Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * KVM page table test
4 *
5 * Copyright (C) 2021, Huawei, Inc.
6 *
7 * Make sure that THP has been enabled or enough HUGETLB pages with specific
8 * page size have been pre-allocated on your system, if you are planning to
9 * use hugepages to back the guest memory for testing.
10 */
11
12#define _GNU_SOURCE /* for program_invocation_name */
13
14#include <stdio.h>
15#include <stdlib.h>
16#include <time.h>
17#include <pthread.h>
18#include <semaphore.h>
19
20#include "test_util.h"
21#include "kvm_util.h"
22#include "processor.h"
23#include "guest_modes.h"
24
25#define TEST_MEM_SLOT_INDEX 1
26
27/* Default size(1GB) of the memory for testing */
28#define DEFAULT_TEST_MEM_SIZE (1 << 30)
29
30/* Default guest test virtual memory offset */
31#define DEFAULT_GUEST_TEST_MEM 0xc0000000
32
33/* Different guest memory accessing stages */
34enum test_stage {
35 KVM_BEFORE_MAPPINGS,
36 KVM_CREATE_MAPPINGS,
37 KVM_UPDATE_MAPPINGS,
38 KVM_ADJUST_MAPPINGS,
39 NUM_TEST_STAGES,
40};
41
42static const char * const test_stage_string[] = {
43 "KVM_BEFORE_MAPPINGS",
44 "KVM_CREATE_MAPPINGS",
45 "KVM_UPDATE_MAPPINGS",
46 "KVM_ADJUST_MAPPINGS",
47};
48
49struct test_args {
50 struct kvm_vm *vm;
51 uint64_t guest_test_virt_mem;
52 uint64_t host_page_size;
53 uint64_t host_num_pages;
54 uint64_t large_page_size;
55 uint64_t large_num_pages;
56 uint64_t host_pages_per_lpage;
57 enum vm_mem_backing_src_type src_type;
58 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
59};
60
61/*
62 * Guest variables. Use addr_gva2hva() if these variables need
63 * to be changed in host.
64 */
65static enum test_stage guest_test_stage;
66
67/* Host variables */
68static uint32_t nr_vcpus = 1;
69static struct test_args test_args;
70static enum test_stage *current_stage;
71static bool host_quit;
72
73/* Whether the test stage is updated, or completed */
74static sem_t test_stage_updated;
75static sem_t test_stage_completed;
76
77/*
78 * Guest physical memory offset of the testing memory slot.
79 * This will be set to the topmost valid physical address minus
80 * the test memory size.
81 */
82static uint64_t guest_test_phys_mem;
83
84/*
85 * Guest virtual memory offset of the testing memory slot.
86 * Must not conflict with identity mapped test code.
87 */
88static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
89
90static void guest_code(bool do_write)
91{
92 struct test_args *p = &test_args;
93 enum test_stage *current_stage = &guest_test_stage;
94 uint64_t addr;
95 int i, j;
96
97 while (true) {
98 addr = p->guest_test_virt_mem;
99
100 switch (READ_ONCE(*current_stage)) {
101 /*
102 * All vCPU threads will be started in this stage,
103 * where guest code of each vCPU will do nothing.
104 */
105 case KVM_BEFORE_MAPPINGS:
106 break;
107
108 /*
109 * Before dirty logging, vCPUs concurrently access the first
110 * 8 bytes of each page (host page/large page) within the same
111 * memory region with different accessing types (read/write).
112 * Then KVM will create normal page mappings or huge block
113 * mappings for them.
114 */
115 case KVM_CREATE_MAPPINGS:
116 for (i = 0; i < p->large_num_pages; i++) {
117 if (do_write)
118 *(uint64_t *)addr = 0x0123456789ABCDEF;
119 else
120 READ_ONCE(*(uint64_t *)addr);
121
122 addr += p->large_page_size;
123 }
124 break;
125
126 /*
127 * During dirty logging, KVM will only update attributes of the
128 * normal page mappings from RO to RW if memory backing src type
129 * is anonymous. In other cases, KVM will split the huge block
130 * mappings into normal page mappings if memory backing src type
131 * is THP or HUGETLB.
132 */
133 case KVM_UPDATE_MAPPINGS:
134 if (p->src_type == VM_MEM_SRC_ANONYMOUS) {
135 for (i = 0; i < p->host_num_pages; i++) {
136 *(uint64_t *)addr = 0x0123456789ABCDEF;
137 addr += p->host_page_size;
138 }
139 break;
140 }
141
142 for (i = 0; i < p->large_num_pages; i++) {
143 /*
144 * Write to the first host page in each large
145 * page region, and triger break of large pages.
146 */
147 *(uint64_t *)addr = 0x0123456789ABCDEF;
148
149 /*
150 * Access the middle host pages in each large
151 * page region. Since dirty logging is enabled,
152 * this will create new mappings at the smallest
153 * granularity.
154 */
155 addr += p->large_page_size / 2;
156 for (j = 0; j < p->host_pages_per_lpage / 2; j++) {
157 READ_ONCE(*(uint64_t *)addr);
158 addr += p->host_page_size;
159 }
160 }
161 break;
162
163 /*
164 * After dirty logging is stopped, vCPUs concurrently read
165 * from every single host page. Then KVM will coalesce the
166 * split page mappings back to block mappings. And a TLB
167 * conflict abort could occur here if TLB entries of the
168 * page mappings are not fully invalidated.
169 */
170 case KVM_ADJUST_MAPPINGS:
171 for (i = 0; i < p->host_num_pages; i++) {
172 READ_ONCE(*(uint64_t *)addr);
173 addr += p->host_page_size;
174 }
175 break;
176
177 default:
178 GUEST_ASSERT(0);
179 }
180
181 GUEST_SYNC(1);
182 }
183}
184
185static void *vcpu_worker(void *data)
186{
187 struct kvm_vcpu *vcpu = data;
188 bool do_write = !(vcpu->id % 2);
189 struct timespec start;
190 struct timespec ts_diff;
191 enum test_stage stage;
192 int ret;
193
194 vcpu_args_set(vcpu, 1, do_write);
195
196 while (!READ_ONCE(host_quit)) {
197 ret = sem_wait(&test_stage_updated);
198 TEST_ASSERT(ret == 0, "Error in sem_wait");
199
200 if (READ_ONCE(host_quit))
201 return NULL;
202
203 clock_gettime(CLOCK_MONOTONIC_RAW, &start);
204 ret = _vcpu_run(vcpu);
205 ts_diff = timespec_elapsed(start);
206
207 TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
208 TEST_ASSERT(get_ucall(vcpu, NULL) == UCALL_SYNC,
209 "Invalid guest sync status: exit_reason=%s\n",
210 exit_reason_str(vcpu->run->exit_reason));
211
212 pr_debug("Got sync event from vCPU %d\n", vcpu->id);
213 stage = READ_ONCE(*current_stage);
214
215 /*
216 * Here we can know the execution time of every
217 * single vcpu running in different test stages.
218 */
219 pr_debug("vCPU %d has completed stage %s\n"
220 "execution time is: %ld.%.9lds\n\n",
221 vcpu->id, test_stage_string[stage],
222 ts_diff.tv_sec, ts_diff.tv_nsec);
223
224 ret = sem_post(&test_stage_completed);
225 TEST_ASSERT(ret == 0, "Error in sem_post");
226 }
227
228 return NULL;
229}
230
231struct test_params {
232 uint64_t phys_offset;
233 uint64_t test_mem_size;
234 enum vm_mem_backing_src_type src_type;
235};
236
237static struct kvm_vm *pre_init_before_test(enum vm_guest_mode mode, void *arg)
238{
239 int ret;
240 struct test_params *p = arg;
241 enum vm_mem_backing_src_type src_type = p->src_type;
242 uint64_t large_page_size = get_backing_src_pagesz(src_type);
243 uint64_t guest_page_size = vm_guest_mode_params[mode].page_size;
244 uint64_t host_page_size = getpagesize();
245 uint64_t test_mem_size = p->test_mem_size;
246 uint64_t guest_num_pages;
247 uint64_t alignment;
248 void *host_test_mem;
249 struct kvm_vm *vm;
250
251 /* Align up the test memory size */
252 alignment = max(large_page_size, guest_page_size);
253 test_mem_size = (test_mem_size + alignment - 1) & ~(alignment - 1);
254
255 /* Create a VM with enough guest pages */
256 guest_num_pages = test_mem_size / guest_page_size;
257 vm = __vm_create_with_vcpus(mode, nr_vcpus, guest_num_pages,
258 guest_code, test_args.vcpus);
259
260 /* Align down GPA of the testing memslot */
261 if (!p->phys_offset)
262 guest_test_phys_mem = (vm->max_gfn - guest_num_pages) *
263 guest_page_size;
264 else
265 guest_test_phys_mem = p->phys_offset;
266#ifdef __s390x__
267 alignment = max(0x100000UL, alignment);
268#endif
269 guest_test_phys_mem = align_down(guest_test_phys_mem, alignment);
270
271 /* Set up the shared data structure test_args */
272 test_args.vm = vm;
273 test_args.guest_test_virt_mem = guest_test_virt_mem;
274 test_args.host_page_size = host_page_size;
275 test_args.host_num_pages = test_mem_size / host_page_size;
276 test_args.large_page_size = large_page_size;
277 test_args.large_num_pages = test_mem_size / large_page_size;
278 test_args.host_pages_per_lpage = large_page_size / host_page_size;
279 test_args.src_type = src_type;
280
281 /* Add an extra memory slot with specified backing src type */
282 vm_userspace_mem_region_add(vm, src_type, guest_test_phys_mem,
283 TEST_MEM_SLOT_INDEX, guest_num_pages, 0);
284
285 /* Do mapping(GVA->GPA) for the testing memory slot */
286 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
287
288 /* Cache the HVA pointer of the region */
289 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
290
291 /* Export shared structure test_args to guest */
292 sync_global_to_guest(vm, test_args);
293
294 ret = sem_init(&test_stage_updated, 0, 0);
295 TEST_ASSERT(ret == 0, "Error in sem_init");
296
297 ret = sem_init(&test_stage_completed, 0, 0);
298 TEST_ASSERT(ret == 0, "Error in sem_init");
299
300 current_stage = addr_gva2hva(vm, (vm_vaddr_t)(&guest_test_stage));
301 *current_stage = NUM_TEST_STAGES;
302
303 pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
304 pr_info("Testing memory backing src type: %s\n",
305 vm_mem_backing_src_alias(src_type)->name);
306 pr_info("Testing memory backing src granularity: 0x%lx\n",
307 large_page_size);
308 pr_info("Testing memory size(aligned): 0x%lx\n", test_mem_size);
309 pr_info("Guest physical test memory offset: 0x%lx\n",
310 guest_test_phys_mem);
311 pr_info("Host virtual test memory offset: 0x%lx\n",
312 (uint64_t)host_test_mem);
313 pr_info("Number of testing vCPUs: %d\n", nr_vcpus);
314
315 return vm;
316}
317
318static void vcpus_complete_new_stage(enum test_stage stage)
319{
320 int ret;
321 int vcpus;
322
323 /* Wake up all the vcpus to run new test stage */
324 for (vcpus = 0; vcpus < nr_vcpus; vcpus++) {
325 ret = sem_post(&test_stage_updated);
326 TEST_ASSERT(ret == 0, "Error in sem_post");
327 }
328 pr_debug("All vcpus have been notified to continue\n");
329
330 /* Wait for all the vcpus to complete new test stage */
331 for (vcpus = 0; vcpus < nr_vcpus; vcpus++) {
332 ret = sem_wait(&test_stage_completed);
333 TEST_ASSERT(ret == 0, "Error in sem_wait");
334
335 pr_debug("%d vcpus have completed stage %s\n",
336 vcpus + 1, test_stage_string[stage]);
337 }
338
339 pr_debug("All vcpus have completed stage %s\n",
340 test_stage_string[stage]);
341}
342
343static void run_test(enum vm_guest_mode mode, void *arg)
344{
345 pthread_t *vcpu_threads;
346 struct kvm_vm *vm;
347 struct timespec start;
348 struct timespec ts_diff;
349 int ret, i;
350
351 /* Create VM with vCPUs and make some pre-initialization */
352 vm = pre_init_before_test(mode, arg);
353
354 vcpu_threads = malloc(nr_vcpus * sizeof(*vcpu_threads));
355 TEST_ASSERT(vcpu_threads, "Memory allocation failed");
356
357 host_quit = false;
358 *current_stage = KVM_BEFORE_MAPPINGS;
359
360 for (i = 0; i < nr_vcpus; i++)
361 pthread_create(&vcpu_threads[i], NULL, vcpu_worker,
362 test_args.vcpus[i]);
363
364 vcpus_complete_new_stage(*current_stage);
365 pr_info("Started all vCPUs successfully\n");
366
367 /* Test the stage of KVM creating mappings */
368 *current_stage = KVM_CREATE_MAPPINGS;
369
370 clock_gettime(CLOCK_MONOTONIC_RAW, &start);
371 vcpus_complete_new_stage(*current_stage);
372 ts_diff = timespec_elapsed(start);
373
374 pr_info("KVM_CREATE_MAPPINGS: total execution time: %ld.%.9lds\n\n",
375 ts_diff.tv_sec, ts_diff.tv_nsec);
376
377 /* Test the stage of KVM updating mappings */
378 vm_mem_region_set_flags(vm, TEST_MEM_SLOT_INDEX,
379 KVM_MEM_LOG_DIRTY_PAGES);
380
381 *current_stage = KVM_UPDATE_MAPPINGS;
382
383 clock_gettime(CLOCK_MONOTONIC_RAW, &start);
384 vcpus_complete_new_stage(*current_stage);
385 ts_diff = timespec_elapsed(start);
386
387 pr_info("KVM_UPDATE_MAPPINGS: total execution time: %ld.%.9lds\n\n",
388 ts_diff.tv_sec, ts_diff.tv_nsec);
389
390 /* Test the stage of KVM adjusting mappings */
391 vm_mem_region_set_flags(vm, TEST_MEM_SLOT_INDEX, 0);
392
393 *current_stage = KVM_ADJUST_MAPPINGS;
394
395 clock_gettime(CLOCK_MONOTONIC_RAW, &start);
396 vcpus_complete_new_stage(*current_stage);
397 ts_diff = timespec_elapsed(start);
398
399 pr_info("KVM_ADJUST_MAPPINGS: total execution time: %ld.%.9lds\n\n",
400 ts_diff.tv_sec, ts_diff.tv_nsec);
401
402 /* Tell the vcpu thread to quit */
403 host_quit = true;
404 for (i = 0; i < nr_vcpus; i++) {
405 ret = sem_post(&test_stage_updated);
406 TEST_ASSERT(ret == 0, "Error in sem_post");
407 }
408
409 for (i = 0; i < nr_vcpus; i++)
410 pthread_join(vcpu_threads[i], NULL);
411
412 ret = sem_destroy(&test_stage_updated);
413 TEST_ASSERT(ret == 0, "Error in sem_destroy");
414
415 ret = sem_destroy(&test_stage_completed);
416 TEST_ASSERT(ret == 0, "Error in sem_destroy");
417
418 free(vcpu_threads);
419 kvm_vm_free(vm);
420}
421
422static void help(char *name)
423{
424 puts("");
425 printf("usage: %s [-h] [-p offset] [-m mode] "
426 "[-b mem-size] [-v vcpus] [-s mem-type]\n", name);
427 puts("");
428 printf(" -p: specify guest physical test memory offset\n"
429 " Warning: a low offset can conflict with the loaded test code.\n");
430 guest_modes_help();
431 printf(" -b: specify size of the memory region for testing. e.g. 10M or 3G.\n"
432 " (default: 1G)\n");
433 printf(" -v: specify the number of vCPUs to run\n"
434 " (default: 1)\n");
435 backing_src_help("-s");
436 puts("");
437}
438
439int main(int argc, char *argv[])
440{
441 int max_vcpus = kvm_check_cap(KVM_CAP_MAX_VCPUS);
442 struct test_params p = {
443 .test_mem_size = DEFAULT_TEST_MEM_SIZE,
444 .src_type = DEFAULT_VM_MEM_SRC,
445 };
446 int opt;
447
448 guest_modes_append_default();
449
450 while ((opt = getopt(argc, argv, "hp:m:b:v:s:")) != -1) {
451 switch (opt) {
452 case 'p':
453 p.phys_offset = strtoull(optarg, NULL, 0);
454 break;
455 case 'm':
456 guest_modes_cmdline(optarg);
457 break;
458 case 'b':
459 p.test_mem_size = parse_size(optarg);
460 break;
461 case 'v':
462 nr_vcpus = atoi_positive("Number of vCPUs", optarg);
463 TEST_ASSERT(nr_vcpus <= max_vcpus,
464 "Invalid number of vcpus, must be between 1 and %d", max_vcpus);
465 break;
466 case 's':
467 p.src_type = parse_backing_src_type(optarg);
468 break;
469 case 'h':
470 default:
471 help(argv[0]);
472 exit(0);
473 }
474 }
475
476 for_each_guest_mode(run_test, &p);
477
478 return 0;
479}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * KVM page table test
4 *
5 * Copyright (C) 2021, Huawei, Inc.
6 *
7 * Make sure that THP has been enabled or enough HUGETLB pages with specific
8 * page size have been pre-allocated on your system, if you are planning to
9 * use hugepages to back the guest memory for testing.
10 */
11#include <stdio.h>
12#include <stdlib.h>
13#include <time.h>
14#include <pthread.h>
15#include <semaphore.h>
16
17#include "test_util.h"
18#include "kvm_util.h"
19#include "processor.h"
20#include "guest_modes.h"
21#include "ucall_common.h"
22
23#define TEST_MEM_SLOT_INDEX 1
24
25/* Default size(1GB) of the memory for testing */
26#define DEFAULT_TEST_MEM_SIZE (1 << 30)
27
28/* Default guest test virtual memory offset */
29#define DEFAULT_GUEST_TEST_MEM 0xc0000000
30
31/* Different guest memory accessing stages */
32enum test_stage {
33 KVM_BEFORE_MAPPINGS,
34 KVM_CREATE_MAPPINGS,
35 KVM_UPDATE_MAPPINGS,
36 KVM_ADJUST_MAPPINGS,
37 NUM_TEST_STAGES,
38};
39
40static const char * const test_stage_string[] = {
41 "KVM_BEFORE_MAPPINGS",
42 "KVM_CREATE_MAPPINGS",
43 "KVM_UPDATE_MAPPINGS",
44 "KVM_ADJUST_MAPPINGS",
45};
46
47struct test_args {
48 struct kvm_vm *vm;
49 uint64_t guest_test_virt_mem;
50 uint64_t host_page_size;
51 uint64_t host_num_pages;
52 uint64_t large_page_size;
53 uint64_t large_num_pages;
54 uint64_t host_pages_per_lpage;
55 enum vm_mem_backing_src_type src_type;
56 struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
57};
58
59/*
60 * Guest variables. Use addr_gva2hva() if these variables need
61 * to be changed in host.
62 */
63static enum test_stage guest_test_stage;
64
65/* Host variables */
66static uint32_t nr_vcpus = 1;
67static struct test_args test_args;
68static enum test_stage *current_stage;
69static bool host_quit;
70
71/* Whether the test stage is updated, or completed */
72static sem_t test_stage_updated;
73static sem_t test_stage_completed;
74
75/*
76 * Guest physical memory offset of the testing memory slot.
77 * This will be set to the topmost valid physical address minus
78 * the test memory size.
79 */
80static uint64_t guest_test_phys_mem;
81
82/*
83 * Guest virtual memory offset of the testing memory slot.
84 * Must not conflict with identity mapped test code.
85 */
86static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
87
88static void guest_code(bool do_write)
89{
90 struct test_args *p = &test_args;
91 enum test_stage *current_stage = &guest_test_stage;
92 uint64_t addr;
93 int i, j;
94
95 while (true) {
96 addr = p->guest_test_virt_mem;
97
98 switch (READ_ONCE(*current_stage)) {
99 /*
100 * All vCPU threads will be started in this stage,
101 * where guest code of each vCPU will do nothing.
102 */
103 case KVM_BEFORE_MAPPINGS:
104 break;
105
106 /*
107 * Before dirty logging, vCPUs concurrently access the first
108 * 8 bytes of each page (host page/large page) within the same
109 * memory region with different accessing types (read/write).
110 * Then KVM will create normal page mappings or huge block
111 * mappings for them.
112 */
113 case KVM_CREATE_MAPPINGS:
114 for (i = 0; i < p->large_num_pages; i++) {
115 if (do_write)
116 *(uint64_t *)addr = 0x0123456789ABCDEF;
117 else
118 READ_ONCE(*(uint64_t *)addr);
119
120 addr += p->large_page_size;
121 }
122 break;
123
124 /*
125 * During dirty logging, KVM will only update attributes of the
126 * normal page mappings from RO to RW if memory backing src type
127 * is anonymous. In other cases, KVM will split the huge block
128 * mappings into normal page mappings if memory backing src type
129 * is THP or HUGETLB.
130 */
131 case KVM_UPDATE_MAPPINGS:
132 if (p->src_type == VM_MEM_SRC_ANONYMOUS) {
133 for (i = 0; i < p->host_num_pages; i++) {
134 *(uint64_t *)addr = 0x0123456789ABCDEF;
135 addr += p->host_page_size;
136 }
137 break;
138 }
139
140 for (i = 0; i < p->large_num_pages; i++) {
141 /*
142 * Write to the first host page in each large
143 * page region, and triger break of large pages.
144 */
145 *(uint64_t *)addr = 0x0123456789ABCDEF;
146
147 /*
148 * Access the middle host pages in each large
149 * page region. Since dirty logging is enabled,
150 * this will create new mappings at the smallest
151 * granularity.
152 */
153 addr += p->large_page_size / 2;
154 for (j = 0; j < p->host_pages_per_lpage / 2; j++) {
155 READ_ONCE(*(uint64_t *)addr);
156 addr += p->host_page_size;
157 }
158 }
159 break;
160
161 /*
162 * After dirty logging is stopped, vCPUs concurrently read
163 * from every single host page. Then KVM will coalesce the
164 * split page mappings back to block mappings. And a TLB
165 * conflict abort could occur here if TLB entries of the
166 * page mappings are not fully invalidated.
167 */
168 case KVM_ADJUST_MAPPINGS:
169 for (i = 0; i < p->host_num_pages; i++) {
170 READ_ONCE(*(uint64_t *)addr);
171 addr += p->host_page_size;
172 }
173 break;
174
175 default:
176 GUEST_ASSERT(0);
177 }
178
179 GUEST_SYNC(1);
180 }
181}
182
183static void *vcpu_worker(void *data)
184{
185 struct kvm_vcpu *vcpu = data;
186 bool do_write = !(vcpu->id % 2);
187 struct timespec start;
188 struct timespec ts_diff;
189 enum test_stage stage;
190 int ret;
191
192 vcpu_args_set(vcpu, 1, do_write);
193
194 while (!READ_ONCE(host_quit)) {
195 ret = sem_wait(&test_stage_updated);
196 TEST_ASSERT(ret == 0, "Error in sem_wait");
197
198 if (READ_ONCE(host_quit))
199 return NULL;
200
201 clock_gettime(CLOCK_MONOTONIC, &start);
202 ret = _vcpu_run(vcpu);
203 ts_diff = timespec_elapsed(start);
204
205 TEST_ASSERT(ret == 0, "vcpu_run failed: %d", ret);
206 TEST_ASSERT(get_ucall(vcpu, NULL) == UCALL_SYNC,
207 "Invalid guest sync status: exit_reason=%s",
208 exit_reason_str(vcpu->run->exit_reason));
209
210 pr_debug("Got sync event from vCPU %d\n", vcpu->id);
211 stage = READ_ONCE(*current_stage);
212
213 /*
214 * Here we can know the execution time of every
215 * single vcpu running in different test stages.
216 */
217 pr_debug("vCPU %d has completed stage %s\n"
218 "execution time is: %ld.%.9lds\n\n",
219 vcpu->id, test_stage_string[stage],
220 ts_diff.tv_sec, ts_diff.tv_nsec);
221
222 ret = sem_post(&test_stage_completed);
223 TEST_ASSERT(ret == 0, "Error in sem_post");
224 }
225
226 return NULL;
227}
228
229struct test_params {
230 uint64_t phys_offset;
231 uint64_t test_mem_size;
232 enum vm_mem_backing_src_type src_type;
233};
234
235static struct kvm_vm *pre_init_before_test(enum vm_guest_mode mode, void *arg)
236{
237 int ret;
238 struct test_params *p = arg;
239 enum vm_mem_backing_src_type src_type = p->src_type;
240 uint64_t large_page_size = get_backing_src_pagesz(src_type);
241 uint64_t guest_page_size = vm_guest_mode_params[mode].page_size;
242 uint64_t host_page_size = getpagesize();
243 uint64_t test_mem_size = p->test_mem_size;
244 uint64_t guest_num_pages;
245 uint64_t alignment;
246 void *host_test_mem;
247 struct kvm_vm *vm;
248
249 /* Align up the test memory size */
250 alignment = max(large_page_size, guest_page_size);
251 test_mem_size = (test_mem_size + alignment - 1) & ~(alignment - 1);
252
253 /* Create a VM with enough guest pages */
254 guest_num_pages = test_mem_size / guest_page_size;
255 vm = __vm_create_with_vcpus(VM_SHAPE(mode), nr_vcpus, guest_num_pages,
256 guest_code, test_args.vcpus);
257
258 /* Align down GPA of the testing memslot */
259 if (!p->phys_offset)
260 guest_test_phys_mem = (vm->max_gfn - guest_num_pages) *
261 guest_page_size;
262 else
263 guest_test_phys_mem = p->phys_offset;
264#ifdef __s390x__
265 alignment = max(0x100000UL, alignment);
266#endif
267 guest_test_phys_mem = align_down(guest_test_phys_mem, alignment);
268
269 /* Set up the shared data structure test_args */
270 test_args.vm = vm;
271 test_args.guest_test_virt_mem = guest_test_virt_mem;
272 test_args.host_page_size = host_page_size;
273 test_args.host_num_pages = test_mem_size / host_page_size;
274 test_args.large_page_size = large_page_size;
275 test_args.large_num_pages = test_mem_size / large_page_size;
276 test_args.host_pages_per_lpage = large_page_size / host_page_size;
277 test_args.src_type = src_type;
278
279 /* Add an extra memory slot with specified backing src type */
280 vm_userspace_mem_region_add(vm, src_type, guest_test_phys_mem,
281 TEST_MEM_SLOT_INDEX, guest_num_pages, 0);
282
283 /* Do mapping(GVA->GPA) for the testing memory slot */
284 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
285
286 /* Cache the HVA pointer of the region */
287 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
288
289 /* Export shared structure test_args to guest */
290 sync_global_to_guest(vm, test_args);
291
292 ret = sem_init(&test_stage_updated, 0, 0);
293 TEST_ASSERT(ret == 0, "Error in sem_init");
294
295 ret = sem_init(&test_stage_completed, 0, 0);
296 TEST_ASSERT(ret == 0, "Error in sem_init");
297
298 current_stage = addr_gva2hva(vm, (vm_vaddr_t)(&guest_test_stage));
299 *current_stage = NUM_TEST_STAGES;
300
301 pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
302 pr_info("Testing memory backing src type: %s\n",
303 vm_mem_backing_src_alias(src_type)->name);
304 pr_info("Testing memory backing src granularity: 0x%lx\n",
305 large_page_size);
306 pr_info("Testing memory size(aligned): 0x%lx\n", test_mem_size);
307 pr_info("Guest physical test memory offset: 0x%lx\n",
308 guest_test_phys_mem);
309 pr_info("Host virtual test memory offset: 0x%lx\n",
310 (uint64_t)host_test_mem);
311 pr_info("Number of testing vCPUs: %d\n", nr_vcpus);
312
313 return vm;
314}
315
316static void vcpus_complete_new_stage(enum test_stage stage)
317{
318 int ret;
319 int vcpus;
320
321 /* Wake up all the vcpus to run new test stage */
322 for (vcpus = 0; vcpus < nr_vcpus; vcpus++) {
323 ret = sem_post(&test_stage_updated);
324 TEST_ASSERT(ret == 0, "Error in sem_post");
325 }
326 pr_debug("All vcpus have been notified to continue\n");
327
328 /* Wait for all the vcpus to complete new test stage */
329 for (vcpus = 0; vcpus < nr_vcpus; vcpus++) {
330 ret = sem_wait(&test_stage_completed);
331 TEST_ASSERT(ret == 0, "Error in sem_wait");
332
333 pr_debug("%d vcpus have completed stage %s\n",
334 vcpus + 1, test_stage_string[stage]);
335 }
336
337 pr_debug("All vcpus have completed stage %s\n",
338 test_stage_string[stage]);
339}
340
341static void run_test(enum vm_guest_mode mode, void *arg)
342{
343 pthread_t *vcpu_threads;
344 struct kvm_vm *vm;
345 struct timespec start;
346 struct timespec ts_diff;
347 int ret, i;
348
349 /* Create VM with vCPUs and make some pre-initialization */
350 vm = pre_init_before_test(mode, arg);
351
352 vcpu_threads = malloc(nr_vcpus * sizeof(*vcpu_threads));
353 TEST_ASSERT(vcpu_threads, "Memory allocation failed");
354
355 host_quit = false;
356 *current_stage = KVM_BEFORE_MAPPINGS;
357
358 for (i = 0; i < nr_vcpus; i++)
359 pthread_create(&vcpu_threads[i], NULL, vcpu_worker,
360 test_args.vcpus[i]);
361
362 vcpus_complete_new_stage(*current_stage);
363 pr_info("Started all vCPUs successfully\n");
364
365 /* Test the stage of KVM creating mappings */
366 *current_stage = KVM_CREATE_MAPPINGS;
367
368 clock_gettime(CLOCK_MONOTONIC, &start);
369 vcpus_complete_new_stage(*current_stage);
370 ts_diff = timespec_elapsed(start);
371
372 pr_info("KVM_CREATE_MAPPINGS: total execution time: %ld.%.9lds\n\n",
373 ts_diff.tv_sec, ts_diff.tv_nsec);
374
375 /* Test the stage of KVM updating mappings */
376 vm_mem_region_set_flags(vm, TEST_MEM_SLOT_INDEX,
377 KVM_MEM_LOG_DIRTY_PAGES);
378
379 *current_stage = KVM_UPDATE_MAPPINGS;
380
381 clock_gettime(CLOCK_MONOTONIC, &start);
382 vcpus_complete_new_stage(*current_stage);
383 ts_diff = timespec_elapsed(start);
384
385 pr_info("KVM_UPDATE_MAPPINGS: total execution time: %ld.%.9lds\n\n",
386 ts_diff.tv_sec, ts_diff.tv_nsec);
387
388 /* Test the stage of KVM adjusting mappings */
389 vm_mem_region_set_flags(vm, TEST_MEM_SLOT_INDEX, 0);
390
391 *current_stage = KVM_ADJUST_MAPPINGS;
392
393 clock_gettime(CLOCK_MONOTONIC, &start);
394 vcpus_complete_new_stage(*current_stage);
395 ts_diff = timespec_elapsed(start);
396
397 pr_info("KVM_ADJUST_MAPPINGS: total execution time: %ld.%.9lds\n\n",
398 ts_diff.tv_sec, ts_diff.tv_nsec);
399
400 /* Tell the vcpu thread to quit */
401 host_quit = true;
402 for (i = 0; i < nr_vcpus; i++) {
403 ret = sem_post(&test_stage_updated);
404 TEST_ASSERT(ret == 0, "Error in sem_post");
405 }
406
407 for (i = 0; i < nr_vcpus; i++)
408 pthread_join(vcpu_threads[i], NULL);
409
410 ret = sem_destroy(&test_stage_updated);
411 TEST_ASSERT(ret == 0, "Error in sem_destroy");
412
413 ret = sem_destroy(&test_stage_completed);
414 TEST_ASSERT(ret == 0, "Error in sem_destroy");
415
416 free(vcpu_threads);
417 kvm_vm_free(vm);
418}
419
420static void help(char *name)
421{
422 puts("");
423 printf("usage: %s [-h] [-p offset] [-m mode] "
424 "[-b mem-size] [-v vcpus] [-s mem-type]\n", name);
425 puts("");
426 printf(" -p: specify guest physical test memory offset\n"
427 " Warning: a low offset can conflict with the loaded test code.\n");
428 guest_modes_help();
429 printf(" -b: specify size of the memory region for testing. e.g. 10M or 3G.\n"
430 " (default: 1G)\n");
431 printf(" -v: specify the number of vCPUs to run\n"
432 " (default: 1)\n");
433 backing_src_help("-s");
434 puts("");
435}
436
437int main(int argc, char *argv[])
438{
439 int max_vcpus = kvm_check_cap(KVM_CAP_MAX_VCPUS);
440 struct test_params p = {
441 .test_mem_size = DEFAULT_TEST_MEM_SIZE,
442 .src_type = DEFAULT_VM_MEM_SRC,
443 };
444 int opt;
445
446 guest_modes_append_default();
447
448 while ((opt = getopt(argc, argv, "hp:m:b:v:s:")) != -1) {
449 switch (opt) {
450 case 'p':
451 p.phys_offset = strtoull(optarg, NULL, 0);
452 break;
453 case 'm':
454 guest_modes_cmdline(optarg);
455 break;
456 case 'b':
457 p.test_mem_size = parse_size(optarg);
458 break;
459 case 'v':
460 nr_vcpus = atoi_positive("Number of vCPUs", optarg);
461 TEST_ASSERT(nr_vcpus <= max_vcpus,
462 "Invalid number of vcpus, must be between 1 and %d", max_vcpus);
463 break;
464 case 's':
465 p.src_type = parse_backing_src_type(optarg);
466 break;
467 case 'h':
468 default:
469 help(argv[0]);
470 exit(0);
471 }
472 }
473
474 for_each_guest_mode(run_test, &p);
475
476 return 0;
477}