Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License as | |
10 | * published by the Free Software Foundation; either version 2 of the | |
11 | * License, or (at your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but | |
14 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 | |
21 | * USA | |
22 | * | |
23 | * | |
24 | * This module implements the virtual ethernet device for iSeries LPAR | |
25 | * Linux. It uses hypervisor message passing to implement an | |
26 | * ethernet-like network device communicating between partitions on | |
27 | * the iSeries. | |
28 | * | |
29 | * The iSeries LPAR hypervisor currently allows for up to 16 different | |
30 | * virtual ethernets. These are all dynamically configurable on | |
31 | * OS/400 partitions, but dynamic configuration is not supported under | |
32 | * Linux yet. An ethXX network device will be created for each | |
33 | * virtual ethernet this partition is connected to. | |
34 | * | |
35 | * - This driver is responsible for routing packets to and from other | |
36 | * partitions. The MAC addresses used by the virtual ethernets | |
37 | * contains meaning and must not be modified. | |
38 | * | |
39 | * - Having 2 virtual ethernets to the same remote partition DOES NOT | |
40 | * double the available bandwidth. The 2 devices will share the | |
41 | * available hypervisor bandwidth. | |
42 | * | |
43 | * - If you send a packet to your own mac address, it will just be | |
44 | * dropped, you won't get it on the receive side. | |
45 | * | |
46 | * - Multicast is implemented by sending the frame frame to every | |
47 | * other partition. It is the responsibility of the receiving | |
48 | * partition to filter the addresses desired. | |
49 | * | |
50 | * Tunable parameters: | |
51 | * | |
52 | * VETH_NUMBUFFERS: This compile time option defaults to 120. It | |
53 | * controls how much memory Linux will allocate per remote partition | |
54 | * it is communicating with. It can be thought of as the maximum | |
55 | * number of packets outstanding to a remote partition at a time. | |
56 | */ | |
57 | ||
58 | #include <linux/config.h> | |
59 | #include <linux/module.h> | |
60 | #include <linux/version.h> | |
61 | #include <linux/types.h> | |
62 | #include <linux/errno.h> | |
63 | #include <linux/ioport.h> | |
64 | #include <linux/kernel.h> | |
65 | #include <linux/netdevice.h> | |
66 | #include <linux/etherdevice.h> | |
67 | #include <linux/skbuff.h> | |
68 | #include <linux/init.h> | |
69 | #include <linux/delay.h> | |
70 | #include <linux/mm.h> | |
71 | #include <linux/ethtool.h> | |
72 | #include <asm/iSeries/mf.h> | |
73 | #include <asm/iSeries/iSeries_pci.h> | |
74 | #include <asm/uaccess.h> | |
75 | ||
76 | #include <asm/iSeries/HvLpConfig.h> | |
77 | #include <asm/iSeries/HvTypes.h> | |
78 | #include <asm/iSeries/HvLpEvent.h> | |
79 | #include <asm/iommu.h> | |
80 | #include <asm/vio.h> | |
81 | ||
61a3c696 ME |
82 | #undef DEBUG |
83 | ||
1da177e4 LT |
84 | #include "iseries_veth.h" |
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_NUMBUFFERS (120) | |
91 | #define VETH_ACKTIMEOUT (1000000) /* microseconds */ | |
92 | #define VETH_MAX_MCAST (12) | |
93 | ||
94 | #define VETH_MAX_MTU (9000) | |
95 | ||
96 | #if VETH_NUMBUFFERS < 10 | |
97 | #define ACK_THRESHOLD (1) | |
98 | #elif VETH_NUMBUFFERS < 20 | |
99 | #define ACK_THRESHOLD (4) | |
100 | #elif VETH_NUMBUFFERS < 40 | |
101 | #define ACK_THRESHOLD (10) | |
102 | #else | |
103 | #define ACK_THRESHOLD (20) | |
104 | #endif | |
105 | ||
106 | #define VETH_STATE_SHUTDOWN (0x0001) | |
107 | #define VETH_STATE_OPEN (0x0002) | |
108 | #define VETH_STATE_RESET (0x0004) | |
109 | #define VETH_STATE_SENTMON (0x0008) | |
110 | #define VETH_STATE_SENTCAPS (0x0010) | |
111 | #define VETH_STATE_GOTCAPACK (0x0020) | |
112 | #define VETH_STATE_GOTCAPS (0x0040) | |
113 | #define VETH_STATE_SENTCAPACK (0x0080) | |
114 | #define VETH_STATE_READY (0x0100) | |
115 | ||
116 | struct veth_msg { | |
117 | struct veth_msg *next; | |
118 | struct VethFramesData data; | |
119 | int token; | |
b08bd5c0 | 120 | int in_use; |
1da177e4 LT |
121 | struct sk_buff *skb; |
122 | struct device *dev; | |
123 | }; | |
124 | ||
125 | struct veth_lpar_connection { | |
126 | HvLpIndex remote_lp; | |
127 | struct work_struct statemachine_wq; | |
128 | struct veth_msg *msgs; | |
129 | int num_events; | |
130 | struct VethCapData local_caps; | |
131 | ||
f0c129ca | 132 | struct kobject kobject; |
1da177e4 LT |
133 | struct timer_list ack_timer; |
134 | ||
24562ffa ME |
135 | struct timer_list reset_timer; |
136 | unsigned int reset_timeout; | |
137 | unsigned long last_contact; | |
138 | int outstanding_tx; | |
139 | ||
1da177e4 LT |
140 | spinlock_t lock; |
141 | unsigned long state; | |
142 | HvLpInstanceId src_inst; | |
143 | HvLpInstanceId dst_inst; | |
144 | struct VethLpEvent cap_event, cap_ack_event; | |
145 | u16 pending_acks[VETH_MAX_ACKS_PER_MSG]; | |
146 | u32 num_pending_acks; | |
147 | ||
148 | int num_ack_events; | |
149 | struct VethCapData remote_caps; | |
150 | u32 ack_timeout; | |
151 | ||
1da177e4 LT |
152 | struct veth_msg *msg_stack_head; |
153 | }; | |
154 | ||
155 | struct veth_port { | |
156 | struct device *dev; | |
157 | struct net_device_stats stats; | |
158 | u64 mac_addr; | |
159 | HvLpIndexMap lpar_map; | |
160 | ||
e0808494 ME |
161 | /* queue_lock protects the stopped_map and dev's queue. */ |
162 | spinlock_t queue_lock; | |
163 | HvLpIndexMap stopped_map; | |
1da177e4 | 164 | |
e0808494 | 165 | /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */ |
1da177e4 LT |
166 | rwlock_t mcast_gate; |
167 | int promiscuous; | |
1da177e4 LT |
168 | int num_mcast; |
169 | u64 mcast_addr[VETH_MAX_MCAST]; | |
170 | }; | |
171 | ||
172 | static HvLpIndex this_lp; | |
173 | static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */ | |
174 | static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */ | |
175 | ||
176 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev); | |
177 | static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *); | |
e0808494 ME |
178 | static void veth_wake_queues(struct veth_lpar_connection *cnx); |
179 | static void veth_stop_queues(struct veth_lpar_connection *cnx); | |
1da177e4 | 180 | static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *); |
f0c129ca | 181 | static void veth_release_connection(struct kobject *kobject); |
24562ffa ME |
182 | static void veth_timed_ack(unsigned long ptr); |
183 | static void veth_timed_reset(unsigned long ptr); | |
f0c129ca ME |
184 | |
185 | static struct kobj_type veth_lpar_connection_ktype = { | |
186 | .release = veth_release_connection | |
187 | }; | |
1da177e4 LT |
188 | |
189 | /* | |
190 | * Utility functions | |
191 | */ | |
192 | ||
61a3c696 ME |
193 | #define veth_info(fmt, args...) \ |
194 | printk(KERN_INFO "iseries_veth: " fmt, ## args) | |
1da177e4 LT |
195 | |
196 | #define veth_error(fmt, args...) \ | |
61a3c696 ME |
197 | printk(KERN_ERR "iseries_veth: Error: " fmt, ## args) |
198 | ||
199 | #ifdef DEBUG | |
200 | #define veth_debug(fmt, args...) \ | |
201 | printk(KERN_DEBUG "iseries_veth: " fmt, ## args) | |
202 | #else | |
203 | #define veth_debug(fmt, args...) do {} while (0) | |
204 | #endif | |
1da177e4 | 205 | |
d7893ddd | 206 | /* You must hold the connection's lock when you call this function. */ |
1da177e4 LT |
207 | static inline void veth_stack_push(struct veth_lpar_connection *cnx, |
208 | struct veth_msg *msg) | |
209 | { | |
1da177e4 LT |
210 | msg->next = cnx->msg_stack_head; |
211 | cnx->msg_stack_head = msg; | |
1da177e4 LT |
212 | } |
213 | ||
d7893ddd | 214 | /* You must hold the connection's lock when you call this function. */ |
1da177e4 LT |
215 | static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx) |
216 | { | |
1da177e4 LT |
217 | struct veth_msg *msg; |
218 | ||
1da177e4 LT |
219 | msg = cnx->msg_stack_head; |
220 | if (msg) | |
221 | cnx->msg_stack_head = cnx->msg_stack_head->next; | |
d7893ddd | 222 | |
1da177e4 LT |
223 | return msg; |
224 | } | |
225 | ||
e0808494 ME |
226 | /* You must hold the connection's lock when you call this function. */ |
227 | static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx) | |
228 | { | |
229 | return cnx->msg_stack_head == NULL; | |
230 | } | |
231 | ||
1da177e4 LT |
232 | static inline HvLpEvent_Rc |
233 | veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype, | |
234 | HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype, | |
235 | u64 token, | |
236 | u64 data1, u64 data2, u64 data3, u64 data4, u64 data5) | |
237 | { | |
238 | return HvCallEvent_signalLpEventFast(cnx->remote_lp, | |
239 | HvLpEvent_Type_VirtualLan, | |
240 | subtype, ackind, acktype, | |
241 | cnx->src_inst, | |
242 | cnx->dst_inst, | |
243 | token, data1, data2, data3, | |
244 | data4, data5); | |
245 | } | |
246 | ||
247 | static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx, | |
248 | u16 subtype, u64 token, void *data) | |
249 | { | |
250 | u64 *p = (u64 *) data; | |
251 | ||
252 | return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck, | |
253 | HvLpEvent_AckType_ImmediateAck, | |
254 | token, p[0], p[1], p[2], p[3], p[4]); | |
255 | } | |
256 | ||
257 | struct veth_allocation { | |
258 | struct completion c; | |
259 | int num; | |
260 | }; | |
261 | ||
262 | static void veth_complete_allocation(void *parm, int number) | |
263 | { | |
264 | struct veth_allocation *vc = (struct veth_allocation *)parm; | |
265 | ||
266 | vc->num = number; | |
267 | complete(&vc->c); | |
268 | } | |
269 | ||
270 | static int veth_allocate_events(HvLpIndex rlp, int number) | |
271 | { | |
272 | struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 }; | |
273 | ||
274 | mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan, | |
275 | sizeof(struct VethLpEvent), number, | |
276 | &veth_complete_allocation, &vc); | |
277 | wait_for_completion(&vc.c); | |
278 | ||
279 | return vc.num; | |
280 | } | |
281 | ||
282 | /* | |
283 | * LPAR connection code | |
284 | */ | |
285 | ||
286 | static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx) | |
287 | { | |
288 | schedule_work(&cnx->statemachine_wq); | |
289 | } | |
290 | ||
291 | static void veth_take_cap(struct veth_lpar_connection *cnx, | |
292 | struct VethLpEvent *event) | |
293 | { | |
294 | unsigned long flags; | |
295 | ||
296 | spin_lock_irqsave(&cnx->lock, flags); | |
297 | /* Receiving caps may mean the other end has just come up, so | |
298 | * we need to reload the instance ID of the far end */ | |
299 | cnx->dst_inst = | |
300 | HvCallEvent_getTargetLpInstanceId(cnx->remote_lp, | |
301 | HvLpEvent_Type_VirtualLan); | |
302 | ||
303 | if (cnx->state & VETH_STATE_GOTCAPS) { | |
61a3c696 | 304 | veth_error("Received a second capabilities from LPAR %d.\n", |
1da177e4 LT |
305 | cnx->remote_lp); |
306 | event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable; | |
307 | HvCallEvent_ackLpEvent((struct HvLpEvent *) event); | |
308 | } else { | |
309 | memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event)); | |
310 | cnx->state |= VETH_STATE_GOTCAPS; | |
311 | veth_kick_statemachine(cnx); | |
312 | } | |
313 | spin_unlock_irqrestore(&cnx->lock, flags); | |
314 | } | |
315 | ||
316 | static void veth_take_cap_ack(struct veth_lpar_connection *cnx, | |
317 | struct VethLpEvent *event) | |
318 | { | |
319 | unsigned long flags; | |
320 | ||
321 | spin_lock_irqsave(&cnx->lock, flags); | |
322 | if (cnx->state & VETH_STATE_GOTCAPACK) { | |
61a3c696 | 323 | veth_error("Received a second capabilities ack from LPAR %d.\n", |
1da177e4 LT |
324 | cnx->remote_lp); |
325 | } else { | |
326 | memcpy(&cnx->cap_ack_event, event, | |
327 | sizeof(&cnx->cap_ack_event)); | |
328 | cnx->state |= VETH_STATE_GOTCAPACK; | |
329 | veth_kick_statemachine(cnx); | |
330 | } | |
331 | spin_unlock_irqrestore(&cnx->lock, flags); | |
332 | } | |
333 | ||
334 | static void veth_take_monitor_ack(struct veth_lpar_connection *cnx, | |
335 | struct VethLpEvent *event) | |
336 | { | |
337 | unsigned long flags; | |
338 | ||
339 | spin_lock_irqsave(&cnx->lock, flags); | |
61a3c696 | 340 | veth_debug("cnx %d: lost connection.\n", cnx->remote_lp); |
58c5900b ME |
341 | |
342 | /* Avoid kicking the statemachine once we're shutdown. | |
343 | * It's unnecessary and it could break veth_stop_connection(). */ | |
344 | ||
345 | if (! (cnx->state & VETH_STATE_SHUTDOWN)) { | |
346 | cnx->state |= VETH_STATE_RESET; | |
347 | veth_kick_statemachine(cnx); | |
348 | } | |
1da177e4 LT |
349 | spin_unlock_irqrestore(&cnx->lock, flags); |
350 | } | |
351 | ||
352 | static void veth_handle_ack(struct VethLpEvent *event) | |
353 | { | |
354 | HvLpIndex rlp = event->base_event.xTargetLp; | |
355 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
356 | ||
357 | BUG_ON(! cnx); | |
358 | ||
359 | switch (event->base_event.xSubtype) { | |
360 | case VethEventTypeCap: | |
361 | veth_take_cap_ack(cnx, event); | |
362 | break; | |
363 | case VethEventTypeMonitor: | |
364 | veth_take_monitor_ack(cnx, event); | |
365 | break; | |
366 | default: | |
61a3c696 ME |
367 | veth_error("Unknown ack type %d from LPAR %d.\n", |
368 | event->base_event.xSubtype, rlp); | |
1da177e4 LT |
369 | }; |
370 | } | |
371 | ||
372 | static void veth_handle_int(struct VethLpEvent *event) | |
373 | { | |
374 | HvLpIndex rlp = event->base_event.xSourceLp; | |
375 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
376 | unsigned long flags; | |
24562ffa | 377 | int i, acked = 0; |
1da177e4 LT |
378 | |
379 | BUG_ON(! cnx); | |
380 | ||
381 | switch (event->base_event.xSubtype) { | |
382 | case VethEventTypeCap: | |
383 | veth_take_cap(cnx, event); | |
384 | break; | |
385 | case VethEventTypeMonitor: | |
386 | /* do nothing... this'll hang out here til we're dead, | |
387 | * and the hypervisor will return it for us. */ | |
388 | break; | |
389 | case VethEventTypeFramesAck: | |
390 | spin_lock_irqsave(&cnx->lock, flags); | |
24562ffa | 391 | |
1da177e4 LT |
392 | for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) { |
393 | u16 msgnum = event->u.frames_ack_data.token[i]; | |
394 | ||
24562ffa | 395 | if (msgnum < VETH_NUMBUFFERS) { |
1da177e4 | 396 | veth_recycle_msg(cnx, cnx->msgs + msgnum); |
24562ffa ME |
397 | cnx->outstanding_tx--; |
398 | acked++; | |
399 | } | |
1da177e4 | 400 | } |
24562ffa | 401 | |
e0808494 | 402 | if (acked > 0) { |
24562ffa | 403 | cnx->last_contact = jiffies; |
e0808494 ME |
404 | veth_wake_queues(cnx); |
405 | } | |
24562ffa | 406 | |
1da177e4 | 407 | spin_unlock_irqrestore(&cnx->lock, flags); |
1da177e4 LT |
408 | break; |
409 | case VethEventTypeFrames: | |
410 | veth_receive(cnx, event); | |
411 | break; | |
412 | default: | |
61a3c696 ME |
413 | veth_error("Unknown interrupt type %d from LPAR %d.\n", |
414 | event->base_event.xSubtype, rlp); | |
1da177e4 LT |
415 | }; |
416 | } | |
417 | ||
418 | static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs) | |
419 | { | |
420 | struct VethLpEvent *veth_event = (struct VethLpEvent *)event; | |
421 | ||
422 | if (event->xFlags.xFunction == HvLpEvent_Function_Ack) | |
423 | veth_handle_ack(veth_event); | |
424 | else if (event->xFlags.xFunction == HvLpEvent_Function_Int) | |
425 | veth_handle_int(veth_event); | |
426 | } | |
427 | ||
428 | static int veth_process_caps(struct veth_lpar_connection *cnx) | |
429 | { | |
430 | struct VethCapData *remote_caps = &cnx->remote_caps; | |
431 | int num_acks_needed; | |
432 | ||
433 | /* Convert timer to jiffies */ | |
434 | cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000; | |
435 | ||
436 | if ( (remote_caps->num_buffers == 0) | |
437 | || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) | |
438 | || (remote_caps->ack_threshold == 0) | |
439 | || (cnx->ack_timeout == 0) ) { | |
61a3c696 ME |
440 | veth_error("Received incompatible capabilities from LPAR %d.\n", |
441 | cnx->remote_lp); | |
1da177e4 LT |
442 | return HvLpEvent_Rc_InvalidSubtypeData; |
443 | } | |
444 | ||
445 | num_acks_needed = (remote_caps->num_buffers | |
446 | / remote_caps->ack_threshold) + 1; | |
447 | ||
448 | /* FIXME: locking on num_ack_events? */ | |
449 | if (cnx->num_ack_events < num_acks_needed) { | |
450 | int num; | |
451 | ||
452 | num = veth_allocate_events(cnx->remote_lp, | |
453 | num_acks_needed-cnx->num_ack_events); | |
454 | if (num > 0) | |
455 | cnx->num_ack_events += num; | |
456 | ||
457 | if (cnx->num_ack_events < num_acks_needed) { | |
61a3c696 ME |
458 | veth_error("Couldn't allocate enough ack events " |
459 | "for LPAR %d.\n", cnx->remote_lp); | |
1da177e4 LT |
460 | |
461 | return HvLpEvent_Rc_BufferNotAvailable; | |
462 | } | |
463 | } | |
464 | ||
465 | ||
466 | return HvLpEvent_Rc_Good; | |
467 | } | |
468 | ||
469 | /* FIXME: The gotos here are a bit dubious */ | |
470 | static void veth_statemachine(void *p) | |
471 | { | |
472 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p; | |
473 | int rlp = cnx->remote_lp; | |
474 | int rc; | |
475 | ||
476 | spin_lock_irq(&cnx->lock); | |
477 | ||
478 | restart: | |
479 | if (cnx->state & VETH_STATE_RESET) { | |
1da177e4 LT |
480 | if (cnx->state & VETH_STATE_OPEN) |
481 | HvCallEvent_closeLpEventPath(cnx->remote_lp, | |
482 | HvLpEvent_Type_VirtualLan); | |
483 | ||
abfda471 ME |
484 | /* |
485 | * Reset ack data. This prevents the ack_timer actually | |
486 | * doing anything, even if it runs one more time when | |
487 | * we drop the lock below. | |
488 | */ | |
1da177e4 LT |
489 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); |
490 | cnx->num_pending_acks = 0; | |
491 | ||
492 | cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON | |
493 | | VETH_STATE_OPEN | VETH_STATE_SENTCAPS | |
494 | | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS | |
495 | | VETH_STATE_SENTCAPACK | VETH_STATE_READY); | |
496 | ||
497 | /* Clean up any leftover messages */ | |
24562ffa ME |
498 | if (cnx->msgs) { |
499 | int i; | |
1da177e4 LT |
500 | for (i = 0; i < VETH_NUMBUFFERS; ++i) |
501 | veth_recycle_msg(cnx, cnx->msgs + i); | |
24562ffa | 502 | } |
e0808494 | 503 | |
24562ffa | 504 | cnx->outstanding_tx = 0; |
e0808494 | 505 | veth_wake_queues(cnx); |
abfda471 ME |
506 | |
507 | /* Drop the lock so we can do stuff that might sleep or | |
508 | * take other locks. */ | |
1da177e4 | 509 | spin_unlock_irq(&cnx->lock); |
abfda471 ME |
510 | |
511 | del_timer_sync(&cnx->ack_timer); | |
24562ffa ME |
512 | del_timer_sync(&cnx->reset_timer); |
513 | ||
1da177e4 | 514 | spin_lock_irq(&cnx->lock); |
abfda471 | 515 | |
1da177e4 LT |
516 | if (cnx->state & VETH_STATE_RESET) |
517 | goto restart; | |
58c5900b ME |
518 | |
519 | /* Hack, wait for the other end to reset itself. */ | |
520 | if (! (cnx->state & VETH_STATE_SHUTDOWN)) { | |
521 | schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ); | |
522 | goto out; | |
523 | } | |
1da177e4 LT |
524 | } |
525 | ||
526 | if (cnx->state & VETH_STATE_SHUTDOWN) | |
527 | /* It's all over, do nothing */ | |
528 | goto out; | |
529 | ||
530 | if ( !(cnx->state & VETH_STATE_OPEN) ) { | |
531 | if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) ) | |
532 | goto cant_cope; | |
533 | ||
534 | HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan); | |
535 | cnx->src_inst = | |
536 | HvCallEvent_getSourceLpInstanceId(rlp, | |
537 | HvLpEvent_Type_VirtualLan); | |
538 | cnx->dst_inst = | |
539 | HvCallEvent_getTargetLpInstanceId(rlp, | |
540 | HvLpEvent_Type_VirtualLan); | |
541 | cnx->state |= VETH_STATE_OPEN; | |
542 | } | |
543 | ||
544 | if ( (cnx->state & VETH_STATE_OPEN) | |
545 | && !(cnx->state & VETH_STATE_SENTMON) ) { | |
546 | rc = veth_signalevent(cnx, VethEventTypeMonitor, | |
547 | HvLpEvent_AckInd_DoAck, | |
548 | HvLpEvent_AckType_DeferredAck, | |
549 | 0, 0, 0, 0, 0, 0); | |
550 | ||
551 | if (rc == HvLpEvent_Rc_Good) { | |
552 | cnx->state |= VETH_STATE_SENTMON; | |
553 | } else { | |
554 | if ( (rc != HvLpEvent_Rc_PartitionDead) | |
555 | && (rc != HvLpEvent_Rc_PathClosed) ) | |
61a3c696 ME |
556 | veth_error("Error sending monitor to LPAR %d, " |
557 | "rc = %d\n", rlp, rc); | |
1da177e4 LT |
558 | |
559 | /* Oh well, hope we get a cap from the other | |
560 | * end and do better when that kicks us */ | |
561 | goto out; | |
562 | } | |
563 | } | |
564 | ||
565 | if ( (cnx->state & VETH_STATE_OPEN) | |
566 | && !(cnx->state & VETH_STATE_SENTCAPS)) { | |
567 | u64 *rawcap = (u64 *)&cnx->local_caps; | |
568 | ||
569 | rc = veth_signalevent(cnx, VethEventTypeCap, | |
570 | HvLpEvent_AckInd_DoAck, | |
571 | HvLpEvent_AckType_ImmediateAck, | |
572 | 0, rawcap[0], rawcap[1], rawcap[2], | |
573 | rawcap[3], rawcap[4]); | |
574 | ||
575 | if (rc == HvLpEvent_Rc_Good) { | |
576 | cnx->state |= VETH_STATE_SENTCAPS; | |
577 | } else { | |
578 | if ( (rc != HvLpEvent_Rc_PartitionDead) | |
579 | && (rc != HvLpEvent_Rc_PathClosed) ) | |
61a3c696 ME |
580 | veth_error("Error sending caps to LPAR %d, " |
581 | "rc = %d\n", rlp, rc); | |
582 | ||
1da177e4 LT |
583 | /* Oh well, hope we get a cap from the other |
584 | * end and do better when that kicks us */ | |
585 | goto out; | |
586 | } | |
587 | } | |
588 | ||
589 | if ((cnx->state & VETH_STATE_GOTCAPS) | |
590 | && !(cnx->state & VETH_STATE_SENTCAPACK)) { | |
591 | struct VethCapData *remote_caps = &cnx->remote_caps; | |
592 | ||
593 | memcpy(remote_caps, &cnx->cap_event.u.caps_data, | |
594 | sizeof(*remote_caps)); | |
595 | ||
596 | spin_unlock_irq(&cnx->lock); | |
597 | rc = veth_process_caps(cnx); | |
598 | spin_lock_irq(&cnx->lock); | |
599 | ||
600 | /* We dropped the lock, so recheck for anything which | |
601 | * might mess us up */ | |
602 | if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN)) | |
603 | goto restart; | |
604 | ||
605 | cnx->cap_event.base_event.xRc = rc; | |
606 | HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event); | |
607 | if (rc == HvLpEvent_Rc_Good) | |
608 | cnx->state |= VETH_STATE_SENTCAPACK; | |
609 | else | |
610 | goto cant_cope; | |
611 | } | |
612 | ||
613 | if ((cnx->state & VETH_STATE_GOTCAPACK) | |
614 | && (cnx->state & VETH_STATE_GOTCAPS) | |
615 | && !(cnx->state & VETH_STATE_READY)) { | |
616 | if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) { | |
617 | /* Start the ACK timer */ | |
618 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; | |
619 | add_timer(&cnx->ack_timer); | |
620 | cnx->state |= VETH_STATE_READY; | |
621 | } else { | |
61a3c696 ME |
622 | veth_error("Caps rejected by LPAR %d, rc = %d\n", |
623 | rlp, cnx->cap_ack_event.base_event.xRc); | |
1da177e4 LT |
624 | goto cant_cope; |
625 | } | |
626 | } | |
627 | ||
628 | out: | |
629 | spin_unlock_irq(&cnx->lock); | |
630 | return; | |
631 | ||
632 | cant_cope: | |
633 | /* FIXME: we get here if something happens we really can't | |
634 | * cope with. The link will never work once we get here, and | |
635 | * all we can do is not lock the rest of the system up */ | |
61a3c696 ME |
636 | veth_error("Unrecoverable error on connection to LPAR %d, shutting down" |
637 | " (state = 0x%04lx)\n", rlp, cnx->state); | |
1da177e4 LT |
638 | cnx->state |= VETH_STATE_SHUTDOWN; |
639 | spin_unlock_irq(&cnx->lock); | |
640 | } | |
641 | ||
642 | static int veth_init_connection(u8 rlp) | |
643 | { | |
644 | struct veth_lpar_connection *cnx; | |
645 | struct veth_msg *msgs; | |
f0c129ca | 646 | int i, rc; |
1da177e4 LT |
647 | |
648 | if ( (rlp == this_lp) | |
649 | || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) ) | |
650 | return 0; | |
651 | ||
652 | cnx = kmalloc(sizeof(*cnx), GFP_KERNEL); | |
653 | if (! cnx) | |
654 | return -ENOMEM; | |
655 | memset(cnx, 0, sizeof(*cnx)); | |
656 | ||
657 | cnx->remote_lp = rlp; | |
658 | spin_lock_init(&cnx->lock); | |
659 | INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx); | |
24562ffa | 660 | |
1da177e4 LT |
661 | init_timer(&cnx->ack_timer); |
662 | cnx->ack_timer.function = veth_timed_ack; | |
663 | cnx->ack_timer.data = (unsigned long) cnx; | |
24562ffa ME |
664 | |
665 | init_timer(&cnx->reset_timer); | |
666 | cnx->reset_timer.function = veth_timed_reset; | |
667 | cnx->reset_timer.data = (unsigned long) cnx; | |
668 | cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000); | |
669 | ||
1da177e4 LT |
670 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); |
671 | ||
672 | veth_cnx[rlp] = cnx; | |
673 | ||
f0c129ca ME |
674 | /* This gets us 1 reference, which is held on behalf of the driver |
675 | * infrastructure. It's released at module unload. */ | |
676 | kobject_init(&cnx->kobject); | |
677 | cnx->kobject.ktype = &veth_lpar_connection_ktype; | |
678 | rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp); | |
679 | if (rc != 0) | |
680 | return rc; | |
681 | ||
1da177e4 LT |
682 | msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL); |
683 | if (! msgs) { | |
61a3c696 | 684 | veth_error("Can't allocate buffers for LPAR %d.\n", rlp); |
1da177e4 LT |
685 | return -ENOMEM; |
686 | } | |
687 | ||
688 | cnx->msgs = msgs; | |
689 | memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg)); | |
1da177e4 LT |
690 | |
691 | for (i = 0; i < VETH_NUMBUFFERS; i++) { | |
692 | msgs[i].token = i; | |
693 | veth_stack_push(cnx, msgs + i); | |
694 | } | |
695 | ||
696 | cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS); | |
697 | ||
698 | if (cnx->num_events < (2 + VETH_NUMBUFFERS)) { | |
61a3c696 | 699 | veth_error("Can't allocate enough events for LPAR %d.\n", rlp); |
1da177e4 LT |
700 | return -ENOMEM; |
701 | } | |
702 | ||
703 | cnx->local_caps.num_buffers = VETH_NUMBUFFERS; | |
704 | cnx->local_caps.ack_threshold = ACK_THRESHOLD; | |
705 | cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT; | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
f0c129ca | 710 | static void veth_stop_connection(struct veth_lpar_connection *cnx) |
1da177e4 | 711 | { |
f0c129ca | 712 | if (!cnx) |
1da177e4 LT |
713 | return; |
714 | ||
715 | spin_lock_irq(&cnx->lock); | |
716 | cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN; | |
717 | veth_kick_statemachine(cnx); | |
718 | spin_unlock_irq(&cnx->lock); | |
719 | ||
58c5900b ME |
720 | /* There's a slim chance the reset code has just queued the |
721 | * statemachine to run in five seconds. If so we need to cancel | |
722 | * that and requeue the work to run now. */ | |
723 | if (cancel_delayed_work(&cnx->statemachine_wq)) { | |
724 | spin_lock_irq(&cnx->lock); | |
725 | veth_kick_statemachine(cnx); | |
726 | spin_unlock_irq(&cnx->lock); | |
727 | } | |
728 | ||
abfda471 | 729 | /* Wait for the state machine to run. */ |
1da177e4 | 730 | flush_scheduled_work(); |
ec60beeb ME |
731 | } |
732 | ||
f0c129ca | 733 | static void veth_destroy_connection(struct veth_lpar_connection *cnx) |
ec60beeb | 734 | { |
f0c129ca | 735 | if (!cnx) |
ec60beeb | 736 | return; |
1da177e4 | 737 | |
1da177e4 LT |
738 | if (cnx->num_events > 0) |
739 | mf_deallocate_lp_events(cnx->remote_lp, | |
740 | HvLpEvent_Type_VirtualLan, | |
741 | cnx->num_events, | |
742 | NULL, NULL); | |
743 | if (cnx->num_ack_events > 0) | |
744 | mf_deallocate_lp_events(cnx->remote_lp, | |
745 | HvLpEvent_Type_VirtualLan, | |
746 | cnx->num_ack_events, | |
747 | NULL, NULL); | |
1da177e4 LT |
748 | |
749 | kfree(cnx->msgs); | |
f0c129ca | 750 | veth_cnx[cnx->remote_lp] = NULL; |
1da177e4 | 751 | kfree(cnx); |
f0c129ca ME |
752 | } |
753 | ||
754 | static void veth_release_connection(struct kobject *kobj) | |
755 | { | |
756 | struct veth_lpar_connection *cnx; | |
757 | cnx = container_of(kobj, struct veth_lpar_connection, kobject); | |
758 | veth_stop_connection(cnx); | |
759 | veth_destroy_connection(cnx); | |
1da177e4 LT |
760 | } |
761 | ||
762 | /* | |
763 | * net_device code | |
764 | */ | |
765 | ||
766 | static int veth_open(struct net_device *dev) | |
767 | { | |
768 | struct veth_port *port = (struct veth_port *) dev->priv; | |
769 | ||
770 | memset(&port->stats, 0, sizeof (port->stats)); | |
771 | netif_start_queue(dev); | |
772 | return 0; | |
773 | } | |
774 | ||
775 | static int veth_close(struct net_device *dev) | |
776 | { | |
777 | netif_stop_queue(dev); | |
778 | return 0; | |
779 | } | |
780 | ||
781 | static struct net_device_stats *veth_get_stats(struct net_device *dev) | |
782 | { | |
783 | struct veth_port *port = (struct veth_port *) dev->priv; | |
784 | ||
785 | return &port->stats; | |
786 | } | |
787 | ||
788 | static int veth_change_mtu(struct net_device *dev, int new_mtu) | |
789 | { | |
790 | if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU)) | |
791 | return -EINVAL; | |
792 | dev->mtu = new_mtu; | |
793 | return 0; | |
794 | } | |
795 | ||
796 | static void veth_set_multicast_list(struct net_device *dev) | |
797 | { | |
798 | struct veth_port *port = (struct veth_port *) dev->priv; | |
799 | unsigned long flags; | |
800 | ||
801 | write_lock_irqsave(&port->mcast_gate, flags); | |
802 | ||
2a5391a1 ME |
803 | if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) || |
804 | (dev->mc_count > VETH_MAX_MCAST)) { | |
1da177e4 | 805 | port->promiscuous = 1; |
1da177e4 LT |
806 | } else { |
807 | struct dev_mc_list *dmi = dev->mc_list; | |
808 | int i; | |
809 | ||
2a5391a1 ME |
810 | port->promiscuous = 0; |
811 | ||
1da177e4 LT |
812 | /* Update table */ |
813 | port->num_mcast = 0; | |
814 | ||
815 | for (i = 0; i < dev->mc_count; i++) { | |
816 | u8 *addr = dmi->dmi_addr; | |
817 | u64 xaddr = 0; | |
818 | ||
819 | if (addr[0] & 0x01) {/* multicast address? */ | |
820 | memcpy(&xaddr, addr, ETH_ALEN); | |
821 | port->mcast_addr[port->num_mcast] = xaddr; | |
822 | port->num_mcast++; | |
823 | } | |
824 | dmi = dmi->next; | |
825 | } | |
826 | } | |
827 | ||
828 | write_unlock_irqrestore(&port->mcast_gate, flags); | |
829 | } | |
830 | ||
831 | static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | |
832 | { | |
833 | strncpy(info->driver, "veth", sizeof(info->driver) - 1); | |
834 | info->driver[sizeof(info->driver) - 1] = '\0'; | |
835 | strncpy(info->version, "1.0", sizeof(info->version) - 1); | |
836 | } | |
837 | ||
838 | static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd) | |
839 | { | |
840 | ecmd->supported = (SUPPORTED_1000baseT_Full | |
841 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); | |
842 | ecmd->advertising = (SUPPORTED_1000baseT_Full | |
843 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); | |
844 | ecmd->port = PORT_FIBRE; | |
845 | ecmd->transceiver = XCVR_INTERNAL; | |
846 | ecmd->phy_address = 0; | |
847 | ecmd->speed = SPEED_1000; | |
848 | ecmd->duplex = DUPLEX_FULL; | |
849 | ecmd->autoneg = AUTONEG_ENABLE; | |
850 | ecmd->maxtxpkt = 120; | |
851 | ecmd->maxrxpkt = 120; | |
852 | return 0; | |
853 | } | |
854 | ||
855 | static u32 veth_get_link(struct net_device *dev) | |
856 | { | |
857 | return 1; | |
858 | } | |
859 | ||
860 | static struct ethtool_ops ops = { | |
861 | .get_drvinfo = veth_get_drvinfo, | |
862 | .get_settings = veth_get_settings, | |
863 | .get_link = veth_get_link, | |
864 | }; | |
865 | ||
1da177e4 LT |
866 | static struct net_device * __init veth_probe_one(int vlan, struct device *vdev) |
867 | { | |
868 | struct net_device *dev; | |
869 | struct veth_port *port; | |
870 | int i, rc; | |
871 | ||
872 | dev = alloc_etherdev(sizeof (struct veth_port)); | |
873 | if (! dev) { | |
874 | veth_error("Unable to allocate net_device structure!\n"); | |
875 | return NULL; | |
876 | } | |
877 | ||
878 | port = (struct veth_port *) dev->priv; | |
879 | ||
e0808494 | 880 | spin_lock_init(&port->queue_lock); |
1da177e4 | 881 | rwlock_init(&port->mcast_gate); |
e0808494 | 882 | port->stopped_map = 0; |
1da177e4 LT |
883 | |
884 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
885 | HvLpVirtualLanIndexMap map; | |
886 | ||
887 | if (i == this_lp) | |
888 | continue; | |
889 | map = HvLpConfig_getVirtualLanIndexMapForLp(i); | |
890 | if (map & (0x8000 >> vlan)) | |
891 | port->lpar_map |= (1 << i); | |
892 | } | |
893 | port->dev = vdev; | |
894 | ||
895 | dev->dev_addr[0] = 0x02; | |
896 | dev->dev_addr[1] = 0x01; | |
897 | dev->dev_addr[2] = 0xff; | |
898 | dev->dev_addr[3] = vlan; | |
899 | dev->dev_addr[4] = 0xff; | |
900 | dev->dev_addr[5] = this_lp; | |
901 | ||
902 | dev->mtu = VETH_MAX_MTU; | |
903 | ||
904 | memcpy(&port->mac_addr, dev->dev_addr, 6); | |
905 | ||
906 | dev->open = veth_open; | |
907 | dev->hard_start_xmit = veth_start_xmit; | |
908 | dev->stop = veth_close; | |
909 | dev->get_stats = veth_get_stats; | |
910 | dev->change_mtu = veth_change_mtu; | |
911 | dev->set_mac_address = NULL; | |
912 | dev->set_multicast_list = veth_set_multicast_list; | |
913 | SET_ETHTOOL_OPS(dev, &ops); | |
914 | ||
1da177e4 LT |
915 | SET_NETDEV_DEV(dev, vdev); |
916 | ||
917 | rc = register_netdev(dev); | |
918 | if (rc != 0) { | |
61a3c696 | 919 | veth_error("Failed registering net device for vlan%d.\n", vlan); |
1da177e4 LT |
920 | free_netdev(dev); |
921 | return NULL; | |
922 | } | |
923 | ||
61a3c696 ME |
924 | veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n", |
925 | dev->name, vlan, port->lpar_map); | |
1da177e4 LT |
926 | |
927 | return dev; | |
928 | } | |
929 | ||
930 | /* | |
931 | * Tx path | |
932 | */ | |
933 | ||
934 | static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp, | |
935 | struct net_device *dev) | |
936 | { | |
937 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
938 | struct veth_port *port = (struct veth_port *) dev->priv; | |
939 | HvLpEvent_Rc rc; | |
1da177e4 | 940 | struct veth_msg *msg = NULL; |
1da177e4 LT |
941 | unsigned long flags; |
942 | ||
db5e8718 | 943 | if (! cnx) |
1da177e4 | 944 | return 0; |
1da177e4 LT |
945 | |
946 | spin_lock_irqsave(&cnx->lock, flags); | |
947 | ||
f27eff1f | 948 | if (! (cnx->state & VETH_STATE_READY)) |
db5e8718 | 949 | goto no_error; |
1da177e4 | 950 | |
db5e8718 | 951 | if ((skb->len - ETH_HLEN) > VETH_MAX_MTU) |
1da177e4 LT |
952 | goto drop; |
953 | ||
954 | msg = veth_stack_pop(cnx); | |
db5e8718 | 955 | if (! msg) |
1da177e4 | 956 | goto drop; |
1da177e4 | 957 | |
b08bd5c0 | 958 | msg->in_use = 1; |
db5e8718 | 959 | msg->skb = skb_get(skb); |
b08bd5c0 | 960 | |
cbf9074c ME |
961 | msg->data.addr[0] = dma_map_single(port->dev, skb->data, |
962 | skb->len, DMA_TO_DEVICE); | |
1da177e4 | 963 | |
cbf9074c | 964 | if (dma_mapping_error(msg->data.addr[0])) |
1da177e4 LT |
965 | goto recycle_and_drop; |
966 | ||
1da177e4 | 967 | msg->dev = port->dev; |
cbf9074c | 968 | msg->data.len[0] = skb->len; |
1da177e4 | 969 | msg->data.eofmask = 1 << VETH_EOF_SHIFT; |
cbf9074c | 970 | |
1da177e4 LT |
971 | rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data); |
972 | ||
973 | if (rc != HvLpEvent_Rc_Good) | |
974 | goto recycle_and_drop; | |
975 | ||
24562ffa ME |
976 | /* If the timer's not already running, start it now. */ |
977 | if (0 == cnx->outstanding_tx) | |
978 | mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout); | |
979 | ||
980 | cnx->last_contact = jiffies; | |
981 | cnx->outstanding_tx++; | |
982 | ||
e0808494 ME |
983 | if (veth_stack_is_empty(cnx)) |
984 | veth_stop_queues(cnx); | |
985 | ||
db5e8718 | 986 | no_error: |
1da177e4 LT |
987 | spin_unlock_irqrestore(&cnx->lock, flags); |
988 | return 0; | |
989 | ||
990 | recycle_and_drop: | |
1da177e4 LT |
991 | veth_recycle_msg(cnx, msg); |
992 | drop: | |
1da177e4 | 993 | spin_unlock_irqrestore(&cnx->lock, flags); |
db5e8718 | 994 | return 1; |
1da177e4 LT |
995 | } |
996 | ||
db5e8718 | 997 | static void veth_transmit_to_many(struct sk_buff *skb, |
1da177e4 LT |
998 | HvLpIndexMap lpmask, |
999 | struct net_device *dev) | |
1000 | { | |
1001 | struct veth_port *port = (struct veth_port *) dev->priv; | |
db5e8718 ME |
1002 | int i, success, error; |
1003 | ||
1004 | success = error = 0; | |
1da177e4 LT |
1005 | |
1006 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
1007 | if ((lpmask & (1 << i)) == 0) | |
1008 | continue; | |
1009 | ||
db5e8718 ME |
1010 | if (veth_transmit_to_one(skb, i, dev)) |
1011 | error = 1; | |
1012 | else | |
1013 | success = 1; | |
1da177e4 LT |
1014 | } |
1015 | ||
db5e8718 ME |
1016 | if (error) |
1017 | port->stats.tx_errors++; | |
1018 | ||
1019 | if (success) { | |
1da177e4 LT |
1020 | port->stats.tx_packets++; |
1021 | port->stats.tx_bytes += skb->len; | |
1022 | } | |
1da177e4 LT |
1023 | } |
1024 | ||
1025 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
1026 | { | |
1027 | unsigned char *frame = skb->data; | |
1028 | struct veth_port *port = (struct veth_port *) dev->priv; | |
1da177e4 LT |
1029 | HvLpIndexMap lpmask; |
1030 | ||
1031 | if (! (frame[0] & 0x01)) { | |
1032 | /* unicast packet */ | |
1033 | HvLpIndex rlp = frame[5]; | |
1034 | ||
1035 | if ( ! ((1 << rlp) & port->lpar_map) ) { | |
1036 | dev_kfree_skb(skb); | |
1037 | return 0; | |
1038 | } | |
1039 | ||
1040 | lpmask = 1 << rlp; | |
1041 | } else { | |
1042 | lpmask = port->lpar_map; | |
1043 | } | |
1044 | ||
e0808494 | 1045 | veth_transmit_to_many(skb, lpmask, dev); |
1da177e4 | 1046 | |
e0808494 | 1047 | dev_kfree_skb(skb); |
1da177e4 LT |
1048 | |
1049 | return 0; | |
1050 | } | |
1051 | ||
b08bd5c0 | 1052 | /* You must hold the connection's lock when you call this function. */ |
1da177e4 LT |
1053 | static void veth_recycle_msg(struct veth_lpar_connection *cnx, |
1054 | struct veth_msg *msg) | |
1055 | { | |
1056 | u32 dma_address, dma_length; | |
1057 | ||
b08bd5c0 ME |
1058 | if (msg->in_use) { |
1059 | msg->in_use = 0; | |
1da177e4 LT |
1060 | dma_address = msg->data.addr[0]; |
1061 | dma_length = msg->data.len[0]; | |
1062 | ||
cbf9074c ME |
1063 | if (!dma_mapping_error(dma_address)) |
1064 | dma_unmap_single(msg->dev, dma_address, dma_length, | |
1065 | DMA_TO_DEVICE); | |
1da177e4 LT |
1066 | |
1067 | if (msg->skb) { | |
1068 | dev_kfree_skb_any(msg->skb); | |
1069 | msg->skb = NULL; | |
1070 | } | |
1071 | ||
1072 | memset(&msg->data, 0, sizeof(msg->data)); | |
1073 | veth_stack_push(cnx, msg); | |
61a3c696 ME |
1074 | } else if (cnx->state & VETH_STATE_OPEN) { |
1075 | veth_error("Non-pending frame (# %d) acked by LPAR %d.\n", | |
1076 | cnx->remote_lp, msg->token); | |
1077 | } | |
1da177e4 LT |
1078 | } |
1079 | ||
e0808494 | 1080 | static void veth_wake_queues(struct veth_lpar_connection *cnx) |
1da177e4 LT |
1081 | { |
1082 | int i; | |
e0808494 | 1083 | |
1da177e4 LT |
1084 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) { |
1085 | struct net_device *dev = veth_dev[i]; | |
1086 | struct veth_port *port; | |
1087 | unsigned long flags; | |
1088 | ||
1089 | if (! dev) | |
1090 | continue; | |
1091 | ||
1092 | port = (struct veth_port *)dev->priv; | |
1093 | ||
1094 | if (! (port->lpar_map & (1<<cnx->remote_lp))) | |
1095 | continue; | |
1096 | ||
e0808494 ME |
1097 | spin_lock_irqsave(&port->queue_lock, flags); |
1098 | ||
1099 | port->stopped_map &= ~(1 << cnx->remote_lp); | |
1100 | ||
1101 | if (0 == port->stopped_map && netif_queue_stopped(dev)) { | |
1102 | veth_debug("cnx %d: woke queue for %s.\n", | |
1103 | cnx->remote_lp, dev->name); | |
1104 | netif_wake_queue(dev); | |
1da177e4 | 1105 | } |
e0808494 ME |
1106 | spin_unlock_irqrestore(&port->queue_lock, flags); |
1107 | } | |
1108 | } | |
1109 | ||
1110 | static void veth_stop_queues(struct veth_lpar_connection *cnx) | |
1111 | { | |
1112 | int i; | |
1113 | ||
1114 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) { | |
1115 | struct net_device *dev = veth_dev[i]; | |
1116 | struct veth_port *port; | |
1117 | ||
1118 | if (! dev) | |
1119 | continue; | |
1120 | ||
1121 | port = (struct veth_port *)dev->priv; | |
1122 | ||
1123 | /* If this cnx is not on the vlan for this port, continue */ | |
1124 | if (! (port->lpar_map & (1 << cnx->remote_lp))) | |
1125 | continue; | |
1126 | ||
1127 | spin_lock(&port->queue_lock); | |
1128 | ||
1129 | netif_stop_queue(dev); | |
1130 | port->stopped_map |= (1 << cnx->remote_lp); | |
1131 | ||
1132 | veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n", | |
1133 | cnx->remote_lp, dev->name, port->stopped_map); | |
1134 | ||
1135 | spin_unlock(&port->queue_lock); | |
1da177e4 LT |
1136 | } |
1137 | } | |
1138 | ||
24562ffa ME |
1139 | static void veth_timed_reset(unsigned long ptr) |
1140 | { | |
1141 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr; | |
1142 | unsigned long trigger_time, flags; | |
1143 | ||
1144 | /* FIXME is it possible this fires after veth_stop_connection()? | |
1145 | * That would reschedule the statemachine for 5 seconds and probably | |
1146 | * execute it after the module's been unloaded. Hmm. */ | |
1147 | ||
1148 | spin_lock_irqsave(&cnx->lock, flags); | |
1149 | ||
1150 | if (cnx->outstanding_tx > 0) { | |
1151 | trigger_time = cnx->last_contact + cnx->reset_timeout; | |
1152 | ||
1153 | if (trigger_time < jiffies) { | |
1154 | cnx->state |= VETH_STATE_RESET; | |
1155 | veth_kick_statemachine(cnx); | |
1156 | veth_error("%d packets not acked by LPAR %d within %d " | |
1157 | "seconds, resetting.\n", | |
1158 | cnx->outstanding_tx, cnx->remote_lp, | |
1159 | cnx->reset_timeout / HZ); | |
1160 | } else { | |
1161 | /* Reschedule the timer */ | |
1162 | trigger_time = jiffies + cnx->reset_timeout; | |
1163 | mod_timer(&cnx->reset_timer, trigger_time); | |
1164 | } | |
1165 | } | |
1166 | ||
1167 | spin_unlock_irqrestore(&cnx->lock, flags); | |
1168 | } | |
1169 | ||
1da177e4 LT |
1170 | /* |
1171 | * Rx path | |
1172 | */ | |
1173 | ||
1174 | static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr) | |
1175 | { | |
1176 | int wanted = 0; | |
1177 | int i; | |
1178 | unsigned long flags; | |
1179 | ||
1180 | if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) ) | |
1181 | return 1; | |
1182 | ||
1da177e4 LT |
1183 | read_lock_irqsave(&port->mcast_gate, flags); |
1184 | ||
2a5391a1 | 1185 | if (port->promiscuous) { |
1da177e4 LT |
1186 | wanted = 1; |
1187 | goto out; | |
1188 | } | |
1189 | ||
1190 | for (i = 0; i < port->num_mcast; ++i) { | |
1191 | if (port->mcast_addr[i] == mac_addr) { | |
1192 | wanted = 1; | |
1193 | break; | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | out: | |
1198 | read_unlock_irqrestore(&port->mcast_gate, flags); | |
1199 | ||
1200 | return wanted; | |
1201 | } | |
1202 | ||
1203 | struct dma_chunk { | |
1204 | u64 addr; | |
1205 | u64 size; | |
1206 | }; | |
1207 | ||
1208 | #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 ) | |
1209 | ||
1210 | static inline void veth_build_dma_list(struct dma_chunk *list, | |
1211 | unsigned char *p, unsigned long length) | |
1212 | { | |
1213 | unsigned long done; | |
1214 | int i = 1; | |
1215 | ||
1216 | /* FIXME: skbs are continguous in real addresses. Do we | |
1217 | * really need to break it into PAGE_SIZE chunks, or can we do | |
1218 | * it just at the granularity of iSeries real->absolute | |
1219 | * mapping? Indeed, given the way the allocator works, can we | |
1220 | * count on them being absolutely contiguous? */ | |
1221 | list[0].addr = ISERIES_HV_ADDR(p); | |
1222 | list[0].size = min(length, | |
1223 | PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK)); | |
1224 | ||
1225 | done = list[0].size; | |
1226 | while (done < length) { | |
1227 | list[i].addr = ISERIES_HV_ADDR(p + done); | |
1228 | list[i].size = min(length-done, PAGE_SIZE); | |
1229 | done += list[i].size; | |
1230 | i++; | |
1231 | } | |
1232 | } | |
1233 | ||
1234 | static void veth_flush_acks(struct veth_lpar_connection *cnx) | |
1235 | { | |
1236 | HvLpEvent_Rc rc; | |
1237 | ||
1238 | rc = veth_signaldata(cnx, VethEventTypeFramesAck, | |
1239 | 0, &cnx->pending_acks); | |
1240 | ||
1241 | if (rc != HvLpEvent_Rc_Good) | |
61a3c696 ME |
1242 | veth_error("Failed acking frames from LPAR %d, rc = %d\n", |
1243 | cnx->remote_lp, (int)rc); | |
1da177e4 LT |
1244 | |
1245 | cnx->num_pending_acks = 0; | |
1246 | memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks)); | |
1247 | } | |
1248 | ||
1249 | static void veth_receive(struct veth_lpar_connection *cnx, | |
1250 | struct VethLpEvent *event) | |
1251 | { | |
1252 | struct VethFramesData *senddata = &event->u.frames_data; | |
1253 | int startchunk = 0; | |
1254 | int nchunks; | |
1255 | unsigned long flags; | |
1256 | HvLpDma_Rc rc; | |
1257 | ||
1258 | do { | |
1259 | u16 length = 0; | |
1260 | struct sk_buff *skb; | |
1261 | struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME]; | |
1262 | struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG]; | |
1263 | u64 dest; | |
1264 | HvLpVirtualLanIndex vlan; | |
1265 | struct net_device *dev; | |
1266 | struct veth_port *port; | |
1267 | ||
1268 | /* FIXME: do we need this? */ | |
1269 | memset(local_list, 0, sizeof(local_list)); | |
1270 | memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG)); | |
1271 | ||
1272 | /* a 0 address marks the end of the valid entries */ | |
1273 | if (senddata->addr[startchunk] == 0) | |
1274 | break; | |
1275 | ||
1276 | /* make sure that we have at least 1 EOF entry in the | |
1277 | * remaining entries */ | |
1278 | if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) { | |
61a3c696 ME |
1279 | veth_error("Missing EOF fragment in event " |
1280 | "eofmask = 0x%x startchunk = %d\n", | |
1281 | (unsigned)senddata->eofmask, | |
1282 | startchunk); | |
1da177e4 LT |
1283 | break; |
1284 | } | |
1285 | ||
1286 | /* build list of chunks in this frame */ | |
1287 | nchunks = 0; | |
1288 | do { | |
1289 | remote_list[nchunks].addr = | |
1290 | (u64) senddata->addr[startchunk+nchunks] << 32; | |
1291 | remote_list[nchunks].size = | |
1292 | senddata->len[startchunk+nchunks]; | |
1293 | length += remote_list[nchunks].size; | |
1294 | } while (! (senddata->eofmask & | |
1295 | (1 << (VETH_EOF_SHIFT + startchunk + nchunks++)))); | |
1296 | ||
1297 | /* length == total length of all chunks */ | |
1298 | /* nchunks == # of chunks in this frame */ | |
1299 | ||
1300 | if ((length - ETH_HLEN) > VETH_MAX_MTU) { | |
61a3c696 ME |
1301 | veth_error("Received oversize frame from LPAR %d " |
1302 | "(length = %d)\n", | |
1303 | cnx->remote_lp, length); | |
1da177e4 LT |
1304 | continue; |
1305 | } | |
1306 | ||
1307 | skb = alloc_skb(length, GFP_ATOMIC); | |
1308 | if (!skb) | |
1309 | continue; | |
1310 | ||
1311 | veth_build_dma_list(local_list, skb->data, length); | |
1312 | ||
1313 | rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan, | |
1314 | event->base_event.xSourceLp, | |
1315 | HvLpDma_Direction_RemoteToLocal, | |
1316 | cnx->src_inst, | |
1317 | cnx->dst_inst, | |
1318 | HvLpDma_AddressType_RealAddress, | |
1319 | HvLpDma_AddressType_TceIndex, | |
1320 | ISERIES_HV_ADDR(&local_list), | |
1321 | ISERIES_HV_ADDR(&remote_list), | |
1322 | length); | |
1323 | if (rc != HvLpDma_Rc_Good) { | |
1324 | dev_kfree_skb_irq(skb); | |
1325 | continue; | |
1326 | } | |
1327 | ||
1328 | vlan = skb->data[9]; | |
1329 | dev = veth_dev[vlan]; | |
41664c03 ME |
1330 | if (! dev) { |
1331 | /* | |
1332 | * Some earlier versions of the driver sent | |
1333 | * broadcasts down all connections, even to lpars | |
1334 | * that weren't on the relevant vlan. So ignore | |
1335 | * packets belonging to a vlan we're not on. | |
1336 | * We can also be here if we receive packets while | |
1337 | * the driver is going down, because then dev is NULL. | |
1338 | */ | |
1339 | dev_kfree_skb_irq(skb); | |
1da177e4 | 1340 | continue; |
41664c03 | 1341 | } |
1da177e4 LT |
1342 | |
1343 | port = (struct veth_port *)dev->priv; | |
1344 | dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000; | |
1345 | ||
1346 | if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) { | |
1347 | dev_kfree_skb_irq(skb); | |
1348 | continue; | |
1349 | } | |
1350 | if (! veth_frame_wanted(port, dest)) { | |
1351 | dev_kfree_skb_irq(skb); | |
1352 | continue; | |
1353 | } | |
1354 | ||
1355 | skb_put(skb, length); | |
1356 | skb->dev = dev; | |
1357 | skb->protocol = eth_type_trans(skb, dev); | |
1358 | skb->ip_summed = CHECKSUM_NONE; | |
1359 | netif_rx(skb); /* send it up */ | |
1360 | port->stats.rx_packets++; | |
1361 | port->stats.rx_bytes += length; | |
1362 | } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG); | |
1363 | ||
1364 | /* Ack it */ | |
1365 | spin_lock_irqsave(&cnx->lock, flags); | |
1366 | BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG); | |
1367 | ||
1368 | cnx->pending_acks[cnx->num_pending_acks++] = | |
1369 | event->base_event.xCorrelationToken; | |
1370 | ||
1371 | if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) | |
1372 | || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) ) | |
1373 | veth_flush_acks(cnx); | |
1374 | ||
1375 | spin_unlock_irqrestore(&cnx->lock, flags); | |
1376 | } | |
1377 | ||
1378 | static void veth_timed_ack(unsigned long ptr) | |
1379 | { | |
1380 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr; | |
1381 | unsigned long flags; | |
1382 | ||
1383 | /* Ack all the events */ | |
1384 | spin_lock_irqsave(&cnx->lock, flags); | |
1385 | if (cnx->num_pending_acks > 0) | |
1386 | veth_flush_acks(cnx); | |
1387 | ||
1388 | /* Reschedule the timer */ | |
1389 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; | |
1390 | add_timer(&cnx->ack_timer); | |
1391 | spin_unlock_irqrestore(&cnx->lock, flags); | |
1392 | } | |
1393 | ||
1394 | static int veth_remove(struct vio_dev *vdev) | |
1395 | { | |
f0c129ca | 1396 | struct veth_lpar_connection *cnx; |
1da177e4 | 1397 | struct net_device *dev; |
f0c129ca ME |
1398 | struct veth_port *port; |
1399 | int i; | |
1da177e4 | 1400 | |
f0c129ca ME |
1401 | dev = veth_dev[vdev->unit_address]; |
1402 | ||
1403 | if (! dev) | |
1404 | return 0; | |
1405 | ||
1406 | port = netdev_priv(dev); | |
1407 | ||
1408 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
1409 | cnx = veth_cnx[i]; | |
1410 | ||
1411 | if (cnx && (port->lpar_map & (1 << i))) { | |
1412 | /* Drop our reference to connections on our VLAN */ | |
1413 | kobject_put(&cnx->kobject); | |
1414 | } | |
1da177e4 | 1415 | } |
f0c129ca ME |
1416 | |
1417 | veth_dev[vdev->unit_address] = NULL; | |
1418 | unregister_netdev(dev); | |
1419 | free_netdev(dev); | |
1420 | ||
1da177e4 LT |
1421 | return 0; |
1422 | } | |
1423 | ||
1424 | static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id) | |
1425 | { | |
1426 | int i = vdev->unit_address; | |
1427 | struct net_device *dev; | |
f0c129ca | 1428 | struct veth_port *port; |
1da177e4 LT |
1429 | |
1430 | dev = veth_probe_one(i, &vdev->dev); | |
1431 | if (dev == NULL) { | |
1432 | veth_remove(vdev); | |
1433 | return 1; | |
1434 | } | |
1435 | veth_dev[i] = dev; | |
1436 | ||
f0c129ca ME |
1437 | port = (struct veth_port*)netdev_priv(dev); |
1438 | ||
1439 | /* Start the state machine on each connection on this vlan. If we're | |
1440 | * the first dev to do so this will commence link negotiation */ | |
1441 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
1442 | struct veth_lpar_connection *cnx; | |
1443 | ||
1444 | if (! (port->lpar_map & (1 << i))) | |
1445 | continue; | |
1446 | ||
1447 | cnx = veth_cnx[i]; | |
1448 | if (!cnx) | |
1449 | continue; | |
1450 | ||
1451 | kobject_get(&cnx->kobject); | |
1452 | veth_kick_statemachine(cnx); | |
1453 | } | |
1da177e4 LT |
1454 | |
1455 | return 0; | |
1456 | } | |
1457 | ||
1458 | /** | |
1459 | * veth_device_table: Used by vio.c to match devices that we | |
1460 | * support. | |
1461 | */ | |
1462 | static struct vio_device_id veth_device_table[] __devinitdata = { | |
1463 | { "vlan", "" }, | |
fb120da6 | 1464 | { "", "" } |
1da177e4 LT |
1465 | }; |
1466 | MODULE_DEVICE_TABLE(vio, veth_device_table); | |
1467 | ||
1468 | static struct vio_driver veth_driver = { | |
1469 | .name = "iseries_veth", | |
1470 | .id_table = veth_device_table, | |
1471 | .probe = veth_probe, | |
1472 | .remove = veth_remove | |
1473 | }; | |
1474 | ||
1475 | /* | |
1476 | * Module initialization/cleanup | |
1477 | */ | |
1478 | ||
1479 | void __exit veth_module_cleanup(void) | |
1480 | { | |
1481 | int i; | |
f0c129ca | 1482 | struct veth_lpar_connection *cnx; |
1da177e4 | 1483 | |
f0c129ca | 1484 | /* Disconnect our "irq" to stop events coming from the Hypervisor. */ |
1da177e4 LT |
1485 | HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan); |
1486 | ||
f0c129ca | 1487 | /* Make sure any work queued from Hypervisor callbacks is finished. */ |
1da177e4 LT |
1488 | flush_scheduled_work(); |
1489 | ||
f0c129ca ME |
1490 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) { |
1491 | cnx = veth_cnx[i]; | |
1492 | ||
1493 | if (!cnx) | |
1494 | continue; | |
b2e0852e | 1495 | |
f0c129ca ME |
1496 | /* Drop the driver's reference to the connection */ |
1497 | kobject_put(&cnx->kobject); | |
1498 | } | |
1da177e4 | 1499 | |
f0c129ca ME |
1500 | /* Unregister the driver, which will close all the netdevs and stop |
1501 | * the connections when they're no longer referenced. */ | |
1502 | vio_unregister_driver(&veth_driver); | |
1da177e4 LT |
1503 | } |
1504 | module_exit(veth_module_cleanup); | |
1505 | ||
1506 | int __init veth_module_init(void) | |
1507 | { | |
1508 | int i; | |
1509 | int rc; | |
1510 | ||
1511 | this_lp = HvLpConfig_getLpIndex_outline(); | |
1512 | ||
1513 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) { | |
1514 | rc = veth_init_connection(i); | |
ec60beeb ME |
1515 | if (rc != 0) |
1516 | goto error; | |
1da177e4 LT |
1517 | } |
1518 | ||
1519 | HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan, | |
1520 | &veth_handle_event); | |
1521 | ||
ec60beeb ME |
1522 | rc = vio_register_driver(&veth_driver); |
1523 | if (rc != 0) | |
1524 | goto error; | |
1525 | ||
1526 | return 0; | |
1527 | ||
1528 | error: | |
1529 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) { | |
f0c129ca | 1530 | veth_destroy_connection(veth_cnx[i]); |
ec60beeb ME |
1531 | } |
1532 | ||
1533 | return rc; | |
1da177e4 LT |
1534 | } |
1535 | module_init(veth_module_init); |