xref: /openbmc/linux/net/smc/smc_llc.c (revision dfc53baa)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  *  Link Layer Control (LLC)
6  *
7  *  Copyright IBM Corp. 2016
8  *
9  *  Author(s):  Klaus Wacker <Klaus.Wacker@de.ibm.com>
10  *              Ursula Braun <ubraun@linux.vnet.ibm.com>
11  */
12 
13 #include <net/tcp.h>
14 #include <rdma/ib_verbs.h>
15 
16 #include "smc.h"
17 #include "smc_core.h"
18 #include "smc_clc.h"
19 #include "smc_llc.h"
20 #include "smc_pnet.h"
21 
22 #define SMC_LLC_DATA_LEN		40
23 
24 struct smc_llc_hdr {
25 	struct smc_wr_rx_hdr common;
26 	u8 length;	/* 44 */
27 #if defined(__BIG_ENDIAN_BITFIELD)
28 	u8 reserved:4,
29 	   add_link_rej_rsn:4;
30 #elif defined(__LITTLE_ENDIAN_BITFIELD)
31 	u8 add_link_rej_rsn:4,
32 	   reserved:4;
33 #endif
34 	u8 flags;
35 };
36 
37 #define SMC_LLC_FLAG_NO_RMBE_EYEC	0x03
38 
39 struct smc_llc_msg_confirm_link {	/* type 0x01 */
40 	struct smc_llc_hdr hd;
41 	u8 sender_mac[ETH_ALEN];
42 	u8 sender_gid[SMC_GID_SIZE];
43 	u8 sender_qp_num[3];
44 	u8 link_num;
45 	u8 link_uid[SMC_LGR_ID_SIZE];
46 	u8 max_links;
47 	u8 reserved[9];
48 };
49 
50 #define SMC_LLC_FLAG_ADD_LNK_REJ	0x40
51 #define SMC_LLC_REJ_RSN_NO_ALT_PATH	1
52 
53 #define SMC_LLC_ADD_LNK_MAX_LINKS	2
54 
55 struct smc_llc_msg_add_link {		/* type 0x02 */
56 	struct smc_llc_hdr hd;
57 	u8 sender_mac[ETH_ALEN];
58 	u8 reserved2[2];
59 	u8 sender_gid[SMC_GID_SIZE];
60 	u8 sender_qp_num[3];
61 	u8 link_num;
62 #if defined(__BIG_ENDIAN_BITFIELD)
63 	u8 reserved3 : 4,
64 	   qp_mtu   : 4;
65 #elif defined(__LITTLE_ENDIAN_BITFIELD)
66 	u8 qp_mtu   : 4,
67 	   reserved3 : 4;
68 #endif
69 	u8 initial_psn[3];
70 	u8 reserved[8];
71 };
72 
73 struct smc_llc_msg_add_link_cont_rt {
74 	__be32 rmb_key;
75 	__be32 rmb_key_new;
76 	__be64 rmb_vaddr_new;
77 };
78 
79 #define SMC_LLC_RKEYS_PER_CONT_MSG	2
80 
81 struct smc_llc_msg_add_link_cont {	/* type 0x03 */
82 	struct smc_llc_hdr hd;
83 	u8 link_num;
84 	u8 num_rkeys;
85 	u8 reserved2[2];
86 	struct smc_llc_msg_add_link_cont_rt rt[SMC_LLC_RKEYS_PER_CONT_MSG];
87 	u8 reserved[4];
88 } __packed;			/* format defined in RFC7609 */
89 
90 #define SMC_LLC_FLAG_DEL_LINK_ALL	0x40
91 #define SMC_LLC_FLAG_DEL_LINK_ORDERLY	0x20
92 
93 struct smc_llc_msg_del_link {		/* type 0x04 */
94 	struct smc_llc_hdr hd;
95 	u8 link_num;
96 	__be32 reason;
97 	u8 reserved[35];
98 } __packed;			/* format defined in RFC7609 */
99 
100 struct smc_llc_msg_test_link {		/* type 0x07 */
101 	struct smc_llc_hdr hd;
102 	u8 user_data[16];
103 	u8 reserved[24];
104 };
105 
106 struct smc_rmb_rtoken {
107 	union {
108 		u8 num_rkeys;	/* first rtoken byte of CONFIRM LINK msg */
109 				/* is actually the num of rtokens, first */
110 				/* rtoken is always for the current link */
111 		u8 link_id;	/* link id of the rtoken */
112 	};
113 	__be32 rmb_key;
114 	__be64 rmb_vaddr;
115 } __packed;			/* format defined in RFC7609 */
116 
117 #define SMC_LLC_RKEYS_PER_MSG	3
118 
119 struct smc_llc_msg_confirm_rkey {	/* type 0x06 */
120 	struct smc_llc_hdr hd;
121 	struct smc_rmb_rtoken rtoken[SMC_LLC_RKEYS_PER_MSG];
122 	u8 reserved;
123 };
124 
125 #define SMC_LLC_DEL_RKEY_MAX	8
126 #define SMC_LLC_FLAG_RKEY_RETRY	0x10
127 #define SMC_LLC_FLAG_RKEY_NEG	0x20
128 
129 struct smc_llc_msg_delete_rkey {	/* type 0x09 */
130 	struct smc_llc_hdr hd;
131 	u8 num_rkeys;
132 	u8 err_mask;
133 	u8 reserved[2];
134 	__be32 rkey[8];
135 	u8 reserved2[4];
136 };
137 
138 union smc_llc_msg {
139 	struct smc_llc_msg_confirm_link confirm_link;
140 	struct smc_llc_msg_add_link add_link;
141 	struct smc_llc_msg_add_link_cont add_link_cont;
142 	struct smc_llc_msg_del_link delete_link;
143 
144 	struct smc_llc_msg_confirm_rkey confirm_rkey;
145 	struct smc_llc_msg_delete_rkey delete_rkey;
146 
147 	struct smc_llc_msg_test_link test_link;
148 	struct {
149 		struct smc_llc_hdr hdr;
150 		u8 data[SMC_LLC_DATA_LEN];
151 	} raw;
152 };
153 
154 #define SMC_LLC_FLAG_RESP		0x80
155 
156 struct smc_llc_qentry {
157 	struct list_head list;
158 	struct smc_link *link;
159 	union smc_llc_msg msg;
160 };
161 
162 static void smc_llc_enqueue(struct smc_link *link, union smc_llc_msg *llc);
163 
164 struct smc_llc_qentry *smc_llc_flow_qentry_clr(struct smc_llc_flow *flow)
165 {
166 	struct smc_llc_qentry *qentry = flow->qentry;
167 
168 	flow->qentry = NULL;
169 	return qentry;
170 }
171 
172 void smc_llc_flow_qentry_del(struct smc_llc_flow *flow)
173 {
174 	struct smc_llc_qentry *qentry;
175 
176 	if (flow->qentry) {
177 		qentry = flow->qentry;
178 		flow->qentry = NULL;
179 		kfree(qentry);
180 	}
181 }
182 
183 static inline void smc_llc_flow_qentry_set(struct smc_llc_flow *flow,
184 					   struct smc_llc_qentry *qentry)
185 {
186 	flow->qentry = qentry;
187 }
188 
189 static void smc_llc_flow_parallel(struct smc_link_group *lgr, u8 flow_type,
190 				  struct smc_llc_qentry *qentry)
191 {
192 	u8 msg_type = qentry->msg.raw.hdr.common.type;
193 
194 	if ((msg_type == SMC_LLC_ADD_LINK || msg_type == SMC_LLC_DELETE_LINK) &&
195 	    flow_type != msg_type && !lgr->delayed_event) {
196 		lgr->delayed_event = qentry;
197 		return;
198 	}
199 	/* drop parallel or already-in-progress llc requests */
200 	if (flow_type != msg_type)
201 		pr_warn_once("smc: SMC-R lg %*phN dropped parallel "
202 			     "LLC msg: msg %d flow %d role %d\n",
203 			     SMC_LGR_ID_SIZE, &lgr->id,
204 			     qentry->msg.raw.hdr.common.type,
205 			     flow_type, lgr->role);
206 	kfree(qentry);
207 }
208 
209 /* try to start a new llc flow, initiated by an incoming llc msg */
210 static bool smc_llc_flow_start(struct smc_llc_flow *flow,
211 			       struct smc_llc_qentry *qentry)
212 {
213 	struct smc_link_group *lgr = qentry->link->lgr;
214 
215 	spin_lock_bh(&lgr->llc_flow_lock);
216 	if (flow->type) {
217 		/* a flow is already active */
218 		smc_llc_flow_parallel(lgr, flow->type, qentry);
219 		spin_unlock_bh(&lgr->llc_flow_lock);
220 		return false;
221 	}
222 	switch (qentry->msg.raw.hdr.common.type) {
223 	case SMC_LLC_ADD_LINK:
224 		flow->type = SMC_LLC_FLOW_ADD_LINK;
225 		break;
226 	case SMC_LLC_DELETE_LINK:
227 		flow->type = SMC_LLC_FLOW_DEL_LINK;
228 		break;
229 	case SMC_LLC_CONFIRM_RKEY:
230 	case SMC_LLC_DELETE_RKEY:
231 		flow->type = SMC_LLC_FLOW_RKEY;
232 		break;
233 	default:
234 		flow->type = SMC_LLC_FLOW_NONE;
235 	}
236 	if (qentry == lgr->delayed_event)
237 		lgr->delayed_event = NULL;
238 	smc_llc_flow_qentry_set(flow, qentry);
239 	spin_unlock_bh(&lgr->llc_flow_lock);
240 	return true;
241 }
242 
243 /* start a new local llc flow, wait till current flow finished */
244 int smc_llc_flow_initiate(struct smc_link_group *lgr,
245 			  enum smc_llc_flowtype type)
246 {
247 	enum smc_llc_flowtype allowed_remote = SMC_LLC_FLOW_NONE;
248 	int rc;
249 
250 	/* all flows except confirm_rkey and delete_rkey are exclusive,
251 	 * confirm/delete rkey flows can run concurrently (local and remote)
252 	 */
253 	if (type == SMC_LLC_FLOW_RKEY)
254 		allowed_remote = SMC_LLC_FLOW_RKEY;
255 again:
256 	if (list_empty(&lgr->list))
257 		return -ENODEV;
258 	spin_lock_bh(&lgr->llc_flow_lock);
259 	if (lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE &&
260 	    (lgr->llc_flow_rmt.type == SMC_LLC_FLOW_NONE ||
261 	     lgr->llc_flow_rmt.type == allowed_remote)) {
262 		lgr->llc_flow_lcl.type = type;
263 		spin_unlock_bh(&lgr->llc_flow_lock);
264 		return 0;
265 	}
266 	spin_unlock_bh(&lgr->llc_flow_lock);
267 	rc = wait_event_timeout(lgr->llc_flow_waiter, (list_empty(&lgr->list) ||
268 				(lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE &&
269 				 (lgr->llc_flow_rmt.type == SMC_LLC_FLOW_NONE ||
270 				  lgr->llc_flow_rmt.type == allowed_remote))),
271 				SMC_LLC_WAIT_TIME * 10);
272 	if (!rc)
273 		return -ETIMEDOUT;
274 	goto again;
275 }
276 
277 /* finish the current llc flow */
278 void smc_llc_flow_stop(struct smc_link_group *lgr, struct smc_llc_flow *flow)
279 {
280 	spin_lock_bh(&lgr->llc_flow_lock);
281 	memset(flow, 0, sizeof(*flow));
282 	flow->type = SMC_LLC_FLOW_NONE;
283 	spin_unlock_bh(&lgr->llc_flow_lock);
284 	if (!list_empty(&lgr->list) && lgr->delayed_event &&
285 	    flow == &lgr->llc_flow_lcl)
286 		schedule_work(&lgr->llc_event_work);
287 	else
288 		wake_up(&lgr->llc_flow_waiter);
289 }
290 
291 /* lnk is optional and used for early wakeup when link goes down, useful in
292  * cases where we wait for a response on the link after we sent a request
293  */
294 struct smc_llc_qentry *smc_llc_wait(struct smc_link_group *lgr,
295 				    struct smc_link *lnk,
296 				    int time_out, u8 exp_msg)
297 {
298 	struct smc_llc_flow *flow = &lgr->llc_flow_lcl;
299 	u8 rcv_msg;
300 
301 	wait_event_timeout(lgr->llc_msg_waiter,
302 			   (flow->qentry ||
303 			    (lnk && !smc_link_usable(lnk)) ||
304 			    list_empty(&lgr->list)),
305 			   time_out);
306 	if (!flow->qentry ||
307 	    (lnk && !smc_link_usable(lnk)) || list_empty(&lgr->list)) {
308 		smc_llc_flow_qentry_del(flow);
309 		goto out;
310 	}
311 	rcv_msg = flow->qentry->msg.raw.hdr.common.type;
312 	if (exp_msg && rcv_msg != exp_msg) {
313 		if (exp_msg == SMC_LLC_ADD_LINK &&
314 		    rcv_msg == SMC_LLC_DELETE_LINK) {
315 			/* flow_start will delay the unexpected msg */
316 			smc_llc_flow_start(&lgr->llc_flow_lcl,
317 					   smc_llc_flow_qentry_clr(flow));
318 			return NULL;
319 		}
320 		pr_warn_once("smc: SMC-R lg %*phN dropped unexpected LLC msg: "
321 			     "msg %d exp %d flow %d role %d flags %x\n",
322 			     SMC_LGR_ID_SIZE, &lgr->id, rcv_msg, exp_msg,
323 			     flow->type, lgr->role,
324 			     flow->qentry->msg.raw.hdr.flags);
325 		smc_llc_flow_qentry_del(flow);
326 	}
327 out:
328 	return flow->qentry;
329 }
330 
331 /********************************** send *************************************/
332 
333 struct smc_llc_tx_pend {
334 };
335 
336 /* handler for send/transmission completion of an LLC msg */
337 static void smc_llc_tx_handler(struct smc_wr_tx_pend_priv *pend,
338 			       struct smc_link *link,
339 			       enum ib_wc_status wc_status)
340 {
341 	/* future work: handle wc_status error for recovery and failover */
342 }
343 
344 /**
345  * smc_llc_add_pending_send() - add LLC control message to pending WQE transmits
346  * @link: Pointer to SMC link used for sending LLC control message.
347  * @wr_buf: Out variable returning pointer to work request payload buffer.
348  * @pend: Out variable returning pointer to private pending WR tracking.
349  *	  It's the context the transmit complete handler will get.
350  *
351  * Reserves and pre-fills an entry for a pending work request send/tx.
352  * Used by mid-level smc_llc_send_msg() to prepare for later actual send/tx.
353  * Can sleep due to smc_get_ctrl_buf (if not in softirq context).
354  *
355  * Return: 0 on success, otherwise an error value.
356  */
357 static int smc_llc_add_pending_send(struct smc_link *link,
358 				    struct smc_wr_buf **wr_buf,
359 				    struct smc_wr_tx_pend_priv **pend)
360 {
361 	int rc;
362 
363 	rc = smc_wr_tx_get_free_slot(link, smc_llc_tx_handler, wr_buf, NULL,
364 				     pend);
365 	if (rc < 0)
366 		return rc;
367 	BUILD_BUG_ON_MSG(
368 		sizeof(union smc_llc_msg) > SMC_WR_BUF_SIZE,
369 		"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_llc_msg)");
370 	BUILD_BUG_ON_MSG(
371 		sizeof(union smc_llc_msg) != SMC_WR_TX_SIZE,
372 		"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_llc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
373 	BUILD_BUG_ON_MSG(
374 		sizeof(struct smc_llc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
375 		"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_llc_tx_pend)");
376 	return 0;
377 }
378 
379 /* high-level API to send LLC confirm link */
380 int smc_llc_send_confirm_link(struct smc_link *link,
381 			      enum smc_llc_reqresp reqresp)
382 {
383 	struct smc_llc_msg_confirm_link *confllc;
384 	struct smc_wr_tx_pend_priv *pend;
385 	struct smc_wr_buf *wr_buf;
386 	int rc;
387 
388 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
389 	if (rc)
390 		return rc;
391 	confllc = (struct smc_llc_msg_confirm_link *)wr_buf;
392 	memset(confllc, 0, sizeof(*confllc));
393 	confllc->hd.common.type = SMC_LLC_CONFIRM_LINK;
394 	confllc->hd.length = sizeof(struct smc_llc_msg_confirm_link);
395 	confllc->hd.flags |= SMC_LLC_FLAG_NO_RMBE_EYEC;
396 	if (reqresp == SMC_LLC_RESP)
397 		confllc->hd.flags |= SMC_LLC_FLAG_RESP;
398 	memcpy(confllc->sender_mac, link->smcibdev->mac[link->ibport - 1],
399 	       ETH_ALEN);
400 	memcpy(confllc->sender_gid, link->gid, SMC_GID_SIZE);
401 	hton24(confllc->sender_qp_num, link->roce_qp->qp_num);
402 	confllc->link_num = link->link_id;
403 	memcpy(confllc->link_uid, link->link_uid, SMC_LGR_ID_SIZE);
404 	confllc->max_links = SMC_LLC_ADD_LNK_MAX_LINKS;
405 	/* send llc message */
406 	rc = smc_wr_tx_send(link, pend);
407 	return rc;
408 }
409 
410 /* send LLC confirm rkey request */
411 static int smc_llc_send_confirm_rkey(struct smc_link *send_link,
412 				     struct smc_buf_desc *rmb_desc)
413 {
414 	struct smc_llc_msg_confirm_rkey *rkeyllc;
415 	struct smc_wr_tx_pend_priv *pend;
416 	struct smc_wr_buf *wr_buf;
417 	struct smc_link *link;
418 	int i, rc, rtok_ix;
419 
420 	rc = smc_llc_add_pending_send(send_link, &wr_buf, &pend);
421 	if (rc)
422 		return rc;
423 	rkeyllc = (struct smc_llc_msg_confirm_rkey *)wr_buf;
424 	memset(rkeyllc, 0, sizeof(*rkeyllc));
425 	rkeyllc->hd.common.type = SMC_LLC_CONFIRM_RKEY;
426 	rkeyllc->hd.length = sizeof(struct smc_llc_msg_confirm_rkey);
427 
428 	rtok_ix = 1;
429 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
430 		link = &send_link->lgr->lnk[i];
431 		if (smc_link_active(link) && link != send_link) {
432 			rkeyllc->rtoken[rtok_ix].link_id = link->link_id;
433 			rkeyllc->rtoken[rtok_ix].rmb_key =
434 				htonl(rmb_desc->mr_rx[link->link_idx]->rkey);
435 			rkeyllc->rtoken[rtok_ix].rmb_vaddr = cpu_to_be64(
436 				(u64)sg_dma_address(
437 					rmb_desc->sgt[link->link_idx].sgl));
438 			rtok_ix++;
439 		}
440 	}
441 	/* rkey of send_link is in rtoken[0] */
442 	rkeyllc->rtoken[0].num_rkeys = rtok_ix - 1;
443 	rkeyllc->rtoken[0].rmb_key =
444 		htonl(rmb_desc->mr_rx[send_link->link_idx]->rkey);
445 	rkeyllc->rtoken[0].rmb_vaddr = cpu_to_be64(
446 		(u64)sg_dma_address(rmb_desc->sgt[send_link->link_idx].sgl));
447 	/* send llc message */
448 	rc = smc_wr_tx_send(send_link, pend);
449 	return rc;
450 }
451 
452 /* send LLC delete rkey request */
453 static int smc_llc_send_delete_rkey(struct smc_link *link,
454 				    struct smc_buf_desc *rmb_desc)
455 {
456 	struct smc_llc_msg_delete_rkey *rkeyllc;
457 	struct smc_wr_tx_pend_priv *pend;
458 	struct smc_wr_buf *wr_buf;
459 	int rc;
460 
461 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
462 	if (rc)
463 		return rc;
464 	rkeyllc = (struct smc_llc_msg_delete_rkey *)wr_buf;
465 	memset(rkeyllc, 0, sizeof(*rkeyllc));
466 	rkeyllc->hd.common.type = SMC_LLC_DELETE_RKEY;
467 	rkeyllc->hd.length = sizeof(struct smc_llc_msg_delete_rkey);
468 	rkeyllc->num_rkeys = 1;
469 	rkeyllc->rkey[0] = htonl(rmb_desc->mr_rx[link->link_idx]->rkey);
470 	/* send llc message */
471 	rc = smc_wr_tx_send(link, pend);
472 	return rc;
473 }
474 
475 /* send ADD LINK request or response */
476 int smc_llc_send_add_link(struct smc_link *link, u8 mac[], u8 gid[],
477 			  struct smc_link *link_new,
478 			  enum smc_llc_reqresp reqresp)
479 {
480 	struct smc_llc_msg_add_link *addllc;
481 	struct smc_wr_tx_pend_priv *pend;
482 	struct smc_wr_buf *wr_buf;
483 	int rc;
484 
485 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
486 	if (rc)
487 		return rc;
488 	addllc = (struct smc_llc_msg_add_link *)wr_buf;
489 
490 	memset(addllc, 0, sizeof(*addllc));
491 	addllc->hd.common.type = SMC_LLC_ADD_LINK;
492 	addllc->hd.length = sizeof(struct smc_llc_msg_add_link);
493 	if (reqresp == SMC_LLC_RESP)
494 		addllc->hd.flags |= SMC_LLC_FLAG_RESP;
495 	memcpy(addllc->sender_mac, mac, ETH_ALEN);
496 	memcpy(addllc->sender_gid, gid, SMC_GID_SIZE);
497 	if (link_new) {
498 		addllc->link_num = link_new->link_id;
499 		hton24(addllc->sender_qp_num, link_new->roce_qp->qp_num);
500 		hton24(addllc->initial_psn, link_new->psn_initial);
501 		if (reqresp == SMC_LLC_REQ)
502 			addllc->qp_mtu = link_new->path_mtu;
503 		else
504 			addllc->qp_mtu = min(link_new->path_mtu,
505 					     link_new->peer_mtu);
506 	}
507 	/* send llc message */
508 	rc = smc_wr_tx_send(link, pend);
509 	return rc;
510 }
511 
512 /* send DELETE LINK request or response */
513 int smc_llc_send_delete_link(struct smc_link *link, u8 link_del_id,
514 			     enum smc_llc_reqresp reqresp, bool orderly,
515 			     u32 reason)
516 {
517 	struct smc_llc_msg_del_link *delllc;
518 	struct smc_wr_tx_pend_priv *pend;
519 	struct smc_wr_buf *wr_buf;
520 	int rc;
521 
522 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
523 	if (rc)
524 		return rc;
525 	delllc = (struct smc_llc_msg_del_link *)wr_buf;
526 
527 	memset(delllc, 0, sizeof(*delllc));
528 	delllc->hd.common.type = SMC_LLC_DELETE_LINK;
529 	delllc->hd.length = sizeof(struct smc_llc_msg_del_link);
530 	if (reqresp == SMC_LLC_RESP)
531 		delllc->hd.flags |= SMC_LLC_FLAG_RESP;
532 	if (orderly)
533 		delllc->hd.flags |= SMC_LLC_FLAG_DEL_LINK_ORDERLY;
534 	if (link_del_id)
535 		delllc->link_num = link_del_id;
536 	else
537 		delllc->hd.flags |= SMC_LLC_FLAG_DEL_LINK_ALL;
538 	delllc->reason = htonl(reason);
539 	/* send llc message */
540 	rc = smc_wr_tx_send(link, pend);
541 	return rc;
542 }
543 
544 /* send LLC test link request */
545 static int smc_llc_send_test_link(struct smc_link *link, u8 user_data[16])
546 {
547 	struct smc_llc_msg_test_link *testllc;
548 	struct smc_wr_tx_pend_priv *pend;
549 	struct smc_wr_buf *wr_buf;
550 	int rc;
551 
552 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
553 	if (rc)
554 		return rc;
555 	testllc = (struct smc_llc_msg_test_link *)wr_buf;
556 	memset(testllc, 0, sizeof(*testllc));
557 	testllc->hd.common.type = SMC_LLC_TEST_LINK;
558 	testllc->hd.length = sizeof(struct smc_llc_msg_test_link);
559 	memcpy(testllc->user_data, user_data, sizeof(testllc->user_data));
560 	/* send llc message */
561 	rc = smc_wr_tx_send(link, pend);
562 	return rc;
563 }
564 
565 /* schedule an llc send on link, may wait for buffers */
566 static int smc_llc_send_message(struct smc_link *link, void *llcbuf)
567 {
568 	struct smc_wr_tx_pend_priv *pend;
569 	struct smc_wr_buf *wr_buf;
570 	int rc;
571 
572 	if (!smc_link_usable(link))
573 		return -ENOLINK;
574 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
575 	if (rc)
576 		return rc;
577 	memcpy(wr_buf, llcbuf, sizeof(union smc_llc_msg));
578 	return smc_wr_tx_send(link, pend);
579 }
580 
581 /* schedule an llc send on link, may wait for buffers,
582  * and wait for send completion notification.
583  * @return 0 on success
584  */
585 static int smc_llc_send_message_wait(struct smc_link *link, void *llcbuf)
586 {
587 	struct smc_wr_tx_pend_priv *pend;
588 	struct smc_wr_buf *wr_buf;
589 	int rc;
590 
591 	if (!smc_link_usable(link))
592 		return -ENOLINK;
593 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
594 	if (rc)
595 		return rc;
596 	memcpy(wr_buf, llcbuf, sizeof(union smc_llc_msg));
597 	return smc_wr_tx_send_wait(link, pend, SMC_LLC_WAIT_TIME);
598 }
599 
600 /********************************* receive ***********************************/
601 
602 static int smc_llc_alloc_alt_link(struct smc_link_group *lgr,
603 				  enum smc_lgr_type lgr_new_t)
604 {
605 	int i;
606 
607 	if (lgr->type == SMC_LGR_SYMMETRIC ||
608 	    (lgr->type != SMC_LGR_SINGLE &&
609 	     (lgr_new_t == SMC_LGR_ASYMMETRIC_LOCAL ||
610 	      lgr_new_t == SMC_LGR_ASYMMETRIC_PEER)))
611 		return -EMLINK;
612 
613 	if (lgr_new_t == SMC_LGR_ASYMMETRIC_LOCAL ||
614 	    lgr_new_t == SMC_LGR_ASYMMETRIC_PEER) {
615 		for (i = SMC_LINKS_PER_LGR_MAX - 1; i >= 0; i--)
616 			if (lgr->lnk[i].state == SMC_LNK_UNUSED)
617 				return i;
618 	} else {
619 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
620 			if (lgr->lnk[i].state == SMC_LNK_UNUSED)
621 				return i;
622 	}
623 	return -EMLINK;
624 }
625 
626 /* return first buffer from any of the next buf lists */
627 static struct smc_buf_desc *_smc_llc_get_next_rmb(struct smc_link_group *lgr,
628 						  int *buf_lst)
629 {
630 	struct smc_buf_desc *buf_pos;
631 
632 	while (*buf_lst < SMC_RMBE_SIZES) {
633 		buf_pos = list_first_entry_or_null(&lgr->rmbs[*buf_lst],
634 						   struct smc_buf_desc, list);
635 		if (buf_pos)
636 			return buf_pos;
637 		(*buf_lst)++;
638 	}
639 	return NULL;
640 }
641 
642 /* return next rmb from buffer lists */
643 static struct smc_buf_desc *smc_llc_get_next_rmb(struct smc_link_group *lgr,
644 						 int *buf_lst,
645 						 struct smc_buf_desc *buf_pos)
646 {
647 	struct smc_buf_desc *buf_next;
648 
649 	if (!buf_pos || list_is_last(&buf_pos->list, &lgr->rmbs[*buf_lst])) {
650 		(*buf_lst)++;
651 		return _smc_llc_get_next_rmb(lgr, buf_lst);
652 	}
653 	buf_next = list_next_entry(buf_pos, list);
654 	return buf_next;
655 }
656 
657 static struct smc_buf_desc *smc_llc_get_first_rmb(struct smc_link_group *lgr,
658 						  int *buf_lst)
659 {
660 	*buf_lst = 0;
661 	return smc_llc_get_next_rmb(lgr, buf_lst, NULL);
662 }
663 
664 /* send one add_link_continue msg */
665 static int smc_llc_add_link_cont(struct smc_link *link,
666 				 struct smc_link *link_new, u8 *num_rkeys_todo,
667 				 int *buf_lst, struct smc_buf_desc **buf_pos)
668 {
669 	struct smc_llc_msg_add_link_cont *addc_llc;
670 	struct smc_link_group *lgr = link->lgr;
671 	int prim_lnk_idx, lnk_idx, i, rc;
672 	struct smc_wr_tx_pend_priv *pend;
673 	struct smc_wr_buf *wr_buf;
674 	struct smc_buf_desc *rmb;
675 	u8 n;
676 
677 	rc = smc_llc_add_pending_send(link, &wr_buf, &pend);
678 	if (rc)
679 		return rc;
680 	addc_llc = (struct smc_llc_msg_add_link_cont *)wr_buf;
681 	memset(addc_llc, 0, sizeof(*addc_llc));
682 
683 	prim_lnk_idx = link->link_idx;
684 	lnk_idx = link_new->link_idx;
685 	addc_llc->link_num = link_new->link_id;
686 	addc_llc->num_rkeys = *num_rkeys_todo;
687 	n = *num_rkeys_todo;
688 	for (i = 0; i < min_t(u8, n, SMC_LLC_RKEYS_PER_CONT_MSG); i++) {
689 		if (!*buf_pos) {
690 			addc_llc->num_rkeys = addc_llc->num_rkeys -
691 					      *num_rkeys_todo;
692 			*num_rkeys_todo = 0;
693 			break;
694 		}
695 		rmb = *buf_pos;
696 
697 		addc_llc->rt[i].rmb_key = htonl(rmb->mr_rx[prim_lnk_idx]->rkey);
698 		addc_llc->rt[i].rmb_key_new = htonl(rmb->mr_rx[lnk_idx]->rkey);
699 		addc_llc->rt[i].rmb_vaddr_new =
700 			cpu_to_be64((u64)sg_dma_address(rmb->sgt[lnk_idx].sgl));
701 
702 		(*num_rkeys_todo)--;
703 		*buf_pos = smc_llc_get_next_rmb(lgr, buf_lst, *buf_pos);
704 		while (*buf_pos && !(*buf_pos)->used)
705 			*buf_pos = smc_llc_get_next_rmb(lgr, buf_lst, *buf_pos);
706 	}
707 	addc_llc->hd.common.type = SMC_LLC_ADD_LINK_CONT;
708 	addc_llc->hd.length = sizeof(struct smc_llc_msg_add_link_cont);
709 	if (lgr->role == SMC_CLNT)
710 		addc_llc->hd.flags |= SMC_LLC_FLAG_RESP;
711 	return smc_wr_tx_send(link, pend);
712 }
713 
714 static int smc_llc_cli_rkey_exchange(struct smc_link *link,
715 				     struct smc_link *link_new)
716 {
717 	struct smc_llc_msg_add_link_cont *addc_llc;
718 	struct smc_link_group *lgr = link->lgr;
719 	u8 max, num_rkeys_send, num_rkeys_recv;
720 	struct smc_llc_qentry *qentry;
721 	struct smc_buf_desc *buf_pos;
722 	int buf_lst;
723 	int rc = 0;
724 	int i;
725 
726 	mutex_lock(&lgr->rmbs_lock);
727 	num_rkeys_send = lgr->conns_num;
728 	buf_pos = smc_llc_get_first_rmb(lgr, &buf_lst);
729 	do {
730 		qentry = smc_llc_wait(lgr, NULL, SMC_LLC_WAIT_TIME,
731 				      SMC_LLC_ADD_LINK_CONT);
732 		if (!qentry) {
733 			rc = -ETIMEDOUT;
734 			break;
735 		}
736 		addc_llc = &qentry->msg.add_link_cont;
737 		num_rkeys_recv = addc_llc->num_rkeys;
738 		max = min_t(u8, num_rkeys_recv, SMC_LLC_RKEYS_PER_CONT_MSG);
739 		for (i = 0; i < max; i++) {
740 			smc_rtoken_set(lgr, link->link_idx, link_new->link_idx,
741 				       addc_llc->rt[i].rmb_key,
742 				       addc_llc->rt[i].rmb_vaddr_new,
743 				       addc_llc->rt[i].rmb_key_new);
744 			num_rkeys_recv--;
745 		}
746 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
747 		rc = smc_llc_add_link_cont(link, link_new, &num_rkeys_send,
748 					   &buf_lst, &buf_pos);
749 		if (rc)
750 			break;
751 	} while (num_rkeys_send || num_rkeys_recv);
752 
753 	mutex_unlock(&lgr->rmbs_lock);
754 	return rc;
755 }
756 
757 /* prepare and send an add link reject response */
758 static int smc_llc_cli_add_link_reject(struct smc_llc_qentry *qentry)
759 {
760 	qentry->msg.raw.hdr.flags |= SMC_LLC_FLAG_RESP;
761 	qentry->msg.raw.hdr.flags |= SMC_LLC_FLAG_ADD_LNK_REJ;
762 	qentry->msg.raw.hdr.add_link_rej_rsn = SMC_LLC_REJ_RSN_NO_ALT_PATH;
763 	return smc_llc_send_message(qentry->link, &qentry->msg);
764 }
765 
766 static int smc_llc_cli_conf_link(struct smc_link *link,
767 				 struct smc_init_info *ini,
768 				 struct smc_link *link_new,
769 				 enum smc_lgr_type lgr_new_t)
770 {
771 	struct smc_link_group *lgr = link->lgr;
772 	struct smc_llc_qentry *qentry = NULL;
773 	int rc = 0;
774 
775 	/* receive CONFIRM LINK request over RoCE fabric */
776 	qentry = smc_llc_wait(lgr, NULL, SMC_LLC_WAIT_FIRST_TIME, 0);
777 	if (!qentry) {
778 		rc = smc_llc_send_delete_link(link, link_new->link_id,
779 					      SMC_LLC_REQ, false,
780 					      SMC_LLC_DEL_LOST_PATH);
781 		return -ENOLINK;
782 	}
783 	if (qentry->msg.raw.hdr.common.type != SMC_LLC_CONFIRM_LINK) {
784 		/* received DELETE_LINK instead */
785 		qentry->msg.raw.hdr.flags |= SMC_LLC_FLAG_RESP;
786 		smc_llc_send_message(link, &qentry->msg);
787 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
788 		return -ENOLINK;
789 	}
790 	smc_llc_save_peer_uid(qentry);
791 	smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
792 
793 	rc = smc_ib_modify_qp_rts(link_new);
794 	if (rc) {
795 		smc_llc_send_delete_link(link, link_new->link_id, SMC_LLC_REQ,
796 					 false, SMC_LLC_DEL_LOST_PATH);
797 		return -ENOLINK;
798 	}
799 	smc_wr_remember_qp_attr(link_new);
800 
801 	rc = smcr_buf_reg_lgr(link_new);
802 	if (rc) {
803 		smc_llc_send_delete_link(link, link_new->link_id, SMC_LLC_REQ,
804 					 false, SMC_LLC_DEL_LOST_PATH);
805 		return -ENOLINK;
806 	}
807 
808 	/* send CONFIRM LINK response over RoCE fabric */
809 	rc = smc_llc_send_confirm_link(link_new, SMC_LLC_RESP);
810 	if (rc) {
811 		smc_llc_send_delete_link(link, link_new->link_id, SMC_LLC_REQ,
812 					 false, SMC_LLC_DEL_LOST_PATH);
813 		return -ENOLINK;
814 	}
815 	smc_llc_link_active(link_new);
816 	if (lgr_new_t == SMC_LGR_ASYMMETRIC_LOCAL ||
817 	    lgr_new_t == SMC_LGR_ASYMMETRIC_PEER)
818 		smcr_lgr_set_type_asym(lgr, lgr_new_t, link_new->link_idx);
819 	else
820 		smcr_lgr_set_type(lgr, lgr_new_t);
821 	return 0;
822 }
823 
824 static void smc_llc_save_add_link_info(struct smc_link *link,
825 				       struct smc_llc_msg_add_link *add_llc)
826 {
827 	link->peer_qpn = ntoh24(add_llc->sender_qp_num);
828 	memcpy(link->peer_gid, add_llc->sender_gid, SMC_GID_SIZE);
829 	memcpy(link->peer_mac, add_llc->sender_mac, ETH_ALEN);
830 	link->peer_psn = ntoh24(add_llc->initial_psn);
831 	link->peer_mtu = add_llc->qp_mtu;
832 }
833 
834 /* as an SMC client, process an add link request */
835 int smc_llc_cli_add_link(struct smc_link *link, struct smc_llc_qentry *qentry)
836 {
837 	struct smc_llc_msg_add_link *llc = &qentry->msg.add_link;
838 	enum smc_lgr_type lgr_new_t = SMC_LGR_SYMMETRIC;
839 	struct smc_link_group *lgr = smc_get_lgr(link);
840 	struct smc_link *lnk_new = NULL;
841 	struct smc_init_info ini;
842 	int lnk_idx, rc = 0;
843 
844 	if (!llc->qp_mtu)
845 		goto out_reject;
846 
847 	ini.vlan_id = lgr->vlan_id;
848 	smc_pnet_find_alt_roce(lgr, &ini, link->smcibdev);
849 	if (!memcmp(llc->sender_gid, link->peer_gid, SMC_GID_SIZE) &&
850 	    !memcmp(llc->sender_mac, link->peer_mac, ETH_ALEN)) {
851 		if (!ini.ib_dev)
852 			goto out_reject;
853 		lgr_new_t = SMC_LGR_ASYMMETRIC_PEER;
854 	}
855 	if (!ini.ib_dev) {
856 		lgr_new_t = SMC_LGR_ASYMMETRIC_LOCAL;
857 		ini.ib_dev = link->smcibdev;
858 		ini.ib_port = link->ibport;
859 	}
860 	lnk_idx = smc_llc_alloc_alt_link(lgr, lgr_new_t);
861 	if (lnk_idx < 0)
862 		goto out_reject;
863 	lnk_new = &lgr->lnk[lnk_idx];
864 	rc = smcr_link_init(lgr, lnk_new, lnk_idx, &ini);
865 	if (rc)
866 		goto out_reject;
867 	smc_llc_save_add_link_info(lnk_new, llc);
868 	lnk_new->link_id = llc->link_num;	/* SMC server assigns link id */
869 	smc_llc_link_set_uid(lnk_new);
870 
871 	rc = smc_ib_ready_link(lnk_new);
872 	if (rc)
873 		goto out_clear_lnk;
874 
875 	rc = smcr_buf_map_lgr(lnk_new);
876 	if (rc)
877 		goto out_clear_lnk;
878 
879 	rc = smc_llc_send_add_link(link,
880 				   lnk_new->smcibdev->mac[ini.ib_port - 1],
881 				   lnk_new->gid, lnk_new, SMC_LLC_RESP);
882 	if (rc)
883 		goto out_clear_lnk;
884 	rc = smc_llc_cli_rkey_exchange(link, lnk_new);
885 	if (rc) {
886 		rc = 0;
887 		goto out_clear_lnk;
888 	}
889 	rc = smc_llc_cli_conf_link(link, &ini, lnk_new, lgr_new_t);
890 	if (!rc)
891 		goto out;
892 out_clear_lnk:
893 	smcr_link_clear(lnk_new, false);
894 out_reject:
895 	smc_llc_cli_add_link_reject(qentry);
896 out:
897 	kfree(qentry);
898 	return rc;
899 }
900 
901 /* as an SMC client, invite server to start the add_link processing */
902 static void smc_llc_cli_add_link_invite(struct smc_link *link,
903 					struct smc_llc_qentry *qentry)
904 {
905 	struct smc_link_group *lgr = smc_get_lgr(link);
906 	struct smc_init_info ini;
907 
908 	if (lgr->type == SMC_LGR_SYMMETRIC ||
909 	    lgr->type == SMC_LGR_ASYMMETRIC_PEER)
910 		goto out;
911 
912 	ini.vlan_id = lgr->vlan_id;
913 	smc_pnet_find_alt_roce(lgr, &ini, link->smcibdev);
914 	if (!ini.ib_dev)
915 		goto out;
916 
917 	smc_llc_send_add_link(link, ini.ib_dev->mac[ini.ib_port - 1],
918 			      ini.ib_gid, NULL, SMC_LLC_REQ);
919 out:
920 	kfree(qentry);
921 }
922 
923 static bool smc_llc_is_empty_llc_message(union smc_llc_msg *llc)
924 {
925 	int i;
926 
927 	for (i = 0; i < ARRAY_SIZE(llc->raw.data); i++)
928 		if (llc->raw.data[i])
929 			return false;
930 	return true;
931 }
932 
933 static bool smc_llc_is_local_add_link(union smc_llc_msg *llc)
934 {
935 	if (llc->raw.hdr.common.type == SMC_LLC_ADD_LINK &&
936 	    smc_llc_is_empty_llc_message(llc))
937 		return true;
938 	return false;
939 }
940 
941 static void smc_llc_process_cli_add_link(struct smc_link_group *lgr)
942 {
943 	struct smc_llc_qentry *qentry;
944 
945 	qentry = smc_llc_flow_qentry_clr(&lgr->llc_flow_lcl);
946 
947 	mutex_lock(&lgr->llc_conf_mutex);
948 	if (smc_llc_is_local_add_link(&qentry->msg))
949 		smc_llc_cli_add_link_invite(qentry->link, qentry);
950 	else
951 		smc_llc_cli_add_link(qentry->link, qentry);
952 	mutex_unlock(&lgr->llc_conf_mutex);
953 }
954 
955 static int smc_llc_active_link_count(struct smc_link_group *lgr)
956 {
957 	int i, link_count = 0;
958 
959 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
960 		if (!smc_link_active(&lgr->lnk[i]))
961 			continue;
962 		link_count++;
963 	}
964 	return link_count;
965 }
966 
967 /* find the asymmetric link when 3 links are established  */
968 static struct smc_link *smc_llc_find_asym_link(struct smc_link_group *lgr)
969 {
970 	int asym_idx = -ENOENT;
971 	int i, j, k;
972 	bool found;
973 
974 	/* determine asymmetric link */
975 	found = false;
976 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
977 		for (j = i + 1; j < SMC_LINKS_PER_LGR_MAX; j++) {
978 			if (!smc_link_usable(&lgr->lnk[i]) ||
979 			    !smc_link_usable(&lgr->lnk[j]))
980 				continue;
981 			if (!memcmp(lgr->lnk[i].gid, lgr->lnk[j].gid,
982 				    SMC_GID_SIZE)) {
983 				found = true;	/* asym_lnk is i or j */
984 				break;
985 			}
986 		}
987 		if (found)
988 			break;
989 	}
990 	if (!found)
991 		goto out; /* no asymmetric link */
992 	for (k = 0; k < SMC_LINKS_PER_LGR_MAX; k++) {
993 		if (!smc_link_usable(&lgr->lnk[k]))
994 			continue;
995 		if (k != i &&
996 		    !memcmp(lgr->lnk[i].peer_gid, lgr->lnk[k].peer_gid,
997 			    SMC_GID_SIZE)) {
998 			asym_idx = i;
999 			break;
1000 		}
1001 		if (k != j &&
1002 		    !memcmp(lgr->lnk[j].peer_gid, lgr->lnk[k].peer_gid,
1003 			    SMC_GID_SIZE)) {
1004 			asym_idx = j;
1005 			break;
1006 		}
1007 	}
1008 out:
1009 	return (asym_idx < 0) ? NULL : &lgr->lnk[asym_idx];
1010 }
1011 
1012 static void smc_llc_delete_asym_link(struct smc_link_group *lgr)
1013 {
1014 	struct smc_link *lnk_new = NULL, *lnk_asym;
1015 	struct smc_llc_qentry *qentry;
1016 	int rc;
1017 
1018 	lnk_asym = smc_llc_find_asym_link(lgr);
1019 	if (!lnk_asym)
1020 		return; /* no asymmetric link */
1021 	if (!smc_link_downing(&lnk_asym->state))
1022 		return;
1023 	lnk_new = smc_switch_conns(lgr, lnk_asym, false);
1024 	smc_wr_tx_wait_no_pending_sends(lnk_asym);
1025 	if (!lnk_new)
1026 		goto out_free;
1027 	/* change flow type from ADD_LINK into DEL_LINK */
1028 	lgr->llc_flow_lcl.type = SMC_LLC_FLOW_DEL_LINK;
1029 	rc = smc_llc_send_delete_link(lnk_new, lnk_asym->link_id, SMC_LLC_REQ,
1030 				      true, SMC_LLC_DEL_NO_ASYM_NEEDED);
1031 	if (rc) {
1032 		smcr_link_down_cond(lnk_new);
1033 		goto out_free;
1034 	}
1035 	qentry = smc_llc_wait(lgr, lnk_new, SMC_LLC_WAIT_TIME,
1036 			      SMC_LLC_DELETE_LINK);
1037 	if (!qentry) {
1038 		smcr_link_down_cond(lnk_new);
1039 		goto out_free;
1040 	}
1041 	smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1042 out_free:
1043 	smcr_link_clear(lnk_asym, true);
1044 }
1045 
1046 static int smc_llc_srv_rkey_exchange(struct smc_link *link,
1047 				     struct smc_link *link_new)
1048 {
1049 	struct smc_llc_msg_add_link_cont *addc_llc;
1050 	struct smc_link_group *lgr = link->lgr;
1051 	u8 max, num_rkeys_send, num_rkeys_recv;
1052 	struct smc_llc_qentry *qentry = NULL;
1053 	struct smc_buf_desc *buf_pos;
1054 	int buf_lst;
1055 	int rc = 0;
1056 	int i;
1057 
1058 	mutex_lock(&lgr->rmbs_lock);
1059 	num_rkeys_send = lgr->conns_num;
1060 	buf_pos = smc_llc_get_first_rmb(lgr, &buf_lst);
1061 	do {
1062 		smc_llc_add_link_cont(link, link_new, &num_rkeys_send,
1063 				      &buf_lst, &buf_pos);
1064 		qentry = smc_llc_wait(lgr, link, SMC_LLC_WAIT_TIME,
1065 				      SMC_LLC_ADD_LINK_CONT);
1066 		if (!qentry) {
1067 			rc = -ETIMEDOUT;
1068 			goto out;
1069 		}
1070 		addc_llc = &qentry->msg.add_link_cont;
1071 		num_rkeys_recv = addc_llc->num_rkeys;
1072 		max = min_t(u8, num_rkeys_recv, SMC_LLC_RKEYS_PER_CONT_MSG);
1073 		for (i = 0; i < max; i++) {
1074 			smc_rtoken_set(lgr, link->link_idx, link_new->link_idx,
1075 				       addc_llc->rt[i].rmb_key,
1076 				       addc_llc->rt[i].rmb_vaddr_new,
1077 				       addc_llc->rt[i].rmb_key_new);
1078 			num_rkeys_recv--;
1079 		}
1080 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1081 	} while (num_rkeys_send || num_rkeys_recv);
1082 out:
1083 	mutex_unlock(&lgr->rmbs_lock);
1084 	return rc;
1085 }
1086 
1087 static int smc_llc_srv_conf_link(struct smc_link *link,
1088 				 struct smc_link *link_new,
1089 				 enum smc_lgr_type lgr_new_t)
1090 {
1091 	struct smc_link_group *lgr = link->lgr;
1092 	struct smc_llc_qentry *qentry = NULL;
1093 	int rc;
1094 
1095 	/* send CONFIRM LINK request over the RoCE fabric */
1096 	rc = smc_llc_send_confirm_link(link_new, SMC_LLC_REQ);
1097 	if (rc)
1098 		return -ENOLINK;
1099 	/* receive CONFIRM LINK response over the RoCE fabric */
1100 	qentry = smc_llc_wait(lgr, link, SMC_LLC_WAIT_FIRST_TIME, 0);
1101 	if (!qentry ||
1102 	    qentry->msg.raw.hdr.common.type != SMC_LLC_CONFIRM_LINK) {
1103 		/* send DELETE LINK */
1104 		smc_llc_send_delete_link(link, link_new->link_id, SMC_LLC_REQ,
1105 					 false, SMC_LLC_DEL_LOST_PATH);
1106 		if (qentry)
1107 			smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1108 		return -ENOLINK;
1109 	}
1110 	smc_llc_save_peer_uid(qentry);
1111 	smc_llc_link_active(link_new);
1112 	if (lgr_new_t == SMC_LGR_ASYMMETRIC_LOCAL ||
1113 	    lgr_new_t == SMC_LGR_ASYMMETRIC_PEER)
1114 		smcr_lgr_set_type_asym(lgr, lgr_new_t, link_new->link_idx);
1115 	else
1116 		smcr_lgr_set_type(lgr, lgr_new_t);
1117 	smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1118 	return 0;
1119 }
1120 
1121 int smc_llc_srv_add_link(struct smc_link *link)
1122 {
1123 	enum smc_lgr_type lgr_new_t = SMC_LGR_SYMMETRIC;
1124 	struct smc_link_group *lgr = link->lgr;
1125 	struct smc_llc_msg_add_link *add_llc;
1126 	struct smc_llc_qentry *qentry = NULL;
1127 	struct smc_link *link_new;
1128 	struct smc_init_info ini;
1129 	int lnk_idx, rc = 0;
1130 
1131 	/* ignore client add link recommendation, start new flow */
1132 	ini.vlan_id = lgr->vlan_id;
1133 	smc_pnet_find_alt_roce(lgr, &ini, link->smcibdev);
1134 	if (!ini.ib_dev) {
1135 		lgr_new_t = SMC_LGR_ASYMMETRIC_LOCAL;
1136 		ini.ib_dev = link->smcibdev;
1137 		ini.ib_port = link->ibport;
1138 	}
1139 	lnk_idx = smc_llc_alloc_alt_link(lgr, lgr_new_t);
1140 	if (lnk_idx < 0)
1141 		return 0;
1142 
1143 	rc = smcr_link_init(lgr, &lgr->lnk[lnk_idx], lnk_idx, &ini);
1144 	if (rc)
1145 		return rc;
1146 	link_new = &lgr->lnk[lnk_idx];
1147 	rc = smc_llc_send_add_link(link,
1148 				   link_new->smcibdev->mac[ini.ib_port - 1],
1149 				   link_new->gid, link_new, SMC_LLC_REQ);
1150 	if (rc)
1151 		goto out_err;
1152 	/* receive ADD LINK response over the RoCE fabric */
1153 	qentry = smc_llc_wait(lgr, link, SMC_LLC_WAIT_TIME, SMC_LLC_ADD_LINK);
1154 	if (!qentry) {
1155 		rc = -ETIMEDOUT;
1156 		goto out_err;
1157 	}
1158 	add_llc = &qentry->msg.add_link;
1159 	if (add_llc->hd.flags & SMC_LLC_FLAG_ADD_LNK_REJ) {
1160 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1161 		rc = -ENOLINK;
1162 		goto out_err;
1163 	}
1164 	if (lgr->type == SMC_LGR_SINGLE &&
1165 	    (!memcmp(add_llc->sender_gid, link->peer_gid, SMC_GID_SIZE) &&
1166 	     !memcmp(add_llc->sender_mac, link->peer_mac, ETH_ALEN))) {
1167 		lgr_new_t = SMC_LGR_ASYMMETRIC_PEER;
1168 	}
1169 	smc_llc_save_add_link_info(link_new, add_llc);
1170 	smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1171 
1172 	rc = smc_ib_ready_link(link_new);
1173 	if (rc)
1174 		goto out_err;
1175 	rc = smcr_buf_map_lgr(link_new);
1176 	if (rc)
1177 		goto out_err;
1178 	rc = smcr_buf_reg_lgr(link_new);
1179 	if (rc)
1180 		goto out_err;
1181 	rc = smc_llc_srv_rkey_exchange(link, link_new);
1182 	if (rc)
1183 		goto out_err;
1184 	rc = smc_llc_srv_conf_link(link, link_new, lgr_new_t);
1185 	if (rc)
1186 		goto out_err;
1187 	return 0;
1188 out_err:
1189 	smcr_link_clear(link_new, false);
1190 	return rc;
1191 }
1192 
1193 static void smc_llc_process_srv_add_link(struct smc_link_group *lgr)
1194 {
1195 	struct smc_link *link = lgr->llc_flow_lcl.qentry->link;
1196 	int rc;
1197 
1198 	smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1199 
1200 	mutex_lock(&lgr->llc_conf_mutex);
1201 	rc = smc_llc_srv_add_link(link);
1202 	if (!rc && lgr->type == SMC_LGR_SYMMETRIC) {
1203 		/* delete any asymmetric link */
1204 		smc_llc_delete_asym_link(lgr);
1205 	}
1206 	mutex_unlock(&lgr->llc_conf_mutex);
1207 }
1208 
1209 /* enqueue a local add_link req to trigger a new add_link flow */
1210 void smc_llc_add_link_local(struct smc_link *link)
1211 {
1212 	struct smc_llc_msg_add_link add_llc = {0};
1213 
1214 	add_llc.hd.length = sizeof(add_llc);
1215 	add_llc.hd.common.type = SMC_LLC_ADD_LINK;
1216 	/* no dev and port needed */
1217 	smc_llc_enqueue(link, (union smc_llc_msg *)&add_llc);
1218 }
1219 
1220 /* worker to process an add link message */
1221 static void smc_llc_add_link_work(struct work_struct *work)
1222 {
1223 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
1224 						  llc_add_link_work);
1225 
1226 	if (list_empty(&lgr->list)) {
1227 		/* link group is terminating */
1228 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1229 		goto out;
1230 	}
1231 
1232 	if (lgr->role == SMC_CLNT)
1233 		smc_llc_process_cli_add_link(lgr);
1234 	else
1235 		smc_llc_process_srv_add_link(lgr);
1236 out:
1237 	smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1238 }
1239 
1240 /* enqueue a local del_link msg to trigger a new del_link flow,
1241  * called only for role SMC_SERV
1242  */
1243 void smc_llc_srv_delete_link_local(struct smc_link *link, u8 del_link_id)
1244 {
1245 	struct smc_llc_msg_del_link del_llc = {0};
1246 
1247 	del_llc.hd.length = sizeof(del_llc);
1248 	del_llc.hd.common.type = SMC_LLC_DELETE_LINK;
1249 	del_llc.link_num = del_link_id;
1250 	del_llc.reason = htonl(SMC_LLC_DEL_LOST_PATH);
1251 	del_llc.hd.flags |= SMC_LLC_FLAG_DEL_LINK_ORDERLY;
1252 	smc_llc_enqueue(link, (union smc_llc_msg *)&del_llc);
1253 }
1254 
1255 static void smc_llc_process_cli_delete_link(struct smc_link_group *lgr)
1256 {
1257 	struct smc_link *lnk_del = NULL, *lnk_asym, *lnk;
1258 	struct smc_llc_msg_del_link *del_llc;
1259 	struct smc_llc_qentry *qentry;
1260 	int active_links;
1261 	int lnk_idx;
1262 
1263 	qentry = smc_llc_flow_qentry_clr(&lgr->llc_flow_lcl);
1264 	lnk = qentry->link;
1265 	del_llc = &qentry->msg.delete_link;
1266 
1267 	if (del_llc->hd.flags & SMC_LLC_FLAG_DEL_LINK_ALL) {
1268 		smc_lgr_terminate_sched(lgr);
1269 		goto out;
1270 	}
1271 	mutex_lock(&lgr->llc_conf_mutex);
1272 	/* delete single link */
1273 	for (lnk_idx = 0; lnk_idx < SMC_LINKS_PER_LGR_MAX; lnk_idx++) {
1274 		if (lgr->lnk[lnk_idx].link_id != del_llc->link_num)
1275 			continue;
1276 		lnk_del = &lgr->lnk[lnk_idx];
1277 		break;
1278 	}
1279 	del_llc->hd.flags |= SMC_LLC_FLAG_RESP;
1280 	if (!lnk_del) {
1281 		/* link was not found */
1282 		del_llc->reason = htonl(SMC_LLC_DEL_NOLNK);
1283 		smc_llc_send_message(lnk, &qentry->msg);
1284 		goto out_unlock;
1285 	}
1286 	lnk_asym = smc_llc_find_asym_link(lgr);
1287 
1288 	del_llc->reason = 0;
1289 	smc_llc_send_message(lnk, &qentry->msg); /* response */
1290 
1291 	if (smc_link_downing(&lnk_del->state)) {
1292 		if (smc_switch_conns(lgr, lnk_del, false))
1293 			smc_wr_tx_wait_no_pending_sends(lnk_del);
1294 	}
1295 	smcr_link_clear(lnk_del, true);
1296 
1297 	active_links = smc_llc_active_link_count(lgr);
1298 	if (lnk_del == lnk_asym) {
1299 		/* expected deletion of asym link, don't change lgr state */
1300 	} else if (active_links == 1) {
1301 		smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1302 	} else if (!active_links) {
1303 		smcr_lgr_set_type(lgr, SMC_LGR_NONE);
1304 		smc_lgr_terminate_sched(lgr);
1305 	}
1306 out_unlock:
1307 	mutex_unlock(&lgr->llc_conf_mutex);
1308 out:
1309 	kfree(qentry);
1310 }
1311 
1312 /* try to send a DELETE LINK ALL request on any active link,
1313  * waiting for send completion
1314  */
1315 void smc_llc_send_link_delete_all(struct smc_link_group *lgr, bool ord, u32 rsn)
1316 {
1317 	struct smc_llc_msg_del_link delllc = {0};
1318 	int i;
1319 
1320 	delllc.hd.common.type = SMC_LLC_DELETE_LINK;
1321 	delllc.hd.length = sizeof(delllc);
1322 	if (ord)
1323 		delllc.hd.flags |= SMC_LLC_FLAG_DEL_LINK_ORDERLY;
1324 	delllc.hd.flags |= SMC_LLC_FLAG_DEL_LINK_ALL;
1325 	delllc.reason = htonl(rsn);
1326 
1327 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1328 		if (!smc_link_usable(&lgr->lnk[i]))
1329 			continue;
1330 		if (!smc_llc_send_message_wait(&lgr->lnk[i], &delllc))
1331 			break;
1332 	}
1333 }
1334 
1335 static void smc_llc_process_srv_delete_link(struct smc_link_group *lgr)
1336 {
1337 	struct smc_llc_msg_del_link *del_llc;
1338 	struct smc_link *lnk, *lnk_del;
1339 	struct smc_llc_qentry *qentry;
1340 	int active_links;
1341 	int i;
1342 
1343 	mutex_lock(&lgr->llc_conf_mutex);
1344 	qentry = smc_llc_flow_qentry_clr(&lgr->llc_flow_lcl);
1345 	lnk = qentry->link;
1346 	del_llc = &qentry->msg.delete_link;
1347 
1348 	if (qentry->msg.delete_link.hd.flags & SMC_LLC_FLAG_DEL_LINK_ALL) {
1349 		/* delete entire lgr */
1350 		smc_llc_send_link_delete_all(lgr, true, ntohl(
1351 					      qentry->msg.delete_link.reason));
1352 		smc_lgr_terminate_sched(lgr);
1353 		goto out;
1354 	}
1355 	/* delete single link */
1356 	lnk_del = NULL;
1357 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1358 		if (lgr->lnk[i].link_id == del_llc->link_num) {
1359 			lnk_del = &lgr->lnk[i];
1360 			break;
1361 		}
1362 	}
1363 	if (!lnk_del)
1364 		goto out; /* asymmetric link already deleted */
1365 
1366 	if (smc_link_downing(&lnk_del->state)) {
1367 		if (smc_switch_conns(lgr, lnk_del, false))
1368 			smc_wr_tx_wait_no_pending_sends(lnk_del);
1369 	}
1370 	if (!list_empty(&lgr->list)) {
1371 		/* qentry is either a request from peer (send it back to
1372 		 * initiate the DELETE_LINK processing), or a locally
1373 		 * enqueued DELETE_LINK request (forward it)
1374 		 */
1375 		if (!smc_llc_send_message(lnk, &qentry->msg)) {
1376 			struct smc_llc_qentry *qentry2;
1377 
1378 			qentry2 = smc_llc_wait(lgr, lnk, SMC_LLC_WAIT_TIME,
1379 					       SMC_LLC_DELETE_LINK);
1380 			if (qentry2)
1381 				smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1382 		}
1383 	}
1384 	smcr_link_clear(lnk_del, true);
1385 
1386 	active_links = smc_llc_active_link_count(lgr);
1387 	if (active_links == 1) {
1388 		smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1389 	} else if (!active_links) {
1390 		smcr_lgr_set_type(lgr, SMC_LGR_NONE);
1391 		smc_lgr_terminate_sched(lgr);
1392 	}
1393 
1394 	if (lgr->type == SMC_LGR_SINGLE && !list_empty(&lgr->list)) {
1395 		/* trigger setup of asymm alt link */
1396 		smc_llc_add_link_local(lnk);
1397 	}
1398 out:
1399 	mutex_unlock(&lgr->llc_conf_mutex);
1400 	kfree(qentry);
1401 }
1402 
1403 static void smc_llc_delete_link_work(struct work_struct *work)
1404 {
1405 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
1406 						  llc_del_link_work);
1407 
1408 	if (list_empty(&lgr->list)) {
1409 		/* link group is terminating */
1410 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1411 		goto out;
1412 	}
1413 
1414 	if (lgr->role == SMC_CLNT)
1415 		smc_llc_process_cli_delete_link(lgr);
1416 	else
1417 		smc_llc_process_srv_delete_link(lgr);
1418 out:
1419 	smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1420 }
1421 
1422 /* process a confirm_rkey request from peer, remote flow */
1423 static void smc_llc_rmt_conf_rkey(struct smc_link_group *lgr)
1424 {
1425 	struct smc_llc_msg_confirm_rkey *llc;
1426 	struct smc_llc_qentry *qentry;
1427 	struct smc_link *link;
1428 	int num_entries;
1429 	int rk_idx;
1430 	int i;
1431 
1432 	qentry = lgr->llc_flow_rmt.qentry;
1433 	llc = &qentry->msg.confirm_rkey;
1434 	link = qentry->link;
1435 
1436 	num_entries = llc->rtoken[0].num_rkeys;
1437 	/* first rkey entry is for receiving link */
1438 	rk_idx = smc_rtoken_add(link,
1439 				llc->rtoken[0].rmb_vaddr,
1440 				llc->rtoken[0].rmb_key);
1441 	if (rk_idx < 0)
1442 		goto out_err;
1443 
1444 	for (i = 1; i <= min_t(u8, num_entries, SMC_LLC_RKEYS_PER_MSG - 1); i++)
1445 		smc_rtoken_set2(lgr, rk_idx, llc->rtoken[i].link_id,
1446 				llc->rtoken[i].rmb_vaddr,
1447 				llc->rtoken[i].rmb_key);
1448 	/* max links is 3 so there is no need to support conf_rkey_cont msgs */
1449 	goto out;
1450 out_err:
1451 	llc->hd.flags |= SMC_LLC_FLAG_RKEY_NEG;
1452 	llc->hd.flags |= SMC_LLC_FLAG_RKEY_RETRY;
1453 out:
1454 	llc->hd.flags |= SMC_LLC_FLAG_RESP;
1455 	smc_llc_send_message(link, &qentry->msg);
1456 	smc_llc_flow_qentry_del(&lgr->llc_flow_rmt);
1457 }
1458 
1459 /* process a delete_rkey request from peer, remote flow */
1460 static void smc_llc_rmt_delete_rkey(struct smc_link_group *lgr)
1461 {
1462 	struct smc_llc_msg_delete_rkey *llc;
1463 	struct smc_llc_qentry *qentry;
1464 	struct smc_link *link;
1465 	u8 err_mask = 0;
1466 	int i, max;
1467 
1468 	qentry = lgr->llc_flow_rmt.qentry;
1469 	llc = &qentry->msg.delete_rkey;
1470 	link = qentry->link;
1471 
1472 	max = min_t(u8, llc->num_rkeys, SMC_LLC_DEL_RKEY_MAX);
1473 	for (i = 0; i < max; i++) {
1474 		if (smc_rtoken_delete(link, llc->rkey[i]))
1475 			err_mask |= 1 << (SMC_LLC_DEL_RKEY_MAX - 1 - i);
1476 	}
1477 	if (err_mask) {
1478 		llc->hd.flags |= SMC_LLC_FLAG_RKEY_NEG;
1479 		llc->err_mask = err_mask;
1480 	}
1481 	llc->hd.flags |= SMC_LLC_FLAG_RESP;
1482 	smc_llc_send_message(link, &qentry->msg);
1483 	smc_llc_flow_qentry_del(&lgr->llc_flow_rmt);
1484 }
1485 
1486 static void smc_llc_protocol_violation(struct smc_link_group *lgr, u8 type)
1487 {
1488 	pr_warn_ratelimited("smc: SMC-R lg %*phN LLC protocol violation: "
1489 			    "llc_type %d\n", SMC_LGR_ID_SIZE, &lgr->id, type);
1490 	smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_PROT_VIOL);
1491 	smc_lgr_terminate_sched(lgr);
1492 }
1493 
1494 /* flush the llc event queue */
1495 static void smc_llc_event_flush(struct smc_link_group *lgr)
1496 {
1497 	struct smc_llc_qentry *qentry, *q;
1498 
1499 	spin_lock_bh(&lgr->llc_event_q_lock);
1500 	list_for_each_entry_safe(qentry, q, &lgr->llc_event_q, list) {
1501 		list_del_init(&qentry->list);
1502 		kfree(qentry);
1503 	}
1504 	spin_unlock_bh(&lgr->llc_event_q_lock);
1505 }
1506 
1507 static void smc_llc_event_handler(struct smc_llc_qentry *qentry)
1508 {
1509 	union smc_llc_msg *llc = &qentry->msg;
1510 	struct smc_link *link = qentry->link;
1511 	struct smc_link_group *lgr = link->lgr;
1512 
1513 	if (!smc_link_usable(link))
1514 		goto out;
1515 
1516 	switch (llc->raw.hdr.common.type) {
1517 	case SMC_LLC_TEST_LINK:
1518 		llc->test_link.hd.flags |= SMC_LLC_FLAG_RESP;
1519 		smc_llc_send_message(link, llc);
1520 		break;
1521 	case SMC_LLC_ADD_LINK:
1522 		if (list_empty(&lgr->list))
1523 			goto out;	/* lgr is terminating */
1524 		if (lgr->role == SMC_CLNT) {
1525 			if (smc_llc_is_local_add_link(llc)) {
1526 				if (lgr->llc_flow_lcl.type ==
1527 				    SMC_LLC_FLOW_ADD_LINK)
1528 					break;	/* add_link in progress */
1529 				if (smc_llc_flow_start(&lgr->llc_flow_lcl,
1530 						       qentry)) {
1531 					schedule_work(&lgr->llc_add_link_work);
1532 				}
1533 				return;
1534 			}
1535 			if (lgr->llc_flow_lcl.type == SMC_LLC_FLOW_ADD_LINK &&
1536 			    !lgr->llc_flow_lcl.qentry) {
1537 				/* a flow is waiting for this message */
1538 				smc_llc_flow_qentry_set(&lgr->llc_flow_lcl,
1539 							qentry);
1540 				wake_up(&lgr->llc_msg_waiter);
1541 			} else if (smc_llc_flow_start(&lgr->llc_flow_lcl,
1542 						      qentry)) {
1543 				schedule_work(&lgr->llc_add_link_work);
1544 			}
1545 		} else if (smc_llc_flow_start(&lgr->llc_flow_lcl, qentry)) {
1546 			/* as smc server, handle client suggestion */
1547 			schedule_work(&lgr->llc_add_link_work);
1548 		}
1549 		return;
1550 	case SMC_LLC_CONFIRM_LINK:
1551 	case SMC_LLC_ADD_LINK_CONT:
1552 		if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) {
1553 			/* a flow is waiting for this message */
1554 			smc_llc_flow_qentry_set(&lgr->llc_flow_lcl, qentry);
1555 			wake_up(&lgr->llc_msg_waiter);
1556 			return;
1557 		}
1558 		break;
1559 	case SMC_LLC_DELETE_LINK:
1560 		if (lgr->llc_flow_lcl.type == SMC_LLC_FLOW_ADD_LINK &&
1561 		    !lgr->llc_flow_lcl.qentry) {
1562 			/* DEL LINK REQ during ADD LINK SEQ */
1563 			smc_llc_flow_qentry_set(&lgr->llc_flow_lcl, qentry);
1564 			wake_up(&lgr->llc_msg_waiter);
1565 		} else if (smc_llc_flow_start(&lgr->llc_flow_lcl, qentry)) {
1566 			schedule_work(&lgr->llc_del_link_work);
1567 		}
1568 		return;
1569 	case SMC_LLC_CONFIRM_RKEY:
1570 		/* new request from remote, assign to remote flow */
1571 		if (smc_llc_flow_start(&lgr->llc_flow_rmt, qentry)) {
1572 			/* process here, does not wait for more llc msgs */
1573 			smc_llc_rmt_conf_rkey(lgr);
1574 			smc_llc_flow_stop(lgr, &lgr->llc_flow_rmt);
1575 		}
1576 		return;
1577 	case SMC_LLC_CONFIRM_RKEY_CONT:
1578 		/* not used because max links is 3, and 3 rkeys fit into
1579 		 * one CONFIRM_RKEY message
1580 		 */
1581 		break;
1582 	case SMC_LLC_DELETE_RKEY:
1583 		/* new request from remote, assign to remote flow */
1584 		if (smc_llc_flow_start(&lgr->llc_flow_rmt, qentry)) {
1585 			/* process here, does not wait for more llc msgs */
1586 			smc_llc_rmt_delete_rkey(lgr);
1587 			smc_llc_flow_stop(lgr, &lgr->llc_flow_rmt);
1588 		}
1589 		return;
1590 	default:
1591 		smc_llc_protocol_violation(lgr, llc->raw.hdr.common.type);
1592 		break;
1593 	}
1594 out:
1595 	kfree(qentry);
1596 }
1597 
1598 /* worker to process llc messages on the event queue */
1599 static void smc_llc_event_work(struct work_struct *work)
1600 {
1601 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
1602 						  llc_event_work);
1603 	struct smc_llc_qentry *qentry;
1604 
1605 	if (!lgr->llc_flow_lcl.type && lgr->delayed_event) {
1606 		if (smc_link_usable(lgr->delayed_event->link)) {
1607 			smc_llc_event_handler(lgr->delayed_event);
1608 		} else {
1609 			qentry = lgr->delayed_event;
1610 			lgr->delayed_event = NULL;
1611 			kfree(qentry);
1612 		}
1613 	}
1614 
1615 again:
1616 	spin_lock_bh(&lgr->llc_event_q_lock);
1617 	if (!list_empty(&lgr->llc_event_q)) {
1618 		qentry = list_first_entry(&lgr->llc_event_q,
1619 					  struct smc_llc_qentry, list);
1620 		list_del_init(&qentry->list);
1621 		spin_unlock_bh(&lgr->llc_event_q_lock);
1622 		smc_llc_event_handler(qentry);
1623 		goto again;
1624 	}
1625 	spin_unlock_bh(&lgr->llc_event_q_lock);
1626 }
1627 
1628 /* process llc responses in tasklet context */
1629 static void smc_llc_rx_response(struct smc_link *link,
1630 				struct smc_llc_qentry *qentry)
1631 {
1632 	enum smc_llc_flowtype flowtype = link->lgr->llc_flow_lcl.type;
1633 	struct smc_llc_flow *flow = &link->lgr->llc_flow_lcl;
1634 	u8 llc_type = qentry->msg.raw.hdr.common.type;
1635 
1636 	switch (llc_type) {
1637 	case SMC_LLC_TEST_LINK:
1638 		if (smc_link_active(link))
1639 			complete(&link->llc_testlink_resp);
1640 		break;
1641 	case SMC_LLC_ADD_LINK:
1642 	case SMC_LLC_ADD_LINK_CONT:
1643 	case SMC_LLC_CONFIRM_LINK:
1644 		if (flowtype != SMC_LLC_FLOW_ADD_LINK || flow->qentry)
1645 			break;	/* drop out-of-flow response */
1646 		goto assign;
1647 	case SMC_LLC_DELETE_LINK:
1648 		if (flowtype != SMC_LLC_FLOW_DEL_LINK || flow->qentry)
1649 			break;	/* drop out-of-flow response */
1650 		goto assign;
1651 	case SMC_LLC_CONFIRM_RKEY:
1652 	case SMC_LLC_DELETE_RKEY:
1653 		if (flowtype != SMC_LLC_FLOW_RKEY || flow->qentry)
1654 			break;	/* drop out-of-flow response */
1655 		goto assign;
1656 	case SMC_LLC_CONFIRM_RKEY_CONT:
1657 		/* not used because max links is 3 */
1658 		break;
1659 	default:
1660 		smc_llc_protocol_violation(link->lgr, llc_type);
1661 		break;
1662 	}
1663 	kfree(qentry);
1664 	return;
1665 assign:
1666 	/* assign responses to the local flow, we requested them */
1667 	smc_llc_flow_qentry_set(&link->lgr->llc_flow_lcl, qentry);
1668 	wake_up(&link->lgr->llc_msg_waiter);
1669 }
1670 
1671 static void smc_llc_enqueue(struct smc_link *link, union smc_llc_msg *llc)
1672 {
1673 	struct smc_link_group *lgr = link->lgr;
1674 	struct smc_llc_qentry *qentry;
1675 	unsigned long flags;
1676 
1677 	qentry = kmalloc(sizeof(*qentry), GFP_ATOMIC);
1678 	if (!qentry)
1679 		return;
1680 	qentry->link = link;
1681 	INIT_LIST_HEAD(&qentry->list);
1682 	memcpy(&qentry->msg, llc, sizeof(union smc_llc_msg));
1683 
1684 	/* process responses immediately */
1685 	if (llc->raw.hdr.flags & SMC_LLC_FLAG_RESP) {
1686 		smc_llc_rx_response(link, qentry);
1687 		return;
1688 	}
1689 
1690 	/* add requests to event queue */
1691 	spin_lock_irqsave(&lgr->llc_event_q_lock, flags);
1692 	list_add_tail(&qentry->list, &lgr->llc_event_q);
1693 	spin_unlock_irqrestore(&lgr->llc_event_q_lock, flags);
1694 	schedule_work(&lgr->llc_event_work);
1695 }
1696 
1697 /* copy received msg and add it to the event queue */
1698 static void smc_llc_rx_handler(struct ib_wc *wc, void *buf)
1699 {
1700 	struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
1701 	union smc_llc_msg *llc = buf;
1702 
1703 	if (wc->byte_len < sizeof(*llc))
1704 		return; /* short message */
1705 	if (llc->raw.hdr.length != sizeof(*llc))
1706 		return; /* invalid message */
1707 
1708 	smc_llc_enqueue(link, llc);
1709 }
1710 
1711 /***************************** worker, utils *********************************/
1712 
1713 static void smc_llc_testlink_work(struct work_struct *work)
1714 {
1715 	struct smc_link *link = container_of(to_delayed_work(work),
1716 					     struct smc_link, llc_testlink_wrk);
1717 	unsigned long next_interval;
1718 	unsigned long expire_time;
1719 	u8 user_data[16] = { 0 };
1720 	int rc;
1721 
1722 	if (!smc_link_active(link))
1723 		return;		/* don't reschedule worker */
1724 	expire_time = link->wr_rx_tstamp + link->llc_testlink_time;
1725 	if (time_is_after_jiffies(expire_time)) {
1726 		next_interval = expire_time - jiffies;
1727 		goto out;
1728 	}
1729 	reinit_completion(&link->llc_testlink_resp);
1730 	smc_llc_send_test_link(link, user_data);
1731 	/* receive TEST LINK response over RoCE fabric */
1732 	rc = wait_for_completion_interruptible_timeout(&link->llc_testlink_resp,
1733 						       SMC_LLC_WAIT_TIME);
1734 	if (!smc_link_active(link))
1735 		return;		/* link state changed */
1736 	if (rc <= 0) {
1737 		smcr_link_down_cond_sched(link);
1738 		return;
1739 	}
1740 	next_interval = link->llc_testlink_time;
1741 out:
1742 	schedule_delayed_work(&link->llc_testlink_wrk, next_interval);
1743 }
1744 
1745 void smc_llc_lgr_init(struct smc_link_group *lgr, struct smc_sock *smc)
1746 {
1747 	struct net *net = sock_net(smc->clcsock->sk);
1748 
1749 	INIT_WORK(&lgr->llc_event_work, smc_llc_event_work);
1750 	INIT_WORK(&lgr->llc_add_link_work, smc_llc_add_link_work);
1751 	INIT_WORK(&lgr->llc_del_link_work, smc_llc_delete_link_work);
1752 	INIT_LIST_HEAD(&lgr->llc_event_q);
1753 	spin_lock_init(&lgr->llc_event_q_lock);
1754 	spin_lock_init(&lgr->llc_flow_lock);
1755 	init_waitqueue_head(&lgr->llc_flow_waiter);
1756 	init_waitqueue_head(&lgr->llc_msg_waiter);
1757 	mutex_init(&lgr->llc_conf_mutex);
1758 	lgr->llc_testlink_time = net->ipv4.sysctl_tcp_keepalive_time;
1759 }
1760 
1761 /* called after lgr was removed from lgr_list */
1762 void smc_llc_lgr_clear(struct smc_link_group *lgr)
1763 {
1764 	smc_llc_event_flush(lgr);
1765 	wake_up_all(&lgr->llc_flow_waiter);
1766 	wake_up_all(&lgr->llc_msg_waiter);
1767 	cancel_work_sync(&lgr->llc_event_work);
1768 	cancel_work_sync(&lgr->llc_add_link_work);
1769 	cancel_work_sync(&lgr->llc_del_link_work);
1770 	if (lgr->delayed_event) {
1771 		kfree(lgr->delayed_event);
1772 		lgr->delayed_event = NULL;
1773 	}
1774 }
1775 
1776 int smc_llc_link_init(struct smc_link *link)
1777 {
1778 	init_completion(&link->llc_testlink_resp);
1779 	INIT_DELAYED_WORK(&link->llc_testlink_wrk, smc_llc_testlink_work);
1780 	return 0;
1781 }
1782 
1783 void smc_llc_link_active(struct smc_link *link)
1784 {
1785 	pr_warn_ratelimited("smc: SMC-R lg %*phN link added: id %*phN, "
1786 			    "peerid %*phN, ibdev %s, ibport %d\n",
1787 			    SMC_LGR_ID_SIZE, &link->lgr->id,
1788 			    SMC_LGR_ID_SIZE, &link->link_uid,
1789 			    SMC_LGR_ID_SIZE, &link->peer_link_uid,
1790 			    link->smcibdev->ibdev->name, link->ibport);
1791 	link->state = SMC_LNK_ACTIVE;
1792 	if (link->lgr->llc_testlink_time) {
1793 		link->llc_testlink_time = link->lgr->llc_testlink_time * HZ;
1794 		schedule_delayed_work(&link->llc_testlink_wrk,
1795 				      link->llc_testlink_time);
1796 	}
1797 }
1798 
1799 /* called in worker context */
1800 void smc_llc_link_clear(struct smc_link *link, bool log)
1801 {
1802 	if (log)
1803 		pr_warn_ratelimited("smc: SMC-R lg %*phN link removed: id %*phN"
1804 				    ", peerid %*phN, ibdev %s, ibport %d\n",
1805 				    SMC_LGR_ID_SIZE, &link->lgr->id,
1806 				    SMC_LGR_ID_SIZE, &link->link_uid,
1807 				    SMC_LGR_ID_SIZE, &link->peer_link_uid,
1808 				    link->smcibdev->ibdev->name, link->ibport);
1809 	complete(&link->llc_testlink_resp);
1810 	cancel_delayed_work_sync(&link->llc_testlink_wrk);
1811 	smc_wr_wakeup_reg_wait(link);
1812 	smc_wr_wakeup_tx_wait(link);
1813 }
1814 
1815 /* register a new rtoken at the remote peer (for all links) */
1816 int smc_llc_do_confirm_rkey(struct smc_link *send_link,
1817 			    struct smc_buf_desc *rmb_desc)
1818 {
1819 	struct smc_link_group *lgr = send_link->lgr;
1820 	struct smc_llc_qentry *qentry = NULL;
1821 	int rc = 0;
1822 
1823 	rc = smc_llc_send_confirm_rkey(send_link, rmb_desc);
1824 	if (rc)
1825 		goto out;
1826 	/* receive CONFIRM RKEY response from server over RoCE fabric */
1827 	qentry = smc_llc_wait(lgr, send_link, SMC_LLC_WAIT_TIME,
1828 			      SMC_LLC_CONFIRM_RKEY);
1829 	if (!qentry || (qentry->msg.raw.hdr.flags & SMC_LLC_FLAG_RKEY_NEG))
1830 		rc = -EFAULT;
1831 out:
1832 	if (qentry)
1833 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1834 	return rc;
1835 }
1836 
1837 /* unregister an rtoken at the remote peer */
1838 int smc_llc_do_delete_rkey(struct smc_link_group *lgr,
1839 			   struct smc_buf_desc *rmb_desc)
1840 {
1841 	struct smc_llc_qentry *qentry = NULL;
1842 	struct smc_link *send_link;
1843 	int rc = 0;
1844 
1845 	send_link = smc_llc_usable_link(lgr);
1846 	if (!send_link)
1847 		return -ENOLINK;
1848 
1849 	/* protected by llc_flow control */
1850 	rc = smc_llc_send_delete_rkey(send_link, rmb_desc);
1851 	if (rc)
1852 		goto out;
1853 	/* receive DELETE RKEY response from server over RoCE fabric */
1854 	qentry = smc_llc_wait(lgr, send_link, SMC_LLC_WAIT_TIME,
1855 			      SMC_LLC_DELETE_RKEY);
1856 	if (!qentry || (qentry->msg.raw.hdr.flags & SMC_LLC_FLAG_RKEY_NEG))
1857 		rc = -EFAULT;
1858 out:
1859 	if (qentry)
1860 		smc_llc_flow_qentry_del(&lgr->llc_flow_lcl);
1861 	return rc;
1862 }
1863 
1864 void smc_llc_link_set_uid(struct smc_link *link)
1865 {
1866 	__be32 link_uid;
1867 
1868 	link_uid = htonl(*((u32 *)link->lgr->id) + link->link_id);
1869 	memcpy(link->link_uid, &link_uid, SMC_LGR_ID_SIZE);
1870 }
1871 
1872 /* save peers link user id, used for debug purposes */
1873 void smc_llc_save_peer_uid(struct smc_llc_qentry *qentry)
1874 {
1875 	memcpy(qentry->link->peer_link_uid, qentry->msg.confirm_link.link_uid,
1876 	       SMC_LGR_ID_SIZE);
1877 }
1878 
1879 /* evaluate confirm link request or response */
1880 int smc_llc_eval_conf_link(struct smc_llc_qentry *qentry,
1881 			   enum smc_llc_reqresp type)
1882 {
1883 	if (type == SMC_LLC_REQ) {	/* SMC server assigns link_id */
1884 		qentry->link->link_id = qentry->msg.confirm_link.link_num;
1885 		smc_llc_link_set_uid(qentry->link);
1886 	}
1887 	if (!(qentry->msg.raw.hdr.flags & SMC_LLC_FLAG_NO_RMBE_EYEC))
1888 		return -ENOTSUPP;
1889 	return 0;
1890 }
1891 
1892 /***************************** init, exit, misc ******************************/
1893 
1894 static struct smc_wr_rx_handler smc_llc_rx_handlers[] = {
1895 	{
1896 		.handler	= smc_llc_rx_handler,
1897 		.type		= SMC_LLC_CONFIRM_LINK
1898 	},
1899 	{
1900 		.handler	= smc_llc_rx_handler,
1901 		.type		= SMC_LLC_TEST_LINK
1902 	},
1903 	{
1904 		.handler	= smc_llc_rx_handler,
1905 		.type		= SMC_LLC_ADD_LINK
1906 	},
1907 	{
1908 		.handler	= smc_llc_rx_handler,
1909 		.type		= SMC_LLC_ADD_LINK_CONT
1910 	},
1911 	{
1912 		.handler	= smc_llc_rx_handler,
1913 		.type		= SMC_LLC_DELETE_LINK
1914 	},
1915 	{
1916 		.handler	= smc_llc_rx_handler,
1917 		.type		= SMC_LLC_CONFIRM_RKEY
1918 	},
1919 	{
1920 		.handler	= smc_llc_rx_handler,
1921 		.type		= SMC_LLC_CONFIRM_RKEY_CONT
1922 	},
1923 	{
1924 		.handler	= smc_llc_rx_handler,
1925 		.type		= SMC_LLC_DELETE_RKEY
1926 	},
1927 	{
1928 		.handler	= NULL,
1929 	}
1930 };
1931 
1932 int __init smc_llc_init(void)
1933 {
1934 	struct smc_wr_rx_handler *handler;
1935 	int rc = 0;
1936 
1937 	for (handler = smc_llc_rx_handlers; handler->handler; handler++) {
1938 		INIT_HLIST_NODE(&handler->list);
1939 		rc = smc_wr_rx_register_handler(handler);
1940 		if (rc)
1941 			break;
1942 	}
1943 	return rc;
1944 }
1945