xref: /openbmc/linux/tools/testing/selftests/kvm/dirty_log_test.c (revision 4f727ecefefbd180de10e25b3e74c03dce3f1e75)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * KVM dirty page logging test
4  *
5  * Copyright (C) 2018, Red Hat, Inc.
6  */
7 
8 #define _GNU_SOURCE /* for program_invocation_name */
9 
10 #include <stdio.h>
11 #include <stdlib.h>
12 #include <unistd.h>
13 #include <time.h>
14 #include <pthread.h>
15 #include <linux/bitmap.h>
16 #include <linux/bitops.h>
17 
18 #include "test_util.h"
19 #include "kvm_util.h"
20 #include "processor.h"
21 
22 #define DEBUG printf
23 
24 #define VCPU_ID				1
25 
26 /* The memory slot index to track dirty pages */
27 #define TEST_MEM_SLOT_INDEX		1
28 
29 /* Default guest test memory offset, 1G */
30 #define DEFAULT_GUEST_TEST_MEM		0x40000000
31 
32 /* How many pages to dirty for each guest loop */
33 #define TEST_PAGES_PER_LOOP		1024
34 
35 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
36 #define TEST_HOST_LOOP_N		32UL
37 
38 /* Interval for each host loop (ms) */
39 #define TEST_HOST_LOOP_INTERVAL		10UL
40 
41 /*
42  * Guest/Host shared variables. Ensure addr_gva2hva() and/or
43  * sync_global_to/from_guest() are used when accessing from
44  * the host. READ/WRITE_ONCE() should also be used with anything
45  * that may change.
46  */
47 static uint64_t host_page_size;
48 static uint64_t guest_page_size;
49 static uint64_t guest_num_pages;
50 static uint64_t random_array[TEST_PAGES_PER_LOOP];
51 static uint64_t iteration;
52 
53 /*
54  * Guest physical memory offset of the testing memory slot.
55  * This will be set to the topmost valid physical address minus
56  * the test memory size.
57  */
58 static uint64_t guest_test_phys_mem;
59 
60 /*
61  * Guest virtual memory offset of the testing memory slot.
62  * Must not conflict with identity mapped test code.
63  */
64 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
65 
66 /*
67  * Continuously write to the first 8 bytes of a random pages within
68  * the testing memory region.
69  */
70 static void guest_code(void)
71 {
72 	int i;
73 
74 	while (true) {
75 		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
76 			uint64_t addr = guest_test_virt_mem;
77 			addr += (READ_ONCE(random_array[i]) % guest_num_pages)
78 				* guest_page_size;
79 			addr &= ~(host_page_size - 1);
80 			*(uint64_t *)addr = READ_ONCE(iteration);
81 		}
82 
83 		/* Tell the host that we need more random numbers */
84 		GUEST_SYNC(1);
85 	}
86 }
87 
88 /* Host variables */
89 static bool host_quit;
90 
91 /* Points to the test VM memory region on which we track dirty logs */
92 static void *host_test_mem;
93 static uint64_t host_num_pages;
94 
95 /* For statistics only */
96 static uint64_t host_dirty_count;
97 static uint64_t host_clear_count;
98 static uint64_t host_track_next_count;
99 
100 /*
101  * We use this bitmap to track some pages that should have its dirty
102  * bit set in the _next_ iteration.  For example, if we detected the
103  * page value changed to current iteration but at the same time the
104  * page bit is cleared in the latest bitmap, then the system must
105  * report that write in the next get dirty log call.
106  */
107 static unsigned long *host_bmap_track;
108 
109 static void generate_random_array(uint64_t *guest_array, uint64_t size)
110 {
111 	uint64_t i;
112 
113 	for (i = 0; i < size; i++)
114 		guest_array[i] = random();
115 }
116 
117 static void *vcpu_worker(void *data)
118 {
119 	int ret;
120 	struct kvm_vm *vm = data;
121 	uint64_t *guest_array;
122 	uint64_t pages_count = 0;
123 	struct kvm_run *run;
124 	struct ucall uc;
125 
126 	run = vcpu_state(vm, VCPU_ID);
127 
128 	guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
129 	generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
130 
131 	while (!READ_ONCE(host_quit)) {
132 		/* Let the guest dirty the random pages */
133 		ret = _vcpu_run(vm, VCPU_ID);
134 		TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
135 		if (get_ucall(vm, VCPU_ID, &uc) == UCALL_SYNC) {
136 			pages_count += TEST_PAGES_PER_LOOP;
137 			generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
138 		} else {
139 			TEST_ASSERT(false,
140 				    "Invalid guest sync status: "
141 				    "exit_reason=%s\n",
142 				    exit_reason_str(run->exit_reason));
143 		}
144 	}
145 
146 	DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
147 
148 	return NULL;
149 }
150 
151 static void vm_dirty_log_verify(unsigned long *bmap)
152 {
153 	uint64_t page;
154 	uint64_t *value_ptr;
155 	uint64_t step = host_page_size >= guest_page_size ? 1 :
156 				guest_page_size / host_page_size;
157 
158 	for (page = 0; page < host_num_pages; page += step) {
159 		value_ptr = host_test_mem + page * host_page_size;
160 
161 		/* If this is a special page that we were tracking... */
162 		if (test_and_clear_bit(page, host_bmap_track)) {
163 			host_track_next_count++;
164 			TEST_ASSERT(test_bit(page, bmap),
165 				    "Page %"PRIu64" should have its dirty bit "
166 				    "set in this iteration but it is missing",
167 				    page);
168 		}
169 
170 		if (test_bit(page, bmap)) {
171 			host_dirty_count++;
172 			/*
173 			 * If the bit is set, the value written onto
174 			 * the corresponding page should be either the
175 			 * previous iteration number or the current one.
176 			 */
177 			TEST_ASSERT(*value_ptr == iteration ||
178 				    *value_ptr == iteration - 1,
179 				    "Set page %"PRIu64" value %"PRIu64
180 				    " incorrect (iteration=%"PRIu64")",
181 				    page, *value_ptr, iteration);
182 		} else {
183 			host_clear_count++;
184 			/*
185 			 * If cleared, the value written can be any
186 			 * value smaller or equals to the iteration
187 			 * number.  Note that the value can be exactly
188 			 * (iteration-1) if that write can happen
189 			 * like this:
190 			 *
191 			 * (1) increase loop count to "iteration-1"
192 			 * (2) write to page P happens (with value
193 			 *     "iteration-1")
194 			 * (3) get dirty log for "iteration-1"; we'll
195 			 *     see that page P bit is set (dirtied),
196 			 *     and not set the bit in host_bmap_track
197 			 * (4) increase loop count to "iteration"
198 			 *     (which is current iteration)
199 			 * (5) get dirty log for current iteration,
200 			 *     we'll see that page P is cleared, with
201 			 *     value "iteration-1".
202 			 */
203 			TEST_ASSERT(*value_ptr <= iteration,
204 				    "Clear page %"PRIu64" value %"PRIu64
205 				    " incorrect (iteration=%"PRIu64")",
206 				    page, *value_ptr, iteration);
207 			if (*value_ptr == iteration) {
208 				/*
209 				 * This page is _just_ modified; it
210 				 * should report its dirtyness in the
211 				 * next run
212 				 */
213 				set_bit(page, host_bmap_track);
214 			}
215 		}
216 	}
217 }
218 
219 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
220 				uint64_t extra_mem_pages, void *guest_code,
221 				unsigned long type)
222 {
223 	struct kvm_vm *vm;
224 	uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
225 
226 	vm = _vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages,
227 			O_RDWR, type);
228 	kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
229 #ifdef __x86_64__
230 	vm_create_irqchip(vm);
231 #endif
232 	vm_vcpu_add_default(vm, vcpuid, guest_code);
233 	return vm;
234 }
235 
236 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
237 		     unsigned long interval, uint64_t phys_offset)
238 {
239 	unsigned int guest_pa_bits, guest_page_shift;
240 	pthread_t vcpu_thread;
241 	struct kvm_vm *vm;
242 	uint64_t max_gfn;
243 	unsigned long *bmap;
244 	unsigned long type = 0;
245 
246 	switch (mode) {
247 	case VM_MODE_P52V48_4K:
248 		guest_pa_bits = 52;
249 		guest_page_shift = 12;
250 		break;
251 	case VM_MODE_P52V48_64K:
252 		guest_pa_bits = 52;
253 		guest_page_shift = 16;
254 		break;
255 	case VM_MODE_P48V48_4K:
256 		guest_pa_bits = 48;
257 		guest_page_shift = 12;
258 		break;
259 	case VM_MODE_P48V48_64K:
260 		guest_pa_bits = 48;
261 		guest_page_shift = 16;
262 		break;
263 	case VM_MODE_P40V48_4K:
264 		guest_pa_bits = 40;
265 		guest_page_shift = 12;
266 		break;
267 	case VM_MODE_P40V48_64K:
268 		guest_pa_bits = 40;
269 		guest_page_shift = 16;
270 		break;
271 	default:
272 		TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", mode);
273 	}
274 
275 	DEBUG("Testing guest mode: %s\n", vm_guest_mode_string(mode));
276 
277 #ifdef __x86_64__
278 	/*
279 	 * FIXME
280 	 * The x86_64 kvm selftests framework currently only supports a
281 	 * single PML4 which restricts the number of physical address
282 	 * bits we can change to 39.
283 	 */
284 	guest_pa_bits = 39;
285 #endif
286 #ifdef __aarch64__
287 	if (guest_pa_bits != 40)
288 		type = KVM_VM_TYPE_ARM_IPA_SIZE(guest_pa_bits);
289 #endif
290 	max_gfn = (1ul << (guest_pa_bits - guest_page_shift)) - 1;
291 	guest_page_size = (1ul << guest_page_shift);
292 	/*
293 	 * A little more than 1G of guest page sized pages.  Cover the
294 	 * case where the size is not aligned to 64 pages.
295 	 */
296 	guest_num_pages = (1ul << (30 - guest_page_shift)) + 16;
297 	host_page_size = getpagesize();
298 	host_num_pages = (guest_num_pages * guest_page_size) / host_page_size +
299 			 !!((guest_num_pages * guest_page_size) % host_page_size);
300 
301 	if (!phys_offset) {
302 		guest_test_phys_mem = (max_gfn - guest_num_pages) * guest_page_size;
303 		guest_test_phys_mem &= ~(host_page_size - 1);
304 	} else {
305 		guest_test_phys_mem = phys_offset;
306 	}
307 
308 	DEBUG("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
309 
310 	bmap = bitmap_alloc(host_num_pages);
311 	host_bmap_track = bitmap_alloc(host_num_pages);
312 
313 	vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code, type);
314 
315 #ifdef USE_CLEAR_DIRTY_LOG
316 	struct kvm_enable_cap cap = {};
317 
318 	cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
319 	cap.args[0] = 1;
320 	vm_enable_cap(vm, &cap);
321 #endif
322 
323 	/* Add an extra memory slot for testing dirty logging */
324 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
325 				    guest_test_phys_mem,
326 				    TEST_MEM_SLOT_INDEX,
327 				    guest_num_pages,
328 				    KVM_MEM_LOG_DIRTY_PAGES);
329 
330 	/* Do mapping for the dirty track memory slot */
331 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem,
332 		 guest_num_pages * guest_page_size, 0);
333 
334 	/* Cache the HVA pointer of the region */
335 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
336 
337 #ifdef __x86_64__
338 	vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
339 #endif
340 #ifdef __aarch64__
341 	ucall_init(vm, UCALL_MMIO, NULL);
342 #endif
343 
344 	/* Export the shared variables to the guest */
345 	sync_global_to_guest(vm, host_page_size);
346 	sync_global_to_guest(vm, guest_page_size);
347 	sync_global_to_guest(vm, guest_test_virt_mem);
348 	sync_global_to_guest(vm, guest_num_pages);
349 
350 	/* Start the iterations */
351 	iteration = 1;
352 	sync_global_to_guest(vm, iteration);
353 	host_quit = false;
354 	host_dirty_count = 0;
355 	host_clear_count = 0;
356 	host_track_next_count = 0;
357 
358 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
359 
360 	while (iteration < iterations) {
361 		/* Give the vcpu thread some time to dirty some pages */
362 		usleep(interval * 1000);
363 		kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
364 #ifdef USE_CLEAR_DIRTY_LOG
365 		kvm_vm_clear_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap, 0,
366 				       host_num_pages);
367 #endif
368 		vm_dirty_log_verify(bmap);
369 		iteration++;
370 		sync_global_to_guest(vm, iteration);
371 	}
372 
373 	/* Tell the vcpu thread to quit */
374 	host_quit = true;
375 	pthread_join(vcpu_thread, NULL);
376 
377 	DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
378 	      "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
379 	      host_track_next_count);
380 
381 	free(bmap);
382 	free(host_bmap_track);
383 	ucall_uninit(vm);
384 	kvm_vm_free(vm);
385 }
386 
387 struct vm_guest_mode_params {
388 	bool supported;
389 	bool enabled;
390 };
391 struct vm_guest_mode_params vm_guest_mode_params[NUM_VM_MODES];
392 
393 #define vm_guest_mode_params_init(mode, supported, enabled)					\
394 ({												\
395 	vm_guest_mode_params[mode] = (struct vm_guest_mode_params){ supported, enabled };	\
396 })
397 
398 static void help(char *name)
399 {
400 	int i;
401 
402 	puts("");
403 	printf("usage: %s [-h] [-i iterations] [-I interval] "
404 	       "[-p offset] [-m mode]\n", name);
405 	puts("");
406 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
407 	       TEST_HOST_LOOP_N);
408 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
409 	       TEST_HOST_LOOP_INTERVAL);
410 	printf(" -p: specify guest physical test memory offset\n"
411 	       "     Warning: a low offset can conflict with the loaded test code.\n");
412 	printf(" -m: specify the guest mode ID to test "
413 	       "(default: test all supported modes)\n"
414 	       "     This option may be used multiple times.\n"
415 	       "     Guest mode IDs:\n");
416 	for (i = 0; i < NUM_VM_MODES; ++i) {
417 		printf("         %d:    %s%s\n", i, vm_guest_mode_string(i),
418 		       vm_guest_mode_params[i].supported ? " (supported)" : "");
419 	}
420 	puts("");
421 	exit(0);
422 }
423 
424 int main(int argc, char *argv[])
425 {
426 	unsigned long iterations = TEST_HOST_LOOP_N;
427 	unsigned long interval = TEST_HOST_LOOP_INTERVAL;
428 	bool mode_selected = false;
429 	uint64_t phys_offset = 0;
430 	unsigned int mode;
431 	int opt, i;
432 #ifdef __aarch64__
433 	unsigned int host_ipa_limit;
434 #endif
435 
436 #ifdef USE_CLEAR_DIRTY_LOG
437 	if (!kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2)) {
438 		fprintf(stderr, "KVM_CLEAR_DIRTY_LOG not available, skipping tests\n");
439 		exit(KSFT_SKIP);
440 	}
441 #endif
442 
443 #ifdef __x86_64__
444 	vm_guest_mode_params_init(VM_MODE_P52V48_4K, true, true);
445 #endif
446 #ifdef __aarch64__
447 	vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
448 	vm_guest_mode_params_init(VM_MODE_P40V48_64K, true, true);
449 
450 	host_ipa_limit = kvm_check_cap(KVM_CAP_ARM_VM_IPA_SIZE);
451 	if (host_ipa_limit >= 52)
452 		vm_guest_mode_params_init(VM_MODE_P52V48_64K, true, true);
453 	if (host_ipa_limit >= 48) {
454 		vm_guest_mode_params_init(VM_MODE_P48V48_4K, true, true);
455 		vm_guest_mode_params_init(VM_MODE_P48V48_64K, true, true);
456 	}
457 #endif
458 
459 	while ((opt = getopt(argc, argv, "hi:I:p:m:")) != -1) {
460 		switch (opt) {
461 		case 'i':
462 			iterations = strtol(optarg, NULL, 10);
463 			break;
464 		case 'I':
465 			interval = strtol(optarg, NULL, 10);
466 			break;
467 		case 'p':
468 			phys_offset = strtoull(optarg, NULL, 0);
469 			break;
470 		case 'm':
471 			if (!mode_selected) {
472 				for (i = 0; i < NUM_VM_MODES; ++i)
473 					vm_guest_mode_params[i].enabled = false;
474 				mode_selected = true;
475 			}
476 			mode = strtoul(optarg, NULL, 10);
477 			TEST_ASSERT(mode < NUM_VM_MODES,
478 				    "Guest mode ID %d too big", mode);
479 			vm_guest_mode_params[mode].enabled = true;
480 			break;
481 		case 'h':
482 		default:
483 			help(argv[0]);
484 			break;
485 		}
486 	}
487 
488 	TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
489 	TEST_ASSERT(interval > 0, "Interval must be greater than zero");
490 
491 	DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
492 	      iterations, interval);
493 
494 	srandom(time(0));
495 
496 	for (i = 0; i < NUM_VM_MODES; ++i) {
497 		if (!vm_guest_mode_params[i].enabled)
498 			continue;
499 		TEST_ASSERT(vm_guest_mode_params[i].supported,
500 			    "Guest mode ID %d (%s) not supported.",
501 			    i, vm_guest_mode_string(i));
502 		run_test(i, iterations, interval, phys_offset);
503 	}
504 
505 	return 0;
506 }
507