migration/dirtyrate.c

   1 /*
   2  * Dirtyrate implement code
   3  *
   4  * Copyright (c) 2020 HUAWEI TECHNOLOGIES CO.,LTD.
   5  *
   6  * Authors:
   7  *  Chuan Zheng <zhengchuan@huawei.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  10  * See the COPYING file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include <zlib.h>
  15 #include "qapi/error.h"
  16 #include "cpu.h"
  17 #include "exec/ramblock.h"
  18 #include "qemu/rcu_queue.h"
  19 #include "qapi/qapi-commands-migration.h"
  20 #include "ram.h"
  21 #include "trace.h"
  22 #include "dirtyrate.h"
  23 #include "monitor/hmp.h"
  24 #include "monitor/monitor.h"
  25 #include "qapi/qmp/qdict.h"
  26
  27 static int CalculatingState = DIRTY_RATE_STATUS_UNSTARTED;
  28 static struct DirtyRateStat DirtyStat;
  29
  30 static int64_t set_sample_page_period(int64_t msec, int64_t initial_time)
  31 {
  32     int64_t current_time;
  33
  34     current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
  35     if ((current_time - initial_time) >= msec) {
  36         msec = current_time - initial_time;
  37     } else {
  38         g_usleep((msec + initial_time - current_time) * 1000);
  39     }
  40
  41     return msec;
  42 }
  43
  44 static bool is_sample_period_valid(int64_t sec)
  45 {
  46     if (sec < MIN_FETCH_DIRTYRATE_TIME_SEC ||
  47         sec > MAX_FETCH_DIRTYRATE_TIME_SEC) {
  48         return false;
  49     }
  50
  51     return true;
  52 }
  53
  54 static bool is_sample_pages_valid(int64_t pages)
  55 {
  56     return pages >= MIN_SAMPLE_PAGE_COUNT &&
  57            pages <= MAX_SAMPLE_PAGE_COUNT;
  58 }
  59
  60 static int dirtyrate_set_state(int *state, int old_state, int new_state)
  61 {
  62     assert(new_state < DIRTY_RATE_STATUS__MAX);
  63     trace_dirtyrate_set_state(DirtyRateStatus_str(new_state));
  64     if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
  65         return 0;
  66     } else {
  67         return -1;
  68     }
  69 }
  70
  71 static struct DirtyRateInfo *query_dirty_rate_info(void)
  72 {
  73     int64_t dirty_rate = DirtyStat.dirty_rate;
  74     struct DirtyRateInfo *info = g_malloc0(sizeof(DirtyRateInfo));
  75
  76     if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) {
  77         info->has_dirty_rate = true;
  78         info->dirty_rate = dirty_rate;
  79     }
  80
  81     info->status = CalculatingState;
  82     info->start_time = DirtyStat.start_time;
  83     info->calc_time = DirtyStat.calc_time;
  84     info->sample_pages = DirtyStat.sample_pages;
  85
  86     trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState));
  87
  88     return info;
  89 }
  90
  91 static void init_dirtyrate_stat(int64_t start_time, int64_t calc_time,
  92                                 uint64_t sample_pages)
  93 {
  94     DirtyStat.total_dirty_samples = 0;
  95     DirtyStat.total_sample_count = 0;
  96     DirtyStat.total_block_mem_MB = 0;
  97     DirtyStat.dirty_rate = -1;
  98     DirtyStat.start_time = start_time;
  99     DirtyStat.calc_time = calc_time;
 100     DirtyStat.sample_pages = sample_pages;
 101 }
 102
 103 static void update_dirtyrate_stat(struct RamblockDirtyInfo *info)
 104 {
 105     DirtyStat.total_dirty_samples += info->sample_dirty_count;
 106     DirtyStat.total_sample_count += info->sample_pages_count;
 107     /* size of total pages in MB */
 108     DirtyStat.total_block_mem_MB += (info->ramblock_pages *
 109                                      TARGET_PAGE_SIZE) >> 20;
 110 }
 111
 112 static void update_dirtyrate(uint64_t msec)
 113 {
 114     uint64_t dirtyrate;
 115     uint64_t total_dirty_samples = DirtyStat.total_dirty_samples;
 116     uint64_t total_sample_count = DirtyStat.total_sample_count;
 117     uint64_t total_block_mem_MB = DirtyStat.total_block_mem_MB;
 118
 119     dirtyrate = total_dirty_samples * total_block_mem_MB *
 120                 1000 / (total_sample_count * msec);
 121
 122     DirtyStat.dirty_rate = dirtyrate;
 123 }
 124
 125 /*
 126  * get hash result for the sampled memory with length of TARGET_PAGE_SIZE
 127  * in ramblock, which starts from ramblock base address.
 128  */
 129 static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info,
 130                                       uint64_t vfn)
 131 {
 132     uint32_t crc;
 133
 134     crc = crc32(0, (info->ramblock_addr +
 135                 vfn * TARGET_PAGE_SIZE), TARGET_PAGE_SIZE);
 136
 137     trace_get_ramblock_vfn_hash(info->idstr, vfn, crc);
 138     return crc;
 139 }
 140
 141 static bool save_ramblock_hash(struct RamblockDirtyInfo *info)
 142 {
 143     unsigned int sample_pages_count;
 144     int i;
 145     GRand *rand;
 146
 147     sample_pages_count = info->sample_pages_count;
 148
 149     /* ramblock size less than one page, return success to skip this ramblock */
 150     if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) {
 151         return true;
 152     }
 153
 154     info->hash_result = g_try_malloc0_n(sample_pages_count,
 155                                         sizeof(uint32_t));
 156     if (!info->hash_result) {
 157         return false;
 158     }
 159
 160     info->sample_page_vfn = g_try_malloc0_n(sample_pages_count,
 161                                             sizeof(uint64_t));
 162     if (!info->sample_page_vfn) {
 163         g_free(info->hash_result);
 164         return false;
 165     }
 166
 167     rand  = g_rand_new();
 168     for (i = 0; i < sample_pages_count; i++) {
 169         info->sample_page_vfn[i] = g_rand_int_range(rand, 0,
 170                                                     info->ramblock_pages - 1);
 171         info->hash_result[i] = get_ramblock_vfn_hash(info,
 172                                                      info->sample_page_vfn[i]);
 173     }
 174     g_rand_free(rand);
 175
 176     return true;
 177 }
 178
 179 static void get_ramblock_dirty_info(RAMBlock *block,
 180                                     struct RamblockDirtyInfo *info,
 181                                     struct DirtyRateConfig *config)
 182 {
 183     uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes;
 184
 185     /* Right shift 30 bits to calc ramblock size in GB */
 186     info->sample_pages_count = (qemu_ram_get_used_length(block) *
 187                                 sample_pages_per_gigabytes) >> 30;
 188     /* Right shift TARGET_PAGE_BITS to calc page count */
 189     info->ramblock_pages = qemu_ram_get_used_length(block) >>
 190                            TARGET_PAGE_BITS;
 191     info->ramblock_addr = qemu_ram_get_host_addr(block);
 192     strcpy(info->idstr, qemu_ram_get_idstr(block));
 193 }
 194
 195 static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count)
 196 {
 197     int i;
 198
 199     if (!infos) {
 200         return;
 201     }
 202
 203     for (i = 0; i < count; i++) {
 204         g_free(infos[i].sample_page_vfn);
 205         g_free(infos[i].hash_result);
 206     }
 207     g_free(infos);
 208 }
 209
 210 static bool skip_sample_ramblock(RAMBlock *block)
 211 {
 212     /*
 213      * Sample only blocks larger than MIN_RAMBLOCK_SIZE.
 214      */
 215     if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) {
 216         trace_skip_sample_ramblock(block->idstr,
 217                                    qemu_ram_get_used_length(block));
 218         return true;
 219     }
 220
 221     return false;
 222 }
 223
 224 static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo,
 225                                       struct DirtyRateConfig config,
 226                                       int *block_count)
 227 {
 228     struct RamblockDirtyInfo *info = NULL;
 229     struct RamblockDirtyInfo *dinfo = NULL;
 230     RAMBlock *block = NULL;
 231     int total_count = 0;
 232     int index = 0;
 233     bool ret = false;
 234
 235     RAMBLOCK_FOREACH_MIGRATABLE(block) {
 236         if (skip_sample_ramblock(block)) {
 237             continue;
 238         }
 239         total_count++;
 240     }
 241
 242     dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo));
 243     if (dinfo == NULL) {
 244         goto out;
 245     }
 246
 247     RAMBLOCK_FOREACH_MIGRATABLE(block) {
 248         if (skip_sample_ramblock(block)) {
 249             continue;
 250         }
 251         if (index >= total_count) {
 252             break;
 253         }
 254         info = &dinfo[index];
 255         get_ramblock_dirty_info(block, info, &config);
 256         if (!save_ramblock_hash(info)) {
 257             goto out;
 258         }
 259         index++;
 260     }
 261     ret = true;
 262
 263 out:
 264     *block_count = index;
 265     *block_dinfo = dinfo;
 266     return ret;
 267 }
 268
 269 static void calc_page_dirty_rate(struct RamblockDirtyInfo *info)
 270 {
 271     uint32_t crc;
 272     int i;
 273
 274     for (i = 0; i < info->sample_pages_count; i++) {
 275         crc = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]);
 276         if (crc != info->hash_result[i]) {
 277             trace_calc_page_dirty_rate(info->idstr, crc, info->hash_result[i]);
 278             info->sample_dirty_count++;
 279         }
 280     }
 281 }
 282
 283 static struct RamblockDirtyInfo *
 284 find_block_matched(RAMBlock *block, int count,
 285                   struct RamblockDirtyInfo *infos)
 286 {
 287     int i;
 288     struct RamblockDirtyInfo *matched;
 289
 290     for (i = 0; i < count; i++) {
 291         if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) {
 292             break;
 293         }
 294     }
 295
 296     if (i == count) {
 297         return NULL;
 298     }
 299
 300     if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) ||
 301         infos[i].ramblock_pages !=
 302             (qemu_ram_get_used_length(block) >> TARGET_PAGE_BITS)) {
 303         trace_find_page_matched(block->idstr);
 304         return NULL;
 305     }
 306
 307     matched = &infos[i];
 308
 309     return matched;
 310 }
 311
 312 static bool compare_page_hash_info(struct RamblockDirtyInfo *info,
 313                                   int block_count)
 314 {
 315     struct RamblockDirtyInfo *block_dinfo = NULL;
 316     RAMBlock *block = NULL;
 317
 318     RAMBLOCK_FOREACH_MIGRATABLE(block) {
 319         if (skip_sample_ramblock(block)) {
 320             continue;
 321         }
 322         block_dinfo = find_block_matched(block, block_count, info);
 323         if (block_dinfo == NULL) {
 324             continue;
 325         }
 326         calc_page_dirty_rate(block_dinfo);
 327         update_dirtyrate_stat(block_dinfo);
 328     }
 329
 330     if (DirtyStat.total_sample_count == 0) {
 331         return false;
 332     }
 333
 334     return true;
 335 }
 336
 337 static void calculate_dirtyrate(struct DirtyRateConfig config)
 338 {
 339     struct RamblockDirtyInfo *block_dinfo = NULL;
 340     int block_count = 0;
 341     int64_t msec = 0;
 342     int64_t initial_time;
 343
 344     rcu_register_thread();
 345     rcu_read_lock();
 346     initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 347     if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) {
 348         goto out;
 349     }
 350     rcu_read_unlock();
 351
 352     msec = config.sample_period_seconds * 1000;
 353     msec = set_sample_page_period(msec, initial_time);
 354     DirtyStat.start_time = initial_time / 1000;
 355     DirtyStat.calc_time = msec / 1000;
 356
 357     rcu_read_lock();
 358     if (!compare_page_hash_info(block_dinfo, block_count)) {
 359         goto out;
 360     }
 361
 362     update_dirtyrate(msec);
 363
 364 out:
 365     rcu_read_unlock();
 366     free_ramblock_dirty_info(block_dinfo, block_count);
 367     rcu_unregister_thread();
 368 }
 369
 370 void *get_dirtyrate_thread(void *arg)
 371 {
 372     struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg;
 373     int ret;
 374     int64_t start_time;
 375     int64_t calc_time;
 376     uint64_t sample_pages;
 377
 378     ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED,
 379                               DIRTY_RATE_STATUS_MEASURING);
 380     if (ret == -1) {
 381         error_report("change dirtyrate state failed.");
 382         return NULL;
 383     }
 384
 385     start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000;
 386     calc_time = config.sample_period_seconds;
 387     sample_pages = config.sample_pages_per_gigabytes;
 388     init_dirtyrate_stat(start_time, calc_time, sample_pages);
 389
 390     calculate_dirtyrate(config);
 391
 392     ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING,
 393                               DIRTY_RATE_STATUS_MEASURED);
 394     if (ret == -1) {
 395         error_report("change dirtyrate state failed.");
 396     }
 397     return NULL;
 398 }
 399
 400 void qmp_calc_dirty_rate(int64_t calc_time, bool has_sample_pages,
 401                          int64_t sample_pages, Error **errp)
 402 {
 403     static struct DirtyRateConfig config;
 404     QemuThread thread;
 405     int ret;
 406
 407     /*
 408      * If the dirty rate is already being measured, don't attempt to start.
 409      */
 410     if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) {
 411         error_setg(errp, "the dirty rate is already being measured.");
 412         return;
 413     }
 414
 415     if (!is_sample_period_valid(calc_time)) {
 416         error_setg(errp, "calc-time is out of range[%d, %d].",
 417                          MIN_FETCH_DIRTYRATE_TIME_SEC,
 418                          MAX_FETCH_DIRTYRATE_TIME_SEC);
 419         return;
 420     }
 421
 422     if (has_sample_pages) {
 423         if (!is_sample_pages_valid(sample_pages)) {
 424             error_setg(errp, "sample-pages is out of range[%d, %d].",
 425                             MIN_SAMPLE_PAGE_COUNT,
 426                             MAX_SAMPLE_PAGE_COUNT);
 427             return;
 428         }
 429     } else {
 430         sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES;
 431     }
 432
 433     /*
 434      * Init calculation state as unstarted.
 435      */
 436     ret = dirtyrate_set_state(&CalculatingState, CalculatingState,
 437                               DIRTY_RATE_STATUS_UNSTARTED);
 438     if (ret == -1) {
 439         error_setg(errp, "init dirty rate calculation state failed.");
 440         return;
 441     }
 442
 443     config.sample_period_seconds = calc_time;
 444     config.sample_pages_per_gigabytes = sample_pages;
 445     qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread,
 446                        (void *)&config, QEMU_THREAD_DETACHED);
 447 }
 448
 449 struct DirtyRateInfo *qmp_query_dirty_rate(Error **errp)
 450 {
 451     return query_dirty_rate_info();
 452 }
 453
 454 void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict)
 455 {
 456     DirtyRateInfo *info = query_dirty_rate_info();
 457
 458     monitor_printf(mon, "Status: %s\n",
 459                    DirtyRateStatus_str(info->status));
 460     monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n",
 461                    info->start_time);
 462     monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n",
 463                    info->sample_pages);
 464     monitor_printf(mon, "Period: %"PRIi64" (sec)\n",
 465                    info->calc_time);
 466     monitor_printf(mon, "Dirty rate: ");
 467     if (info->has_dirty_rate) {
 468         monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate);
 469     } else {
 470         monitor_printf(mon, "(not ready)\n");
 471     }
 472     g_free(info);
 473 }
 474
 475 void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict)
 476 {
 477     int64_t sec = qdict_get_try_int(qdict, "second", 0);
 478     int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1);
 479     bool has_sample_pages = (sample_pages != -1);
 480     Error *err = NULL;
 481
 482     if (!sec) {
 483         monitor_printf(mon, "Incorrect period length specified!\n");
 484         return;
 485     }
 486
 487     qmp_calc_dirty_rate(sec, has_sample_pages, sample_pages, &err);
 488     if (err) {
 489         hmp_handle_error(mon, err);
 490         return;
 491     }
 492
 493     monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64
 494                    " seconds\n", sec);
 495     monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n");
 496 }