xref: /openbmc/linux/net/smc/smc_core.c (revision 34fa67e7)
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 (!lgr)
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 	conn->lgr = NULL;
222 }
223 
224 int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb)
225 {
226 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
227 	char hostname[SMC_MAX_HOSTNAME_LEN + 1];
228 	char smc_seid[SMC_MAX_EID_LEN + 1];
229 	struct nlattr *attrs;
230 	u8 *seid = NULL;
231 	u8 *host = NULL;
232 	void *nlh;
233 
234 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
235 			  &smc_gen_nl_family, NLM_F_MULTI,
236 			  SMC_NETLINK_GET_SYS_INFO);
237 	if (!nlh)
238 		goto errmsg;
239 	if (cb_ctx->pos[0])
240 		goto errout;
241 	attrs = nla_nest_start(skb, SMC_GEN_SYS_INFO);
242 	if (!attrs)
243 		goto errout;
244 	if (nla_put_u8(skb, SMC_NLA_SYS_VER, SMC_V2))
245 		goto errattr;
246 	if (nla_put_u8(skb, SMC_NLA_SYS_REL, SMC_RELEASE))
247 		goto errattr;
248 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_ISM_V2, smc_ism_is_v2_capable()))
249 		goto errattr;
250 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_SMCR_V2, true))
251 		goto errattr;
252 	smc_clc_get_hostname(&host);
253 	if (host) {
254 		memcpy(hostname, host, SMC_MAX_HOSTNAME_LEN);
255 		hostname[SMC_MAX_HOSTNAME_LEN] = 0;
256 		if (nla_put_string(skb, SMC_NLA_SYS_LOCAL_HOST, hostname))
257 			goto errattr;
258 	}
259 	if (smc_ism_is_v2_capable()) {
260 		smc_ism_get_system_eid(&seid);
261 		memcpy(smc_seid, seid, SMC_MAX_EID_LEN);
262 		smc_seid[SMC_MAX_EID_LEN] = 0;
263 		if (nla_put_string(skb, SMC_NLA_SYS_SEID, smc_seid))
264 			goto errattr;
265 	}
266 	nla_nest_end(skb, attrs);
267 	genlmsg_end(skb, nlh);
268 	cb_ctx->pos[0] = 1;
269 	return skb->len;
270 
271 errattr:
272 	nla_nest_cancel(skb, attrs);
273 errout:
274 	genlmsg_cancel(skb, nlh);
275 errmsg:
276 	return skb->len;
277 }
278 
279 /* Fill SMC_NLA_LGR_D_V2_COMMON/SMC_NLA_LGR_R_V2_COMMON nested attributes */
280 static int smc_nl_fill_lgr_v2_common(struct smc_link_group *lgr,
281 				     struct sk_buff *skb,
282 				     struct netlink_callback *cb,
283 				     struct nlattr *v2_attrs)
284 {
285 	char smc_host[SMC_MAX_HOSTNAME_LEN + 1];
286 	char smc_eid[SMC_MAX_EID_LEN + 1];
287 
288 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_VER, lgr->smc_version))
289 		goto errv2attr;
290 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_REL, lgr->peer_smc_release))
291 		goto errv2attr;
292 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_OS, lgr->peer_os))
293 		goto errv2attr;
294 	memcpy(smc_host, lgr->peer_hostname, SMC_MAX_HOSTNAME_LEN);
295 	smc_host[SMC_MAX_HOSTNAME_LEN] = 0;
296 	if (nla_put_string(skb, SMC_NLA_LGR_V2_PEER_HOST, smc_host))
297 		goto errv2attr;
298 	memcpy(smc_eid, lgr->negotiated_eid, SMC_MAX_EID_LEN);
299 	smc_eid[SMC_MAX_EID_LEN] = 0;
300 	if (nla_put_string(skb, SMC_NLA_LGR_V2_NEG_EID, smc_eid))
301 		goto errv2attr;
302 
303 	nla_nest_end(skb, v2_attrs);
304 	return 0;
305 
306 errv2attr:
307 	nla_nest_cancel(skb, v2_attrs);
308 	return -EMSGSIZE;
309 }
310 
311 static int smc_nl_fill_smcr_lgr_v2(struct smc_link_group *lgr,
312 				   struct sk_buff *skb,
313 				   struct netlink_callback *cb)
314 {
315 	struct nlattr *v2_attrs;
316 
317 	v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2);
318 	if (!v2_attrs)
319 		goto errattr;
320 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_DIRECT, !lgr->uses_gateway))
321 		goto errv2attr;
322 
323 	nla_nest_end(skb, v2_attrs);
324 	return 0;
325 
326 errv2attr:
327 	nla_nest_cancel(skb, v2_attrs);
328 errattr:
329 	return -EMSGSIZE;
330 }
331 
332 static int smc_nl_fill_lgr(struct smc_link_group *lgr,
333 			   struct sk_buff *skb,
334 			   struct netlink_callback *cb)
335 {
336 	char smc_target[SMC_MAX_PNETID_LEN + 1];
337 	struct nlattr *attrs, *v2_attrs;
338 
339 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCR);
340 	if (!attrs)
341 		goto errout;
342 
343 	if (nla_put_u32(skb, SMC_NLA_LGR_R_ID, *((u32 *)&lgr->id)))
344 		goto errattr;
345 	if (nla_put_u32(skb, SMC_NLA_LGR_R_CONNS_NUM, lgr->conns_num))
346 		goto errattr;
347 	if (nla_put_u8(skb, SMC_NLA_LGR_R_ROLE, lgr->role))
348 		goto errattr;
349 	if (nla_put_u8(skb, SMC_NLA_LGR_R_TYPE, lgr->type))
350 		goto errattr;
351 	if (nla_put_u8(skb, SMC_NLA_LGR_R_VLAN_ID, lgr->vlan_id))
352 		goto errattr;
353 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_R_NET_COOKIE,
354 			      lgr->net->net_cookie, SMC_NLA_LGR_R_PAD))
355 		goto errattr;
356 	memcpy(smc_target, lgr->pnet_id, SMC_MAX_PNETID_LEN);
357 	smc_target[SMC_MAX_PNETID_LEN] = 0;
358 	if (nla_put_string(skb, SMC_NLA_LGR_R_PNETID, smc_target))
359 		goto errattr;
360 	if (lgr->smc_version > SMC_V1) {
361 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2_COMMON);
362 		if (!v2_attrs)
363 			goto errattr;
364 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
365 			goto errattr;
366 		if (smc_nl_fill_smcr_lgr_v2(lgr, skb, cb))
367 			goto errattr;
368 	}
369 
370 	nla_nest_end(skb, attrs);
371 	return 0;
372 errattr:
373 	nla_nest_cancel(skb, attrs);
374 errout:
375 	return -EMSGSIZE;
376 }
377 
378 static int smc_nl_fill_lgr_link(struct smc_link_group *lgr,
379 				struct smc_link *link,
380 				struct sk_buff *skb,
381 				struct netlink_callback *cb)
382 {
383 	char smc_ibname[IB_DEVICE_NAME_MAX];
384 	u8 smc_gid_target[41];
385 	struct nlattr *attrs;
386 	u32 link_uid = 0;
387 	void *nlh;
388 
389 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
390 			  &smc_gen_nl_family, NLM_F_MULTI,
391 			  SMC_NETLINK_GET_LINK_SMCR);
392 	if (!nlh)
393 		goto errmsg;
394 
395 	attrs = nla_nest_start(skb, SMC_GEN_LINK_SMCR);
396 	if (!attrs)
397 		goto errout;
398 
399 	if (nla_put_u8(skb, SMC_NLA_LINK_ID, link->link_id))
400 		goto errattr;
401 	if (nla_put_u32(skb, SMC_NLA_LINK_STATE, link->state))
402 		goto errattr;
403 	if (nla_put_u32(skb, SMC_NLA_LINK_CONN_CNT,
404 			atomic_read(&link->conn_cnt)))
405 		goto errattr;
406 	if (nla_put_u8(skb, SMC_NLA_LINK_IB_PORT, link->ibport))
407 		goto errattr;
408 	if (nla_put_u32(skb, SMC_NLA_LINK_NET_DEV, link->ndev_ifidx))
409 		goto errattr;
410 	snprintf(smc_ibname, sizeof(smc_ibname), "%s", link->ibname);
411 	if (nla_put_string(skb, SMC_NLA_LINK_IB_DEV, smc_ibname))
412 		goto errattr;
413 	memcpy(&link_uid, link->link_uid, sizeof(link_uid));
414 	if (nla_put_u32(skb, SMC_NLA_LINK_UID, link_uid))
415 		goto errattr;
416 	memcpy(&link_uid, link->peer_link_uid, sizeof(link_uid));
417 	if (nla_put_u32(skb, SMC_NLA_LINK_PEER_UID, link_uid))
418 		goto errattr;
419 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
420 	smc_gid_be16_convert(smc_gid_target, link->gid);
421 	if (nla_put_string(skb, SMC_NLA_LINK_GID, smc_gid_target))
422 		goto errattr;
423 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
424 	smc_gid_be16_convert(smc_gid_target, link->peer_gid);
425 	if (nla_put_string(skb, SMC_NLA_LINK_PEER_GID, smc_gid_target))
426 		goto errattr;
427 
428 	nla_nest_end(skb, attrs);
429 	genlmsg_end(skb, nlh);
430 	return 0;
431 errattr:
432 	nla_nest_cancel(skb, attrs);
433 errout:
434 	genlmsg_cancel(skb, nlh);
435 errmsg:
436 	return -EMSGSIZE;
437 }
438 
439 static int smc_nl_handle_lgr(struct smc_link_group *lgr,
440 			     struct sk_buff *skb,
441 			     struct netlink_callback *cb,
442 			     bool list_links)
443 {
444 	void *nlh;
445 	int i;
446 
447 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
448 			  &smc_gen_nl_family, NLM_F_MULTI,
449 			  SMC_NETLINK_GET_LGR_SMCR);
450 	if (!nlh)
451 		goto errmsg;
452 	if (smc_nl_fill_lgr(lgr, skb, cb))
453 		goto errout;
454 
455 	genlmsg_end(skb, nlh);
456 	if (!list_links)
457 		goto out;
458 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
459 		if (!smc_link_usable(&lgr->lnk[i]))
460 			continue;
461 		if (smc_nl_fill_lgr_link(lgr, &lgr->lnk[i], skb, cb))
462 			goto errout;
463 	}
464 out:
465 	return 0;
466 
467 errout:
468 	genlmsg_cancel(skb, nlh);
469 errmsg:
470 	return -EMSGSIZE;
471 }
472 
473 static void smc_nl_fill_lgr_list(struct smc_lgr_list *smc_lgr,
474 				 struct sk_buff *skb,
475 				 struct netlink_callback *cb,
476 				 bool list_links)
477 {
478 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
479 	struct smc_link_group *lgr;
480 	int snum = cb_ctx->pos[0];
481 	int num = 0;
482 
483 	spin_lock_bh(&smc_lgr->lock);
484 	list_for_each_entry(lgr, &smc_lgr->list, list) {
485 		if (num < snum)
486 			goto next;
487 		if (smc_nl_handle_lgr(lgr, skb, cb, list_links))
488 			goto errout;
489 next:
490 		num++;
491 	}
492 errout:
493 	spin_unlock_bh(&smc_lgr->lock);
494 	cb_ctx->pos[0] = num;
495 }
496 
497 static int smc_nl_fill_smcd_lgr(struct smc_link_group *lgr,
498 				struct sk_buff *skb,
499 				struct netlink_callback *cb)
500 {
501 	char smc_pnet[SMC_MAX_PNETID_LEN + 1];
502 	struct nlattr *attrs;
503 	void *nlh;
504 
505 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
506 			  &smc_gen_nl_family, NLM_F_MULTI,
507 			  SMC_NETLINK_GET_LGR_SMCD);
508 	if (!nlh)
509 		goto errmsg;
510 
511 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCD);
512 	if (!attrs)
513 		goto errout;
514 
515 	if (nla_put_u32(skb, SMC_NLA_LGR_D_ID, *((u32 *)&lgr->id)))
516 		goto errattr;
517 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_GID, lgr->smcd->local_gid,
518 			      SMC_NLA_LGR_D_PAD))
519 		goto errattr;
520 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_GID, lgr->peer_gid,
521 			      SMC_NLA_LGR_D_PAD))
522 		goto errattr;
523 	if (nla_put_u8(skb, SMC_NLA_LGR_D_VLAN_ID, lgr->vlan_id))
524 		goto errattr;
525 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CONNS_NUM, lgr->conns_num))
526 		goto errattr;
527 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CHID, smc_ism_get_chid(lgr->smcd)))
528 		goto errattr;
529 	memcpy(smc_pnet, lgr->smcd->pnetid, SMC_MAX_PNETID_LEN);
530 	smc_pnet[SMC_MAX_PNETID_LEN] = 0;
531 	if (nla_put_string(skb, SMC_NLA_LGR_D_PNETID, smc_pnet))
532 		goto errattr;
533 	if (lgr->smc_version > SMC_V1) {
534 		struct nlattr *v2_attrs;
535 
536 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_D_V2_COMMON);
537 		if (!v2_attrs)
538 			goto errattr;
539 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
540 			goto errattr;
541 	}
542 	nla_nest_end(skb, attrs);
543 	genlmsg_end(skb, nlh);
544 	return 0;
545 
546 errattr:
547 	nla_nest_cancel(skb, attrs);
548 errout:
549 	genlmsg_cancel(skb, nlh);
550 errmsg:
551 	return -EMSGSIZE;
552 }
553 
554 static int smc_nl_handle_smcd_lgr(struct smcd_dev *dev,
555 				  struct sk_buff *skb,
556 				  struct netlink_callback *cb)
557 {
558 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
559 	struct smc_link_group *lgr;
560 	int snum = cb_ctx->pos[1];
561 	int rc = 0, num = 0;
562 
563 	spin_lock_bh(&dev->lgr_lock);
564 	list_for_each_entry(lgr, &dev->lgr_list, list) {
565 		if (!lgr->is_smcd)
566 			continue;
567 		if (num < snum)
568 			goto next;
569 		rc = smc_nl_fill_smcd_lgr(lgr, skb, cb);
570 		if (rc)
571 			goto errout;
572 next:
573 		num++;
574 	}
575 errout:
576 	spin_unlock_bh(&dev->lgr_lock);
577 	cb_ctx->pos[1] = num;
578 	return rc;
579 }
580 
581 static int smc_nl_fill_smcd_dev(struct smcd_dev_list *dev_list,
582 				struct sk_buff *skb,
583 				struct netlink_callback *cb)
584 {
585 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
586 	struct smcd_dev *smcd_dev;
587 	int snum = cb_ctx->pos[0];
588 	int rc = 0, num = 0;
589 
590 	mutex_lock(&dev_list->mutex);
591 	list_for_each_entry(smcd_dev, &dev_list->list, list) {
592 		if (list_empty(&smcd_dev->lgr_list))
593 			continue;
594 		if (num < snum)
595 			goto next;
596 		rc = smc_nl_handle_smcd_lgr(smcd_dev, skb, cb);
597 		if (rc)
598 			goto errout;
599 next:
600 		num++;
601 	}
602 errout:
603 	mutex_unlock(&dev_list->mutex);
604 	cb_ctx->pos[0] = num;
605 	return rc;
606 }
607 
608 int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
609 {
610 	bool list_links = false;
611 
612 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
613 	return skb->len;
614 }
615 
616 int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb)
617 {
618 	bool list_links = true;
619 
620 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
621 	return skb->len;
622 }
623 
624 int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
625 {
626 	smc_nl_fill_smcd_dev(&smcd_dev_list, skb, cb);
627 	return skb->len;
628 }
629 
630 void smc_lgr_cleanup_early(struct smc_link_group *lgr)
631 {
632 	spinlock_t *lgr_lock;
633 
634 	if (!lgr)
635 		return;
636 
637 	smc_lgr_list_head(lgr, &lgr_lock);
638 	spin_lock_bh(lgr_lock);
639 	/* do not use this link group for new connections */
640 	if (!list_empty(&lgr->list))
641 		list_del_init(&lgr->list);
642 	spin_unlock_bh(lgr_lock);
643 	__smc_lgr_terminate(lgr, true);
644 }
645 
646 static void smcr_lgr_link_deactivate_all(struct smc_link_group *lgr)
647 {
648 	int i;
649 
650 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
651 		struct smc_link *lnk = &lgr->lnk[i];
652 
653 		if (smc_link_sendable(lnk))
654 			lnk->state = SMC_LNK_INACTIVE;
655 	}
656 	wake_up_all(&lgr->llc_msg_waiter);
657 	wake_up_all(&lgr->llc_flow_waiter);
658 }
659 
660 static void smc_lgr_free(struct smc_link_group *lgr);
661 
662 static void smc_lgr_free_work(struct work_struct *work)
663 {
664 	struct smc_link_group *lgr = container_of(to_delayed_work(work),
665 						  struct smc_link_group,
666 						  free_work);
667 	spinlock_t *lgr_lock;
668 	bool conns;
669 
670 	smc_lgr_list_head(lgr, &lgr_lock);
671 	spin_lock_bh(lgr_lock);
672 	if (lgr->freeing) {
673 		spin_unlock_bh(lgr_lock);
674 		return;
675 	}
676 	read_lock_bh(&lgr->conns_lock);
677 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
678 	read_unlock_bh(&lgr->conns_lock);
679 	if (!conns) { /* number of lgr connections is no longer zero */
680 		spin_unlock_bh(lgr_lock);
681 		return;
682 	}
683 	list_del_init(&lgr->list); /* remove from smc_lgr_list */
684 	lgr->freeing = 1; /* this instance does the freeing, no new schedule */
685 	spin_unlock_bh(lgr_lock);
686 	cancel_delayed_work(&lgr->free_work);
687 
688 	if (!lgr->is_smcd && !lgr->terminating)
689 		smc_llc_send_link_delete_all(lgr, true,
690 					     SMC_LLC_DEL_PROG_INIT_TERM);
691 	if (lgr->is_smcd && !lgr->terminating)
692 		smc_ism_signal_shutdown(lgr);
693 	if (!lgr->is_smcd)
694 		smcr_lgr_link_deactivate_all(lgr);
695 	smc_lgr_free(lgr);
696 }
697 
698 static void smc_lgr_terminate_work(struct work_struct *work)
699 {
700 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
701 						  terminate_work);
702 
703 	__smc_lgr_terminate(lgr, true);
704 }
705 
706 /* return next unique link id for the lgr */
707 static u8 smcr_next_link_id(struct smc_link_group *lgr)
708 {
709 	u8 link_id;
710 	int i;
711 
712 	while (1) {
713 again:
714 		link_id = ++lgr->next_link_id;
715 		if (!link_id)	/* skip zero as link_id */
716 			link_id = ++lgr->next_link_id;
717 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
718 			if (smc_link_usable(&lgr->lnk[i]) &&
719 			    lgr->lnk[i].link_id == link_id)
720 				goto again;
721 		}
722 		break;
723 	}
724 	return link_id;
725 }
726 
727 static void smcr_copy_dev_info_to_link(struct smc_link *link)
728 {
729 	struct smc_ib_device *smcibdev = link->smcibdev;
730 
731 	snprintf(link->ibname, sizeof(link->ibname), "%s",
732 		 smcibdev->ibdev->name);
733 	link->ndev_ifidx = smcibdev->ndev_ifidx[link->ibport - 1];
734 }
735 
736 int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk,
737 		   u8 link_idx, struct smc_init_info *ini)
738 {
739 	struct smc_ib_device *smcibdev;
740 	u8 rndvec[3];
741 	int rc;
742 
743 	if (lgr->smc_version == SMC_V2) {
744 		lnk->smcibdev = ini->smcrv2.ib_dev_v2;
745 		lnk->ibport = ini->smcrv2.ib_port_v2;
746 	} else {
747 		lnk->smcibdev = ini->ib_dev;
748 		lnk->ibport = ini->ib_port;
749 	}
750 	get_device(&lnk->smcibdev->ibdev->dev);
751 	atomic_inc(&lnk->smcibdev->lnk_cnt);
752 	lnk->path_mtu = lnk->smcibdev->pattr[lnk->ibport - 1].active_mtu;
753 	lnk->link_id = smcr_next_link_id(lgr);
754 	lnk->lgr = lgr;
755 	lnk->link_idx = link_idx;
756 	smc_ibdev_cnt_inc(lnk);
757 	smcr_copy_dev_info_to_link(lnk);
758 	atomic_set(&lnk->conn_cnt, 0);
759 	smc_llc_link_set_uid(lnk);
760 	INIT_WORK(&lnk->link_down_wrk, smc_link_down_work);
761 	if (!lnk->smcibdev->initialized) {
762 		rc = (int)smc_ib_setup_per_ibdev(lnk->smcibdev);
763 		if (rc)
764 			goto out;
765 	}
766 	get_random_bytes(rndvec, sizeof(rndvec));
767 	lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) +
768 		(rndvec[2] << 16);
769 	rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport,
770 				  ini->vlan_id, lnk->gid, &lnk->sgid_index,
771 				  lgr->smc_version == SMC_V2 ?
772 						  &ini->smcrv2 : NULL);
773 	if (rc)
774 		goto out;
775 	rc = smc_llc_link_init(lnk);
776 	if (rc)
777 		goto out;
778 	rc = smc_wr_alloc_link_mem(lnk);
779 	if (rc)
780 		goto clear_llc_lnk;
781 	rc = smc_ib_create_protection_domain(lnk);
782 	if (rc)
783 		goto free_link_mem;
784 	rc = smc_ib_create_queue_pair(lnk);
785 	if (rc)
786 		goto dealloc_pd;
787 	rc = smc_wr_create_link(lnk);
788 	if (rc)
789 		goto destroy_qp;
790 	lnk->state = SMC_LNK_ACTIVATING;
791 	return 0;
792 
793 destroy_qp:
794 	smc_ib_destroy_queue_pair(lnk);
795 dealloc_pd:
796 	smc_ib_dealloc_protection_domain(lnk);
797 free_link_mem:
798 	smc_wr_free_link_mem(lnk);
799 clear_llc_lnk:
800 	smc_llc_link_clear(lnk, false);
801 out:
802 	smc_ibdev_cnt_dec(lnk);
803 	put_device(&lnk->smcibdev->ibdev->dev);
804 	smcibdev = lnk->smcibdev;
805 	memset(lnk, 0, sizeof(struct smc_link));
806 	lnk->state = SMC_LNK_UNUSED;
807 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
808 		wake_up(&smcibdev->lnks_deleted);
809 	return rc;
810 }
811 
812 /* create a new SMC link group */
813 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
814 {
815 	struct smc_link_group *lgr;
816 	struct list_head *lgr_list;
817 	struct smc_link *lnk;
818 	spinlock_t *lgr_lock;
819 	u8 link_idx;
820 	int rc = 0;
821 	int i;
822 
823 	if (ini->is_smcd && ini->vlan_id) {
824 		if (smc_ism_get_vlan(ini->ism_dev[ini->ism_selected],
825 				     ini->vlan_id)) {
826 			rc = SMC_CLC_DECL_ISMVLANERR;
827 			goto out;
828 		}
829 	}
830 
831 	lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
832 	if (!lgr) {
833 		rc = SMC_CLC_DECL_MEM;
834 		goto ism_put_vlan;
835 	}
836 	lgr->tx_wq = alloc_workqueue("smc_tx_wq-%*phN", 0, 0,
837 				     SMC_LGR_ID_SIZE, &lgr->id);
838 	if (!lgr->tx_wq) {
839 		rc = -ENOMEM;
840 		goto free_lgr;
841 	}
842 	lgr->is_smcd = ini->is_smcd;
843 	lgr->sync_err = 0;
844 	lgr->terminating = 0;
845 	lgr->freeing = 0;
846 	lgr->vlan_id = ini->vlan_id;
847 	mutex_init(&lgr->sndbufs_lock);
848 	mutex_init(&lgr->rmbs_lock);
849 	rwlock_init(&lgr->conns_lock);
850 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
851 		INIT_LIST_HEAD(&lgr->sndbufs[i]);
852 		INIT_LIST_HEAD(&lgr->rmbs[i]);
853 	}
854 	lgr->next_link_id = 0;
855 	smc_lgr_list.num += SMC_LGR_NUM_INCR;
856 	memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE);
857 	INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
858 	INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work);
859 	lgr->conns_all = RB_ROOT;
860 	if (ini->is_smcd) {
861 		/* SMC-D specific settings */
862 		get_device(&ini->ism_dev[ini->ism_selected]->dev);
863 		lgr->peer_gid = ini->ism_peer_gid[ini->ism_selected];
864 		lgr->smcd = ini->ism_dev[ini->ism_selected];
865 		lgr_list = &ini->ism_dev[ini->ism_selected]->lgr_list;
866 		lgr_lock = &lgr->smcd->lgr_lock;
867 		lgr->smc_version = ini->smcd_version;
868 		lgr->peer_shutdown = 0;
869 		atomic_inc(&ini->ism_dev[ini->ism_selected]->lgr_cnt);
870 	} else {
871 		/* SMC-R specific settings */
872 		struct smc_ib_device *ibdev;
873 		int ibport;
874 
875 		lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
876 		lgr->smc_version = ini->smcr_version;
877 		memcpy(lgr->peer_systemid, ini->peer_systemid,
878 		       SMC_SYSTEMID_LEN);
879 		if (lgr->smc_version == SMC_V2) {
880 			ibdev = ini->smcrv2.ib_dev_v2;
881 			ibport = ini->smcrv2.ib_port_v2;
882 			lgr->saddr = ini->smcrv2.saddr;
883 			lgr->uses_gateway = ini->smcrv2.uses_gateway;
884 			memcpy(lgr->nexthop_mac, ini->smcrv2.nexthop_mac,
885 			       ETH_ALEN);
886 		} else {
887 			ibdev = ini->ib_dev;
888 			ibport = ini->ib_port;
889 		}
890 		memcpy(lgr->pnet_id, ibdev->pnetid[ibport - 1],
891 		       SMC_MAX_PNETID_LEN);
892 		if (smc_wr_alloc_lgr_mem(lgr))
893 			goto free_wq;
894 		smc_llc_lgr_init(lgr, smc);
895 
896 		link_idx = SMC_SINGLE_LINK;
897 		lnk = &lgr->lnk[link_idx];
898 		rc = smcr_link_init(lgr, lnk, link_idx, ini);
899 		if (rc) {
900 			smc_wr_free_lgr_mem(lgr);
901 			goto free_wq;
902 		}
903 		lgr->net = smc_ib_net(lnk->smcibdev);
904 		lgr_list = &smc_lgr_list.list;
905 		lgr_lock = &smc_lgr_list.lock;
906 		atomic_inc(&lgr_cnt);
907 	}
908 	smc->conn.lgr = lgr;
909 	spin_lock_bh(lgr_lock);
910 	list_add_tail(&lgr->list, lgr_list);
911 	spin_unlock_bh(lgr_lock);
912 	return 0;
913 
914 free_wq:
915 	destroy_workqueue(lgr->tx_wq);
916 free_lgr:
917 	kfree(lgr);
918 ism_put_vlan:
919 	if (ini->is_smcd && ini->vlan_id)
920 		smc_ism_put_vlan(ini->ism_dev[ini->ism_selected], ini->vlan_id);
921 out:
922 	if (rc < 0) {
923 		if (rc == -ENOMEM)
924 			rc = SMC_CLC_DECL_MEM;
925 		else
926 			rc = SMC_CLC_DECL_INTERR;
927 	}
928 	return rc;
929 }
930 
931 static int smc_write_space(struct smc_connection *conn)
932 {
933 	int buffer_len = conn->peer_rmbe_size;
934 	union smc_host_cursor prod;
935 	union smc_host_cursor cons;
936 	int space;
937 
938 	smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
939 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
940 	/* determine rx_buf space */
941 	space = buffer_len - smc_curs_diff(buffer_len, &cons, &prod);
942 	return space;
943 }
944 
945 static int smc_switch_cursor(struct smc_sock *smc, struct smc_cdc_tx_pend *pend,
946 			     struct smc_wr_buf *wr_buf)
947 {
948 	struct smc_connection *conn = &smc->conn;
949 	union smc_host_cursor cons, fin;
950 	int rc = 0;
951 	int diff;
952 
953 	smc_curs_copy(&conn->tx_curs_sent, &conn->tx_curs_fin, conn);
954 	smc_curs_copy(&fin, &conn->local_tx_ctrl_fin, conn);
955 	/* set prod cursor to old state, enforce tx_rdma_writes() */
956 	smc_curs_copy(&conn->local_tx_ctrl.prod, &fin, conn);
957 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
958 
959 	if (smc_curs_comp(conn->peer_rmbe_size, &cons, &fin) < 0) {
960 		/* cons cursor advanced more than fin, and prod was set
961 		 * fin above, so now prod is smaller than cons. Fix that.
962 		 */
963 		diff = smc_curs_diff(conn->peer_rmbe_size, &fin, &cons);
964 		smc_curs_add(conn->sndbuf_desc->len,
965 			     &conn->tx_curs_sent, diff);
966 		smc_curs_add(conn->sndbuf_desc->len,
967 			     &conn->tx_curs_fin, diff);
968 
969 		smp_mb__before_atomic();
970 		atomic_add(diff, &conn->sndbuf_space);
971 		smp_mb__after_atomic();
972 
973 		smc_curs_add(conn->peer_rmbe_size,
974 			     &conn->local_tx_ctrl.prod, diff);
975 		smc_curs_add(conn->peer_rmbe_size,
976 			     &conn->local_tx_ctrl_fin, diff);
977 	}
978 	/* recalculate, value is used by tx_rdma_writes() */
979 	atomic_set(&smc->conn.peer_rmbe_space, smc_write_space(conn));
980 
981 	if (smc->sk.sk_state != SMC_INIT &&
982 	    smc->sk.sk_state != SMC_CLOSED) {
983 		rc = smcr_cdc_msg_send_validation(conn, pend, wr_buf);
984 		if (!rc) {
985 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work, 0);
986 			smc->sk.sk_data_ready(&smc->sk);
987 		}
988 	} else {
989 		smc_wr_tx_put_slot(conn->lnk,
990 				   (struct smc_wr_tx_pend_priv *)pend);
991 	}
992 	return rc;
993 }
994 
995 void smc_switch_link_and_count(struct smc_connection *conn,
996 			       struct smc_link *to_lnk)
997 {
998 	atomic_dec(&conn->lnk->conn_cnt);
999 	conn->lnk = to_lnk;
1000 	atomic_inc(&conn->lnk->conn_cnt);
1001 }
1002 
1003 struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
1004 				  struct smc_link *from_lnk, bool is_dev_err)
1005 {
1006 	struct smc_link *to_lnk = NULL;
1007 	struct smc_cdc_tx_pend *pend;
1008 	struct smc_connection *conn;
1009 	struct smc_wr_buf *wr_buf;
1010 	struct smc_sock *smc;
1011 	struct rb_node *node;
1012 	int i, rc = 0;
1013 
1014 	/* link is inactive, wake up tx waiters */
1015 	smc_wr_wakeup_tx_wait(from_lnk);
1016 
1017 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1018 		if (!smc_link_active(&lgr->lnk[i]) || i == from_lnk->link_idx)
1019 			continue;
1020 		if (is_dev_err && from_lnk->smcibdev == lgr->lnk[i].smcibdev &&
1021 		    from_lnk->ibport == lgr->lnk[i].ibport) {
1022 			continue;
1023 		}
1024 		to_lnk = &lgr->lnk[i];
1025 		break;
1026 	}
1027 	if (!to_lnk || !smc_wr_tx_link_hold(to_lnk)) {
1028 		smc_lgr_terminate_sched(lgr);
1029 		return NULL;
1030 	}
1031 again:
1032 	read_lock_bh(&lgr->conns_lock);
1033 	for (node = rb_first(&lgr->conns_all); node; node = rb_next(node)) {
1034 		conn = rb_entry(node, struct smc_connection, alert_node);
1035 		if (conn->lnk != from_lnk)
1036 			continue;
1037 		smc = container_of(conn, struct smc_sock, conn);
1038 		/* conn->lnk not yet set in SMC_INIT state */
1039 		if (smc->sk.sk_state == SMC_INIT)
1040 			continue;
1041 		if (smc->sk.sk_state == SMC_CLOSED ||
1042 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT1 ||
1043 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT2 ||
1044 		    smc->sk.sk_state == SMC_APPFINCLOSEWAIT ||
1045 		    smc->sk.sk_state == SMC_APPCLOSEWAIT1 ||
1046 		    smc->sk.sk_state == SMC_APPCLOSEWAIT2 ||
1047 		    smc->sk.sk_state == SMC_PEERFINCLOSEWAIT ||
1048 		    smc->sk.sk_state == SMC_PEERABORTWAIT ||
1049 		    smc->sk.sk_state == SMC_PROCESSABORT) {
1050 			spin_lock_bh(&conn->send_lock);
1051 			smc_switch_link_and_count(conn, to_lnk);
1052 			spin_unlock_bh(&conn->send_lock);
1053 			continue;
1054 		}
1055 		sock_hold(&smc->sk);
1056 		read_unlock_bh(&lgr->conns_lock);
1057 		/* pre-fetch buffer outside of send_lock, might sleep */
1058 		rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
1059 		if (rc)
1060 			goto err_out;
1061 		/* avoid race with smcr_tx_sndbuf_nonempty() */
1062 		spin_lock_bh(&conn->send_lock);
1063 		smc_switch_link_and_count(conn, to_lnk);
1064 		rc = smc_switch_cursor(smc, pend, wr_buf);
1065 		spin_unlock_bh(&conn->send_lock);
1066 		sock_put(&smc->sk);
1067 		if (rc)
1068 			goto err_out;
1069 		goto again;
1070 	}
1071 	read_unlock_bh(&lgr->conns_lock);
1072 	smc_wr_tx_link_put(to_lnk);
1073 	return to_lnk;
1074 
1075 err_out:
1076 	smcr_link_down_cond_sched(to_lnk);
1077 	smc_wr_tx_link_put(to_lnk);
1078 	return NULL;
1079 }
1080 
1081 static void smcr_buf_unuse(struct smc_buf_desc *rmb_desc,
1082 			   struct smc_link_group *lgr)
1083 {
1084 	int rc;
1085 
1086 	if (rmb_desc->is_conf_rkey && !list_empty(&lgr->list)) {
1087 		/* unregister rmb with peer */
1088 		rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
1089 		if (!rc) {
1090 			/* protect against smc_llc_cli_rkey_exchange() */
1091 			mutex_lock(&lgr->llc_conf_mutex);
1092 			smc_llc_do_delete_rkey(lgr, rmb_desc);
1093 			rmb_desc->is_conf_rkey = false;
1094 			mutex_unlock(&lgr->llc_conf_mutex);
1095 			smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1096 		}
1097 	}
1098 
1099 	if (rmb_desc->is_reg_err) {
1100 		/* buf registration failed, reuse not possible */
1101 		mutex_lock(&lgr->rmbs_lock);
1102 		list_del(&rmb_desc->list);
1103 		mutex_unlock(&lgr->rmbs_lock);
1104 
1105 		smc_buf_free(lgr, true, rmb_desc);
1106 	} else {
1107 		rmb_desc->used = 0;
1108 		memset(rmb_desc->cpu_addr, 0, rmb_desc->len);
1109 	}
1110 }
1111 
1112 static void smc_buf_unuse(struct smc_connection *conn,
1113 			  struct smc_link_group *lgr)
1114 {
1115 	if (conn->sndbuf_desc) {
1116 		conn->sndbuf_desc->used = 0;
1117 		memset(conn->sndbuf_desc->cpu_addr, 0, conn->sndbuf_desc->len);
1118 	}
1119 	if (conn->rmb_desc && lgr->is_smcd) {
1120 		conn->rmb_desc->used = 0;
1121 		memset(conn->rmb_desc->cpu_addr, 0, conn->rmb_desc->len +
1122 		       sizeof(struct smcd_cdc_msg));
1123 	} else if (conn->rmb_desc) {
1124 		smcr_buf_unuse(conn->rmb_desc, lgr);
1125 	}
1126 }
1127 
1128 /* remove a finished connection from its link group */
1129 void smc_conn_free(struct smc_connection *conn)
1130 {
1131 	struct smc_link_group *lgr = conn->lgr;
1132 
1133 	if (!lgr)
1134 		return;
1135 	if (lgr->is_smcd) {
1136 		if (!list_empty(&lgr->list))
1137 			smc_ism_unset_conn(conn);
1138 		tasklet_kill(&conn->rx_tsklet);
1139 	} else {
1140 		smc_cdc_wait_pend_tx_wr(conn);
1141 		if (current_work() != &conn->abort_work)
1142 			cancel_work_sync(&conn->abort_work);
1143 	}
1144 	if (!list_empty(&lgr->list)) {
1145 		smc_lgr_unregister_conn(conn);
1146 		smc_buf_unuse(conn, lgr); /* allow buffer reuse */
1147 	}
1148 
1149 	if (!lgr->conns_num)
1150 		smc_lgr_schedule_free_work(lgr);
1151 }
1152 
1153 /* unregister a link from a buf_desc */
1154 static void smcr_buf_unmap_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1155 				struct smc_link *lnk)
1156 {
1157 	if (is_rmb)
1158 		buf_desc->is_reg_mr[lnk->link_idx] = false;
1159 	if (!buf_desc->is_map_ib[lnk->link_idx])
1160 		return;
1161 	if (is_rmb) {
1162 		if (buf_desc->mr_rx[lnk->link_idx]) {
1163 			smc_ib_put_memory_region(
1164 					buf_desc->mr_rx[lnk->link_idx]);
1165 			buf_desc->mr_rx[lnk->link_idx] = NULL;
1166 		}
1167 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_FROM_DEVICE);
1168 	} else {
1169 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_TO_DEVICE);
1170 	}
1171 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1172 	buf_desc->is_map_ib[lnk->link_idx] = false;
1173 }
1174 
1175 /* unmap all buffers of lgr for a deleted link */
1176 static void smcr_buf_unmap_lgr(struct smc_link *lnk)
1177 {
1178 	struct smc_link_group *lgr = lnk->lgr;
1179 	struct smc_buf_desc *buf_desc, *bf;
1180 	int i;
1181 
1182 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1183 		mutex_lock(&lgr->rmbs_lock);
1184 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list)
1185 			smcr_buf_unmap_link(buf_desc, true, lnk);
1186 		mutex_unlock(&lgr->rmbs_lock);
1187 		mutex_lock(&lgr->sndbufs_lock);
1188 		list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i],
1189 					 list)
1190 			smcr_buf_unmap_link(buf_desc, false, lnk);
1191 		mutex_unlock(&lgr->sndbufs_lock);
1192 	}
1193 }
1194 
1195 static void smcr_rtoken_clear_link(struct smc_link *lnk)
1196 {
1197 	struct smc_link_group *lgr = lnk->lgr;
1198 	int i;
1199 
1200 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
1201 		lgr->rtokens[i][lnk->link_idx].rkey = 0;
1202 		lgr->rtokens[i][lnk->link_idx].dma_addr = 0;
1203 	}
1204 }
1205 
1206 /* must be called under lgr->llc_conf_mutex lock */
1207 void smcr_link_clear(struct smc_link *lnk, bool log)
1208 {
1209 	struct smc_ib_device *smcibdev;
1210 
1211 	if (!lnk->lgr || lnk->state == SMC_LNK_UNUSED)
1212 		return;
1213 	lnk->peer_qpn = 0;
1214 	smc_llc_link_clear(lnk, log);
1215 	smcr_buf_unmap_lgr(lnk);
1216 	smcr_rtoken_clear_link(lnk);
1217 	smc_ib_modify_qp_error(lnk);
1218 	smc_wr_free_link(lnk);
1219 	smc_ib_destroy_queue_pair(lnk);
1220 	smc_ib_dealloc_protection_domain(lnk);
1221 	smc_wr_free_link_mem(lnk);
1222 	smc_ibdev_cnt_dec(lnk);
1223 	put_device(&lnk->smcibdev->ibdev->dev);
1224 	smcibdev = lnk->smcibdev;
1225 	memset(lnk, 0, sizeof(struct smc_link));
1226 	lnk->state = SMC_LNK_UNUSED;
1227 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
1228 		wake_up(&smcibdev->lnks_deleted);
1229 }
1230 
1231 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb,
1232 			  struct smc_buf_desc *buf_desc)
1233 {
1234 	int i;
1235 
1236 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1237 		smcr_buf_unmap_link(buf_desc, is_rmb, &lgr->lnk[i]);
1238 
1239 	if (buf_desc->pages)
1240 		__free_pages(buf_desc->pages, buf_desc->order);
1241 	kfree(buf_desc);
1242 }
1243 
1244 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb,
1245 			  struct smc_buf_desc *buf_desc)
1246 {
1247 	if (is_dmb) {
1248 		/* restore original buf len */
1249 		buf_desc->len += sizeof(struct smcd_cdc_msg);
1250 		smc_ism_unregister_dmb(lgr->smcd, buf_desc);
1251 	} else {
1252 		kfree(buf_desc->cpu_addr);
1253 	}
1254 	kfree(buf_desc);
1255 }
1256 
1257 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
1258 			 struct smc_buf_desc *buf_desc)
1259 {
1260 	if (lgr->is_smcd)
1261 		smcd_buf_free(lgr, is_rmb, buf_desc);
1262 	else
1263 		smcr_buf_free(lgr, is_rmb, buf_desc);
1264 }
1265 
1266 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb)
1267 {
1268 	struct smc_buf_desc *buf_desc, *bf_desc;
1269 	struct list_head *buf_list;
1270 	int i;
1271 
1272 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1273 		if (is_rmb)
1274 			buf_list = &lgr->rmbs[i];
1275 		else
1276 			buf_list = &lgr->sndbufs[i];
1277 		list_for_each_entry_safe(buf_desc, bf_desc, buf_list,
1278 					 list) {
1279 			list_del(&buf_desc->list);
1280 			smc_buf_free(lgr, is_rmb, buf_desc);
1281 		}
1282 	}
1283 }
1284 
1285 static void smc_lgr_free_bufs(struct smc_link_group *lgr)
1286 {
1287 	/* free send buffers */
1288 	__smc_lgr_free_bufs(lgr, false);
1289 	/* free rmbs */
1290 	__smc_lgr_free_bufs(lgr, true);
1291 }
1292 
1293 /* remove a link group */
1294 static void smc_lgr_free(struct smc_link_group *lgr)
1295 {
1296 	int i;
1297 
1298 	if (!lgr->is_smcd) {
1299 		mutex_lock(&lgr->llc_conf_mutex);
1300 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1301 			if (lgr->lnk[i].state != SMC_LNK_UNUSED)
1302 				smcr_link_clear(&lgr->lnk[i], false);
1303 		}
1304 		mutex_unlock(&lgr->llc_conf_mutex);
1305 		smc_llc_lgr_clear(lgr);
1306 	}
1307 
1308 	smc_lgr_free_bufs(lgr);
1309 	destroy_workqueue(lgr->tx_wq);
1310 	if (lgr->is_smcd) {
1311 		smc_ism_put_vlan(lgr->smcd, lgr->vlan_id);
1312 		put_device(&lgr->smcd->dev);
1313 		if (!atomic_dec_return(&lgr->smcd->lgr_cnt))
1314 			wake_up(&lgr->smcd->lgrs_deleted);
1315 	} else {
1316 		smc_wr_free_lgr_mem(lgr);
1317 		if (!atomic_dec_return(&lgr_cnt))
1318 			wake_up(&lgrs_deleted);
1319 	}
1320 	kfree(lgr);
1321 }
1322 
1323 static void smc_sk_wake_ups(struct smc_sock *smc)
1324 {
1325 	smc->sk.sk_write_space(&smc->sk);
1326 	smc->sk.sk_data_ready(&smc->sk);
1327 	smc->sk.sk_state_change(&smc->sk);
1328 }
1329 
1330 /* kill a connection */
1331 static void smc_conn_kill(struct smc_connection *conn, bool soft)
1332 {
1333 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1334 
1335 	if (conn->lgr->is_smcd && conn->lgr->peer_shutdown)
1336 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
1337 	else
1338 		smc_close_abort(conn);
1339 	conn->killed = 1;
1340 	smc->sk.sk_err = ECONNABORTED;
1341 	smc_sk_wake_ups(smc);
1342 	if (conn->lgr->is_smcd) {
1343 		smc_ism_unset_conn(conn);
1344 		if (soft)
1345 			tasklet_kill(&conn->rx_tsklet);
1346 		else
1347 			tasklet_unlock_wait(&conn->rx_tsklet);
1348 	} else {
1349 		smc_cdc_wait_pend_tx_wr(conn);
1350 	}
1351 	smc_lgr_unregister_conn(conn);
1352 	smc_close_active_abort(smc);
1353 }
1354 
1355 static void smc_lgr_cleanup(struct smc_link_group *lgr)
1356 {
1357 	if (lgr->is_smcd) {
1358 		smc_ism_signal_shutdown(lgr);
1359 	} else {
1360 		u32 rsn = lgr->llc_termination_rsn;
1361 
1362 		if (!rsn)
1363 			rsn = SMC_LLC_DEL_PROG_INIT_TERM;
1364 		smc_llc_send_link_delete_all(lgr, false, rsn);
1365 		smcr_lgr_link_deactivate_all(lgr);
1366 	}
1367 }
1368 
1369 /* terminate link group
1370  * @soft: true if link group shutdown can take its time
1371  *	  false if immediate link group shutdown is required
1372  */
1373 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft)
1374 {
1375 	struct smc_connection *conn;
1376 	struct smc_sock *smc;
1377 	struct rb_node *node;
1378 
1379 	if (lgr->terminating)
1380 		return;	/* lgr already terminating */
1381 	/* cancel free_work sync, will terminate when lgr->freeing is set */
1382 	cancel_delayed_work_sync(&lgr->free_work);
1383 	lgr->terminating = 1;
1384 
1385 	/* kill remaining link group connections */
1386 	read_lock_bh(&lgr->conns_lock);
1387 	node = rb_first(&lgr->conns_all);
1388 	while (node) {
1389 		read_unlock_bh(&lgr->conns_lock);
1390 		conn = rb_entry(node, struct smc_connection, alert_node);
1391 		smc = container_of(conn, struct smc_sock, conn);
1392 		sock_hold(&smc->sk); /* sock_put below */
1393 		lock_sock(&smc->sk);
1394 		smc_conn_kill(conn, soft);
1395 		release_sock(&smc->sk);
1396 		sock_put(&smc->sk); /* sock_hold above */
1397 		read_lock_bh(&lgr->conns_lock);
1398 		node = rb_first(&lgr->conns_all);
1399 	}
1400 	read_unlock_bh(&lgr->conns_lock);
1401 	smc_lgr_cleanup(lgr);
1402 	smc_lgr_free(lgr);
1403 }
1404 
1405 /* unlink link group and schedule termination */
1406 void smc_lgr_terminate_sched(struct smc_link_group *lgr)
1407 {
1408 	spinlock_t *lgr_lock;
1409 
1410 	smc_lgr_list_head(lgr, &lgr_lock);
1411 	spin_lock_bh(lgr_lock);
1412 	if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) {
1413 		spin_unlock_bh(lgr_lock);
1414 		return;	/* lgr already terminating */
1415 	}
1416 	list_del_init(&lgr->list);
1417 	lgr->freeing = 1;
1418 	spin_unlock_bh(lgr_lock);
1419 	schedule_work(&lgr->terminate_work);
1420 }
1421 
1422 /* Called when peer lgr shutdown (regularly or abnormally) is received */
1423 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan)
1424 {
1425 	struct smc_link_group *lgr, *l;
1426 	LIST_HEAD(lgr_free_list);
1427 
1428 	/* run common cleanup function and build free list */
1429 	spin_lock_bh(&dev->lgr_lock);
1430 	list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) {
1431 		if ((!peer_gid || lgr->peer_gid == peer_gid) &&
1432 		    (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) {
1433 			if (peer_gid) /* peer triggered termination */
1434 				lgr->peer_shutdown = 1;
1435 			list_move(&lgr->list, &lgr_free_list);
1436 			lgr->freeing = 1;
1437 		}
1438 	}
1439 	spin_unlock_bh(&dev->lgr_lock);
1440 
1441 	/* cancel the regular free workers and actually free lgrs */
1442 	list_for_each_entry_safe(lgr, l, &lgr_free_list, list) {
1443 		list_del_init(&lgr->list);
1444 		schedule_work(&lgr->terminate_work);
1445 	}
1446 }
1447 
1448 /* Called when an SMCD device is removed or the smc module is unloaded */
1449 void smc_smcd_terminate_all(struct smcd_dev *smcd)
1450 {
1451 	struct smc_link_group *lgr, *lg;
1452 	LIST_HEAD(lgr_free_list);
1453 
1454 	spin_lock_bh(&smcd->lgr_lock);
1455 	list_splice_init(&smcd->lgr_list, &lgr_free_list);
1456 	list_for_each_entry(lgr, &lgr_free_list, list)
1457 		lgr->freeing = 1;
1458 	spin_unlock_bh(&smcd->lgr_lock);
1459 
1460 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1461 		list_del_init(&lgr->list);
1462 		__smc_lgr_terminate(lgr, false);
1463 	}
1464 
1465 	if (atomic_read(&smcd->lgr_cnt))
1466 		wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt));
1467 }
1468 
1469 /* Called when an SMCR device is removed or the smc module is unloaded.
1470  * If smcibdev is given, all SMCR link groups using this device are terminated.
1471  * If smcibdev is NULL, all SMCR link groups are terminated.
1472  *
1473  * We must wait here for QPs been destroyed before we destroy the CQs,
1474  * or we won't received any CQEs and cdc_pend_tx_wr cannot reach 0 thus
1475  * smc_sock cannot be released.
1476  */
1477 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev)
1478 {
1479 	struct smc_link_group *lgr, *lg;
1480 	LIST_HEAD(lgr_free_list);
1481 	LIST_HEAD(lgr_linkdown_list);
1482 	int i;
1483 
1484 	spin_lock_bh(&smc_lgr_list.lock);
1485 	if (!smcibdev) {
1486 		list_splice_init(&smc_lgr_list.list, &lgr_free_list);
1487 		list_for_each_entry(lgr, &lgr_free_list, list)
1488 			lgr->freeing = 1;
1489 	} else {
1490 		list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) {
1491 			for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1492 				if (lgr->lnk[i].smcibdev == smcibdev)
1493 					list_move_tail(&lgr->list, &lgr_linkdown_list);
1494 			}
1495 		}
1496 	}
1497 	spin_unlock_bh(&smc_lgr_list.lock);
1498 
1499 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1500 		list_del_init(&lgr->list);
1501 		smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_OP_INIT_TERM);
1502 		__smc_lgr_terminate(lgr, false);
1503 	}
1504 
1505 	list_for_each_entry_safe(lgr, lg, &lgr_linkdown_list, list) {
1506 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1507 			if (lgr->lnk[i].smcibdev == smcibdev) {
1508 				mutex_lock(&lgr->llc_conf_mutex);
1509 				smcr_link_down_cond(&lgr->lnk[i]);
1510 				mutex_unlock(&lgr->llc_conf_mutex);
1511 			}
1512 		}
1513 	}
1514 
1515 	if (smcibdev) {
1516 		if (atomic_read(&smcibdev->lnk_cnt))
1517 			wait_event(smcibdev->lnks_deleted,
1518 				   !atomic_read(&smcibdev->lnk_cnt));
1519 	} else {
1520 		if (atomic_read(&lgr_cnt))
1521 			wait_event(lgrs_deleted, !atomic_read(&lgr_cnt));
1522 	}
1523 }
1524 
1525 /* set new lgr type and clear all asymmetric link tagging */
1526 void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type)
1527 {
1528 	char *lgr_type = "";
1529 	int i;
1530 
1531 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1532 		if (smc_link_usable(&lgr->lnk[i]))
1533 			lgr->lnk[i].link_is_asym = false;
1534 	if (lgr->type == new_type)
1535 		return;
1536 	lgr->type = new_type;
1537 
1538 	switch (lgr->type) {
1539 	case SMC_LGR_NONE:
1540 		lgr_type = "NONE";
1541 		break;
1542 	case SMC_LGR_SINGLE:
1543 		lgr_type = "SINGLE";
1544 		break;
1545 	case SMC_LGR_SYMMETRIC:
1546 		lgr_type = "SYMMETRIC";
1547 		break;
1548 	case SMC_LGR_ASYMMETRIC_PEER:
1549 		lgr_type = "ASYMMETRIC_PEER";
1550 		break;
1551 	case SMC_LGR_ASYMMETRIC_LOCAL:
1552 		lgr_type = "ASYMMETRIC_LOCAL";
1553 		break;
1554 	}
1555 	pr_warn_ratelimited("smc: SMC-R lg %*phN net %llu state changed: "
1556 			    "%s, pnetid %.16s\n", SMC_LGR_ID_SIZE, &lgr->id,
1557 			    lgr->net->net_cookie, lgr_type, lgr->pnet_id);
1558 }
1559 
1560 /* set new lgr type and tag a link as asymmetric */
1561 void smcr_lgr_set_type_asym(struct smc_link_group *lgr,
1562 			    enum smc_lgr_type new_type, int asym_lnk_idx)
1563 {
1564 	smcr_lgr_set_type(lgr, new_type);
1565 	lgr->lnk[asym_lnk_idx].link_is_asym = true;
1566 }
1567 
1568 /* abort connection, abort_work scheduled from tasklet context */
1569 static void smc_conn_abort_work(struct work_struct *work)
1570 {
1571 	struct smc_connection *conn = container_of(work,
1572 						   struct smc_connection,
1573 						   abort_work);
1574 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1575 
1576 	lock_sock(&smc->sk);
1577 	smc_conn_kill(conn, true);
1578 	release_sock(&smc->sk);
1579 	sock_put(&smc->sk); /* sock_hold done by schedulers of abort_work */
1580 }
1581 
1582 void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport)
1583 {
1584 	struct smc_link_group *lgr, *n;
1585 
1586 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1587 		struct smc_link *link;
1588 
1589 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1590 			    SMC_MAX_PNETID_LEN) ||
1591 		    lgr->type == SMC_LGR_SYMMETRIC ||
1592 		    lgr->type == SMC_LGR_ASYMMETRIC_PEER ||
1593 		    !rdma_dev_access_netns(smcibdev->ibdev, lgr->net))
1594 			continue;
1595 
1596 		/* trigger local add link processing */
1597 		link = smc_llc_usable_link(lgr);
1598 		if (link)
1599 			smc_llc_add_link_local(link);
1600 	}
1601 }
1602 
1603 /* link is down - switch connections to alternate link,
1604  * must be called under lgr->llc_conf_mutex lock
1605  */
1606 static void smcr_link_down(struct smc_link *lnk)
1607 {
1608 	struct smc_link_group *lgr = lnk->lgr;
1609 	struct smc_link *to_lnk;
1610 	int del_link_id;
1611 
1612 	if (!lgr || lnk->state == SMC_LNK_UNUSED || list_empty(&lgr->list))
1613 		return;
1614 
1615 	to_lnk = smc_switch_conns(lgr, lnk, true);
1616 	if (!to_lnk) { /* no backup link available */
1617 		smcr_link_clear(lnk, true);
1618 		return;
1619 	}
1620 	smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1621 	del_link_id = lnk->link_id;
1622 
1623 	if (lgr->role == SMC_SERV) {
1624 		/* trigger local delete link processing */
1625 		smc_llc_srv_delete_link_local(to_lnk, del_link_id);
1626 	} else {
1627 		if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) {
1628 			/* another llc task is ongoing */
1629 			mutex_unlock(&lgr->llc_conf_mutex);
1630 			wait_event_timeout(lgr->llc_flow_waiter,
1631 				(list_empty(&lgr->list) ||
1632 				 lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE),
1633 				SMC_LLC_WAIT_TIME);
1634 			mutex_lock(&lgr->llc_conf_mutex);
1635 		}
1636 		if (!list_empty(&lgr->list)) {
1637 			smc_llc_send_delete_link(to_lnk, del_link_id,
1638 						 SMC_LLC_REQ, true,
1639 						 SMC_LLC_DEL_LOST_PATH);
1640 			smcr_link_clear(lnk, true);
1641 		}
1642 		wake_up(&lgr->llc_flow_waiter);	/* wake up next waiter */
1643 	}
1644 }
1645 
1646 /* must be called under lgr->llc_conf_mutex lock */
1647 void smcr_link_down_cond(struct smc_link *lnk)
1648 {
1649 	if (smc_link_downing(&lnk->state)) {
1650 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1651 		smcr_link_down(lnk);
1652 	}
1653 }
1654 
1655 /* will get the lgr->llc_conf_mutex lock */
1656 void smcr_link_down_cond_sched(struct smc_link *lnk)
1657 {
1658 	if (smc_link_downing(&lnk->state)) {
1659 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1660 		schedule_work(&lnk->link_down_wrk);
1661 	}
1662 }
1663 
1664 void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport)
1665 {
1666 	struct smc_link_group *lgr, *n;
1667 	int i;
1668 
1669 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1670 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1671 			    SMC_MAX_PNETID_LEN))
1672 			continue; /* lgr is not affected */
1673 		if (list_empty(&lgr->list))
1674 			continue;
1675 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1676 			struct smc_link *lnk = &lgr->lnk[i];
1677 
1678 			if (smc_link_usable(lnk) &&
1679 			    lnk->smcibdev == smcibdev && lnk->ibport == ibport)
1680 				smcr_link_down_cond_sched(lnk);
1681 		}
1682 	}
1683 }
1684 
1685 static void smc_link_down_work(struct work_struct *work)
1686 {
1687 	struct smc_link *link = container_of(work, struct smc_link,
1688 					     link_down_wrk);
1689 	struct smc_link_group *lgr = link->lgr;
1690 
1691 	if (list_empty(&lgr->list))
1692 		return;
1693 	wake_up_all(&lgr->llc_msg_waiter);
1694 	mutex_lock(&lgr->llc_conf_mutex);
1695 	smcr_link_down(link);
1696 	mutex_unlock(&lgr->llc_conf_mutex);
1697 }
1698 
1699 static int smc_vlan_by_tcpsk_walk(struct net_device *lower_dev,
1700 				  struct netdev_nested_priv *priv)
1701 {
1702 	unsigned short *vlan_id = (unsigned short *)priv->data;
1703 
1704 	if (is_vlan_dev(lower_dev)) {
1705 		*vlan_id = vlan_dev_vlan_id(lower_dev);
1706 		return 1;
1707 	}
1708 
1709 	return 0;
1710 }
1711 
1712 /* Determine vlan of internal TCP socket. */
1713 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini)
1714 {
1715 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
1716 	struct netdev_nested_priv priv;
1717 	struct net_device *ndev;
1718 	int rc = 0;
1719 
1720 	ini->vlan_id = 0;
1721 	if (!dst) {
1722 		rc = -ENOTCONN;
1723 		goto out;
1724 	}
1725 	if (!dst->dev) {
1726 		rc = -ENODEV;
1727 		goto out_rel;
1728 	}
1729 
1730 	ndev = dst->dev;
1731 	if (is_vlan_dev(ndev)) {
1732 		ini->vlan_id = vlan_dev_vlan_id(ndev);
1733 		goto out_rel;
1734 	}
1735 
1736 	priv.data = (void *)&ini->vlan_id;
1737 	rtnl_lock();
1738 	netdev_walk_all_lower_dev(ndev, smc_vlan_by_tcpsk_walk, &priv);
1739 	rtnl_unlock();
1740 
1741 out_rel:
1742 	dst_release(dst);
1743 out:
1744 	return rc;
1745 }
1746 
1747 static bool smcr_lgr_match(struct smc_link_group *lgr, u8 smcr_version,
1748 			   u8 peer_systemid[],
1749 			   u8 peer_gid[],
1750 			   u8 peer_mac_v1[],
1751 			   enum smc_lgr_role role, u32 clcqpn,
1752 			   struct net *net)
1753 {
1754 	struct smc_link *lnk;
1755 	int i;
1756 
1757 	if (memcmp(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN) ||
1758 	    lgr->role != role)
1759 		return false;
1760 
1761 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1762 		lnk = &lgr->lnk[i];
1763 
1764 		if (!smc_link_active(lnk))
1765 			continue;
1766 		/* use verbs API to check netns, instead of lgr->net */
1767 		if (!rdma_dev_access_netns(lnk->smcibdev->ibdev, net))
1768 			return false;
1769 		if ((lgr->role == SMC_SERV || lnk->peer_qpn == clcqpn) &&
1770 		    !memcmp(lnk->peer_gid, peer_gid, SMC_GID_SIZE) &&
1771 		    (smcr_version == SMC_V2 ||
1772 		     !memcmp(lnk->peer_mac, peer_mac_v1, ETH_ALEN)))
1773 			return true;
1774 	}
1775 	return false;
1776 }
1777 
1778 static bool smcd_lgr_match(struct smc_link_group *lgr,
1779 			   struct smcd_dev *smcismdev, u64 peer_gid)
1780 {
1781 	return lgr->peer_gid == peer_gid && lgr->smcd == smcismdev;
1782 }
1783 
1784 /* create a new SMC connection (and a new link group if necessary) */
1785 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
1786 {
1787 	struct smc_connection *conn = &smc->conn;
1788 	struct net *net = sock_net(&smc->sk);
1789 	struct list_head *lgr_list;
1790 	struct smc_link_group *lgr;
1791 	enum smc_lgr_role role;
1792 	spinlock_t *lgr_lock;
1793 	int rc = 0;
1794 
1795 	lgr_list = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_list :
1796 				  &smc_lgr_list.list;
1797 	lgr_lock = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_lock :
1798 				  &smc_lgr_list.lock;
1799 	ini->first_contact_local = 1;
1800 	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
1801 	if (role == SMC_CLNT && ini->first_contact_peer)
1802 		/* create new link group as well */
1803 		goto create;
1804 
1805 	/* determine if an existing link group can be reused */
1806 	spin_lock_bh(lgr_lock);
1807 	list_for_each_entry(lgr, lgr_list, list) {
1808 		write_lock_bh(&lgr->conns_lock);
1809 		if ((ini->is_smcd ?
1810 		     smcd_lgr_match(lgr, ini->ism_dev[ini->ism_selected],
1811 				    ini->ism_peer_gid[ini->ism_selected]) :
1812 		     smcr_lgr_match(lgr, ini->smcr_version,
1813 				    ini->peer_systemid,
1814 				    ini->peer_gid, ini->peer_mac, role,
1815 				    ini->ib_clcqpn, net)) &&
1816 		    !lgr->sync_err &&
1817 		    (ini->smcd_version == SMC_V2 ||
1818 		     lgr->vlan_id == ini->vlan_id) &&
1819 		    (role == SMC_CLNT || ini->is_smcd ||
1820 		     lgr->conns_num < SMC_RMBS_PER_LGR_MAX)) {
1821 			/* link group found */
1822 			ini->first_contact_local = 0;
1823 			conn->lgr = lgr;
1824 			rc = smc_lgr_register_conn(conn, false);
1825 			write_unlock_bh(&lgr->conns_lock);
1826 			if (!rc && delayed_work_pending(&lgr->free_work))
1827 				cancel_delayed_work(&lgr->free_work);
1828 			break;
1829 		}
1830 		write_unlock_bh(&lgr->conns_lock);
1831 	}
1832 	spin_unlock_bh(lgr_lock);
1833 	if (rc)
1834 		return rc;
1835 
1836 	if (role == SMC_CLNT && !ini->first_contact_peer &&
1837 	    ini->first_contact_local) {
1838 		/* Server reuses a link group, but Client wants to start
1839 		 * a new one
1840 		 * send out_of_sync decline, reason synchr. error
1841 		 */
1842 		return SMC_CLC_DECL_SYNCERR;
1843 	}
1844 
1845 create:
1846 	if (ini->first_contact_local) {
1847 		rc = smc_lgr_create(smc, ini);
1848 		if (rc)
1849 			goto out;
1850 		lgr = conn->lgr;
1851 		write_lock_bh(&lgr->conns_lock);
1852 		rc = smc_lgr_register_conn(conn, true);
1853 		write_unlock_bh(&lgr->conns_lock);
1854 		if (rc) {
1855 			smc_lgr_cleanup_early(lgr);
1856 			goto out;
1857 		}
1858 	}
1859 	conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
1860 	conn->local_tx_ctrl.len = SMC_WR_TX_SIZE;
1861 	conn->urg_state = SMC_URG_READ;
1862 	init_waitqueue_head(&conn->cdc_pend_tx_wq);
1863 	INIT_WORK(&smc->conn.abort_work, smc_conn_abort_work);
1864 	if (ini->is_smcd) {
1865 		conn->rx_off = sizeof(struct smcd_cdc_msg);
1866 		smcd_cdc_rx_init(conn); /* init tasklet for this conn */
1867 	} else {
1868 		conn->rx_off = 0;
1869 	}
1870 #ifndef KERNEL_HAS_ATOMIC64
1871 	spin_lock_init(&conn->acurs_lock);
1872 #endif
1873 
1874 out:
1875 	return rc;
1876 }
1877 
1878 #define SMCD_DMBE_SIZES		6 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */
1879 #define SMCR_RMBE_SIZES		5 /* 0 -> 16KB, 1 -> 32KB, .. 5 -> 512KB */
1880 
1881 /* convert the RMB size into the compressed notation (minimum 16K, see
1882  * SMCD/R_DMBE_SIZES.
1883  * In contrast to plain ilog2, this rounds towards the next power of 2,
1884  * so the socket application gets at least its desired sndbuf / rcvbuf size.
1885  */
1886 static u8 smc_compress_bufsize(int size, bool is_smcd, bool is_rmb)
1887 {
1888 	const unsigned int max_scat = SG_MAX_SINGLE_ALLOC * PAGE_SIZE;
1889 	u8 compressed;
1890 
1891 	if (size <= SMC_BUF_MIN_SIZE)
1892 		return 0;
1893 
1894 	size = (size - 1) >> 14;  /* convert to 16K multiple */
1895 	compressed = min_t(u8, ilog2(size) + 1,
1896 			   is_smcd ? SMCD_DMBE_SIZES : SMCR_RMBE_SIZES);
1897 
1898 	if (!is_smcd && is_rmb)
1899 		/* RMBs are backed by & limited to max size of scatterlists */
1900 		compressed = min_t(u8, compressed, ilog2(max_scat >> 14));
1901 
1902 	return compressed;
1903 }
1904 
1905 /* convert the RMB size from compressed notation into integer */
1906 int smc_uncompress_bufsize(u8 compressed)
1907 {
1908 	u32 size;
1909 
1910 	size = 0x00000001 << (((int)compressed) + 14);
1911 	return (int)size;
1912 }
1913 
1914 /* try to reuse a sndbuf or rmb description slot for a certain
1915  * buffer size; if not available, return NULL
1916  */
1917 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize,
1918 					     struct mutex *lock,
1919 					     struct list_head *buf_list)
1920 {
1921 	struct smc_buf_desc *buf_slot;
1922 
1923 	mutex_lock(lock);
1924 	list_for_each_entry(buf_slot, buf_list, list) {
1925 		if (cmpxchg(&buf_slot->used, 0, 1) == 0) {
1926 			mutex_unlock(lock);
1927 			return buf_slot;
1928 		}
1929 	}
1930 	mutex_unlock(lock);
1931 	return NULL;
1932 }
1933 
1934 /* one of the conditions for announcing a receiver's current window size is
1935  * that it "results in a minimum increase in the window size of 10% of the
1936  * receive buffer space" [RFC7609]
1937  */
1938 static inline int smc_rmb_wnd_update_limit(int rmbe_size)
1939 {
1940 	return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
1941 }
1942 
1943 /* map an rmb buf to a link */
1944 static int smcr_buf_map_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1945 			     struct smc_link *lnk)
1946 {
1947 	int rc;
1948 
1949 	if (buf_desc->is_map_ib[lnk->link_idx])
1950 		return 0;
1951 
1952 	rc = sg_alloc_table(&buf_desc->sgt[lnk->link_idx], 1, GFP_KERNEL);
1953 	if (rc)
1954 		return rc;
1955 	sg_set_buf(buf_desc->sgt[lnk->link_idx].sgl,
1956 		   buf_desc->cpu_addr, buf_desc->len);
1957 
1958 	/* map sg table to DMA address */
1959 	rc = smc_ib_buf_map_sg(lnk, buf_desc,
1960 			       is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
1961 	/* SMC protocol depends on mapping to one DMA address only */
1962 	if (rc != 1) {
1963 		rc = -EAGAIN;
1964 		goto free_table;
1965 	}
1966 
1967 	/* create a new memory region for the RMB */
1968 	if (is_rmb) {
1969 		rc = smc_ib_get_memory_region(lnk->roce_pd,
1970 					      IB_ACCESS_REMOTE_WRITE |
1971 					      IB_ACCESS_LOCAL_WRITE,
1972 					      buf_desc, lnk->link_idx);
1973 		if (rc)
1974 			goto buf_unmap;
1975 		smc_ib_sync_sg_for_device(lnk, buf_desc, DMA_FROM_DEVICE);
1976 	}
1977 	buf_desc->is_map_ib[lnk->link_idx] = true;
1978 	return 0;
1979 
1980 buf_unmap:
1981 	smc_ib_buf_unmap_sg(lnk, buf_desc,
1982 			    is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
1983 free_table:
1984 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1985 	return rc;
1986 }
1987 
1988 /* register a new rmb on IB device,
1989  * must be called under lgr->llc_conf_mutex lock
1990  */
1991 int smcr_link_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc)
1992 {
1993 	if (list_empty(&link->lgr->list))
1994 		return -ENOLINK;
1995 	if (!rmb_desc->is_reg_mr[link->link_idx]) {
1996 		/* register memory region for new rmb */
1997 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[link->link_idx])) {
1998 			rmb_desc->is_reg_err = true;
1999 			return -EFAULT;
2000 		}
2001 		rmb_desc->is_reg_mr[link->link_idx] = true;
2002 	}
2003 	return 0;
2004 }
2005 
2006 static int _smcr_buf_map_lgr(struct smc_link *lnk, struct mutex *lock,
2007 			     struct list_head *lst, bool is_rmb)
2008 {
2009 	struct smc_buf_desc *buf_desc, *bf;
2010 	int rc = 0;
2011 
2012 	mutex_lock(lock);
2013 	list_for_each_entry_safe(buf_desc, bf, lst, list) {
2014 		if (!buf_desc->used)
2015 			continue;
2016 		rc = smcr_buf_map_link(buf_desc, is_rmb, lnk);
2017 		if (rc)
2018 			goto out;
2019 	}
2020 out:
2021 	mutex_unlock(lock);
2022 	return rc;
2023 }
2024 
2025 /* map all used buffers of lgr for a new link */
2026 int smcr_buf_map_lgr(struct smc_link *lnk)
2027 {
2028 	struct smc_link_group *lgr = lnk->lgr;
2029 	int i, rc = 0;
2030 
2031 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2032 		rc = _smcr_buf_map_lgr(lnk, &lgr->rmbs_lock,
2033 				       &lgr->rmbs[i], true);
2034 		if (rc)
2035 			return rc;
2036 		rc = _smcr_buf_map_lgr(lnk, &lgr->sndbufs_lock,
2037 				       &lgr->sndbufs[i], false);
2038 		if (rc)
2039 			return rc;
2040 	}
2041 	return 0;
2042 }
2043 
2044 /* register all used buffers of lgr for a new link,
2045  * must be called under lgr->llc_conf_mutex lock
2046  */
2047 int smcr_buf_reg_lgr(struct smc_link *lnk)
2048 {
2049 	struct smc_link_group *lgr = lnk->lgr;
2050 	struct smc_buf_desc *buf_desc, *bf;
2051 	int i, rc = 0;
2052 
2053 	mutex_lock(&lgr->rmbs_lock);
2054 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2055 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) {
2056 			if (!buf_desc->used)
2057 				continue;
2058 			rc = smcr_link_reg_rmb(lnk, buf_desc);
2059 			if (rc)
2060 				goto out;
2061 		}
2062 	}
2063 out:
2064 	mutex_unlock(&lgr->rmbs_lock);
2065 	return rc;
2066 }
2067 
2068 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr,
2069 						bool is_rmb, int bufsize)
2070 {
2071 	struct smc_buf_desc *buf_desc;
2072 
2073 	/* try to alloc a new buffer */
2074 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2075 	if (!buf_desc)
2076 		return ERR_PTR(-ENOMEM);
2077 
2078 	buf_desc->order = get_order(bufsize);
2079 	buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN |
2080 				      __GFP_NOMEMALLOC | __GFP_COMP |
2081 				      __GFP_NORETRY | __GFP_ZERO,
2082 				      buf_desc->order);
2083 	if (!buf_desc->pages) {
2084 		kfree(buf_desc);
2085 		return ERR_PTR(-EAGAIN);
2086 	}
2087 	buf_desc->cpu_addr = (void *)page_address(buf_desc->pages);
2088 	buf_desc->len = bufsize;
2089 	return buf_desc;
2090 }
2091 
2092 /* map buf_desc on all usable links,
2093  * unused buffers stay mapped as long as the link is up
2094  */
2095 static int smcr_buf_map_usable_links(struct smc_link_group *lgr,
2096 				     struct smc_buf_desc *buf_desc, bool is_rmb)
2097 {
2098 	int i, rc = 0;
2099 
2100 	/* protect against parallel link reconfiguration */
2101 	mutex_lock(&lgr->llc_conf_mutex);
2102 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2103 		struct smc_link *lnk = &lgr->lnk[i];
2104 
2105 		if (!smc_link_usable(lnk))
2106 			continue;
2107 		if (smcr_buf_map_link(buf_desc, is_rmb, lnk)) {
2108 			rc = -ENOMEM;
2109 			goto out;
2110 		}
2111 	}
2112 out:
2113 	mutex_unlock(&lgr->llc_conf_mutex);
2114 	return rc;
2115 }
2116 
2117 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr,
2118 						bool is_dmb, int bufsize)
2119 {
2120 	struct smc_buf_desc *buf_desc;
2121 	int rc;
2122 
2123 	/* try to alloc a new DMB */
2124 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2125 	if (!buf_desc)
2126 		return ERR_PTR(-ENOMEM);
2127 	if (is_dmb) {
2128 		rc = smc_ism_register_dmb(lgr, bufsize, buf_desc);
2129 		if (rc) {
2130 			kfree(buf_desc);
2131 			if (rc == -ENOMEM)
2132 				return ERR_PTR(-EAGAIN);
2133 			if (rc == -ENOSPC)
2134 				return ERR_PTR(-ENOSPC);
2135 			return ERR_PTR(-EIO);
2136 		}
2137 		buf_desc->pages = virt_to_page(buf_desc->cpu_addr);
2138 		/* CDC header stored in buf. So, pretend it was smaller */
2139 		buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg);
2140 	} else {
2141 		buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL |
2142 					     __GFP_NOWARN | __GFP_NORETRY |
2143 					     __GFP_NOMEMALLOC);
2144 		if (!buf_desc->cpu_addr) {
2145 			kfree(buf_desc);
2146 			return ERR_PTR(-EAGAIN);
2147 		}
2148 		buf_desc->len = bufsize;
2149 	}
2150 	return buf_desc;
2151 }
2152 
2153 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb)
2154 {
2155 	struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM);
2156 	struct smc_connection *conn = &smc->conn;
2157 	struct smc_link_group *lgr = conn->lgr;
2158 	struct list_head *buf_list;
2159 	int bufsize, bufsize_short;
2160 	bool is_dgraded = false;
2161 	struct mutex *lock;	/* lock buffer list */
2162 	int sk_buf_size;
2163 
2164 	if (is_rmb)
2165 		/* use socket recv buffer size (w/o overhead) as start value */
2166 		sk_buf_size = smc->sk.sk_rcvbuf / 2;
2167 	else
2168 		/* use socket send buffer size (w/o overhead) as start value */
2169 		sk_buf_size = smc->sk.sk_sndbuf / 2;
2170 
2171 	for (bufsize_short = smc_compress_bufsize(sk_buf_size, is_smcd, is_rmb);
2172 	     bufsize_short >= 0; bufsize_short--) {
2173 		if (is_rmb) {
2174 			lock = &lgr->rmbs_lock;
2175 			buf_list = &lgr->rmbs[bufsize_short];
2176 		} else {
2177 			lock = &lgr->sndbufs_lock;
2178 			buf_list = &lgr->sndbufs[bufsize_short];
2179 		}
2180 		bufsize = smc_uncompress_bufsize(bufsize_short);
2181 
2182 		/* check for reusable slot in the link group */
2183 		buf_desc = smc_buf_get_slot(bufsize_short, lock, buf_list);
2184 		if (buf_desc) {
2185 			SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2186 			SMC_STAT_BUF_REUSE(smc, is_smcd, is_rmb);
2187 			break; /* found reusable slot */
2188 		}
2189 
2190 		if (is_smcd)
2191 			buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize);
2192 		else
2193 			buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize);
2194 
2195 		if (PTR_ERR(buf_desc) == -ENOMEM)
2196 			break;
2197 		if (IS_ERR(buf_desc)) {
2198 			if (!is_dgraded) {
2199 				is_dgraded = true;
2200 				SMC_STAT_RMB_DOWNGRADED(smc, is_smcd, is_rmb);
2201 			}
2202 			continue;
2203 		}
2204 
2205 		SMC_STAT_RMB_ALLOC(smc, is_smcd, is_rmb);
2206 		SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2207 		buf_desc->used = 1;
2208 		mutex_lock(lock);
2209 		list_add(&buf_desc->list, buf_list);
2210 		mutex_unlock(lock);
2211 		break; /* found */
2212 	}
2213 
2214 	if (IS_ERR(buf_desc))
2215 		return PTR_ERR(buf_desc);
2216 
2217 	if (!is_smcd) {
2218 		if (smcr_buf_map_usable_links(lgr, buf_desc, is_rmb)) {
2219 			smcr_buf_unuse(buf_desc, lgr);
2220 			return -ENOMEM;
2221 		}
2222 	}
2223 
2224 	if (is_rmb) {
2225 		conn->rmb_desc = buf_desc;
2226 		conn->rmbe_size_short = bufsize_short;
2227 		smc->sk.sk_rcvbuf = bufsize * 2;
2228 		atomic_set(&conn->bytes_to_rcv, 0);
2229 		conn->rmbe_update_limit =
2230 			smc_rmb_wnd_update_limit(buf_desc->len);
2231 		if (is_smcd)
2232 			smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */
2233 	} else {
2234 		conn->sndbuf_desc = buf_desc;
2235 		smc->sk.sk_sndbuf = bufsize * 2;
2236 		atomic_set(&conn->sndbuf_space, bufsize);
2237 	}
2238 	return 0;
2239 }
2240 
2241 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn)
2242 {
2243 	if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk))
2244 		return;
2245 	smc_ib_sync_sg_for_cpu(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2246 }
2247 
2248 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn)
2249 {
2250 	if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk))
2251 		return;
2252 	smc_ib_sync_sg_for_device(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2253 }
2254 
2255 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn)
2256 {
2257 	int i;
2258 
2259 	if (!conn->lgr || conn->lgr->is_smcd)
2260 		return;
2261 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2262 		if (!smc_link_active(&conn->lgr->lnk[i]))
2263 			continue;
2264 		smc_ib_sync_sg_for_cpu(&conn->lgr->lnk[i], conn->rmb_desc,
2265 				       DMA_FROM_DEVICE);
2266 	}
2267 }
2268 
2269 void smc_rmb_sync_sg_for_device(struct smc_connection *conn)
2270 {
2271 	int i;
2272 
2273 	if (!conn->lgr || conn->lgr->is_smcd)
2274 		return;
2275 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2276 		if (!smc_link_active(&conn->lgr->lnk[i]))
2277 			continue;
2278 		smc_ib_sync_sg_for_device(&conn->lgr->lnk[i], conn->rmb_desc,
2279 					  DMA_FROM_DEVICE);
2280 	}
2281 }
2282 
2283 /* create the send and receive buffer for an SMC socket;
2284  * receive buffers are called RMBs;
2285  * (even though the SMC protocol allows more than one RMB-element per RMB,
2286  * the Linux implementation uses just one RMB-element per RMB, i.e. uses an
2287  * extra RMB for every connection in a link group
2288  */
2289 int smc_buf_create(struct smc_sock *smc, bool is_smcd)
2290 {
2291 	int rc;
2292 
2293 	/* create send buffer */
2294 	rc = __smc_buf_create(smc, is_smcd, false);
2295 	if (rc)
2296 		return rc;
2297 	/* create rmb */
2298 	rc = __smc_buf_create(smc, is_smcd, true);
2299 	if (rc) {
2300 		mutex_lock(&smc->conn.lgr->sndbufs_lock);
2301 		list_del(&smc->conn.sndbuf_desc->list);
2302 		mutex_unlock(&smc->conn.lgr->sndbufs_lock);
2303 		smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc);
2304 		smc->conn.sndbuf_desc = NULL;
2305 	}
2306 	return rc;
2307 }
2308 
2309 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
2310 {
2311 	int i;
2312 
2313 	for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
2314 		if (!test_and_set_bit(i, lgr->rtokens_used_mask))
2315 			return i;
2316 	}
2317 	return -ENOSPC;
2318 }
2319 
2320 static int smc_rtoken_find_by_link(struct smc_link_group *lgr, int lnk_idx,
2321 				   u32 rkey)
2322 {
2323 	int i;
2324 
2325 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2326 		if (test_bit(i, lgr->rtokens_used_mask) &&
2327 		    lgr->rtokens[i][lnk_idx].rkey == rkey)
2328 			return i;
2329 	}
2330 	return -ENOENT;
2331 }
2332 
2333 /* set rtoken for a new link to an existing rmb */
2334 void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new,
2335 		    __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey)
2336 {
2337 	int rtok_idx;
2338 
2339 	rtok_idx = smc_rtoken_find_by_link(lgr, link_idx, ntohl(nw_rkey_known));
2340 	if (rtok_idx == -ENOENT)
2341 		return;
2342 	lgr->rtokens[rtok_idx][link_idx_new].rkey = ntohl(nw_rkey);
2343 	lgr->rtokens[rtok_idx][link_idx_new].dma_addr = be64_to_cpu(nw_vaddr);
2344 }
2345 
2346 /* set rtoken for a new link whose link_id is given */
2347 void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id,
2348 		     __be64 nw_vaddr, __be32 nw_rkey)
2349 {
2350 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2351 	u32 rkey = ntohl(nw_rkey);
2352 	bool found = false;
2353 	int link_idx;
2354 
2355 	for (link_idx = 0; link_idx < SMC_LINKS_PER_LGR_MAX; link_idx++) {
2356 		if (lgr->lnk[link_idx].link_id == link_id) {
2357 			found = true;
2358 			break;
2359 		}
2360 	}
2361 	if (!found)
2362 		return;
2363 	lgr->rtokens[rtok_idx][link_idx].rkey = rkey;
2364 	lgr->rtokens[rtok_idx][link_idx].dma_addr = dma_addr;
2365 }
2366 
2367 /* add a new rtoken from peer */
2368 int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey)
2369 {
2370 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2371 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2372 	u32 rkey = ntohl(nw_rkey);
2373 	int i;
2374 
2375 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2376 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2377 		    lgr->rtokens[i][lnk->link_idx].dma_addr == dma_addr &&
2378 		    test_bit(i, lgr->rtokens_used_mask)) {
2379 			/* already in list */
2380 			return i;
2381 		}
2382 	}
2383 	i = smc_rmb_reserve_rtoken_idx(lgr);
2384 	if (i < 0)
2385 		return i;
2386 	lgr->rtokens[i][lnk->link_idx].rkey = rkey;
2387 	lgr->rtokens[i][lnk->link_idx].dma_addr = dma_addr;
2388 	return i;
2389 }
2390 
2391 /* delete an rtoken from all links */
2392 int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey)
2393 {
2394 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2395 	u32 rkey = ntohl(nw_rkey);
2396 	int i, j;
2397 
2398 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2399 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2400 		    test_bit(i, lgr->rtokens_used_mask)) {
2401 			for (j = 0; j < SMC_LINKS_PER_LGR_MAX; j++) {
2402 				lgr->rtokens[i][j].rkey = 0;
2403 				lgr->rtokens[i][j].dma_addr = 0;
2404 			}
2405 			clear_bit(i, lgr->rtokens_used_mask);
2406 			return 0;
2407 		}
2408 	}
2409 	return -ENOENT;
2410 }
2411 
2412 /* save rkey and dma_addr received from peer during clc handshake */
2413 int smc_rmb_rtoken_handling(struct smc_connection *conn,
2414 			    struct smc_link *lnk,
2415 			    struct smc_clc_msg_accept_confirm *clc)
2416 {
2417 	conn->rtoken_idx = smc_rtoken_add(lnk, clc->r0.rmb_dma_addr,
2418 					  clc->r0.rmb_rkey);
2419 	if (conn->rtoken_idx < 0)
2420 		return conn->rtoken_idx;
2421 	return 0;
2422 }
2423 
2424 static void smc_core_going_away(void)
2425 {
2426 	struct smc_ib_device *smcibdev;
2427 	struct smcd_dev *smcd;
2428 
2429 	mutex_lock(&smc_ib_devices.mutex);
2430 	list_for_each_entry(smcibdev, &smc_ib_devices.list, list) {
2431 		int i;
2432 
2433 		for (i = 0; i < SMC_MAX_PORTS; i++)
2434 			set_bit(i, smcibdev->ports_going_away);
2435 	}
2436 	mutex_unlock(&smc_ib_devices.mutex);
2437 
2438 	mutex_lock(&smcd_dev_list.mutex);
2439 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2440 		smcd->going_away = 1;
2441 	}
2442 	mutex_unlock(&smcd_dev_list.mutex);
2443 }
2444 
2445 /* Clean up all SMC link groups */
2446 static void smc_lgrs_shutdown(void)
2447 {
2448 	struct smcd_dev *smcd;
2449 
2450 	smc_core_going_away();
2451 
2452 	smc_smcr_terminate_all(NULL);
2453 
2454 	mutex_lock(&smcd_dev_list.mutex);
2455 	list_for_each_entry(smcd, &smcd_dev_list.list, list)
2456 		smc_smcd_terminate_all(smcd);
2457 	mutex_unlock(&smcd_dev_list.mutex);
2458 }
2459 
2460 static int smc_core_reboot_event(struct notifier_block *this,
2461 				 unsigned long event, void *ptr)
2462 {
2463 	smc_lgrs_shutdown();
2464 	smc_ib_unregister_client();
2465 	return 0;
2466 }
2467 
2468 static struct notifier_block smc_reboot_notifier = {
2469 	.notifier_call = smc_core_reboot_event,
2470 };
2471 
2472 int __init smc_core_init(void)
2473 {
2474 	return register_reboot_notifier(&smc_reboot_notifier);
2475 }
2476 
2477 /* Called (from smc_exit) when module is removed */
2478 void smc_core_exit(void)
2479 {
2480 	unregister_reboot_notifier(&smc_reboot_notifier);
2481 	smc_lgrs_shutdown();
2482 }
2483