summaryrefslogtreecommitdiff
path: root/tools/testing/selftests/kvm/memslot_perf_test.c
blob: 8698d1ab60d00f72399571b62b49314f3a8c401f (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
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
// SPDX-License-Identifier: GPL-2.0
/*
 * A memslot-related performance benchmark.
 *
 * Copyright (C) 2021 Oracle and/or its affiliates.
 *
 * Basic guest setup / host vCPU thread code lifted from set_memory_region_test.
 */
#include <pthread.h>
#include <sched.h>
#include <semaphore.h>
#include <stdatomic.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <time.h>
#include <unistd.h>

#include <linux/compiler.h>
#include <linux/sizes.h>

#include <test_util.h>
#include <kvm_util.h>
#include <processor.h>

#define MEM_EXTRA_SIZE		SZ_64K

#define MEM_SIZE		(SZ_512M + MEM_EXTRA_SIZE)
#define MEM_GPA			SZ_256M
#define MEM_AUX_GPA		MEM_GPA
#define MEM_SYNC_GPA		MEM_AUX_GPA
#define MEM_TEST_GPA		(MEM_AUX_GPA + MEM_EXTRA_SIZE)
#define MEM_TEST_SIZE		(MEM_SIZE - MEM_EXTRA_SIZE)

/*
 * 32 MiB is max size that gets well over 100 iterations on 509 slots.
 * Considering that each slot needs to have at least one page up to
 * 8194 slots in use can then be tested (although with slightly
 * limited resolution).
 */
#define MEM_SIZE_MAP		(SZ_32M + MEM_EXTRA_SIZE)
#define MEM_TEST_MAP_SIZE	(MEM_SIZE_MAP - MEM_EXTRA_SIZE)

/*
 * 128 MiB is min size that fills 32k slots with at least one page in each
 * while at the same time gets 100+ iterations in such test
 *
 * 2 MiB chunk size like a typical huge page
 */
#define MEM_TEST_UNMAP_SIZE		SZ_128M
#define MEM_TEST_UNMAP_CHUNK_SIZE	SZ_2M

/*
 * For the move active test the middle of the test area is placed on
 * a memslot boundary: half lies in the memslot being moved, half in
 * other memslot(s).
 *
 * We have different number of memory slots, excluding the reserved
 * memory slot 0, on various architectures and configurations. The
 * memory size in this test is calculated by picking the maximal
 * last memory slot's memory size, with alignment to the largest
 * supported page size (64KB). In this way, the selected memory
 * size for this test is compatible with test_memslot_move_prepare().
 *
 * architecture   slots    memory-per-slot    memory-on-last-slot
 * --------------------------------------------------------------
 * x86-4KB        32763    16KB               160KB
 * arm64-4KB      32766    16KB               112KB
 * arm64-16KB     32766    16KB               112KB
 * arm64-64KB     8192     64KB               128KB
 */
#define MEM_TEST_MOVE_SIZE		(3 * SZ_64K)
#define MEM_TEST_MOVE_GPA_DEST		(MEM_GPA + MEM_SIZE)
static_assert(MEM_TEST_MOVE_SIZE <= MEM_TEST_SIZE,
	      "invalid move test region size");

#define MEM_TEST_VAL_1 0x1122334455667788
#define MEM_TEST_VAL_2 0x99AABBCCDDEEFF00

struct vm_data {
	struct kvm_vm *vm;
	struct kvm_vcpu *vcpu;
	pthread_t vcpu_thread;
	uint32_t nslots;
	uint64_t npages;
	uint64_t pages_per_slot;
	void **hva_slots;
	bool mmio_ok;
	uint64_t mmio_gpa_min;
	uint64_t mmio_gpa_max;
};

struct sync_area {
	uint32_t    guest_page_size;
	atomic_bool start_flag;
	atomic_bool exit_flag;
	atomic_bool sync_flag;
	void *move_area_ptr;
};

/*
 * Technically, we need also for the atomic bool to be address-free, which
 * is recommended, but not strictly required, by C11 for lockless
 * implementations.
 * However, in practice both GCC and Clang fulfill this requirement on
 * all KVM-supported platforms.
 */
static_assert(ATOMIC_BOOL_LOCK_FREE == 2, "atomic bool is not lockless");

static sem_t vcpu_ready;

static bool map_unmap_verify;

static bool verbose;
#define pr_info_v(...)				\
	do {					\
		if (verbose)			\
			pr_info(__VA_ARGS__);	\
	} while (0)

static void check_mmio_access(struct vm_data *data, struct kvm_run *run)
{
	TEST_ASSERT(data->mmio_ok, "Unexpected mmio exit");
	TEST_ASSERT(run->mmio.is_write, "Unexpected mmio read");
	TEST_ASSERT(run->mmio.len == 8,
		    "Unexpected exit mmio size = %u", run->mmio.len);
	TEST_ASSERT(run->mmio.phys_addr >= data->mmio_gpa_min &&
		    run->mmio.phys_addr <= data->mmio_gpa_max,
		    "Unexpected exit mmio address = 0x%llx",
		    run->mmio.phys_addr);
}

static void *vcpu_worker(void *__data)
{
	struct vm_data *data = __data;
	struct kvm_vcpu *vcpu = data->vcpu;
	struct kvm_run *run = vcpu->run;
	struct ucall uc;

	while (1) {
		vcpu_run(vcpu);

		switch (get_ucall(vcpu, &uc)) {
		case UCALL_SYNC:
			TEST_ASSERT(uc.args[1] == 0,
				"Unexpected sync ucall, got %lx",
				(ulong)uc.args[1]);
			sem_post(&vcpu_ready);
			continue;
		case UCALL_NONE:
			if (run->exit_reason == KVM_EXIT_MMIO)
				check_mmio_access(data, run);
			else
				goto done;
			break;
		case UCALL_ABORT:
			REPORT_GUEST_ASSERT(uc);
			break;
		case UCALL_DONE:
			goto done;
		default:
			TEST_FAIL("Unknown ucall %lu", uc.cmd);
		}
	}

done:
	return NULL;
}

static void wait_for_vcpu(void)
{
	struct timespec ts;

	TEST_ASSERT(!clock_gettime(CLOCK_REALTIME, &ts),
		    "clock_gettime() failed: %d\n", errno);

	ts.tv_sec += 2;
	TEST_ASSERT(!sem_timedwait(&vcpu_ready, &ts),
		    "sem_timedwait() failed: %d\n", errno);
}

static void *vm_gpa2hva(struct vm_data *data, uint64_t gpa, uint64_t *rempages)
{
	uint64_t gpage, pgoffs;
	uint32_t slot, slotoffs;
	void *base;
	uint32_t guest_page_size = data->vm->page_size;

	TEST_ASSERT(gpa >= MEM_GPA, "Too low gpa to translate");
	TEST_ASSERT(gpa < MEM_GPA + data->npages * guest_page_size,
		    "Too high gpa to translate");
	gpa -= MEM_GPA;

	gpage = gpa / guest_page_size;
	pgoffs = gpa % guest_page_size;
	slot = min(gpage / data->pages_per_slot, (uint64_t)data->nslots - 1);
	slotoffs = gpage - (slot * data->pages_per_slot);

	if (rempages) {
		uint64_t slotpages;

		if (slot == data->nslots - 1)
			slotpages = data->npages - slot * data->pages_per_slot;
		else
			slotpages = data->pages_per_slot;

		TEST_ASSERT(!pgoffs,
			    "Asking for remaining pages in slot but gpa not page aligned");
		*rempages = slotpages - slotoffs;
	}

	base = data->hva_slots[slot];
	return (uint8_t *)base + slotoffs * guest_page_size + pgoffs;
}

static uint64_t vm_slot2gpa(struct vm_data *data, uint32_t slot)
{
	uint32_t guest_page_size = data->vm->page_size;

	TEST_ASSERT(slot < data->nslots, "Too high slot number");

	return MEM_GPA + slot * data->pages_per_slot * guest_page_size;
}

static struct vm_data *alloc_vm(void)
{
	struct vm_data *data;

	data = malloc(sizeof(*data));
	TEST_ASSERT(data, "malloc(vmdata) failed");

	data->vm = NULL;
	data->vcpu = NULL;
	data->hva_slots = NULL;

	return data;
}

static bool check_slot_pages(uint32_t host_page_size, uint32_t guest_page_size,
			     uint64_t pages_per_slot, uint64_t rempages)
{
	if (!pages_per_slot)
		return false;

	if ((pages_per_slot * guest_page_size) % host_page_size)
		return false;

	if ((rempages * guest_page_size) % host_page_size)
		return false;

	return true;
}


static uint64_t get_max_slots(struct vm_data *data, uint32_t host_page_size)
{
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t mempages, pages_per_slot, rempages;
	uint64_t slots;

	mempages = data->npages;
	slots = data->nslots;
	while (--slots > 1) {
		pages_per_slot = mempages / slots;
		if (!pages_per_slot)
			continue;

		rempages = mempages % pages_per_slot;
		if (check_slot_pages(host_page_size, guest_page_size,
				     pages_per_slot, rempages))
			return slots + 1;	/* slot 0 is reserved */
	}

	return 0;
}

static bool prepare_vm(struct vm_data *data, int nslots, uint64_t *maxslots,
		       void *guest_code, uint64_t mem_size,
		       struct timespec *slot_runtime)
{
	uint64_t mempages, rempages;
	uint64_t guest_addr;
	uint32_t slot, host_page_size, guest_page_size;
	struct timespec tstart;
	struct sync_area *sync;

	host_page_size = getpagesize();
	guest_page_size = vm_guest_mode_params[VM_MODE_DEFAULT].page_size;
	mempages = mem_size / guest_page_size;

	data->vm = __vm_create_with_one_vcpu(&data->vcpu, mempages, guest_code);
	TEST_ASSERT(data->vm->page_size == guest_page_size, "Invalid VM page size");

	data->npages = mempages;
	TEST_ASSERT(data->npages > 1, "Can't test without any memory");
	data->nslots = nslots;
	data->pages_per_slot = data->npages / data->nslots;
	rempages = data->npages % data->nslots;
	if (!check_slot_pages(host_page_size, guest_page_size,
			      data->pages_per_slot, rempages)) {
		*maxslots = get_max_slots(data, host_page_size);
		return false;
	}

	data->hva_slots = malloc(sizeof(*data->hva_slots) * data->nslots);
	TEST_ASSERT(data->hva_slots, "malloc() fail");

	pr_info_v("Adding slots 1..%i, each slot with %"PRIu64" pages + %"PRIu64" extra pages last\n",
		data->nslots, data->pages_per_slot, rempages);

	clock_gettime(CLOCK_MONOTONIC, &tstart);
	for (slot = 1, guest_addr = MEM_GPA; slot <= data->nslots; slot++) {
		uint64_t npages;

		npages = data->pages_per_slot;
		if (slot == data->nslots)
			npages += rempages;

		vm_userspace_mem_region_add(data->vm, VM_MEM_SRC_ANONYMOUS,
					    guest_addr, slot, npages,
					    0);
		guest_addr += npages * guest_page_size;
	}
	*slot_runtime = timespec_elapsed(tstart);

	for (slot = 1, guest_addr = MEM_GPA; slot <= data->nslots; slot++) {
		uint64_t npages;
		uint64_t gpa;

		npages = data->pages_per_slot;
		if (slot == data->nslots)
			npages += rempages;

		gpa = vm_phy_pages_alloc(data->vm, npages, guest_addr, slot);
		TEST_ASSERT(gpa == guest_addr,
			    "vm_phy_pages_alloc() failed\n");

		data->hva_slots[slot - 1] = addr_gpa2hva(data->vm, guest_addr);
		memset(data->hva_slots[slot - 1], 0, npages * guest_page_size);

		guest_addr += npages * guest_page_size;
	}

	virt_map(data->vm, MEM_GPA, MEM_GPA, data->npages);

	sync = (typeof(sync))vm_gpa2hva(data, MEM_SYNC_GPA, NULL);
	sync->guest_page_size = data->vm->page_size;
	atomic_init(&sync->start_flag, false);
	atomic_init(&sync->exit_flag, false);
	atomic_init(&sync->sync_flag, false);

	data->mmio_ok = false;

	return true;
}

static void launch_vm(struct vm_data *data)
{
	pr_info_v("Launching the test VM\n");

	pthread_create(&data->vcpu_thread, NULL, vcpu_worker, data);

	/* Ensure the guest thread is spun up. */
	wait_for_vcpu();
}

static void free_vm(struct vm_data *data)
{
	kvm_vm_free(data->vm);
	free(data->hva_slots);
	free(data);
}

static void wait_guest_exit(struct vm_data *data)
{
	pthread_join(data->vcpu_thread, NULL);
}

static void let_guest_run(struct sync_area *sync)
{
	atomic_store_explicit(&sync->start_flag, true, memory_order_release);
}

static void guest_spin_until_start(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;

	while (!atomic_load_explicit(&sync->start_flag, memory_order_acquire))
		;
}

static void make_guest_exit(struct sync_area *sync)
{
	atomic_store_explicit(&sync->exit_flag, true, memory_order_release);
}

static bool _guest_should_exit(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;

	return atomic_load_explicit(&sync->exit_flag, memory_order_acquire);
}

#define guest_should_exit() unlikely(_guest_should_exit())

/*
 * noinline so we can easily see how much time the host spends waiting
 * for the guest.
 * For the same reason use alarm() instead of polling clock_gettime()
 * to implement a wait timeout.
 */
static noinline void host_perform_sync(struct sync_area *sync)
{
	alarm(2);

	atomic_store_explicit(&sync->sync_flag, true, memory_order_release);
	while (atomic_load_explicit(&sync->sync_flag, memory_order_acquire))
		;

	alarm(0);
}

static bool guest_perform_sync(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;
	bool expected;

	do {
		if (guest_should_exit())
			return false;

		expected = true;
	} while (!atomic_compare_exchange_weak_explicit(&sync->sync_flag,
							&expected, false,
							memory_order_acq_rel,
							memory_order_relaxed));

	return true;
}

static void guest_code_test_memslot_move(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;
	uint32_t page_size = (typeof(page_size))READ_ONCE(sync->guest_page_size);
	uintptr_t base = (typeof(base))READ_ONCE(sync->move_area_ptr);

	GUEST_SYNC(0);

	guest_spin_until_start();

	while (!guest_should_exit()) {
		uintptr_t ptr;

		for (ptr = base; ptr < base + MEM_TEST_MOVE_SIZE;
		     ptr += page_size)
			*(uint64_t *)ptr = MEM_TEST_VAL_1;

		/*
		 * No host sync here since the MMIO exits are so expensive
		 * that the host would spend most of its time waiting for
		 * the guest and so instead of measuring memslot move
		 * performance we would measure the performance and
		 * likelihood of MMIO exits
		 */
	}

	GUEST_DONE();
}

static void guest_code_test_memslot_map(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;
	uint32_t page_size = (typeof(page_size))READ_ONCE(sync->guest_page_size);

	GUEST_SYNC(0);

	guest_spin_until_start();

	while (1) {
		uintptr_t ptr;

		for (ptr = MEM_TEST_GPA;
		     ptr < MEM_TEST_GPA + MEM_TEST_MAP_SIZE / 2;
		     ptr += page_size)
			*(uint64_t *)ptr = MEM_TEST_VAL_1;

		if (!guest_perform_sync())
			break;

		for (ptr = MEM_TEST_GPA + MEM_TEST_MAP_SIZE / 2;
		     ptr < MEM_TEST_GPA + MEM_TEST_MAP_SIZE;
		     ptr += page_size)
			*(uint64_t *)ptr = MEM_TEST_VAL_2;

		if (!guest_perform_sync())
			break;
	}

	GUEST_DONE();
}

static void guest_code_test_memslot_unmap(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;

	GUEST_SYNC(0);

	guest_spin_until_start();

	while (1) {
		uintptr_t ptr = MEM_TEST_GPA;

		/*
		 * We can afford to access (map) just a small number of pages
		 * per host sync as otherwise the host will spend
		 * a significant amount of its time waiting for the guest
		 * (instead of doing unmap operations), so this will
		 * effectively turn this test into a map performance test.
		 *
		 * Just access a single page to be on the safe side.
		 */
		*(uint64_t *)ptr = MEM_TEST_VAL_1;

		if (!guest_perform_sync())
			break;

		ptr += MEM_TEST_UNMAP_SIZE / 2;
		*(uint64_t *)ptr = MEM_TEST_VAL_2;

		if (!guest_perform_sync())
			break;
	}

	GUEST_DONE();
}

static void guest_code_test_memslot_rw(void)
{
	struct sync_area *sync = (typeof(sync))MEM_SYNC_GPA;
	uint32_t page_size = (typeof(page_size))READ_ONCE(sync->guest_page_size);

	GUEST_SYNC(0);

	guest_spin_until_start();

	while (1) {
		uintptr_t ptr;

		for (ptr = MEM_TEST_GPA;
		     ptr < MEM_TEST_GPA + MEM_TEST_SIZE; ptr += page_size)
			*(uint64_t *)ptr = MEM_TEST_VAL_1;

		if (!guest_perform_sync())
			break;

		for (ptr = MEM_TEST_GPA + page_size / 2;
		     ptr < MEM_TEST_GPA + MEM_TEST_SIZE; ptr += page_size) {
			uint64_t val = *(uint64_t *)ptr;

			GUEST_ASSERT_EQ(val, MEM_TEST_VAL_2);
			*(uint64_t *)ptr = 0;
		}

		if (!guest_perform_sync())
			break;
	}

	GUEST_DONE();
}

static bool test_memslot_move_prepare(struct vm_data *data,
				      struct sync_area *sync,
				      uint64_t *maxslots, bool isactive)
{
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t movesrcgpa, movetestgpa;

	movesrcgpa = vm_slot2gpa(data, data->nslots - 1);

	if (isactive) {
		uint64_t lastpages;

		vm_gpa2hva(data, movesrcgpa, &lastpages);
		if (lastpages * guest_page_size < MEM_TEST_MOVE_SIZE / 2) {
			*maxslots = 0;
			return false;
		}
	}

	movetestgpa = movesrcgpa - (MEM_TEST_MOVE_SIZE / (isactive ? 2 : 1));
	sync->move_area_ptr = (void *)movetestgpa;

	if (isactive) {
		data->mmio_ok = true;
		data->mmio_gpa_min = movesrcgpa;
		data->mmio_gpa_max = movesrcgpa + MEM_TEST_MOVE_SIZE / 2 - 1;
	}

	return true;
}

static bool test_memslot_move_prepare_active(struct vm_data *data,
					     struct sync_area *sync,
					     uint64_t *maxslots)
{
	return test_memslot_move_prepare(data, sync, maxslots, true);
}

static bool test_memslot_move_prepare_inactive(struct vm_data *data,
					       struct sync_area *sync,
					       uint64_t *maxslots)
{
	return test_memslot_move_prepare(data, sync, maxslots, false);
}

static void test_memslot_move_loop(struct vm_data *data, struct sync_area *sync)
{
	uint64_t movesrcgpa;

	movesrcgpa = vm_slot2gpa(data, data->nslots - 1);
	vm_mem_region_move(data->vm, data->nslots - 1 + 1,
			   MEM_TEST_MOVE_GPA_DEST);
	vm_mem_region_move(data->vm, data->nslots - 1 + 1, movesrcgpa);
}

static void test_memslot_do_unmap(struct vm_data *data,
				  uint64_t offsp, uint64_t count)
{
	uint64_t gpa, ctr;
	uint32_t guest_page_size = data->vm->page_size;

	for (gpa = MEM_TEST_GPA + offsp * guest_page_size, ctr = 0; ctr < count; ) {
		uint64_t npages;
		void *hva;
		int ret;

		hva = vm_gpa2hva(data, gpa, &npages);
		TEST_ASSERT(npages, "Empty memory slot at gptr 0x%"PRIx64, gpa);
		npages = min(npages, count - ctr);
		ret = madvise(hva, npages * guest_page_size, MADV_DONTNEED);
		TEST_ASSERT(!ret,
			    "madvise(%p, MADV_DONTNEED) on VM memory should not fail for gptr 0x%"PRIx64,
			    hva, gpa);
		ctr += npages;
		gpa += npages * guest_page_size;
	}
	TEST_ASSERT(ctr == count,
		    "madvise(MADV_DONTNEED) should exactly cover all of the requested area");
}

static void test_memslot_map_unmap_check(struct vm_data *data,
					 uint64_t offsp, uint64_t valexp)
{
	uint64_t gpa;
	uint64_t *val;
	uint32_t guest_page_size = data->vm->page_size;

	if (!map_unmap_verify)
		return;

	gpa = MEM_TEST_GPA + offsp * guest_page_size;
	val = (typeof(val))vm_gpa2hva(data, gpa, NULL);
	TEST_ASSERT(*val == valexp,
		    "Guest written values should read back correctly before unmap (%"PRIu64" vs %"PRIu64" @ %"PRIx64")",
		    *val, valexp, gpa);
	*val = 0;
}

static void test_memslot_map_loop(struct vm_data *data, struct sync_area *sync)
{
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t guest_pages = MEM_TEST_MAP_SIZE / guest_page_size;

	/*
	 * Unmap the second half of the test area while guest writes to (maps)
	 * the first half.
	 */
	test_memslot_do_unmap(data, guest_pages / 2, guest_pages / 2);

	/*
	 * Wait for the guest to finish writing the first half of the test
	 * area, verify the written value on the first and the last page of
	 * this area and then unmap it.
	 * Meanwhile, the guest is writing to (mapping) the second half of
	 * the test area.
	 */
	host_perform_sync(sync);
	test_memslot_map_unmap_check(data, 0, MEM_TEST_VAL_1);
	test_memslot_map_unmap_check(data, guest_pages / 2 - 1, MEM_TEST_VAL_1);
	test_memslot_do_unmap(data, 0, guest_pages / 2);


	/*
	 * Wait for the guest to finish writing the second half of the test
	 * area and verify the written value on the first and the last page
	 * of this area.
	 * The area will be unmapped at the beginning of the next loop
	 * iteration.
	 * Meanwhile, the guest is writing to (mapping) the first half of
	 * the test area.
	 */
	host_perform_sync(sync);
	test_memslot_map_unmap_check(data, guest_pages / 2, MEM_TEST_VAL_2);
	test_memslot_map_unmap_check(data, guest_pages - 1, MEM_TEST_VAL_2);
}

static void test_memslot_unmap_loop_common(struct vm_data *data,
					   struct sync_area *sync,
					   uint64_t chunk)
{
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t guest_pages = MEM_TEST_UNMAP_SIZE / guest_page_size;
	uint64_t ctr;

	/*
	 * Wait for the guest to finish mapping page(s) in the first half
	 * of the test area, verify the written value and then perform unmap
	 * of this area.
	 * Meanwhile, the guest is writing to (mapping) page(s) in the second
	 * half of the test area.
	 */
	host_perform_sync(sync);
	test_memslot_map_unmap_check(data, 0, MEM_TEST_VAL_1);
	for (ctr = 0; ctr < guest_pages / 2; ctr += chunk)
		test_memslot_do_unmap(data, ctr, chunk);

	/* Likewise, but for the opposite host / guest areas */
	host_perform_sync(sync);
	test_memslot_map_unmap_check(data, guest_pages / 2, MEM_TEST_VAL_2);
	for (ctr = guest_pages / 2; ctr < guest_pages; ctr += chunk)
		test_memslot_do_unmap(data, ctr, chunk);
}

static void test_memslot_unmap_loop(struct vm_data *data,
				    struct sync_area *sync)
{
	uint32_t host_page_size = getpagesize();
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t guest_chunk_pages = guest_page_size >= host_page_size ?
					1 : host_page_size / guest_page_size;

	test_memslot_unmap_loop_common(data, sync, guest_chunk_pages);
}

static void test_memslot_unmap_loop_chunked(struct vm_data *data,
					    struct sync_area *sync)
{
	uint32_t guest_page_size = data->vm->page_size;
	uint64_t guest_chunk_pages = MEM_TEST_UNMAP_CHUNK_SIZE / guest_page_size;

	test_memslot_unmap_loop_common(data, sync, guest_chunk_pages);
}

static void test_memslot_rw_loop(struct vm_data *data, struct sync_area *sync)
{
	uint64_t gptr;
	uint32_t guest_page_size = data->vm->page_size;

	for (gptr = MEM_TEST_GPA + guest_page_size / 2;
	     gptr < MEM_TEST_GPA + MEM_TEST_SIZE; gptr += guest_page_size)
		*(uint64_t *)vm_gpa2hva(data, gptr, NULL) = MEM_TEST_VAL_2;

	host_perform_sync(sync);

	for (gptr = MEM_TEST_GPA;
	     gptr < MEM_TEST_GPA + MEM_TEST_SIZE; gptr += guest_page_size) {
		uint64_t *vptr = (typeof(vptr))vm_gpa2hva(data, gptr, NULL);
		uint64_t val = *vptr;

		TEST_ASSERT(val == MEM_TEST_VAL_1,
			    "Guest written values should read back correctly (is %"PRIu64" @ %"PRIx64")",
			    val, gptr);
		*vptr = 0;
	}

	host_perform_sync(sync);
}

struct test_data {
	const char *name;
	uint64_t mem_size;
	void (*guest_code)(void);
	bool (*prepare)(struct vm_data *data, struct sync_area *sync,
			uint64_t *maxslots);
	void (*loop)(struct vm_data *data, struct sync_area *sync);
};

static bool test_execute(int nslots, uint64_t *maxslots,
			 unsigned int maxtime,
			 const struct test_data *tdata,
			 uint64_t *nloops,
			 struct timespec *slot_runtime,
			 struct timespec *guest_runtime)
{
	uint64_t mem_size = tdata->mem_size ? : MEM_SIZE;
	struct vm_data *data;
	struct sync_area *sync;
	struct timespec tstart;
	bool ret = true;

	data = alloc_vm();
	if (!prepare_vm(data, nslots, maxslots, tdata->guest_code,
			mem_size, slot_runtime)) {
		ret = false;
		goto exit_free;
	}

	sync = (typeof(sync))vm_gpa2hva(data, MEM_SYNC_GPA, NULL);
	if (tdata->prepare &&
	    !tdata->prepare(data, sync, maxslots)) {
		ret = false;
		goto exit_free;
	}

	launch_vm(data);

	clock_gettime(CLOCK_MONOTONIC, &tstart);
	let_guest_run(sync);

	while (1) {
		*guest_runtime = timespec_elapsed(tstart);
		if (guest_runtime->tv_sec >= maxtime)
			break;

		tdata->loop(data, sync);

		(*nloops)++;
	}

	make_guest_exit(sync);
	wait_guest_exit(data);

exit_free:
	free_vm(data);

	return ret;
}

static const struct test_data tests[] = {
	{
		.name = "map",
		.mem_size = MEM_SIZE_MAP,
		.guest_code = guest_code_test_memslot_map,
		.loop = test_memslot_map_loop,
	},
	{
		.name = "unmap",
		.mem_size = MEM_TEST_UNMAP_SIZE + MEM_EXTRA_SIZE,
		.guest_code = guest_code_test_memslot_unmap,
		.loop = test_memslot_unmap_loop,
	},
	{
		.name = "unmap chunked",
		.mem_size = MEM_TEST_UNMAP_SIZE + MEM_EXTRA_SIZE,
		.guest_code = guest_code_test_memslot_unmap,
		.loop = test_memslot_unmap_loop_chunked,
	},
	{
		.name = "move active area",
		.guest_code = guest_code_test_memslot_move,
		.prepare = test_memslot_move_prepare_active,
		.loop = test_memslot_move_loop,
	},
	{
		.name = "move inactive area",
		.guest_code = guest_code_test_memslot_move,
		.prepare = test_memslot_move_prepare_inactive,
		.loop = test_memslot_move_loop,
	},
	{
		.name = "RW",
		.guest_code = guest_code_test_memslot_rw,
		.loop = test_memslot_rw_loop
	},
};

#define NTESTS ARRAY_SIZE(tests)

struct test_args {
	int tfirst;
	int tlast;
	int nslots;
	int seconds;
	int runs;
};

static void help(char *name, struct test_args *targs)
{
	int ctr;

	pr_info("usage: %s [-h] [-v] [-d] [-s slots] [-f first_test] [-e last_test] [-l test_length] [-r run_count]\n",
		name);
	pr_info(" -h: print this help screen.\n");
	pr_info(" -v: enable verbose mode (not for benchmarking).\n");
	pr_info(" -d: enable extra debug checks.\n");
	pr_info(" -s: specify memslot count cap (-1 means no cap; currently: %i)\n",
		targs->nslots);
	pr_info(" -f: specify the first test to run (currently: %i; max %zu)\n",
		targs->tfirst, NTESTS - 1);
	pr_info(" -e: specify the last test to run (currently: %i; max %zu)\n",
		targs->tlast, NTESTS - 1);
	pr_info(" -l: specify the test length in seconds (currently: %i)\n",
		targs->seconds);
	pr_info(" -r: specify the number of runs per test (currently: %i)\n",
		targs->runs);

	pr_info("\nAvailable tests:\n");
	for (ctr = 0; ctr < NTESTS; ctr++)
		pr_info("%d: %s\n", ctr, tests[ctr].name);
}

static bool check_memory_sizes(void)
{
	uint32_t host_page_size = getpagesize();
	uint32_t guest_page_size = vm_guest_mode_params[VM_MODE_DEFAULT].page_size;

	if (host_page_size > SZ_64K || guest_page_size > SZ_64K) {
		pr_info("Unsupported page size on host (0x%x) or guest (0x%x)\n",
			host_page_size, guest_page_size);
		return false;
	}

	if (MEM_SIZE % guest_page_size ||
	    MEM_TEST_SIZE % guest_page_size) {
		pr_info("invalid MEM_SIZE or MEM_TEST_SIZE\n");
		return false;
	}

	if (MEM_SIZE_MAP % guest_page_size		||
	    MEM_TEST_MAP_SIZE % guest_page_size		||
	    (MEM_TEST_MAP_SIZE / guest_page_size) <= 2	||
	    (MEM_TEST_MAP_SIZE / guest_page_size) % 2) {
		pr_info("invalid MEM_SIZE_MAP or MEM_TEST_MAP_SIZE\n");
		return false;
	}

	if (MEM_TEST_UNMAP_SIZE > MEM_TEST_SIZE		||
	    MEM_TEST_UNMAP_SIZE % guest_page_size	||
	    (MEM_TEST_UNMAP_SIZE / guest_page_size) %
	    (2 * MEM_TEST_UNMAP_CHUNK_SIZE / guest_page_size)) {
		pr_info("invalid MEM_TEST_UNMAP_SIZE or MEM_TEST_UNMAP_CHUNK_SIZE\n");
		return false;
	}

	return true;
}

static bool parse_args(int argc, char *argv[],
		       struct test_args *targs)
{
	uint32_t max_mem_slots;
	int opt;

	while ((opt = getopt(argc, argv, "hvds:f:e:l:r:")) != -1) {
		switch (opt) {
		case 'h':
		default:
			help(argv[0], targs);
			return false;
		case 'v':
			verbose = true;
			break;
		case 'd':
			map_unmap_verify = true;
			break;
		case 's':
			targs->nslots = atoi_paranoid(optarg);
			if (targs->nslots <= 1 && targs->nslots != -1) {
				pr_info("Slot count cap must be larger than 1 or -1 for no cap\n");
				return false;
			}
			break;
		case 'f':
			targs->tfirst = atoi_non_negative("First test", optarg);
			break;
		case 'e':
			targs->tlast = atoi_non_negative("Last test", optarg);
			if (targs->tlast >= NTESTS) {
				pr_info("Last test to run has to be non-negative and less than %zu\n",
					NTESTS);
				return false;
			}
			break;
		case 'l':
			targs->seconds = atoi_non_negative("Test length", optarg);
			break;
		case 'r':
			targs->runs = atoi_positive("Runs per test", optarg);
			break;
		}
	}

	if (optind < argc) {
		help(argv[0], targs);
		return false;
	}

	if (targs->tfirst > targs->tlast) {
		pr_info("First test to run cannot be greater than the last test to run\n");
		return false;
	}

	max_mem_slots = kvm_check_cap(KVM_CAP_NR_MEMSLOTS);
	if (max_mem_slots <= 1) {
		pr_info("KVM_CAP_NR_MEMSLOTS should be greater than 1\n");
		return false;
	}

	/* Memory slot 0 is reserved */
	if (targs->nslots == -1)
		targs->nslots = max_mem_slots - 1;
	else
		targs->nslots = min_t(int, targs->nslots, max_mem_slots) - 1;

	pr_info_v("Allowed Number of memory slots: %"PRIu32"\n",
		  targs->nslots + 1);

	return true;
}

struct test_result {
	struct timespec slot_runtime, guest_runtime, iter_runtime;
	int64_t slottimens, runtimens;
	uint64_t nloops;
};

static bool test_loop(const struct test_data *data,
		      const struct test_args *targs,
		      struct test_result *rbestslottime,
		      struct test_result *rbestruntime)
{
	uint64_t maxslots;
	struct test_result result = {};

	if (!test_execute(targs->nslots, &maxslots, targs->seconds, data,
			  &result.nloops,
			  &result.slot_runtime, &result.guest_runtime)) {
		if (maxslots)
			pr_info("Memslot count too high for this test, decrease the cap (max is %"PRIu64")\n",
				maxslots);
		else
			pr_info("Memslot count may be too high for this test, try adjusting the cap\n");

		return false;
	}

	pr_info("Test took %ld.%.9lds for slot setup + %ld.%.9lds all iterations\n",
		result.slot_runtime.tv_sec, result.slot_runtime.tv_nsec,
		result.guest_runtime.tv_sec, result.guest_runtime.tv_nsec);
	if (!result.nloops) {
		pr_info("No full loops done - too short test time or system too loaded?\n");
		return true;
	}

	result.iter_runtime = timespec_div(result.guest_runtime,
					   result.nloops);
	pr_info("Done %"PRIu64" iterations, avg %ld.%.9lds each\n",
		result.nloops,
		result.iter_runtime.tv_sec,
		result.iter_runtime.tv_nsec);
	result.slottimens = timespec_to_ns(result.slot_runtime);
	result.runtimens = timespec_to_ns(result.iter_runtime);

	/*
	 * Only rank the slot setup time for tests using the whole test memory
	 * area so they are comparable
	 */
	if (!data->mem_size &&
	    (!rbestslottime->slottimens ||
	     result.slottimens < rbestslottime->slottimens))
		*rbestslottime = result;
	if (!rbestruntime->runtimens ||
	    result.runtimens < rbestruntime->runtimens)
		*rbestruntime = result;

	return true;
}

int main(int argc, char *argv[])
{
	struct test_args targs = {
		.tfirst = 0,
		.tlast = NTESTS - 1,
		.nslots = -1,
		.seconds = 5,
		.runs = 1,
	};
	struct test_result rbestslottime = {};
	int tctr;

	if (!check_memory_sizes())
		return -1;

	if (!parse_args(argc, argv, &targs))
		return -1;

	for (tctr = targs.tfirst; tctr <= targs.tlast; tctr++) {
		const struct test_data *data = &tests[tctr];
		unsigned int runctr;
		struct test_result rbestruntime = {};

		if (tctr > targs.tfirst)
			pr_info("\n");

		pr_info("Testing %s performance with %i runs, %d seconds each\n",
			data->name, targs.runs, targs.seconds);

		for (runctr = 0; runctr < targs.runs; runctr++)
			if (!test_loop(data, &targs,
				       &rbestslottime, &rbestruntime))
				break;

		if (rbestruntime.runtimens)
			pr_info("Best runtime result was %ld.%.9lds per iteration (with %"PRIu64" iterations)\n",
				rbestruntime.iter_runtime.tv_sec,
				rbestruntime.iter_runtime.tv_nsec,
				rbestruntime.nloops);
	}

	if (rbestslottime.slottimens)
		pr_info("Best slot setup time for the whole test area was %ld.%.9lds\n",
			rbestslottime.slot_runtime.tv_sec,
			rbestslottime.slot_runtime.tv_nsec);

	return 0;
}