summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/hfi1/user_exp_rcv.c
blob: dbe7d14a5c76d18f23ddf13e264f97c56d086d86 (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
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
/*
 * Copyright(c) 2015-2018 Intel Corporation.
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.
 *
 * GPL LICENSE SUMMARY
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * BSD LICENSE
 *
 * 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.
 *  - Neither the name of Intel Corporation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */
#include <asm/page.h>
#include <linux/string.h>

#include "mmu_rb.h"
#include "user_exp_rcv.h"
#include "trace.h"

static void unlock_exp_tids(struct hfi1_ctxtdata *uctxt,
			    struct exp_tid_set *set,
			    struct hfi1_filedata *fd);
static u32 find_phys_blocks(struct tid_user_buf *tidbuf, unsigned int npages);
static int set_rcvarray_entry(struct hfi1_filedata *fd,
			      struct tid_user_buf *tbuf,
			      u32 rcventry, struct tid_group *grp,
			      u16 pageidx, unsigned int npages);
static int tid_rb_insert(void *arg, struct mmu_rb_node *node);
static void cacheless_tid_rb_remove(struct hfi1_filedata *fdata,
				    struct tid_rb_node *tnode);
static void tid_rb_remove(void *arg, struct mmu_rb_node *node);
static int tid_rb_invalidate(void *arg, struct mmu_rb_node *mnode);
static int program_rcvarray(struct hfi1_filedata *fd, struct tid_user_buf *,
			    struct tid_group *grp,
			    unsigned int start, u16 count,
			    u32 *tidlist, unsigned int *tididx,
			    unsigned int *pmapped);
static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo,
			      struct tid_group **grp);
static void clear_tid_node(struct hfi1_filedata *fd, struct tid_rb_node *node);

static struct mmu_rb_ops tid_rb_ops = {
	.insert = tid_rb_insert,
	.remove = tid_rb_remove,
	.invalidate = tid_rb_invalidate
};

/*
 * Initialize context and file private data needed for Expected
 * receive caching. This needs to be done after the context has
 * been configured with the eager/expected RcvEntry counts.
 */
int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd,
			   struct hfi1_ctxtdata *uctxt)
{
	struct hfi1_devdata *dd = uctxt->dd;
	int ret = 0;

	spin_lock_init(&fd->tid_lock);
	spin_lock_init(&fd->invalid_lock);

	fd->entry_to_rb = kcalloc(uctxt->expected_count,
				  sizeof(struct rb_node *),
				  GFP_KERNEL);
	if (!fd->entry_to_rb)
		return -ENOMEM;

	if (!HFI1_CAP_UGET_MASK(uctxt->flags, TID_UNMAP)) {
		fd->invalid_tid_idx = 0;
		fd->invalid_tids = kcalloc(uctxt->expected_count,
					   sizeof(*fd->invalid_tids),
					   GFP_KERNEL);
		if (!fd->invalid_tids) {
			kfree(fd->entry_to_rb);
			fd->entry_to_rb = NULL;
			return -ENOMEM;
		}

		/*
		 * Register MMU notifier callbacks. If the registration
		 * fails, continue without TID caching for this context.
		 */
		ret = hfi1_mmu_rb_register(fd, fd->mm, &tid_rb_ops,
					   dd->pport->hfi1_wq,
					   &fd->handler);
		if (ret) {
			dd_dev_info(dd,
				    "Failed MMU notifier registration %d\n",
				    ret);
			ret = 0;
		}
	}

	/*
	 * PSM does not have a good way to separate, count, and
	 * effectively enforce a limit on RcvArray entries used by
	 * subctxts (when context sharing is used) when TID caching
	 * is enabled. To help with that, we calculate a per-process
	 * RcvArray entry share and enforce that.
	 * If TID caching is not in use, PSM deals with usage on its
	 * own. In that case, we allow any subctxt to take all of the
	 * entries.
	 *
	 * Make sure that we set the tid counts only after successful
	 * init.
	 */
	spin_lock(&fd->tid_lock);
	if (uctxt->subctxt_cnt && fd->handler) {
		u16 remainder;

		fd->tid_limit = uctxt->expected_count / uctxt->subctxt_cnt;
		remainder = uctxt->expected_count % uctxt->subctxt_cnt;
		if (remainder && fd->subctxt < remainder)
			fd->tid_limit++;
	} else {
		fd->tid_limit = uctxt->expected_count;
	}
	spin_unlock(&fd->tid_lock);

	return ret;
}

void hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;

	/*
	 * The notifier would have been removed when the process'es mm
	 * was freed.
	 */
	if (fd->handler) {
		hfi1_mmu_rb_unregister(fd->handler);
	} else {
		if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list))
			unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd);
		if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list))
			unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd);
	}

	kfree(fd->invalid_tids);
	fd->invalid_tids = NULL;

	kfree(fd->entry_to_rb);
	fd->entry_to_rb = NULL;
}

/**
 * Release pinned receive buffer pages.
 *
 * @mapped - true if the pages have been DMA mapped. false otherwise.
 * @idx - Index of the first page to unpin.
 * @npages - No of pages to unpin.
 *
 * If the pages have been DMA mapped (indicated by mapped parameter), their
 * info will be passed via a struct tid_rb_node. If they haven't been mapped,
 * their info will be passed via a struct tid_user_buf.
 */
static void unpin_rcv_pages(struct hfi1_filedata *fd,
			    struct tid_user_buf *tidbuf,
			    struct tid_rb_node *node,
			    unsigned int idx,
			    unsigned int npages,
			    bool mapped)
{
	struct page **pages;
	struct hfi1_devdata *dd = fd->uctxt->dd;

	if (mapped) {
		pci_unmap_single(dd->pcidev, node->dma_addr,
				 node->mmu.len, PCI_DMA_FROMDEVICE);
		pages = &node->pages[idx];
	} else {
		pages = &tidbuf->pages[idx];
	}
	hfi1_release_user_pages(fd->mm, pages, npages, mapped);
	fd->tid_n_pinned -= npages;
}

/**
 * Pin receive buffer pages.
 */
static int pin_rcv_pages(struct hfi1_filedata *fd, struct tid_user_buf *tidbuf)
{
	int pinned;
	unsigned int npages;
	unsigned long vaddr = tidbuf->vaddr;
	struct page **pages = NULL;
	struct hfi1_devdata *dd = fd->uctxt->dd;

	/* Get the number of pages the user buffer spans */
	npages = num_user_pages(vaddr, tidbuf->length);
	if (!npages)
		return -EINVAL;

	if (npages > fd->uctxt->expected_count) {
		dd_dev_err(dd, "Expected buffer too big\n");
		return -EINVAL;
	}

	/* Verify that access is OK for the user buffer */
	if (!access_ok(VERIFY_WRITE, (void __user *)vaddr,
		       npages * PAGE_SIZE)) {
		dd_dev_err(dd, "Fail vaddr %p, %u pages, !access_ok\n",
			   (void *)vaddr, npages);
		return -EFAULT;
	}
	/* Allocate the array of struct page pointers needed for pinning */
	pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
	if (!pages)
		return -ENOMEM;

	/*
	 * Pin all the pages of the user buffer. If we can't pin all the
	 * pages, accept the amount pinned so far and program only that.
	 * User space knows how to deal with partially programmed buffers.
	 */
	if (!hfi1_can_pin_pages(dd, fd->mm, fd->tid_n_pinned, npages)) {
		kfree(pages);
		return -ENOMEM;
	}

	pinned = hfi1_acquire_user_pages(fd->mm, vaddr, npages, true, pages);
	if (pinned <= 0) {
		kfree(pages);
		return pinned;
	}
	tidbuf->pages = pages;
	tidbuf->npages = npages;
	fd->tid_n_pinned += pinned;
	return pinned;
}

/*
 * RcvArray entry allocation for Expected Receives is done by the
 * following algorithm:
 *
 * The context keeps 3 lists of groups of RcvArray entries:
 *   1. List of empty groups - tid_group_list
 *      This list is created during user context creation and
 *      contains elements which describe sets (of 8) of empty
 *      RcvArray entries.
 *   2. List of partially used groups - tid_used_list
 *      This list contains sets of RcvArray entries which are
 *      not completely used up. Another mapping request could
 *      use some of all of the remaining entries.
 *   3. List of full groups - tid_full_list
 *      This is the list where sets that are completely used
 *      up go.
 *
 * An attempt to optimize the usage of RcvArray entries is
 * made by finding all sets of physically contiguous pages in a
 * user's buffer.
 * These physically contiguous sets are further split into
 * sizes supported by the receive engine of the HFI. The
 * resulting sets of pages are stored in struct tid_pageset,
 * which describes the sets as:
 *    * .count - number of pages in this set
 *    * .idx - starting index into struct page ** array
 *                    of this set
 *
 * From this point on, the algorithm deals with the page sets
 * described above. The number of pagesets is divided by the
 * RcvArray group size to produce the number of full groups
 * needed.
 *
 * Groups from the 3 lists are manipulated using the following
 * rules:
 *   1. For each set of 8 pagesets, a complete group from
 *      tid_group_list is taken, programmed, and moved to
 *      the tid_full_list list.
 *   2. For all remaining pagesets:
 *      2.1 If the tid_used_list is empty and the tid_group_list
 *          is empty, stop processing pageset and return only
 *          what has been programmed up to this point.
 *      2.2 If the tid_used_list is empty and the tid_group_list
 *          is not empty, move a group from tid_group_list to
 *          tid_used_list.
 *      2.3 For each group is tid_used_group, program as much as
 *          can fit into the group. If the group becomes fully
 *          used, move it to tid_full_list.
 */
int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd,
			    struct hfi1_tid_info *tinfo)
{
	int ret = 0, need_group = 0, pinned;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	unsigned int ngroups, pageidx = 0, pageset_count,
		tididx = 0, mapped, mapped_pages = 0;
	u32 *tidlist = NULL;
	struct tid_user_buf *tidbuf;

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

	tidbuf->vaddr = tinfo->vaddr;
	tidbuf->length = tinfo->length;
	tidbuf->psets = kcalloc(uctxt->expected_count, sizeof(*tidbuf->psets),
				GFP_KERNEL);
	if (!tidbuf->psets) {
		kfree(tidbuf);
		return -ENOMEM;
	}

	pinned = pin_rcv_pages(fd, tidbuf);
	if (pinned <= 0) {
		kfree(tidbuf->psets);
		kfree(tidbuf);
		return pinned;
	}

	/* Find sets of physically contiguous pages */
	tidbuf->n_psets = find_phys_blocks(tidbuf, pinned);

	/*
	 * We don't need to access this under a lock since tid_used is per
	 * process and the same process cannot be in hfi1_user_exp_rcv_clear()
	 * and hfi1_user_exp_rcv_setup() at the same time.
	 */
	spin_lock(&fd->tid_lock);
	if (fd->tid_used + tidbuf->n_psets > fd->tid_limit)
		pageset_count = fd->tid_limit - fd->tid_used;
	else
		pageset_count = tidbuf->n_psets;
	spin_unlock(&fd->tid_lock);

	if (!pageset_count)
		goto bail;

	ngroups = pageset_count / dd->rcv_entries.group_size;
	tidlist = kcalloc(pageset_count, sizeof(*tidlist), GFP_KERNEL);
	if (!tidlist) {
		ret = -ENOMEM;
		goto nomem;
	}

	tididx = 0;

	/*
	 * From this point on, we are going to be using shared (between master
	 * and subcontexts) context resources. We need to take the lock.
	 */
	mutex_lock(&uctxt->exp_mutex);
	/*
	 * The first step is to program the RcvArray entries which are complete
	 * groups.
	 */
	while (ngroups && uctxt->tid_group_list.count) {
		struct tid_group *grp =
			tid_group_pop(&uctxt->tid_group_list);

		ret = program_rcvarray(fd, tidbuf, grp,
				       pageidx, dd->rcv_entries.group_size,
				       tidlist, &tididx, &mapped);
		/*
		 * If there was a failure to program the RcvArray
		 * entries for the entire group, reset the grp fields
		 * and add the grp back to the free group list.
		 */
		if (ret <= 0) {
			tid_group_add_tail(grp, &uctxt->tid_group_list);
			hfi1_cdbg(TID,
				  "Failed to program RcvArray group %d", ret);
			goto unlock;
		}

		tid_group_add_tail(grp, &uctxt->tid_full_list);
		ngroups--;
		pageidx += ret;
		mapped_pages += mapped;
	}

	while (pageidx < pageset_count) {
		struct tid_group *grp, *ptr;
		/*
		 * If we don't have any partially used tid groups, check
		 * if we have empty groups. If so, take one from there and
		 * put in the partially used list.
		 */
		if (!uctxt->tid_used_list.count || need_group) {
			if (!uctxt->tid_group_list.count)
				goto unlock;

			grp = tid_group_pop(&uctxt->tid_group_list);
			tid_group_add_tail(grp, &uctxt->tid_used_list);
			need_group = 0;
		}
		/*
		 * There is an optimization opportunity here - instead of
		 * fitting as many page sets as we can, check for a group
		 * later on in the list that could fit all of them.
		 */
		list_for_each_entry_safe(grp, ptr, &uctxt->tid_used_list.list,
					 list) {
			unsigned use = min_t(unsigned, pageset_count - pageidx,
					     grp->size - grp->used);

			ret = program_rcvarray(fd, tidbuf, grp,
					       pageidx, use, tidlist,
					       &tididx, &mapped);
			if (ret < 0) {
				hfi1_cdbg(TID,
					  "Failed to program RcvArray entries %d",
					  ret);
				goto unlock;
			} else if (ret > 0) {
				if (grp->used == grp->size)
					tid_group_move(grp,
						       &uctxt->tid_used_list,
						       &uctxt->tid_full_list);
				pageidx += ret;
				mapped_pages += mapped;
				need_group = 0;
				/* Check if we are done so we break out early */
				if (pageidx >= pageset_count)
					break;
			} else if (WARN_ON(ret == 0)) {
				/*
				 * If ret is 0, we did not program any entries
				 * into this group, which can only happen if
				 * we've screwed up the accounting somewhere.
				 * Warn and try to continue.
				 */
				need_group = 1;
			}
		}
	}
unlock:
	mutex_unlock(&uctxt->exp_mutex);
nomem:
	hfi1_cdbg(TID, "total mapped: tidpairs:%u pages:%u (%d)", tididx,
		  mapped_pages, ret);
	if (tididx) {
		spin_lock(&fd->tid_lock);
		fd->tid_used += tididx;
		spin_unlock(&fd->tid_lock);
		tinfo->tidcnt = tididx;
		tinfo->length = mapped_pages * PAGE_SIZE;

		if (copy_to_user(u64_to_user_ptr(tinfo->tidlist),
				 tidlist, sizeof(tidlist[0]) * tididx)) {
			/*
			 * On failure to copy to the user level, we need to undo
			 * everything done so far so we don't leak resources.
			 */
			tinfo->tidlist = (unsigned long)&tidlist;
			hfi1_user_exp_rcv_clear(fd, tinfo);
			tinfo->tidlist = 0;
			ret = -EFAULT;
			goto bail;
		}
	}

	/*
	 * If not everything was mapped (due to insufficient RcvArray entries,
	 * for example), unpin all unmapped pages so we can pin them nex time.
	 */
	if (mapped_pages != pinned)
		unpin_rcv_pages(fd, tidbuf, NULL, mapped_pages,
				(pinned - mapped_pages), false);
bail:
	kfree(tidbuf->psets);
	kfree(tidlist);
	kfree(tidbuf->pages);
	kfree(tidbuf);
	return ret > 0 ? 0 : ret;
}

int hfi1_user_exp_rcv_clear(struct hfi1_filedata *fd,
			    struct hfi1_tid_info *tinfo)
{
	int ret = 0;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	u32 *tidinfo;
	unsigned tididx;

	if (unlikely(tinfo->tidcnt > fd->tid_used))
		return -EINVAL;

	tidinfo = memdup_user(u64_to_user_ptr(tinfo->tidlist),
			      sizeof(tidinfo[0]) * tinfo->tidcnt);
	if (IS_ERR(tidinfo))
		return PTR_ERR(tidinfo);

	mutex_lock(&uctxt->exp_mutex);
	for (tididx = 0; tididx < tinfo->tidcnt; tididx++) {
		ret = unprogram_rcvarray(fd, tidinfo[tididx], NULL);
		if (ret) {
			hfi1_cdbg(TID, "Failed to unprogram rcv array %d",
				  ret);
			break;
		}
	}
	spin_lock(&fd->tid_lock);
	fd->tid_used -= tididx;
	spin_unlock(&fd->tid_lock);
	tinfo->tidcnt = tididx;
	mutex_unlock(&uctxt->exp_mutex);

	kfree(tidinfo);
	return ret;
}

int hfi1_user_exp_rcv_invalid(struct hfi1_filedata *fd,
			      struct hfi1_tid_info *tinfo)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	unsigned long *ev = uctxt->dd->events +
		(uctxt_offset(uctxt) + fd->subctxt);
	u32 *array;
	int ret = 0;

	/*
	 * copy_to_user() can sleep, which will leave the invalid_lock
	 * locked and cause the MMU notifier to be blocked on the lock
	 * for a long time.
	 * Copy the data to a local buffer so we can release the lock.
	 */
	array = kcalloc(uctxt->expected_count, sizeof(*array), GFP_KERNEL);
	if (!array)
		return -EFAULT;

	spin_lock(&fd->invalid_lock);
	if (fd->invalid_tid_idx) {
		memcpy(array, fd->invalid_tids, sizeof(*array) *
		       fd->invalid_tid_idx);
		memset(fd->invalid_tids, 0, sizeof(*fd->invalid_tids) *
		       fd->invalid_tid_idx);
		tinfo->tidcnt = fd->invalid_tid_idx;
		fd->invalid_tid_idx = 0;
		/*
		 * Reset the user flag while still holding the lock.
		 * Otherwise, PSM can miss events.
		 */
		clear_bit(_HFI1_EVENT_TID_MMU_NOTIFY_BIT, ev);
	} else {
		tinfo->tidcnt = 0;
	}
	spin_unlock(&fd->invalid_lock);

	if (tinfo->tidcnt) {
		if (copy_to_user((void __user *)tinfo->tidlist,
				 array, sizeof(*array) * tinfo->tidcnt))
			ret = -EFAULT;
	}
	kfree(array);

	return ret;
}

static u32 find_phys_blocks(struct tid_user_buf *tidbuf, unsigned int npages)
{
	unsigned pagecount, pageidx, setcount = 0, i;
	unsigned long pfn, this_pfn;
	struct page **pages = tidbuf->pages;
	struct tid_pageset *list = tidbuf->psets;

	if (!npages)
		return 0;

	/*
	 * Look for sets of physically contiguous pages in the user buffer.
	 * This will allow us to optimize Expected RcvArray entry usage by
	 * using the bigger supported sizes.
	 */
	pfn = page_to_pfn(pages[0]);
	for (pageidx = 0, pagecount = 1, i = 1; i <= npages; i++) {
		this_pfn = i < npages ? page_to_pfn(pages[i]) : 0;

		/*
		 * If the pfn's are not sequential, pages are not physically
		 * contiguous.
		 */
		if (this_pfn != ++pfn) {
			/*
			 * At this point we have to loop over the set of
			 * physically contiguous pages and break them down it
			 * sizes supported by the HW.
			 * There are two main constraints:
			 *     1. The max buffer size is MAX_EXPECTED_BUFFER.
			 *        If the total set size is bigger than that
			 *        program only a MAX_EXPECTED_BUFFER chunk.
			 *     2. The buffer size has to be a power of two. If
			 *        it is not, round down to the closes power of
			 *        2 and program that size.
			 */
			while (pagecount) {
				int maxpages = pagecount;
				u32 bufsize = pagecount * PAGE_SIZE;

				if (bufsize > MAX_EXPECTED_BUFFER)
					maxpages =
						MAX_EXPECTED_BUFFER >>
						PAGE_SHIFT;
				else if (!is_power_of_2(bufsize))
					maxpages =
						rounddown_pow_of_two(bufsize) >>
						PAGE_SHIFT;

				list[setcount].idx = pageidx;
				list[setcount].count = maxpages;
				pagecount -= maxpages;
				pageidx += maxpages;
				setcount++;
			}
			pageidx = i;
			pagecount = 1;
			pfn = this_pfn;
		} else {
			pagecount++;
		}
	}
	return setcount;
}

/**
 * program_rcvarray() - program an RcvArray group with receive buffers
 * @fd: filedata pointer
 * @tbuf: pointer to struct tid_user_buf that has the user buffer starting
 *	  virtual address, buffer length, page pointers, pagesets (array of
 *	  struct tid_pageset holding information on physically contiguous
 *	  chunks from the user buffer), and other fields.
 * @grp: RcvArray group
 * @start: starting index into sets array
 * @count: number of struct tid_pageset's to program
 * @tidlist: the array of u32 elements when the information about the
 *           programmed RcvArray entries is to be encoded.
 * @tididx: starting offset into tidlist
 * @pmapped: (output parameter) number of pages programmed into the RcvArray
 *           entries.
 *
 * This function will program up to 'count' number of RcvArray entries from the
 * group 'grp'. To make best use of write-combining writes, the function will
 * perform writes to the unused RcvArray entries which will be ignored by the
 * HW. Each RcvArray entry will be programmed with a physically contiguous
 * buffer chunk from the user's virtual buffer.
 *
 * Return:
 * -EINVAL if the requested count is larger than the size of the group,
 * -ENOMEM or -EFAULT on error from set_rcvarray_entry(), or
 * number of RcvArray entries programmed.
 */
static int program_rcvarray(struct hfi1_filedata *fd, struct tid_user_buf *tbuf,
			    struct tid_group *grp,
			    unsigned int start, u16 count,
			    u32 *tidlist, unsigned int *tididx,
			    unsigned int *pmapped)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	u16 idx;
	u32 tidinfo = 0, rcventry, useidx = 0;
	int mapped = 0;

	/* Count should never be larger than the group size */
	if (count > grp->size)
		return -EINVAL;

	/* Find the first unused entry in the group */
	for (idx = 0; idx < grp->size; idx++) {
		if (!(grp->map & (1 << idx))) {
			useidx = idx;
			break;
		}
		rcv_array_wc_fill(dd, grp->base + idx);
	}

	idx = 0;
	while (idx < count) {
		u16 npages, pageidx, setidx = start + idx;
		int ret = 0;

		/*
		 * If this entry in the group is used, move to the next one.
		 * If we go past the end of the group, exit the loop.
		 */
		if (useidx >= grp->size) {
			break;
		} else if (grp->map & (1 << useidx)) {
			rcv_array_wc_fill(dd, grp->base + useidx);
			useidx++;
			continue;
		}

		rcventry = grp->base + useidx;
		npages = tbuf->psets[setidx].count;
		pageidx = tbuf->psets[setidx].idx;

		ret = set_rcvarray_entry(fd, tbuf,
					 rcventry, grp, pageidx,
					 npages);
		if (ret)
			return ret;
		mapped += npages;

		tidinfo = rcventry2tidinfo(rcventry - uctxt->expected_base) |
			EXP_TID_SET(LEN, npages);
		tidlist[(*tididx)++] = tidinfo;
		grp->used++;
		grp->map |= 1 << useidx++;
		idx++;
	}

	/* Fill the rest of the group with "blank" writes */
	for (; useidx < grp->size; useidx++)
		rcv_array_wc_fill(dd, grp->base + useidx);
	*pmapped = mapped;
	return idx;
}

static int set_rcvarray_entry(struct hfi1_filedata *fd,
			      struct tid_user_buf *tbuf,
			      u32 rcventry, struct tid_group *grp,
			      u16 pageidx, unsigned int npages)
{
	int ret;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct tid_rb_node *node;
	struct hfi1_devdata *dd = uctxt->dd;
	dma_addr_t phys;
	struct page **pages = tbuf->pages + pageidx;

	/*
	 * Allocate the node first so we can handle a potential
	 * failure before we've programmed anything.
	 */
	node = kzalloc(sizeof(*node) + (sizeof(struct page *) * npages),
		       GFP_KERNEL);
	if (!node)
		return -ENOMEM;

	phys = pci_map_single(dd->pcidev,
			      __va(page_to_phys(pages[0])),
			      npages * PAGE_SIZE, PCI_DMA_FROMDEVICE);
	if (dma_mapping_error(&dd->pcidev->dev, phys)) {
		dd_dev_err(dd, "Failed to DMA map Exp Rcv pages 0x%llx\n",
			   phys);
		kfree(node);
		return -EFAULT;
	}

	node->mmu.addr = tbuf->vaddr + (pageidx * PAGE_SIZE);
	node->mmu.len = npages * PAGE_SIZE;
	node->phys = page_to_phys(pages[0]);
	node->npages = npages;
	node->rcventry = rcventry;
	node->dma_addr = phys;
	node->grp = grp;
	node->freed = false;
	memcpy(node->pages, pages, sizeof(struct page *) * npages);

	if (!fd->handler)
		ret = tid_rb_insert(fd, &node->mmu);
	else
		ret = hfi1_mmu_rb_insert(fd->handler, &node->mmu);

	if (ret) {
		hfi1_cdbg(TID, "Failed to insert RB node %u 0x%lx, 0x%lx %d",
			  node->rcventry, node->mmu.addr, node->phys, ret);
		pci_unmap_single(dd->pcidev, phys, npages * PAGE_SIZE,
				 PCI_DMA_FROMDEVICE);
		kfree(node);
		return -EFAULT;
	}
	hfi1_put_tid(dd, rcventry, PT_EXPECTED, phys, ilog2(npages) + 1);
	trace_hfi1_exp_tid_reg(uctxt->ctxt, fd->subctxt, rcventry, npages,
			       node->mmu.addr, node->phys, phys);
	return 0;
}

static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo,
			      struct tid_group **grp)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	struct tid_rb_node *node;
	u8 tidctrl = EXP_TID_GET(tidinfo, CTRL);
	u32 tididx = EXP_TID_GET(tidinfo, IDX) << 1, rcventry;

	if (tididx >= uctxt->expected_count) {
		dd_dev_err(dd, "Invalid RcvArray entry (%u) index for ctxt %u\n",
			   tididx, uctxt->ctxt);
		return -EINVAL;
	}

	if (tidctrl == 0x3)
		return -EINVAL;

	rcventry = tididx + (tidctrl - 1);

	node = fd->entry_to_rb[rcventry];
	if (!node || node->rcventry != (uctxt->expected_base + rcventry))
		return -EBADF;

	if (grp)
		*grp = node->grp;

	if (!fd->handler)
		cacheless_tid_rb_remove(fd, node);
	else
		hfi1_mmu_rb_remove(fd->handler, &node->mmu);

	return 0;
}

static void clear_tid_node(struct hfi1_filedata *fd, struct tid_rb_node *node)
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;

	trace_hfi1_exp_tid_unreg(uctxt->ctxt, fd->subctxt, node->rcventry,
				 node->npages, node->mmu.addr, node->phys,
				 node->dma_addr);

	/*
	 * Make sure device has seen the write before we unpin the
	 * pages.
	 */
	hfi1_put_tid(dd, node->rcventry, PT_INVALID_FLUSH, 0, 0);

	unpin_rcv_pages(fd, NULL, node, 0, node->npages, true);

	node->grp->used--;
	node->grp->map &= ~(1 << (node->rcventry - node->grp->base));

	if (node->grp->used == node->grp->size - 1)
		tid_group_move(node->grp, &uctxt->tid_full_list,
			       &uctxt->tid_used_list);
	else if (!node->grp->used)
		tid_group_move(node->grp, &uctxt->tid_used_list,
			       &uctxt->tid_group_list);
	kfree(node);
}

/*
 * As a simple helper for hfi1_user_exp_rcv_free, this function deals with
 * clearing nodes in the non-cached case.
 */
static void unlock_exp_tids(struct hfi1_ctxtdata *uctxt,
			    struct exp_tid_set *set,
			    struct hfi1_filedata *fd)
{
	struct tid_group *grp, *ptr;
	int i;

	list_for_each_entry_safe(grp, ptr, &set->list, list) {
		list_del_init(&grp->list);

		for (i = 0; i < grp->size; i++) {
			if (grp->map & (1 << i)) {
				u16 rcventry = grp->base + i;
				struct tid_rb_node *node;

				node = fd->entry_to_rb[rcventry -
							  uctxt->expected_base];
				if (!node || node->rcventry != rcventry)
					continue;

				cacheless_tid_rb_remove(fd, node);
			}
		}
	}
}

/*
 * Always return 0 from this function.  A non-zero return indicates that the
 * remove operation will be called and that memory should be unpinned.
 * However, the driver cannot unpin out from under PSM.  Instead, retain the
 * memory (by returning 0) and inform PSM that the memory is going away.  PSM
 * will call back later when it has removed the memory from its list.
 */
static int tid_rb_invalidate(void *arg, struct mmu_rb_node *mnode)
{
	struct hfi1_filedata *fdata = arg;
	struct hfi1_ctxtdata *uctxt = fdata->uctxt;
	struct tid_rb_node *node =
		container_of(mnode, struct tid_rb_node, mmu);

	if (node->freed)
		return 0;

	trace_hfi1_exp_tid_inval(uctxt->ctxt, fdata->subctxt, node->mmu.addr,
				 node->rcventry, node->npages, node->dma_addr);
	node->freed = true;

	spin_lock(&fdata->invalid_lock);
	if (fdata->invalid_tid_idx < uctxt->expected_count) {
		fdata->invalid_tids[fdata->invalid_tid_idx] =
			rcventry2tidinfo(node->rcventry - uctxt->expected_base);
		fdata->invalid_tids[fdata->invalid_tid_idx] |=
			EXP_TID_SET(LEN, node->npages);
		if (!fdata->invalid_tid_idx) {
			unsigned long *ev;

			/*
			 * hfi1_set_uevent_bits() sets a user event flag
			 * for all processes. Because calling into the
			 * driver to process TID cache invalidations is
			 * expensive and TID cache invalidations are
			 * handled on a per-process basis, we can
			 * optimize this to set the flag only for the
			 * process in question.
			 */
			ev = uctxt->dd->events +
				(uctxt_offset(uctxt) + fdata->subctxt);
			set_bit(_HFI1_EVENT_TID_MMU_NOTIFY_BIT, ev);
		}
		fdata->invalid_tid_idx++;
	}
	spin_unlock(&fdata->invalid_lock);
	return 0;
}

static int tid_rb_insert(void *arg, struct mmu_rb_node *node)
{
	struct hfi1_filedata *fdata = arg;
	struct tid_rb_node *tnode =
		container_of(node, struct tid_rb_node, mmu);
	u32 base = fdata->uctxt->expected_base;

	fdata->entry_to_rb[tnode->rcventry - base] = tnode;
	return 0;
}

static void cacheless_tid_rb_remove(struct hfi1_filedata *fdata,
				    struct tid_rb_node *tnode)
{
	u32 base = fdata->uctxt->expected_base;

	fdata->entry_to_rb[tnode->rcventry - base] = NULL;
	clear_tid_node(fdata, tnode);
}

static void tid_rb_remove(void *arg, struct mmu_rb_node *node)
{
	struct hfi1_filedata *fdata = arg;
	struct tid_rb_node *tnode =
		container_of(node, struct tid_rb_node, mmu);

	cacheless_tid_rb_remove(fdata, tnode);
}