xref: /freebsd/sys/ofed/drivers/infiniband/core/ib_mad_rmpp.c (revision 687057e712edbbc31e33613d99d9f8cc436cfa1f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3  *
4  * Copyright (c) 2005 Intel Inc. All rights reserved.
5  * Copyright (c) 2005-2006 Voltaire, Inc. All rights reserved.
6  * Copyright (c) 2014 Intel Corporation.  All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 #include <sys/cdefs.h>
38 #include <linux/slab.h>
39 
40 #include "mad_priv.h"
41 #include "mad_rmpp.h"
42 
43 enum rmpp_state {
44 	RMPP_STATE_ACTIVE,
45 	RMPP_STATE_TIMEOUT,
46 	RMPP_STATE_COMPLETE,
47 	RMPP_STATE_CANCELING
48 };
49 
50 struct mad_rmpp_recv {
51 	struct ib_mad_agent_private *agent;
52 	struct list_head list;
53 	struct delayed_work timeout_work;
54 	struct delayed_work cleanup_work;
55 	struct completion comp;
56 	enum rmpp_state state;
57 	spinlock_t lock;
58 	atomic_t refcount;
59 
60 	struct ib_ah *ah;
61 	struct ib_mad_recv_wc *rmpp_wc;
62 	struct ib_mad_recv_buf *cur_seg_buf;
63 	int last_ack;
64 	int seg_num;
65 	int newwin;
66 	int repwin;
67 
68 	__be64 tid;
69 	u32 src_qp;
70 	u32 slid;
71 	u8 mgmt_class;
72 	u8 class_version;
73 	u8 method;
74 	u8 base_version;
75 };
76 
77 static inline void deref_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
78 {
79 	if (atomic_dec_and_test(&rmpp_recv->refcount))
80 		complete(&rmpp_recv->comp);
81 }
82 
83 static void destroy_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
84 {
85 	deref_rmpp_recv(rmpp_recv);
86 	wait_for_completion(&rmpp_recv->comp);
87 	rdma_destroy_ah(rmpp_recv->ah, RDMA_DESTROY_AH_SLEEPABLE);
88 	kfree(rmpp_recv);
89 }
90 
91 void ib_cancel_rmpp_recvs(struct ib_mad_agent_private *agent)
92 {
93 	struct mad_rmpp_recv *rmpp_recv, *temp_rmpp_recv;
94 	unsigned long flags;
95 
96 	spin_lock_irqsave(&agent->lock, flags);
97 	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
98 		if (rmpp_recv->state != RMPP_STATE_COMPLETE)
99 			ib_free_recv_mad(rmpp_recv->rmpp_wc);
100 		rmpp_recv->state = RMPP_STATE_CANCELING;
101 	}
102 	spin_unlock_irqrestore(&agent->lock, flags);
103 
104 	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
105 		cancel_delayed_work_sync(&rmpp_recv->timeout_work);
106 		cancel_delayed_work_sync(&rmpp_recv->cleanup_work);
107 	}
108 
109 	flush_workqueue(agent->qp_info->port_priv->wq);
110 
111 	list_for_each_entry_safe(rmpp_recv, temp_rmpp_recv,
112 				 &agent->rmpp_list, list) {
113 		list_del(&rmpp_recv->list);
114 		destroy_rmpp_recv(rmpp_recv);
115 	}
116 }
117 
118 static void format_ack(struct ib_mad_send_buf *msg,
119 		       struct ib_rmpp_mad *data,
120 		       struct mad_rmpp_recv *rmpp_recv)
121 {
122 	struct ib_rmpp_mad *ack = msg->mad;
123 	unsigned long flags;
124 
125 	memcpy(ack, &data->mad_hdr, msg->hdr_len);
126 
127 	ack->mad_hdr.method ^= IB_MGMT_METHOD_RESP;
128 	ack->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_ACK;
129 	ib_set_rmpp_flags(&ack->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
130 
131 	spin_lock_irqsave(&rmpp_recv->lock, flags);
132 	rmpp_recv->last_ack = rmpp_recv->seg_num;
133 	ack->rmpp_hdr.seg_num = cpu_to_be32(rmpp_recv->seg_num);
134 	ack->rmpp_hdr.paylen_newwin = cpu_to_be32(rmpp_recv->newwin);
135 	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
136 }
137 
138 static void ack_recv(struct mad_rmpp_recv *rmpp_recv,
139 		     struct ib_mad_recv_wc *recv_wc)
140 {
141 	struct ib_mad_send_buf *msg;
142 	int ret, hdr_len;
143 
144 	hdr_len = ib_get_mad_data_offset(recv_wc->recv_buf.mad->mad_hdr.mgmt_class);
145 	msg = ib_create_send_mad(&rmpp_recv->agent->agent, recv_wc->wc->src_qp,
146 				 recv_wc->wc->pkey_index, 1, hdr_len,
147 				 0, GFP_KERNEL,
148 				 IB_MGMT_BASE_VERSION);
149 	if (IS_ERR(msg))
150 		return;
151 
152 	format_ack(msg, (struct ib_rmpp_mad *) recv_wc->recv_buf.mad, rmpp_recv);
153 	msg->ah = rmpp_recv->ah;
154 	ret = ib_post_send_mad(msg, NULL);
155 	if (ret)
156 		ib_free_send_mad(msg);
157 }
158 
159 static struct ib_mad_send_buf *alloc_response_msg(struct ib_mad_agent *agent,
160 						  struct ib_mad_recv_wc *recv_wc)
161 {
162 	struct ib_mad_send_buf *msg;
163 	struct ib_ah *ah;
164 	int hdr_len;
165 
166 	ah = ib_create_ah_from_wc(agent->qp->pd, recv_wc->wc,
167 				  recv_wc->recv_buf.grh, agent->port_num);
168 	if (IS_ERR(ah))
169 		return (void *) ah;
170 
171 	hdr_len = ib_get_mad_data_offset(recv_wc->recv_buf.mad->mad_hdr.mgmt_class);
172 	msg = ib_create_send_mad(agent, recv_wc->wc->src_qp,
173 				 recv_wc->wc->pkey_index, 1,
174 				 hdr_len, 0, GFP_KERNEL,
175 				 IB_MGMT_BASE_VERSION);
176 	if (IS_ERR(msg))
177 		rdma_destroy_ah(ah, RDMA_DESTROY_AH_SLEEPABLE);
178 	else {
179 		msg->ah = ah;
180 		msg->context[0] = ah;
181 	}
182 
183 	return msg;
184 }
185 
186 static void ack_ds_ack(struct ib_mad_agent_private *agent,
187 		       struct ib_mad_recv_wc *recv_wc)
188 {
189 	struct ib_mad_send_buf *msg;
190 	struct ib_rmpp_mad *rmpp_mad;
191 	int ret;
192 
193 	msg = alloc_response_msg(&agent->agent, recv_wc);
194 	if (IS_ERR(msg))
195 		return;
196 
197 	rmpp_mad = msg->mad;
198 	memcpy(rmpp_mad, recv_wc->recv_buf.mad, msg->hdr_len);
199 
200 	rmpp_mad->mad_hdr.method ^= IB_MGMT_METHOD_RESP;
201 	ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
202 	rmpp_mad->rmpp_hdr.seg_num = 0;
203 	rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(1);
204 
205 	ret = ib_post_send_mad(msg, NULL);
206 	if (ret) {
207 		rdma_destroy_ah(msg->ah, RDMA_DESTROY_AH_SLEEPABLE);
208 		ib_free_send_mad(msg);
209 	}
210 }
211 
212 void ib_rmpp_send_handler(struct ib_mad_send_wc *mad_send_wc)
213 {
214 	if (mad_send_wc->send_buf->context[0] == mad_send_wc->send_buf->ah)
215 		rdma_destroy_ah(mad_send_wc->send_buf->ah,
216 				RDMA_DESTROY_AH_SLEEPABLE);
217 	ib_free_send_mad(mad_send_wc->send_buf);
218 }
219 
220 static void nack_recv(struct ib_mad_agent_private *agent,
221 		      struct ib_mad_recv_wc *recv_wc, u8 rmpp_status)
222 {
223 	struct ib_mad_send_buf *msg;
224 	struct ib_rmpp_mad *rmpp_mad;
225 	int ret;
226 
227 	msg = alloc_response_msg(&agent->agent, recv_wc);
228 	if (IS_ERR(msg))
229 		return;
230 
231 	rmpp_mad = msg->mad;
232 	memcpy(rmpp_mad, recv_wc->recv_buf.mad, msg->hdr_len);
233 
234 	rmpp_mad->mad_hdr.method ^= IB_MGMT_METHOD_RESP;
235 	rmpp_mad->rmpp_hdr.rmpp_version = IB_MGMT_RMPP_VERSION;
236 	rmpp_mad->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_ABORT;
237 	ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
238 	rmpp_mad->rmpp_hdr.rmpp_status = rmpp_status;
239 	rmpp_mad->rmpp_hdr.seg_num = 0;
240 	rmpp_mad->rmpp_hdr.paylen_newwin = 0;
241 
242 	ret = ib_post_send_mad(msg, NULL);
243 	if (ret) {
244 		rdma_destroy_ah(msg->ah, RDMA_DESTROY_AH_SLEEPABLE);
245 		ib_free_send_mad(msg);
246 	}
247 }
248 
249 static void recv_timeout_handler(struct work_struct *work)
250 {
251 	struct mad_rmpp_recv *rmpp_recv =
252 		container_of(work, struct mad_rmpp_recv, timeout_work.work);
253 	struct ib_mad_recv_wc *rmpp_wc;
254 	unsigned long flags;
255 
256 	spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
257 	if (rmpp_recv->state != RMPP_STATE_ACTIVE) {
258 		spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
259 		return;
260 	}
261 	rmpp_recv->state = RMPP_STATE_TIMEOUT;
262 	list_del(&rmpp_recv->list);
263 	spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
264 
265 	rmpp_wc = rmpp_recv->rmpp_wc;
266 	nack_recv(rmpp_recv->agent, rmpp_wc, IB_MGMT_RMPP_STATUS_T2L);
267 	destroy_rmpp_recv(rmpp_recv);
268 	ib_free_recv_mad(rmpp_wc);
269 }
270 
271 static void recv_cleanup_handler(struct work_struct *work)
272 {
273 	struct mad_rmpp_recv *rmpp_recv =
274 		container_of(work, struct mad_rmpp_recv, cleanup_work.work);
275 	unsigned long flags;
276 
277 	spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
278 	if (rmpp_recv->state == RMPP_STATE_CANCELING) {
279 		spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
280 		return;
281 	}
282 	list_del(&rmpp_recv->list);
283 	spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
284 	destroy_rmpp_recv(rmpp_recv);
285 }
286 
287 static struct mad_rmpp_recv *
288 create_rmpp_recv(struct ib_mad_agent_private *agent,
289 		 struct ib_mad_recv_wc *mad_recv_wc)
290 {
291 	struct mad_rmpp_recv *rmpp_recv;
292 	struct ib_mad_hdr *mad_hdr;
293 
294 	rmpp_recv = kmalloc(sizeof *rmpp_recv, GFP_KERNEL);
295 	if (!rmpp_recv)
296 		return NULL;
297 
298 	rmpp_recv->ah = ib_create_ah_from_wc(agent->agent.qp->pd,
299 					     mad_recv_wc->wc,
300 					     mad_recv_wc->recv_buf.grh,
301 					     agent->agent.port_num);
302 	if (IS_ERR(rmpp_recv->ah))
303 		goto error;
304 
305 	rmpp_recv->agent = agent;
306 	init_completion(&rmpp_recv->comp);
307 	INIT_DELAYED_WORK(&rmpp_recv->timeout_work, recv_timeout_handler);
308 	INIT_DELAYED_WORK(&rmpp_recv->cleanup_work, recv_cleanup_handler);
309 	spin_lock_init(&rmpp_recv->lock);
310 	rmpp_recv->state = RMPP_STATE_ACTIVE;
311 	atomic_set(&rmpp_recv->refcount, 1);
312 
313 	rmpp_recv->rmpp_wc = mad_recv_wc;
314 	rmpp_recv->cur_seg_buf = &mad_recv_wc->recv_buf;
315 	rmpp_recv->newwin = 1;
316 	rmpp_recv->seg_num = 1;
317 	rmpp_recv->last_ack = 0;
318 	rmpp_recv->repwin = 1;
319 
320 	mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
321 	rmpp_recv->tid = mad_hdr->tid;
322 	rmpp_recv->src_qp = mad_recv_wc->wc->src_qp;
323 	rmpp_recv->slid = mad_recv_wc->wc->slid;
324 	rmpp_recv->mgmt_class = mad_hdr->mgmt_class;
325 	rmpp_recv->class_version = mad_hdr->class_version;
326 	rmpp_recv->method  = mad_hdr->method;
327 	rmpp_recv->base_version  = mad_hdr->base_version;
328 	return rmpp_recv;
329 
330 error:	kfree(rmpp_recv);
331 	return NULL;
332 }
333 
334 static struct mad_rmpp_recv *
335 find_rmpp_recv(struct ib_mad_agent_private *agent,
336 	       struct ib_mad_recv_wc *mad_recv_wc)
337 {
338 	struct mad_rmpp_recv *rmpp_recv;
339 	struct ib_mad_hdr *mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
340 
341 	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
342 		if (rmpp_recv->tid == mad_hdr->tid &&
343 		    rmpp_recv->src_qp == mad_recv_wc->wc->src_qp &&
344 		    rmpp_recv->slid == mad_recv_wc->wc->slid &&
345 		    rmpp_recv->mgmt_class == mad_hdr->mgmt_class &&
346 		    rmpp_recv->class_version == mad_hdr->class_version &&
347 		    rmpp_recv->method == mad_hdr->method)
348 			return rmpp_recv;
349 	}
350 	return NULL;
351 }
352 
353 static struct mad_rmpp_recv *
354 acquire_rmpp_recv(struct ib_mad_agent_private *agent,
355 		  struct ib_mad_recv_wc *mad_recv_wc)
356 {
357 	struct mad_rmpp_recv *rmpp_recv;
358 	unsigned long flags;
359 
360 	spin_lock_irqsave(&agent->lock, flags);
361 	rmpp_recv = find_rmpp_recv(agent, mad_recv_wc);
362 	if (rmpp_recv)
363 		atomic_inc(&rmpp_recv->refcount);
364 	spin_unlock_irqrestore(&agent->lock, flags);
365 	return rmpp_recv;
366 }
367 
368 static struct mad_rmpp_recv *
369 insert_rmpp_recv(struct ib_mad_agent_private *agent,
370 		 struct mad_rmpp_recv *rmpp_recv)
371 {
372 	struct mad_rmpp_recv *cur_rmpp_recv;
373 
374 	cur_rmpp_recv = find_rmpp_recv(agent, rmpp_recv->rmpp_wc);
375 	if (!cur_rmpp_recv)
376 		list_add_tail(&rmpp_recv->list, &agent->rmpp_list);
377 
378 	return cur_rmpp_recv;
379 }
380 
381 static inline int get_last_flag(struct ib_mad_recv_buf *seg)
382 {
383 	struct ib_rmpp_mad *rmpp_mad;
384 
385 	rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
386 	return ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_LAST;
387 }
388 
389 static inline int get_seg_num(struct ib_mad_recv_buf *seg)
390 {
391 	struct ib_rmpp_mad *rmpp_mad;
392 
393 	rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
394 	return be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
395 }
396 
397 static inline struct ib_mad_recv_buf * get_next_seg(struct list_head *rmpp_list,
398 						    struct ib_mad_recv_buf *seg)
399 {
400 	if (seg->list.next == rmpp_list)
401 		return NULL;
402 
403 	return container_of(seg->list.next, struct ib_mad_recv_buf, list);
404 }
405 
406 static inline int window_size(struct ib_mad_agent_private *agent)
407 {
408 	return max(agent->qp_info->recv_queue.max_active >> 3, 1);
409 }
410 
411 static struct ib_mad_recv_buf * find_seg_location(struct list_head *rmpp_list,
412 						  int seg_num)
413 {
414 	struct ib_mad_recv_buf *seg_buf;
415 	int cur_seg_num;
416 
417 	list_for_each_entry_reverse(seg_buf, rmpp_list, list) {
418 		cur_seg_num = get_seg_num(seg_buf);
419 		if (seg_num > cur_seg_num)
420 			return seg_buf;
421 		if (seg_num == cur_seg_num)
422 			break;
423 	}
424 	return NULL;
425 }
426 
427 static void update_seg_num(struct mad_rmpp_recv *rmpp_recv,
428 			   struct ib_mad_recv_buf *new_buf)
429 {
430 	struct list_head *rmpp_list = &rmpp_recv->rmpp_wc->rmpp_list;
431 
432 	while (new_buf && (get_seg_num(new_buf) == rmpp_recv->seg_num + 1)) {
433 		rmpp_recv->cur_seg_buf = new_buf;
434 		rmpp_recv->seg_num++;
435 		new_buf = get_next_seg(rmpp_list, new_buf);
436 	}
437 }
438 
439 static inline int get_mad_len(struct mad_rmpp_recv *rmpp_recv)
440 {
441 	struct ib_rmpp_mad *rmpp_mad;
442 	int hdr_size, data_size, pad;
443 	bool opa = rdma_cap_opa_mad(rmpp_recv->agent->qp_info->port_priv->device,
444 				    rmpp_recv->agent->qp_info->port_priv->port_num);
445 
446 	rmpp_mad = (struct ib_rmpp_mad *)rmpp_recv->cur_seg_buf->mad;
447 
448 	hdr_size = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
449 	if (opa && rmpp_recv->base_version == OPA_MGMT_BASE_VERSION) {
450 		data_size = sizeof(struct opa_rmpp_mad) - hdr_size;
451 		pad = OPA_MGMT_RMPP_DATA - be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
452 		if (pad > OPA_MGMT_RMPP_DATA || pad < 0)
453 			pad = 0;
454 	} else {
455 		data_size = sizeof(struct ib_rmpp_mad) - hdr_size;
456 		pad = IB_MGMT_RMPP_DATA - be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
457 		if (pad > IB_MGMT_RMPP_DATA || pad < 0)
458 			pad = 0;
459 	}
460 
461 	return hdr_size + rmpp_recv->seg_num * data_size - pad;
462 }
463 
464 static struct ib_mad_recv_wc * complete_rmpp(struct mad_rmpp_recv *rmpp_recv)
465 {
466 	struct ib_mad_recv_wc *rmpp_wc;
467 
468 	ack_recv(rmpp_recv, rmpp_recv->rmpp_wc);
469 	if (rmpp_recv->seg_num > 1)
470 		cancel_delayed_work(&rmpp_recv->timeout_work);
471 
472 	rmpp_wc = rmpp_recv->rmpp_wc;
473 	rmpp_wc->mad_len = get_mad_len(rmpp_recv);
474 	/* 10 seconds until we can find the packet lifetime */
475 	queue_delayed_work(rmpp_recv->agent->qp_info->port_priv->wq,
476 			   &rmpp_recv->cleanup_work, msecs_to_jiffies(10000));
477 	return rmpp_wc;
478 }
479 
480 static struct ib_mad_recv_wc *
481 continue_rmpp(struct ib_mad_agent_private *agent,
482 	      struct ib_mad_recv_wc *mad_recv_wc)
483 {
484 	struct mad_rmpp_recv *rmpp_recv;
485 	struct ib_mad_recv_buf *prev_buf;
486 	struct ib_mad_recv_wc *done_wc;
487 	int seg_num;
488 	unsigned long flags;
489 
490 	rmpp_recv = acquire_rmpp_recv(agent, mad_recv_wc);
491 	if (!rmpp_recv)
492 		goto drop1;
493 
494 	seg_num = get_seg_num(&mad_recv_wc->recv_buf);
495 
496 	spin_lock_irqsave(&rmpp_recv->lock, flags);
497 	if ((rmpp_recv->state == RMPP_STATE_TIMEOUT) ||
498 	    (seg_num > rmpp_recv->newwin))
499 		goto drop3;
500 
501 	if ((seg_num <= rmpp_recv->last_ack) ||
502 	    (rmpp_recv->state == RMPP_STATE_COMPLETE)) {
503 		spin_unlock_irqrestore(&rmpp_recv->lock, flags);
504 		ack_recv(rmpp_recv, mad_recv_wc);
505 		goto drop2;
506 	}
507 
508 	prev_buf = find_seg_location(&rmpp_recv->rmpp_wc->rmpp_list, seg_num);
509 	if (!prev_buf)
510 		goto drop3;
511 
512 	done_wc = NULL;
513 	list_add(&mad_recv_wc->recv_buf.list, &prev_buf->list);
514 	if (rmpp_recv->cur_seg_buf == prev_buf) {
515 		update_seg_num(rmpp_recv, &mad_recv_wc->recv_buf);
516 		if (get_last_flag(rmpp_recv->cur_seg_buf)) {
517 			rmpp_recv->state = RMPP_STATE_COMPLETE;
518 			spin_unlock_irqrestore(&rmpp_recv->lock, flags);
519 			done_wc = complete_rmpp(rmpp_recv);
520 			goto out;
521 		} else if (rmpp_recv->seg_num == rmpp_recv->newwin) {
522 			rmpp_recv->newwin += window_size(agent);
523 			spin_unlock_irqrestore(&rmpp_recv->lock, flags);
524 			ack_recv(rmpp_recv, mad_recv_wc);
525 			goto out;
526 		}
527 	}
528 	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
529 out:
530 	deref_rmpp_recv(rmpp_recv);
531 	return done_wc;
532 
533 drop3:	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
534 drop2:	deref_rmpp_recv(rmpp_recv);
535 drop1:	ib_free_recv_mad(mad_recv_wc);
536 	return NULL;
537 }
538 
539 static struct ib_mad_recv_wc *
540 start_rmpp(struct ib_mad_agent_private *agent,
541 	   struct ib_mad_recv_wc *mad_recv_wc)
542 {
543 	struct mad_rmpp_recv *rmpp_recv;
544 	unsigned long flags;
545 
546 	rmpp_recv = create_rmpp_recv(agent, mad_recv_wc);
547 	if (!rmpp_recv) {
548 		ib_free_recv_mad(mad_recv_wc);
549 		return NULL;
550 	}
551 
552 	spin_lock_irqsave(&agent->lock, flags);
553 	if (insert_rmpp_recv(agent, rmpp_recv)) {
554 		spin_unlock_irqrestore(&agent->lock, flags);
555 		/* duplicate first MAD */
556 		destroy_rmpp_recv(rmpp_recv);
557 		return continue_rmpp(agent, mad_recv_wc);
558 	}
559 	atomic_inc(&rmpp_recv->refcount);
560 
561 	if (get_last_flag(&mad_recv_wc->recv_buf)) {
562 		rmpp_recv->state = RMPP_STATE_COMPLETE;
563 		spin_unlock_irqrestore(&agent->lock, flags);
564 		complete_rmpp(rmpp_recv);
565 	} else {
566 		spin_unlock_irqrestore(&agent->lock, flags);
567 		/* 40 seconds until we can find the packet lifetimes */
568 		queue_delayed_work(agent->qp_info->port_priv->wq,
569 				   &rmpp_recv->timeout_work,
570 				   msecs_to_jiffies(40000));
571 		rmpp_recv->newwin += window_size(agent);
572 		ack_recv(rmpp_recv, mad_recv_wc);
573 		mad_recv_wc = NULL;
574 	}
575 	deref_rmpp_recv(rmpp_recv);
576 	return mad_recv_wc;
577 }
578 
579 static int send_next_seg(struct ib_mad_send_wr_private *mad_send_wr)
580 {
581 	struct ib_rmpp_mad *rmpp_mad;
582 	int timeout;
583 	u32 paylen = 0;
584 
585 	rmpp_mad = mad_send_wr->send_buf.mad;
586 	ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
587 	rmpp_mad->rmpp_hdr.seg_num = cpu_to_be32(++mad_send_wr->seg_num);
588 
589 	if (mad_send_wr->seg_num == 1) {
590 		rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_FIRST;
591 		paylen = (mad_send_wr->send_buf.seg_count *
592 			  mad_send_wr->send_buf.seg_rmpp_size) -
593 			  mad_send_wr->pad;
594 	}
595 
596 	if (mad_send_wr->seg_num == mad_send_wr->send_buf.seg_count) {
597 		rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_LAST;
598 		paylen = mad_send_wr->send_buf.seg_rmpp_size - mad_send_wr->pad;
599 	}
600 	rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(paylen);
601 
602 	/* 2 seconds for an ACK until we can find the packet lifetime */
603 	timeout = mad_send_wr->send_buf.timeout_ms;
604 	if (!timeout || timeout > 2000)
605 		mad_send_wr->timeout = msecs_to_jiffies(2000);
606 
607 	return ib_send_mad(mad_send_wr);
608 }
609 
610 static void abort_send(struct ib_mad_agent_private *agent,
611 		       struct ib_mad_recv_wc *mad_recv_wc, u8 rmpp_status)
612 {
613 	struct ib_mad_send_wr_private *mad_send_wr;
614 	struct ib_mad_send_wc wc;
615 	unsigned long flags;
616 
617 	spin_lock_irqsave(&agent->lock, flags);
618 	mad_send_wr = ib_find_send_mad(agent, mad_recv_wc);
619 	if (!mad_send_wr)
620 		goto out;	/* Unmatched send */
621 
622 	if ((mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) ||
623 	    (!mad_send_wr->timeout) || (mad_send_wr->status != IB_WC_SUCCESS))
624 		goto out;	/* Send is already done */
625 
626 	ib_mark_mad_done(mad_send_wr);
627 	spin_unlock_irqrestore(&agent->lock, flags);
628 
629 	wc.status = IB_WC_REM_ABORT_ERR;
630 	wc.vendor_err = rmpp_status;
631 	wc.send_buf = &mad_send_wr->send_buf;
632 	ib_mad_complete_send_wr(mad_send_wr, &wc);
633 	return;
634 out:
635 	spin_unlock_irqrestore(&agent->lock, flags);
636 }
637 
638 static inline void adjust_last_ack(struct ib_mad_send_wr_private *wr,
639 				   int seg_num)
640 {
641 	struct list_head *list;
642 
643 	wr->last_ack = seg_num;
644 	list = &wr->last_ack_seg->list;
645 	list_for_each_entry(wr->last_ack_seg, list, list)
646 		if (wr->last_ack_seg->num == seg_num)
647 			break;
648 }
649 
650 static void process_ds_ack(struct ib_mad_agent_private *agent,
651 			   struct ib_mad_recv_wc *mad_recv_wc, int newwin)
652 {
653 	struct mad_rmpp_recv *rmpp_recv;
654 
655 	rmpp_recv = find_rmpp_recv(agent, mad_recv_wc);
656 	if (rmpp_recv && rmpp_recv->state == RMPP_STATE_COMPLETE)
657 		rmpp_recv->repwin = newwin;
658 }
659 
660 static void process_rmpp_ack(struct ib_mad_agent_private *agent,
661 			     struct ib_mad_recv_wc *mad_recv_wc)
662 {
663 	struct ib_mad_send_wr_private *mad_send_wr;
664 	struct ib_rmpp_mad *rmpp_mad;
665 	unsigned long flags;
666 	int seg_num, newwin, ret;
667 
668 	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
669 	if (rmpp_mad->rmpp_hdr.rmpp_status) {
670 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
671 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
672 		return;
673 	}
674 
675 	seg_num = be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
676 	newwin = be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
677 	if (newwin < seg_num) {
678 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_W2S);
679 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_W2S);
680 		return;
681 	}
682 
683 	spin_lock_irqsave(&agent->lock, flags);
684 	mad_send_wr = ib_find_send_mad(agent, mad_recv_wc);
685 	if (!mad_send_wr) {
686 		if (!seg_num)
687 			process_ds_ack(agent, mad_recv_wc, newwin);
688 		goto out;	/* Unmatched or DS RMPP ACK */
689 	}
690 
691 	if ((mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) &&
692 	    (mad_send_wr->timeout)) {
693 		spin_unlock_irqrestore(&agent->lock, flags);
694 		ack_ds_ack(agent, mad_recv_wc);
695 		return;		/* Repeated ACK for DS RMPP transaction */
696 	}
697 
698 	if ((mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) ||
699 	    (!mad_send_wr->timeout) || (mad_send_wr->status != IB_WC_SUCCESS))
700 		goto out;	/* Send is already done */
701 
702 	if (seg_num > mad_send_wr->send_buf.seg_count ||
703 	    seg_num > mad_send_wr->newwin) {
704 		spin_unlock_irqrestore(&agent->lock, flags);
705 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_S2B);
706 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_S2B);
707 		return;
708 	}
709 
710 	if (newwin < mad_send_wr->newwin || seg_num < mad_send_wr->last_ack)
711 		goto out;	/* Old ACK */
712 
713 	if (seg_num > mad_send_wr->last_ack) {
714 		adjust_last_ack(mad_send_wr, seg_num);
715 		mad_send_wr->retries_left = mad_send_wr->max_retries;
716 	}
717 	mad_send_wr->newwin = newwin;
718 	if (mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) {
719 		/* If no response is expected, the ACK completes the send */
720 		if (!mad_send_wr->send_buf.timeout_ms) {
721 			struct ib_mad_send_wc wc;
722 
723 			ib_mark_mad_done(mad_send_wr);
724 			spin_unlock_irqrestore(&agent->lock, flags);
725 
726 			wc.status = IB_WC_SUCCESS;
727 			wc.vendor_err = 0;
728 			wc.send_buf = &mad_send_wr->send_buf;
729 			ib_mad_complete_send_wr(mad_send_wr, &wc);
730 			return;
731 		}
732 		if (mad_send_wr->refcount == 1)
733 			ib_reset_mad_timeout(mad_send_wr,
734 					     mad_send_wr->send_buf.timeout_ms);
735 		spin_unlock_irqrestore(&agent->lock, flags);
736 		ack_ds_ack(agent, mad_recv_wc);
737 		return;
738 	} else if (mad_send_wr->refcount == 1 &&
739 		   mad_send_wr->seg_num < mad_send_wr->newwin &&
740 		   mad_send_wr->seg_num < mad_send_wr->send_buf.seg_count) {
741 		/* Send failure will just result in a timeout/retry */
742 		ret = send_next_seg(mad_send_wr);
743 		if (ret)
744 			goto out;
745 
746 		mad_send_wr->refcount++;
747 		list_move_tail(&mad_send_wr->agent_list,
748 			      &mad_send_wr->mad_agent_priv->send_list);
749 	}
750 out:
751 	spin_unlock_irqrestore(&agent->lock, flags);
752 }
753 
754 static struct ib_mad_recv_wc *
755 process_rmpp_data(struct ib_mad_agent_private *agent,
756 		  struct ib_mad_recv_wc *mad_recv_wc)
757 {
758 	struct ib_rmpp_hdr *rmpp_hdr;
759 	u8 rmpp_status;
760 
761 	rmpp_hdr = &((struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad)->rmpp_hdr;
762 
763 	if (rmpp_hdr->rmpp_status) {
764 		rmpp_status = IB_MGMT_RMPP_STATUS_BAD_STATUS;
765 		goto bad;
766 	}
767 
768 	if (rmpp_hdr->seg_num == cpu_to_be32(1)) {
769 		if (!(ib_get_rmpp_flags(rmpp_hdr) & IB_MGMT_RMPP_FLAG_FIRST)) {
770 			rmpp_status = IB_MGMT_RMPP_STATUS_BAD_SEG;
771 			goto bad;
772 		}
773 		return start_rmpp(agent, mad_recv_wc);
774 	} else {
775 		if (ib_get_rmpp_flags(rmpp_hdr) & IB_MGMT_RMPP_FLAG_FIRST) {
776 			rmpp_status = IB_MGMT_RMPP_STATUS_BAD_SEG;
777 			goto bad;
778 		}
779 		return continue_rmpp(agent, mad_recv_wc);
780 	}
781 bad:
782 	nack_recv(agent, mad_recv_wc, rmpp_status);
783 	ib_free_recv_mad(mad_recv_wc);
784 	return NULL;
785 }
786 
787 static void process_rmpp_stop(struct ib_mad_agent_private *agent,
788 			      struct ib_mad_recv_wc *mad_recv_wc)
789 {
790 	struct ib_rmpp_mad *rmpp_mad;
791 
792 	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
793 
794 	if (rmpp_mad->rmpp_hdr.rmpp_status != IB_MGMT_RMPP_STATUS_RESX) {
795 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
796 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
797 	} else
798 		abort_send(agent, mad_recv_wc, rmpp_mad->rmpp_hdr.rmpp_status);
799 }
800 
801 static void process_rmpp_abort(struct ib_mad_agent_private *agent,
802 			       struct ib_mad_recv_wc *mad_recv_wc)
803 {
804 	struct ib_rmpp_mad *rmpp_mad;
805 
806 	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
807 
808 	if (rmpp_mad->rmpp_hdr.rmpp_status < IB_MGMT_RMPP_STATUS_ABORT_MIN ||
809 	    rmpp_mad->rmpp_hdr.rmpp_status > IB_MGMT_RMPP_STATUS_ABORT_MAX) {
810 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
811 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BAD_STATUS);
812 	} else
813 		abort_send(agent, mad_recv_wc, rmpp_mad->rmpp_hdr.rmpp_status);
814 }
815 
816 struct ib_mad_recv_wc *
817 ib_process_rmpp_recv_wc(struct ib_mad_agent_private *agent,
818 			struct ib_mad_recv_wc *mad_recv_wc)
819 {
820 	struct ib_rmpp_mad *rmpp_mad;
821 
822 	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
823 	if (!(rmpp_mad->rmpp_hdr.rmpp_rtime_flags & IB_MGMT_RMPP_FLAG_ACTIVE))
824 		return mad_recv_wc;
825 
826 	if (rmpp_mad->rmpp_hdr.rmpp_version != IB_MGMT_RMPP_VERSION) {
827 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_UNV);
828 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_UNV);
829 		goto out;
830 	}
831 
832 	switch (rmpp_mad->rmpp_hdr.rmpp_type) {
833 	case IB_MGMT_RMPP_TYPE_DATA:
834 		return process_rmpp_data(agent, mad_recv_wc);
835 	case IB_MGMT_RMPP_TYPE_ACK:
836 		process_rmpp_ack(agent, mad_recv_wc);
837 		break;
838 	case IB_MGMT_RMPP_TYPE_STOP:
839 		process_rmpp_stop(agent, mad_recv_wc);
840 		break;
841 	case IB_MGMT_RMPP_TYPE_ABORT:
842 		process_rmpp_abort(agent, mad_recv_wc);
843 		break;
844 	default:
845 		abort_send(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BADT);
846 		nack_recv(agent, mad_recv_wc, IB_MGMT_RMPP_STATUS_BADT);
847 		break;
848 	}
849 out:
850 	ib_free_recv_mad(mad_recv_wc);
851 	return NULL;
852 }
853 
854 static int init_newwin(struct ib_mad_send_wr_private *mad_send_wr)
855 {
856 	struct ib_mad_agent_private *agent = mad_send_wr->mad_agent_priv;
857 	struct ib_mad_hdr *mad_hdr = mad_send_wr->send_buf.mad;
858 	struct mad_rmpp_recv *rmpp_recv;
859 	struct rdma_ah_attr ah_attr;
860 	unsigned long flags;
861 	int newwin = 1;
862 
863 	if (!(mad_hdr->method & IB_MGMT_METHOD_RESP))
864 		goto out;
865 
866 	spin_lock_irqsave(&agent->lock, flags);
867 	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
868 		if (rmpp_recv->tid != mad_hdr->tid ||
869 		    rmpp_recv->mgmt_class != mad_hdr->mgmt_class ||
870 		    rmpp_recv->class_version != mad_hdr->class_version ||
871 		    (rmpp_recv->method & IB_MGMT_METHOD_RESP))
872 			continue;
873 
874 		if (rdma_query_ah(mad_send_wr->send_buf.ah, &ah_attr))
875 			continue;
876 
877 		if (rmpp_recv->slid == rdma_ah_get_dlid(&ah_attr)) {
878 			newwin = rmpp_recv->repwin;
879 			break;
880 		}
881 	}
882 	spin_unlock_irqrestore(&agent->lock, flags);
883 out:
884 	return newwin;
885 }
886 
887 int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr)
888 {
889 	struct ib_rmpp_mad *rmpp_mad;
890 	int ret;
891 
892 	rmpp_mad = mad_send_wr->send_buf.mad;
893 	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
894 	      IB_MGMT_RMPP_FLAG_ACTIVE))
895 		return IB_RMPP_RESULT_UNHANDLED;
896 
897 	if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA) {
898 		mad_send_wr->seg_num = 1;
899 		return IB_RMPP_RESULT_INTERNAL;
900 	}
901 
902 	mad_send_wr->newwin = init_newwin(mad_send_wr);
903 
904 	/* We need to wait for the final ACK even if there isn't a response */
905 	mad_send_wr->refcount += (mad_send_wr->timeout == 0);
906 	ret = send_next_seg(mad_send_wr);
907 	if (!ret)
908 		return IB_RMPP_RESULT_CONSUMED;
909 	return ret;
910 }
911 
912 int ib_process_rmpp_send_wc(struct ib_mad_send_wr_private *mad_send_wr,
913 			    struct ib_mad_send_wc *mad_send_wc)
914 {
915 	struct ib_rmpp_mad *rmpp_mad;
916 	int ret;
917 
918 	rmpp_mad = mad_send_wr->send_buf.mad;
919 	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
920 	      IB_MGMT_RMPP_FLAG_ACTIVE))
921 		return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
922 
923 	if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA)
924 		return IB_RMPP_RESULT_INTERNAL;	 /* ACK, STOP, or ABORT */
925 
926 	if (mad_send_wc->status != IB_WC_SUCCESS ||
927 	    mad_send_wr->status != IB_WC_SUCCESS)
928 		return IB_RMPP_RESULT_PROCESSED; /* Canceled or send error */
929 
930 	if (!mad_send_wr->timeout)
931 		return IB_RMPP_RESULT_PROCESSED; /* Response received */
932 
933 	if (mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) {
934 		mad_send_wr->timeout =
935 			msecs_to_jiffies(mad_send_wr->send_buf.timeout_ms);
936 		return IB_RMPP_RESULT_PROCESSED; /* Send done */
937 	}
938 
939 	if (mad_send_wr->seg_num == mad_send_wr->newwin ||
940 	    mad_send_wr->seg_num == mad_send_wr->send_buf.seg_count)
941 		return IB_RMPP_RESULT_PROCESSED; /* Wait for ACK */
942 
943 	ret = send_next_seg(mad_send_wr);
944 	if (ret) {
945 		mad_send_wc->status = IB_WC_GENERAL_ERR;
946 		return IB_RMPP_RESULT_PROCESSED;
947 	}
948 	return IB_RMPP_RESULT_CONSUMED;
949 }
950 
951 int ib_retry_rmpp(struct ib_mad_send_wr_private *mad_send_wr)
952 {
953 	struct ib_rmpp_mad *rmpp_mad;
954 	int ret;
955 
956 	rmpp_mad = mad_send_wr->send_buf.mad;
957 	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
958 	      IB_MGMT_RMPP_FLAG_ACTIVE))
959 		return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
960 
961 	if (mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count)
962 		return IB_RMPP_RESULT_PROCESSED;
963 
964 	mad_send_wr->seg_num = mad_send_wr->last_ack;
965 	mad_send_wr->cur_seg = mad_send_wr->last_ack_seg;
966 
967 	ret = send_next_seg(mad_send_wr);
968 	if (ret)
969 		return IB_RMPP_RESULT_PROCESSED;
970 
971 	return IB_RMPP_RESULT_CONSUMED;
972 }
973