xref: /openbmc/linux/net/smc/smc_core.c (revision b7b3c35e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  *  Basic Transport Functions exploiting Infiniband API
6  *
7  *  Copyright IBM Corp. 2016
8  *
9  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
10  */
11 
12 #include <linux/socket.h>
13 #include <linux/if_vlan.h>
14 #include <linux/random.h>
15 #include <linux/workqueue.h>
16 #include <linux/wait.h>
17 #include <linux/reboot.h>
18 #include <linux/mutex.h>
19 #include <linux/list.h>
20 #include <linux/smc.h>
21 #include <net/tcp.h>
22 #include <net/sock.h>
23 #include <rdma/ib_verbs.h>
24 #include <rdma/ib_cache.h>
25 
26 #include "smc.h"
27 #include "smc_clc.h"
28 #include "smc_core.h"
29 #include "smc_ib.h"
30 #include "smc_wr.h"
31 #include "smc_llc.h"
32 #include "smc_cdc.h"
33 #include "smc_close.h"
34 #include "smc_ism.h"
35 #include "smc_netlink.h"
36 #include "smc_stats.h"
37 #include "smc_tracepoint.h"
38 
39 #define SMC_LGR_NUM_INCR		256
40 #define SMC_LGR_FREE_DELAY_SERV		(600 * HZ)
41 #define SMC_LGR_FREE_DELAY_CLNT		(SMC_LGR_FREE_DELAY_SERV + 10 * HZ)
42 
43 struct smc_lgr_list smc_lgr_list = {	/* established link groups */
44 	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
45 	.list = LIST_HEAD_INIT(smc_lgr_list.list),
46 	.num = 0,
47 };
48 
49 static atomic_t lgr_cnt = ATOMIC_INIT(0); /* number of existing link groups */
50 static DECLARE_WAIT_QUEUE_HEAD(lgrs_deleted);
51 
52 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
53 			 struct smc_buf_desc *buf_desc);
54 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft);
55 
56 static void smc_link_down_work(struct work_struct *work);
57 
58 /* return head of link group list and its lock for a given link group */
59 static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr,
60 						  spinlock_t **lgr_lock)
61 {
62 	if (lgr->is_smcd) {
63 		*lgr_lock = &lgr->smcd->lgr_lock;
64 		return &lgr->smcd->lgr_list;
65 	}
66 
67 	*lgr_lock = &smc_lgr_list.lock;
68 	return &smc_lgr_list.list;
69 }
70 
71 static void smc_ibdev_cnt_inc(struct smc_link *lnk)
72 {
73 	atomic_inc(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
74 }
75 
76 static void smc_ibdev_cnt_dec(struct smc_link *lnk)
77 {
78 	atomic_dec(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
79 }
80 
81 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr)
82 {
83 	/* client link group creation always follows the server link group
84 	 * creation. For client use a somewhat higher removal delay time,
85 	 * otherwise there is a risk of out-of-sync link groups.
86 	 */
87 	if (!lgr->freeing) {
88 		mod_delayed_work(system_wq, &lgr->free_work,
89 				 (!lgr->is_smcd && lgr->role == SMC_CLNT) ?
90 						SMC_LGR_FREE_DELAY_CLNT :
91 						SMC_LGR_FREE_DELAY_SERV);
92 	}
93 }
94 
95 /* Register connection's alert token in our lookup structure.
96  * To use rbtrees we have to implement our own insert core.
97  * Requires @conns_lock
98  * @smc		connection to register
99  * Returns 0 on success, != otherwise.
100  */
101 static void smc_lgr_add_alert_token(struct smc_connection *conn)
102 {
103 	struct rb_node **link, *parent = NULL;
104 	u32 token = conn->alert_token_local;
105 
106 	link = &conn->lgr->conns_all.rb_node;
107 	while (*link) {
108 		struct smc_connection *cur = rb_entry(*link,
109 					struct smc_connection, alert_node);
110 
111 		parent = *link;
112 		if (cur->alert_token_local > token)
113 			link = &parent->rb_left;
114 		else
115 			link = &parent->rb_right;
116 	}
117 	/* Put the new node there */
118 	rb_link_node(&conn->alert_node, parent, link);
119 	rb_insert_color(&conn->alert_node, &conn->lgr->conns_all);
120 }
121 
122 /* assign an SMC-R link to the connection */
123 static int smcr_lgr_conn_assign_link(struct smc_connection *conn, bool first)
124 {
125 	enum smc_link_state expected = first ? SMC_LNK_ACTIVATING :
126 				       SMC_LNK_ACTIVE;
127 	int i, j;
128 
129 	/* do link balancing */
130 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
131 		struct smc_link *lnk = &conn->lgr->lnk[i];
132 
133 		if (lnk->state != expected || lnk->link_is_asym)
134 			continue;
135 		if (conn->lgr->role == SMC_CLNT) {
136 			conn->lnk = lnk; /* temporary, SMC server assigns link*/
137 			break;
138 		}
139 		if (conn->lgr->conns_num % 2) {
140 			for (j = i + 1; j < SMC_LINKS_PER_LGR_MAX; j++) {
141 				struct smc_link *lnk2;
142 
143 				lnk2 = &conn->lgr->lnk[j];
144 				if (lnk2->state == expected &&
145 				    !lnk2->link_is_asym) {
146 					conn->lnk = lnk2;
147 					break;
148 				}
149 			}
150 		}
151 		if (!conn->lnk)
152 			conn->lnk = lnk;
153 		break;
154 	}
155 	if (!conn->lnk)
156 		return SMC_CLC_DECL_NOACTLINK;
157 	atomic_inc(&conn->lnk->conn_cnt);
158 	return 0;
159 }
160 
161 /* Register connection in link group by assigning an alert token
162  * registered in a search tree.
163  * Requires @conns_lock
164  * Note that '0' is a reserved value and not assigned.
165  */
166 static int smc_lgr_register_conn(struct smc_connection *conn, bool first)
167 {
168 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
169 	static atomic_t nexttoken = ATOMIC_INIT(0);
170 	int rc;
171 
172 	if (!conn->lgr->is_smcd) {
173 		rc = smcr_lgr_conn_assign_link(conn, first);
174 		if (rc) {
175 			conn->lgr = NULL;
176 			return rc;
177 		}
178 	}
179 	/* find a new alert_token_local value not yet used by some connection
180 	 * in this link group
181 	 */
182 	sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */
183 	while (!conn->alert_token_local) {
184 		conn->alert_token_local = atomic_inc_return(&nexttoken);
185 		if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr))
186 			conn->alert_token_local = 0;
187 	}
188 	smc_lgr_add_alert_token(conn);
189 	conn->lgr->conns_num++;
190 	return 0;
191 }
192 
193 /* Unregister connection and reset the alert token of the given connection<
194  */
195 static void __smc_lgr_unregister_conn(struct smc_connection *conn)
196 {
197 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
198 	struct smc_link_group *lgr = conn->lgr;
199 
200 	rb_erase(&conn->alert_node, &lgr->conns_all);
201 	if (conn->lnk)
202 		atomic_dec(&conn->lnk->conn_cnt);
203 	lgr->conns_num--;
204 	conn->alert_token_local = 0;
205 	sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */
206 }
207 
208 /* Unregister connection from lgr
209  */
210 static void smc_lgr_unregister_conn(struct smc_connection *conn)
211 {
212 	struct smc_link_group *lgr = conn->lgr;
213 
214 	if (!smc_conn_lgr_valid(conn))
215 		return;
216 	write_lock_bh(&lgr->conns_lock);
217 	if (conn->alert_token_local) {
218 		__smc_lgr_unregister_conn(conn);
219 	}
220 	write_unlock_bh(&lgr->conns_lock);
221 }
222 
223 int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb)
224 {
225 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
226 	char hostname[SMC_MAX_HOSTNAME_LEN + 1];
227 	char smc_seid[SMC_MAX_EID_LEN + 1];
228 	struct nlattr *attrs;
229 	u8 *seid = NULL;
230 	u8 *host = NULL;
231 	void *nlh;
232 
233 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
234 			  &smc_gen_nl_family, NLM_F_MULTI,
235 			  SMC_NETLINK_GET_SYS_INFO);
236 	if (!nlh)
237 		goto errmsg;
238 	if (cb_ctx->pos[0])
239 		goto errout;
240 	attrs = nla_nest_start(skb, SMC_GEN_SYS_INFO);
241 	if (!attrs)
242 		goto errout;
243 	if (nla_put_u8(skb, SMC_NLA_SYS_VER, SMC_V2))
244 		goto errattr;
245 	if (nla_put_u8(skb, SMC_NLA_SYS_REL, SMC_RELEASE))
246 		goto errattr;
247 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_ISM_V2, smc_ism_is_v2_capable()))
248 		goto errattr;
249 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_SMCR_V2, true))
250 		goto errattr;
251 	smc_clc_get_hostname(&host);
252 	if (host) {
253 		memcpy(hostname, host, SMC_MAX_HOSTNAME_LEN);
254 		hostname[SMC_MAX_HOSTNAME_LEN] = 0;
255 		if (nla_put_string(skb, SMC_NLA_SYS_LOCAL_HOST, hostname))
256 			goto errattr;
257 	}
258 	if (smc_ism_is_v2_capable()) {
259 		smc_ism_get_system_eid(&seid);
260 		memcpy(smc_seid, seid, SMC_MAX_EID_LEN);
261 		smc_seid[SMC_MAX_EID_LEN] = 0;
262 		if (nla_put_string(skb, SMC_NLA_SYS_SEID, smc_seid))
263 			goto errattr;
264 	}
265 	nla_nest_end(skb, attrs);
266 	genlmsg_end(skb, nlh);
267 	cb_ctx->pos[0] = 1;
268 	return skb->len;
269 
270 errattr:
271 	nla_nest_cancel(skb, attrs);
272 errout:
273 	genlmsg_cancel(skb, nlh);
274 errmsg:
275 	return skb->len;
276 }
277 
278 /* Fill SMC_NLA_LGR_D_V2_COMMON/SMC_NLA_LGR_R_V2_COMMON nested attributes */
279 static int smc_nl_fill_lgr_v2_common(struct smc_link_group *lgr,
280 				     struct sk_buff *skb,
281 				     struct netlink_callback *cb,
282 				     struct nlattr *v2_attrs)
283 {
284 	char smc_host[SMC_MAX_HOSTNAME_LEN + 1];
285 	char smc_eid[SMC_MAX_EID_LEN + 1];
286 
287 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_VER, lgr->smc_version))
288 		goto errv2attr;
289 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_REL, lgr->peer_smc_release))
290 		goto errv2attr;
291 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_OS, lgr->peer_os))
292 		goto errv2attr;
293 	memcpy(smc_host, lgr->peer_hostname, SMC_MAX_HOSTNAME_LEN);
294 	smc_host[SMC_MAX_HOSTNAME_LEN] = 0;
295 	if (nla_put_string(skb, SMC_NLA_LGR_V2_PEER_HOST, smc_host))
296 		goto errv2attr;
297 	memcpy(smc_eid, lgr->negotiated_eid, SMC_MAX_EID_LEN);
298 	smc_eid[SMC_MAX_EID_LEN] = 0;
299 	if (nla_put_string(skb, SMC_NLA_LGR_V2_NEG_EID, smc_eid))
300 		goto errv2attr;
301 
302 	nla_nest_end(skb, v2_attrs);
303 	return 0;
304 
305 errv2attr:
306 	nla_nest_cancel(skb, v2_attrs);
307 	return -EMSGSIZE;
308 }
309 
310 static int smc_nl_fill_smcr_lgr_v2(struct smc_link_group *lgr,
311 				   struct sk_buff *skb,
312 				   struct netlink_callback *cb)
313 {
314 	struct nlattr *v2_attrs;
315 
316 	v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2);
317 	if (!v2_attrs)
318 		goto errattr;
319 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_DIRECT, !lgr->uses_gateway))
320 		goto errv2attr;
321 
322 	nla_nest_end(skb, v2_attrs);
323 	return 0;
324 
325 errv2attr:
326 	nla_nest_cancel(skb, v2_attrs);
327 errattr:
328 	return -EMSGSIZE;
329 }
330 
331 static int smc_nl_fill_lgr(struct smc_link_group *lgr,
332 			   struct sk_buff *skb,
333 			   struct netlink_callback *cb)
334 {
335 	char smc_target[SMC_MAX_PNETID_LEN + 1];
336 	struct nlattr *attrs, *v2_attrs;
337 
338 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCR);
339 	if (!attrs)
340 		goto errout;
341 
342 	if (nla_put_u32(skb, SMC_NLA_LGR_R_ID, *((u32 *)&lgr->id)))
343 		goto errattr;
344 	if (nla_put_u32(skb, SMC_NLA_LGR_R_CONNS_NUM, lgr->conns_num))
345 		goto errattr;
346 	if (nla_put_u8(skb, SMC_NLA_LGR_R_ROLE, lgr->role))
347 		goto errattr;
348 	if (nla_put_u8(skb, SMC_NLA_LGR_R_TYPE, lgr->type))
349 		goto errattr;
350 	if (nla_put_u8(skb, SMC_NLA_LGR_R_VLAN_ID, lgr->vlan_id))
351 		goto errattr;
352 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_R_NET_COOKIE,
353 			      lgr->net->net_cookie, SMC_NLA_LGR_R_PAD))
354 		goto errattr;
355 	memcpy(smc_target, lgr->pnet_id, SMC_MAX_PNETID_LEN);
356 	smc_target[SMC_MAX_PNETID_LEN] = 0;
357 	if (nla_put_string(skb, SMC_NLA_LGR_R_PNETID, smc_target))
358 		goto errattr;
359 	if (lgr->smc_version > SMC_V1) {
360 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2_COMMON);
361 		if (!v2_attrs)
362 			goto errattr;
363 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
364 			goto errattr;
365 		if (smc_nl_fill_smcr_lgr_v2(lgr, skb, cb))
366 			goto errattr;
367 	}
368 
369 	nla_nest_end(skb, attrs);
370 	return 0;
371 errattr:
372 	nla_nest_cancel(skb, attrs);
373 errout:
374 	return -EMSGSIZE;
375 }
376 
377 static int smc_nl_fill_lgr_link(struct smc_link_group *lgr,
378 				struct smc_link *link,
379 				struct sk_buff *skb,
380 				struct netlink_callback *cb)
381 {
382 	char smc_ibname[IB_DEVICE_NAME_MAX];
383 	u8 smc_gid_target[41];
384 	struct nlattr *attrs;
385 	u32 link_uid = 0;
386 	void *nlh;
387 
388 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
389 			  &smc_gen_nl_family, NLM_F_MULTI,
390 			  SMC_NETLINK_GET_LINK_SMCR);
391 	if (!nlh)
392 		goto errmsg;
393 
394 	attrs = nla_nest_start(skb, SMC_GEN_LINK_SMCR);
395 	if (!attrs)
396 		goto errout;
397 
398 	if (nla_put_u8(skb, SMC_NLA_LINK_ID, link->link_id))
399 		goto errattr;
400 	if (nla_put_u32(skb, SMC_NLA_LINK_STATE, link->state))
401 		goto errattr;
402 	if (nla_put_u32(skb, SMC_NLA_LINK_CONN_CNT,
403 			atomic_read(&link->conn_cnt)))
404 		goto errattr;
405 	if (nla_put_u8(skb, SMC_NLA_LINK_IB_PORT, link->ibport))
406 		goto errattr;
407 	if (nla_put_u32(skb, SMC_NLA_LINK_NET_DEV, link->ndev_ifidx))
408 		goto errattr;
409 	snprintf(smc_ibname, sizeof(smc_ibname), "%s", link->ibname);
410 	if (nla_put_string(skb, SMC_NLA_LINK_IB_DEV, smc_ibname))
411 		goto errattr;
412 	memcpy(&link_uid, link->link_uid, sizeof(link_uid));
413 	if (nla_put_u32(skb, SMC_NLA_LINK_UID, link_uid))
414 		goto errattr;
415 	memcpy(&link_uid, link->peer_link_uid, sizeof(link_uid));
416 	if (nla_put_u32(skb, SMC_NLA_LINK_PEER_UID, link_uid))
417 		goto errattr;
418 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
419 	smc_gid_be16_convert(smc_gid_target, link->gid);
420 	if (nla_put_string(skb, SMC_NLA_LINK_GID, smc_gid_target))
421 		goto errattr;
422 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
423 	smc_gid_be16_convert(smc_gid_target, link->peer_gid);
424 	if (nla_put_string(skb, SMC_NLA_LINK_PEER_GID, smc_gid_target))
425 		goto errattr;
426 
427 	nla_nest_end(skb, attrs);
428 	genlmsg_end(skb, nlh);
429 	return 0;
430 errattr:
431 	nla_nest_cancel(skb, attrs);
432 errout:
433 	genlmsg_cancel(skb, nlh);
434 errmsg:
435 	return -EMSGSIZE;
436 }
437 
438 static int smc_nl_handle_lgr(struct smc_link_group *lgr,
439 			     struct sk_buff *skb,
440 			     struct netlink_callback *cb,
441 			     bool list_links)
442 {
443 	void *nlh;
444 	int i;
445 
446 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
447 			  &smc_gen_nl_family, NLM_F_MULTI,
448 			  SMC_NETLINK_GET_LGR_SMCR);
449 	if (!nlh)
450 		goto errmsg;
451 	if (smc_nl_fill_lgr(lgr, skb, cb))
452 		goto errout;
453 
454 	genlmsg_end(skb, nlh);
455 	if (!list_links)
456 		goto out;
457 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
458 		if (!smc_link_usable(&lgr->lnk[i]))
459 			continue;
460 		if (smc_nl_fill_lgr_link(lgr, &lgr->lnk[i], skb, cb))
461 			goto errout;
462 	}
463 out:
464 	return 0;
465 
466 errout:
467 	genlmsg_cancel(skb, nlh);
468 errmsg:
469 	return -EMSGSIZE;
470 }
471 
472 static void smc_nl_fill_lgr_list(struct smc_lgr_list *smc_lgr,
473 				 struct sk_buff *skb,
474 				 struct netlink_callback *cb,
475 				 bool list_links)
476 {
477 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
478 	struct smc_link_group *lgr;
479 	int snum = cb_ctx->pos[0];
480 	int num = 0;
481 
482 	spin_lock_bh(&smc_lgr->lock);
483 	list_for_each_entry(lgr, &smc_lgr->list, list) {
484 		if (num < snum)
485 			goto next;
486 		if (smc_nl_handle_lgr(lgr, skb, cb, list_links))
487 			goto errout;
488 next:
489 		num++;
490 	}
491 errout:
492 	spin_unlock_bh(&smc_lgr->lock);
493 	cb_ctx->pos[0] = num;
494 }
495 
496 static int smc_nl_fill_smcd_lgr(struct smc_link_group *lgr,
497 				struct sk_buff *skb,
498 				struct netlink_callback *cb)
499 {
500 	char smc_pnet[SMC_MAX_PNETID_LEN + 1];
501 	struct nlattr *attrs;
502 	void *nlh;
503 
504 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
505 			  &smc_gen_nl_family, NLM_F_MULTI,
506 			  SMC_NETLINK_GET_LGR_SMCD);
507 	if (!nlh)
508 		goto errmsg;
509 
510 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCD);
511 	if (!attrs)
512 		goto errout;
513 
514 	if (nla_put_u32(skb, SMC_NLA_LGR_D_ID, *((u32 *)&lgr->id)))
515 		goto errattr;
516 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_GID, lgr->smcd->local_gid,
517 			      SMC_NLA_LGR_D_PAD))
518 		goto errattr;
519 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_GID, lgr->peer_gid,
520 			      SMC_NLA_LGR_D_PAD))
521 		goto errattr;
522 	if (nla_put_u8(skb, SMC_NLA_LGR_D_VLAN_ID, lgr->vlan_id))
523 		goto errattr;
524 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CONNS_NUM, lgr->conns_num))
525 		goto errattr;
526 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CHID, smc_ism_get_chid(lgr->smcd)))
527 		goto errattr;
528 	memcpy(smc_pnet, lgr->smcd->pnetid, SMC_MAX_PNETID_LEN);
529 	smc_pnet[SMC_MAX_PNETID_LEN] = 0;
530 	if (nla_put_string(skb, SMC_NLA_LGR_D_PNETID, smc_pnet))
531 		goto errattr;
532 	if (lgr->smc_version > SMC_V1) {
533 		struct nlattr *v2_attrs;
534 
535 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_D_V2_COMMON);
536 		if (!v2_attrs)
537 			goto errattr;
538 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
539 			goto errattr;
540 	}
541 	nla_nest_end(skb, attrs);
542 	genlmsg_end(skb, nlh);
543 	return 0;
544 
545 errattr:
546 	nla_nest_cancel(skb, attrs);
547 errout:
548 	genlmsg_cancel(skb, nlh);
549 errmsg:
550 	return -EMSGSIZE;
551 }
552 
553 static int smc_nl_handle_smcd_lgr(struct smcd_dev *dev,
554 				  struct sk_buff *skb,
555 				  struct netlink_callback *cb)
556 {
557 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
558 	struct smc_link_group *lgr;
559 	int snum = cb_ctx->pos[1];
560 	int rc = 0, num = 0;
561 
562 	spin_lock_bh(&dev->lgr_lock);
563 	list_for_each_entry(lgr, &dev->lgr_list, list) {
564 		if (!lgr->is_smcd)
565 			continue;
566 		if (num < snum)
567 			goto next;
568 		rc = smc_nl_fill_smcd_lgr(lgr, skb, cb);
569 		if (rc)
570 			goto errout;
571 next:
572 		num++;
573 	}
574 errout:
575 	spin_unlock_bh(&dev->lgr_lock);
576 	cb_ctx->pos[1] = num;
577 	return rc;
578 }
579 
580 static int smc_nl_fill_smcd_dev(struct smcd_dev_list *dev_list,
581 				struct sk_buff *skb,
582 				struct netlink_callback *cb)
583 {
584 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
585 	struct smcd_dev *smcd_dev;
586 	int snum = cb_ctx->pos[0];
587 	int rc = 0, num = 0;
588 
589 	mutex_lock(&dev_list->mutex);
590 	list_for_each_entry(smcd_dev, &dev_list->list, list) {
591 		if (list_empty(&smcd_dev->lgr_list))
592 			continue;
593 		if (num < snum)
594 			goto next;
595 		rc = smc_nl_handle_smcd_lgr(smcd_dev, skb, cb);
596 		if (rc)
597 			goto errout;
598 next:
599 		num++;
600 	}
601 errout:
602 	mutex_unlock(&dev_list->mutex);
603 	cb_ctx->pos[0] = num;
604 	return rc;
605 }
606 
607 int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
608 {
609 	bool list_links = false;
610 
611 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
612 	return skb->len;
613 }
614 
615 int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb)
616 {
617 	bool list_links = true;
618 
619 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
620 	return skb->len;
621 }
622 
623 int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
624 {
625 	smc_nl_fill_smcd_dev(&smcd_dev_list, skb, cb);
626 	return skb->len;
627 }
628 
629 void smc_lgr_cleanup_early(struct smc_link_group *lgr)
630 {
631 	spinlock_t *lgr_lock;
632 
633 	if (!lgr)
634 		return;
635 
636 	smc_lgr_list_head(lgr, &lgr_lock);
637 	spin_lock_bh(lgr_lock);
638 	/* do not use this link group for new connections */
639 	if (!list_empty(&lgr->list))
640 		list_del_init(&lgr->list);
641 	spin_unlock_bh(lgr_lock);
642 	__smc_lgr_terminate(lgr, true);
643 }
644 
645 static void smcr_lgr_link_deactivate_all(struct smc_link_group *lgr)
646 {
647 	int i;
648 
649 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
650 		struct smc_link *lnk = &lgr->lnk[i];
651 
652 		if (smc_link_sendable(lnk))
653 			lnk->state = SMC_LNK_INACTIVE;
654 	}
655 	wake_up_all(&lgr->llc_msg_waiter);
656 	wake_up_all(&lgr->llc_flow_waiter);
657 }
658 
659 static void smc_lgr_free(struct smc_link_group *lgr);
660 
661 static void smc_lgr_free_work(struct work_struct *work)
662 {
663 	struct smc_link_group *lgr = container_of(to_delayed_work(work),
664 						  struct smc_link_group,
665 						  free_work);
666 	spinlock_t *lgr_lock;
667 	bool conns;
668 
669 	smc_lgr_list_head(lgr, &lgr_lock);
670 	spin_lock_bh(lgr_lock);
671 	if (lgr->freeing) {
672 		spin_unlock_bh(lgr_lock);
673 		return;
674 	}
675 	read_lock_bh(&lgr->conns_lock);
676 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
677 	read_unlock_bh(&lgr->conns_lock);
678 	if (!conns) { /* number of lgr connections is no longer zero */
679 		spin_unlock_bh(lgr_lock);
680 		return;
681 	}
682 	list_del_init(&lgr->list); /* remove from smc_lgr_list */
683 	lgr->freeing = 1; /* this instance does the freeing, no new schedule */
684 	spin_unlock_bh(lgr_lock);
685 	cancel_delayed_work(&lgr->free_work);
686 
687 	if (!lgr->is_smcd && !lgr->terminating)
688 		smc_llc_send_link_delete_all(lgr, true,
689 					     SMC_LLC_DEL_PROG_INIT_TERM);
690 	if (lgr->is_smcd && !lgr->terminating)
691 		smc_ism_signal_shutdown(lgr);
692 	if (!lgr->is_smcd)
693 		smcr_lgr_link_deactivate_all(lgr);
694 	smc_lgr_free(lgr);
695 }
696 
697 static void smc_lgr_terminate_work(struct work_struct *work)
698 {
699 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
700 						  terminate_work);
701 
702 	__smc_lgr_terminate(lgr, true);
703 }
704 
705 /* return next unique link id for the lgr */
706 static u8 smcr_next_link_id(struct smc_link_group *lgr)
707 {
708 	u8 link_id;
709 	int i;
710 
711 	while (1) {
712 again:
713 		link_id = ++lgr->next_link_id;
714 		if (!link_id)	/* skip zero as link_id */
715 			link_id = ++lgr->next_link_id;
716 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
717 			if (smc_link_usable(&lgr->lnk[i]) &&
718 			    lgr->lnk[i].link_id == link_id)
719 				goto again;
720 		}
721 		break;
722 	}
723 	return link_id;
724 }
725 
726 static void smcr_copy_dev_info_to_link(struct smc_link *link)
727 {
728 	struct smc_ib_device *smcibdev = link->smcibdev;
729 
730 	snprintf(link->ibname, sizeof(link->ibname), "%s",
731 		 smcibdev->ibdev->name);
732 	link->ndev_ifidx = smcibdev->ndev_ifidx[link->ibport - 1];
733 }
734 
735 int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk,
736 		   u8 link_idx, struct smc_init_info *ini)
737 {
738 	struct smc_ib_device *smcibdev;
739 	u8 rndvec[3];
740 	int rc;
741 
742 	if (lgr->smc_version == SMC_V2) {
743 		lnk->smcibdev = ini->smcrv2.ib_dev_v2;
744 		lnk->ibport = ini->smcrv2.ib_port_v2;
745 	} else {
746 		lnk->smcibdev = ini->ib_dev;
747 		lnk->ibport = ini->ib_port;
748 	}
749 	get_device(&lnk->smcibdev->ibdev->dev);
750 	atomic_inc(&lnk->smcibdev->lnk_cnt);
751 	refcount_set(&lnk->refcnt, 1); /* link refcnt is set to 1 */
752 	lnk->clearing = 0;
753 	lnk->path_mtu = lnk->smcibdev->pattr[lnk->ibport - 1].active_mtu;
754 	lnk->link_id = smcr_next_link_id(lgr);
755 	lnk->lgr = lgr;
756 	smc_lgr_hold(lgr); /* lgr_put in smcr_link_clear() */
757 	lnk->link_idx = link_idx;
758 	smc_ibdev_cnt_inc(lnk);
759 	smcr_copy_dev_info_to_link(lnk);
760 	atomic_set(&lnk->conn_cnt, 0);
761 	smc_llc_link_set_uid(lnk);
762 	INIT_WORK(&lnk->link_down_wrk, smc_link_down_work);
763 	if (!lnk->smcibdev->initialized) {
764 		rc = (int)smc_ib_setup_per_ibdev(lnk->smcibdev);
765 		if (rc)
766 			goto out;
767 	}
768 	get_random_bytes(rndvec, sizeof(rndvec));
769 	lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) +
770 		(rndvec[2] << 16);
771 	rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport,
772 				  ini->vlan_id, lnk->gid, &lnk->sgid_index,
773 				  lgr->smc_version == SMC_V2 ?
774 						  &ini->smcrv2 : NULL);
775 	if (rc)
776 		goto out;
777 	rc = smc_llc_link_init(lnk);
778 	if (rc)
779 		goto out;
780 	rc = smc_wr_alloc_link_mem(lnk);
781 	if (rc)
782 		goto clear_llc_lnk;
783 	rc = smc_ib_create_protection_domain(lnk);
784 	if (rc)
785 		goto free_link_mem;
786 	rc = smc_ib_create_queue_pair(lnk);
787 	if (rc)
788 		goto dealloc_pd;
789 	rc = smc_wr_create_link(lnk);
790 	if (rc)
791 		goto destroy_qp;
792 	lnk->state = SMC_LNK_ACTIVATING;
793 	return 0;
794 
795 destroy_qp:
796 	smc_ib_destroy_queue_pair(lnk);
797 dealloc_pd:
798 	smc_ib_dealloc_protection_domain(lnk);
799 free_link_mem:
800 	smc_wr_free_link_mem(lnk);
801 clear_llc_lnk:
802 	smc_llc_link_clear(lnk, false);
803 out:
804 	smc_ibdev_cnt_dec(lnk);
805 	put_device(&lnk->smcibdev->ibdev->dev);
806 	smcibdev = lnk->smcibdev;
807 	memset(lnk, 0, sizeof(struct smc_link));
808 	lnk->state = SMC_LNK_UNUSED;
809 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
810 		wake_up(&smcibdev->lnks_deleted);
811 	smc_lgr_put(lgr); /* lgr_hold above */
812 	return rc;
813 }
814 
815 /* create a new SMC link group */
816 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
817 {
818 	struct smc_link_group *lgr;
819 	struct list_head *lgr_list;
820 	struct smc_link *lnk;
821 	spinlock_t *lgr_lock;
822 	u8 link_idx;
823 	int rc = 0;
824 	int i;
825 
826 	if (ini->is_smcd && ini->vlan_id) {
827 		if (smc_ism_get_vlan(ini->ism_dev[ini->ism_selected],
828 				     ini->vlan_id)) {
829 			rc = SMC_CLC_DECL_ISMVLANERR;
830 			goto out;
831 		}
832 	}
833 
834 	lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
835 	if (!lgr) {
836 		rc = SMC_CLC_DECL_MEM;
837 		goto ism_put_vlan;
838 	}
839 	lgr->tx_wq = alloc_workqueue("smc_tx_wq-%*phN", 0, 0,
840 				     SMC_LGR_ID_SIZE, &lgr->id);
841 	if (!lgr->tx_wq) {
842 		rc = -ENOMEM;
843 		goto free_lgr;
844 	}
845 	lgr->is_smcd = ini->is_smcd;
846 	lgr->sync_err = 0;
847 	lgr->terminating = 0;
848 	lgr->freeing = 0;
849 	lgr->vlan_id = ini->vlan_id;
850 	refcount_set(&lgr->refcnt, 1); /* set lgr refcnt to 1 */
851 	mutex_init(&lgr->sndbufs_lock);
852 	mutex_init(&lgr->rmbs_lock);
853 	rwlock_init(&lgr->conns_lock);
854 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
855 		INIT_LIST_HEAD(&lgr->sndbufs[i]);
856 		INIT_LIST_HEAD(&lgr->rmbs[i]);
857 	}
858 	lgr->next_link_id = 0;
859 	smc_lgr_list.num += SMC_LGR_NUM_INCR;
860 	memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE);
861 	INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
862 	INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work);
863 	lgr->conns_all = RB_ROOT;
864 	if (ini->is_smcd) {
865 		/* SMC-D specific settings */
866 		get_device(&ini->ism_dev[ini->ism_selected]->dev);
867 		lgr->peer_gid = ini->ism_peer_gid[ini->ism_selected];
868 		lgr->smcd = ini->ism_dev[ini->ism_selected];
869 		lgr_list = &ini->ism_dev[ini->ism_selected]->lgr_list;
870 		lgr_lock = &lgr->smcd->lgr_lock;
871 		lgr->smc_version = ini->smcd_version;
872 		lgr->peer_shutdown = 0;
873 		atomic_inc(&ini->ism_dev[ini->ism_selected]->lgr_cnt);
874 	} else {
875 		/* SMC-R specific settings */
876 		struct smc_ib_device *ibdev;
877 		int ibport;
878 
879 		lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
880 		lgr->smc_version = ini->smcr_version;
881 		memcpy(lgr->peer_systemid, ini->peer_systemid,
882 		       SMC_SYSTEMID_LEN);
883 		if (lgr->smc_version == SMC_V2) {
884 			ibdev = ini->smcrv2.ib_dev_v2;
885 			ibport = ini->smcrv2.ib_port_v2;
886 			lgr->saddr = ini->smcrv2.saddr;
887 			lgr->uses_gateway = ini->smcrv2.uses_gateway;
888 			memcpy(lgr->nexthop_mac, ini->smcrv2.nexthop_mac,
889 			       ETH_ALEN);
890 		} else {
891 			ibdev = ini->ib_dev;
892 			ibport = ini->ib_port;
893 		}
894 		memcpy(lgr->pnet_id, ibdev->pnetid[ibport - 1],
895 		       SMC_MAX_PNETID_LEN);
896 		if (smc_wr_alloc_lgr_mem(lgr))
897 			goto free_wq;
898 		smc_llc_lgr_init(lgr, smc);
899 
900 		link_idx = SMC_SINGLE_LINK;
901 		lnk = &lgr->lnk[link_idx];
902 		rc = smcr_link_init(lgr, lnk, link_idx, ini);
903 		if (rc) {
904 			smc_wr_free_lgr_mem(lgr);
905 			goto free_wq;
906 		}
907 		lgr->net = smc_ib_net(lnk->smcibdev);
908 		lgr_list = &smc_lgr_list.list;
909 		lgr_lock = &smc_lgr_list.lock;
910 		atomic_inc(&lgr_cnt);
911 	}
912 	smc->conn.lgr = lgr;
913 	spin_lock_bh(lgr_lock);
914 	list_add_tail(&lgr->list, lgr_list);
915 	spin_unlock_bh(lgr_lock);
916 	return 0;
917 
918 free_wq:
919 	destroy_workqueue(lgr->tx_wq);
920 free_lgr:
921 	kfree(lgr);
922 ism_put_vlan:
923 	if (ini->is_smcd && ini->vlan_id)
924 		smc_ism_put_vlan(ini->ism_dev[ini->ism_selected], ini->vlan_id);
925 out:
926 	if (rc < 0) {
927 		if (rc == -ENOMEM)
928 			rc = SMC_CLC_DECL_MEM;
929 		else
930 			rc = SMC_CLC_DECL_INTERR;
931 	}
932 	return rc;
933 }
934 
935 static int smc_write_space(struct smc_connection *conn)
936 {
937 	int buffer_len = conn->peer_rmbe_size;
938 	union smc_host_cursor prod;
939 	union smc_host_cursor cons;
940 	int space;
941 
942 	smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
943 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
944 	/* determine rx_buf space */
945 	space = buffer_len - smc_curs_diff(buffer_len, &cons, &prod);
946 	return space;
947 }
948 
949 static int smc_switch_cursor(struct smc_sock *smc, struct smc_cdc_tx_pend *pend,
950 			     struct smc_wr_buf *wr_buf)
951 {
952 	struct smc_connection *conn = &smc->conn;
953 	union smc_host_cursor cons, fin;
954 	int rc = 0;
955 	int diff;
956 
957 	smc_curs_copy(&conn->tx_curs_sent, &conn->tx_curs_fin, conn);
958 	smc_curs_copy(&fin, &conn->local_tx_ctrl_fin, conn);
959 	/* set prod cursor to old state, enforce tx_rdma_writes() */
960 	smc_curs_copy(&conn->local_tx_ctrl.prod, &fin, conn);
961 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
962 
963 	if (smc_curs_comp(conn->peer_rmbe_size, &cons, &fin) < 0) {
964 		/* cons cursor advanced more than fin, and prod was set
965 		 * fin above, so now prod is smaller than cons. Fix that.
966 		 */
967 		diff = smc_curs_diff(conn->peer_rmbe_size, &fin, &cons);
968 		smc_curs_add(conn->sndbuf_desc->len,
969 			     &conn->tx_curs_sent, diff);
970 		smc_curs_add(conn->sndbuf_desc->len,
971 			     &conn->tx_curs_fin, diff);
972 
973 		smp_mb__before_atomic();
974 		atomic_add(diff, &conn->sndbuf_space);
975 		smp_mb__after_atomic();
976 
977 		smc_curs_add(conn->peer_rmbe_size,
978 			     &conn->local_tx_ctrl.prod, diff);
979 		smc_curs_add(conn->peer_rmbe_size,
980 			     &conn->local_tx_ctrl_fin, diff);
981 	}
982 	/* recalculate, value is used by tx_rdma_writes() */
983 	atomic_set(&smc->conn.peer_rmbe_space, smc_write_space(conn));
984 
985 	if (smc->sk.sk_state != SMC_INIT &&
986 	    smc->sk.sk_state != SMC_CLOSED) {
987 		rc = smcr_cdc_msg_send_validation(conn, pend, wr_buf);
988 		if (!rc) {
989 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work, 0);
990 			smc->sk.sk_data_ready(&smc->sk);
991 		}
992 	} else {
993 		smc_wr_tx_put_slot(conn->lnk,
994 				   (struct smc_wr_tx_pend_priv *)pend);
995 	}
996 	return rc;
997 }
998 
999 void smc_switch_link_and_count(struct smc_connection *conn,
1000 			       struct smc_link *to_lnk)
1001 {
1002 	atomic_dec(&conn->lnk->conn_cnt);
1003 	/* link_hold in smc_conn_create() */
1004 	smcr_link_put(conn->lnk);
1005 	conn->lnk = to_lnk;
1006 	atomic_inc(&conn->lnk->conn_cnt);
1007 	/* link_put in smc_conn_free() */
1008 	smcr_link_hold(conn->lnk);
1009 }
1010 
1011 struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
1012 				  struct smc_link *from_lnk, bool is_dev_err)
1013 {
1014 	struct smc_link *to_lnk = NULL;
1015 	struct smc_cdc_tx_pend *pend;
1016 	struct smc_connection *conn;
1017 	struct smc_wr_buf *wr_buf;
1018 	struct smc_sock *smc;
1019 	struct rb_node *node;
1020 	int i, rc = 0;
1021 
1022 	/* link is inactive, wake up tx waiters */
1023 	smc_wr_wakeup_tx_wait(from_lnk);
1024 
1025 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1026 		if (!smc_link_active(&lgr->lnk[i]) || i == from_lnk->link_idx)
1027 			continue;
1028 		if (is_dev_err && from_lnk->smcibdev == lgr->lnk[i].smcibdev &&
1029 		    from_lnk->ibport == lgr->lnk[i].ibport) {
1030 			continue;
1031 		}
1032 		to_lnk = &lgr->lnk[i];
1033 		break;
1034 	}
1035 	if (!to_lnk || !smc_wr_tx_link_hold(to_lnk)) {
1036 		smc_lgr_terminate_sched(lgr);
1037 		return NULL;
1038 	}
1039 again:
1040 	read_lock_bh(&lgr->conns_lock);
1041 	for (node = rb_first(&lgr->conns_all); node; node = rb_next(node)) {
1042 		conn = rb_entry(node, struct smc_connection, alert_node);
1043 		if (conn->lnk != from_lnk)
1044 			continue;
1045 		smc = container_of(conn, struct smc_sock, conn);
1046 		/* conn->lnk not yet set in SMC_INIT state */
1047 		if (smc->sk.sk_state == SMC_INIT)
1048 			continue;
1049 		if (smc->sk.sk_state == SMC_CLOSED ||
1050 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT1 ||
1051 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT2 ||
1052 		    smc->sk.sk_state == SMC_APPFINCLOSEWAIT ||
1053 		    smc->sk.sk_state == SMC_APPCLOSEWAIT1 ||
1054 		    smc->sk.sk_state == SMC_APPCLOSEWAIT2 ||
1055 		    smc->sk.sk_state == SMC_PEERFINCLOSEWAIT ||
1056 		    smc->sk.sk_state == SMC_PEERABORTWAIT ||
1057 		    smc->sk.sk_state == SMC_PROCESSABORT) {
1058 			spin_lock_bh(&conn->send_lock);
1059 			smc_switch_link_and_count(conn, to_lnk);
1060 			spin_unlock_bh(&conn->send_lock);
1061 			continue;
1062 		}
1063 		sock_hold(&smc->sk);
1064 		read_unlock_bh(&lgr->conns_lock);
1065 		/* pre-fetch buffer outside of send_lock, might sleep */
1066 		rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
1067 		if (rc)
1068 			goto err_out;
1069 		/* avoid race with smcr_tx_sndbuf_nonempty() */
1070 		spin_lock_bh(&conn->send_lock);
1071 		smc_switch_link_and_count(conn, to_lnk);
1072 		rc = smc_switch_cursor(smc, pend, wr_buf);
1073 		spin_unlock_bh(&conn->send_lock);
1074 		sock_put(&smc->sk);
1075 		if (rc)
1076 			goto err_out;
1077 		goto again;
1078 	}
1079 	read_unlock_bh(&lgr->conns_lock);
1080 	smc_wr_tx_link_put(to_lnk);
1081 	return to_lnk;
1082 
1083 err_out:
1084 	smcr_link_down_cond_sched(to_lnk);
1085 	smc_wr_tx_link_put(to_lnk);
1086 	return NULL;
1087 }
1088 
1089 static void smcr_buf_unuse(struct smc_buf_desc *rmb_desc,
1090 			   struct smc_link_group *lgr)
1091 {
1092 	int rc;
1093 
1094 	if (rmb_desc->is_conf_rkey && !list_empty(&lgr->list)) {
1095 		/* unregister rmb with peer */
1096 		rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
1097 		if (!rc) {
1098 			/* protect against smc_llc_cli_rkey_exchange() */
1099 			mutex_lock(&lgr->llc_conf_mutex);
1100 			smc_llc_do_delete_rkey(lgr, rmb_desc);
1101 			rmb_desc->is_conf_rkey = false;
1102 			mutex_unlock(&lgr->llc_conf_mutex);
1103 			smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1104 		}
1105 	}
1106 
1107 	if (rmb_desc->is_reg_err) {
1108 		/* buf registration failed, reuse not possible */
1109 		mutex_lock(&lgr->rmbs_lock);
1110 		list_del(&rmb_desc->list);
1111 		mutex_unlock(&lgr->rmbs_lock);
1112 
1113 		smc_buf_free(lgr, true, rmb_desc);
1114 	} else {
1115 		rmb_desc->used = 0;
1116 		memset(rmb_desc->cpu_addr, 0, rmb_desc->len);
1117 	}
1118 }
1119 
1120 static void smc_buf_unuse(struct smc_connection *conn,
1121 			  struct smc_link_group *lgr)
1122 {
1123 	if (conn->sndbuf_desc) {
1124 		conn->sndbuf_desc->used = 0;
1125 		memset(conn->sndbuf_desc->cpu_addr, 0, conn->sndbuf_desc->len);
1126 	}
1127 	if (conn->rmb_desc && lgr->is_smcd) {
1128 		conn->rmb_desc->used = 0;
1129 		memset(conn->rmb_desc->cpu_addr, 0, conn->rmb_desc->len +
1130 		       sizeof(struct smcd_cdc_msg));
1131 	} else if (conn->rmb_desc) {
1132 		smcr_buf_unuse(conn->rmb_desc, lgr);
1133 	}
1134 }
1135 
1136 /* remove a finished connection from its link group */
1137 void smc_conn_free(struct smc_connection *conn)
1138 {
1139 	struct smc_link_group *lgr = conn->lgr;
1140 
1141 	if (!lgr || conn->freed)
1142 		/* Connection has never been registered in a
1143 		 * link group, or has already been freed.
1144 		 */
1145 		return;
1146 
1147 	conn->freed = 1;
1148 	if (!smc_conn_lgr_valid(conn))
1149 		/* Connection has already unregistered from
1150 		 * link group.
1151 		 */
1152 		goto lgr_put;
1153 
1154 	if (lgr->is_smcd) {
1155 		if (!list_empty(&lgr->list))
1156 			smc_ism_unset_conn(conn);
1157 		tasklet_kill(&conn->rx_tsklet);
1158 	} else {
1159 		smc_cdc_wait_pend_tx_wr(conn);
1160 		if (current_work() != &conn->abort_work)
1161 			cancel_work_sync(&conn->abort_work);
1162 	}
1163 	if (!list_empty(&lgr->list)) {
1164 		smc_lgr_unregister_conn(conn);
1165 		smc_buf_unuse(conn, lgr); /* allow buffer reuse */
1166 	}
1167 
1168 	if (!lgr->conns_num)
1169 		smc_lgr_schedule_free_work(lgr);
1170 lgr_put:
1171 	if (!lgr->is_smcd)
1172 		smcr_link_put(conn->lnk); /* link_hold in smc_conn_create() */
1173 	smc_lgr_put(lgr); /* lgr_hold in smc_conn_create() */
1174 }
1175 
1176 /* unregister a link from a buf_desc */
1177 static void smcr_buf_unmap_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1178 				struct smc_link *lnk)
1179 {
1180 	if (is_rmb)
1181 		buf_desc->is_reg_mr[lnk->link_idx] = false;
1182 	if (!buf_desc->is_map_ib[lnk->link_idx])
1183 		return;
1184 	if (is_rmb) {
1185 		if (buf_desc->mr_rx[lnk->link_idx]) {
1186 			smc_ib_put_memory_region(
1187 					buf_desc->mr_rx[lnk->link_idx]);
1188 			buf_desc->mr_rx[lnk->link_idx] = NULL;
1189 		}
1190 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_FROM_DEVICE);
1191 	} else {
1192 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_TO_DEVICE);
1193 	}
1194 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1195 	buf_desc->is_map_ib[lnk->link_idx] = false;
1196 }
1197 
1198 /* unmap all buffers of lgr for a deleted link */
1199 static void smcr_buf_unmap_lgr(struct smc_link *lnk)
1200 {
1201 	struct smc_link_group *lgr = lnk->lgr;
1202 	struct smc_buf_desc *buf_desc, *bf;
1203 	int i;
1204 
1205 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1206 		mutex_lock(&lgr->rmbs_lock);
1207 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list)
1208 			smcr_buf_unmap_link(buf_desc, true, lnk);
1209 		mutex_unlock(&lgr->rmbs_lock);
1210 		mutex_lock(&lgr->sndbufs_lock);
1211 		list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i],
1212 					 list)
1213 			smcr_buf_unmap_link(buf_desc, false, lnk);
1214 		mutex_unlock(&lgr->sndbufs_lock);
1215 	}
1216 }
1217 
1218 static void smcr_rtoken_clear_link(struct smc_link *lnk)
1219 {
1220 	struct smc_link_group *lgr = lnk->lgr;
1221 	int i;
1222 
1223 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
1224 		lgr->rtokens[i][lnk->link_idx].rkey = 0;
1225 		lgr->rtokens[i][lnk->link_idx].dma_addr = 0;
1226 	}
1227 }
1228 
1229 static void __smcr_link_clear(struct smc_link *lnk)
1230 {
1231 	struct smc_link_group *lgr = lnk->lgr;
1232 	struct smc_ib_device *smcibdev;
1233 
1234 	smc_wr_free_link_mem(lnk);
1235 	smc_ibdev_cnt_dec(lnk);
1236 	put_device(&lnk->smcibdev->ibdev->dev);
1237 	smcibdev = lnk->smcibdev;
1238 	memset(lnk, 0, sizeof(struct smc_link));
1239 	lnk->state = SMC_LNK_UNUSED;
1240 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
1241 		wake_up(&smcibdev->lnks_deleted);
1242 	smc_lgr_put(lgr); /* lgr_hold in smcr_link_init() */
1243 }
1244 
1245 /* must be called under lgr->llc_conf_mutex lock */
1246 void smcr_link_clear(struct smc_link *lnk, bool log)
1247 {
1248 	if (!lnk->lgr || lnk->clearing ||
1249 	    lnk->state == SMC_LNK_UNUSED)
1250 		return;
1251 	lnk->clearing = 1;
1252 	lnk->peer_qpn = 0;
1253 	smc_llc_link_clear(lnk, log);
1254 	smcr_buf_unmap_lgr(lnk);
1255 	smcr_rtoken_clear_link(lnk);
1256 	smc_ib_modify_qp_error(lnk);
1257 	smc_wr_free_link(lnk);
1258 	smc_ib_destroy_queue_pair(lnk);
1259 	smc_ib_dealloc_protection_domain(lnk);
1260 	smcr_link_put(lnk); /* theoretically last link_put */
1261 }
1262 
1263 void smcr_link_hold(struct smc_link *lnk)
1264 {
1265 	refcount_inc(&lnk->refcnt);
1266 }
1267 
1268 void smcr_link_put(struct smc_link *lnk)
1269 {
1270 	if (refcount_dec_and_test(&lnk->refcnt))
1271 		__smcr_link_clear(lnk);
1272 }
1273 
1274 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb,
1275 			  struct smc_buf_desc *buf_desc)
1276 {
1277 	int i;
1278 
1279 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1280 		smcr_buf_unmap_link(buf_desc, is_rmb, &lgr->lnk[i]);
1281 
1282 	if (buf_desc->pages)
1283 		__free_pages(buf_desc->pages, buf_desc->order);
1284 	kfree(buf_desc);
1285 }
1286 
1287 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb,
1288 			  struct smc_buf_desc *buf_desc)
1289 {
1290 	if (is_dmb) {
1291 		/* restore original buf len */
1292 		buf_desc->len += sizeof(struct smcd_cdc_msg);
1293 		smc_ism_unregister_dmb(lgr->smcd, buf_desc);
1294 	} else {
1295 		kfree(buf_desc->cpu_addr);
1296 	}
1297 	kfree(buf_desc);
1298 }
1299 
1300 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
1301 			 struct smc_buf_desc *buf_desc)
1302 {
1303 	if (lgr->is_smcd)
1304 		smcd_buf_free(lgr, is_rmb, buf_desc);
1305 	else
1306 		smcr_buf_free(lgr, is_rmb, buf_desc);
1307 }
1308 
1309 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb)
1310 {
1311 	struct smc_buf_desc *buf_desc, *bf_desc;
1312 	struct list_head *buf_list;
1313 	int i;
1314 
1315 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1316 		if (is_rmb)
1317 			buf_list = &lgr->rmbs[i];
1318 		else
1319 			buf_list = &lgr->sndbufs[i];
1320 		list_for_each_entry_safe(buf_desc, bf_desc, buf_list,
1321 					 list) {
1322 			list_del(&buf_desc->list);
1323 			smc_buf_free(lgr, is_rmb, buf_desc);
1324 		}
1325 	}
1326 }
1327 
1328 static void smc_lgr_free_bufs(struct smc_link_group *lgr)
1329 {
1330 	/* free send buffers */
1331 	__smc_lgr_free_bufs(lgr, false);
1332 	/* free rmbs */
1333 	__smc_lgr_free_bufs(lgr, true);
1334 }
1335 
1336 /* won't be freed until no one accesses to lgr anymore */
1337 static void __smc_lgr_free(struct smc_link_group *lgr)
1338 {
1339 	smc_lgr_free_bufs(lgr);
1340 	if (lgr->is_smcd) {
1341 		if (!atomic_dec_return(&lgr->smcd->lgr_cnt))
1342 			wake_up(&lgr->smcd->lgrs_deleted);
1343 	} else {
1344 		smc_wr_free_lgr_mem(lgr);
1345 		if (!atomic_dec_return(&lgr_cnt))
1346 			wake_up(&lgrs_deleted);
1347 	}
1348 	kfree(lgr);
1349 }
1350 
1351 /* remove a link group */
1352 static void smc_lgr_free(struct smc_link_group *lgr)
1353 {
1354 	int i;
1355 
1356 	if (!lgr->is_smcd) {
1357 		mutex_lock(&lgr->llc_conf_mutex);
1358 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1359 			if (lgr->lnk[i].state != SMC_LNK_UNUSED)
1360 				smcr_link_clear(&lgr->lnk[i], false);
1361 		}
1362 		mutex_unlock(&lgr->llc_conf_mutex);
1363 		smc_llc_lgr_clear(lgr);
1364 	}
1365 
1366 	destroy_workqueue(lgr->tx_wq);
1367 	if (lgr->is_smcd) {
1368 		smc_ism_put_vlan(lgr->smcd, lgr->vlan_id);
1369 		put_device(&lgr->smcd->dev);
1370 	}
1371 	smc_lgr_put(lgr); /* theoretically last lgr_put */
1372 }
1373 
1374 void smc_lgr_hold(struct smc_link_group *lgr)
1375 {
1376 	refcount_inc(&lgr->refcnt);
1377 }
1378 
1379 void smc_lgr_put(struct smc_link_group *lgr)
1380 {
1381 	if (refcount_dec_and_test(&lgr->refcnt))
1382 		__smc_lgr_free(lgr);
1383 }
1384 
1385 static void smc_sk_wake_ups(struct smc_sock *smc)
1386 {
1387 	smc->sk.sk_write_space(&smc->sk);
1388 	smc->sk.sk_data_ready(&smc->sk);
1389 	smc->sk.sk_state_change(&smc->sk);
1390 }
1391 
1392 /* kill a connection */
1393 static void smc_conn_kill(struct smc_connection *conn, bool soft)
1394 {
1395 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1396 
1397 	if (conn->lgr->is_smcd && conn->lgr->peer_shutdown)
1398 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
1399 	else
1400 		smc_close_abort(conn);
1401 	conn->killed = 1;
1402 	smc->sk.sk_err = ECONNABORTED;
1403 	smc_sk_wake_ups(smc);
1404 	if (conn->lgr->is_smcd) {
1405 		smc_ism_unset_conn(conn);
1406 		if (soft)
1407 			tasklet_kill(&conn->rx_tsklet);
1408 		else
1409 			tasklet_unlock_wait(&conn->rx_tsklet);
1410 	} else {
1411 		smc_cdc_wait_pend_tx_wr(conn);
1412 	}
1413 	smc_lgr_unregister_conn(conn);
1414 	smc_close_active_abort(smc);
1415 }
1416 
1417 static void smc_lgr_cleanup(struct smc_link_group *lgr)
1418 {
1419 	if (lgr->is_smcd) {
1420 		smc_ism_signal_shutdown(lgr);
1421 	} else {
1422 		u32 rsn = lgr->llc_termination_rsn;
1423 
1424 		if (!rsn)
1425 			rsn = SMC_LLC_DEL_PROG_INIT_TERM;
1426 		smc_llc_send_link_delete_all(lgr, false, rsn);
1427 		smcr_lgr_link_deactivate_all(lgr);
1428 	}
1429 }
1430 
1431 /* terminate link group
1432  * @soft: true if link group shutdown can take its time
1433  *	  false if immediate link group shutdown is required
1434  */
1435 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft)
1436 {
1437 	struct smc_connection *conn;
1438 	struct smc_sock *smc;
1439 	struct rb_node *node;
1440 
1441 	if (lgr->terminating)
1442 		return;	/* lgr already terminating */
1443 	/* cancel free_work sync, will terminate when lgr->freeing is set */
1444 	cancel_delayed_work_sync(&lgr->free_work);
1445 	lgr->terminating = 1;
1446 
1447 	/* kill remaining link group connections */
1448 	read_lock_bh(&lgr->conns_lock);
1449 	node = rb_first(&lgr->conns_all);
1450 	while (node) {
1451 		read_unlock_bh(&lgr->conns_lock);
1452 		conn = rb_entry(node, struct smc_connection, alert_node);
1453 		smc = container_of(conn, struct smc_sock, conn);
1454 		sock_hold(&smc->sk); /* sock_put below */
1455 		lock_sock(&smc->sk);
1456 		smc_conn_kill(conn, soft);
1457 		release_sock(&smc->sk);
1458 		sock_put(&smc->sk); /* sock_hold above */
1459 		read_lock_bh(&lgr->conns_lock);
1460 		node = rb_first(&lgr->conns_all);
1461 	}
1462 	read_unlock_bh(&lgr->conns_lock);
1463 	smc_lgr_cleanup(lgr);
1464 	smc_lgr_free(lgr);
1465 }
1466 
1467 /* unlink link group and schedule termination */
1468 void smc_lgr_terminate_sched(struct smc_link_group *lgr)
1469 {
1470 	spinlock_t *lgr_lock;
1471 
1472 	smc_lgr_list_head(lgr, &lgr_lock);
1473 	spin_lock_bh(lgr_lock);
1474 	if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) {
1475 		spin_unlock_bh(lgr_lock);
1476 		return;	/* lgr already terminating */
1477 	}
1478 	list_del_init(&lgr->list);
1479 	lgr->freeing = 1;
1480 	spin_unlock_bh(lgr_lock);
1481 	schedule_work(&lgr->terminate_work);
1482 }
1483 
1484 /* Called when peer lgr shutdown (regularly or abnormally) is received */
1485 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan)
1486 {
1487 	struct smc_link_group *lgr, *l;
1488 	LIST_HEAD(lgr_free_list);
1489 
1490 	/* run common cleanup function and build free list */
1491 	spin_lock_bh(&dev->lgr_lock);
1492 	list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) {
1493 		if ((!peer_gid || lgr->peer_gid == peer_gid) &&
1494 		    (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) {
1495 			if (peer_gid) /* peer triggered termination */
1496 				lgr->peer_shutdown = 1;
1497 			list_move(&lgr->list, &lgr_free_list);
1498 			lgr->freeing = 1;
1499 		}
1500 	}
1501 	spin_unlock_bh(&dev->lgr_lock);
1502 
1503 	/* cancel the regular free workers and actually free lgrs */
1504 	list_for_each_entry_safe(lgr, l, &lgr_free_list, list) {
1505 		list_del_init(&lgr->list);
1506 		schedule_work(&lgr->terminate_work);
1507 	}
1508 }
1509 
1510 /* Called when an SMCD device is removed or the smc module is unloaded */
1511 void smc_smcd_terminate_all(struct smcd_dev *smcd)
1512 {
1513 	struct smc_link_group *lgr, *lg;
1514 	LIST_HEAD(lgr_free_list);
1515 
1516 	spin_lock_bh(&smcd->lgr_lock);
1517 	list_splice_init(&smcd->lgr_list, &lgr_free_list);
1518 	list_for_each_entry(lgr, &lgr_free_list, list)
1519 		lgr->freeing = 1;
1520 	spin_unlock_bh(&smcd->lgr_lock);
1521 
1522 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1523 		list_del_init(&lgr->list);
1524 		__smc_lgr_terminate(lgr, false);
1525 	}
1526 
1527 	if (atomic_read(&smcd->lgr_cnt))
1528 		wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt));
1529 }
1530 
1531 /* Called when an SMCR device is removed or the smc module is unloaded.
1532  * If smcibdev is given, all SMCR link groups using this device are terminated.
1533  * If smcibdev is NULL, all SMCR link groups are terminated.
1534  */
1535 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev)
1536 {
1537 	struct smc_link_group *lgr, *lg;
1538 	LIST_HEAD(lgr_free_list);
1539 	int i;
1540 
1541 	spin_lock_bh(&smc_lgr_list.lock);
1542 	if (!smcibdev) {
1543 		list_splice_init(&smc_lgr_list.list, &lgr_free_list);
1544 		list_for_each_entry(lgr, &lgr_free_list, list)
1545 			lgr->freeing = 1;
1546 	} else {
1547 		list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) {
1548 			for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1549 				if (lgr->lnk[i].smcibdev == smcibdev)
1550 					smcr_link_down_cond_sched(&lgr->lnk[i]);
1551 			}
1552 		}
1553 	}
1554 	spin_unlock_bh(&smc_lgr_list.lock);
1555 
1556 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1557 		list_del_init(&lgr->list);
1558 		smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_OP_INIT_TERM);
1559 		__smc_lgr_terminate(lgr, false);
1560 	}
1561 
1562 	if (smcibdev) {
1563 		if (atomic_read(&smcibdev->lnk_cnt))
1564 			wait_event(smcibdev->lnks_deleted,
1565 				   !atomic_read(&smcibdev->lnk_cnt));
1566 	} else {
1567 		if (atomic_read(&lgr_cnt))
1568 			wait_event(lgrs_deleted, !atomic_read(&lgr_cnt));
1569 	}
1570 }
1571 
1572 /* set new lgr type and clear all asymmetric link tagging */
1573 void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type)
1574 {
1575 	char *lgr_type = "";
1576 	int i;
1577 
1578 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1579 		if (smc_link_usable(&lgr->lnk[i]))
1580 			lgr->lnk[i].link_is_asym = false;
1581 	if (lgr->type == new_type)
1582 		return;
1583 	lgr->type = new_type;
1584 
1585 	switch (lgr->type) {
1586 	case SMC_LGR_NONE:
1587 		lgr_type = "NONE";
1588 		break;
1589 	case SMC_LGR_SINGLE:
1590 		lgr_type = "SINGLE";
1591 		break;
1592 	case SMC_LGR_SYMMETRIC:
1593 		lgr_type = "SYMMETRIC";
1594 		break;
1595 	case SMC_LGR_ASYMMETRIC_PEER:
1596 		lgr_type = "ASYMMETRIC_PEER";
1597 		break;
1598 	case SMC_LGR_ASYMMETRIC_LOCAL:
1599 		lgr_type = "ASYMMETRIC_LOCAL";
1600 		break;
1601 	}
1602 	pr_warn_ratelimited("smc: SMC-R lg %*phN net %llu state changed: "
1603 			    "%s, pnetid %.16s\n", SMC_LGR_ID_SIZE, &lgr->id,
1604 			    lgr->net->net_cookie, lgr_type, lgr->pnet_id);
1605 }
1606 
1607 /* set new lgr type and tag a link as asymmetric */
1608 void smcr_lgr_set_type_asym(struct smc_link_group *lgr,
1609 			    enum smc_lgr_type new_type, int asym_lnk_idx)
1610 {
1611 	smcr_lgr_set_type(lgr, new_type);
1612 	lgr->lnk[asym_lnk_idx].link_is_asym = true;
1613 }
1614 
1615 /* abort connection, abort_work scheduled from tasklet context */
1616 static void smc_conn_abort_work(struct work_struct *work)
1617 {
1618 	struct smc_connection *conn = container_of(work,
1619 						   struct smc_connection,
1620 						   abort_work);
1621 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1622 
1623 	lock_sock(&smc->sk);
1624 	smc_conn_kill(conn, true);
1625 	release_sock(&smc->sk);
1626 	sock_put(&smc->sk); /* sock_hold done by schedulers of abort_work */
1627 }
1628 
1629 void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport)
1630 {
1631 	struct smc_link_group *lgr, *n;
1632 
1633 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1634 		struct smc_link *link;
1635 
1636 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1637 			    SMC_MAX_PNETID_LEN) ||
1638 		    lgr->type == SMC_LGR_SYMMETRIC ||
1639 		    lgr->type == SMC_LGR_ASYMMETRIC_PEER ||
1640 		    !rdma_dev_access_netns(smcibdev->ibdev, lgr->net))
1641 			continue;
1642 
1643 		/* trigger local add link processing */
1644 		link = smc_llc_usable_link(lgr);
1645 		if (link)
1646 			smc_llc_add_link_local(link);
1647 	}
1648 }
1649 
1650 /* link is down - switch connections to alternate link,
1651  * must be called under lgr->llc_conf_mutex lock
1652  */
1653 static void smcr_link_down(struct smc_link *lnk)
1654 {
1655 	struct smc_link_group *lgr = lnk->lgr;
1656 	struct smc_link *to_lnk;
1657 	int del_link_id;
1658 
1659 	if (!lgr || lnk->state == SMC_LNK_UNUSED || list_empty(&lgr->list))
1660 		return;
1661 
1662 	to_lnk = smc_switch_conns(lgr, lnk, true);
1663 	if (!to_lnk) { /* no backup link available */
1664 		smcr_link_clear(lnk, true);
1665 		return;
1666 	}
1667 	smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1668 	del_link_id = lnk->link_id;
1669 
1670 	if (lgr->role == SMC_SERV) {
1671 		/* trigger local delete link processing */
1672 		smc_llc_srv_delete_link_local(to_lnk, del_link_id);
1673 	} else {
1674 		if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) {
1675 			/* another llc task is ongoing */
1676 			mutex_unlock(&lgr->llc_conf_mutex);
1677 			wait_event_timeout(lgr->llc_flow_waiter,
1678 				(list_empty(&lgr->list) ||
1679 				 lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE),
1680 				SMC_LLC_WAIT_TIME);
1681 			mutex_lock(&lgr->llc_conf_mutex);
1682 		}
1683 		if (!list_empty(&lgr->list)) {
1684 			smc_llc_send_delete_link(to_lnk, del_link_id,
1685 						 SMC_LLC_REQ, true,
1686 						 SMC_LLC_DEL_LOST_PATH);
1687 			smcr_link_clear(lnk, true);
1688 		}
1689 		wake_up(&lgr->llc_flow_waiter);	/* wake up next waiter */
1690 	}
1691 }
1692 
1693 /* must be called under lgr->llc_conf_mutex lock */
1694 void smcr_link_down_cond(struct smc_link *lnk)
1695 {
1696 	if (smc_link_downing(&lnk->state)) {
1697 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1698 		smcr_link_down(lnk);
1699 	}
1700 }
1701 
1702 /* will get the lgr->llc_conf_mutex lock */
1703 void smcr_link_down_cond_sched(struct smc_link *lnk)
1704 {
1705 	if (smc_link_downing(&lnk->state)) {
1706 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1707 		schedule_work(&lnk->link_down_wrk);
1708 	}
1709 }
1710 
1711 void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport)
1712 {
1713 	struct smc_link_group *lgr, *n;
1714 	int i;
1715 
1716 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1717 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1718 			    SMC_MAX_PNETID_LEN))
1719 			continue; /* lgr is not affected */
1720 		if (list_empty(&lgr->list))
1721 			continue;
1722 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1723 			struct smc_link *lnk = &lgr->lnk[i];
1724 
1725 			if (smc_link_usable(lnk) &&
1726 			    lnk->smcibdev == smcibdev && lnk->ibport == ibport)
1727 				smcr_link_down_cond_sched(lnk);
1728 		}
1729 	}
1730 }
1731 
1732 static void smc_link_down_work(struct work_struct *work)
1733 {
1734 	struct smc_link *link = container_of(work, struct smc_link,
1735 					     link_down_wrk);
1736 	struct smc_link_group *lgr = link->lgr;
1737 
1738 	if (list_empty(&lgr->list))
1739 		return;
1740 	wake_up_all(&lgr->llc_msg_waiter);
1741 	mutex_lock(&lgr->llc_conf_mutex);
1742 	smcr_link_down(link);
1743 	mutex_unlock(&lgr->llc_conf_mutex);
1744 }
1745 
1746 static int smc_vlan_by_tcpsk_walk(struct net_device *lower_dev,
1747 				  struct netdev_nested_priv *priv)
1748 {
1749 	unsigned short *vlan_id = (unsigned short *)priv->data;
1750 
1751 	if (is_vlan_dev(lower_dev)) {
1752 		*vlan_id = vlan_dev_vlan_id(lower_dev);
1753 		return 1;
1754 	}
1755 
1756 	return 0;
1757 }
1758 
1759 /* Determine vlan of internal TCP socket. */
1760 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini)
1761 {
1762 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
1763 	struct netdev_nested_priv priv;
1764 	struct net_device *ndev;
1765 	int rc = 0;
1766 
1767 	ini->vlan_id = 0;
1768 	if (!dst) {
1769 		rc = -ENOTCONN;
1770 		goto out;
1771 	}
1772 	if (!dst->dev) {
1773 		rc = -ENODEV;
1774 		goto out_rel;
1775 	}
1776 
1777 	ndev = dst->dev;
1778 	if (is_vlan_dev(ndev)) {
1779 		ini->vlan_id = vlan_dev_vlan_id(ndev);
1780 		goto out_rel;
1781 	}
1782 
1783 	priv.data = (void *)&ini->vlan_id;
1784 	rtnl_lock();
1785 	netdev_walk_all_lower_dev(ndev, smc_vlan_by_tcpsk_walk, &priv);
1786 	rtnl_unlock();
1787 
1788 out_rel:
1789 	dst_release(dst);
1790 out:
1791 	return rc;
1792 }
1793 
1794 static bool smcr_lgr_match(struct smc_link_group *lgr, u8 smcr_version,
1795 			   u8 peer_systemid[],
1796 			   u8 peer_gid[],
1797 			   u8 peer_mac_v1[],
1798 			   enum smc_lgr_role role, u32 clcqpn,
1799 			   struct net *net)
1800 {
1801 	struct smc_link *lnk;
1802 	int i;
1803 
1804 	if (memcmp(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN) ||
1805 	    lgr->role != role)
1806 		return false;
1807 
1808 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1809 		lnk = &lgr->lnk[i];
1810 
1811 		if (!smc_link_active(lnk))
1812 			continue;
1813 		/* use verbs API to check netns, instead of lgr->net */
1814 		if (!rdma_dev_access_netns(lnk->smcibdev->ibdev, net))
1815 			return false;
1816 		if ((lgr->role == SMC_SERV || lnk->peer_qpn == clcqpn) &&
1817 		    !memcmp(lnk->peer_gid, peer_gid, SMC_GID_SIZE) &&
1818 		    (smcr_version == SMC_V2 ||
1819 		     !memcmp(lnk->peer_mac, peer_mac_v1, ETH_ALEN)))
1820 			return true;
1821 	}
1822 	return false;
1823 }
1824 
1825 static bool smcd_lgr_match(struct smc_link_group *lgr,
1826 			   struct smcd_dev *smcismdev, u64 peer_gid)
1827 {
1828 	return lgr->peer_gid == peer_gid && lgr->smcd == smcismdev;
1829 }
1830 
1831 /* create a new SMC connection (and a new link group if necessary) */
1832 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
1833 {
1834 	struct smc_connection *conn = &smc->conn;
1835 	struct net *net = sock_net(&smc->sk);
1836 	struct list_head *lgr_list;
1837 	struct smc_link_group *lgr;
1838 	enum smc_lgr_role role;
1839 	spinlock_t *lgr_lock;
1840 	int rc = 0;
1841 
1842 	lgr_list = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_list :
1843 				  &smc_lgr_list.list;
1844 	lgr_lock = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_lock :
1845 				  &smc_lgr_list.lock;
1846 	ini->first_contact_local = 1;
1847 	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
1848 	if (role == SMC_CLNT && ini->first_contact_peer)
1849 		/* create new link group as well */
1850 		goto create;
1851 
1852 	/* determine if an existing link group can be reused */
1853 	spin_lock_bh(lgr_lock);
1854 	list_for_each_entry(lgr, lgr_list, list) {
1855 		write_lock_bh(&lgr->conns_lock);
1856 		if ((ini->is_smcd ?
1857 		     smcd_lgr_match(lgr, ini->ism_dev[ini->ism_selected],
1858 				    ini->ism_peer_gid[ini->ism_selected]) :
1859 		     smcr_lgr_match(lgr, ini->smcr_version,
1860 				    ini->peer_systemid,
1861 				    ini->peer_gid, ini->peer_mac, role,
1862 				    ini->ib_clcqpn, net)) &&
1863 		    !lgr->sync_err &&
1864 		    (ini->smcd_version == SMC_V2 ||
1865 		     lgr->vlan_id == ini->vlan_id) &&
1866 		    (role == SMC_CLNT || ini->is_smcd ||
1867 		     lgr->conns_num < SMC_RMBS_PER_LGR_MAX)) {
1868 			/* link group found */
1869 			ini->first_contact_local = 0;
1870 			conn->lgr = lgr;
1871 			rc = smc_lgr_register_conn(conn, false);
1872 			write_unlock_bh(&lgr->conns_lock);
1873 			if (!rc && delayed_work_pending(&lgr->free_work))
1874 				cancel_delayed_work(&lgr->free_work);
1875 			break;
1876 		}
1877 		write_unlock_bh(&lgr->conns_lock);
1878 	}
1879 	spin_unlock_bh(lgr_lock);
1880 	if (rc)
1881 		return rc;
1882 
1883 	if (role == SMC_CLNT && !ini->first_contact_peer &&
1884 	    ini->first_contact_local) {
1885 		/* Server reuses a link group, but Client wants to start
1886 		 * a new one
1887 		 * send out_of_sync decline, reason synchr. error
1888 		 */
1889 		return SMC_CLC_DECL_SYNCERR;
1890 	}
1891 
1892 create:
1893 	if (ini->first_contact_local) {
1894 		rc = smc_lgr_create(smc, ini);
1895 		if (rc)
1896 			goto out;
1897 		lgr = conn->lgr;
1898 		write_lock_bh(&lgr->conns_lock);
1899 		rc = smc_lgr_register_conn(conn, true);
1900 		write_unlock_bh(&lgr->conns_lock);
1901 		if (rc) {
1902 			smc_lgr_cleanup_early(lgr);
1903 			goto out;
1904 		}
1905 	}
1906 	smc_lgr_hold(conn->lgr); /* lgr_put in smc_conn_free() */
1907 	if (!conn->lgr->is_smcd)
1908 		smcr_link_hold(conn->lnk); /* link_put in smc_conn_free() */
1909 	conn->freed = 0;
1910 	conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
1911 	conn->local_tx_ctrl.len = SMC_WR_TX_SIZE;
1912 	conn->urg_state = SMC_URG_READ;
1913 	init_waitqueue_head(&conn->cdc_pend_tx_wq);
1914 	INIT_WORK(&smc->conn.abort_work, smc_conn_abort_work);
1915 	if (ini->is_smcd) {
1916 		conn->rx_off = sizeof(struct smcd_cdc_msg);
1917 		smcd_cdc_rx_init(conn); /* init tasklet for this conn */
1918 	} else {
1919 		conn->rx_off = 0;
1920 	}
1921 #ifndef KERNEL_HAS_ATOMIC64
1922 	spin_lock_init(&conn->acurs_lock);
1923 #endif
1924 
1925 out:
1926 	return rc;
1927 }
1928 
1929 #define SMCD_DMBE_SIZES		6 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */
1930 #define SMCR_RMBE_SIZES		5 /* 0 -> 16KB, 1 -> 32KB, .. 5 -> 512KB */
1931 
1932 /* convert the RMB size into the compressed notation (minimum 16K, see
1933  * SMCD/R_DMBE_SIZES.
1934  * In contrast to plain ilog2, this rounds towards the next power of 2,
1935  * so the socket application gets at least its desired sndbuf / rcvbuf size.
1936  */
1937 static u8 smc_compress_bufsize(int size, bool is_smcd, bool is_rmb)
1938 {
1939 	const unsigned int max_scat = SG_MAX_SINGLE_ALLOC * PAGE_SIZE;
1940 	u8 compressed;
1941 
1942 	if (size <= SMC_BUF_MIN_SIZE)
1943 		return 0;
1944 
1945 	size = (size - 1) >> 14;  /* convert to 16K multiple */
1946 	compressed = min_t(u8, ilog2(size) + 1,
1947 			   is_smcd ? SMCD_DMBE_SIZES : SMCR_RMBE_SIZES);
1948 
1949 	if (!is_smcd && is_rmb)
1950 		/* RMBs are backed by & limited to max size of scatterlists */
1951 		compressed = min_t(u8, compressed, ilog2(max_scat >> 14));
1952 
1953 	return compressed;
1954 }
1955 
1956 /* convert the RMB size from compressed notation into integer */
1957 int smc_uncompress_bufsize(u8 compressed)
1958 {
1959 	u32 size;
1960 
1961 	size = 0x00000001 << (((int)compressed) + 14);
1962 	return (int)size;
1963 }
1964 
1965 /* try to reuse a sndbuf or rmb description slot for a certain
1966  * buffer size; if not available, return NULL
1967  */
1968 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize,
1969 					     struct mutex *lock,
1970 					     struct list_head *buf_list)
1971 {
1972 	struct smc_buf_desc *buf_slot;
1973 
1974 	mutex_lock(lock);
1975 	list_for_each_entry(buf_slot, buf_list, list) {
1976 		if (cmpxchg(&buf_slot->used, 0, 1) == 0) {
1977 			mutex_unlock(lock);
1978 			return buf_slot;
1979 		}
1980 	}
1981 	mutex_unlock(lock);
1982 	return NULL;
1983 }
1984 
1985 /* one of the conditions for announcing a receiver's current window size is
1986  * that it "results in a minimum increase in the window size of 10% of the
1987  * receive buffer space" [RFC7609]
1988  */
1989 static inline int smc_rmb_wnd_update_limit(int rmbe_size)
1990 {
1991 	return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
1992 }
1993 
1994 /* map an rmb buf to a link */
1995 static int smcr_buf_map_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1996 			     struct smc_link *lnk)
1997 {
1998 	int rc;
1999 
2000 	if (buf_desc->is_map_ib[lnk->link_idx])
2001 		return 0;
2002 
2003 	rc = sg_alloc_table(&buf_desc->sgt[lnk->link_idx], 1, GFP_KERNEL);
2004 	if (rc)
2005 		return rc;
2006 	sg_set_buf(buf_desc->sgt[lnk->link_idx].sgl,
2007 		   buf_desc->cpu_addr, buf_desc->len);
2008 
2009 	/* map sg table to DMA address */
2010 	rc = smc_ib_buf_map_sg(lnk, buf_desc,
2011 			       is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2012 	/* SMC protocol depends on mapping to one DMA address only */
2013 	if (rc != 1) {
2014 		rc = -EAGAIN;
2015 		goto free_table;
2016 	}
2017 
2018 	/* create a new memory region for the RMB */
2019 	if (is_rmb) {
2020 		rc = smc_ib_get_memory_region(lnk->roce_pd,
2021 					      IB_ACCESS_REMOTE_WRITE |
2022 					      IB_ACCESS_LOCAL_WRITE,
2023 					      buf_desc, lnk->link_idx);
2024 		if (rc)
2025 			goto buf_unmap;
2026 		smc_ib_sync_sg_for_device(lnk, buf_desc, DMA_FROM_DEVICE);
2027 	}
2028 	buf_desc->is_map_ib[lnk->link_idx] = true;
2029 	return 0;
2030 
2031 buf_unmap:
2032 	smc_ib_buf_unmap_sg(lnk, buf_desc,
2033 			    is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2034 free_table:
2035 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
2036 	return rc;
2037 }
2038 
2039 /* register a new rmb on IB device,
2040  * must be called under lgr->llc_conf_mutex lock
2041  */
2042 int smcr_link_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc)
2043 {
2044 	if (list_empty(&link->lgr->list))
2045 		return -ENOLINK;
2046 	if (!rmb_desc->is_reg_mr[link->link_idx]) {
2047 		/* register memory region for new rmb */
2048 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[link->link_idx])) {
2049 			rmb_desc->is_reg_err = true;
2050 			return -EFAULT;
2051 		}
2052 		rmb_desc->is_reg_mr[link->link_idx] = true;
2053 	}
2054 	return 0;
2055 }
2056 
2057 static int _smcr_buf_map_lgr(struct smc_link *lnk, struct mutex *lock,
2058 			     struct list_head *lst, bool is_rmb)
2059 {
2060 	struct smc_buf_desc *buf_desc, *bf;
2061 	int rc = 0;
2062 
2063 	mutex_lock(lock);
2064 	list_for_each_entry_safe(buf_desc, bf, lst, list) {
2065 		if (!buf_desc->used)
2066 			continue;
2067 		rc = smcr_buf_map_link(buf_desc, is_rmb, lnk);
2068 		if (rc)
2069 			goto out;
2070 	}
2071 out:
2072 	mutex_unlock(lock);
2073 	return rc;
2074 }
2075 
2076 /* map all used buffers of lgr for a new link */
2077 int smcr_buf_map_lgr(struct smc_link *lnk)
2078 {
2079 	struct smc_link_group *lgr = lnk->lgr;
2080 	int i, rc = 0;
2081 
2082 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2083 		rc = _smcr_buf_map_lgr(lnk, &lgr->rmbs_lock,
2084 				       &lgr->rmbs[i], true);
2085 		if (rc)
2086 			return rc;
2087 		rc = _smcr_buf_map_lgr(lnk, &lgr->sndbufs_lock,
2088 				       &lgr->sndbufs[i], false);
2089 		if (rc)
2090 			return rc;
2091 	}
2092 	return 0;
2093 }
2094 
2095 /* register all used buffers of lgr for a new link,
2096  * must be called under lgr->llc_conf_mutex lock
2097  */
2098 int smcr_buf_reg_lgr(struct smc_link *lnk)
2099 {
2100 	struct smc_link_group *lgr = lnk->lgr;
2101 	struct smc_buf_desc *buf_desc, *bf;
2102 	int i, rc = 0;
2103 
2104 	mutex_lock(&lgr->rmbs_lock);
2105 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2106 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) {
2107 			if (!buf_desc->used)
2108 				continue;
2109 			rc = smcr_link_reg_rmb(lnk, buf_desc);
2110 			if (rc)
2111 				goto out;
2112 		}
2113 	}
2114 out:
2115 	mutex_unlock(&lgr->rmbs_lock);
2116 	return rc;
2117 }
2118 
2119 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr,
2120 						bool is_rmb, int bufsize)
2121 {
2122 	struct smc_buf_desc *buf_desc;
2123 
2124 	/* try to alloc a new buffer */
2125 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2126 	if (!buf_desc)
2127 		return ERR_PTR(-ENOMEM);
2128 
2129 	buf_desc->order = get_order(bufsize);
2130 	buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN |
2131 				      __GFP_NOMEMALLOC | __GFP_COMP |
2132 				      __GFP_NORETRY | __GFP_ZERO,
2133 				      buf_desc->order);
2134 	if (!buf_desc->pages) {
2135 		kfree(buf_desc);
2136 		return ERR_PTR(-EAGAIN);
2137 	}
2138 	buf_desc->cpu_addr = (void *)page_address(buf_desc->pages);
2139 	buf_desc->len = bufsize;
2140 	return buf_desc;
2141 }
2142 
2143 /* map buf_desc on all usable links,
2144  * unused buffers stay mapped as long as the link is up
2145  */
2146 static int smcr_buf_map_usable_links(struct smc_link_group *lgr,
2147 				     struct smc_buf_desc *buf_desc, bool is_rmb)
2148 {
2149 	int i, rc = 0;
2150 
2151 	/* protect against parallel link reconfiguration */
2152 	mutex_lock(&lgr->llc_conf_mutex);
2153 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2154 		struct smc_link *lnk = &lgr->lnk[i];
2155 
2156 		if (!smc_link_usable(lnk))
2157 			continue;
2158 		if (smcr_buf_map_link(buf_desc, is_rmb, lnk)) {
2159 			rc = -ENOMEM;
2160 			goto out;
2161 		}
2162 	}
2163 out:
2164 	mutex_unlock(&lgr->llc_conf_mutex);
2165 	return rc;
2166 }
2167 
2168 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr,
2169 						bool is_dmb, int bufsize)
2170 {
2171 	struct smc_buf_desc *buf_desc;
2172 	int rc;
2173 
2174 	/* try to alloc a new DMB */
2175 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2176 	if (!buf_desc)
2177 		return ERR_PTR(-ENOMEM);
2178 	if (is_dmb) {
2179 		rc = smc_ism_register_dmb(lgr, bufsize, buf_desc);
2180 		if (rc) {
2181 			kfree(buf_desc);
2182 			if (rc == -ENOMEM)
2183 				return ERR_PTR(-EAGAIN);
2184 			if (rc == -ENOSPC)
2185 				return ERR_PTR(-ENOSPC);
2186 			return ERR_PTR(-EIO);
2187 		}
2188 		buf_desc->pages = virt_to_page(buf_desc->cpu_addr);
2189 		/* CDC header stored in buf. So, pretend it was smaller */
2190 		buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg);
2191 	} else {
2192 		buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL |
2193 					     __GFP_NOWARN | __GFP_NORETRY |
2194 					     __GFP_NOMEMALLOC);
2195 		if (!buf_desc->cpu_addr) {
2196 			kfree(buf_desc);
2197 			return ERR_PTR(-EAGAIN);
2198 		}
2199 		buf_desc->len = bufsize;
2200 	}
2201 	return buf_desc;
2202 }
2203 
2204 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb)
2205 {
2206 	struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM);
2207 	struct smc_connection *conn = &smc->conn;
2208 	struct smc_link_group *lgr = conn->lgr;
2209 	struct list_head *buf_list;
2210 	int bufsize, bufsize_short;
2211 	bool is_dgraded = false;
2212 	struct mutex *lock;	/* lock buffer list */
2213 	int sk_buf_size;
2214 
2215 	if (is_rmb)
2216 		/* use socket recv buffer size (w/o overhead) as start value */
2217 		sk_buf_size = smc->sk.sk_rcvbuf / 2;
2218 	else
2219 		/* use socket send buffer size (w/o overhead) as start value */
2220 		sk_buf_size = smc->sk.sk_sndbuf / 2;
2221 
2222 	for (bufsize_short = smc_compress_bufsize(sk_buf_size, is_smcd, is_rmb);
2223 	     bufsize_short >= 0; bufsize_short--) {
2224 		if (is_rmb) {
2225 			lock = &lgr->rmbs_lock;
2226 			buf_list = &lgr->rmbs[bufsize_short];
2227 		} else {
2228 			lock = &lgr->sndbufs_lock;
2229 			buf_list = &lgr->sndbufs[bufsize_short];
2230 		}
2231 		bufsize = smc_uncompress_bufsize(bufsize_short);
2232 
2233 		/* check for reusable slot in the link group */
2234 		buf_desc = smc_buf_get_slot(bufsize_short, lock, buf_list);
2235 		if (buf_desc) {
2236 			SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2237 			SMC_STAT_BUF_REUSE(smc, is_smcd, is_rmb);
2238 			break; /* found reusable slot */
2239 		}
2240 
2241 		if (is_smcd)
2242 			buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize);
2243 		else
2244 			buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize);
2245 
2246 		if (PTR_ERR(buf_desc) == -ENOMEM)
2247 			break;
2248 		if (IS_ERR(buf_desc)) {
2249 			if (!is_dgraded) {
2250 				is_dgraded = true;
2251 				SMC_STAT_RMB_DOWNGRADED(smc, is_smcd, is_rmb);
2252 			}
2253 			continue;
2254 		}
2255 
2256 		SMC_STAT_RMB_ALLOC(smc, is_smcd, is_rmb);
2257 		SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2258 		buf_desc->used = 1;
2259 		mutex_lock(lock);
2260 		list_add(&buf_desc->list, buf_list);
2261 		mutex_unlock(lock);
2262 		break; /* found */
2263 	}
2264 
2265 	if (IS_ERR(buf_desc))
2266 		return PTR_ERR(buf_desc);
2267 
2268 	if (!is_smcd) {
2269 		if (smcr_buf_map_usable_links(lgr, buf_desc, is_rmb)) {
2270 			smcr_buf_unuse(buf_desc, lgr);
2271 			return -ENOMEM;
2272 		}
2273 	}
2274 
2275 	if (is_rmb) {
2276 		conn->rmb_desc = buf_desc;
2277 		conn->rmbe_size_short = bufsize_short;
2278 		smc->sk.sk_rcvbuf = bufsize * 2;
2279 		atomic_set(&conn->bytes_to_rcv, 0);
2280 		conn->rmbe_update_limit =
2281 			smc_rmb_wnd_update_limit(buf_desc->len);
2282 		if (is_smcd)
2283 			smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */
2284 	} else {
2285 		conn->sndbuf_desc = buf_desc;
2286 		smc->sk.sk_sndbuf = bufsize * 2;
2287 		atomic_set(&conn->sndbuf_space, bufsize);
2288 	}
2289 	return 0;
2290 }
2291 
2292 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn)
2293 {
2294 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd ||
2295 	    !smc_link_active(conn->lnk))
2296 		return;
2297 	smc_ib_sync_sg_for_cpu(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2298 }
2299 
2300 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn)
2301 {
2302 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd ||
2303 	    !smc_link_active(conn->lnk))
2304 		return;
2305 	smc_ib_sync_sg_for_device(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2306 }
2307 
2308 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn)
2309 {
2310 	int i;
2311 
2312 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd)
2313 		return;
2314 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2315 		if (!smc_link_active(&conn->lgr->lnk[i]))
2316 			continue;
2317 		smc_ib_sync_sg_for_cpu(&conn->lgr->lnk[i], conn->rmb_desc,
2318 				       DMA_FROM_DEVICE);
2319 	}
2320 }
2321 
2322 void smc_rmb_sync_sg_for_device(struct smc_connection *conn)
2323 {
2324 	int i;
2325 
2326 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd)
2327 		return;
2328 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2329 		if (!smc_link_active(&conn->lgr->lnk[i]))
2330 			continue;
2331 		smc_ib_sync_sg_for_device(&conn->lgr->lnk[i], conn->rmb_desc,
2332 					  DMA_FROM_DEVICE);
2333 	}
2334 }
2335 
2336 /* create the send and receive buffer for an SMC socket;
2337  * receive buffers are called RMBs;
2338  * (even though the SMC protocol allows more than one RMB-element per RMB,
2339  * the Linux implementation uses just one RMB-element per RMB, i.e. uses an
2340  * extra RMB for every connection in a link group
2341  */
2342 int smc_buf_create(struct smc_sock *smc, bool is_smcd)
2343 {
2344 	int rc;
2345 
2346 	/* create send buffer */
2347 	rc = __smc_buf_create(smc, is_smcd, false);
2348 	if (rc)
2349 		return rc;
2350 	/* create rmb */
2351 	rc = __smc_buf_create(smc, is_smcd, true);
2352 	if (rc) {
2353 		mutex_lock(&smc->conn.lgr->sndbufs_lock);
2354 		list_del(&smc->conn.sndbuf_desc->list);
2355 		mutex_unlock(&smc->conn.lgr->sndbufs_lock);
2356 		smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc);
2357 		smc->conn.sndbuf_desc = NULL;
2358 	}
2359 	return rc;
2360 }
2361 
2362 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
2363 {
2364 	int i;
2365 
2366 	for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
2367 		if (!test_and_set_bit(i, lgr->rtokens_used_mask))
2368 			return i;
2369 	}
2370 	return -ENOSPC;
2371 }
2372 
2373 static int smc_rtoken_find_by_link(struct smc_link_group *lgr, int lnk_idx,
2374 				   u32 rkey)
2375 {
2376 	int i;
2377 
2378 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2379 		if (test_bit(i, lgr->rtokens_used_mask) &&
2380 		    lgr->rtokens[i][lnk_idx].rkey == rkey)
2381 			return i;
2382 	}
2383 	return -ENOENT;
2384 }
2385 
2386 /* set rtoken for a new link to an existing rmb */
2387 void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new,
2388 		    __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey)
2389 {
2390 	int rtok_idx;
2391 
2392 	rtok_idx = smc_rtoken_find_by_link(lgr, link_idx, ntohl(nw_rkey_known));
2393 	if (rtok_idx == -ENOENT)
2394 		return;
2395 	lgr->rtokens[rtok_idx][link_idx_new].rkey = ntohl(nw_rkey);
2396 	lgr->rtokens[rtok_idx][link_idx_new].dma_addr = be64_to_cpu(nw_vaddr);
2397 }
2398 
2399 /* set rtoken for a new link whose link_id is given */
2400 void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id,
2401 		     __be64 nw_vaddr, __be32 nw_rkey)
2402 {
2403 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2404 	u32 rkey = ntohl(nw_rkey);
2405 	bool found = false;
2406 	int link_idx;
2407 
2408 	for (link_idx = 0; link_idx < SMC_LINKS_PER_LGR_MAX; link_idx++) {
2409 		if (lgr->lnk[link_idx].link_id == link_id) {
2410 			found = true;
2411 			break;
2412 		}
2413 	}
2414 	if (!found)
2415 		return;
2416 	lgr->rtokens[rtok_idx][link_idx].rkey = rkey;
2417 	lgr->rtokens[rtok_idx][link_idx].dma_addr = dma_addr;
2418 }
2419 
2420 /* add a new rtoken from peer */
2421 int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey)
2422 {
2423 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2424 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2425 	u32 rkey = ntohl(nw_rkey);
2426 	int i;
2427 
2428 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2429 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2430 		    lgr->rtokens[i][lnk->link_idx].dma_addr == dma_addr &&
2431 		    test_bit(i, lgr->rtokens_used_mask)) {
2432 			/* already in list */
2433 			return i;
2434 		}
2435 	}
2436 	i = smc_rmb_reserve_rtoken_idx(lgr);
2437 	if (i < 0)
2438 		return i;
2439 	lgr->rtokens[i][lnk->link_idx].rkey = rkey;
2440 	lgr->rtokens[i][lnk->link_idx].dma_addr = dma_addr;
2441 	return i;
2442 }
2443 
2444 /* delete an rtoken from all links */
2445 int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey)
2446 {
2447 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2448 	u32 rkey = ntohl(nw_rkey);
2449 	int i, j;
2450 
2451 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2452 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2453 		    test_bit(i, lgr->rtokens_used_mask)) {
2454 			for (j = 0; j < SMC_LINKS_PER_LGR_MAX; j++) {
2455 				lgr->rtokens[i][j].rkey = 0;
2456 				lgr->rtokens[i][j].dma_addr = 0;
2457 			}
2458 			clear_bit(i, lgr->rtokens_used_mask);
2459 			return 0;
2460 		}
2461 	}
2462 	return -ENOENT;
2463 }
2464 
2465 /* save rkey and dma_addr received from peer during clc handshake */
2466 int smc_rmb_rtoken_handling(struct smc_connection *conn,
2467 			    struct smc_link *lnk,
2468 			    struct smc_clc_msg_accept_confirm *clc)
2469 {
2470 	conn->rtoken_idx = smc_rtoken_add(lnk, clc->r0.rmb_dma_addr,
2471 					  clc->r0.rmb_rkey);
2472 	if (conn->rtoken_idx < 0)
2473 		return conn->rtoken_idx;
2474 	return 0;
2475 }
2476 
2477 static void smc_core_going_away(void)
2478 {
2479 	struct smc_ib_device *smcibdev;
2480 	struct smcd_dev *smcd;
2481 
2482 	mutex_lock(&smc_ib_devices.mutex);
2483 	list_for_each_entry(smcibdev, &smc_ib_devices.list, list) {
2484 		int i;
2485 
2486 		for (i = 0; i < SMC_MAX_PORTS; i++)
2487 			set_bit(i, smcibdev->ports_going_away);
2488 	}
2489 	mutex_unlock(&smc_ib_devices.mutex);
2490 
2491 	mutex_lock(&smcd_dev_list.mutex);
2492 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2493 		smcd->going_away = 1;
2494 	}
2495 	mutex_unlock(&smcd_dev_list.mutex);
2496 }
2497 
2498 /* Clean up all SMC link groups */
2499 static void smc_lgrs_shutdown(void)
2500 {
2501 	struct smcd_dev *smcd;
2502 
2503 	smc_core_going_away();
2504 
2505 	smc_smcr_terminate_all(NULL);
2506 
2507 	mutex_lock(&smcd_dev_list.mutex);
2508 	list_for_each_entry(smcd, &smcd_dev_list.list, list)
2509 		smc_smcd_terminate_all(smcd);
2510 	mutex_unlock(&smcd_dev_list.mutex);
2511 }
2512 
2513 static int smc_core_reboot_event(struct notifier_block *this,
2514 				 unsigned long event, void *ptr)
2515 {
2516 	smc_lgrs_shutdown();
2517 	smc_ib_unregister_client();
2518 	return 0;
2519 }
2520 
2521 static struct notifier_block smc_reboot_notifier = {
2522 	.notifier_call = smc_core_reboot_event,
2523 };
2524 
2525 int __init smc_core_init(void)
2526 {
2527 	return register_reboot_notifier(&smc_reboot_notifier);
2528 }
2529 
2530 /* Called (from smc_exit) when module is removed */
2531 void smc_core_exit(void)
2532 {
2533 	unregister_reboot_notifier(&smc_reboot_notifier);
2534 	smc_lgrs_shutdown();
2535 }
2536