summaryrefslogtreecommitdiffstats
path: root/net/rds/ib_rdma.c
blob: a2340748ec8673d478df5961329734a07d0b9ca6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
/*
 * Copyright (c) 2006 Oracle.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/rculist.h>
#include <linux/llist.h>

#include "rds.h"
#include "ib.h"

static DEFINE_PER_CPU(unsigned long, clean_list_grace);
#define CLEAN_LIST_BUSY_BIT 0

/*
 * This is stored as mr->r_trans_private.
 */
struct rds_ib_mr {
	struct rds_ib_device	*device;
	struct rds_ib_mr_pool	*pool;
	struct ib_fmr		*fmr;

	struct llist_node	llnode;

	/* unmap_list is for freeing */
	struct list_head	unmap_list;
	unsigned int		remap_count;

	struct scatterlist	*sg;
	unsigned int		sg_len;
	u64			*dma;
	int			sg_dma_len;
};

/*
 * Our own little FMR pool
 */
struct rds_ib_mr_pool {
	unsigned int            pool_type;
	struct mutex		flush_lock;		/* serialize fmr invalidate */
	struct delayed_work	flush_worker;		/* flush worker */

	atomic_t		item_count;		/* total # of MRs */
	atomic_t		dirty_count;		/* # dirty of MRs */

	struct llist_head	drop_list;		/* MRs that have reached their max_maps limit */
	struct llist_head	free_list;		/* unused MRs */
	struct llist_head	clean_list;		/* global unused & unamapped MRs */
	wait_queue_head_t	flush_wait;

	atomic_t		free_pinned;		/* memory pinned by free MRs */
	unsigned long		max_items;
	unsigned long		max_items_soft;
	unsigned long		max_free_pinned;
	struct ib_fmr_attr	fmr_attr;
};

static struct workqueue_struct *rds_ib_fmr_wq;

int rds_ib_fmr_init(void)
{
	rds_ib_fmr_wq = create_workqueue("rds_fmr_flushd");
	if (!rds_ib_fmr_wq)
		return -ENOMEM;
	return 0;
}

/* By the time this is called all the IB devices should have been torn down and
 * had their pools freed.  As each pool is freed its work struct is waited on,
 * so the pool flushing work queue should be idle by the time we get here.
 */
void rds_ib_fmr_exit(void)
{
	destroy_workqueue(rds_ib_fmr_wq);
}

static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool, int free_all, struct rds_ib_mr **);
static void rds_ib_teardown_mr(struct rds_ib_mr *ibmr);
static void rds_ib_mr_pool_flush_worker(struct work_struct *work);

static struct rds_ib_device *rds_ib_get_device(__be32 ipaddr)
{
	struct rds_ib_device *rds_ibdev;
	struct rds_ib_ipaddr *i_ipaddr;

	rcu_read_lock();
	list_for_each_entry_rcu(rds_ibdev, &rds_ib_devices, list) {
		list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
			if (i_ipaddr->ipaddr == ipaddr) {
				atomic_inc(&rds_ibdev->refcount);
				rcu_read_unlock();
				return rds_ibdev;
			}
		}
	}
	rcu_read_unlock();

	return NULL;
}

static int rds_ib_add_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
{
	struct rds_ib_ipaddr *i_ipaddr;

	i_ipaddr = kmalloc(sizeof *i_ipaddr, GFP_KERNEL);
	if (!i_ipaddr)
		return -ENOMEM;

	i_ipaddr->ipaddr = ipaddr;

	spin_lock_irq(&rds_ibdev->spinlock);
	list_add_tail_rcu(&i_ipaddr->list, &rds_ibdev->ipaddr_list);
	spin_unlock_irq(&rds_ibdev->spinlock);

	return 0;
}

static void rds_ib_remove_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
{
	struct rds_ib_ipaddr *i_ipaddr;
	struct rds_ib_ipaddr *to_free = NULL;


	spin_lock_irq(&rds_ibdev->spinlock);
	list_for_each_entry_rcu(i_ipaddr, &rds_ibdev->ipaddr_list, list) {
		if (i_ipaddr->ipaddr == ipaddr) {
			list_del_rcu(&i_ipaddr->list);
			to_free = i_ipaddr;
			break;
		}
	}
	spin_unlock_irq(&rds_ibdev->spinlock);

	if (to_free)
		kfree_rcu(to_free, rcu);
}

int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr)
{
	struct rds_ib_device *rds_ibdev_old;

	rds_ibdev_old = rds_ib_get_device(ipaddr);
	if (!rds_ibdev_old)
		return rds_ib_add_ipaddr(rds_ibdev, ipaddr);

	if (rds_ibdev_old != rds_ibdev) {
		rds_ib_remove_ipaddr(rds_ibdev_old, ipaddr);
		rds_ib_dev_put(rds_ibdev_old);
		return rds_ib_add_ipaddr(rds_ibdev, ipaddr);
	}
	rds_ib_dev_put(rds_ibdev_old);

	return 0;
}

void rds_ib_add_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
{
	struct rds_ib_connection *ic = conn->c_transport_data;

	/* conn was previously on the nodev_conns_list */
	spin_lock_irq(&ib_nodev_conns_lock);
	BUG_ON(list_empty(&ib_nodev_conns));
	BUG_ON(list_empty(&ic->ib_node));
	list_del(&ic->ib_node);

	spin_lock(&rds_ibdev->spinlock);
	list_add_tail(&ic->ib_node, &rds_ibdev->conn_list);
	spin_unlock(&rds_ibdev->spinlock);
	spin_unlock_irq(&ib_nodev_conns_lock);

	ic->rds_ibdev = rds_ibdev;
	atomic_inc(&rds_ibdev->refcount);
}

void rds_ib_remove_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn)
{
	struct rds_ib_connection *ic = conn->c_transport_data;

	/* place conn on nodev_conns_list */
	spin_lock(&ib_nodev_conns_lock);

	spin_lock_irq(&rds_ibdev->spinlock);
	BUG_ON(list_empty(&ic->ib_node));
	list_del(&ic->ib_node);
	spin_unlock_irq(&rds_ibdev->spinlock);

	list_add_tail(&ic->ib_node, &ib_nodev_conns);

	spin_unlock(&ib_nodev_conns_lock);

	ic->rds_ibdev = NULL;
	rds_ib_dev_put(rds_ibdev);
}

void rds_ib_destroy_nodev_conns(void)
{
	struct rds_ib_connection *ic, *_ic;
	LIST_HEAD(tmp_list);

	/* avoid calling conn_destroy with irqs off */
	spin_lock_irq(&ib_nodev_conns_lock);
	list_splice(&ib_nodev_conns, &tmp_list);
	spin_unlock_irq(&ib_nodev_conns_lock);

	list_for_each_entry_safe(ic, _ic, &tmp_list, ib_node)
		rds_conn_destroy(ic->conn);
}

struct rds_ib_mr_pool *rds_ib_create_mr_pool(struct rds_ib_device *rds_ibdev,
					     int pool_type)
{
	struct rds_ib_mr_pool *pool;

	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
	if (!pool)
		return ERR_PTR(-ENOMEM);

	pool->pool_type = pool_type;
	init_llist_head(&pool->free_list);
	init_llist_head(&pool->drop_list);
	init_llist_head(&pool->clean_list);
	mutex_init(&pool->flush_lock);
	init_waitqueue_head(&pool->flush_wait);
	INIT_DELAYED_WORK(&pool->flush_worker, rds_ib_mr_pool_flush_worker);

	if (pool_type == RDS_IB_MR_1M_POOL) {
		/* +1 allows for unaligned MRs */
		pool->fmr_attr.max_pages = RDS_FMR_1M_MSG_SIZE + 1;
		pool->max_items = RDS_FMR_1M_POOL_SIZE;
	} else {
		/* pool_type == RDS_IB_MR_8K_POOL */
		pool->fmr_attr.max_pages = RDS_FMR_8K_MSG_SIZE + 1;
		pool->max_items = RDS_FMR_8K_POOL_SIZE;
	}

	pool->max_free_pinned = pool->max_items * pool->fmr_attr.max_pages / 4;
	pool->fmr_attr.max_maps = rds_ibdev->fmr_max_remaps;
	pool->fmr_attr.page_shift = PAGE_SHIFT;
	pool->max_items_soft = rds_ibdev->max_fmrs * 3 / 4;

	return pool;
}

void rds_ib_get_mr_info(struct rds_ib_device *rds_ibdev, struct rds_info_rdma_connection *iinfo)
{
	struct rds_ib_mr_pool *pool_1m = rds_ibdev->mr_1m_pool;

	iinfo->rdma_mr_max = pool_1m->max_items;
	iinfo->rdma_mr_size = pool_1m->fmr_attr.max_pages;
}

void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *pool)
{
	cancel_delayed_work_sync(&pool->flush_worker);
	rds_ib_flush_mr_pool(pool, 1, NULL);
	WARN_ON(atomic_read(&pool->item_count));
	WARN_ON(atomic_read(&pool->free_pinned));
	kfree(pool);
}

static inline struct rds_ib_mr *rds_ib_reuse_fmr(struct rds_ib_mr_pool *pool)
{
	struct rds_ib_mr *ibmr = NULL;
	struct llist_node *ret;
	unsigned long *flag;

	preempt_disable();
	flag = this_cpu_ptr(&clean_list_grace);
	set_bit(CLEAN_LIST_BUSY_BIT, flag);
	ret = llist_del_first(&pool->clean_list);
	if (ret)
		ibmr = llist_entry(ret, struct rds_ib_mr, llnode);

	clear_bit(CLEAN_LIST_BUSY_BIT, flag);
	preempt_enable();
	return ibmr;
}

static inline void wait_clean_list_grace(void)
{
	int cpu;
	unsigned long *flag;

	for_each_online_cpu(cpu) {
		flag = &per_cpu(clean_list_grace, cpu);
		while (test_bit(CLEAN_LIST_BUSY_BIT, flag))
			cpu_relax();
	}
}

static struct rds_ib_mr *rds_ib_alloc_fmr(struct rds_ib_device *rds_ibdev,
					  int npages)
{
	struct rds_ib_mr_pool *pool;
	struct rds_ib_mr *ibmr = NULL;
	int err = 0, iter = 0;

	if (npages <= RDS_FMR_8K_MSG_SIZE)
		pool = rds_ibdev->mr_8k_pool;
	else
		pool = rds_ibdev->mr_1m_pool;

	if (atomic_read(&pool->dirty_count) >= pool->max_items / 10)
		queue_delayed_work(rds_ib_fmr_wq, &pool->flush_worker, 10);

	/* Switch pools if one of the pool is reaching upper limit */
	if (atomic_read(&pool->dirty_count) >=  pool->max_items * 9 / 10) {
		if (pool->pool_type == RDS_IB_MR_8K_POOL)
			pool = rds_ibdev->mr_1m_pool;
		else
			pool = rds_ibdev->mr_8k_pool;
	}

	while (1) {
		ibmr = rds_ib_reuse_fmr(pool);
		if (ibmr)
			return ibmr;

		/* No clean MRs - now we have the choice of either
		 * allocating a fresh MR up to the limit imposed by the
		 * driver, or flush any dirty unused MRs.
		 * We try to avoid stalling in the send path if possible,
		 * so we allocate as long as we're allowed to.
		 *
		 * We're fussy with enforcing the FMR limit, though. If the driver
		 * tells us we can't use more than N fmrs, we shouldn't start
		 * arguing with it */
		if (atomic_inc_return(&pool->item_count) <= pool->max_items)
			break;

		atomic_dec(&pool->item_count);

		if (++iter > 2) {
			if (pool->pool_type == RDS_IB_MR_8K_POOL)
				rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_depleted);
			else
				rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_depleted);
			return ERR_PTR(-EAGAIN);
		}

		/* We do have some empty MRs. Flush them out. */
		if (pool->pool_type == RDS_IB_MR_8K_POOL)
			rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_wait);
		else
			rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_wait);
		rds_ib_flush_mr_pool(pool, 0, &ibmr);
		if (ibmr)
			return ibmr;
	}

	ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL, rdsibdev_to_node(rds_ibdev));
	if (!ibmr) {
		err = -ENOMEM;
		goto out_no_cigar;
	}

	ibmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
			(IB_ACCESS_LOCAL_WRITE |
			 IB_ACCESS_REMOTE_READ |
			 IB_ACCESS_REMOTE_WRITE|
			 IB_ACCESS_REMOTE_ATOMIC),
			&pool->fmr_attr);
	if (IS_ERR(ibmr->fmr)) {
		err = PTR_ERR(ibmr->fmr);
		ibmr->fmr = NULL;
		printk(KERN_WARNING "RDS/IB: ib_alloc_fmr failed (err=%d)\n", err);
		goto out_no_cigar;
	}

	ibmr->pool = pool;
	if (pool->pool_type == RDS_IB_MR_8K_POOL)
		rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
	else
		rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);

	return ibmr;

out_no_cigar:
	if (ibmr) {
		if (ibmr->fmr)
			ib_dealloc_fmr(ibmr->fmr);
		kfree(ibmr);
	}
	atomic_dec(&pool->item_count);
	return ERR_PTR(err);
}

static int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr,
	       struct scatterlist *sg, unsigned int nents)
{
	struct ib_device *dev = rds_ibdev->dev;
	struct scatterlist *scat = sg;
	u64 io_addr = 0;
	u64 *dma_pages;
	u32 len;
	int page_cnt, sg_dma_len;
	int i, j;
	int ret;

	sg_dma_len = ib_dma_map_sg(dev, sg, nents,
				 DMA_BIDIRECTIONAL);
	if (unlikely(!sg_dma_len)) {
		printk(KERN_WARNING "RDS/IB: dma_map_sg failed!\n");
		return -EBUSY;
	}

	len = 0;
	page_cnt = 0;

	for (i = 0; i < sg_dma_len; ++i) {
		unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
		u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);

		if (dma_addr & ~PAGE_MASK) {
			if (i > 0)
				return -EINVAL;
			else
				++page_cnt;
		}
		if ((dma_addr + dma_len) & ~PAGE_MASK) {
			if (i < sg_dma_len - 1)
				return -EINVAL;
			else
				++page_cnt;
		}

		len += dma_len;
	}

	page_cnt += len >> PAGE_SHIFT;
	if (page_cnt > ibmr->pool->fmr_attr.max_pages)
		return -EINVAL;

	dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
				 rdsibdev_to_node(rds_ibdev));
	if (!dma_pages)
		return -ENOMEM;

	page_cnt = 0;
	for (i = 0; i < sg_dma_len; ++i) {
		unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
		u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);

		for (j = 0; j < dma_len; j += PAGE_SIZE)
			dma_pages[page_cnt++] =
				(dma_addr & PAGE_MASK) + j;
	}

	ret = ib_map_phys_fmr(ibmr->fmr,
				   dma_pages, page_cnt, io_addr);
	if (ret)
		goto out;

	/* Success - we successfully remapped the MR, so we can
	 * safely tear down the old mapping. */
	rds_ib_teardown_mr(ibmr);

	ibmr->sg = scat;
	ibmr->sg_len = nents;
	ibmr->sg_dma_len = sg_dma_len;
	ibmr->remap_count++;

	if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
		rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
	else
		rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
	ret = 0;

out:
	kfree(dma_pages);

	return ret;
}

void rds_ib_sync_mr(void *trans_private, int direction)
{
	struct rds_ib_mr *ibmr = trans_private;
	struct rds_ib_device *rds_ibdev = ibmr->device;

	switch (direction) {
	case DMA_FROM_DEVICE:
		ib_dma_sync_sg_for_cpu(rds_ibdev->dev, ibmr->sg,
			ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
		break;
	case DMA_TO_DEVICE:
		ib_dma_sync_sg_for_device(rds_ibdev->dev, ibmr->sg,
			ibmr->sg_dma_len, DMA_BIDIRECTIONAL);
		break;
	}
}

static void __rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
{
	struct rds_ib_device *rds_ibdev = ibmr->device;

	if (ibmr->sg_dma_len) {
		ib_dma_unmap_sg(rds_ibdev->dev,
				ibmr->sg, ibmr->sg_len,
				DMA_BIDIRECTIONAL);
		ibmr->sg_dma_len = 0;
	}

	/* Release the s/g list */
	if (ibmr->sg_len) {
		unsigned int i;

		for (i = 0; i < ibmr->sg_len; ++i) {
			struct page *page = sg_page(&ibmr->sg[i]);

			/* FIXME we need a way to tell a r/w MR
			 * from a r/o MR */
			WARN_ON(!page->mapping && irqs_disabled());
			set_page_dirty(page);
			put_page(page);
		}
		kfree(ibmr->sg);

		ibmr->sg = NULL;
		ibmr->sg_len = 0;
	}
}

static void rds_ib_teardown_mr(struct rds_ib_mr *ibmr)
{
	unsigned int pinned = ibmr->sg_len;

	__rds_ib_teardown_mr(ibmr);
	if (pinned) {
		struct rds_ib_mr_pool *pool = ibmr->pool;

		atomic_sub(pinned, &pool->free_pinned);
	}
}

static inline unsigned int rds_ib_flush_goal(struct rds_ib_mr_pool *pool, int free_all)
{
	unsigned int item_count;

	item_count = atomic_read(&pool->item_count);
	if (free_all)
		return item_count;

	return 0;
}

/*
 * given an llist of mrs, put them all into the list_head for more processing
 */
static unsigned int llist_append_to_list(struct llist_head *llist,
					 struct list_head *list)
{
	struct rds_ib_mr *ibmr;
	struct llist_node *node;
	struct llist_node *next;
	unsigned int count = 0;

	node = llist_del_all(llist);
	while (node) {
		next = node->next;
		ibmr = llist_entry(node, struct rds_ib_mr, llnode);
		list_add_tail(&ibmr->unmap_list, list);
		node = next;
		count++;
	}
	return count;
}

/*
 * this takes a list head of mrs and turns it into linked llist nodes
 * of clusters.  Each cluster has linked llist nodes of
 * MR_CLUSTER_SIZE mrs that are ready for reuse.
 */
static void list_to_llist_nodes(struct rds_ib_mr_pool *pool,
				struct list_head *list,
				struct llist_node **nodes_head,
				struct llist_node **nodes_tail)
{
	struct rds_ib_mr *ibmr;
	struct llist_node *cur = NULL;
	struct llist_node **next = nodes_head;

	list_for_each_entry(ibmr, list, unmap_list) {
		cur = &ibmr->llnode;
		*next = cur;
		next = &cur->next;
	}
	*next = NULL;
	*nodes_tail = cur;
}

/*
 * Flush our pool of MRs.
 * At a minimum, all currently unused MRs are unmapped.
 * If the number of MRs allocated exceeds the limit, we also try
 * to free as many MRs as needed to get back to this limit.
 */
static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
				int free_all, struct rds_ib_mr **ibmr_ret)
{
	struct rds_ib_mr *ibmr, *next;
	struct llist_node *clean_nodes;
	struct llist_node *clean_tail;
	LIST_HEAD(unmap_list);
	LIST_HEAD(fmr_list);
	unsigned long unpinned = 0;
	unsigned int nfreed = 0, dirty_to_clean = 0, free_goal;
	int ret = 0;

	if (pool->pool_type == RDS_IB_MR_8K_POOL)
		rds_ib_stats_inc(s_ib_rdma_mr_8k_pool_flush);
	else
		rds_ib_stats_inc(s_ib_rdma_mr_1m_pool_flush);

	if (ibmr_ret) {
		DEFINE_WAIT(wait);
		while (!mutex_trylock(&pool->flush_lock)) {
			ibmr = rds_ib_reuse_fmr(pool);
			if (ibmr) {
				*ibmr_ret = ibmr;
				finish_wait(&pool->flush_wait, &wait);
				goto out_nolock;
			}

			prepare_to_wait(&pool->flush_wait, &wait,
					TASK_UNINTERRUPTIBLE);
			if (llist_empty(&pool->clean_list))
				schedule();

			ibmr = rds_ib_reuse_fmr(pool);
			if (ibmr) {
				*ibmr_ret = ibmr;
				finish_wait(&pool->flush_wait, &wait);
				goto out_nolock;
			}
		}
		finish_wait(&pool->flush_wait, &wait);
	} else
		mutex_lock(&pool->flush_lock);

	if (ibmr_ret) {
		ibmr = rds_ib_reuse_fmr(pool);
		if (ibmr) {
			*ibmr_ret = ibmr;
			goto out;
		}
	}

	/* Get the list of all MRs to be dropped. Ordering matters -
	 * we want to put drop_list ahead of free_list.
	 */
	dirty_to_clean = llist_append_to_list(&pool->drop_list, &unmap_list);
	dirty_to_clean += llist_append_to_list(&pool->free_list, &unmap_list);
	if (free_all)
		llist_append_to_list(&pool->clean_list, &unmap_list);

	free_goal = rds_ib_flush_goal(pool, free_all);

	if (list_empty(&unmap_list))
		goto out;

	/* String all ib_mr's onto one list and hand them to ib_unmap_fmr */
	list_for_each_entry(ibmr, &unmap_list, unmap_list)
		list_add(&ibmr->fmr->list, &fmr_list);

	ret = ib_unmap_fmr(&fmr_list);
	if (ret)
		printk(KERN_WARNING "RDS/IB: ib_unmap_fmr failed (err=%d)\n", ret);

	/* Now we can destroy the DMA mapping and unpin any pages */
	list_for_each_entry_safe(ibmr, next, &unmap_list, unmap_list) {
		unpinned += ibmr->sg_len;
		__rds_ib_teardown_mr(ibmr);
		if (nfreed < free_goal ||
		    ibmr->remap_count >= pool->fmr_attr.max_maps) {
			if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
				rds_ib_stats_inc(s_ib_rdma_mr_8k_free);
			else
				rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
			list_del(&ibmr->unmap_list);
			ib_dealloc_fmr(ibmr->fmr);
			kfree(ibmr);
			nfreed++;
		}
	}

	if (!list_empty(&unmap_list)) {
		/* we have to make sure that none of the things we're about
		 * to put on the clean list would race with other cpus trying
		 * to pull items off.  The llist would explode if we managed to
		 * remove something from the clean list and then add it back again
		 * while another CPU was spinning on that same item in llist_del_first.
		 *
		 * This is pretty unlikely, but just in case  wait for an llist grace period
		 * here before adding anything back into the clean list.
		 */
		wait_clean_list_grace();

		list_to_llist_nodes(pool, &unmap_list, &clean_nodes, &clean_tail);
		if (ibmr_ret)
			*ibmr_ret = llist_entry(clean_nodes, struct rds_ib_mr, llnode);

		/* more than one entry in llist nodes */
		if (clean_nodes->next)
			llist_add_batch(clean_nodes->next, clean_tail, &pool->clean_list);

	}

	atomic_sub(unpinned, &pool->free_pinned);
	atomic_sub(dirty_to_clean, &pool->dirty_count);
	atomic_sub(nfreed, &pool->item_count);

out:
	mutex_unlock(&pool->flush_lock);
	if (waitqueue_active(&pool->flush_wait))
		wake_up(&pool->flush_wait);
out_nolock:
	return ret;
}

static void rds_ib_mr_pool_flush_worker(struct work_struct *work)
{
	struct rds_ib_mr_pool *pool = container_of(work, struct rds_ib_mr_pool, flush_worker.work);

	rds_ib_flush_mr_pool(pool, 0, NULL);
}

void rds_ib_free_mr(void *trans_private, int invalidate)
{
	struct rds_ib_mr *ibmr = trans_private;
	struct rds_ib_mr_pool *pool = ibmr->pool;
	struct rds_ib_device *rds_ibdev = ibmr->device;

	rdsdebug("RDS/IB: free_mr nents %u\n", ibmr->sg_len);

	/* Return it to the pool's free list */
	if (ibmr->remap_count >= pool->fmr_attr.max_maps)
		llist_add(&ibmr->llnode, &pool->drop_list);
	else
		llist_add(&ibmr->llnode, &pool->free_list);

	atomic_add(ibmr->sg_len, &pool->free_pinned);
	atomic_inc(&pool->dirty_count);

	/* If we've pinned too many pages, request a flush */
	if (atomic_read(&pool->free_pinned) >= pool->max_free_pinned ||
	    atomic_read(&pool->dirty_count) >= pool->max_items / 5)
		queue_delayed_work(rds_ib_fmr_wq, &pool->flush_worker, 10);

	if (invalidate) {
		if (likely(!in_interrupt())) {
			rds_ib_flush_mr_pool(pool, 0, NULL);
		} else {
			/* We get here if the user created a MR marked
			 * as use_once and invalidate at the same time.
			 */
			queue_delayed_work(rds_ib_fmr_wq,
					   &pool->flush_worker, 10);
		}
	}

	rds_ib_dev_put(rds_ibdev);
}

void rds_ib_flush_mrs(void)
{
	struct rds_ib_device *rds_ibdev;

	down_read(&rds_ib_devices_lock);
	list_for_each_entry(rds_ibdev, &rds_ib_devices, list) {
		if (rds_ibdev->mr_8k_pool)
			rds_ib_flush_mr_pool(rds_ibdev->mr_8k_pool, 0, NULL);

		if (rds_ibdev->mr_1m_pool)
			rds_ib_flush_mr_pool(rds_ibdev->mr_1m_pool, 0, NULL);
	}
	up_read(&rds_ib_devices_lock);
}

void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
		    struct rds_sock *rs, u32 *key_ret)
{
	struct rds_ib_device *rds_ibdev;
	struct rds_ib_mr *ibmr = NULL;
	int ret;

	rds_ibdev = rds_ib_get_device(rs->rs_bound_addr);
	if (!rds_ibdev) {
		ret = -ENODEV;
		goto out;
	}

	if (!rds_ibdev->mr_8k_pool || !rds_ibdev->mr_1m_pool) {
		ret = -ENODEV;
		goto out;
	}

	ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
	if (IS_ERR(ibmr)) {
		rds_ib_dev_put(rds_ibdev);
		return ibmr;
	}

	ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
	if (ret == 0)
		*key_ret = ibmr->fmr->rkey;
	else
		printk(KERN_WARNING "RDS/IB: map_fmr failed (errno=%d)\n", ret);

	ibmr->device = rds_ibdev;
	rds_ibdev = NULL;

 out:
	if (ret) {
		if (ibmr)
			rds_ib_free_mr(ibmr, 0);
		ibmr = ERR_PTR(ret);
	}
	if (rds_ibdev)
		rds_ib_dev_put(rds_ibdev);
	return ibmr;
}