xref: /openbmc/linux/net/rds/ib_rdma.c (revision 239480ab)
1 /*
2  * Copyright (c) 2006 Oracle.  All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 #include <linux/kernel.h>
34 #include <linux/slab.h>
35 #include <linux/rculist.h>
36 #include <linux/llist.h>
37 
38 #include "rds_single_path.h"
39 #include "ib_mr.h"
40 
41 struct workqueue_struct *rds_ib_mr_wq;
42 
43 static DEFINE_PER_CPU(unsigned long, clean_list_grace);
44 #define CLEAN_LIST_BUSY_BIT 0
45 
46 static struct rds_ib_device *rds_ib_get_device(__be32 ipaddr)
47 {
48 	struct rds_ib_device *rds_ibdev;
49 	struct rds_ib_ipaddr *i_ipaddr;
50 
51 	rcu_read_lock();
52 	list_for_each_entry_rcu(rds_ibdev, &rds_ib_devices, list) {
53 		list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
54 			if (i_ipaddr->ipaddr == ipaddr) {
55 				atomic_inc(&rds_ibdev->refcount);
56 				rcu_read_unlock();
57 				return rds_ibdev;
58 			}
59 		}
60 	}
61 	rcu_read_unlock();
62 
63 	return NULL;
64 }
65 
66 static int rds_ib_add_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
67 {
68 	struct rds_ib_ipaddr *i_ipaddr;
69 
70 	i_ipaddr = kmalloc(sizeof *i_ipaddr, GFP_KERNEL);
71 	if (!i_ipaddr)
72 		return -ENOMEM;
73 
74 	i_ipaddr->ipaddr = ipaddr;
75 
76 	spin_lock_irq(&rds_ibdev->spinlock);
77 	list_add_tail_rcu(&i_ipaddr->list, &rds_ibdev->ipaddr_list);
78 	spin_unlock_irq(&rds_ibdev->spinlock);
79 
80 	return 0;
81 }
82 
83 static void rds_ib_remove_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
84 {
85 	struct rds_ib_ipaddr *i_ipaddr;
86 	struct rds_ib_ipaddr *to_free = NULL;
87 
88 
89 	spin_lock_irq(&rds_ibdev->spinlock);
90 	list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
91 		if (i_ipaddr->ipaddr == ipaddr) {
92 			list_del_rcu(&i_ipaddr->list);
93 			to_free = i_ipaddr;
94 			break;
95 		}
96 	}
97 	spin_unlock_irq(&rds_ibdev->spinlock);
98 
99 	if (to_free)
100 		kfree_rcu(to_free, rcu);
101 }
102 
103 int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
104 {
105 	struct rds_ib_device *rds_ibdev_old;
106 
107 	rds_ibdev_old = rds_ib_get_device(ipaddr);
108 	if (!rds_ibdev_old)
109 		return rds_ib_add_ipaddr(rds_ibdev, ipaddr);
110 
111 	if (rds_ibdev_old != rds_ibdev) {
112 		rds_ib_remove_ipaddr(rds_ibdev_old, ipaddr);
113 		rds_ib_dev_put(rds_ibdev_old);
114 		return rds_ib_add_ipaddr(rds_ibdev, ipaddr);
115 	}
116 	rds_ib_dev_put(rds_ibdev_old);
117 
118 	return 0;
119 }
120 
121 void rds_ib_add_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
122 {
123 	struct rds_ib_connection *ic = conn->c_transport_data;
124 
125 	/* conn was previously on the nodev_conns_list */
126 	spin_lock_irq(&ib_nodev_conns_lock);
127 	BUG_ON(list_empty(&ib_nodev_conns));
128 	BUG_ON(list_empty(&ic->ib_node));
129 	list_del(&ic->ib_node);
130 
131 	spin_lock(&rds_ibdev->spinlock);
132 	list_add_tail(&ic->ib_node, &rds_ibdev->conn_list);
133 	spin_unlock(&rds_ibdev->spinlock);
134 	spin_unlock_irq(&ib_nodev_conns_lock);
135 
136 	ic->rds_ibdev = rds_ibdev;
137 	atomic_inc(&rds_ibdev->refcount);
138 }
139 
140 void rds_ib_remove_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
141 {
142 	struct rds_ib_connection *ic = conn->c_transport_data;
143 
144 	/* place conn on nodev_conns_list */
145 	spin_lock(&ib_nodev_conns_lock);
146 
147 	spin_lock_irq(&rds_ibdev->spinlock);
148 	BUG_ON(list_empty(&ic->ib_node));
149 	list_del(&ic->ib_node);
150 	spin_unlock_irq(&rds_ibdev->spinlock);
151 
152 	list_add_tail(&ic->ib_node, &ib_nodev_conns);
153 
154 	spin_unlock(&ib_nodev_conns_lock);
155 
156 	ic->rds_ibdev = NULL;
157 	rds_ib_dev_put(rds_ibdev);
158 }
159 
160 void rds_ib_destroy_nodev_conns(void)
161 {
162 	struct rds_ib_connection *ic, *_ic;
163 	LIST_HEAD(tmp_list);
164 
165 	/* avoid calling conn_destroy with irqs off */
166 	spin_lock_irq(&ib_nodev_conns_lock);
167 	list_splice(&ib_nodev_conns, &tmp_list);
168 	spin_unlock_irq(&ib_nodev_conns_lock);
169 
170 	list_for_each_entry_safe(ic, _ic, &tmp_list, ib_node)
171 		rds_conn_destroy(ic->conn);
172 }
173 
174 void rds_ib_get_mr_info(struct rds_ib_device *rds_ibdev, struct rds_info_rdma_connection *iinfo)
175 {
176 	struct rds_ib_mr_pool *pool_1m = rds_ibdev->mr_1m_pool;
177 
178 	iinfo->rdma_mr_max = pool_1m->max_items;
179 	iinfo->rdma_mr_size = pool_1m->fmr_attr.max_pages;
180 }
181 
182 struct rds_ib_mr *rds_ib_reuse_mr(struct rds_ib_mr_pool *pool)
183 {
184 	struct rds_ib_mr *ibmr = NULL;
185 	struct llist_node *ret;
186 	unsigned long *flag;
187 
188 	preempt_disable();
189 	flag = this_cpu_ptr(&clean_list_grace);
190 	set_bit(CLEAN_LIST_BUSY_BIT, flag);
191 	ret = llist_del_first(&pool->clean_list);
192 	if (ret) {
193 		ibmr = llist_entry(ret, struct rds_ib_mr, llnode);
194 		if (pool->pool_type == RDS_IB_MR_8K_POOL)
195 			rds_ib_stats_inc(s_ib_rdma_mr_8k_reused);
196 		else
197 			rds_ib_stats_inc(s_ib_rdma_mr_1m_reused);
198 	}
199 
200 	clear_bit(CLEAN_LIST_BUSY_BIT, flag);
201 	preempt_enable();
202 	return ibmr;
203 }
204 
205 static inline void wait_clean_list_grace(void)
206 {
207 	int cpu;
208 	unsigned long *flag;
209 
210 	for_each_online_cpu(cpu) {
211 		flag = &per_cpu(clean_list_grace, cpu);
212 		while (test_bit(CLEAN_LIST_BUSY_BIT, flag))
213 			cpu_relax();
214 	}
215 }
216 
217 void rds_ib_sync_mr(void *trans_private, int direction)
218 {
219 	struct rds_ib_mr *ibmr = trans_private;
220 	struct rds_ib_device *rds_ibdev = ibmr->device;
221 
222 	switch (direction) {
223 	case DMA_FROM_DEVICE:
224 		ib_dma_sync_sg_for_cpu(rds_ibdev->dev, ibmr->sg,
225 			ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
226 		break;
227 	case DMA_TO_DEVICE:
228 		ib_dma_sync_sg_for_device(rds_ibdev->dev, ibmr->sg,
229 			ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
230 		break;
231 	}
232 }
233 
234 void __rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
235 {
236 	struct rds_ib_device *rds_ibdev = ibmr->device;
237 
238 	if (ibmr->sg_dma_len) {
239 		ib_dma_unmap_sg(rds_ibdev->dev,
240 				ibmr->sg, ibmr->sg_len,
241 				DMA_BIDIRECTIONAL);
242 		ibmr->sg_dma_len = 0;
243 	}
244 
245 	/* Release the s/g list */
246 	if (ibmr->sg_len) {
247 		unsigned int i;
248 
249 		for (i = 0; i < ibmr->sg_len; ++i) {
250 			struct page *page = sg_page(&ibmr->sg[i]);
251 
252 			/* FIXME we need a way to tell a r/w MR
253 			 * from a r/o MR */
254 			WARN_ON(!page->mapping && irqs_disabled());
255 			set_page_dirty(page);
256 			put_page(page);
257 		}
258 		kfree(ibmr->sg);
259 
260 		ibmr->sg = NULL;
261 		ibmr->sg_len = 0;
262 	}
263 }
264 
265 void rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
266 {
267 	unsigned int pinned = ibmr->sg_len;
268 
269 	__rds_ib_teardown_mr(ibmr);
270 	if (pinned) {
271 		struct rds_ib_mr_pool *pool = ibmr->pool;
272 
273 		atomic_sub(pinned, &pool->free_pinned);
274 	}
275 }
276 
277 static inline unsigned int rds_ib_flush_goal(struct rds_ib_mr_pool *pool, int free_all)
278 {
279 	unsigned int item_count;
280 
281 	item_count = atomic_read(&pool->item_count);
282 	if (free_all)
283 		return item_count;
284 
285 	return 0;
286 }
287 
288 /*
289  * given an llist of mrs, put them all into the list_head for more processing
290  */
291 static unsigned int llist_append_to_list(struct llist_head *llist,
292 					 struct list_head *list)
293 {
294 	struct rds_ib_mr *ibmr;
295 	struct llist_node *node;
296 	struct llist_node *next;
297 	unsigned int count = 0;
298 
299 	node = llist_del_all(llist);
300 	while (node) {
301 		next = node->next;
302 		ibmr = llist_entry(node, struct rds_ib_mr, llnode);
303 		list_add_tail(&ibmr->unmap_list, list);
304 		node = next;
305 		count++;
306 	}
307 	return count;
308 }
309 
310 /*
311  * this takes a list head of mrs and turns it into linked llist nodes
312  * of clusters.  Each cluster has linked llist nodes of
313  * MR_CLUSTER_SIZE mrs that are ready for reuse.
314  */
315 static void list_to_llist_nodes(struct rds_ib_mr_pool *pool,
316 				struct list_head *list,
317 				struct llist_node **nodes_head,
318 				struct llist_node **nodes_tail)
319 {
320 	struct rds_ib_mr *ibmr;
321 	struct llist_node *cur = NULL;
322 	struct llist_node **next = nodes_head;
323 
324 	list_for_each_entry(ibmr, list, unmap_list) {
325 		cur = &ibmr->llnode;
326 		*next = cur;
327 		next = &cur->next;
328 	}
329 	*next = NULL;
330 	*nodes_tail = cur;
331 }
332 
333 /*
334  * Flush our pool of MRs.
335  * At a minimum, all currently unused MRs are unmapped.
336  * If the number of MRs allocated exceeds the limit, we also try
337  * to free as many MRs as needed to get back to this limit.
338  */
339 int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
340 			 int free_all, struct rds_ib_mr **ibmr_ret)
341 {
342 	struct rds_ib_mr *ibmr;
343 	struct llist_node *clean_nodes;
344 	struct llist_node *clean_tail;
345 	LIST_HEAD(unmap_list);
346 	unsigned long unpinned = 0;
347 	unsigned int nfreed = 0, dirty_to_clean = 0, free_goal;
348 
349 	if (pool->pool_type == RDS_IB_MR_8K_POOL)
350 		rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_flush);
351 	else
352 		rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_flush);
353 
354 	if (ibmr_ret) {
355 		DEFINE_WAIT(wait);
356 		while (!mutex_trylock(&pool->flush_lock)) {
357 			ibmr = rds_ib_reuse_mr(pool);
358 			if (ibmr) {
359 				*ibmr_ret = ibmr;
360 				finish_wait(&pool->flush_wait, &wait);
361 				goto out_nolock;
362 			}
363 
364 			prepare_to_wait(&pool->flush_wait, &wait,
365 					TASK_UNINTERRUPTIBLE);
366 			if (llist_empty(&pool->clean_list))
367 				schedule();
368 
369 			ibmr = rds_ib_reuse_mr(pool);
370 			if (ibmr) {
371 				*ibmr_ret = ibmr;
372 				finish_wait(&pool->flush_wait, &wait);
373 				goto out_nolock;
374 			}
375 		}
376 		finish_wait(&pool->flush_wait, &wait);
377 	} else
378 		mutex_lock(&pool->flush_lock);
379 
380 	if (ibmr_ret) {
381 		ibmr = rds_ib_reuse_mr(pool);
382 		if (ibmr) {
383 			*ibmr_ret = ibmr;
384 			goto out;
385 		}
386 	}
387 
388 	/* Get the list of all MRs to be dropped. Ordering matters -
389 	 * we want to put drop_list ahead of free_list.
390 	 */
391 	dirty_to_clean = llist_append_to_list(&pool->drop_list, &unmap_list);
392 	dirty_to_clean += llist_append_to_list(&pool->free_list, &unmap_list);
393 	if (free_all)
394 		llist_append_to_list(&pool->clean_list, &unmap_list);
395 
396 	free_goal = rds_ib_flush_goal(pool, free_all);
397 
398 	if (list_empty(&unmap_list))
399 		goto out;
400 
401 	if (pool->use_fastreg)
402 		rds_ib_unreg_frmr(&unmap_list, &nfreed, &unpinned, free_goal);
403 	else
404 		rds_ib_unreg_fmr(&unmap_list, &nfreed, &unpinned, free_goal);
405 
406 	if (!list_empty(&unmap_list)) {
407 		/* we have to make sure that none of the things we're about
408 		 * to put on the clean list would race with other cpus trying
409 		 * to pull items off.  The llist would explode if we managed to
410 		 * remove something from the clean list and then add it back again
411 		 * while another CPU was spinning on that same item in llist_del_first.
412 		 *
413 		 * This is pretty unlikely, but just in case  wait for an llist grace period
414 		 * here before adding anything back into the clean list.
415 		 */
416 		wait_clean_list_grace();
417 
418 		list_to_llist_nodes(pool, &unmap_list, &clean_nodes, &clean_tail);
419 		if (ibmr_ret)
420 			*ibmr_ret = llist_entry(clean_nodes, struct rds_ib_mr, llnode);
421 
422 		/* more than one entry in llist nodes */
423 		if (clean_nodes->next)
424 			llist_add_batch(clean_nodes->next, clean_tail, &pool->clean_list);
425 
426 	}
427 
428 	atomic_sub(unpinned, &pool->free_pinned);
429 	atomic_sub(dirty_to_clean, &pool->dirty_count);
430 	atomic_sub(nfreed, &pool->item_count);
431 
432 out:
433 	mutex_unlock(&pool->flush_lock);
434 	if (waitqueue_active(&pool->flush_wait))
435 		wake_up(&pool->flush_wait);
436 out_nolock:
437 	return 0;
438 }
439 
440 struct rds_ib_mr *rds_ib_try_reuse_ibmr(struct rds_ib_mr_pool *pool)
441 {
442 	struct rds_ib_mr *ibmr = NULL;
443 	int iter = 0;
444 
445 	if (atomic_read(&pool->dirty_count) >= pool->max_items_soft / 10)
446 		queue_delayed_work(rds_ib_mr_wq, &pool->flush_worker, 10);
447 
448 	while (1) {
449 		ibmr = rds_ib_reuse_mr(pool);
450 		if (ibmr)
451 			return ibmr;
452 
453 		if (atomic_inc_return(&pool->item_count) <= pool->max_items)
454 			break;
455 
456 		atomic_dec(&pool->item_count);
457 
458 		if (++iter > 2) {
459 			if (pool->pool_type == RDS_IB_MR_8K_POOL)
460 				rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_depleted);
461 			else
462 				rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_depleted);
463 			return ERR_PTR(-EAGAIN);
464 		}
465 
466 		/* We do have some empty MRs. Flush them out. */
467 		if (pool->pool_type == RDS_IB_MR_8K_POOL)
468 			rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_wait);
469 		else
470 			rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_wait);
471 
472 		rds_ib_flush_mr_pool(pool, 0, &ibmr);
473 		if (ibmr)
474 			return ibmr;
475 	}
476 
477 	return ibmr;
478 }
479 
480 static void rds_ib_mr_pool_flush_worker(struct work_struct *work)
481 {
482 	struct rds_ib_mr_pool *pool = container_of(work, struct rds_ib_mr_pool, flush_worker.work);
483 
484 	rds_ib_flush_mr_pool(pool, 0, NULL);
485 }
486 
487 void rds_ib_free_mr(void *trans_private, int invalidate)
488 {
489 	struct rds_ib_mr *ibmr = trans_private;
490 	struct rds_ib_mr_pool *pool = ibmr->pool;
491 	struct rds_ib_device *rds_ibdev = ibmr->device;
492 
493 	rdsdebug("RDS/IB: free_mr nents %u\n", ibmr->sg_len);
494 
495 	/* Return it to the pool's free list */
496 	if (rds_ibdev->use_fastreg)
497 		rds_ib_free_frmr_list(ibmr);
498 	else
499 		rds_ib_free_fmr_list(ibmr);
500 
501 	atomic_add(ibmr->sg_len, &pool->free_pinned);
502 	atomic_inc(&pool->dirty_count);
503 
504 	/* If we've pinned too many pages, request a flush */
505 	if (atomic_read(&pool->free_pinned) >= pool->max_free_pinned ||
506 	    atomic_read(&pool->dirty_count) >= pool->max_items / 5)
507 		queue_delayed_work(rds_ib_mr_wq, &pool->flush_worker, 10);
508 
509 	if (invalidate) {
510 		if (likely(!in_interrupt())) {
511 			rds_ib_flush_mr_pool(pool, 0, NULL);
512 		} else {
513 			/* We get here if the user created a MR marked
514 			 * as use_once and invalidate at the same time.
515 			 */
516 			queue_delayed_work(rds_ib_mr_wq,
517 					   &pool->flush_worker, 10);
518 		}
519 	}
520 
521 	rds_ib_dev_put(rds_ibdev);
522 }
523 
524 void rds_ib_flush_mrs(void)
525 {
526 	struct rds_ib_device *rds_ibdev;
527 
528 	down_read(&rds_ib_devices_lock);
529 	list_for_each_entry(rds_ibdev, &rds_ib_devices, list) {
530 		if (rds_ibdev->mr_8k_pool)
531 			rds_ib_flush_mr_pool(rds_ibdev->mr_8k_pool, 0, NULL);
532 
533 		if (rds_ibdev->mr_1m_pool)
534 			rds_ib_flush_mr_pool(rds_ibdev->mr_1m_pool, 0, NULL);
535 	}
536 	up_read(&rds_ib_devices_lock);
537 }
538 
539 void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
540 		    struct rds_sock *rs, u32 *key_ret)
541 {
542 	struct rds_ib_device *rds_ibdev;
543 	struct rds_ib_mr *ibmr = NULL;
544 	struct rds_ib_connection *ic = rs->rs_conn->c_transport_data;
545 	int ret;
546 
547 	rds_ibdev = rds_ib_get_device(rs->rs_bound_addr);
548 	if (!rds_ibdev) {
549 		ret = -ENODEV;
550 		goto out;
551 	}
552 
553 	if (!rds_ibdev->mr_8k_pool || !rds_ibdev->mr_1m_pool) {
554 		ret = -ENODEV;
555 		goto out;
556 	}
557 
558 	if (rds_ibdev->use_fastreg)
559 		ibmr = rds_ib_reg_frmr(rds_ibdev, ic, sg, nents, key_ret);
560 	else
561 		ibmr = rds_ib_reg_fmr(rds_ibdev, sg, nents, key_ret);
562 	if (ibmr)
563 		rds_ibdev = NULL;
564 
565  out:
566 	if (!ibmr)
567 		pr_warn("RDS/IB: rds_ib_get_mr failed (errno=%d)\n", ret);
568 
569 	if (rds_ibdev)
570 		rds_ib_dev_put(rds_ibdev);
571 
572 	return ibmr;
573 }
574 
575 void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *pool)
576 {
577 	cancel_delayed_work_sync(&pool->flush_worker);
578 	rds_ib_flush_mr_pool(pool, 1, NULL);
579 	WARN_ON(atomic_read(&pool->item_count));
580 	WARN_ON(atomic_read(&pool->free_pinned));
581 	kfree(pool);
582 }
583 
584 struct rds_ib_mr_pool *rds_ib_create_mr_pool(struct rds_ib_device *rds_ibdev,
585 					     int pool_type)
586 {
587 	struct rds_ib_mr_pool *pool;
588 
589 	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
590 	if (!pool)
591 		return ERR_PTR(-ENOMEM);
592 
593 	pool->pool_type = pool_type;
594 	init_llist_head(&pool->free_list);
595 	init_llist_head(&pool->drop_list);
596 	init_llist_head(&pool->clean_list);
597 	mutex_init(&pool->flush_lock);
598 	init_waitqueue_head(&pool->flush_wait);
599 	INIT_DELAYED_WORK(&pool->flush_worker, rds_ib_mr_pool_flush_worker);
600 
601 	if (pool_type == RDS_IB_MR_1M_POOL) {
602 		/* +1 allows for unaligned MRs */
603 		pool->fmr_attr.max_pages = RDS_MR_1M_MSG_SIZE + 1;
604 		pool->max_items = RDS_MR_1M_POOL_SIZE;
605 	} else {
606 		/* pool_type == RDS_IB_MR_8K_POOL */
607 		pool->fmr_attr.max_pages = RDS_MR_8K_MSG_SIZE + 1;
608 		pool->max_items = RDS_MR_8K_POOL_SIZE;
609 	}
610 
611 	pool->max_free_pinned = pool->max_items * pool->fmr_attr.max_pages / 4;
612 	pool->fmr_attr.max_maps = rds_ibdev->fmr_max_remaps;
613 	pool->fmr_attr.page_shift = PAGE_SHIFT;
614 	pool->max_items_soft = rds_ibdev->max_mrs * 3 / 4;
615 	pool->use_fastreg = rds_ibdev->use_fastreg;
616 
617 	return pool;
618 }
619 
620 int rds_ib_mr_init(void)
621 {
622 	rds_ib_mr_wq = alloc_workqueue("rds_mr_flushd", WQ_MEM_RECLAIM, 0);
623 	if (!rds_ib_mr_wq)
624 		return -ENOMEM;
625 	return 0;
626 }
627 
628 /* By the time this is called all the IB devices should have been torn down and
629  * had their pools freed.  As each pool is freed its work struct is waited on,
630  * so the pool flushing work queue should be idle by the time we get here.
631  */
632 void rds_ib_mr_exit(void)
633 {
634 	destroy_workqueue(rds_ib_mr_wq);
635 }
636