]>
Commit | Line | Data |
---|---|---|
4240dcee CZ |
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" | |
cc37d98b | 14 | #include "qemu/error-report.h" |
662770af | 15 | #include <zlib.h> |
1e05888a | 16 | #include "hw/core/cpu.h" |
4240dcee | 17 | #include "qapi/error.h" |
4240dcee | 18 | #include "exec/ramblock.h" |
beeda9b7 | 19 | #include "exec/target_page.h" |
4240dcee | 20 | #include "qemu/rcu_queue.h" |
0e21bf24 | 21 | #include "qemu/main-loop.h" |
4240dcee | 22 | #include "qapi/qapi-commands-migration.h" |
3ded54b1 | 23 | #include "ram.h" |
3c0b5dff | 24 | #include "trace.h" |
4240dcee | 25 | #include "dirtyrate.h" |
a4a571d9 PX |
26 | #include "monitor/hmp.h" |
27 | #include "monitor/monitor.h" | |
28 | #include "qapi/qmp/qdict.h" | |
0e21bf24 HH |
29 | #include "sysemu/kvm.h" |
30 | #include "sysemu/runstate.h" | |
31 | #include "exec/memory.h" | |
00a3f9c6 | 32 | #include "qemu/xxhash.h" |
0e21bf24 | 33 | |
4998a37e HH |
34 | /* |
35 | * total_dirty_pages is procted by BQL and is used | |
36 | * to stat dirty pages during the period of two | |
37 | * memory_global_dirty_log_sync | |
38 | */ | |
39 | uint64_t total_dirty_pages; | |
40 | ||
0e21bf24 HH |
41 | typedef struct DirtyPageRecord { |
42 | uint64_t start_pages; | |
43 | uint64_t end_pages; | |
44 | } DirtyPageRecord; | |
4240dcee | 45 | |
7df3aa30 | 46 | static int CalculatingState = DIRTY_RATE_STATUS_UNSTARTED; |
c9a58d71 | 47 | static struct DirtyRateStat DirtyStat; |
0e21bf24 HH |
48 | static DirtyRateMeasureMode dirtyrate_mode = |
49 | DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
7df3aa30 | 50 | |
8244166d | 51 | static int64_t dirty_stat_wait(int64_t msec, int64_t initial_time) |
eca58224 CZ |
52 | { |
53 | int64_t current_time; | |
54 | ||
55 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
56 | if ((current_time - initial_time) >= msec) { | |
57 | msec = current_time - initial_time; | |
58 | } else { | |
59 | g_usleep((msec + initial_time - current_time) * 1000); | |
60 | } | |
61 | ||
62 | return msec; | |
63 | } | |
64 | ||
8244166d HH |
65 | static inline void record_dirtypages(DirtyPageRecord *dirty_pages, |
66 | CPUState *cpu, bool start) | |
67 | { | |
68 | if (start) { | |
69 | dirty_pages[cpu->cpu_index].start_pages = cpu->dirty_pages; | |
70 | } else { | |
71 | dirty_pages[cpu->cpu_index].end_pages = cpu->dirty_pages; | |
72 | } | |
73 | } | |
74 | ||
75 | static int64_t do_calculate_dirtyrate(DirtyPageRecord dirty_pages, | |
76 | int64_t calc_time_ms) | |
77 | { | |
8244166d HH |
78 | uint64_t increased_dirty_pages = |
79 | dirty_pages.end_pages - dirty_pages.start_pages; | |
beeda9b7 | 80 | uint64_t memory_size_MiB = qemu_target_pages_to_MiB(increased_dirty_pages); |
8244166d | 81 | |
beeda9b7 | 82 | return memory_size_MiB * 1000 / calc_time_ms; |
8244166d HH |
83 | } |
84 | ||
85 | void global_dirty_log_change(unsigned int flag, bool start) | |
86 | { | |
87 | qemu_mutex_lock_iothread(); | |
88 | if (start) { | |
89 | memory_global_dirty_log_start(flag); | |
90 | } else { | |
91 | memory_global_dirty_log_stop(flag); | |
92 | } | |
93 | qemu_mutex_unlock_iothread(); | |
94 | } | |
95 | ||
96 | /* | |
97 | * global_dirty_log_sync | |
98 | * 1. sync dirty log from kvm | |
99 | * 2. stop dirty tracking if needed. | |
100 | */ | |
101 | static void global_dirty_log_sync(unsigned int flag, bool one_shot) | |
102 | { | |
103 | qemu_mutex_lock_iothread(); | |
104 | memory_global_dirty_log_sync(); | |
105 | if (one_shot) { | |
106 | memory_global_dirty_log_stop(flag); | |
107 | } | |
108 | qemu_mutex_unlock_iothread(); | |
109 | } | |
110 | ||
111 | static DirtyPageRecord *vcpu_dirty_stat_alloc(VcpuStat *stat) | |
112 | { | |
113 | CPUState *cpu; | |
8244166d HH |
114 | int nvcpu = 0; |
115 | ||
116 | CPU_FOREACH(cpu) { | |
117 | nvcpu++; | |
118 | } | |
119 | ||
120 | stat->nvcpu = nvcpu; | |
c5e8d518 | 121 | stat->rates = g_new0(DirtyRateVcpu, nvcpu); |
8244166d | 122 | |
66997c42 | 123 | return g_new0(DirtyPageRecord, nvcpu); |
8244166d HH |
124 | } |
125 | ||
126 | static void vcpu_dirty_stat_collect(VcpuStat *stat, | |
127 | DirtyPageRecord *records, | |
128 | bool start) | |
129 | { | |
130 | CPUState *cpu; | |
131 | ||
132 | CPU_FOREACH(cpu) { | |
133 | record_dirtypages(records, cpu, start); | |
134 | } | |
135 | } | |
136 | ||
137 | int64_t vcpu_calculate_dirtyrate(int64_t calc_time_ms, | |
138 | VcpuStat *stat, | |
139 | unsigned int flag, | |
140 | bool one_shot) | |
141 | { | |
142 | DirtyPageRecord *records; | |
143 | int64_t init_time_ms; | |
144 | int64_t duration; | |
145 | int64_t dirtyrate; | |
146 | int i = 0; | |
147 | unsigned int gen_id; | |
148 | ||
149 | retry: | |
150 | init_time_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
151 | ||
370ed600 JI |
152 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
153 | gen_id = cpu_list_generation_id_get(); | |
154 | records = vcpu_dirty_stat_alloc(stat); | |
155 | vcpu_dirty_stat_collect(stat, records, true); | |
156 | } | |
8244166d HH |
157 | |
158 | duration = dirty_stat_wait(calc_time_ms, init_time_ms); | |
159 | ||
160 | global_dirty_log_sync(flag, one_shot); | |
161 | ||
370ed600 JI |
162 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
163 | if (gen_id != cpu_list_generation_id_get()) { | |
164 | g_free(records); | |
165 | g_free(stat->rates); | |
166 | cpu_list_unlock(); | |
167 | goto retry; | |
168 | } | |
169 | vcpu_dirty_stat_collect(stat, records, false); | |
8244166d | 170 | } |
8244166d HH |
171 | |
172 | for (i = 0; i < stat->nvcpu; i++) { | |
173 | dirtyrate = do_calculate_dirtyrate(records[i], duration); | |
174 | ||
175 | stat->rates[i].id = i; | |
176 | stat->rates[i].dirty_rate = dirtyrate; | |
177 | ||
178 | trace_dirtyrate_do_calculate_vcpu(i, dirtyrate); | |
179 | } | |
180 | ||
181 | g_free(records); | |
182 | ||
183 | return duration; | |
184 | } | |
185 | ||
eca58224 CZ |
186 | static bool is_sample_period_valid(int64_t sec) |
187 | { | |
188 | if (sec < MIN_FETCH_DIRTYRATE_TIME_SEC || | |
189 | sec > MAX_FETCH_DIRTYRATE_TIME_SEC) { | |
190 | return false; | |
191 | } | |
192 | ||
193 | return true; | |
194 | } | |
195 | ||
7afa08cd HH |
196 | static bool is_sample_pages_valid(int64_t pages) |
197 | { | |
198 | return pages >= MIN_SAMPLE_PAGE_COUNT && | |
199 | pages <= MAX_SAMPLE_PAGE_COUNT; | |
200 | } | |
201 | ||
7df3aa30 CZ |
202 | static int dirtyrate_set_state(int *state, int old_state, int new_state) |
203 | { | |
204 | assert(new_state < DIRTY_RATE_STATUS__MAX); | |
3c0b5dff | 205 | trace_dirtyrate_set_state(DirtyRateStatus_str(new_state)); |
7df3aa30 CZ |
206 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
207 | return 0; | |
208 | } else { | |
209 | return -1; | |
210 | } | |
211 | } | |
212 | ||
4c437254 CZ |
213 | static struct DirtyRateInfo *query_dirty_rate_info(void) |
214 | { | |
0e21bf24 | 215 | int i; |
4c437254 | 216 | int64_t dirty_rate = DirtyStat.dirty_rate; |
b21e2380 | 217 | struct DirtyRateInfo *info = g_new0(DirtyRateInfo, 1); |
0e21bf24 | 218 | DirtyRateVcpuList *head = NULL, **tail = &head; |
4c437254 CZ |
219 | |
220 | info->status = CalculatingState; | |
221 | info->start_time = DirtyStat.start_time; | |
222 | info->calc_time = DirtyStat.calc_time; | |
7afa08cd | 223 | info->sample_pages = DirtyStat.sample_pages; |
0e21bf24 HH |
224 | info->mode = dirtyrate_mode; |
225 | ||
226 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) { | |
227 | info->has_dirty_rate = true; | |
228 | info->dirty_rate = dirty_rate; | |
229 | ||
230 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
231 | /* | |
232 | * set sample_pages with 0 to indicate page sampling | |
233 | * isn't enabled | |
234 | **/ | |
235 | info->sample_pages = 0; | |
236 | info->has_vcpu_dirty_rate = true; | |
237 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { | |
b21e2380 | 238 | DirtyRateVcpu *rate = g_new0(DirtyRateVcpu, 1); |
0e21bf24 HH |
239 | rate->id = DirtyStat.dirty_ring.rates[i].id; |
240 | rate->dirty_rate = DirtyStat.dirty_ring.rates[i].dirty_rate; | |
241 | QAPI_LIST_APPEND(tail, rate); | |
242 | } | |
243 | info->vcpu_dirty_rate = head; | |
244 | } | |
826b8bc8 HH |
245 | |
246 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { | |
247 | info->sample_pages = 0; | |
248 | } | |
0e21bf24 | 249 | } |
4c437254 | 250 | |
3c0b5dff CZ |
251 | trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState)); |
252 | ||
4c437254 CZ |
253 | return info; |
254 | } | |
255 | ||
71864ead HH |
256 | static void init_dirtyrate_stat(int64_t start_time, |
257 | struct DirtyRateConfig config) | |
c9a58d71 | 258 | { |
c9a58d71 | 259 | DirtyStat.dirty_rate = -1; |
aa84b506 | 260 | DirtyStat.start_time = start_time; |
71864ead HH |
261 | DirtyStat.calc_time = config.sample_period_seconds; |
262 | DirtyStat.sample_pages = config.sample_pages_per_gigabytes; | |
263 | ||
264 | switch (config.mode) { | |
265 | case DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING: | |
266 | DirtyStat.page_sampling.total_dirty_samples = 0; | |
267 | DirtyStat.page_sampling.total_sample_count = 0; | |
268 | DirtyStat.page_sampling.total_block_mem_MB = 0; | |
269 | break; | |
270 | case DIRTY_RATE_MEASURE_MODE_DIRTY_RING: | |
271 | DirtyStat.dirty_ring.nvcpu = -1; | |
272 | DirtyStat.dirty_ring.rates = NULL; | |
273 | break; | |
274 | default: | |
275 | break; | |
276 | } | |
c9a58d71 CZ |
277 | } |
278 | ||
0e21bf24 HH |
279 | static void cleanup_dirtyrate_stat(struct DirtyRateConfig config) |
280 | { | |
281 | /* last calc-dirty-rate qmp use dirty ring mode */ | |
282 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
283 | free(DirtyStat.dirty_ring.rates); | |
284 | DirtyStat.dirty_ring.rates = NULL; | |
285 | } | |
286 | } | |
287 | ||
c9a58d71 CZ |
288 | static void update_dirtyrate_stat(struct RamblockDirtyInfo *info) |
289 | { | |
71864ead HH |
290 | DirtyStat.page_sampling.total_dirty_samples += info->sample_dirty_count; |
291 | DirtyStat.page_sampling.total_sample_count += info->sample_pages_count; | |
c9a58d71 | 292 | /* size of total pages in MB */ |
beeda9b7 JQ |
293 | DirtyStat.page_sampling.total_block_mem_MB += |
294 | qemu_target_pages_to_MiB(info->ramblock_pages); | |
c9a58d71 CZ |
295 | } |
296 | ||
297 | static void update_dirtyrate(uint64_t msec) | |
298 | { | |
299 | uint64_t dirtyrate; | |
71864ead HH |
300 | uint64_t total_dirty_samples = DirtyStat.page_sampling.total_dirty_samples; |
301 | uint64_t total_sample_count = DirtyStat.page_sampling.total_sample_count; | |
302 | uint64_t total_block_mem_MB = DirtyStat.page_sampling.total_block_mem_MB; | |
c9a58d71 CZ |
303 | |
304 | dirtyrate = total_dirty_samples * total_block_mem_MB * | |
305 | 1000 / (total_sample_count * msec); | |
306 | ||
307 | DirtyStat.dirty_rate = dirtyrate; | |
308 | } | |
7df3aa30 | 309 | |
00a3f9c6 AG |
310 | /* |
311 | * Compute hash of a single page of size TARGET_PAGE_SIZE. | |
312 | */ | |
313 | static uint32_t compute_page_hash(void *ptr) | |
314 | { | |
edd83a70 | 315 | size_t page_size = qemu_target_page_size(); |
00a3f9c6 AG |
316 | uint32_t i; |
317 | uint64_t v1, v2, v3, v4; | |
318 | uint64_t res; | |
319 | const uint64_t *p = ptr; | |
320 | ||
321 | v1 = QEMU_XXHASH_SEED + XXH_PRIME64_1 + XXH_PRIME64_2; | |
322 | v2 = QEMU_XXHASH_SEED + XXH_PRIME64_2; | |
323 | v3 = QEMU_XXHASH_SEED + 0; | |
324 | v4 = QEMU_XXHASH_SEED - XXH_PRIME64_1; | |
edd83a70 | 325 | for (i = 0; i < page_size / 8; i += 4) { |
00a3f9c6 AG |
326 | v1 = XXH64_round(v1, p[i + 0]); |
327 | v2 = XXH64_round(v2, p[i + 1]); | |
328 | v3 = XXH64_round(v3, p[i + 2]); | |
329 | v4 = XXH64_round(v4, p[i + 3]); | |
330 | } | |
331 | res = XXH64_mergerounds(v1, v2, v3, v4); | |
edd83a70 | 332 | res += page_size; |
00a3f9c6 AG |
333 | res = XXH64_avalanche(res); |
334 | return (uint32_t)(res & UINT32_MAX); | |
335 | } | |
336 | ||
337 | ||
ba0e519f CZ |
338 | /* |
339 | * get hash result for the sampled memory with length of TARGET_PAGE_SIZE | |
340 | * in ramblock, which starts from ramblock base address. | |
341 | */ | |
342 | static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info, | |
343 | uint64_t vfn) | |
344 | { | |
00a3f9c6 | 345 | uint32_t hash; |
ba0e519f | 346 | |
edd83a70 JQ |
347 | hash = compute_page_hash(info->ramblock_addr + |
348 | vfn * qemu_target_page_size()); | |
ba0e519f | 349 | |
00a3f9c6 AG |
350 | trace_get_ramblock_vfn_hash(info->idstr, vfn, hash); |
351 | return hash; | |
ba0e519f CZ |
352 | } |
353 | ||
354 | static bool save_ramblock_hash(struct RamblockDirtyInfo *info) | |
355 | { | |
356 | unsigned int sample_pages_count; | |
357 | int i; | |
358 | GRand *rand; | |
359 | ||
360 | sample_pages_count = info->sample_pages_count; | |
361 | ||
362 | /* ramblock size less than one page, return success to skip this ramblock */ | |
363 | if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) { | |
364 | return true; | |
365 | } | |
366 | ||
367 | info->hash_result = g_try_malloc0_n(sample_pages_count, | |
368 | sizeof(uint32_t)); | |
369 | if (!info->hash_result) { | |
370 | return false; | |
371 | } | |
372 | ||
373 | info->sample_page_vfn = g_try_malloc0_n(sample_pages_count, | |
374 | sizeof(uint64_t)); | |
375 | if (!info->sample_page_vfn) { | |
376 | g_free(info->hash_result); | |
377 | return false; | |
378 | } | |
379 | ||
380 | rand = g_rand_new(); | |
381 | for (i = 0; i < sample_pages_count; i++) { | |
382 | info->sample_page_vfn[i] = g_rand_int_range(rand, 0, | |
383 | info->ramblock_pages - 1); | |
384 | info->hash_result[i] = get_ramblock_vfn_hash(info, | |
385 | info->sample_page_vfn[i]); | |
386 | } | |
387 | g_rand_free(rand); | |
388 | ||
389 | return true; | |
390 | } | |
391 | ||
392 | static void get_ramblock_dirty_info(RAMBlock *block, | |
393 | struct RamblockDirtyInfo *info, | |
394 | struct DirtyRateConfig *config) | |
395 | { | |
396 | uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes; | |
397 | ||
398 | /* Right shift 30 bits to calc ramblock size in GB */ | |
399 | info->sample_pages_count = (qemu_ram_get_used_length(block) * | |
400 | sample_pages_per_gigabytes) >> 30; | |
401 | /* Right shift TARGET_PAGE_BITS to calc page count */ | |
402 | info->ramblock_pages = qemu_ram_get_used_length(block) >> | |
148b1ad8 | 403 | qemu_target_page_bits(); |
ba0e519f CZ |
404 | info->ramblock_addr = qemu_ram_get_host_addr(block); |
405 | strcpy(info->idstr, qemu_ram_get_idstr(block)); | |
406 | } | |
407 | ||
cf0bbb49 CZ |
408 | static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count) |
409 | { | |
410 | int i; | |
411 | ||
412 | if (!infos) { | |
413 | return; | |
414 | } | |
415 | ||
416 | for (i = 0; i < count; i++) { | |
417 | g_free(infos[i].sample_page_vfn); | |
418 | g_free(infos[i].hash_result); | |
419 | } | |
420 | g_free(infos); | |
421 | } | |
422 | ||
f82583cd CZ |
423 | static bool skip_sample_ramblock(RAMBlock *block) |
424 | { | |
425 | /* | |
426 | * Sample only blocks larger than MIN_RAMBLOCK_SIZE. | |
427 | */ | |
428 | if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) { | |
3c0b5dff CZ |
429 | trace_skip_sample_ramblock(block->idstr, |
430 | qemu_ram_get_used_length(block)); | |
f82583cd CZ |
431 | return true; |
432 | } | |
433 | ||
434 | return false; | |
435 | } | |
436 | ||
ba0e519f CZ |
437 | static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo, |
438 | struct DirtyRateConfig config, | |
439 | int *block_count) | |
440 | { | |
441 | struct RamblockDirtyInfo *info = NULL; | |
442 | struct RamblockDirtyInfo *dinfo = NULL; | |
443 | RAMBlock *block = NULL; | |
444 | int total_count = 0; | |
445 | int index = 0; | |
446 | bool ret = false; | |
447 | ||
448 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
449 | if (skip_sample_ramblock(block)) { |
450 | continue; | |
451 | } | |
ba0e519f CZ |
452 | total_count++; |
453 | } | |
454 | ||
455 | dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo)); | |
456 | if (dinfo == NULL) { | |
457 | goto out; | |
458 | } | |
459 | ||
460 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
461 | if (skip_sample_ramblock(block)) { |
462 | continue; | |
463 | } | |
ba0e519f CZ |
464 | if (index >= total_count) { |
465 | break; | |
466 | } | |
467 | info = &dinfo[index]; | |
468 | get_ramblock_dirty_info(block, info, &config); | |
469 | if (!save_ramblock_hash(info)) { | |
470 | goto out; | |
471 | } | |
472 | index++; | |
473 | } | |
474 | ret = true; | |
475 | ||
476 | out: | |
477 | *block_count = index; | |
478 | *block_dinfo = dinfo; | |
479 | return ret; | |
480 | } | |
481 | ||
9c04387b CZ |
482 | static void calc_page_dirty_rate(struct RamblockDirtyInfo *info) |
483 | { | |
00a3f9c6 | 484 | uint32_t hash; |
9c04387b CZ |
485 | int i; |
486 | ||
487 | for (i = 0; i < info->sample_pages_count; i++) { | |
00a3f9c6 AG |
488 | hash = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]); |
489 | if (hash != info->hash_result[i]) { | |
490 | trace_calc_page_dirty_rate(info->idstr, hash, info->hash_result[i]); | |
9c04387b CZ |
491 | info->sample_dirty_count++; |
492 | } | |
493 | } | |
494 | } | |
495 | ||
496 | static struct RamblockDirtyInfo * | |
497 | find_block_matched(RAMBlock *block, int count, | |
498 | struct RamblockDirtyInfo *infos) | |
499 | { | |
500 | int i; | |
9c04387b CZ |
501 | |
502 | for (i = 0; i < count; i++) { | |
503 | if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) { | |
504 | break; | |
505 | } | |
506 | } | |
507 | ||
508 | if (i == count) { | |
509 | return NULL; | |
510 | } | |
511 | ||
512 | if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) || | |
513 | infos[i].ramblock_pages != | |
148b1ad8 | 514 | (qemu_ram_get_used_length(block) >> qemu_target_page_bits())) { |
3c0b5dff | 515 | trace_find_page_matched(block->idstr); |
9c04387b CZ |
516 | return NULL; |
517 | } | |
518 | ||
66997c42 | 519 | return &infos[i]; |
9c04387b CZ |
520 | } |
521 | ||
522 | static bool compare_page_hash_info(struct RamblockDirtyInfo *info, | |
523 | int block_count) | |
524 | { | |
525 | struct RamblockDirtyInfo *block_dinfo = NULL; | |
526 | RAMBlock *block = NULL; | |
527 | ||
528 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
529 | if (skip_sample_ramblock(block)) { |
530 | continue; | |
531 | } | |
9c04387b CZ |
532 | block_dinfo = find_block_matched(block, block_count, info); |
533 | if (block_dinfo == NULL) { | |
534 | continue; | |
535 | } | |
536 | calc_page_dirty_rate(block_dinfo); | |
537 | update_dirtyrate_stat(block_dinfo); | |
538 | } | |
539 | ||
71864ead | 540 | if (DirtyStat.page_sampling.total_sample_count == 0) { |
9c04387b CZ |
541 | return false; |
542 | } | |
543 | ||
544 | return true; | |
545 | } | |
546 | ||
826b8bc8 HH |
547 | static inline void record_dirtypages_bitmap(DirtyPageRecord *dirty_pages, |
548 | bool start) | |
549 | { | |
550 | if (start) { | |
551 | dirty_pages->start_pages = total_dirty_pages; | |
552 | } else { | |
553 | dirty_pages->end_pages = total_dirty_pages; | |
554 | } | |
555 | } | |
556 | ||
826b8bc8 HH |
557 | static inline void dirtyrate_manual_reset_protect(void) |
558 | { | |
559 | RAMBlock *block = NULL; | |
560 | ||
561 | WITH_RCU_READ_LOCK_GUARD() { | |
562 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
563 | memory_region_clear_dirty_bitmap(block->mr, 0, | |
564 | block->used_length); | |
565 | } | |
566 | } | |
567 | } | |
568 | ||
569 | static void calculate_dirtyrate_dirty_bitmap(struct DirtyRateConfig config) | |
570 | { | |
571 | int64_t msec = 0; | |
572 | int64_t start_time; | |
573 | DirtyPageRecord dirty_pages; | |
574 | ||
575 | qemu_mutex_lock_iothread(); | |
576 | memory_global_dirty_log_start(GLOBAL_DIRTY_DIRTY_RATE); | |
577 | ||
578 | /* | |
579 | * 1'round of log sync may return all 1 bits with | |
580 | * KVM_DIRTY_LOG_INITIALLY_SET enable | |
581 | * skip it unconditionally and start dirty tracking | |
582 | * from 2'round of log sync | |
583 | */ | |
584 | memory_global_dirty_log_sync(); | |
585 | ||
586 | /* | |
587 | * reset page protect manually and unconditionally. | |
588 | * this make sure kvm dirty log be cleared if | |
589 | * KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE cap is enabled. | |
590 | */ | |
591 | dirtyrate_manual_reset_protect(); | |
592 | qemu_mutex_unlock_iothread(); | |
593 | ||
594 | record_dirtypages_bitmap(&dirty_pages, true); | |
595 | ||
596 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
597 | DirtyStat.start_time = start_time / 1000; | |
598 | ||
599 | msec = config.sample_period_seconds * 1000; | |
8244166d | 600 | msec = dirty_stat_wait(msec, start_time); |
826b8bc8 HH |
601 | DirtyStat.calc_time = msec / 1000; |
602 | ||
603 | /* | |
8244166d | 604 | * do two things. |
826b8bc8 HH |
605 | * 1. fetch dirty bitmap from kvm |
606 | * 2. stop dirty tracking | |
607 | */ | |
8244166d | 608 | global_dirty_log_sync(GLOBAL_DIRTY_DIRTY_RATE, true); |
826b8bc8 HH |
609 | |
610 | record_dirtypages_bitmap(&dirty_pages, false); | |
611 | ||
8244166d | 612 | DirtyStat.dirty_rate = do_calculate_dirtyrate(dirty_pages, msec); |
826b8bc8 HH |
613 | } |
614 | ||
0e21bf24 HH |
615 | static void calculate_dirtyrate_dirty_ring(struct DirtyRateConfig config) |
616 | { | |
8244166d | 617 | int64_t duration; |
0e21bf24 HH |
618 | uint64_t dirtyrate = 0; |
619 | uint64_t dirtyrate_sum = 0; | |
0e21bf24 HH |
620 | int i = 0; |
621 | ||
8244166d HH |
622 | /* start log sync */ |
623 | global_dirty_log_change(GLOBAL_DIRTY_DIRTY_RATE, true); | |
0e21bf24 | 624 | |
8244166d | 625 | DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; |
0e21bf24 | 626 | |
8244166d HH |
627 | /* calculate vcpu dirtyrate */ |
628 | duration = vcpu_calculate_dirtyrate(config.sample_period_seconds * 1000, | |
629 | &DirtyStat.dirty_ring, | |
630 | GLOBAL_DIRTY_DIRTY_RATE, | |
631 | true); | |
0e21bf24 | 632 | |
8244166d | 633 | DirtyStat.calc_time = duration / 1000; |
0e21bf24 | 634 | |
8244166d | 635 | /* calculate vm dirtyrate */ |
0e21bf24 | 636 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { |
8244166d | 637 | dirtyrate = DirtyStat.dirty_ring.rates[i].dirty_rate; |
0e21bf24 HH |
638 | DirtyStat.dirty_ring.rates[i].dirty_rate = dirtyrate; |
639 | dirtyrate_sum += dirtyrate; | |
640 | } | |
641 | ||
642 | DirtyStat.dirty_rate = dirtyrate_sum; | |
0e21bf24 HH |
643 | } |
644 | ||
645 | static void calculate_dirtyrate_sample_vm(struct DirtyRateConfig config) | |
4240dcee | 646 | { |
cf0bbb49 CZ |
647 | struct RamblockDirtyInfo *block_dinfo = NULL; |
648 | int block_count = 0; | |
649 | int64_t msec = 0; | |
650 | int64_t initial_time; | |
651 | ||
cf0bbb49 CZ |
652 | rcu_read_lock(); |
653 | initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
654 | if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) { | |
655 | goto out; | |
656 | } | |
657 | rcu_read_unlock(); | |
658 | ||
659 | msec = config.sample_period_seconds * 1000; | |
8244166d | 660 | msec = dirty_stat_wait(msec, initial_time); |
4c437254 CZ |
661 | DirtyStat.start_time = initial_time / 1000; |
662 | DirtyStat.calc_time = msec / 1000; | |
cf0bbb49 CZ |
663 | |
664 | rcu_read_lock(); | |
665 | if (!compare_page_hash_info(block_dinfo, block_count)) { | |
666 | goto out; | |
667 | } | |
668 | ||
669 | update_dirtyrate(msec); | |
670 | ||
671 | out: | |
672 | rcu_read_unlock(); | |
673 | free_ramblock_dirty_info(block_dinfo, block_count); | |
4240dcee CZ |
674 | } |
675 | ||
0e21bf24 HH |
676 | static void calculate_dirtyrate(struct DirtyRateConfig config) |
677 | { | |
826b8bc8 HH |
678 | if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { |
679 | calculate_dirtyrate_dirty_bitmap(config); | |
680 | } else if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
0e21bf24 HH |
681 | calculate_dirtyrate_dirty_ring(config); |
682 | } else { | |
683 | calculate_dirtyrate_sample_vm(config); | |
684 | } | |
685 | ||
686 | trace_dirtyrate_calculate(DirtyStat.dirty_rate); | |
687 | } | |
688 | ||
4240dcee CZ |
689 | void *get_dirtyrate_thread(void *arg) |
690 | { | |
691 | struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg; | |
7df3aa30 | 692 | int ret; |
15eb2d64 | 693 | rcu_register_thread(); |
7df3aa30 CZ |
694 | |
695 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED, | |
696 | DIRTY_RATE_STATUS_MEASURING); | |
697 | if (ret == -1) { | |
698 | error_report("change dirtyrate state failed."); | |
699 | return NULL; | |
700 | } | |
4240dcee CZ |
701 | |
702 | calculate_dirtyrate(config); | |
703 | ||
7df3aa30 CZ |
704 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING, |
705 | DIRTY_RATE_STATUS_MEASURED); | |
706 | if (ret == -1) { | |
707 | error_report("change dirtyrate state failed."); | |
708 | } | |
15eb2d64 HH |
709 | |
710 | rcu_unregister_thread(); | |
4240dcee CZ |
711 | return NULL; |
712 | } | |
4c437254 | 713 | |
0e21bf24 HH |
714 | void qmp_calc_dirty_rate(int64_t calc_time, |
715 | bool has_sample_pages, | |
716 | int64_t sample_pages, | |
717 | bool has_mode, | |
718 | DirtyRateMeasureMode mode, | |
719 | Error **errp) | |
4c437254 CZ |
720 | { |
721 | static struct DirtyRateConfig config; | |
722 | QemuThread thread; | |
723 | int ret; | |
9865d0f6 | 724 | int64_t start_time; |
4c437254 CZ |
725 | |
726 | /* | |
727 | * If the dirty rate is already being measured, don't attempt to start. | |
728 | */ | |
729 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) { | |
730 | error_setg(errp, "the dirty rate is already being measured."); | |
731 | return; | |
732 | } | |
733 | ||
734 | if (!is_sample_period_valid(calc_time)) { | |
735 | error_setg(errp, "calc-time is out of range[%d, %d].", | |
736 | MIN_FETCH_DIRTYRATE_TIME_SEC, | |
737 | MAX_FETCH_DIRTYRATE_TIME_SEC); | |
738 | return; | |
739 | } | |
740 | ||
0e21bf24 HH |
741 | if (!has_mode) { |
742 | mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
743 | } | |
744 | ||
bd9510d3 ZD |
745 | if (has_sample_pages && mode != DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
746 | error_setg(errp, "sample-pages is used only in page-sampling mode"); | |
0e21bf24 HH |
747 | return; |
748 | } | |
749 | ||
7afa08cd HH |
750 | if (has_sample_pages) { |
751 | if (!is_sample_pages_valid(sample_pages)) { | |
752 | error_setg(errp, "sample-pages is out of range[%d, %d].", | |
753 | MIN_SAMPLE_PAGE_COUNT, | |
754 | MAX_SAMPLE_PAGE_COUNT); | |
755 | return; | |
756 | } | |
757 | } else { | |
758 | sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES; | |
759 | } | |
760 | ||
0e21bf24 HH |
761 | /* |
762 | * dirty ring mode only works when kvm dirty ring is enabled. | |
826b8bc8 | 763 | * on the contrary, dirty bitmap mode is not. |
0e21bf24 | 764 | */ |
826b8bc8 HH |
765 | if (((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) && |
766 | !kvm_dirty_ring_enabled()) || | |
767 | ((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) && | |
768 | kvm_dirty_ring_enabled())) { | |
769 | error_setg(errp, "mode %s is not enabled, use other method instead.", | |
770 | DirtyRateMeasureMode_str(mode)); | |
771 | return; | |
0e21bf24 HH |
772 | } |
773 | ||
4c437254 CZ |
774 | /* |
775 | * Init calculation state as unstarted. | |
776 | */ | |
777 | ret = dirtyrate_set_state(&CalculatingState, CalculatingState, | |
778 | DIRTY_RATE_STATUS_UNSTARTED); | |
779 | if (ret == -1) { | |
780 | error_setg(errp, "init dirty rate calculation state failed."); | |
781 | return; | |
782 | } | |
783 | ||
784 | config.sample_period_seconds = calc_time; | |
7afa08cd | 785 | config.sample_pages_per_gigabytes = sample_pages; |
0e21bf24 HH |
786 | config.mode = mode; |
787 | ||
788 | cleanup_dirtyrate_stat(config); | |
789 | ||
790 | /* | |
791 | * update dirty rate mode so that we can figure out what mode has | |
792 | * been used in last calculation | |
793 | **/ | |
794 | dirtyrate_mode = mode; | |
9865d0f6 HH |
795 | |
796 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; | |
797 | init_dirtyrate_stat(start_time, config); | |
798 | ||
4c437254 CZ |
799 | qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread, |
800 | (void *)&config, QEMU_THREAD_DETACHED); | |
801 | } | |
802 | ||
803 | struct DirtyRateInfo *qmp_query_dirty_rate(Error **errp) | |
804 | { | |
805 | return query_dirty_rate_info(); | |
806 | } | |
a4a571d9 PX |
807 | |
808 | void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict) | |
809 | { | |
810 | DirtyRateInfo *info = query_dirty_rate_info(); | |
811 | ||
812 | monitor_printf(mon, "Status: %s\n", | |
813 | DirtyRateStatus_str(info->status)); | |
814 | monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n", | |
815 | info->start_time); | |
bd9510d3 ZD |
816 | if (info->mode == DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
817 | monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n", | |
818 | info->sample_pages); | |
819 | } | |
a4a571d9 PX |
820 | monitor_printf(mon, "Period: %"PRIi64" (sec)\n", |
821 | info->calc_time); | |
0e21bf24 HH |
822 | monitor_printf(mon, "Mode: %s\n", |
823 | DirtyRateMeasureMode_str(info->mode)); | |
a4a571d9 PX |
824 | monitor_printf(mon, "Dirty rate: "); |
825 | if (info->has_dirty_rate) { | |
826 | monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate); | |
0e21bf24 HH |
827 | if (info->has_vcpu_dirty_rate) { |
828 | DirtyRateVcpuList *rate, *head = info->vcpu_dirty_rate; | |
829 | for (rate = head; rate != NULL; rate = rate->next) { | |
830 | monitor_printf(mon, "vcpu[%"PRIi64"], Dirty rate: %"PRIi64 | |
831 | " (MB/s)\n", rate->value->id, | |
832 | rate->value->dirty_rate); | |
833 | } | |
834 | } | |
a4a571d9 PX |
835 | } else { |
836 | monitor_printf(mon, "(not ready)\n"); | |
837 | } | |
0e21bf24 HH |
838 | |
839 | qapi_free_DirtyRateVcpuList(info->vcpu_dirty_rate); | |
a4a571d9 PX |
840 | g_free(info); |
841 | } | |
842 | ||
843 | void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict) | |
844 | { | |
845 | int64_t sec = qdict_get_try_int(qdict, "second", 0); | |
846 | int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1); | |
847 | bool has_sample_pages = (sample_pages != -1); | |
0e21bf24 | 848 | bool dirty_ring = qdict_get_try_bool(qdict, "dirty_ring", false); |
826b8bc8 HH |
849 | bool dirty_bitmap = qdict_get_try_bool(qdict, "dirty_bitmap", false); |
850 | DirtyRateMeasureMode mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
a4a571d9 PX |
851 | Error *err = NULL; |
852 | ||
853 | if (!sec) { | |
854 | monitor_printf(mon, "Incorrect period length specified!\n"); | |
855 | return; | |
856 | } | |
857 | ||
826b8bc8 HH |
858 | if (dirty_ring && dirty_bitmap) { |
859 | monitor_printf(mon, "Either dirty ring or dirty bitmap " | |
860 | "can be specified!\n"); | |
861 | return; | |
862 | } | |
863 | ||
864 | if (dirty_bitmap) { | |
865 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP; | |
866 | } else if (dirty_ring) { | |
867 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_RING; | |
868 | } | |
869 | ||
0e21bf24 HH |
870 | qmp_calc_dirty_rate(sec, has_sample_pages, sample_pages, true, |
871 | mode, &err); | |
a4a571d9 PX |
872 | if (err) { |
873 | hmp_handle_error(mon, err); | |
874 | return; | |
875 | } | |
876 | ||
877 | monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64 | |
878 | " seconds\n", sec); | |
879 | monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n"); | |
880 | } |