]>
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; | |
120 | unsigned long in_use; | |
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 | ||
132 | struct timer_list ack_timer; | |
133 | ||
134 | spinlock_t lock; | |
135 | unsigned long state; | |
136 | HvLpInstanceId src_inst; | |
137 | HvLpInstanceId dst_inst; | |
138 | struct VethLpEvent cap_event, cap_ack_event; | |
139 | u16 pending_acks[VETH_MAX_ACKS_PER_MSG]; | |
140 | u32 num_pending_acks; | |
141 | ||
142 | int num_ack_events; | |
143 | struct VethCapData remote_caps; | |
144 | u32 ack_timeout; | |
145 | ||
146 | spinlock_t msg_stack_lock; | |
147 | struct veth_msg *msg_stack_head; | |
148 | }; | |
149 | ||
150 | struct veth_port { | |
151 | struct device *dev; | |
152 | struct net_device_stats stats; | |
153 | u64 mac_addr; | |
154 | HvLpIndexMap lpar_map; | |
155 | ||
156 | spinlock_t pending_gate; | |
157 | struct sk_buff *pending_skb; | |
158 | HvLpIndexMap pending_lpmask; | |
159 | ||
160 | rwlock_t mcast_gate; | |
161 | int promiscuous; | |
162 | int all_mcast; | |
163 | int num_mcast; | |
164 | u64 mcast_addr[VETH_MAX_MCAST]; | |
165 | }; | |
166 | ||
167 | static HvLpIndex this_lp; | |
168 | static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */ | |
169 | static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */ | |
170 | ||
171 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev); | |
172 | static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *); | |
173 | static void veth_flush_pending(struct veth_lpar_connection *cnx); | |
174 | static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *); | |
175 | static void veth_timed_ack(unsigned long connectionPtr); | |
176 | ||
177 | /* | |
178 | * Utility functions | |
179 | */ | |
180 | ||
61a3c696 ME |
181 | #define veth_info(fmt, args...) \ |
182 | printk(KERN_INFO "iseries_veth: " fmt, ## args) | |
1da177e4 LT |
183 | |
184 | #define veth_error(fmt, args...) \ | |
61a3c696 ME |
185 | printk(KERN_ERR "iseries_veth: Error: " fmt, ## args) |
186 | ||
187 | #ifdef DEBUG | |
188 | #define veth_debug(fmt, args...) \ | |
189 | printk(KERN_DEBUG "iseries_veth: " fmt, ## args) | |
190 | #else | |
191 | #define veth_debug(fmt, args...) do {} while (0) | |
192 | #endif | |
1da177e4 LT |
193 | |
194 | static inline void veth_stack_push(struct veth_lpar_connection *cnx, | |
195 | struct veth_msg *msg) | |
196 | { | |
197 | unsigned long flags; | |
198 | ||
199 | spin_lock_irqsave(&cnx->msg_stack_lock, flags); | |
200 | msg->next = cnx->msg_stack_head; | |
201 | cnx->msg_stack_head = msg; | |
202 | spin_unlock_irqrestore(&cnx->msg_stack_lock, flags); | |
203 | } | |
204 | ||
205 | static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx) | |
206 | { | |
207 | unsigned long flags; | |
208 | struct veth_msg *msg; | |
209 | ||
210 | spin_lock_irqsave(&cnx->msg_stack_lock, flags); | |
211 | msg = cnx->msg_stack_head; | |
212 | if (msg) | |
213 | cnx->msg_stack_head = cnx->msg_stack_head->next; | |
214 | spin_unlock_irqrestore(&cnx->msg_stack_lock, flags); | |
215 | return msg; | |
216 | } | |
217 | ||
218 | static inline HvLpEvent_Rc | |
219 | veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype, | |
220 | HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype, | |
221 | u64 token, | |
222 | u64 data1, u64 data2, u64 data3, u64 data4, u64 data5) | |
223 | { | |
224 | return HvCallEvent_signalLpEventFast(cnx->remote_lp, | |
225 | HvLpEvent_Type_VirtualLan, | |
226 | subtype, ackind, acktype, | |
227 | cnx->src_inst, | |
228 | cnx->dst_inst, | |
229 | token, data1, data2, data3, | |
230 | data4, data5); | |
231 | } | |
232 | ||
233 | static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx, | |
234 | u16 subtype, u64 token, void *data) | |
235 | { | |
236 | u64 *p = (u64 *) data; | |
237 | ||
238 | return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck, | |
239 | HvLpEvent_AckType_ImmediateAck, | |
240 | token, p[0], p[1], p[2], p[3], p[4]); | |
241 | } | |
242 | ||
243 | struct veth_allocation { | |
244 | struct completion c; | |
245 | int num; | |
246 | }; | |
247 | ||
248 | static void veth_complete_allocation(void *parm, int number) | |
249 | { | |
250 | struct veth_allocation *vc = (struct veth_allocation *)parm; | |
251 | ||
252 | vc->num = number; | |
253 | complete(&vc->c); | |
254 | } | |
255 | ||
256 | static int veth_allocate_events(HvLpIndex rlp, int number) | |
257 | { | |
258 | struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 }; | |
259 | ||
260 | mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan, | |
261 | sizeof(struct VethLpEvent), number, | |
262 | &veth_complete_allocation, &vc); | |
263 | wait_for_completion(&vc.c); | |
264 | ||
265 | return vc.num; | |
266 | } | |
267 | ||
268 | /* | |
269 | * LPAR connection code | |
270 | */ | |
271 | ||
272 | static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx) | |
273 | { | |
274 | schedule_work(&cnx->statemachine_wq); | |
275 | } | |
276 | ||
277 | static void veth_take_cap(struct veth_lpar_connection *cnx, | |
278 | struct VethLpEvent *event) | |
279 | { | |
280 | unsigned long flags; | |
281 | ||
282 | spin_lock_irqsave(&cnx->lock, flags); | |
283 | /* Receiving caps may mean the other end has just come up, so | |
284 | * we need to reload the instance ID of the far end */ | |
285 | cnx->dst_inst = | |
286 | HvCallEvent_getTargetLpInstanceId(cnx->remote_lp, | |
287 | HvLpEvent_Type_VirtualLan); | |
288 | ||
289 | if (cnx->state & VETH_STATE_GOTCAPS) { | |
61a3c696 | 290 | veth_error("Received a second capabilities from LPAR %d.\n", |
1da177e4 LT |
291 | cnx->remote_lp); |
292 | event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable; | |
293 | HvCallEvent_ackLpEvent((struct HvLpEvent *) event); | |
294 | } else { | |
295 | memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event)); | |
296 | cnx->state |= VETH_STATE_GOTCAPS; | |
297 | veth_kick_statemachine(cnx); | |
298 | } | |
299 | spin_unlock_irqrestore(&cnx->lock, flags); | |
300 | } | |
301 | ||
302 | static void veth_take_cap_ack(struct veth_lpar_connection *cnx, | |
303 | struct VethLpEvent *event) | |
304 | { | |
305 | unsigned long flags; | |
306 | ||
307 | spin_lock_irqsave(&cnx->lock, flags); | |
308 | if (cnx->state & VETH_STATE_GOTCAPACK) { | |
61a3c696 | 309 | veth_error("Received a second capabilities ack from LPAR %d.\n", |
1da177e4 LT |
310 | cnx->remote_lp); |
311 | } else { | |
312 | memcpy(&cnx->cap_ack_event, event, | |
313 | sizeof(&cnx->cap_ack_event)); | |
314 | cnx->state |= VETH_STATE_GOTCAPACK; | |
315 | veth_kick_statemachine(cnx); | |
316 | } | |
317 | spin_unlock_irqrestore(&cnx->lock, flags); | |
318 | } | |
319 | ||
320 | static void veth_take_monitor_ack(struct veth_lpar_connection *cnx, | |
321 | struct VethLpEvent *event) | |
322 | { | |
323 | unsigned long flags; | |
324 | ||
325 | spin_lock_irqsave(&cnx->lock, flags); | |
61a3c696 | 326 | veth_debug("cnx %d: lost connection.\n", cnx->remote_lp); |
1da177e4 LT |
327 | cnx->state |= VETH_STATE_RESET; |
328 | veth_kick_statemachine(cnx); | |
329 | spin_unlock_irqrestore(&cnx->lock, flags); | |
330 | } | |
331 | ||
332 | static void veth_handle_ack(struct VethLpEvent *event) | |
333 | { | |
334 | HvLpIndex rlp = event->base_event.xTargetLp; | |
335 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
336 | ||
337 | BUG_ON(! cnx); | |
338 | ||
339 | switch (event->base_event.xSubtype) { | |
340 | case VethEventTypeCap: | |
341 | veth_take_cap_ack(cnx, event); | |
342 | break; | |
343 | case VethEventTypeMonitor: | |
344 | veth_take_monitor_ack(cnx, event); | |
345 | break; | |
346 | default: | |
61a3c696 ME |
347 | veth_error("Unknown ack type %d from LPAR %d.\n", |
348 | event->base_event.xSubtype, rlp); | |
1da177e4 LT |
349 | }; |
350 | } | |
351 | ||
352 | static void veth_handle_int(struct VethLpEvent *event) | |
353 | { | |
354 | HvLpIndex rlp = event->base_event.xSourceLp; | |
355 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
356 | unsigned long flags; | |
357 | int i; | |
358 | ||
359 | BUG_ON(! cnx); | |
360 | ||
361 | switch (event->base_event.xSubtype) { | |
362 | case VethEventTypeCap: | |
363 | veth_take_cap(cnx, event); | |
364 | break; | |
365 | case VethEventTypeMonitor: | |
366 | /* do nothing... this'll hang out here til we're dead, | |
367 | * and the hypervisor will return it for us. */ | |
368 | break; | |
369 | case VethEventTypeFramesAck: | |
370 | spin_lock_irqsave(&cnx->lock, flags); | |
371 | for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) { | |
372 | u16 msgnum = event->u.frames_ack_data.token[i]; | |
373 | ||
374 | if (msgnum < VETH_NUMBUFFERS) | |
375 | veth_recycle_msg(cnx, cnx->msgs + msgnum); | |
376 | } | |
377 | spin_unlock_irqrestore(&cnx->lock, flags); | |
378 | veth_flush_pending(cnx); | |
379 | break; | |
380 | case VethEventTypeFrames: | |
381 | veth_receive(cnx, event); | |
382 | break; | |
383 | default: | |
61a3c696 ME |
384 | veth_error("Unknown interrupt type %d from LPAR %d.\n", |
385 | event->base_event.xSubtype, rlp); | |
1da177e4 LT |
386 | }; |
387 | } | |
388 | ||
389 | static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs) | |
390 | { | |
391 | struct VethLpEvent *veth_event = (struct VethLpEvent *)event; | |
392 | ||
393 | if (event->xFlags.xFunction == HvLpEvent_Function_Ack) | |
394 | veth_handle_ack(veth_event); | |
395 | else if (event->xFlags.xFunction == HvLpEvent_Function_Int) | |
396 | veth_handle_int(veth_event); | |
397 | } | |
398 | ||
399 | static int veth_process_caps(struct veth_lpar_connection *cnx) | |
400 | { | |
401 | struct VethCapData *remote_caps = &cnx->remote_caps; | |
402 | int num_acks_needed; | |
403 | ||
404 | /* Convert timer to jiffies */ | |
405 | cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000; | |
406 | ||
407 | if ( (remote_caps->num_buffers == 0) | |
408 | || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) | |
409 | || (remote_caps->ack_threshold == 0) | |
410 | || (cnx->ack_timeout == 0) ) { | |
61a3c696 ME |
411 | veth_error("Received incompatible capabilities from LPAR %d.\n", |
412 | cnx->remote_lp); | |
1da177e4 LT |
413 | return HvLpEvent_Rc_InvalidSubtypeData; |
414 | } | |
415 | ||
416 | num_acks_needed = (remote_caps->num_buffers | |
417 | / remote_caps->ack_threshold) + 1; | |
418 | ||
419 | /* FIXME: locking on num_ack_events? */ | |
420 | if (cnx->num_ack_events < num_acks_needed) { | |
421 | int num; | |
422 | ||
423 | num = veth_allocate_events(cnx->remote_lp, | |
424 | num_acks_needed-cnx->num_ack_events); | |
425 | if (num > 0) | |
426 | cnx->num_ack_events += num; | |
427 | ||
428 | if (cnx->num_ack_events < num_acks_needed) { | |
61a3c696 ME |
429 | veth_error("Couldn't allocate enough ack events " |
430 | "for LPAR %d.\n", cnx->remote_lp); | |
1da177e4 LT |
431 | |
432 | return HvLpEvent_Rc_BufferNotAvailable; | |
433 | } | |
434 | } | |
435 | ||
436 | ||
437 | return HvLpEvent_Rc_Good; | |
438 | } | |
439 | ||
440 | /* FIXME: The gotos here are a bit dubious */ | |
441 | static void veth_statemachine(void *p) | |
442 | { | |
443 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p; | |
444 | int rlp = cnx->remote_lp; | |
445 | int rc; | |
446 | ||
447 | spin_lock_irq(&cnx->lock); | |
448 | ||
449 | restart: | |
450 | if (cnx->state & VETH_STATE_RESET) { | |
451 | int i; | |
452 | ||
453 | del_timer(&cnx->ack_timer); | |
454 | ||
455 | if (cnx->state & VETH_STATE_OPEN) | |
456 | HvCallEvent_closeLpEventPath(cnx->remote_lp, | |
457 | HvLpEvent_Type_VirtualLan); | |
458 | ||
459 | /* reset ack data */ | |
460 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); | |
461 | cnx->num_pending_acks = 0; | |
462 | ||
463 | cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON | |
464 | | VETH_STATE_OPEN | VETH_STATE_SENTCAPS | |
465 | | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS | |
466 | | VETH_STATE_SENTCAPACK | VETH_STATE_READY); | |
467 | ||
468 | /* Clean up any leftover messages */ | |
469 | if (cnx->msgs) | |
470 | for (i = 0; i < VETH_NUMBUFFERS; ++i) | |
471 | veth_recycle_msg(cnx, cnx->msgs + i); | |
472 | spin_unlock_irq(&cnx->lock); | |
473 | veth_flush_pending(cnx); | |
474 | spin_lock_irq(&cnx->lock); | |
475 | if (cnx->state & VETH_STATE_RESET) | |
476 | goto restart; | |
477 | } | |
478 | ||
479 | if (cnx->state & VETH_STATE_SHUTDOWN) | |
480 | /* It's all over, do nothing */ | |
481 | goto out; | |
482 | ||
483 | if ( !(cnx->state & VETH_STATE_OPEN) ) { | |
484 | if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) ) | |
485 | goto cant_cope; | |
486 | ||
487 | HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan); | |
488 | cnx->src_inst = | |
489 | HvCallEvent_getSourceLpInstanceId(rlp, | |
490 | HvLpEvent_Type_VirtualLan); | |
491 | cnx->dst_inst = | |
492 | HvCallEvent_getTargetLpInstanceId(rlp, | |
493 | HvLpEvent_Type_VirtualLan); | |
494 | cnx->state |= VETH_STATE_OPEN; | |
495 | } | |
496 | ||
497 | if ( (cnx->state & VETH_STATE_OPEN) | |
498 | && !(cnx->state & VETH_STATE_SENTMON) ) { | |
499 | rc = veth_signalevent(cnx, VethEventTypeMonitor, | |
500 | HvLpEvent_AckInd_DoAck, | |
501 | HvLpEvent_AckType_DeferredAck, | |
502 | 0, 0, 0, 0, 0, 0); | |
503 | ||
504 | if (rc == HvLpEvent_Rc_Good) { | |
505 | cnx->state |= VETH_STATE_SENTMON; | |
506 | } else { | |
507 | if ( (rc != HvLpEvent_Rc_PartitionDead) | |
508 | && (rc != HvLpEvent_Rc_PathClosed) ) | |
61a3c696 ME |
509 | veth_error("Error sending monitor to LPAR %d, " |
510 | "rc = %d\n", rlp, rc); | |
1da177e4 LT |
511 | |
512 | /* Oh well, hope we get a cap from the other | |
513 | * end and do better when that kicks us */ | |
514 | goto out; | |
515 | } | |
516 | } | |
517 | ||
518 | if ( (cnx->state & VETH_STATE_OPEN) | |
519 | && !(cnx->state & VETH_STATE_SENTCAPS)) { | |
520 | u64 *rawcap = (u64 *)&cnx->local_caps; | |
521 | ||
522 | rc = veth_signalevent(cnx, VethEventTypeCap, | |
523 | HvLpEvent_AckInd_DoAck, | |
524 | HvLpEvent_AckType_ImmediateAck, | |
525 | 0, rawcap[0], rawcap[1], rawcap[2], | |
526 | rawcap[3], rawcap[4]); | |
527 | ||
528 | if (rc == HvLpEvent_Rc_Good) { | |
529 | cnx->state |= VETH_STATE_SENTCAPS; | |
530 | } else { | |
531 | if ( (rc != HvLpEvent_Rc_PartitionDead) | |
532 | && (rc != HvLpEvent_Rc_PathClosed) ) | |
61a3c696 ME |
533 | veth_error("Error sending caps to LPAR %d, " |
534 | "rc = %d\n", rlp, rc); | |
535 | ||
1da177e4 LT |
536 | /* Oh well, hope we get a cap from the other |
537 | * end and do better when that kicks us */ | |
538 | goto out; | |
539 | } | |
540 | } | |
541 | ||
542 | if ((cnx->state & VETH_STATE_GOTCAPS) | |
543 | && !(cnx->state & VETH_STATE_SENTCAPACK)) { | |
544 | struct VethCapData *remote_caps = &cnx->remote_caps; | |
545 | ||
546 | memcpy(remote_caps, &cnx->cap_event.u.caps_data, | |
547 | sizeof(*remote_caps)); | |
548 | ||
549 | spin_unlock_irq(&cnx->lock); | |
550 | rc = veth_process_caps(cnx); | |
551 | spin_lock_irq(&cnx->lock); | |
552 | ||
553 | /* We dropped the lock, so recheck for anything which | |
554 | * might mess us up */ | |
555 | if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN)) | |
556 | goto restart; | |
557 | ||
558 | cnx->cap_event.base_event.xRc = rc; | |
559 | HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event); | |
560 | if (rc == HvLpEvent_Rc_Good) | |
561 | cnx->state |= VETH_STATE_SENTCAPACK; | |
562 | else | |
563 | goto cant_cope; | |
564 | } | |
565 | ||
566 | if ((cnx->state & VETH_STATE_GOTCAPACK) | |
567 | && (cnx->state & VETH_STATE_GOTCAPS) | |
568 | && !(cnx->state & VETH_STATE_READY)) { | |
569 | if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) { | |
570 | /* Start the ACK timer */ | |
571 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; | |
572 | add_timer(&cnx->ack_timer); | |
573 | cnx->state |= VETH_STATE_READY; | |
574 | } else { | |
61a3c696 ME |
575 | veth_error("Caps rejected by LPAR %d, rc = %d\n", |
576 | rlp, cnx->cap_ack_event.base_event.xRc); | |
1da177e4 LT |
577 | goto cant_cope; |
578 | } | |
579 | } | |
580 | ||
581 | out: | |
582 | spin_unlock_irq(&cnx->lock); | |
583 | return; | |
584 | ||
585 | cant_cope: | |
586 | /* FIXME: we get here if something happens we really can't | |
587 | * cope with. The link will never work once we get here, and | |
588 | * all we can do is not lock the rest of the system up */ | |
61a3c696 ME |
589 | veth_error("Unrecoverable error on connection to LPAR %d, shutting down" |
590 | " (state = 0x%04lx)\n", rlp, cnx->state); | |
1da177e4 LT |
591 | cnx->state |= VETH_STATE_SHUTDOWN; |
592 | spin_unlock_irq(&cnx->lock); | |
593 | } | |
594 | ||
595 | static int veth_init_connection(u8 rlp) | |
596 | { | |
597 | struct veth_lpar_connection *cnx; | |
598 | struct veth_msg *msgs; | |
599 | int i; | |
600 | ||
601 | if ( (rlp == this_lp) | |
602 | || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) ) | |
603 | return 0; | |
604 | ||
605 | cnx = kmalloc(sizeof(*cnx), GFP_KERNEL); | |
606 | if (! cnx) | |
607 | return -ENOMEM; | |
608 | memset(cnx, 0, sizeof(*cnx)); | |
609 | ||
610 | cnx->remote_lp = rlp; | |
611 | spin_lock_init(&cnx->lock); | |
612 | INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx); | |
613 | init_timer(&cnx->ack_timer); | |
614 | cnx->ack_timer.function = veth_timed_ack; | |
615 | cnx->ack_timer.data = (unsigned long) cnx; | |
616 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); | |
617 | ||
618 | veth_cnx[rlp] = cnx; | |
619 | ||
620 | msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL); | |
621 | if (! msgs) { | |
61a3c696 | 622 | veth_error("Can't allocate buffers for LPAR %d.\n", rlp); |
1da177e4 LT |
623 | return -ENOMEM; |
624 | } | |
625 | ||
626 | cnx->msgs = msgs; | |
627 | memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg)); | |
628 | spin_lock_init(&cnx->msg_stack_lock); | |
629 | ||
630 | for (i = 0; i < VETH_NUMBUFFERS; i++) { | |
631 | msgs[i].token = i; | |
632 | veth_stack_push(cnx, msgs + i); | |
633 | } | |
634 | ||
635 | cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS); | |
636 | ||
637 | if (cnx->num_events < (2 + VETH_NUMBUFFERS)) { | |
61a3c696 | 638 | veth_error("Can't allocate enough events for LPAR %d.\n", rlp); |
1da177e4 LT |
639 | return -ENOMEM; |
640 | } | |
641 | ||
642 | cnx->local_caps.num_buffers = VETH_NUMBUFFERS; | |
643 | cnx->local_caps.ack_threshold = ACK_THRESHOLD; | |
644 | cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT; | |
645 | ||
646 | return 0; | |
647 | } | |
648 | ||
649 | static void veth_stop_connection(u8 rlp) | |
650 | { | |
651 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
652 | ||
653 | if (! cnx) | |
654 | return; | |
655 | ||
656 | spin_lock_irq(&cnx->lock); | |
657 | cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN; | |
658 | veth_kick_statemachine(cnx); | |
659 | spin_unlock_irq(&cnx->lock); | |
660 | ||
661 | flush_scheduled_work(); | |
662 | ||
663 | /* FIXME: not sure if this is necessary - will already have | |
664 | * been deleted by the state machine, just want to make sure | |
665 | * its not running any more */ | |
666 | del_timer_sync(&cnx->ack_timer); | |
667 | ||
668 | if (cnx->num_events > 0) | |
669 | mf_deallocate_lp_events(cnx->remote_lp, | |
670 | HvLpEvent_Type_VirtualLan, | |
671 | cnx->num_events, | |
672 | NULL, NULL); | |
673 | if (cnx->num_ack_events > 0) | |
674 | mf_deallocate_lp_events(cnx->remote_lp, | |
675 | HvLpEvent_Type_VirtualLan, | |
676 | cnx->num_ack_events, | |
677 | NULL, NULL); | |
678 | } | |
679 | ||
680 | static void veth_destroy_connection(u8 rlp) | |
681 | { | |
682 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
683 | ||
684 | if (! cnx) | |
685 | return; | |
686 | ||
687 | kfree(cnx->msgs); | |
688 | kfree(cnx); | |
689 | veth_cnx[rlp] = NULL; | |
690 | } | |
691 | ||
692 | /* | |
693 | * net_device code | |
694 | */ | |
695 | ||
696 | static int veth_open(struct net_device *dev) | |
697 | { | |
698 | struct veth_port *port = (struct veth_port *) dev->priv; | |
699 | ||
700 | memset(&port->stats, 0, sizeof (port->stats)); | |
701 | netif_start_queue(dev); | |
702 | return 0; | |
703 | } | |
704 | ||
705 | static int veth_close(struct net_device *dev) | |
706 | { | |
707 | netif_stop_queue(dev); | |
708 | return 0; | |
709 | } | |
710 | ||
711 | static struct net_device_stats *veth_get_stats(struct net_device *dev) | |
712 | { | |
713 | struct veth_port *port = (struct veth_port *) dev->priv; | |
714 | ||
715 | return &port->stats; | |
716 | } | |
717 | ||
718 | static int veth_change_mtu(struct net_device *dev, int new_mtu) | |
719 | { | |
720 | if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU)) | |
721 | return -EINVAL; | |
722 | dev->mtu = new_mtu; | |
723 | return 0; | |
724 | } | |
725 | ||
726 | static void veth_set_multicast_list(struct net_device *dev) | |
727 | { | |
728 | struct veth_port *port = (struct veth_port *) dev->priv; | |
729 | unsigned long flags; | |
730 | ||
731 | write_lock_irqsave(&port->mcast_gate, flags); | |
732 | ||
733 | if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */ | |
734 | printk(KERN_INFO "%s: Promiscuous mode enabled.\n", | |
735 | dev->name); | |
736 | port->promiscuous = 1; | |
737 | } else if ( (dev->flags & IFF_ALLMULTI) | |
738 | || (dev->mc_count > VETH_MAX_MCAST) ) { | |
739 | port->all_mcast = 1; | |
740 | } else { | |
741 | struct dev_mc_list *dmi = dev->mc_list; | |
742 | int i; | |
743 | ||
744 | /* Update table */ | |
745 | port->num_mcast = 0; | |
746 | ||
747 | for (i = 0; i < dev->mc_count; i++) { | |
748 | u8 *addr = dmi->dmi_addr; | |
749 | u64 xaddr = 0; | |
750 | ||
751 | if (addr[0] & 0x01) {/* multicast address? */ | |
752 | memcpy(&xaddr, addr, ETH_ALEN); | |
753 | port->mcast_addr[port->num_mcast] = xaddr; | |
754 | port->num_mcast++; | |
755 | } | |
756 | dmi = dmi->next; | |
757 | } | |
758 | } | |
759 | ||
760 | write_unlock_irqrestore(&port->mcast_gate, flags); | |
761 | } | |
762 | ||
763 | static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | |
764 | { | |
765 | strncpy(info->driver, "veth", sizeof(info->driver) - 1); | |
766 | info->driver[sizeof(info->driver) - 1] = '\0'; | |
767 | strncpy(info->version, "1.0", sizeof(info->version) - 1); | |
768 | } | |
769 | ||
770 | static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd) | |
771 | { | |
772 | ecmd->supported = (SUPPORTED_1000baseT_Full | |
773 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); | |
774 | ecmd->advertising = (SUPPORTED_1000baseT_Full | |
775 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); | |
776 | ecmd->port = PORT_FIBRE; | |
777 | ecmd->transceiver = XCVR_INTERNAL; | |
778 | ecmd->phy_address = 0; | |
779 | ecmd->speed = SPEED_1000; | |
780 | ecmd->duplex = DUPLEX_FULL; | |
781 | ecmd->autoneg = AUTONEG_ENABLE; | |
782 | ecmd->maxtxpkt = 120; | |
783 | ecmd->maxrxpkt = 120; | |
784 | return 0; | |
785 | } | |
786 | ||
787 | static u32 veth_get_link(struct net_device *dev) | |
788 | { | |
789 | return 1; | |
790 | } | |
791 | ||
792 | static struct ethtool_ops ops = { | |
793 | .get_drvinfo = veth_get_drvinfo, | |
794 | .get_settings = veth_get_settings, | |
795 | .get_link = veth_get_link, | |
796 | }; | |
797 | ||
798 | static void veth_tx_timeout(struct net_device *dev) | |
799 | { | |
800 | struct veth_port *port = (struct veth_port *)dev->priv; | |
801 | struct net_device_stats *stats = &port->stats; | |
802 | unsigned long flags; | |
803 | int i; | |
804 | ||
805 | stats->tx_errors++; | |
806 | ||
807 | spin_lock_irqsave(&port->pending_gate, flags); | |
808 | ||
243cd55e ME |
809 | if (!port->pending_lpmask) { |
810 | spin_unlock_irqrestore(&port->pending_gate, flags); | |
811 | return; | |
812 | } | |
813 | ||
1da177e4 LT |
814 | printk(KERN_WARNING "%s: Tx timeout! Resetting lp connections: %08x\n", |
815 | dev->name, port->pending_lpmask); | |
816 | ||
1da177e4 LT |
817 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { |
818 | struct veth_lpar_connection *cnx = veth_cnx[i]; | |
819 | ||
820 | if (! (port->pending_lpmask & (1<<i))) | |
821 | continue; | |
822 | ||
823 | /* If we're pending on it, we must be connected to it, | |
824 | * so we should certainly have a structure for it. */ | |
825 | BUG_ON(! cnx); | |
826 | ||
827 | /* Theoretically we could be kicking a connection | |
828 | * which doesn't deserve it, but in practice if we've | |
829 | * had a Tx timeout, the pending_lpmask will have | |
830 | * exactly one bit set - the connection causing the | |
831 | * problem. */ | |
832 | spin_lock(&cnx->lock); | |
833 | cnx->state |= VETH_STATE_RESET; | |
834 | veth_kick_statemachine(cnx); | |
835 | spin_unlock(&cnx->lock); | |
836 | } | |
837 | ||
838 | spin_unlock_irqrestore(&port->pending_gate, flags); | |
839 | } | |
840 | ||
841 | static struct net_device * __init veth_probe_one(int vlan, struct device *vdev) | |
842 | { | |
843 | struct net_device *dev; | |
844 | struct veth_port *port; | |
845 | int i, rc; | |
846 | ||
847 | dev = alloc_etherdev(sizeof (struct veth_port)); | |
848 | if (! dev) { | |
849 | veth_error("Unable to allocate net_device structure!\n"); | |
850 | return NULL; | |
851 | } | |
852 | ||
853 | port = (struct veth_port *) dev->priv; | |
854 | ||
855 | spin_lock_init(&port->pending_gate); | |
856 | rwlock_init(&port->mcast_gate); | |
857 | ||
858 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
859 | HvLpVirtualLanIndexMap map; | |
860 | ||
861 | if (i == this_lp) | |
862 | continue; | |
863 | map = HvLpConfig_getVirtualLanIndexMapForLp(i); | |
864 | if (map & (0x8000 >> vlan)) | |
865 | port->lpar_map |= (1 << i); | |
866 | } | |
867 | port->dev = vdev; | |
868 | ||
869 | dev->dev_addr[0] = 0x02; | |
870 | dev->dev_addr[1] = 0x01; | |
871 | dev->dev_addr[2] = 0xff; | |
872 | dev->dev_addr[3] = vlan; | |
873 | dev->dev_addr[4] = 0xff; | |
874 | dev->dev_addr[5] = this_lp; | |
875 | ||
876 | dev->mtu = VETH_MAX_MTU; | |
877 | ||
878 | memcpy(&port->mac_addr, dev->dev_addr, 6); | |
879 | ||
880 | dev->open = veth_open; | |
881 | dev->hard_start_xmit = veth_start_xmit; | |
882 | dev->stop = veth_close; | |
883 | dev->get_stats = veth_get_stats; | |
884 | dev->change_mtu = veth_change_mtu; | |
885 | dev->set_mac_address = NULL; | |
886 | dev->set_multicast_list = veth_set_multicast_list; | |
887 | SET_ETHTOOL_OPS(dev, &ops); | |
888 | ||
889 | dev->watchdog_timeo = 2 * (VETH_ACKTIMEOUT * HZ / 1000000); | |
890 | dev->tx_timeout = veth_tx_timeout; | |
891 | ||
892 | SET_NETDEV_DEV(dev, vdev); | |
893 | ||
894 | rc = register_netdev(dev); | |
895 | if (rc != 0) { | |
61a3c696 | 896 | veth_error("Failed registering net device for vlan%d.\n", vlan); |
1da177e4 LT |
897 | free_netdev(dev); |
898 | return NULL; | |
899 | } | |
900 | ||
61a3c696 ME |
901 | veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n", |
902 | dev->name, vlan, port->lpar_map); | |
1da177e4 LT |
903 | |
904 | return dev; | |
905 | } | |
906 | ||
907 | /* | |
908 | * Tx path | |
909 | */ | |
910 | ||
911 | static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp, | |
912 | struct net_device *dev) | |
913 | { | |
914 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; | |
915 | struct veth_port *port = (struct veth_port *) dev->priv; | |
916 | HvLpEvent_Rc rc; | |
917 | u32 dma_address, dma_length; | |
918 | struct veth_msg *msg = NULL; | |
919 | int err = 0; | |
920 | unsigned long flags; | |
921 | ||
922 | if (! cnx) { | |
923 | port->stats.tx_errors++; | |
924 | dev_kfree_skb(skb); | |
925 | return 0; | |
926 | } | |
927 | ||
928 | spin_lock_irqsave(&cnx->lock, flags); | |
929 | ||
f27eff1f | 930 | if (! (cnx->state & VETH_STATE_READY)) |
1da177e4 LT |
931 | goto drop; |
932 | ||
933 | if ((skb->len - 14) > VETH_MAX_MTU) | |
934 | goto drop; | |
935 | ||
936 | msg = veth_stack_pop(cnx); | |
937 | ||
938 | if (! msg) { | |
939 | err = 1; | |
940 | goto drop; | |
941 | } | |
942 | ||
943 | dma_length = skb->len; | |
944 | dma_address = dma_map_single(port->dev, skb->data, | |
945 | dma_length, DMA_TO_DEVICE); | |
946 | ||
947 | if (dma_mapping_error(dma_address)) | |
948 | goto recycle_and_drop; | |
949 | ||
950 | /* Is it really necessary to check the length and address | |
951 | * fields of the first entry here? */ | |
952 | msg->skb = skb; | |
953 | msg->dev = port->dev; | |
954 | msg->data.addr[0] = dma_address; | |
955 | msg->data.len[0] = dma_length; | |
956 | msg->data.eofmask = 1 << VETH_EOF_SHIFT; | |
957 | set_bit(0, &(msg->in_use)); | |
958 | rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data); | |
959 | ||
960 | if (rc != HvLpEvent_Rc_Good) | |
961 | goto recycle_and_drop; | |
962 | ||
963 | spin_unlock_irqrestore(&cnx->lock, flags); | |
964 | return 0; | |
965 | ||
966 | recycle_and_drop: | |
967 | msg->skb = NULL; | |
968 | /* need to set in use to make veth_recycle_msg in case this | |
969 | * was a mapping failure */ | |
970 | set_bit(0, &msg->in_use); | |
971 | veth_recycle_msg(cnx, msg); | |
972 | drop: | |
973 | port->stats.tx_errors++; | |
974 | dev_kfree_skb(skb); | |
975 | spin_unlock_irqrestore(&cnx->lock, flags); | |
976 | return err; | |
977 | } | |
978 | ||
979 | static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb, | |
980 | HvLpIndexMap lpmask, | |
981 | struct net_device *dev) | |
982 | { | |
983 | struct veth_port *port = (struct veth_port *) dev->priv; | |
984 | int i; | |
985 | int rc; | |
986 | ||
987 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { | |
988 | if ((lpmask & (1 << i)) == 0) | |
989 | continue; | |
990 | ||
991 | rc = veth_transmit_to_one(skb_get(skb), i, dev); | |
992 | if (! rc) | |
993 | lpmask &= ~(1<<i); | |
994 | } | |
995 | ||
996 | if (! lpmask) { | |
997 | port->stats.tx_packets++; | |
998 | port->stats.tx_bytes += skb->len; | |
999 | } | |
1000 | ||
1001 | return lpmask; | |
1002 | } | |
1003 | ||
1004 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
1005 | { | |
1006 | unsigned char *frame = skb->data; | |
1007 | struct veth_port *port = (struct veth_port *) dev->priv; | |
1008 | unsigned long flags; | |
1009 | HvLpIndexMap lpmask; | |
1010 | ||
1011 | if (! (frame[0] & 0x01)) { | |
1012 | /* unicast packet */ | |
1013 | HvLpIndex rlp = frame[5]; | |
1014 | ||
1015 | if ( ! ((1 << rlp) & port->lpar_map) ) { | |
1016 | dev_kfree_skb(skb); | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | lpmask = 1 << rlp; | |
1021 | } else { | |
1022 | lpmask = port->lpar_map; | |
1023 | } | |
1024 | ||
1025 | spin_lock_irqsave(&port->pending_gate, flags); | |
1026 | ||
1027 | lpmask = veth_transmit_to_many(skb, lpmask, dev); | |
1028 | ||
eb235aef ME |
1029 | dev->trans_start = jiffies; |
1030 | ||
1da177e4 LT |
1031 | if (! lpmask) { |
1032 | dev_kfree_skb(skb); | |
1033 | } else { | |
1034 | if (port->pending_skb) { | |
61a3c696 | 1035 | veth_error("%s: TX while skb was pending!\n", |
1da177e4 LT |
1036 | dev->name); |
1037 | dev_kfree_skb(skb); | |
1038 | spin_unlock_irqrestore(&port->pending_gate, flags); | |
1039 | return 1; | |
1040 | } | |
1041 | ||
1042 | port->pending_skb = skb; | |
1043 | port->pending_lpmask = lpmask; | |
1044 | netif_stop_queue(dev); | |
1045 | } | |
1046 | ||
1047 | spin_unlock_irqrestore(&port->pending_gate, flags); | |
1048 | ||
1049 | return 0; | |
1050 | } | |
1051 | ||
1052 | static void veth_recycle_msg(struct veth_lpar_connection *cnx, | |
1053 | struct veth_msg *msg) | |
1054 | { | |
1055 | u32 dma_address, dma_length; | |
1056 | ||
1057 | if (test_and_clear_bit(0, &msg->in_use)) { | |
1058 | dma_address = msg->data.addr[0]; | |
1059 | dma_length = msg->data.len[0]; | |
1060 | ||
1061 | dma_unmap_single(msg->dev, dma_address, dma_length, | |
1062 | DMA_TO_DEVICE); | |
1063 | ||
1064 | if (msg->skb) { | |
1065 | dev_kfree_skb_any(msg->skb); | |
1066 | msg->skb = NULL; | |
1067 | } | |
1068 | ||
1069 | memset(&msg->data, 0, sizeof(msg->data)); | |
1070 | veth_stack_push(cnx, msg); | |
61a3c696 ME |
1071 | } else if (cnx->state & VETH_STATE_OPEN) { |
1072 | veth_error("Non-pending frame (# %d) acked by LPAR %d.\n", | |
1073 | cnx->remote_lp, msg->token); | |
1074 | } | |
1da177e4 LT |
1075 | } |
1076 | ||
1077 | static void veth_flush_pending(struct veth_lpar_connection *cnx) | |
1078 | { | |
1079 | int i; | |
1080 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) { | |
1081 | struct net_device *dev = veth_dev[i]; | |
1082 | struct veth_port *port; | |
1083 | unsigned long flags; | |
1084 | ||
1085 | if (! dev) | |
1086 | continue; | |
1087 | ||
1088 | port = (struct veth_port *)dev->priv; | |
1089 | ||
1090 | if (! (port->lpar_map & (1<<cnx->remote_lp))) | |
1091 | continue; | |
1092 | ||
1093 | spin_lock_irqsave(&port->pending_gate, flags); | |
1094 | if (port->pending_skb) { | |
1095 | port->pending_lpmask = | |
1096 | veth_transmit_to_many(port->pending_skb, | |
1097 | port->pending_lpmask, | |
1098 | dev); | |
1099 | if (! port->pending_lpmask) { | |
1100 | dev_kfree_skb_any(port->pending_skb); | |
1101 | port->pending_skb = NULL; | |
1102 | netif_wake_queue(dev); | |
1103 | } | |
1104 | } | |
1105 | spin_unlock_irqrestore(&port->pending_gate, flags); | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Rx path | |
1111 | */ | |
1112 | ||
1113 | static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr) | |
1114 | { | |
1115 | int wanted = 0; | |
1116 | int i; | |
1117 | unsigned long flags; | |
1118 | ||
1119 | if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) ) | |
1120 | return 1; | |
1121 | ||
1122 | if (! (((char *) &mac_addr)[0] & 0x01)) | |
1123 | return 0; | |
1124 | ||
1125 | read_lock_irqsave(&port->mcast_gate, flags); | |
1126 | ||
1127 | if (port->promiscuous || port->all_mcast) { | |
1128 | wanted = 1; | |
1129 | goto out; | |
1130 | } | |
1131 | ||
1132 | for (i = 0; i < port->num_mcast; ++i) { | |
1133 | if (port->mcast_addr[i] == mac_addr) { | |
1134 | wanted = 1; | |
1135 | break; | |
1136 | } | |
1137 | } | |
1138 | ||
1139 | out: | |
1140 | read_unlock_irqrestore(&port->mcast_gate, flags); | |
1141 | ||
1142 | return wanted; | |
1143 | } | |
1144 | ||
1145 | struct dma_chunk { | |
1146 | u64 addr; | |
1147 | u64 size; | |
1148 | }; | |
1149 | ||
1150 | #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 ) | |
1151 | ||
1152 | static inline void veth_build_dma_list(struct dma_chunk *list, | |
1153 | unsigned char *p, unsigned long length) | |
1154 | { | |
1155 | unsigned long done; | |
1156 | int i = 1; | |
1157 | ||
1158 | /* FIXME: skbs are continguous in real addresses. Do we | |
1159 | * really need to break it into PAGE_SIZE chunks, or can we do | |
1160 | * it just at the granularity of iSeries real->absolute | |
1161 | * mapping? Indeed, given the way the allocator works, can we | |
1162 | * count on them being absolutely contiguous? */ | |
1163 | list[0].addr = ISERIES_HV_ADDR(p); | |
1164 | list[0].size = min(length, | |
1165 | PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK)); | |
1166 | ||
1167 | done = list[0].size; | |
1168 | while (done < length) { | |
1169 | list[i].addr = ISERIES_HV_ADDR(p + done); | |
1170 | list[i].size = min(length-done, PAGE_SIZE); | |
1171 | done += list[i].size; | |
1172 | i++; | |
1173 | } | |
1174 | } | |
1175 | ||
1176 | static void veth_flush_acks(struct veth_lpar_connection *cnx) | |
1177 | { | |
1178 | HvLpEvent_Rc rc; | |
1179 | ||
1180 | rc = veth_signaldata(cnx, VethEventTypeFramesAck, | |
1181 | 0, &cnx->pending_acks); | |
1182 | ||
1183 | if (rc != HvLpEvent_Rc_Good) | |
61a3c696 ME |
1184 | veth_error("Failed acking frames from LPAR %d, rc = %d\n", |
1185 | cnx->remote_lp, (int)rc); | |
1da177e4 LT |
1186 | |
1187 | cnx->num_pending_acks = 0; | |
1188 | memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks)); | |
1189 | } | |
1190 | ||
1191 | static void veth_receive(struct veth_lpar_connection *cnx, | |
1192 | struct VethLpEvent *event) | |
1193 | { | |
1194 | struct VethFramesData *senddata = &event->u.frames_data; | |
1195 | int startchunk = 0; | |
1196 | int nchunks; | |
1197 | unsigned long flags; | |
1198 | HvLpDma_Rc rc; | |
1199 | ||
1200 | do { | |
1201 | u16 length = 0; | |
1202 | struct sk_buff *skb; | |
1203 | struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME]; | |
1204 | struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG]; | |
1205 | u64 dest; | |
1206 | HvLpVirtualLanIndex vlan; | |
1207 | struct net_device *dev; | |
1208 | struct veth_port *port; | |
1209 | ||
1210 | /* FIXME: do we need this? */ | |
1211 | memset(local_list, 0, sizeof(local_list)); | |
1212 | memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG)); | |
1213 | ||
1214 | /* a 0 address marks the end of the valid entries */ | |
1215 | if (senddata->addr[startchunk] == 0) | |
1216 | break; | |
1217 | ||
1218 | /* make sure that we have at least 1 EOF entry in the | |
1219 | * remaining entries */ | |
1220 | if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) { | |
61a3c696 ME |
1221 | veth_error("Missing EOF fragment in event " |
1222 | "eofmask = 0x%x startchunk = %d\n", | |
1223 | (unsigned)senddata->eofmask, | |
1224 | startchunk); | |
1da177e4 LT |
1225 | break; |
1226 | } | |
1227 | ||
1228 | /* build list of chunks in this frame */ | |
1229 | nchunks = 0; | |
1230 | do { | |
1231 | remote_list[nchunks].addr = | |
1232 | (u64) senddata->addr[startchunk+nchunks] << 32; | |
1233 | remote_list[nchunks].size = | |
1234 | senddata->len[startchunk+nchunks]; | |
1235 | length += remote_list[nchunks].size; | |
1236 | } while (! (senddata->eofmask & | |
1237 | (1 << (VETH_EOF_SHIFT + startchunk + nchunks++)))); | |
1238 | ||
1239 | /* length == total length of all chunks */ | |
1240 | /* nchunks == # of chunks in this frame */ | |
1241 | ||
1242 | if ((length - ETH_HLEN) > VETH_MAX_MTU) { | |
61a3c696 ME |
1243 | veth_error("Received oversize frame from LPAR %d " |
1244 | "(length = %d)\n", | |
1245 | cnx->remote_lp, length); | |
1da177e4 LT |
1246 | continue; |
1247 | } | |
1248 | ||
1249 | skb = alloc_skb(length, GFP_ATOMIC); | |
1250 | if (!skb) | |
1251 | continue; | |
1252 | ||
1253 | veth_build_dma_list(local_list, skb->data, length); | |
1254 | ||
1255 | rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan, | |
1256 | event->base_event.xSourceLp, | |
1257 | HvLpDma_Direction_RemoteToLocal, | |
1258 | cnx->src_inst, | |
1259 | cnx->dst_inst, | |
1260 | HvLpDma_AddressType_RealAddress, | |
1261 | HvLpDma_AddressType_TceIndex, | |
1262 | ISERIES_HV_ADDR(&local_list), | |
1263 | ISERIES_HV_ADDR(&remote_list), | |
1264 | length); | |
1265 | if (rc != HvLpDma_Rc_Good) { | |
1266 | dev_kfree_skb_irq(skb); | |
1267 | continue; | |
1268 | } | |
1269 | ||
1270 | vlan = skb->data[9]; | |
1271 | dev = veth_dev[vlan]; | |
41664c03 ME |
1272 | if (! dev) { |
1273 | /* | |
1274 | * Some earlier versions of the driver sent | |
1275 | * broadcasts down all connections, even to lpars | |
1276 | * that weren't on the relevant vlan. So ignore | |
1277 | * packets belonging to a vlan we're not on. | |
1278 | * We can also be here if we receive packets while | |
1279 | * the driver is going down, because then dev is NULL. | |
1280 | */ | |
1281 | dev_kfree_skb_irq(skb); | |
1da177e4 | 1282 | continue; |
41664c03 | 1283 | } |
1da177e4 LT |
1284 | |
1285 | port = (struct veth_port *)dev->priv; | |
1286 | dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000; | |
1287 | ||
1288 | if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) { | |
1289 | dev_kfree_skb_irq(skb); | |
1290 | continue; | |
1291 | } | |
1292 | if (! veth_frame_wanted(port, dest)) { | |
1293 | dev_kfree_skb_irq(skb); | |
1294 | continue; | |
1295 | } | |
1296 | ||
1297 | skb_put(skb, length); | |
1298 | skb->dev = dev; | |
1299 | skb->protocol = eth_type_trans(skb, dev); | |
1300 | skb->ip_summed = CHECKSUM_NONE; | |
1301 | netif_rx(skb); /* send it up */ | |
1302 | port->stats.rx_packets++; | |
1303 | port->stats.rx_bytes += length; | |
1304 | } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG); | |
1305 | ||
1306 | /* Ack it */ | |
1307 | spin_lock_irqsave(&cnx->lock, flags); | |
1308 | BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG); | |
1309 | ||
1310 | cnx->pending_acks[cnx->num_pending_acks++] = | |
1311 | event->base_event.xCorrelationToken; | |
1312 | ||
1313 | if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) | |
1314 | || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) ) | |
1315 | veth_flush_acks(cnx); | |
1316 | ||
1317 | spin_unlock_irqrestore(&cnx->lock, flags); | |
1318 | } | |
1319 | ||
1320 | static void veth_timed_ack(unsigned long ptr) | |
1321 | { | |
1322 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr; | |
1323 | unsigned long flags; | |
1324 | ||
1325 | /* Ack all the events */ | |
1326 | spin_lock_irqsave(&cnx->lock, flags); | |
1327 | if (cnx->num_pending_acks > 0) | |
1328 | veth_flush_acks(cnx); | |
1329 | ||
1330 | /* Reschedule the timer */ | |
1331 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; | |
1332 | add_timer(&cnx->ack_timer); | |
1333 | spin_unlock_irqrestore(&cnx->lock, flags); | |
1334 | } | |
1335 | ||
1336 | static int veth_remove(struct vio_dev *vdev) | |
1337 | { | |
1338 | int i = vdev->unit_address; | |
1339 | struct net_device *dev; | |
1340 | ||
1341 | dev = veth_dev[i]; | |
1342 | if (dev != NULL) { | |
1343 | veth_dev[i] = NULL; | |
1344 | unregister_netdev(dev); | |
1345 | free_netdev(dev); | |
1346 | } | |
1347 | return 0; | |
1348 | } | |
1349 | ||
1350 | static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id) | |
1351 | { | |
1352 | int i = vdev->unit_address; | |
1353 | struct net_device *dev; | |
1354 | ||
1355 | dev = veth_probe_one(i, &vdev->dev); | |
1356 | if (dev == NULL) { | |
1357 | veth_remove(vdev); | |
1358 | return 1; | |
1359 | } | |
1360 | veth_dev[i] = dev; | |
1361 | ||
1362 | /* Start the state machine on each connection, to commence | |
1363 | * link negotiation */ | |
1364 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) | |
1365 | if (veth_cnx[i]) | |
1366 | veth_kick_statemachine(veth_cnx[i]); | |
1367 | ||
1368 | return 0; | |
1369 | } | |
1370 | ||
1371 | /** | |
1372 | * veth_device_table: Used by vio.c to match devices that we | |
1373 | * support. | |
1374 | */ | |
1375 | static struct vio_device_id veth_device_table[] __devinitdata = { | |
1376 | { "vlan", "" }, | |
fb120da6 | 1377 | { "", "" } |
1da177e4 LT |
1378 | }; |
1379 | MODULE_DEVICE_TABLE(vio, veth_device_table); | |
1380 | ||
1381 | static struct vio_driver veth_driver = { | |
1382 | .name = "iseries_veth", | |
1383 | .id_table = veth_device_table, | |
1384 | .probe = veth_probe, | |
1385 | .remove = veth_remove | |
1386 | }; | |
1387 | ||
1388 | /* | |
1389 | * Module initialization/cleanup | |
1390 | */ | |
1391 | ||
1392 | void __exit veth_module_cleanup(void) | |
1393 | { | |
1394 | int i; | |
1395 | ||
b2e0852e ME |
1396 | /* Stop the queues first to stop any new packets being sent. */ |
1397 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) | |
1398 | if (veth_dev[i]) | |
1399 | netif_stop_queue(veth_dev[i]); | |
1da177e4 | 1400 | |
b2e0852e ME |
1401 | /* Stop the connections before we unregister the driver. This |
1402 | * ensures there's no skbs lying around holding the device open. */ | |
1da177e4 LT |
1403 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) |
1404 | veth_stop_connection(i); | |
1405 | ||
1406 | HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan); | |
1407 | ||
1408 | /* Hypervisor callbacks may have scheduled more work while we | |
b2e0852e | 1409 | * were stoping connections. Now that we've disconnected from |
1da177e4 LT |
1410 | * the hypervisor make sure everything's finished. */ |
1411 | flush_scheduled_work(); | |
1412 | ||
b2e0852e ME |
1413 | vio_unregister_driver(&veth_driver); |
1414 | ||
1da177e4 LT |
1415 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) |
1416 | veth_destroy_connection(i); | |
1417 | ||
1418 | } | |
1419 | module_exit(veth_module_cleanup); | |
1420 | ||
1421 | int __init veth_module_init(void) | |
1422 | { | |
1423 | int i; | |
1424 | int rc; | |
1425 | ||
1426 | this_lp = HvLpConfig_getLpIndex_outline(); | |
1427 | ||
1428 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) { | |
1429 | rc = veth_init_connection(i); | |
1430 | if (rc != 0) { | |
1431 | veth_module_cleanup(); | |
1432 | return rc; | |
1433 | } | |
1434 | } | |
1435 | ||
1436 | HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan, | |
1437 | &veth_handle_event); | |
1438 | ||
1439 | return vio_register_driver(&veth_driver); | |
1440 | } | |
1441 | module_init(veth_module_init); |