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"
26 /* The memory slot index to track dirty pages */
27 #define TEST_MEM_SLOT_INDEX 1
29 /* Default guest test virtual memory offset */
30 #define DEFAULT_GUEST_TEST_MEM 0xc0000000
32 /* How many pages to dirty for each guest loop */
33 #define TEST_PAGES_PER_LOOP 1024
35 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
36 #define TEST_HOST_LOOP_N 32UL
38 /* Interval for each host loop (ms) */
39 #define TEST_HOST_LOOP_INTERVAL 10UL
41 /* Dirty bitmaps are always little endian, so we need to swap on big endian */
42 #if defined(__s390x__)
43 # define BITOP_LE_SWIZZLE ((BITS_PER_LONG-1) & ~0x7)
44 # define test_bit_le(nr, addr) \
45 test_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
46 # define set_bit_le(nr, addr) \
47 set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
48 # define clear_bit_le(nr, addr) \
49 clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
50 # define test_and_set_bit_le(nr, addr) \
51 test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
52 # define test_and_clear_bit_le(nr, addr) \
53 test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
55 # define test_bit_le test_bit
56 # define set_bit_le set_bit
57 # define clear_bit_le clear_bit
58 # define test_and_set_bit_le test_and_set_bit
59 # define test_and_clear_bit_le test_and_clear_bit
63 * Guest/Host shared variables. Ensure addr_gva2hva() and/or
64 * sync_global_to/from_guest() are used when accessing from
65 * the host. READ/WRITE_ONCE() should also be used with anything
68 static uint64_t host_page_size;
69 static uint64_t guest_page_size;
70 static uint64_t guest_num_pages;
71 static uint64_t random_array[TEST_PAGES_PER_LOOP];
72 static uint64_t iteration;
75 * Guest physical memory offset of the testing memory slot.
76 * This will be set to the topmost valid physical address minus
77 * the test memory size.
79 static uint64_t guest_test_phys_mem;
82 * Guest virtual memory offset of the testing memory slot.
83 * Must not conflict with identity mapped test code.
85 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
88 * Continuously write to the first 8 bytes of a random pages within
89 * the testing memory region.
91 static void guest_code(void)
97 * On s390x, all pages of a 1M segment are initially marked as dirty
98 * when a page of the segment is written to for the very first time.
99 * To compensate this specialty in this test, we need to touch all
100 * pages during the first iteration.
102 for (i = 0; i < guest_num_pages; i++) {
103 addr = guest_test_virt_mem + i * guest_page_size;
104 *(uint64_t *)addr = READ_ONCE(iteration);
108 for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
109 addr = guest_test_virt_mem;
110 addr += (READ_ONCE(random_array[i]) % guest_num_pages)
112 addr &= ~(host_page_size - 1);
113 *(uint64_t *)addr = READ_ONCE(iteration);
116 /* Tell the host that we need more random numbers */
122 static bool host_quit;
124 /* Points to the test VM memory region on which we track dirty logs */
125 static void *host_test_mem;
126 static uint64_t host_num_pages;
128 /* For statistics only */
129 static uint64_t host_dirty_count;
130 static uint64_t host_clear_count;
131 static uint64_t host_track_next_count;
134 * We use this bitmap to track some pages that should have its dirty
135 * bit set in the _next_ iteration. For example, if we detected the
136 * page value changed to current iteration but at the same time the
137 * page bit is cleared in the latest bitmap, then the system must
138 * report that write in the next get dirty log call.
140 static unsigned long *host_bmap_track;
142 static void generate_random_array(uint64_t *guest_array, uint64_t size)
146 for (i = 0; i < size; i++)
147 guest_array[i] = random();
150 static void *vcpu_worker(void *data)
153 struct kvm_vm *vm = data;
154 uint64_t *guest_array;
155 uint64_t pages_count = 0;
158 run = vcpu_state(vm, VCPU_ID);
160 guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
161 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
163 while (!READ_ONCE(host_quit)) {
164 /* Let the guest dirty the random pages */
165 ret = _vcpu_run(vm, VCPU_ID);
166 TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
167 if (get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC) {
168 pages_count += TEST_PAGES_PER_LOOP;
169 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
172 "Invalid guest sync status: "
174 exit_reason_str(run->exit_reason));
178 DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
183 static void vm_dirty_log_verify(unsigned long *bmap)
187 uint64_t step = host_page_size >= guest_page_size ? 1 :
188 guest_page_size / host_page_size;
190 for (page = 0; page < host_num_pages; page += step) {
191 value_ptr = host_test_mem + page * host_page_size;
193 /* If this is a special page that we were tracking... */
194 if (test_and_clear_bit_le(page, host_bmap_track)) {
195 host_track_next_count++;
196 TEST_ASSERT(test_bit_le(page, bmap),
197 "Page %"PRIu64" should have its dirty bit "
198 "set in this iteration but it is missing",
202 if (test_bit_le(page, bmap)) {
205 * If the bit is set, the value written onto
206 * the corresponding page should be either the
207 * previous iteration number or the current one.
209 TEST_ASSERT(*value_ptr == iteration ||
210 *value_ptr == iteration - 1,
211 "Set page %"PRIu64" value %"PRIu64
212 " incorrect (iteration=%"PRIu64")",
213 page, *value_ptr, iteration);
217 * If cleared, the value written can be any
218 * value smaller or equals to the iteration
219 * number. Note that the value can be exactly
220 * (iteration-1) if that write can happen
223 * (1) increase loop count to "iteration-1"
224 * (2) write to page P happens (with value
226 * (3) get dirty log for "iteration-1"; we'll
227 * see that page P bit is set (dirtied),
228 * and not set the bit in host_bmap_track
229 * (4) increase loop count to "iteration"
230 * (which is current iteration)
231 * (5) get dirty log for current iteration,
232 * we'll see that page P is cleared, with
233 * value "iteration-1".
235 TEST_ASSERT(*value_ptr <= iteration,
236 "Clear page %"PRIu64" value %"PRIu64
237 " incorrect (iteration=%"PRIu64")",
238 page, *value_ptr, iteration);
239 if (*value_ptr == iteration) {
241 * This page is _just_ modified; it
242 * should report its dirtyness in the
245 set_bit_le(page, host_bmap_track);
251 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
252 uint64_t extra_mem_pages, void *guest_code)
255 uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
257 vm = _vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
258 kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
260 vm_create_irqchip(vm);
262 vm_vcpu_add_default(vm, vcpuid, guest_code);
266 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
267 unsigned long interval, uint64_t phys_offset)
269 unsigned int guest_pa_bits, guest_page_shift;
270 pthread_t vcpu_thread;
276 case VM_MODE_P52V48_4K:
278 guest_page_shift = 12;
280 case VM_MODE_P52V48_64K:
282 guest_page_shift = 16;
284 case VM_MODE_P48V48_4K:
286 guest_page_shift = 12;
288 case VM_MODE_P48V48_64K:
290 guest_page_shift = 16;
292 case VM_MODE_P40V48_4K:
294 guest_page_shift = 12;
296 case VM_MODE_P40V48_64K:
298 guest_page_shift = 16;
301 TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", mode);
304 DEBUG("Testing guest mode: %s\n", vm_guest_mode_string(mode));
309 * The x86_64 kvm selftests framework currently only supports a
310 * single PML4 which restricts the number of physical address
311 * bits we can change to 39.
315 max_gfn = (1ul << (guest_pa_bits - guest_page_shift)) - 1;
316 guest_page_size = (1ul << guest_page_shift);
318 * A little more than 1G of guest page sized pages. Cover the
319 * case where the size is not aligned to 64 pages.
321 guest_num_pages = (1ul << (30 - guest_page_shift)) + 16;
323 /* Round up to multiple of 1M (segment size) */
324 guest_num_pages = (guest_num_pages + 0xff) & ~0xffUL;
326 host_page_size = getpagesize();
327 host_num_pages = (guest_num_pages * guest_page_size) / host_page_size +
328 !!((guest_num_pages * guest_page_size) % host_page_size);
331 guest_test_phys_mem = (max_gfn - guest_num_pages) * guest_page_size;
332 guest_test_phys_mem &= ~(host_page_size - 1);
334 guest_test_phys_mem = phys_offset;
338 /* Align to 1M (segment size) */
339 guest_test_phys_mem &= ~((1 << 20) - 1);
342 DEBUG("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
344 bmap = bitmap_alloc(host_num_pages);
345 host_bmap_track = bitmap_alloc(host_num_pages);
347 vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code);
349 #ifdef USE_CLEAR_DIRTY_LOG
350 struct kvm_enable_cap cap = {};
352 cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
354 vm_enable_cap(vm, &cap);
357 /* Add an extra memory slot for testing dirty logging */
358 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
362 KVM_MEM_LOG_DIRTY_PAGES);
364 /* Do mapping for the dirty track memory slot */
365 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem,
366 guest_num_pages * guest_page_size, 0);
368 /* Cache the HVA pointer of the region */
369 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
372 vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
375 ucall_init(vm, NULL);
378 /* Export the shared variables to the guest */
379 sync_global_to_guest(vm, host_page_size);
380 sync_global_to_guest(vm, guest_page_size);
381 sync_global_to_guest(vm, guest_test_virt_mem);
382 sync_global_to_guest(vm, guest_num_pages);
384 /* Start the iterations */
386 sync_global_to_guest(vm, iteration);
388 host_dirty_count = 0;
389 host_clear_count = 0;
390 host_track_next_count = 0;
392 pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
394 while (iteration < iterations) {
395 /* Give the vcpu thread some time to dirty some pages */
396 usleep(interval * 1000);
397 kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
398 #ifdef USE_CLEAR_DIRTY_LOG
399 kvm_vm_clear_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap, 0,
402 vm_dirty_log_verify(bmap);
404 sync_global_to_guest(vm, iteration);
407 /* Tell the vcpu thread to quit */
409 pthread_join(vcpu_thread, NULL);
411 DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
412 "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
413 host_track_next_count);
416 free(host_bmap_track);
421 struct vm_guest_mode_params {
425 struct vm_guest_mode_params vm_guest_mode_params[NUM_VM_MODES];
427 #define vm_guest_mode_params_init(mode, supported, enabled) \
429 vm_guest_mode_params[mode] = (struct vm_guest_mode_params){ supported, enabled }; \
432 static void help(char *name)
437 printf("usage: %s [-h] [-i iterations] [-I interval] "
438 "[-p offset] [-m mode]\n", name);
440 printf(" -i: specify iteration counts (default: %"PRIu64")\n",
442 printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
443 TEST_HOST_LOOP_INTERVAL);
444 printf(" -p: specify guest physical test memory offset\n"
445 " Warning: a low offset can conflict with the loaded test code.\n");
446 printf(" -m: specify the guest mode ID to test "
447 "(default: test all supported modes)\n"
448 " This option may be used multiple times.\n"
449 " Guest mode IDs:\n");
450 for (i = 0; i < NUM_VM_MODES; ++i) {
451 printf(" %d: %s%s\n", i, vm_guest_mode_string(i),
452 vm_guest_mode_params[i].supported ? " (supported)" : "");
458 int main(int argc, char *argv[])
460 unsigned long iterations = TEST_HOST_LOOP_N;
461 unsigned long interval = TEST_HOST_LOOP_INTERVAL;
462 bool mode_selected = false;
463 uint64_t phys_offset = 0;
467 unsigned int host_ipa_limit;
470 #ifdef USE_CLEAR_DIRTY_LOG
471 if (!kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2)) {
472 fprintf(stderr, "KVM_CLEAR_DIRTY_LOG not available, skipping tests\n");
478 vm_guest_mode_params_init(VM_MODE_P52V48_4K, true, true);
481 vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
482 vm_guest_mode_params_init(VM_MODE_P40V48_64K, true, true);
484 host_ipa_limit = kvm_check_cap(KVM_CAP_ARM_VM_IPA_SIZE);
485 if (host_ipa_limit >= 52)
486 vm_guest_mode_params_init(VM_MODE_P52V48_64K, true, true);
487 if (host_ipa_limit >= 48) {
488 vm_guest_mode_params_init(VM_MODE_P48V48_4K, true, true);
489 vm_guest_mode_params_init(VM_MODE_P48V48_64K, true, true);
493 vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
496 while ((opt = getopt(argc, argv, "hi:I:p:m:")) != -1) {
499 iterations = strtol(optarg, NULL, 10);
502 interval = strtol(optarg, NULL, 10);
505 phys_offset = strtoull(optarg, NULL, 0);
508 if (!mode_selected) {
509 for (i = 0; i < NUM_VM_MODES; ++i)
510 vm_guest_mode_params[i].enabled = false;
511 mode_selected = true;
513 mode = strtoul(optarg, NULL, 10);
514 TEST_ASSERT(mode < NUM_VM_MODES,
515 "Guest mode ID %d too big", mode);
516 vm_guest_mode_params[mode].enabled = true;
525 TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
526 TEST_ASSERT(interval > 0, "Interval must be greater than zero");
528 DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
529 iterations, interval);
533 for (i = 0; i < NUM_VM_MODES; ++i) {
534 if (!vm_guest_mode_params[i].enabled)
536 TEST_ASSERT(vm_guest_mode_params[i].supported,
537 "Guest mode ID %d (%s) not supported.",
538 i, vm_guest_mode_string(i));
539 run_test(i, iterations, interval, phys_offset);