1 // SPDX-License-Identifier: GPL-2.0
3 * KVM dirty page logging test
5 * Copyright (C) 2018, Red Hat, Inc.
8 #define _GNU_SOURCE /* for program_invocation_name */
15 #include <linux/bitmap.h>
16 #include <linux/bitops.h>
18 #include "test_util.h"
20 #include "processor.h"
24 /* The memory slot index to track dirty pages */
25 #define TEST_MEM_SLOT_INDEX 1
27 /* Default guest test virtual memory offset */
28 #define DEFAULT_GUEST_TEST_MEM 0xc0000000
30 /* How many pages to dirty for each guest loop */
31 #define TEST_PAGES_PER_LOOP 1024
33 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
34 #define TEST_HOST_LOOP_N 32UL
36 /* Interval for each host loop (ms) */
37 #define TEST_HOST_LOOP_INTERVAL 10UL
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)
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
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
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;
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.
77 static uint64_t guest_test_phys_mem;
80 * Guest virtual memory offset of the testing memory slot.
81 * Must not conflict with identity mapped test code.
83 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
86 * Continuously write to the first 8 bytes of a random pages within
87 * the testing memory region.
89 static void guest_code(void)
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.
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);
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)
110 addr &= ~(host_page_size - 1);
111 *(uint64_t *)addr = READ_ONCE(iteration);
114 /* Tell the host that we need more random numbers */
120 static bool host_quit;
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;
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;
132 /* Only use KVM_GET_DIRTY_LOG for logging */
133 LOG_MODE_DIRTY_LOG = 0,
135 /* Use both KVM_[GET|CLEAR]_DIRTY_LOG for logging */
136 LOG_MODE_CLEAR_LOG = 1,
140 /* Run all supported modes */
141 LOG_MODE_ALL = LOG_MODE_NUM,
144 /* Mode of logging to test. Default is to run all supported modes */
145 static enum log_mode_t host_log_mode_option = LOG_MODE_ALL;
146 /* Logging mode for current run */
147 static enum log_mode_t host_log_mode;
149 static bool clear_log_supported(void)
151 return kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
154 static void clear_log_create_vm_done(struct kvm_vm *vm)
156 struct kvm_enable_cap cap = {};
159 manual_caps = kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
160 TEST_ASSERT(manual_caps, "MANUAL_CAPS is zero!");
161 manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE |
162 KVM_DIRTY_LOG_INITIALLY_SET);
163 cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
164 cap.args[0] = manual_caps;
165 vm_enable_cap(vm, &cap);
168 static void dirty_log_collect_dirty_pages(struct kvm_vm *vm, int slot,
169 void *bitmap, uint32_t num_pages)
171 kvm_vm_get_dirty_log(vm, slot, bitmap);
174 static void clear_log_collect_dirty_pages(struct kvm_vm *vm, int slot,
175 void *bitmap, uint32_t num_pages)
177 kvm_vm_get_dirty_log(vm, slot, bitmap);
178 kvm_vm_clear_dirty_log(vm, slot, bitmap, 0, num_pages);
181 static void default_after_vcpu_run(struct kvm_vm *vm)
183 struct kvm_run *run = vcpu_state(vm, VCPU_ID);
185 TEST_ASSERT(get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC,
186 "Invalid guest sync status: exit_reason=%s\n",
187 exit_reason_str(run->exit_reason));
192 /* Return true if this mode is supported, otherwise false */
193 bool (*supported)(void);
194 /* Hook when the vm creation is done (before vcpu creation) */
195 void (*create_vm_done)(struct kvm_vm *vm);
196 /* Hook to collect the dirty pages into the bitmap provided */
197 void (*collect_dirty_pages) (struct kvm_vm *vm, int slot,
198 void *bitmap, uint32_t num_pages);
199 /* Hook to call when after each vcpu run */
200 void (*after_vcpu_run)(struct kvm_vm *vm);
201 } log_modes[LOG_MODE_NUM] = {
204 .collect_dirty_pages = dirty_log_collect_dirty_pages,
205 .after_vcpu_run = default_after_vcpu_run,
209 .supported = clear_log_supported,
210 .create_vm_done = clear_log_create_vm_done,
211 .collect_dirty_pages = clear_log_collect_dirty_pages,
212 .after_vcpu_run = default_after_vcpu_run,
217 * We use this bitmap to track some pages that should have its dirty
218 * bit set in the _next_ iteration. For example, if we detected the
219 * page value changed to current iteration but at the same time the
220 * page bit is cleared in the latest bitmap, then the system must
221 * report that write in the next get dirty log call.
223 static unsigned long *host_bmap_track;
225 static void log_modes_dump(void)
230 for (i = 0; i < LOG_MODE_NUM; i++)
231 printf(", %s", log_modes[i].name);
235 static bool log_mode_supported(void)
237 struct log_mode *mode = &log_modes[host_log_mode];
240 return mode->supported();
245 static void log_mode_create_vm_done(struct kvm_vm *vm)
247 struct log_mode *mode = &log_modes[host_log_mode];
249 if (mode->create_vm_done)
250 mode->create_vm_done(vm);
253 static void log_mode_collect_dirty_pages(struct kvm_vm *vm, int slot,
254 void *bitmap, uint32_t num_pages)
256 struct log_mode *mode = &log_modes[host_log_mode];
258 TEST_ASSERT(mode->collect_dirty_pages != NULL,
259 "collect_dirty_pages() is required for any log mode!");
260 mode->collect_dirty_pages(vm, slot, bitmap, num_pages);
263 static void log_mode_after_vcpu_run(struct kvm_vm *vm)
265 struct log_mode *mode = &log_modes[host_log_mode];
267 if (mode->after_vcpu_run)
268 mode->after_vcpu_run(vm);
271 static void generate_random_array(uint64_t *guest_array, uint64_t size)
275 for (i = 0; i < size; i++)
276 guest_array[i] = random();
279 static void *vcpu_worker(void *data)
282 struct kvm_vm *vm = data;
283 uint64_t *guest_array;
284 uint64_t pages_count = 0;
286 guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
288 while (!READ_ONCE(host_quit)) {
289 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
290 pages_count += TEST_PAGES_PER_LOOP;
291 /* Let the guest dirty the random pages */
292 ret = _vcpu_run(vm, VCPU_ID);
293 TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
294 log_mode_after_vcpu_run(vm);
297 pr_info("Dirtied %"PRIu64" pages\n", pages_count);
302 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap)
304 uint64_t step = vm_num_host_pages(mode, 1);
308 for (page = 0; page < host_num_pages; page += step) {
309 value_ptr = host_test_mem + page * host_page_size;
311 /* If this is a special page that we were tracking... */
312 if (test_and_clear_bit_le(page, host_bmap_track)) {
313 host_track_next_count++;
314 TEST_ASSERT(test_bit_le(page, bmap),
315 "Page %"PRIu64" should have its dirty bit "
316 "set in this iteration but it is missing",
320 if (test_and_clear_bit_le(page, bmap)) {
323 * If the bit is set, the value written onto
324 * the corresponding page should be either the
325 * previous iteration number or the current one.
327 TEST_ASSERT(*value_ptr == iteration ||
328 *value_ptr == iteration - 1,
329 "Set page %"PRIu64" value %"PRIu64
330 " incorrect (iteration=%"PRIu64")",
331 page, *value_ptr, iteration);
335 * If cleared, the value written can be any
336 * value smaller or equals to the iteration
337 * number. Note that the value can be exactly
338 * (iteration-1) if that write can happen
341 * (1) increase loop count to "iteration-1"
342 * (2) write to page P happens (with value
344 * (3) get dirty log for "iteration-1"; we'll
345 * see that page P bit is set (dirtied),
346 * and not set the bit in host_bmap_track
347 * (4) increase loop count to "iteration"
348 * (which is current iteration)
349 * (5) get dirty log for current iteration,
350 * we'll see that page P is cleared, with
351 * value "iteration-1".
353 TEST_ASSERT(*value_ptr <= iteration,
354 "Clear page %"PRIu64" value %"PRIu64
355 " incorrect (iteration=%"PRIu64")",
356 page, *value_ptr, iteration);
357 if (*value_ptr == iteration) {
359 * This page is _just_ modified; it
360 * should report its dirtyness in the
363 set_bit_le(page, host_bmap_track);
369 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
370 uint64_t extra_mem_pages, void *guest_code)
373 uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
375 pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
377 vm = vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
378 kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
380 vm_create_irqchip(vm);
382 log_mode_create_vm_done(vm);
383 vm_vcpu_add_default(vm, vcpuid, guest_code);
387 #define DIRTY_MEM_BITS 30 /* 1G */
388 #define PAGE_SHIFT_4K 12
390 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
391 unsigned long interval, uint64_t phys_offset)
393 pthread_t vcpu_thread;
397 if (!log_mode_supported()) {
398 print_skip("Log mode '%s' not supported",
399 log_modes[host_log_mode].name);
404 * We reserve page table for 2 times of extra dirty mem which
405 * will definitely cover the original (1G+) test range. Here
406 * we do the calculation with 4K page size which is the
407 * smallest so the page number will be enough for all archs
408 * (e.g., 64K page size guest will need even less memory for
411 vm = create_vm(mode, VCPU_ID,
412 2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K),
415 guest_page_size = vm_get_page_size(vm);
417 * A little more than 1G of guest page sized pages. Cover the
418 * case where the size is not aligned to 64 pages.
420 guest_num_pages = (1ul << (DIRTY_MEM_BITS -
421 vm_get_page_shift(vm))) + 3;
422 guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
424 host_page_size = getpagesize();
425 host_num_pages = vm_num_host_pages(mode, guest_num_pages);
428 guest_test_phys_mem = (vm_get_max_gfn(vm) -
429 guest_num_pages) * guest_page_size;
430 guest_test_phys_mem &= ~(host_page_size - 1);
432 guest_test_phys_mem = phys_offset;
436 /* Align to 1M (segment size) */
437 guest_test_phys_mem &= ~((1 << 20) - 1);
440 pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
442 bmap = bitmap_alloc(host_num_pages);
443 host_bmap_track = bitmap_alloc(host_num_pages);
445 /* Add an extra memory slot for testing dirty logging */
446 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
450 KVM_MEM_LOG_DIRTY_PAGES);
452 /* Do mapping for the dirty track memory slot */
453 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages, 0);
455 /* Cache the HVA pointer of the region */
456 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
459 vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
461 ucall_init(vm, NULL);
463 /* Export the shared variables to the guest */
464 sync_global_to_guest(vm, host_page_size);
465 sync_global_to_guest(vm, guest_page_size);
466 sync_global_to_guest(vm, guest_test_virt_mem);
467 sync_global_to_guest(vm, guest_num_pages);
469 /* Start the iterations */
471 sync_global_to_guest(vm, iteration);
473 host_dirty_count = 0;
474 host_clear_count = 0;
475 host_track_next_count = 0;
477 pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
479 while (iteration < iterations) {
480 /* Give the vcpu thread some time to dirty some pages */
481 usleep(interval * 1000);
482 log_mode_collect_dirty_pages(vm, TEST_MEM_SLOT_INDEX,
483 bmap, host_num_pages);
484 vm_dirty_log_verify(mode, bmap);
486 sync_global_to_guest(vm, iteration);
489 /* Tell the vcpu thread to quit */
491 pthread_join(vcpu_thread, NULL);
493 pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
494 "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
495 host_track_next_count);
498 free(host_bmap_track);
507 static struct guest_mode guest_modes[NUM_VM_MODES];
509 #define guest_mode_init(mode, supported, enabled) ({ \
510 guest_modes[mode] = (struct guest_mode){ supported, enabled }; \
513 static void help(char *name)
518 printf("usage: %s [-h] [-i iterations] [-I interval] "
519 "[-p offset] [-m mode]\n", name);
521 printf(" -i: specify iteration counts (default: %"PRIu64")\n",
523 printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
524 TEST_HOST_LOOP_INTERVAL);
525 printf(" -p: specify guest physical test memory offset\n"
526 " Warning: a low offset can conflict with the loaded test code.\n");
527 printf(" -M: specify the host logging mode "
528 "(default: run all log modes). Supported modes: \n\t");
530 printf(" -m: specify the guest mode ID to test "
531 "(default: test all supported modes)\n"
532 " This option may be used multiple times.\n"
533 " Guest mode IDs:\n");
534 for (i = 0; i < NUM_VM_MODES; ++i) {
535 printf(" %d: %s%s\n", i, vm_guest_mode_string(i),
536 guest_modes[i].supported ? " (supported)" : "");
542 int main(int argc, char *argv[])
544 unsigned long iterations = TEST_HOST_LOOP_N;
545 unsigned long interval = TEST_HOST_LOOP_INTERVAL;
546 bool mode_selected = false;
547 uint64_t phys_offset = 0;
552 guest_mode_init(VM_MODE_PXXV48_4K, true, true);
555 guest_mode_init(VM_MODE_P40V48_4K, true, true);
556 guest_mode_init(VM_MODE_P40V48_64K, true, true);
559 unsigned int limit = kvm_check_cap(KVM_CAP_ARM_VM_IPA_SIZE);
562 guest_mode_init(VM_MODE_P52V48_64K, true, true);
564 guest_mode_init(VM_MODE_P48V48_4K, true, true);
565 guest_mode_init(VM_MODE_P48V48_64K, true, true);
570 guest_mode_init(VM_MODE_P40V48_4K, true, true);
573 while ((opt = getopt(argc, argv, "hi:I:p:m:M:")) != -1) {
576 iterations = strtol(optarg, NULL, 10);
579 interval = strtol(optarg, NULL, 10);
582 phys_offset = strtoull(optarg, NULL, 0);
585 if (!mode_selected) {
586 for (i = 0; i < NUM_VM_MODES; ++i)
587 guest_modes[i].enabled = false;
588 mode_selected = true;
590 mode = strtoul(optarg, NULL, 10);
591 TEST_ASSERT(mode < NUM_VM_MODES,
592 "Guest mode ID %d too big", mode);
593 guest_modes[mode].enabled = true;
596 if (!strcmp(optarg, "all")) {
597 host_log_mode_option = LOG_MODE_ALL;
600 for (i = 0; i < LOG_MODE_NUM; i++) {
601 if (!strcmp(optarg, log_modes[i].name)) {
602 pr_info("Setting log mode to: '%s'\n",
604 host_log_mode_option = i;
608 if (i == LOG_MODE_NUM) {
609 printf("Log mode '%s' invalid. Please choose "
622 TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
623 TEST_ASSERT(interval > 0, "Interval must be greater than zero");
625 pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
626 iterations, interval);
630 for (i = 0; i < NUM_VM_MODES; ++i) {
631 if (!guest_modes[i].enabled)
633 TEST_ASSERT(guest_modes[i].supported,
634 "Guest mode ID %d (%s) not supported.",
635 i, vm_guest_mode_string(i));
636 if (host_log_mode_option == LOG_MODE_ALL) {
637 /* Run each log mode */
638 for (j = 0; j < LOG_MODE_NUM; j++) {
639 pr_info("Testing Log Mode '%s'\n",
642 run_test(i, iterations, interval, phys_offset);
645 host_log_mode = host_log_mode_option;
646 run_test(i, iterations, interval, phys_offset);