dev *vio_dev)
1019 {
1020 struct net_device *dev;
1021 struct veth_port *port;
1022 struct device *vdev = &vio_dev->dev;
1023 int i, rc;
1024 const unsigned char *mac_addr;
1025
1026 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1027 if (mac_addr == NULL)
1028 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1029 if (mac_addr == NULL) {
1030 veth_error("Unable to fetch MAC address from device tree.\n");
1031 return NULL;
1032 }
1033
1034 dev = alloc_etherdev(sizeof (struct veth_port));
1035 if (! dev) {
1036 veth_error("Unable to allocate net_device structure!\n");
1037 return NULL;
1038 }
1039
1040 port = netdev_priv(dev);
1041
1042 spin_lock_init(&port->queue_lock);
1043 rwlock_init(&port->mcast_gate);
1044 port->stopped_map = 0;
1045
1046 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1047 HvLpVirtualLanIndexMap map;
1048
1049 if (i == this_lp)
1050 continue;
1051 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1052 if (map & (0x8000 >> vlan))
1053 port->lpar_map |= (1 << i);
1054 }
1055 port->dev = vdev;
1056
1057 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1058
1059 dev->mtu = VETH_MAX_MTU;
1060
1061 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1062
1063 dev->netdev_ops = &veth_netdev_ops;
1064 SET_ETHTOOL_OPS(dev, &ops);
1065
1066 SET_NETDEV_DEV(dev, vdev);
1067
1068 rc = register_netdev(dev);
1069 if (rc != 0) {
1070 veth_error("Failed registering net device for vlan%d.\n", vlan);
1071 free_netdev(dev);
1072 return NULL;
1073 }
1074
1075 kobject_init(&port->kobject, &veth_port_ktype);
1076 if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
1077 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1078
1079 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1080 dev->name, vlan, port->lpar_map);
1081
1082 return dev;
1083 }
1084
1085 /*
1086 * Tx path
1087 */
1088
veth_transmit_to_one(struct sk_buff * skb,HvLpIndex rlp,struct net_device * dev)1089 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1090 struct net_device *dev)
1091 {
1092 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1093 struct veth_port *port = netdev_priv(dev);
1094 HvLpEvent_Rc rc;
1095 struct veth_msg *msg = NULL;
1096 unsigned long flags;
1097
1098 if (! cnx)
1099 return 0;
1100
1101 spin_lock_irqsave(&cnx->lock, flags);
1102
1103 if (! (cnx->state & VETH_STATE_READY))
1104 goto no_error;
1105
1106 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1107 goto drop;
1108
1109 msg = veth_stack_pop(cnx);
1110 if (! msg)
1111 goto drop;
1112
1113 msg->in_use = 1;
1114 msg->skb = skb_get(skb);
1115
1116 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1117 skb->len, DMA_TO_DEVICE);
1118
1119 if (dma_mapping_error(port->dev, msg->data.addr[0]))
1120 goto recycle_and_drop;
1121
1122 msg->dev = port->dev;
1123 msg->data.len[0] = skb->len;
1124 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1125
1126 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1127
1128 if (rc != HvLpEvent_Rc_Good)
1129 goto recycle_and_drop;
1130
1131 /* If the timer's not already running, start it now. */
1132 if (0 == cnx->outstanding_tx)
1133 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1134
1135 cnx->last_contact = jiffies;
1136 cnx->outstanding_tx++;
1137
1138 if (veth_stack_is_empty(cnx))
1139 veth_stop_queues(cnx);
1140
1141 no_error:
1142 spin_unlock_irqrestore(&cnx->lock, flags);
1143 return 0;
1144
1145 recycle_and_drop:
1146 veth_recycle_msg(cnx, msg);
1147 drop:
1148 spin_unlock_irqrestore(&cnx->lock, flags);
1149 return 1;
1150 }
1151
veth_transmit_to_many(struct sk_buff * skb,HvLpIndexMap lpmask,struct net_device * dev)1152 static void veth_transmit_to_many(struct sk_buff *skb,
1153 HvLpIndexMap lpmask,
1154 struct net_device *dev)
1155 {
1156 int i, success, error;
1157
1158 success = error = 0;
1159
1160 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1161 if ((lpmask & (1 << i)) == 0)
1162 continue;
1163
1164 if (veth_transmit_to_one(skb, i, dev))
1165 error = 1;
1166 else
1167 success = 1;
1168 }
1169
1170 if (error)
1171 dev->stats.tx_errors++;
1172
1173 if (success) {
1174 dev->stats.tx_packets++;
1175 dev->stats.tx_bytes += skb->len;
1176 }
1177 }
1178
veth_start_xmit(struct sk_buff * skb,struct net_device * dev)1179 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1180 {
1181 unsigned char *frame = skb->data;
1182 struct veth_port *port = netdev_priv(dev);
1183 HvLpIndexMap lpmask;
1184
1185 if (is_unicast_ether_addr(frame)) {
1186 /* unicast packet */
1187 HvLpIndex rlp = frame[5];
1188
1189 if ( ! ((1 << rlp) & port->lpar_map) ) {
1190 dev_kfree_skb(skb);
1191 return NETDEV_TX_OK;
1192 }
1193
1194 lpmask = 1 << rlp;
1195 } else {
1196 lpmask = port->lpar_map;
1197 }
1198
1199 veth_transmit_to_many(skb, lpmask, dev);
1200
1201 dev_kfree_skb(skb);
1202
1203 return NETDEV_TX_OK;
1204 }
1205
1206 /* You must hold the connection's lock when you call this function. */
veth_recycle_msg(struct veth_lpar_connection * cnx,struct veth_msg * msg)1207 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1208 struct veth_msg *msg)
1209 {
1210 u32 dma_address, dma_length;
1211
1212 if (msg->in_use) {
1213 msg->in_use = 0;
1214 dma_address = msg->data.addr[0];
1215 dma_length = msg->data.len[0];
1216
1217 if (!dma_mapping_error(msg->dev, dma_address))
1218 dma_unmap_single(msg->dev, dma_address, dma_length,
1219 DMA_TO_DEVICE);
1220
1221 if (msg->skb) {
1222 dev_kfree_skb_any(msg->skb);
1223 msg->skb = NULL;
1224 }
1225
1226 memset(&msg->data, 0, sizeof(msg->data));
1227 veth_stack_push(cnx, msg);
1228 } else if (cnx->state & VETH_STATE_OPEN) {
1229 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1230 cnx->remote_lp, msg->token);
1231 }
1232 }
1233
veth_wake_queues(struct veth_lpar_connection * cnx)1234 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1235 {
1236 int i;
1237
1238 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1239 struct net_device *dev = veth_dev[i];
1240 struct veth_port *port;
1241 unsigned long flags;
1242
1243 if (! dev)
1244 continue;
1245
1246 port = netdev_priv(dev);
1247
1248 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1249 continue;
1250
1251 spin_lock_irqsave(&port->queue_lock, flags);
1252
1253 port->stopped_map &= ~(1 << cnx->remote_lp);
1254
1255 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1256 veth_debug("cnx %d: woke queue for %s.\n",
1257 cnx->remote_lp, dev->name);
1258 netif_wake_queue(dev);
1259 }
1260 spin_unlock_irqrestore(&port->queue_lock, flags);
1261 }
1262 }
1263
veth_stop_queues(struct veth_lpar_connection * cnx)1264 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1265 {
1266 int i;
1267
1268 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1269 struct net_device *dev = veth_dev[i];
1270 struct veth_port *port;
1271
1272 if (! dev)
1273 continue;
1274
1275 port = netdev_priv(dev);
1276
1277 /* If this cnx is not on the vlan for this port, continue */
1278 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1279 continue;
1280
1281 spin_lock(&port->queue_lock);
1282
1283 netif_stop_queue(dev);
1284 port->stopped_map |= (1 << cnx->remote_lp);
1285
1286 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1287 cnx->remote_lp, dev->name, port->stopped_map);
1288
1289 spin_unlock(&port->queue_lock);
1290 }
1291 }
1292
veth_timed_reset(unsigned long ptr)1293 static void veth_timed_reset(unsigned long ptr)
1294 {
1295 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1296 unsigned long trigger_time, flags;
1297
1298 /* FIXME is it possible this fires after veth_stop_connection()?
1299 * That would reschedule the statemachine for 5 seconds and probably
1300 * execute it after the module's been unloaded. Hmm. */
1301
1302 spin_lock_irqsave(&cnx->lock, flags);
1303
1304 if (cnx->outstanding_tx > 0) {
1305 trigger_time = cnx->last_contact + cnx->reset_timeout;
1306
1307 if (trigger_time < jiffies) {
1308 cnx->state |= VETH_STATE_RESET;
1309 veth_kick_statemachine(cnx);
1310 veth_error("%d packets not acked by LPAR %d within %d "
1311 "seconds, resetting.\n",
1312 cnx->outstanding_tx, cnx->remote_lp,
1313 cnx->reset_timeout / HZ);
1314 } else {
1315 /* Reschedule the timer */
1316 trigger_time = jiffies + cnx->reset_timeout;
1317 mod_timer(&cnx->reset_timer, trigger_time);
1318 }
1319 }
1320
1321 spin_unlock_irqrestore(&cnx->lock, flags);
1322 }
1323
1324 /*
1325 * Rx path
1326 */
1327
veth_frame_wanted(struct veth_port * port,u64 mac_addr)1328 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1329 {
1330 int wanted = 0;
1331 int i;
1332 unsigned long flags;
1333
1334 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1335 return 1;
1336
1337 read_lock_irqsave(&port->mcast_gate, flags);
1338
1339 if (port->promiscuous) {
1340 wanted = 1;
1341 goto out;
1342 }
1343
1344 for (i = 0; i < port->num_mcast; ++i) {
1345 if (port->mcast_addr[i] == mac_addr) {
1346 wanted = 1;
1347 break;
1348 }
1349 }
1350
1351 out:
1352 read_unlock_irqrestore(&port->mcast_gate, flags);
1353
1354 return wanted;
1355 }
1356
1357 struct dma_chunk {
1358 u64 addr;
1359 u64 size;
1360 };
1361
1362 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1363
veth_build_dma_list(struct dma_chunk * list,unsigned char * p,unsigned long length)1364 static inline void veth_build_dma_list(struct dma_chunk *list,
1365 unsigned char *p, unsigned long length)
1366 {
1367 unsigned long done;
1368 int i = 1;
1369
1370 /* FIXME: skbs are contiguous in real addresses. Do we
1371 * really need to break it into PAGE_SIZE chunks, or can we do
1372 * it just at the granularity of iSeries real->absolute
1373 * mapping? Indeed, given the way the allocator works, can we
1374 * count on them being absolutely contiguous? */
1375 list[0].addr = iseries_hv_addr(p);
1376 list[0].size = min(length,
1377 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1378
1379 done = list[0].size;
1380 while (done < length) {
1381 list[i].addr = iseries_hv_addr(p + done);
1382 list[i].size = min(length-done, PAGE_SIZE);
1383 done += list[i].size;
1384 i++;
1385 }
1386 }
1387
veth_flush_acks(struct veth_lpar_connection * cnx)1388 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1389 {
1390 HvLpEvent_Rc rc;
1391
1392 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1393 0, &cnx->pending_acks);
1394
1395 if (rc != HvLpEvent_Rc_Good)
1396 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1397 cnx->remote_lp, (int)rc);
1398
1399 cnx->num_pending_acks = 0;
1400 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1401 }
1402
veth_receive(struct veth_lpar_connection * cnx,struct veth_lpevent * event)1403 static void veth_receive(struct veth_lpar_connection *cnx,
1404 struct veth_lpevent *event)
1405 {
1406 struct veth_frames_data *senddata = &event->u.frames_data;
1407 int startchunk = 0;
1408 int nchunks;
1409 unsigned long flags;
1410 HvLpDma_Rc rc;
1411
1412 do {
1413 u16 length = 0;
1414 struct sk_buff *skb;
1415 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1416 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1417 u64 dest;
1418 HvLpVirtualLanIndex vlan;
1419 struct net_device *dev;
1420 struct veth_port *port;
1421
1422 /* FIXME: do we need this? */
1423 memset(local_list, 0, sizeof(local_list));
1424 memset(remote_list, 0, sizeof(remote_list));
1425
1426 /* a 0 address marks the end of the valid entries */
1427 if (senddata->addr[startchunk] == 0)
1428 break;
1429
1430 /* make sure that we have at least 1 EOF entry in the
1431 * remaining entries */
1432 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1433 veth_error("Missing EOF fragment in event "
1434 "eofmask = 0x%x startchunk = %d\n",
1435 (unsigned)senddata->eofmask,
1436 startchunk);
1437 break;
1438 }
1439
1440 /* build list of chunks in this frame */
1441 nchunks = 0;
1442 do {
1443 remote_list[nchunks].addr =
1444 (u64) senddata->addr[startchunk+nchunks] << 32;
1445 remote_list[nchunks].size =
1446 senddata->len[startchunk+nchunks];
1447 length += remote_list[nchunks].size;
1448 } while (! (senddata->eofmask &
1449 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1450
1451 /* length == total length of all chunks */
1452 /* nchunks == # of chunks in this frame */
1453
1454 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1455 veth_error("Received oversize frame from LPAR %d "
1456 "(length = %d)\n",
1457 cnx->remote_lp, length);
1458 continue;
1459 }
1460
1461 skb = alloc_skb(length, GFP_ATOMIC);
1462 if (!skb)
1463 continue;
1464
1465 veth_build_dma_list(local_list, skb->data, length);
1466
1467 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1468 event->base_event.xSourceLp,
1469 HvLpDma_Direction_RemoteToLocal,
1470 cnx->src_inst,
1471 cnx->dst_inst,
1472 HvLpDma_AddressType_RealAddress,
1473 HvLpDma_AddressType_TceIndex,
1474 iseries_hv_addr(&local_list),
1475 iseries_hv_addr(&remote_list),
1476 length);
1477 if (rc != HvLpDma_Rc_Good) {
1478 dev_kfree_skb_irq(skb);
1479 continue;
1480 }
1481
1482 vlan = skb->data[9];
1483 dev = veth_dev[vlan];
1484 if (! dev) {
1485 /*
1486 * Some earlier versions of the driver sent
1487 * broadcasts down all connections, even to lpars
1488 * that weren't on the relevant vlan. So ignore
1489 * packets belonging to a vlan we're not on.
1490 * We can also be here if we receive packets while
1491 * the driver is going down, because then dev is NULL.
1492 */
1493 dev_kfree_skb_irq(skb);
1494 continue;
1495 }
1496
1497 port = netdev_priv(dev);
1498 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1499
1500 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1501 dev_kfree_skb_irq(skb);
1502 continue;
1503 }
1504 if (! veth_frame_wanted(port, dest)) {
1505 dev_kfree_skb_irq(skb);
1506 continue;
1507 }
1508
1509 skb_put(skb, length);
1510 skb->protocol = eth_type_trans(skb, dev);
1511 skb_checksum_none_assert(skb);
1512 netif_rx(skb); /* send it up */
1513 dev->stats.rx_packets++;
1514 dev->stats.rx_bytes += length;
1515 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1516
1517 /* Ack it */
1518 spin_lock_irqsave(&cnx->lock, flags);
1519 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1520
1521 cnx->pending_acks[cnx->num_pending_acks++] =
1522 event->base_event.xCorrelationToken;
1523
1524 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1525 (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1526 veth_flush_acks(cnx);
1527
1528 spin_unlock_irqrestore(&cnx->lock, flags);
1529 }
1530
veth_timed_ack(unsigned long ptr)1531 static void veth_timed_ack(unsigned long ptr)
1532 {
1533 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1534 unsigned long flags;
1535
1536 /* Ack all the events */
1537 spin_lock_irqsave(&cnx->lock, flags);
1538 if (cnx->num_pending_acks > 0)
1539 veth_flush_acks(cnx);
1540
1541 /* Reschedule the timer */
1542 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1543 add_timer(&cnx->ack_timer);
1544 spin_unlock_irqrestore(&cnx->lock, flags);
1545 }
1546
veth_remove(struct vio_dev * vdev)1547 static int veth_remove(struct vio_dev *vdev)
1548 {
1549 struct veth_lpar_connection *cnx;
1550 struct net_device *dev;
1551 struct veth_port *port;
1552 int i;
1553
1554 dev = veth_dev[vdev->unit_address];
1555
1556 if (! dev)
1557 return 0;
1558
1559 port = netdev_priv(dev);
1560
1561 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1562 cnx = veth_cnx[i];
1563
1564 if (cnx && (port->lpar_map & (1 << i))) {
1565 /* Drop our reference to connections on our VLAN */
1566 kobject_put(&cnx->kobject);
1567 }
1568 }
1569
1570 veth_dev[vdev->unit_address] = NULL;
1571 kobject_del(&port->kobject);
1572 kobject_put(&port->kobject);
1573 unregister_netdev(dev);
1574 free_netdev(dev);
1575
1576 return 0;
1577 }
1578
veth_probe(struct vio_dev * vdev,const struct vio_device_id * id)1579 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1580 {
1581 int i = vdev->unit_address;
1582 struct net_device *dev;
1583 struct veth_port *port;
1584
1585 dev = veth_probe_one(i, vdev);
1586 if (dev == NULL) {
1587 veth_remove(vdev);
1588 return 1;
1589 }
1590 veth_dev[i] = dev;
1591
1592 port = netdev_priv(dev);
1593
1594 /* Start the state machine on each connection on this vlan. If we're
1595 * the first dev to do so this will commence link negotiation */
1596 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1597 struct veth_lpar_connection *cnx;
1598
1599 if (! (port->lpar_map & (1 << i)))
1600 continue;
1601
1602 cnx = veth_cnx[i];
1603 if (!cnx)
1604 continue;
1605
1606 kobject_get(&cnx->kobject);
1607 veth_kick_statemachine(cnx);
1608 }
1609
1610 return 0;
1611 }
1612
1613 /**
1614 * veth_device_table: Used by vio.c to match devices that we
1615 * support.
1616 */
1617 static struct vio_device_id veth_device_table[] __devinitdata = {
1618 { "network", "IBM,iSeries-l-lan" },
1619 { "", "" }
1620 };
1621 MODULE_DEVICE_TABLE(vio, veth_device_table);
1622
1623 static struct vio_driver veth_driver = {
1624 .id_table = veth_device_table,
1625 .probe = veth_probe,
1626 .remove = veth_remove,
1627 .driver = {
1628 .name = DRV_NAME,
1629 .owner = THIS_MODULE,
1630 }
1631 };
1632
1633 /*
1634 * Module initialization/cleanup
1635 */
1636
veth_module_cleanup(void)1637 static void __exit veth_module_cleanup(void)
1638 {
1639 int i;
1640 struct veth_lpar_connection *cnx;
1641
1642 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1643 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1644
1645 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1646 cnx = veth_cnx[i];
1647
1648 if (!cnx)
1649 continue;
1650
1651 /* Cancel work queued from Hypervisor callbacks */
1652 cancel_delayed_work_sync(&cnx->statemachine_wq);
1653 /* Remove the connection from sysfs */
1654 kobject_del(&cnx->kobject);
1655 /* Drop the driver's reference to the connection */
1656 kobject_put(&cnx->kobject);
1657 }
1658
1659 /* Unregister the driver, which will close all the netdevs and stop
1660 * the connections when they're no longer referenced. */
1661 vio_unregister_driver(&veth_driver);
1662 }
1663 module_exit(veth_module_cleanup);
1664
veth_module_init(void)1665 static int __init veth_module_init(void)
1666 {
1667 int i;
1668 int rc;
1669
1670 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1671 return -ENODEV;
1672
1673 this_lp = HvLpConfig_getLpIndex_outline();
1674
1675 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1676 rc = veth_init_connection(i);
1677 if (rc != 0)
1678 goto error;
1679 }
1680
1681 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1682 &veth_handle_event);
1683
1684 rc = vio_register_driver(&veth_driver);
1685 if (rc != 0)
1686 goto error;
1687
1688 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1689 struct kobject *kobj;
1690
1691 if (!veth_cnx[i])
1692 continue;
1693
1694 kobj = &veth_cnx[i]->kobject;
1695 /* If the add failes, complain but otherwise continue */
1696 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1697 "cnx%.2d", veth_cnx[i]->remote_lp))
1698 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1699 }
1700
1701 return 0;
1702
1703 error:
1704 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1705 veth_destroy_connection(veth_cnx[i]);
1706 }
1707
1708 return rc;
1709 }
1710 module_init(veth_module_init);
1711