tools/testing/selftests/kvm/dirty_log_test.c

   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 /*
  30  * GPA offset of the testing memory slot. Must be bigger than the
  31  * default vm mem slot, which is DEFAULT_GUEST_PHY_PAGES.
  32  */
  33 #define TEST_MEM_OFFSET                 (1ul << 30) /* 1G */
  34
  35 /* How many pages to dirty for each guest loop */
  36 #define TEST_PAGES_PER_LOOP             1024
  37
  38 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
  39 #define TEST_HOST_LOOP_N                32
  40
  41 /* Interval for each host loop (ms) */
  42 #define TEST_HOST_LOOP_INTERVAL         10
  43
  44 /*
  45  * Guest/Host shared variables. Ensure addr_gva2hva() and/or
  46  * sync_global_to/from_guest() are used when accessing from
  47  * the host. READ/WRITE_ONCE() should also be used with anything
  48  * that may change.
  49  */
  50 static uint64_t host_page_size;
  51 static uint64_t guest_page_size;
  52 static uint64_t guest_num_pages;
  53 static uint64_t random_array[TEST_PAGES_PER_LOOP];
  54 static uint64_t iteration;
  55
  56 /*
  57  * Continuously write to the first 8 bytes of a random pages within
  58  * the testing memory region.
  59  */
  60 static void guest_code(void)
  61 {
  62         int i;
  63
  64         while (true) {
  65                 for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
  66                         uint64_t addr = TEST_MEM_OFFSET;
  67                         addr += (READ_ONCE(random_array[i]) % guest_num_pages)
  68                                 * guest_page_size;
  69                         addr &= ~(host_page_size - 1);
  70                         *(uint64_t *)addr = READ_ONCE(iteration);
  71                 }
  72
  73                 /* Tell the host that we need more random numbers */
  74                 GUEST_SYNC(1);
  75         }
  76 }
  77
  78 /* Host variables */
  79 static bool host_quit;
  80
  81 /* Points to the test VM memory region on which we track dirty logs */
  82 static void *host_test_mem;
  83 static uint64_t host_num_pages;
  84
  85 /* For statistics only */
  86 static uint64_t host_dirty_count;
  87 static uint64_t host_clear_count;
  88 static uint64_t host_track_next_count;
  89
  90 /*
  91  * We use this bitmap to track some pages that should have its dirty
  92  * bit set in the _next_ iteration.  For example, if we detected the
  93  * page value changed to current iteration but at the same time the
  94  * page bit is cleared in the latest bitmap, then the system must
  95  * report that write in the next get dirty log call.
  96  */
  97 static unsigned long *host_bmap_track;
  98
  99 static void generate_random_array(uint64_t *guest_array, uint64_t size)
 100 {
 101         uint64_t i;
 102
 103         for (i = 0; i < size; i++)
 104                 guest_array[i] = random();
 105 }
 106
 107 static void *vcpu_worker(void *data)
 108 {
 109         int ret;
 110         struct kvm_vm *vm = data;
 111         uint64_t *guest_array;
 112         uint64_t pages_count = 0;
 113         struct kvm_run *run;
 114         struct ucall uc;
 115
 116         run = vcpu_state(vm, VCPU_ID);
 117
 118         guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
 119         generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
 120
 121         while (!READ_ONCE(host_quit)) {
 122                 /* Let the guest dirty the random pages */
 123                 ret = _vcpu_run(vm, VCPU_ID);
 124                 if (get_ucall(vm, VCPU_ID, &uc) == UCALL_SYNC) {
 125                         pages_count += TEST_PAGES_PER_LOOP;
 126                         generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
 127                 } else {
 128                         TEST_ASSERT(false,
 129                                     "Invalid guest sync status: "
 130                                     "exit_reason=%s\n",
 131                                     exit_reason_str(run->exit_reason));
 132                 }
 133         }
 134
 135         DEBUG("Dirtied %"PRIu64" pages\n", pages_count);
 136
 137         return NULL;
 138 }
 139
 140 static void vm_dirty_log_verify(unsigned long *bmap)
 141 {
 142         uint64_t page;
 143         uint64_t *value_ptr;
 144         uint64_t step = host_page_size >= guest_page_size ? 1 :
 145                                 guest_page_size / host_page_size;
 146
 147         for (page = 0; page < host_num_pages; page += step) {
 148                 value_ptr = host_test_mem + page * host_page_size;
 149
 150                 /* If this is a special page that we were tracking... */
 151                 if (test_and_clear_bit(page, host_bmap_track)) {
 152                         host_track_next_count++;
 153                         TEST_ASSERT(test_bit(page, bmap),
 154                                     "Page %"PRIu64" should have its dirty bit "
 155                                     "set in this iteration but it is missing",
 156                                     page);
 157                 }
 158
 159                 if (test_bit(page, bmap)) {
 160                         host_dirty_count++;
 161                         /*
 162                          * If the bit is set, the value written onto
 163                          * the corresponding page should be either the
 164                          * previous iteration number or the current one.
 165                          */
 166                         TEST_ASSERT(*value_ptr == iteration ||
 167                                     *value_ptr == iteration - 1,
 168                                     "Set page %"PRIu64" value %"PRIu64
 169                                     " incorrect (iteration=%"PRIu64")",
 170                                     page, *value_ptr, iteration);
 171                 } else {
 172                         host_clear_count++;
 173                         /*
 174                          * If cleared, the value written can be any
 175                          * value smaller or equals to the iteration
 176                          * number.  Note that the value can be exactly
 177                          * (iteration-1) if that write can happen
 178                          * like this:
 179                          *
 180                          * (1) increase loop count to "iteration-1"
 181                          * (2) write to page P happens (with value
 182                          *     "iteration-1")
 183                          * (3) get dirty log for "iteration-1"; we'll
 184                          *     see that page P bit is set (dirtied),
 185                          *     and not set the bit in host_bmap_track
 186                          * (4) increase loop count to "iteration"
 187                          *     (which is current iteration)
 188                          * (5) get dirty log for current iteration,
 189                          *     we'll see that page P is cleared, with
 190                          *     value "iteration-1".
 191                          */
 192                         TEST_ASSERT(*value_ptr <= iteration,
 193                                     "Clear page %"PRIu64" value %"PRIu64
 194                                     " incorrect (iteration=%"PRIu64")",
 195                                     page, *value_ptr, iteration);
 196                         if (*value_ptr == iteration) {
 197                                 /*
 198                                  * This page is _just_ modified; it
 199                                  * should report its dirtyness in the
 200                                  * next run
 201                                  */
 202                                 set_bit(page, host_bmap_track);
 203                         }
 204                 }
 205         }
 206 }
 207
 208 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid,
 209                                 uint64_t extra_mem_pages, void *guest_code)
 210 {
 211         struct kvm_vm *vm;
 212         uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
 213
 214         vm = vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
 215         kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
 216 #ifdef __x86_64__
 217         vm_create_irqchip(vm);
 218 #endif
 219         vm_vcpu_add_default(vm, vcpuid, guest_code);
 220         return vm;
 221 }
 222
 223 static void run_test(enum vm_guest_mode mode, unsigned long iterations,
 224                      unsigned long interval)
 225 {
 226         unsigned int guest_page_shift;
 227         pthread_t vcpu_thread;
 228         struct kvm_vm *vm;
 229         unsigned long *bmap;
 230
 231         switch (mode) {
 232         case VM_MODE_P52V48_4K:
 233                 guest_page_shift = 12;
 234                 break;
 235         case VM_MODE_P52V48_64K:
 236                 guest_page_shift = 16;
 237                 break;
 238         default:
 239                 TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", mode);
 240         }
 241
 242         DEBUG("Testing guest mode: %s\n", vm_guest_mode_string(mode));
 243
 244         guest_page_size = (1ul << guest_page_shift);
 245         /* 1G of guest page sized pages */
 246         guest_num_pages = (1ul << (30 - guest_page_shift));
 247         host_page_size = getpagesize();
 248         host_num_pages = (guest_num_pages * guest_page_size) / host_page_size +
 249                          !!((guest_num_pages * guest_page_size) % host_page_size);
 250
 251         bmap = bitmap_alloc(host_num_pages);
 252         host_bmap_track = bitmap_alloc(host_num_pages);
 253
 254         vm = create_vm(mode, VCPU_ID, guest_num_pages, guest_code);
 255
 256         /* Add an extra memory slot for testing dirty logging */
 257         vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
 258                                     TEST_MEM_OFFSET,
 259                                     TEST_MEM_SLOT_INDEX,
 260                                     guest_num_pages,
 261                                     KVM_MEM_LOG_DIRTY_PAGES);
 262
 263         /* Do 1:1 mapping for the dirty track memory slot */
 264         virt_map(vm, TEST_MEM_OFFSET, TEST_MEM_OFFSET,
 265                  guest_num_pages * guest_page_size, 0);
 266
 267         /* Cache the HVA pointer of the region */
 268         host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)TEST_MEM_OFFSET);
 269
 270 #ifdef __x86_64__
 271         vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
 272 #endif
 273 #ifdef __aarch64__
 274         ucall_init(vm, UCALL_MMIO, NULL);
 275 #endif
 276
 277         /* Export the shared variables to the guest */
 278         sync_global_to_guest(vm, host_page_size);
 279         sync_global_to_guest(vm, guest_page_size);
 280         sync_global_to_guest(vm, guest_num_pages);
 281
 282         /* Start the iterations */
 283         iteration = 1;
 284         sync_global_to_guest(vm, iteration);
 285         host_quit = false;
 286         host_dirty_count = 0;
 287         host_clear_count = 0;
 288         host_track_next_count = 0;
 289
 290         pthread_create(&vcpu_thread, NULL, vcpu_worker, vm);
 291
 292         while (iteration < iterations) {
 293                 /* Give the vcpu thread some time to dirty some pages */
 294                 usleep(interval * 1000);
 295                 kvm_vm_get_dirty_log(vm, TEST_MEM_SLOT_INDEX, bmap);
 296                 vm_dirty_log_verify(bmap);
 297                 iteration++;
 298                 sync_global_to_guest(vm, iteration);
 299         }
 300
 301         /* Tell the vcpu thread to quit */
 302         host_quit = true;
 303         pthread_join(vcpu_thread, NULL);
 304
 305         DEBUG("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
 306               "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
 307               host_track_next_count);
 308
 309         free(bmap);
 310         free(host_bmap_track);
 311         ucall_uninit(vm);
 312         kvm_vm_free(vm);
 313 }
 314
 315 static struct vm_guest_modes {
 316         enum vm_guest_mode mode;
 317         bool supported;
 318         bool enabled;
 319 } vm_guest_modes[NUM_VM_MODES] = {
 320         { VM_MODE_P52V48_4K,    1, 1, },
 321 #ifdef __aarch64__
 322         { VM_MODE_P52V48_64K,   1, 1, },
 323 #else
 324         { VM_MODE_P52V48_64K,   0, 0, },
 325 #endif
 326 };
 327
 328 static void help(char *name)
 329 {
 330         int i;
 331
 332         puts("");
 333         printf("usage: %s [-h] [-i iterations] [-I interval] [-m mode]\n", name);
 334         puts("");
 335         printf(" -i: specify iteration counts (default: %"PRIu64")\n",
 336                TEST_HOST_LOOP_N);
 337         printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
 338                TEST_HOST_LOOP_INTERVAL);
 339         printf(" -m: specify the guest mode ID to test "
 340                "(default: test all supported modes)\n"
 341                "     This option may be used multiple times.\n"
 342                "     Guest mode IDs:\n");
 343         for (i = 0; i < NUM_VM_MODES; ++i) {
 344                 printf("         %d:    %s%s\n",
 345                        vm_guest_modes[i].mode,
 346                        vm_guest_mode_string(vm_guest_modes[i].mode),
 347                        vm_guest_modes[i].supported ? " (supported)" : "");
 348         }
 349         puts("");
 350         exit(0);
 351 }
 352
 353 int main(int argc, char *argv[])
 354 {
 355         unsigned long iterations = TEST_HOST_LOOP_N;
 356         unsigned long interval = TEST_HOST_LOOP_INTERVAL;
 357         bool mode_selected = false;
 358         unsigned int mode;
 359         int opt, i;
 360
 361         while ((opt = getopt(argc, argv, "hi:I:m:")) != -1) {
 362                 switch (opt) {
 363                 case 'i':
 364                         iterations = strtol(optarg, NULL, 10);
 365                         break;
 366                 case 'I':
 367                         interval = strtol(optarg, NULL, 10);
 368                         break;
 369                 case 'm':
 370                         if (!mode_selected) {
 371                                 for (i = 0; i < NUM_VM_MODES; ++i)
 372                                         vm_guest_modes[i].enabled = 0;
 373                                 mode_selected = true;
 374                         }
 375                         mode = strtoul(optarg, NULL, 10);
 376                         TEST_ASSERT(mode < NUM_VM_MODES,
 377                                     "Guest mode ID %d too big", mode);
 378                         vm_guest_modes[mode].enabled = 1;
 379                         break;
 380                 case 'h':
 381                 default:
 382                         help(argv[0]);
 383                         break;
 384                 }
 385         }
 386
 387         TEST_ASSERT(iterations > 2, "Iterations must be greater than two");
 388         TEST_ASSERT(interval > 0, "Interval must be greater than zero");
 389
 390         DEBUG("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
 391               iterations, interval);
 392
 393         srandom(time(0));
 394
 395         for (i = 0; i < NUM_VM_MODES; ++i) {
 396                 if (!vm_guest_modes[i].enabled)
 397                         continue;
 398                 TEST_ASSERT(vm_guest_modes[i].supported,
 399                             "Guest mode ID %d (%s) not supported.",
 400                             vm_guest_modes[i].mode,
 401                             vm_guest_mode_string(vm_guest_modes[i].mode));
 402                 run_test(vm_guest_modes[i].mode, iterations, interval);
 403         }
 404
 405         return 0;
 406 }