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 VCPU_ID				1
23 
24 /* The memory slot index to track dirty pages */
25 #define TEST_MEM_SLOT_INDEX		1
26 
27 /* Default guest test virtual memory offset */
28 #define DEFAULT_GUEST_TEST_MEM		0xc0000000
29 
30 /* How many pages to dirty for each guest loop */
31 #define TEST_PAGES_PER_LOOP		1024
32 
33 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
34 #define TEST_HOST_LOOP_N		32UL
35 
36 /* Interval for each host loop (ms) */
37 #define TEST_HOST_LOOP_INTERVAL		10UL
38 
39 /* Dirty bitmaps are always little endian, so we need to swap on big endian */
40 #if defined(__s390x__)
41 # define BITOP_LE_SWIZZLE	((BITS_PER_LONG-1) & ~0x7)
42 # define test_bit_le(nr, addr) \
43 	test_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
44 # define set_bit_le(nr, addr) \
45 	set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
46 # define clear_bit_le(nr, addr) \
47 	clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
48 # define test_and_set_bit_le(nr, addr) \
49 	test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
50 # define test_and_clear_bit_le(nr, addr) \
51 	test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
52 #else
53 # define test_bit_le		test_bit
54 # define set_bit_le		set_bit
55 # define clear_bit_le		clear_bit
56 # define test_and_set_bit_le	test_and_set_bit
57 # define test_and_clear_bit_le	test_and_clear_bit
58 #endif
59 
60 /*
61  * Guest/Host shared variables. Ensure addr_gva2hva() and/or
62  * sync_global_to/from_guest() are used when accessing from
63  * the host. READ/WRITE_ONCE() should also be used with anything
64  * that may change.
65  */
66 static uint64_t host_page_size;
67 static uint64_t guest_page_size;
68 static uint64_t guest_num_pages;
69 static uint64_t random_array[TEST_PAGES_PER_LOOP];
70 static uint64_t iteration;
71 
72 /*
73  * Guest physical memory offset of the testing memory slot.
74  * This will be set to the topmost valid physical address minus
75  * the test memory size.
76  */
77 static uint64_t guest_test_phys_mem;
78 
79 /*
80  * Guest virtual memory offset of the testing memory slot.
81  * Must not conflict with identity mapped test code.
82  */
83 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
84 
85 /*
86  * Continuously write to the first 8 bytes of a random pages within
87  * the testing memory region.
88  */
89 static void guest_code(void)
90 {
91 	uint64_t addr;
92 	int i;
93 
94 	/*
95 	 * On s390x, all pages of a 1M segment are initially marked as dirty
96 	 * when a page of the segment is written to for the very first time.
97 	 * To compensate this specialty in this test, we need to touch all
98 	 * pages during the first iteration.
99 	 */
100 	for (i = 0; i < guest_num_pages; i++) {
101 		addr = guest_test_virt_mem + i * guest_page_size;
102 		*(uint64_t *)addr = READ_ONCE(iteration);
103 	}
104 
105 	while (true) {
106 		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
107 			addr = guest_test_virt_mem;
108 			addr += (READ_ONCE(random_array[i]) % guest_num_pages)
109 				* guest_page_size;
110 			addr &= ~(host_page_size - 1);
111 			*(uint64_t *)addr = READ_ONCE(iteration);
112 		}
113 
114 		/* Tell the host that we need more random numbers */
115 		GUEST_SYNC(1);
116 	}
117 }
118 
119 /* Host variables */
120 static bool host_quit;
121 
122 /* Points to the test VM memory region on which we track dirty logs */
123 static void *host_test_mem;
124 static uint64_t host_num_pages;
125 
126 /* For statistics only */
127 static uint64_t host_dirty_count;
128 static uint64_t host_clear_count;
129 static uint64_t host_track_next_count;
130 
131 /*
132  * We use this bitmap to track some pages that should have its dirty
133  * bit set in the _next_ iteration.  For example, if we detected the
134  * page value changed to current iteration but at the same time the
135  * page bit is cleared in the latest bitmap, then the system must
136  * report that write in the next get dirty log call.
137  */
138 static unsigned long *host_bmap_track;
139 
140 static void generate_random_array(uint64_t *guest_array, uint64_t size)
141 {
142 	uint64_t i;
143 
144 	for (i = 0; i < size; i++)
145 		guest_array[i] = random();
146 }
147 
148 static void *vcpu_worker(void *data)
149 {
150 	int ret;
151 	struct kvm_vm *vm = data;
152 	uint64_t *guest_array;
153 	uint64_t pages_count = 0;
154 	struct kvm_run *run;
155 
156 	run = vcpu_state(vm, VCPU_ID);
157 
158 	guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
159 	generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
160 
161 	while (!READ_ONCE(host_quit)) {
162 		/* Let the guest dirty the random pages */
163 		ret = _vcpu_run(vm, VCPU_ID);
164 		TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
165 		if (get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC) {
166 			pages_count += TEST_PAGES_PER_LOOP;
167 			generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
168 		} else {
169 			TEST_ASSERT(false,
170 				    "Invalid guest sync status: "
171 				    "exit_reason=%s\n",
172 				    exit_reason_str(run->exit_reason));
173 		}
174 	}
175 
176 	DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
177 
178 	return NULL;
179 }
180 
181 static void vm_dirty_log_verify(unsigned long *bmap)
182 {
183 	uint64_t page;
184 	uint64_t *value_ptr;
185 	uint64_t step = host_page_size >= guest_page_size ? 1 :
186 				guest_page_size / host_page_size;
187 
188 	for (page = 0; page < host_num_pages; page += step) {
189 		value_ptr = host_test_mem + page * host_page_size;
190 
191 		/* If this is a special page that we were tracking... */
192 		if (test_and_clear_bit_le(page, host_bmap_track)) {
193 			host_track_next_count++;
194 			TEST_ASSERT(test_bit_le(page, bmap),
195 				    "Page %"PRIu64" should have its dirty bit "
196 				    "set in this iteration but it is missing",
197 				    page);
198 		}
199 
200 		if (test_bit_le(page, bmap)) {
201 			host_dirty_count++;
202 			/*
203 			 * If the bit is set, the value written onto
204 			 * the corresponding page should be either the
205 			 * previous iteration number or the current one.
206 			 */
207 			TEST_ASSERT(*value_ptr == iteration ||
208 				    *value_ptr == iteration - 1,
209 				    "Set page %"PRIu64" value %"PRIu64
210 				    " incorrect (iteration=%"PRIu64")",
211 				    page, *value_ptr, iteration);
212 		} else {
213 			host_clear_count++;
214 			/*
215 			 * If cleared, the value written can be any
216 			 * value smaller or equals to the iteration
217 			 * number.  Note that the value can be exactly
218 			 * (iteration-1) if that write can happen
219 			 * like this:
220 			 *
221 			 * (1) increase loop count to "iteration-1"
222 			 * (2) write to page P happens (with value
223 			 *     "iteration-1")
224 			 * (3) get dirty log for "iteration-1"; we'll
225 			 *     see that page P bit is set (dirtied),
226 			 *     and not set the bit in host_bmap_track
227 			 * (4) increase loop count to "iteration"
228 			 *     (which is current iteration)
229 			 * (5) get dirty log for current iteration,
230 			 *     we'll see that page P is cleared, with
231 			 *     value "iteration-1".
232 			 */
233 			TEST_ASSERT(*value_ptr <= iteration,
234 				    "Clear page %"PRIu64" value %"PRIu64
235 				    " incorrect (iteration=%"PRIu64")",
236 				    page, *value_ptr, iteration);
237 			if (*value_ptr == iteration) {
238 				/*
239 				 * This page is _just_ modified; it
240 				 * should report its dirtyness in the
241 				 * next run
242 				 */
243 				set_bit_le(page, host_bmap_track);
244 			}
245 		}
246 	}
247 }
248 
249 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
250 				uint64_t extra_mem_pages, void *guest_code)
251 {
252 	struct kvm_vm *vm;
253 	uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
254 
255 	vm = _vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
256 	kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
257 #ifdef __x86_64__
258 	vm_create_irqchip(vm);
259 #endif
260 	vm_vcpu_add_default(vm, vcpuid, guest_code);
261 	return vm;
262 }
263 
264 #define DIRTY_MEM_BITS 30 /* 1G */
265 #define PAGE_SHIFT_4K  12
266 
267 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
268 		     unsigned long interval, uint64_t phys_offset)
269 {
270 	pthread_t vcpu_thread;
271 	struct kvm_vm *vm;
272 	unsigned long *bmap;
273 
274 	/*
275 	 * We reserve page table for 2 times of extra dirty mem which
276 	 * will definitely cover the original (1G+) test range.  Here
277 	 * we do the calculation with 4K page size which is the
278 	 * smallest so the page number will be enough for all archs
279 	 * (e.g., 64K page size guest will need even less memory for
280 	 * page tables).
281 	 */
282 	vm = create_vm(mode, VCPU_ID,
283 		       2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K),
284 		       guest_code);
285 
286 	guest_page_size = vm_get_page_size(vm);
287 	/*
288 	 * A little more than 1G of guest page sized pages.  Cover the
289 	 * case where the size is not aligned to 64 pages.
290 	 */
291 	guest_num_pages = (1ul << (DIRTY_MEM_BITS -
292 				   vm_get_page_shift(vm))) + 16;
293 #ifdef __s390x__
294 	/* Round up to multiple of 1M (segment size) */
295 	guest_num_pages = (guest_num_pages + 0xff) & ~0xffUL;
296 #endif
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 = (vm_get_max_gfn(vm) -
303 				       guest_num_pages) * guest_page_size;
304 		guest_test_phys_mem &= ~(host_page_size - 1);
305 	} else {
306 		guest_test_phys_mem = phys_offset;
307 	}
308 
309 #ifdef __s390x__
310 	/* Align to 1M (segment size) */
311 	guest_test_phys_mem &= ~((1 << 20) - 1);
312 #endif
313 
314 	DEBUG("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
315 
316 	bmap = bitmap_alloc(host_num_pages);
317 	host_bmap_track = bitmap_alloc(host_num_pages);
318 
319 #ifdef USE_CLEAR_DIRTY_LOG
320 	struct kvm_enable_cap cap = {};
321 
322 	cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
323 	cap.args[0] = 1;
324 	vm_enable_cap(vm, &cap);
325 #endif
326 
327 	/* Add an extra memory slot for testing dirty logging */
328 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
329 				    guest_test_phys_mem,
330 				    TEST_MEM_SLOT_INDEX,
331 				    guest_num_pages,
332 				    KVM_MEM_LOG_DIRTY_PAGES);
333 
334 	/* Do mapping for the dirty track memory slot */
335 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem,
336 		 guest_num_pages * guest_page_size, 0);
337 
338 	/* Cache the HVA pointer of the region */
339 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
340 
341 #ifdef __x86_64__
342 	vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
343 #endif
344 #ifdef __aarch64__
345 	ucall_init(vm, NULL);
346 #endif
347 
348 	/* Export the shared variables to the guest */
349 	sync_global_to_guest(vm, host_page_size);
350 	sync_global_to_guest(vm, guest_page_size);
351 	sync_global_to_guest(vm, guest_test_virt_mem);
352 	sync_global_to_guest(vm, guest_num_pages);
353 
354 	/* Start the iterations */
355 	iteration = 1;
356 	sync_global_to_guest(vm, iteration);
357 	host_quit = false;
358 	host_dirty_count = 0;
359 	host_clear_count = 0;
360 	host_track_next_count = 0;
361 
362 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
363 
364 	while (iteration < iterations) {
365 		/* Give the vcpu thread some time to dirty some pages */
366 		usleep(interval * 1000);
367 		kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
368 #ifdef USE_CLEAR_DIRTY_LOG
369 		kvm_vm_clear_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap, 0,
370 				       host_num_pages);
371 #endif
372 		vm_dirty_log_verify(bmap);
373 		iteration++;
374 		sync_global_to_guest(vm, iteration);
375 	}
376 
377 	/* Tell the vcpu thread to quit */
378 	host_quit = true;
379 	pthread_join(vcpu_thread, NULL);
380 
381 	DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
382 	      "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
383 	      host_track_next_count);
384 
385 	free(bmap);
386 	free(host_bmap_track);
387 	ucall_uninit(vm);
388 	kvm_vm_free(vm);
389 }
390 
391 struct vm_guest_mode_params {
392 	bool supported;
393 	bool enabled;
394 };
395 struct vm_guest_mode_params vm_guest_mode_params[NUM_VM_MODES];
396 
397 #define vm_guest_mode_params_init(mode, supported, enabled)					\
398 ({												\
399 	vm_guest_mode_params[mode] = (struct vm_guest_mode_params){ supported, enabled };	\
400 })
401 
402 static void help(char *name)
403 {
404 	int i;
405 
406 	puts("");
407 	printf("usage: %s [-h] [-i iterations] [-I interval] "
408 	       "[-p offset] [-m mode]\n", name);
409 	puts("");
410 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
411 	       TEST_HOST_LOOP_N);
412 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
413 	       TEST_HOST_LOOP_INTERVAL);
414 	printf(" -p: specify guest physical test memory offset\n"
415 	       "     Warning: a low offset can conflict with the loaded test code.\n");
416 	printf(" -m: specify the guest mode ID to test "
417 	       "(default: test all supported modes)\n"
418 	       "     This option may be used multiple times.\n"
419 	       "     Guest mode IDs:\n");
420 	for (i = 0; i < NUM_VM_MODES; ++i) {
421 		printf("         %d:    %s%s\n", i, vm_guest_mode_string(i),
422 		       vm_guest_mode_params[i].supported ? " (supported)" : "");
423 	}
424 	puts("");
425 	exit(0);
426 }
427 
428 int main(int argc, char *argv[])
429 {
430 	unsigned long iterations = TEST_HOST_LOOP_N;
431 	unsigned long interval = TEST_HOST_LOOP_INTERVAL;
432 	bool mode_selected = false;
433 	uint64_t phys_offset = 0;
434 	unsigned int mode;
435 	int opt, i;
436 #ifdef __aarch64__
437 	unsigned int host_ipa_limit;
438 #endif
439 
440 #ifdef USE_CLEAR_DIRTY_LOG
441 	if (!kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2)) {
442 		fprintf(stderr, "KVM_CLEAR_DIRTY_LOG not available, skipping tests\n");
443 		exit(KSFT_SKIP);
444 	}
445 #endif
446 
447 #ifdef __x86_64__
448 	vm_guest_mode_params_init(VM_MODE_PXXV48_4K, true, true);
449 #endif
450 #ifdef __aarch64__
451 	vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
452 	vm_guest_mode_params_init(VM_MODE_P40V48_64K, true, true);
453 
454 	host_ipa_limit = kvm_check_cap(KVM_CAP_ARM_VM_IPA_SIZE);
455 	if (host_ipa_limit >= 52)
456 		vm_guest_mode_params_init(VM_MODE_P52V48_64K, true, true);
457 	if (host_ipa_limit >= 48) {
458 		vm_guest_mode_params_init(VM_MODE_P48V48_4K, true, true);
459 		vm_guest_mode_params_init(VM_MODE_P48V48_64K, true, true);
460 	}
461 #endif
462 #ifdef __s390x__
463 	vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
464 #endif
465 
466 	while ((opt = getopt(argc, argv, "hi:I:p:m:")) != -1) {
467 		switch (opt) {
468 		case 'i':
469 			iterations = strtol(optarg, NULL, 10);
470 			break;
471 		case 'I':
472 			interval = strtol(optarg, NULL, 10);
473 			break;
474 		case 'p':
475 			phys_offset = strtoull(optarg, NULL, 0);
476 			break;
477 		case 'm':
478 			if (!mode_selected) {
479 				for (i = 0; i < NUM_VM_MODES; ++i)
480 					vm_guest_mode_params[i].enabled = false;
481 				mode_selected = true;
482 			}
483 			mode = strtoul(optarg, NULL, 10);
484 			TEST_ASSERT(mode < NUM_VM_MODES,
485 				    "Guest mode ID %d too big", mode);
486 			vm_guest_mode_params[mode].enabled = true;
487 			break;
488 		case 'h':
489 		default:
490 			help(argv[0]);
491 			break;
492 		}
493 	}
494 
495 	TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
496 	TEST_ASSERT(interval > 0, "Interval must be greater than zero");
497 
498 	DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
499 	      iterations, interval);
500 
501 	srandom(time(0));
502 
503 	for (i = 0; i < NUM_VM_MODES; ++i) {
504 		if (!vm_guest_mode_params[i].enabled)
505 			continue;
506 		TEST_ASSERT(vm_guest_mode_params[i].supported,
507 			    "Guest mode ID %d (%s) not supported.",
508 			    i, vm_guest_mode_string(i));
509 		run_test(i, iterations, interval, phys_offset);
510 	}
511 
512 	return 0;
513 }
514