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 memory offset, 1G */
30 #define DEFAULT_GUEST_TEST_MEM 0x40000000
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
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
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;
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.
58 static uint64_t guest_test_phys_mem;
61 * Guest virtual memory offset of the testing memory slot.
62 * Must not conflict with identity mapped test code.
64 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
67 * Continuously write to the first 8 bytes of a random pages within
68 * the testing memory region.
70 static void guest_code(void)
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)
79 addr &= ~(host_page_size - 1);
80 *(uint64_t *)addr = READ_ONCE(iteration);
83 /* Tell the host that we need more random numbers */
89 static bool host_quit;
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;
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;
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.
107 static unsigned long *host_bmap_track;
109 static void generate_random_array(uint64_t *guest_array, uint64_t size)
113 for (i = 0; i < size; i++)
114 guest_array[i] = random();
117 static void *vcpu_worker(void *data)
120 struct kvm_vm *vm = data;
121 uint64_t *guest_array;
122 uint64_t pages_count = 0;
125 run = vcpu_state(vm, VCPU_ID);
127 guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
128 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
130 while (!READ_ONCE(host_quit)) {
131 /* Let the guest dirty the random pages */
132 ret = _vcpu_run(vm, VCPU_ID);
133 TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
134 if (get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC) {
135 pages_count += TEST_PAGES_PER_LOOP;
136 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
139 "Invalid guest sync status: "
141 exit_reason_str(run->exit_reason));
145 DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
150 static void vm_dirty_log_verify(unsigned long *bmap)
154 uint64_t step = host_page_size >= guest_page_size ? 1 :
155 guest_page_size / host_page_size;
157 for (page = 0; page < host_num_pages; page += step) {
158 value_ptr = host_test_mem + page * host_page_size;
160 /* If this is a special page that we were tracking... */
161 if (test_and_clear_bit(page, host_bmap_track)) {
162 host_track_next_count++;
163 TEST_ASSERT(test_bit(page, bmap),
164 "Page %"PRIu64" should have its dirty bit "
165 "set in this iteration but it is missing",
169 if (test_bit(page, bmap)) {
172 * If the bit is set, the value written onto
173 * the corresponding page should be either the
174 * previous iteration number or the current one.
176 TEST_ASSERT(*value_ptr == iteration ||
177 *value_ptr == iteration - 1,
178 "Set page %"PRIu64" value %"PRIu64
179 " incorrect (iteration=%"PRIu64")",
180 page, *value_ptr, iteration);
184 * If cleared, the value written can be any
185 * value smaller or equals to the iteration
186 * number. Note that the value can be exactly
187 * (iteration-1) if that write can happen
190 * (1) increase loop count to "iteration-1"
191 * (2) write to page P happens (with value
193 * (3) get dirty log for "iteration-1"; we'll
194 * see that page P bit is set (dirtied),
195 * and not set the bit in host_bmap_track
196 * (4) increase loop count to "iteration"
197 * (which is current iteration)
198 * (5) get dirty log for current iteration,
199 * we'll see that page P is cleared, with
200 * value "iteration-1".
202 TEST_ASSERT(*value_ptr <= iteration,
203 "Clear page %"PRIu64" value %"PRIu64
204 " incorrect (iteration=%"PRIu64")",
205 page, *value_ptr, iteration);
206 if (*value_ptr == iteration) {
208 * This page is _just_ modified; it
209 * should report its dirtyness in the
212 set_bit(page, host_bmap_track);
218 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
219 uint64_t extra_mem_pages, void *guest_code,
223 uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
225 vm = _vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages,
227 kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
229 vm_create_irqchip(vm);
231 vm_vcpu_add_default(vm, vcpuid, guest_code);
235 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
236 unsigned long interval, uint64_t phys_offset)
238 unsigned int guest_pa_bits, guest_page_shift;
239 pthread_t vcpu_thread;
243 unsigned long type = 0;
246 case VM_MODE_P52V48_4K:
248 guest_page_shift = 12;
250 case VM_MODE_P52V48_64K:
252 guest_page_shift = 16;
254 case VM_MODE_P48V48_4K:
256 guest_page_shift = 12;
258 case VM_MODE_P48V48_64K:
260 guest_page_shift = 16;
262 case VM_MODE_P40V48_4K:
264 guest_page_shift = 12;
266 case VM_MODE_P40V48_64K:
268 guest_page_shift = 16;
271 TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", mode);
274 DEBUG("Testing guest mode: %s\n", vm_guest_mode_string(mode));
279 * The x86_64 kvm selftests framework currently only supports a
280 * single PML4 which restricts the number of physical address
281 * bits we can change to 39.
286 if (guest_pa_bits != 40)
287 type = KVM_VM_TYPE_ARM_IPA_SIZE(guest_pa_bits);
289 max_gfn = (1ul << (guest_pa_bits - guest_page_shift)) - 1;
290 guest_page_size = (1ul << guest_page_shift);
292 * A little more than 1G of guest page sized pages. Cover the
293 * case where the size is not aligned to 64 pages.
295 guest_num_pages = (1ul << (30 - guest_page_shift)) + 16;
296 host_page_size = getpagesize();
297 host_num_pages = (guest_num_pages * guest_page_size) / host_page_size +
298 !!((guest_num_pages * guest_page_size) % host_page_size);
301 guest_test_phys_mem = (max_gfn - guest_num_pages) * guest_page_size;
302 guest_test_phys_mem &= ~(host_page_size - 1);
304 guest_test_phys_mem = phys_offset;
307 DEBUG("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
309 bmap = bitmap_alloc(host_num_pages);
310 host_bmap_track = bitmap_alloc(host_num_pages);
312 vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code, type);
314 #ifdef USE_CLEAR_DIRTY_LOG
315 struct kvm_enable_cap cap = {};
317 cap.cap = KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2;
319 vm_enable_cap(vm, &cap);
322 /* Add an extra memory slot for testing dirty logging */
323 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
327 KVM_MEM_LOG_DIRTY_PAGES);
329 /* Do mapping for the dirty track memory slot */
330 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem,
331 guest_num_pages * guest_page_size, 0);
333 /* Cache the HVA pointer of the region */
334 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
337 vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
340 ucall_init(vm, UCALL_MMIO, NULL);
343 /* Export the shared variables to the guest */
344 sync_global_to_guest(vm, host_page_size);
345 sync_global_to_guest(vm, guest_page_size);
346 sync_global_to_guest(vm, guest_test_virt_mem);
347 sync_global_to_guest(vm, guest_num_pages);
349 /* Start the iterations */
351 sync_global_to_guest(vm, iteration);
353 host_dirty_count = 0;
354 host_clear_count = 0;
355 host_track_next_count = 0;
357 pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
359 while (iteration < iterations) {
360 /* Give the vcpu thread some time to dirty some pages */
361 usleep(interval * 1000);
362 kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
363 #ifdef USE_CLEAR_DIRTY_LOG
364 kvm_vm_clear_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap, 0,
367 vm_dirty_log_verify(bmap);
369 sync_global_to_guest(vm, iteration);
372 /* Tell the vcpu thread to quit */
374 pthread_join(vcpu_thread, NULL);
376 DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
377 "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
378 host_track_next_count);
381 free(host_bmap_track);
386 struct vm_guest_mode_params {
390 struct vm_guest_mode_params vm_guest_mode_params[NUM_VM_MODES];
392 #define vm_guest_mode_params_init(mode, supported, enabled) \
394 vm_guest_mode_params[mode] = (struct vm_guest_mode_params){ supported, enabled }; \
397 static void help(char *name)
402 printf("usage: %s [-h] [-i iterations] [-I interval] "
403 "[-p offset] [-m mode]\n", name);
405 printf(" -i: specify iteration counts (default: %"PRIu64")\n",
407 printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
408 TEST_HOST_LOOP_INTERVAL);
409 printf(" -p: specify guest physical test memory offset\n"
410 " Warning: a low offset can conflict with the loaded test code.\n");
411 printf(" -m: specify the guest mode ID to test "
412 "(default: test all supported modes)\n"
413 " This option may be used multiple times.\n"
414 " Guest mode IDs:\n");
415 for (i = 0; i < NUM_VM_MODES; ++i) {
416 printf(" %d: %s%s\n", i, vm_guest_mode_string(i),
417 vm_guest_mode_params[i].supported ? " (supported)" : "");
423 int main(int argc, char *argv[])
425 unsigned long iterations = TEST_HOST_LOOP_N;
426 unsigned long interval = TEST_HOST_LOOP_INTERVAL;
427 bool mode_selected = false;
428 uint64_t phys_offset = 0;
432 unsigned int host_ipa_limit;
435 #ifdef USE_CLEAR_DIRTY_LOG
436 if (!kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2)) {
437 fprintf(stderr, "KVM_CLEAR_DIRTY_LOG not available, skipping tests\n");
443 vm_guest_mode_params_init(VM_MODE_P52V48_4K, true, true);
446 vm_guest_mode_params_init(VM_MODE_P40V48_4K, true, true);
447 vm_guest_mode_params_init(VM_MODE_P40V48_64K, true, true);
449 host_ipa_limit = kvm_check_cap(KVM_CAP_ARM_VM_IPA_SIZE);
450 if (host_ipa_limit >= 52)
451 vm_guest_mode_params_init(VM_MODE_P52V48_64K, true, true);
452 if (host_ipa_limit >= 48) {
453 vm_guest_mode_params_init(VM_MODE_P48V48_4K, true, true);
454 vm_guest_mode_params_init(VM_MODE_P48V48_64K, true, true);
458 while ((opt = getopt(argc, argv, "hi:I:p:m:")) != -1) {
461 iterations = strtol(optarg, NULL, 10);
464 interval = strtol(optarg, NULL, 10);
467 phys_offset = strtoull(optarg, NULL, 0);
470 if (!mode_selected) {
471 for (i = 0; i < NUM_VM_MODES; ++i)
472 vm_guest_mode_params[i].enabled = false;
473 mode_selected = true;
475 mode = strtoul(optarg, NULL, 10);
476 TEST_ASSERT(mode < NUM_VM_MODES,
477 "Guest mode ID %d too big", mode);
478 vm_guest_mode_params[mode].enabled = true;
487 TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
488 TEST_ASSERT(interval > 0, "Interval must be greater than zero");
490 DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
491 iterations, interval);
495 for (i = 0; i < NUM_VM_MODES; ++i) {
496 if (!vm_guest_mode_params[i].enabled)
498 TEST_ASSERT(vm_guest_mode_params[i].supported,
499 "Guest mode ID %d (%s) not supported.",
500 i, vm_guest_mode_string(i));
501 run_test(i, iterations, interval, phys_offset);