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 * GPA offset of the testing memory slot. Must be bigger than
55 * DEFAULT_GUEST_PHY_PAGES.
56 */
57 static uint64_t guest_test_mem = DEFAULT_GUEST_TEST_MEM;
58
59 /*
60 * Continuously write to the first 8 bytes of a random pages within
61 * the testing memory region.
62 */
guest_code(void)63 static void guest_code(void)
64 {
65 int i;
66
67 while (true) {
68 for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
69 uint64_t addr = guest_test_mem;
70 addr += (READ_ONCE(random_array[i]) % guest_num_pages)
71 * guest_page_size;
72 addr &= ~(host_page_size - 1);
73 *(uint64_t *)addr = READ_ONCE(iteration);
74 }
75
76 /* Tell the host that we need more random numbers */
77 GUEST_SYNC(1);
78 }
79 }
80
81 /* Host variables */
82 static bool host_quit;
83
84 /* Points to the test VM memory region on which we track dirty logs */
85 static void *host_test_mem;
86 static uint64_t host_num_pages;
87
88 /* For statistics only */
89 static uint64_t host_dirty_count;
90 static uint64_t host_clear_count;
91 static uint64_t host_track_next_count;
92
93 /*
94 * We use this bitmap to track some pages that should have its dirty
95 * bit set in the _next_ iteration. For example, if we detected the
96 * page value changed to current iteration but at the same time the
97 * page bit is cleared in the latest bitmap, then the system must
98 * report that write in the next get dirty log call.
99 */
100 static unsigned long *host_bmap_track;
101
generate_random_array(uint64_t * guest_array,uint64_t size)102 static void generate_random_array(uint64_t *guest_array, uint64_t size)
103 {
104 uint64_t i;
105
106 for (i = 0; i < size; i++)
107 guest_array[i] = random();
108 }
109
vcpu_worker(void * data)110 static void *vcpu_worker(void *data)
111 {
112 int ret;
113 struct kvm_vm *vm = data;
114 uint64_t *guest_array;
115 uint64_t pages_count = 0;
116 struct kvm_run *run;
117 struct ucall uc;
118
119 run = vcpu_state(vm, VCPU_ID);
120
121 guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
122 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
123
124 while (!READ_ONCE(host_quit)) {
125 /* Let the guest dirty the random pages */
126 ret = _vcpu_run(vm, VCPU_ID);
127 if (get_ucall(vm, VCPU_ID, &uc) == UCALL_SYNC) {
128 pages_count += TEST_PAGES_PER_LOOP;
129 generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
130 } else {
131 TEST_ASSERT(false,
132 "Invalid guest sync status: "
133 "exit_reason=%s\n",
134 exit_reason_str(run->exit_reason));
135 }
136 }
137
138 DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
139
140 return NULL;
141 }
142
vm_dirty_log_verify(unsigned long * bmap)143 static void vm_dirty_log_verify(unsigned long *bmap)
144 {
145 uint64_t page;
146 uint64_t *value_ptr;
147 uint64_t step = host_page_size >= guest_page_size ? 1 :
148 guest_page_size / host_page_size;
149
150 for (page = 0; page < host_num_pages; page += step) {
151 value_ptr = host_test_mem + page * host_page_size;
152
153 /* If this is a special page that we were tracking... */
154 if (test_and_clear_bit(page, host_bmap_track)) {
155 host_track_next_count++;
156 TEST_ASSERT(test_bit(page, bmap),
157 "Page %"PRIu64" should have its dirty bit "
158 "set in this iteration but it is missing",
159 page);
160 }
161
162 if (test_bit(page, bmap)) {
163 host_dirty_count++;
164 /*
165 * If the bit is set, the value written onto
166 * the corresponding page should be either the
167 * previous iteration number or the current one.
168 */
169 TEST_ASSERT(*value_ptr == iteration ||
170 *value_ptr == iteration - 1,
171 "Set page %"PRIu64" value %"PRIu64
172 " incorrect (iteration=%"PRIu64")",
173 page, *value_ptr, iteration);
174 } else {
175 host_clear_count++;
176 /*
177 * If cleared, the value written can be any
178 * value smaller or equals to the iteration
179 * number. Note that the value can be exactly
180 * (iteration-1) if that write can happen
181 * like this:
182 *
183 * (1) increase loop count to "iteration-1"
184 * (2) write to page P happens (with value
185 * "iteration-1")
186 * (3) get dirty log for "iteration-1"; we'll
187 * see that page P bit is set (dirtied),
188 * and not set the bit in host_bmap_track
189 * (4) increase loop count to "iteration"
190 * (which is current iteration)
191 * (5) get dirty log for current iteration,
192 * we'll see that page P is cleared, with
193 * value "iteration-1".
194 */
195 TEST_ASSERT(*value_ptr <= iteration,
196 "Clear page %"PRIu64" value %"PRIu64
197 " incorrect (iteration=%"PRIu64")",
198 page, *value_ptr, iteration);
199 if (*value_ptr == iteration) {
200 /*
201 * This page is _just_ modified; it
202 * should report its dirtyness in the
203 * next run
204 */
205 set_bit(page, host_bmap_track);
206 }
207 }
208 }
209 }
210
create_vm(enum vm_guest_mode mode,uint32_t vcpuid,uint64_t extra_mem_pages,void * guest_code)211 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
212 uint64_t extra_mem_pages, void *guest_code)
213 {
214 struct kvm_vm *vm;
215 uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
216
217 vm = vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
218 kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
219 #ifdef __x86_64__
220 vm_create_irqchip(vm);
221 #endif
222 vm_vcpu_add_default(vm, vcpuid, guest_code);
223 return vm;
224 }
225
run_test(enum vm_guest_mode mode,unsigned long iterations,unsigned long interval,bool top_offset)226 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
227 unsigned long interval, bool top_offset)
228 {
229 unsigned int guest_pa_bits, guest_page_shift;
230 pthread_t vcpu_thread;
231 struct kvm_vm *vm;
232 uint64_t max_gfn;
233 unsigned long *bmap;
234
235 switch (mode) {
236 case VM_MODE_P52V48_4K:
237 guest_pa_bits = 52;
238 guest_page_shift = 12;
239 break;
240 case VM_MODE_P52V48_64K:
241 guest_pa_bits = 52;
242 guest_page_shift = 16;
243 break;
244 case VM_MODE_P40V48_4K:
245 guest_pa_bits = 40;
246 guest_page_shift = 12;
247 break;
248 case VM_MODE_P40V48_64K:
249 guest_pa_bits = 40;
250 guest_page_shift = 16;
251 break;
252 default:
253 TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", mode);
254 }
255
256 DEBUG("Testing guest mode: %s\n", vm_guest_mode_string(mode));
257
258 max_gfn = (1ul << (guest_pa_bits - guest_page_shift)) - 1;
259 guest_page_size = (1ul << guest_page_shift);
260 /* 1G of guest page sized pages */
261 guest_num_pages = (1ul << (30 - guest_page_shift));
262 host_page_size = getpagesize();
263 host_num_pages = (guest_num_pages * guest_page_size) / host_page_size +
264 !!((guest_num_pages * guest_page_size) % host_page_size);
265
266 if (top_offset) {
267 guest_test_mem = (max_gfn - guest_num_pages) * guest_page_size;
268 guest_test_mem &= ~(host_page_size - 1);
269 }
270
271 DEBUG("guest test mem offset: 0x%lx\n", guest_test_mem);
272
273 bmap = bitmap_alloc(host_num_pages);
274 host_bmap_track = bitmap_alloc(host_num_pages);
275
276 vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code);
277
278 /* Add an extra memory slot for testing dirty logging */
279 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
280 guest_test_mem,
281 TEST_MEM_SLOT_INDEX,
282 guest_num_pages,
283 KVM_MEM_LOG_DIRTY_PAGES);
284
285 /* Do 1:1 mapping for the dirty track memory slot */
286 virt_map(vm, guest_test_mem, guest_test_mem,
287 guest_num_pages * guest_page_size, 0);
288
289 /* Cache the HVA pointer of the region */
290 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_mem);
291
292 #ifdef __x86_64__
293 vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
294 #endif
295 #ifdef __aarch64__
296 ucall_init(vm, UCALL_MMIO, NULL);
297 #endif
298
299 /* Export the shared variables to the guest */
300 sync_global_to_guest(vm, host_page_size);
301 sync_global_to_guest(vm, guest_page_size);
302 sync_global_to_guest(vm, guest_test_mem);
303 sync_global_to_guest(vm, guest_num_pages);
304
305 /* Start the iterations */
306 iteration = 1;
307 sync_global_to_guest(vm, iteration);
308 host_quit = false;
309 host_dirty_count = 0;
310 host_clear_count = 0;
311 host_track_next_count = 0;
312
313 pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
314
315 while (iteration < iterations) {
316 /* Give the vcpu thread some time to dirty some pages */
317 usleep(interval * 1000);
318 kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
319 vm_dirty_log_verify(bmap);
320 iteration++;
321 sync_global_to_guest(vm, iteration);
322 }
323
324 /* Tell the vcpu thread to quit */
325 host_quit = true;
326 pthread_join(vcpu_thread, NULL);
327
328 DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
329 "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
330 host_track_next_count);
331
332 free(bmap);
333 free(host_bmap_track);
334 ucall_uninit(vm);
335 kvm_vm_free(vm);
336 }
337
338 static struct vm_guest_modes {
339 enum vm_guest_mode mode;
340 bool supported;
341 bool enabled;
342 } vm_guest_modes[NUM_VM_MODES] = {
343 #if defined(__x86_64__)
344 { VM_MODE_P52V48_4K, 1, 1, },
345 { VM_MODE_P52V48_64K, 0, 0, },
346 { VM_MODE_P40V48_4K, 0, 0, },
347 { VM_MODE_P40V48_64K, 0, 0, },
348 #elif defined(__aarch64__)
349 { VM_MODE_P52V48_4K, 0, 0, },
350 { VM_MODE_P52V48_64K, 0, 0, },
351 { VM_MODE_P40V48_4K, 1, 1, },
352 { VM_MODE_P40V48_64K, 1, 1, },
353 #endif
354 };
355
help(char * name)356 static void help(char *name)
357 {
358 int i;
359
360 puts("");
361 printf("usage: %s [-h] [-i iterations] [-I interval] "
362 "[-o offset] [-t] [-m mode]\n", name);
363 puts("");
364 printf(" -i: specify iteration counts (default: %"PRIu64")\n",
365 TEST_HOST_LOOP_N);
366 printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
367 TEST_HOST_LOOP_INTERVAL);
368 printf(" -o: guest test memory offset (default: 0x%lx)\n",
369 DEFAULT_GUEST_TEST_MEM);
370 printf(" -t: map guest test memory at the top of the allowed "
371 "physical address range\n");
372 printf(" -m: specify the guest mode ID to test "
373 "(default: test all supported modes)\n"
374 " This option may be used multiple times.\n"
375 " Guest mode IDs:\n");
376 for (i = 0; i < NUM_VM_MODES; ++i) {
377 printf(" %d: %s%s\n",
378 vm_guest_modes[i].mode,
379 vm_guest_mode_string(vm_guest_modes[i].mode),
380 vm_guest_modes[i].supported ? " (supported)" : "");
381 }
382 puts("");
383 exit(0);
384 }
385
main(int argc,char * argv[])386 int main(int argc, char *argv[])
387 {
388 unsigned long iterations = TEST_HOST_LOOP_N;
389 unsigned long interval = TEST_HOST_LOOP_INTERVAL;
390 bool mode_selected = false;
391 bool top_offset = false;
392 unsigned int mode;
393 int opt, i;
394
395 while ((opt = getopt(argc, argv, "hi:I:o:tm:")) != -1) {
396 switch (opt) {
397 case 'i':
398 iterations = strtol(optarg, NULL, 10);
399 break;
400 case 'I':
401 interval = strtol(optarg, NULL, 10);
402 break;
403 case 'o':
404 guest_test_mem = strtoull(optarg, NULL, 0);
405 break;
406 case 't':
407 top_offset = true;
408 break;
409 case 'm':
410 if (!mode_selected) {
411 for (i = 0; i < NUM_VM_MODES; ++i)
412 vm_guest_modes[i].enabled = 0;
413 mode_selected = true;
414 }
415 mode = strtoul(optarg, NULL, 10);
416 TEST_ASSERT(mode < NUM_VM_MODES,
417 "Guest mode ID %d too big", mode);
418 vm_guest_modes[mode].enabled = 1;
419 break;
420 case 'h':
421 default:
422 help(argv[0]);
423 break;
424 }
425 }
426
427 TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
428 TEST_ASSERT(interval > 0, "Interval must be greater than zero");
429 TEST_ASSERT(!top_offset || guest_test_mem == DEFAULT_GUEST_TEST_MEM,
430 "Cannot use both -o [offset] and -t at the same time");
431
432 DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
433 iterations, interval);
434
435 srandom(time(0));
436
437 for (i = 0; i < NUM_VM_MODES; ++i) {
438 if (!vm_guest_modes[i].enabled)
439 continue;
440 TEST_ASSERT(vm_guest_modes[i].supported,
441 "Guest mode ID %d (%s) not supported.",
442 vm_guest_modes[i].mode,
443 vm_guest_mode_string(vm_guest_modes[i].mode));
444 run_test(vm_guest_modes[i].mode, iterations, interval, top_offset);
445 }
446
447 return 0;
448 }
449