]>
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); | |
3eb82637 AG |
60 | /* g_usleep may overshoot */ |
61 | msec = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - initial_time; | |
eca58224 CZ |
62 | } |
63 | ||
64 | return msec; | |
65 | } | |
66 | ||
8244166d HH |
67 | static inline void record_dirtypages(DirtyPageRecord *dirty_pages, |
68 | CPUState *cpu, bool start) | |
69 | { | |
70 | if (start) { | |
71 | dirty_pages[cpu->cpu_index].start_pages = cpu->dirty_pages; | |
72 | } else { | |
73 | dirty_pages[cpu->cpu_index].end_pages = cpu->dirty_pages; | |
74 | } | |
75 | } | |
76 | ||
77 | static int64_t do_calculate_dirtyrate(DirtyPageRecord dirty_pages, | |
78 | int64_t calc_time_ms) | |
79 | { | |
8244166d HH |
80 | uint64_t increased_dirty_pages = |
81 | dirty_pages.end_pages - dirty_pages.start_pages; | |
82 | ||
3eb82637 AG |
83 | /* |
84 | * multiply by 1000ms/s _before_ converting down to megabytes | |
85 | * to avoid losing precision | |
86 | */ | |
87 | return qemu_target_pages_to_MiB(increased_dirty_pages * 1000) / | |
88 | calc_time_ms; | |
8244166d HH |
89 | } |
90 | ||
91 | void global_dirty_log_change(unsigned int flag, bool start) | |
92 | { | |
93 | qemu_mutex_lock_iothread(); | |
94 | if (start) { | |
95 | memory_global_dirty_log_start(flag); | |
96 | } else { | |
97 | memory_global_dirty_log_stop(flag); | |
98 | } | |
99 | qemu_mutex_unlock_iothread(); | |
100 | } | |
101 | ||
102 | /* | |
103 | * global_dirty_log_sync | |
104 | * 1. sync dirty log from kvm | |
105 | * 2. stop dirty tracking if needed. | |
106 | */ | |
107 | static void global_dirty_log_sync(unsigned int flag, bool one_shot) | |
108 | { | |
109 | qemu_mutex_lock_iothread(); | |
1e493be5 | 110 | memory_global_dirty_log_sync(false); |
8244166d HH |
111 | if (one_shot) { |
112 | memory_global_dirty_log_stop(flag); | |
113 | } | |
114 | qemu_mutex_unlock_iothread(); | |
115 | } | |
116 | ||
117 | static DirtyPageRecord *vcpu_dirty_stat_alloc(VcpuStat *stat) | |
118 | { | |
119 | CPUState *cpu; | |
8244166d HH |
120 | int nvcpu = 0; |
121 | ||
122 | CPU_FOREACH(cpu) { | |
123 | nvcpu++; | |
124 | } | |
125 | ||
126 | stat->nvcpu = nvcpu; | |
c5e8d518 | 127 | stat->rates = g_new0(DirtyRateVcpu, nvcpu); |
8244166d | 128 | |
66997c42 | 129 | return g_new0(DirtyPageRecord, nvcpu); |
8244166d HH |
130 | } |
131 | ||
132 | static void vcpu_dirty_stat_collect(VcpuStat *stat, | |
133 | DirtyPageRecord *records, | |
134 | bool start) | |
135 | { | |
136 | CPUState *cpu; | |
137 | ||
138 | CPU_FOREACH(cpu) { | |
139 | record_dirtypages(records, cpu, start); | |
140 | } | |
141 | } | |
142 | ||
143 | int64_t vcpu_calculate_dirtyrate(int64_t calc_time_ms, | |
144 | VcpuStat *stat, | |
145 | unsigned int flag, | |
146 | bool one_shot) | |
147 | { | |
148 | DirtyPageRecord *records; | |
149 | int64_t init_time_ms; | |
150 | int64_t duration; | |
151 | int64_t dirtyrate; | |
152 | int i = 0; | |
153 | unsigned int gen_id; | |
154 | ||
155 | retry: | |
156 | init_time_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
157 | ||
370ed600 JI |
158 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
159 | gen_id = cpu_list_generation_id_get(); | |
160 | records = vcpu_dirty_stat_alloc(stat); | |
161 | vcpu_dirty_stat_collect(stat, records, true); | |
162 | } | |
8244166d HH |
163 | |
164 | duration = dirty_stat_wait(calc_time_ms, init_time_ms); | |
165 | ||
166 | global_dirty_log_sync(flag, one_shot); | |
167 | ||
370ed600 JI |
168 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
169 | if (gen_id != cpu_list_generation_id_get()) { | |
170 | g_free(records); | |
171 | g_free(stat->rates); | |
172 | cpu_list_unlock(); | |
173 | goto retry; | |
174 | } | |
175 | vcpu_dirty_stat_collect(stat, records, false); | |
8244166d | 176 | } |
8244166d HH |
177 | |
178 | for (i = 0; i < stat->nvcpu; i++) { | |
179 | dirtyrate = do_calculate_dirtyrate(records[i], duration); | |
180 | ||
181 | stat->rates[i].id = i; | |
182 | stat->rates[i].dirty_rate = dirtyrate; | |
183 | ||
184 | trace_dirtyrate_do_calculate_vcpu(i, dirtyrate); | |
185 | } | |
186 | ||
187 | g_free(records); | |
188 | ||
189 | return duration; | |
190 | } | |
191 | ||
eca58224 CZ |
192 | static bool is_sample_period_valid(int64_t sec) |
193 | { | |
194 | if (sec < MIN_FETCH_DIRTYRATE_TIME_SEC || | |
195 | sec > MAX_FETCH_DIRTYRATE_TIME_SEC) { | |
196 | return false; | |
197 | } | |
198 | ||
199 | return true; | |
200 | } | |
201 | ||
7afa08cd HH |
202 | static bool is_sample_pages_valid(int64_t pages) |
203 | { | |
204 | return pages >= MIN_SAMPLE_PAGE_COUNT && | |
205 | pages <= MAX_SAMPLE_PAGE_COUNT; | |
206 | } | |
207 | ||
7df3aa30 CZ |
208 | static int dirtyrate_set_state(int *state, int old_state, int new_state) |
209 | { | |
210 | assert(new_state < DIRTY_RATE_STATUS__MAX); | |
3c0b5dff | 211 | trace_dirtyrate_set_state(DirtyRateStatus_str(new_state)); |
7df3aa30 CZ |
212 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
213 | return 0; | |
214 | } else { | |
215 | return -1; | |
216 | } | |
217 | } | |
218 | ||
4c437254 CZ |
219 | static struct DirtyRateInfo *query_dirty_rate_info(void) |
220 | { | |
0e21bf24 | 221 | int i; |
4c437254 | 222 | int64_t dirty_rate = DirtyStat.dirty_rate; |
b21e2380 | 223 | struct DirtyRateInfo *info = g_new0(DirtyRateInfo, 1); |
0e21bf24 | 224 | DirtyRateVcpuList *head = NULL, **tail = &head; |
4c437254 CZ |
225 | |
226 | info->status = CalculatingState; | |
227 | info->start_time = DirtyStat.start_time; | |
228 | info->calc_time = DirtyStat.calc_time; | |
7afa08cd | 229 | info->sample_pages = DirtyStat.sample_pages; |
0e21bf24 HH |
230 | info->mode = dirtyrate_mode; |
231 | ||
232 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) { | |
233 | info->has_dirty_rate = true; | |
234 | info->dirty_rate = dirty_rate; | |
235 | ||
236 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
237 | /* | |
238 | * set sample_pages with 0 to indicate page sampling | |
239 | * isn't enabled | |
240 | **/ | |
241 | info->sample_pages = 0; | |
242 | info->has_vcpu_dirty_rate = true; | |
243 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { | |
b21e2380 | 244 | DirtyRateVcpu *rate = g_new0(DirtyRateVcpu, 1); |
0e21bf24 HH |
245 | rate->id = DirtyStat.dirty_ring.rates[i].id; |
246 | rate->dirty_rate = DirtyStat.dirty_ring.rates[i].dirty_rate; | |
247 | QAPI_LIST_APPEND(tail, rate); | |
248 | } | |
249 | info->vcpu_dirty_rate = head; | |
250 | } | |
826b8bc8 HH |
251 | |
252 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { | |
253 | info->sample_pages = 0; | |
254 | } | |
0e21bf24 | 255 | } |
4c437254 | 256 | |
3c0b5dff CZ |
257 | trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState)); |
258 | ||
4c437254 CZ |
259 | return info; |
260 | } | |
261 | ||
71864ead HH |
262 | static void init_dirtyrate_stat(int64_t start_time, |
263 | struct DirtyRateConfig config) | |
c9a58d71 | 264 | { |
c9a58d71 | 265 | DirtyStat.dirty_rate = -1; |
aa84b506 | 266 | DirtyStat.start_time = start_time; |
71864ead HH |
267 | DirtyStat.calc_time = config.sample_period_seconds; |
268 | DirtyStat.sample_pages = config.sample_pages_per_gigabytes; | |
269 | ||
270 | switch (config.mode) { | |
271 | case DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING: | |
272 | DirtyStat.page_sampling.total_dirty_samples = 0; | |
273 | DirtyStat.page_sampling.total_sample_count = 0; | |
274 | DirtyStat.page_sampling.total_block_mem_MB = 0; | |
275 | break; | |
276 | case DIRTY_RATE_MEASURE_MODE_DIRTY_RING: | |
277 | DirtyStat.dirty_ring.nvcpu = -1; | |
278 | DirtyStat.dirty_ring.rates = NULL; | |
279 | break; | |
280 | default: | |
281 | break; | |
282 | } | |
c9a58d71 CZ |
283 | } |
284 | ||
0e21bf24 HH |
285 | static void cleanup_dirtyrate_stat(struct DirtyRateConfig config) |
286 | { | |
287 | /* last calc-dirty-rate qmp use dirty ring mode */ | |
288 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
289 | free(DirtyStat.dirty_ring.rates); | |
290 | DirtyStat.dirty_ring.rates = NULL; | |
291 | } | |
292 | } | |
293 | ||
c9a58d71 CZ |
294 | static void update_dirtyrate_stat(struct RamblockDirtyInfo *info) |
295 | { | |
71864ead HH |
296 | DirtyStat.page_sampling.total_dirty_samples += info->sample_dirty_count; |
297 | DirtyStat.page_sampling.total_sample_count += info->sample_pages_count; | |
c9a58d71 | 298 | /* size of total pages in MB */ |
beeda9b7 JQ |
299 | DirtyStat.page_sampling.total_block_mem_MB += |
300 | qemu_target_pages_to_MiB(info->ramblock_pages); | |
c9a58d71 CZ |
301 | } |
302 | ||
303 | static void update_dirtyrate(uint64_t msec) | |
304 | { | |
305 | uint64_t dirtyrate; | |
71864ead HH |
306 | uint64_t total_dirty_samples = DirtyStat.page_sampling.total_dirty_samples; |
307 | uint64_t total_sample_count = DirtyStat.page_sampling.total_sample_count; | |
308 | uint64_t total_block_mem_MB = DirtyStat.page_sampling.total_block_mem_MB; | |
c9a58d71 CZ |
309 | |
310 | dirtyrate = total_dirty_samples * total_block_mem_MB * | |
311 | 1000 / (total_sample_count * msec); | |
312 | ||
313 | DirtyStat.dirty_rate = dirtyrate; | |
314 | } | |
7df3aa30 | 315 | |
00a3f9c6 AG |
316 | /* |
317 | * Compute hash of a single page of size TARGET_PAGE_SIZE. | |
318 | */ | |
319 | static uint32_t compute_page_hash(void *ptr) | |
320 | { | |
edd83a70 | 321 | size_t page_size = qemu_target_page_size(); |
00a3f9c6 AG |
322 | uint32_t i; |
323 | uint64_t v1, v2, v3, v4; | |
324 | uint64_t res; | |
325 | const uint64_t *p = ptr; | |
326 | ||
327 | v1 = QEMU_XXHASH_SEED + XXH_PRIME64_1 + XXH_PRIME64_2; | |
328 | v2 = QEMU_XXHASH_SEED + XXH_PRIME64_2; | |
329 | v3 = QEMU_XXHASH_SEED + 0; | |
330 | v4 = QEMU_XXHASH_SEED - XXH_PRIME64_1; | |
edd83a70 | 331 | for (i = 0; i < page_size / 8; i += 4) { |
00a3f9c6 AG |
332 | v1 = XXH64_round(v1, p[i + 0]); |
333 | v2 = XXH64_round(v2, p[i + 1]); | |
334 | v3 = XXH64_round(v3, p[i + 2]); | |
335 | v4 = XXH64_round(v4, p[i + 3]); | |
336 | } | |
337 | res = XXH64_mergerounds(v1, v2, v3, v4); | |
edd83a70 | 338 | res += page_size; |
00a3f9c6 AG |
339 | res = XXH64_avalanche(res); |
340 | return (uint32_t)(res & UINT32_MAX); | |
341 | } | |
342 | ||
343 | ||
ba0e519f CZ |
344 | /* |
345 | * get hash result for the sampled memory with length of TARGET_PAGE_SIZE | |
346 | * in ramblock, which starts from ramblock base address. | |
347 | */ | |
348 | static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info, | |
349 | uint64_t vfn) | |
350 | { | |
00a3f9c6 | 351 | uint32_t hash; |
ba0e519f | 352 | |
edd83a70 JQ |
353 | hash = compute_page_hash(info->ramblock_addr + |
354 | vfn * qemu_target_page_size()); | |
ba0e519f | 355 | |
00a3f9c6 AG |
356 | trace_get_ramblock_vfn_hash(info->idstr, vfn, hash); |
357 | return hash; | |
ba0e519f CZ |
358 | } |
359 | ||
360 | static bool save_ramblock_hash(struct RamblockDirtyInfo *info) | |
361 | { | |
362 | unsigned int sample_pages_count; | |
363 | int i; | |
364 | GRand *rand; | |
365 | ||
366 | sample_pages_count = info->sample_pages_count; | |
367 | ||
368 | /* ramblock size less than one page, return success to skip this ramblock */ | |
369 | if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) { | |
370 | return true; | |
371 | } | |
372 | ||
373 | info->hash_result = g_try_malloc0_n(sample_pages_count, | |
374 | sizeof(uint32_t)); | |
375 | if (!info->hash_result) { | |
376 | return false; | |
377 | } | |
378 | ||
379 | info->sample_page_vfn = g_try_malloc0_n(sample_pages_count, | |
380 | sizeof(uint64_t)); | |
381 | if (!info->sample_page_vfn) { | |
382 | g_free(info->hash_result); | |
383 | return false; | |
384 | } | |
385 | ||
386 | rand = g_rand_new(); | |
387 | for (i = 0; i < sample_pages_count; i++) { | |
388 | info->sample_page_vfn[i] = g_rand_int_range(rand, 0, | |
389 | info->ramblock_pages - 1); | |
390 | info->hash_result[i] = get_ramblock_vfn_hash(info, | |
391 | info->sample_page_vfn[i]); | |
392 | } | |
393 | g_rand_free(rand); | |
394 | ||
395 | return true; | |
396 | } | |
397 | ||
398 | static void get_ramblock_dirty_info(RAMBlock *block, | |
399 | struct RamblockDirtyInfo *info, | |
400 | struct DirtyRateConfig *config) | |
401 | { | |
402 | uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes; | |
403 | ||
404 | /* Right shift 30 bits to calc ramblock size in GB */ | |
405 | info->sample_pages_count = (qemu_ram_get_used_length(block) * | |
406 | sample_pages_per_gigabytes) >> 30; | |
407 | /* Right shift TARGET_PAGE_BITS to calc page count */ | |
408 | info->ramblock_pages = qemu_ram_get_used_length(block) >> | |
148b1ad8 | 409 | qemu_target_page_bits(); |
ba0e519f CZ |
410 | info->ramblock_addr = qemu_ram_get_host_addr(block); |
411 | strcpy(info->idstr, qemu_ram_get_idstr(block)); | |
412 | } | |
413 | ||
cf0bbb49 CZ |
414 | static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count) |
415 | { | |
416 | int i; | |
417 | ||
418 | if (!infos) { | |
419 | return; | |
420 | } | |
421 | ||
422 | for (i = 0; i < count; i++) { | |
423 | g_free(infos[i].sample_page_vfn); | |
424 | g_free(infos[i].hash_result); | |
425 | } | |
426 | g_free(infos); | |
427 | } | |
428 | ||
f82583cd CZ |
429 | static bool skip_sample_ramblock(RAMBlock *block) |
430 | { | |
431 | /* | |
432 | * Sample only blocks larger than MIN_RAMBLOCK_SIZE. | |
433 | */ | |
434 | if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) { | |
3c0b5dff CZ |
435 | trace_skip_sample_ramblock(block->idstr, |
436 | qemu_ram_get_used_length(block)); | |
f82583cd CZ |
437 | return true; |
438 | } | |
439 | ||
440 | return false; | |
441 | } | |
442 | ||
ba0e519f CZ |
443 | static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo, |
444 | struct DirtyRateConfig config, | |
445 | int *block_count) | |
446 | { | |
447 | struct RamblockDirtyInfo *info = NULL; | |
448 | struct RamblockDirtyInfo *dinfo = NULL; | |
449 | RAMBlock *block = NULL; | |
450 | int total_count = 0; | |
451 | int index = 0; | |
452 | bool ret = false; | |
453 | ||
454 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
455 | if (skip_sample_ramblock(block)) { |
456 | continue; | |
457 | } | |
ba0e519f CZ |
458 | total_count++; |
459 | } | |
460 | ||
461 | dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo)); | |
462 | if (dinfo == NULL) { | |
463 | goto out; | |
464 | } | |
465 | ||
466 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
467 | if (skip_sample_ramblock(block)) { |
468 | continue; | |
469 | } | |
ba0e519f CZ |
470 | if (index >= total_count) { |
471 | break; | |
472 | } | |
473 | info = &dinfo[index]; | |
474 | get_ramblock_dirty_info(block, info, &config); | |
475 | if (!save_ramblock_hash(info)) { | |
476 | goto out; | |
477 | } | |
478 | index++; | |
479 | } | |
480 | ret = true; | |
481 | ||
482 | out: | |
483 | *block_count = index; | |
484 | *block_dinfo = dinfo; | |
485 | return ret; | |
486 | } | |
487 | ||
9c04387b CZ |
488 | static void calc_page_dirty_rate(struct RamblockDirtyInfo *info) |
489 | { | |
00a3f9c6 | 490 | uint32_t hash; |
9c04387b CZ |
491 | int i; |
492 | ||
493 | for (i = 0; i < info->sample_pages_count; i++) { | |
00a3f9c6 AG |
494 | hash = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]); |
495 | if (hash != info->hash_result[i]) { | |
496 | trace_calc_page_dirty_rate(info->idstr, hash, info->hash_result[i]); | |
9c04387b CZ |
497 | info->sample_dirty_count++; |
498 | } | |
499 | } | |
500 | } | |
501 | ||
502 | static struct RamblockDirtyInfo * | |
503 | find_block_matched(RAMBlock *block, int count, | |
504 | struct RamblockDirtyInfo *infos) | |
505 | { | |
506 | int i; | |
9c04387b CZ |
507 | |
508 | for (i = 0; i < count; i++) { | |
509 | if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) { | |
510 | break; | |
511 | } | |
512 | } | |
513 | ||
514 | if (i == count) { | |
515 | return NULL; | |
516 | } | |
517 | ||
518 | if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) || | |
519 | infos[i].ramblock_pages != | |
148b1ad8 | 520 | (qemu_ram_get_used_length(block) >> qemu_target_page_bits())) { |
3c0b5dff | 521 | trace_find_page_matched(block->idstr); |
9c04387b CZ |
522 | return NULL; |
523 | } | |
524 | ||
66997c42 | 525 | return &infos[i]; |
9c04387b CZ |
526 | } |
527 | ||
528 | static bool compare_page_hash_info(struct RamblockDirtyInfo *info, | |
529 | int block_count) | |
530 | { | |
531 | struct RamblockDirtyInfo *block_dinfo = NULL; | |
532 | RAMBlock *block = NULL; | |
533 | ||
534 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
535 | if (skip_sample_ramblock(block)) { |
536 | continue; | |
537 | } | |
9c04387b CZ |
538 | block_dinfo = find_block_matched(block, block_count, info); |
539 | if (block_dinfo == NULL) { | |
540 | continue; | |
541 | } | |
542 | calc_page_dirty_rate(block_dinfo); | |
543 | update_dirtyrate_stat(block_dinfo); | |
544 | } | |
545 | ||
71864ead | 546 | if (DirtyStat.page_sampling.total_sample_count == 0) { |
9c04387b CZ |
547 | return false; |
548 | } | |
549 | ||
550 | return true; | |
551 | } | |
552 | ||
826b8bc8 HH |
553 | static inline void record_dirtypages_bitmap(DirtyPageRecord *dirty_pages, |
554 | bool start) | |
555 | { | |
556 | if (start) { | |
557 | dirty_pages->start_pages = total_dirty_pages; | |
558 | } else { | |
559 | dirty_pages->end_pages = total_dirty_pages; | |
560 | } | |
561 | } | |
562 | ||
826b8bc8 HH |
563 | static inline void dirtyrate_manual_reset_protect(void) |
564 | { | |
565 | RAMBlock *block = NULL; | |
566 | ||
567 | WITH_RCU_READ_LOCK_GUARD() { | |
568 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
569 | memory_region_clear_dirty_bitmap(block->mr, 0, | |
570 | block->used_length); | |
571 | } | |
572 | } | |
573 | } | |
574 | ||
575 | static void calculate_dirtyrate_dirty_bitmap(struct DirtyRateConfig config) | |
576 | { | |
577 | int64_t msec = 0; | |
578 | int64_t start_time; | |
579 | DirtyPageRecord dirty_pages; | |
580 | ||
581 | qemu_mutex_lock_iothread(); | |
582 | memory_global_dirty_log_start(GLOBAL_DIRTY_DIRTY_RATE); | |
583 | ||
584 | /* | |
585 | * 1'round of log sync may return all 1 bits with | |
586 | * KVM_DIRTY_LOG_INITIALLY_SET enable | |
587 | * skip it unconditionally and start dirty tracking | |
588 | * from 2'round of log sync | |
589 | */ | |
1e493be5 | 590 | memory_global_dirty_log_sync(false); |
826b8bc8 HH |
591 | |
592 | /* | |
593 | * reset page protect manually and unconditionally. | |
594 | * this make sure kvm dirty log be cleared if | |
595 | * KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE cap is enabled. | |
596 | */ | |
597 | dirtyrate_manual_reset_protect(); | |
598 | qemu_mutex_unlock_iothread(); | |
599 | ||
600 | record_dirtypages_bitmap(&dirty_pages, true); | |
601 | ||
602 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
603 | DirtyStat.start_time = start_time / 1000; | |
604 | ||
605 | msec = config.sample_period_seconds * 1000; | |
8244166d | 606 | msec = dirty_stat_wait(msec, start_time); |
826b8bc8 HH |
607 | DirtyStat.calc_time = msec / 1000; |
608 | ||
609 | /* | |
8244166d | 610 | * do two things. |
826b8bc8 HH |
611 | * 1. fetch dirty bitmap from kvm |
612 | * 2. stop dirty tracking | |
613 | */ | |
8244166d | 614 | global_dirty_log_sync(GLOBAL_DIRTY_DIRTY_RATE, true); |
826b8bc8 HH |
615 | |
616 | record_dirtypages_bitmap(&dirty_pages, false); | |
617 | ||
8244166d | 618 | DirtyStat.dirty_rate = do_calculate_dirtyrate(dirty_pages, msec); |
826b8bc8 HH |
619 | } |
620 | ||
0e21bf24 HH |
621 | static void calculate_dirtyrate_dirty_ring(struct DirtyRateConfig config) |
622 | { | |
8244166d | 623 | int64_t duration; |
0e21bf24 HH |
624 | uint64_t dirtyrate = 0; |
625 | uint64_t dirtyrate_sum = 0; | |
0e21bf24 HH |
626 | int i = 0; |
627 | ||
8244166d HH |
628 | /* start log sync */ |
629 | global_dirty_log_change(GLOBAL_DIRTY_DIRTY_RATE, true); | |
0e21bf24 | 630 | |
8244166d | 631 | DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; |
0e21bf24 | 632 | |
8244166d HH |
633 | /* calculate vcpu dirtyrate */ |
634 | duration = vcpu_calculate_dirtyrate(config.sample_period_seconds * 1000, | |
635 | &DirtyStat.dirty_ring, | |
636 | GLOBAL_DIRTY_DIRTY_RATE, | |
637 | true); | |
0e21bf24 | 638 | |
8244166d | 639 | DirtyStat.calc_time = duration / 1000; |
0e21bf24 | 640 | |
8244166d | 641 | /* calculate vm dirtyrate */ |
0e21bf24 | 642 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { |
8244166d | 643 | dirtyrate = DirtyStat.dirty_ring.rates[i].dirty_rate; |
0e21bf24 HH |
644 | DirtyStat.dirty_ring.rates[i].dirty_rate = dirtyrate; |
645 | dirtyrate_sum += dirtyrate; | |
646 | } | |
647 | ||
648 | DirtyStat.dirty_rate = dirtyrate_sum; | |
0e21bf24 HH |
649 | } |
650 | ||
651 | static void calculate_dirtyrate_sample_vm(struct DirtyRateConfig config) | |
4240dcee | 652 | { |
cf0bbb49 CZ |
653 | struct RamblockDirtyInfo *block_dinfo = NULL; |
654 | int block_count = 0; | |
655 | int64_t msec = 0; | |
656 | int64_t initial_time; | |
657 | ||
cf0bbb49 CZ |
658 | rcu_read_lock(); |
659 | initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
660 | if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) { | |
661 | goto out; | |
662 | } | |
663 | rcu_read_unlock(); | |
664 | ||
665 | msec = config.sample_period_seconds * 1000; | |
8244166d | 666 | msec = dirty_stat_wait(msec, initial_time); |
4c437254 CZ |
667 | DirtyStat.start_time = initial_time / 1000; |
668 | DirtyStat.calc_time = msec / 1000; | |
cf0bbb49 CZ |
669 | |
670 | rcu_read_lock(); | |
671 | if (!compare_page_hash_info(block_dinfo, block_count)) { | |
672 | goto out; | |
673 | } | |
674 | ||
675 | update_dirtyrate(msec); | |
676 | ||
677 | out: | |
678 | rcu_read_unlock(); | |
679 | free_ramblock_dirty_info(block_dinfo, block_count); | |
4240dcee CZ |
680 | } |
681 | ||
0e21bf24 HH |
682 | static void calculate_dirtyrate(struct DirtyRateConfig config) |
683 | { | |
826b8bc8 HH |
684 | if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { |
685 | calculate_dirtyrate_dirty_bitmap(config); | |
686 | } else if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
0e21bf24 HH |
687 | calculate_dirtyrate_dirty_ring(config); |
688 | } else { | |
689 | calculate_dirtyrate_sample_vm(config); | |
690 | } | |
691 | ||
692 | trace_dirtyrate_calculate(DirtyStat.dirty_rate); | |
693 | } | |
694 | ||
4240dcee CZ |
695 | void *get_dirtyrate_thread(void *arg) |
696 | { | |
697 | struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg; | |
7df3aa30 | 698 | int ret; |
15eb2d64 | 699 | rcu_register_thread(); |
7df3aa30 CZ |
700 | |
701 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED, | |
702 | DIRTY_RATE_STATUS_MEASURING); | |
703 | if (ret == -1) { | |
704 | error_report("change dirtyrate state failed."); | |
705 | return NULL; | |
706 | } | |
4240dcee CZ |
707 | |
708 | calculate_dirtyrate(config); | |
709 | ||
7df3aa30 CZ |
710 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING, |
711 | DIRTY_RATE_STATUS_MEASURED); | |
712 | if (ret == -1) { | |
713 | error_report("change dirtyrate state failed."); | |
714 | } | |
15eb2d64 HH |
715 | |
716 | rcu_unregister_thread(); | |
4240dcee CZ |
717 | return NULL; |
718 | } | |
4c437254 | 719 | |
0e21bf24 HH |
720 | void qmp_calc_dirty_rate(int64_t calc_time, |
721 | bool has_sample_pages, | |
722 | int64_t sample_pages, | |
723 | bool has_mode, | |
724 | DirtyRateMeasureMode mode, | |
725 | Error **errp) | |
4c437254 CZ |
726 | { |
727 | static struct DirtyRateConfig config; | |
728 | QemuThread thread; | |
729 | int ret; | |
9865d0f6 | 730 | int64_t start_time; |
4c437254 CZ |
731 | |
732 | /* | |
733 | * If the dirty rate is already being measured, don't attempt to start. | |
734 | */ | |
735 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) { | |
736 | error_setg(errp, "the dirty rate is already being measured."); | |
737 | return; | |
738 | } | |
739 | ||
740 | if (!is_sample_period_valid(calc_time)) { | |
741 | error_setg(errp, "calc-time is out of range[%d, %d].", | |
742 | MIN_FETCH_DIRTYRATE_TIME_SEC, | |
743 | MAX_FETCH_DIRTYRATE_TIME_SEC); | |
744 | return; | |
745 | } | |
746 | ||
0e21bf24 HH |
747 | if (!has_mode) { |
748 | mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
749 | } | |
750 | ||
bd9510d3 ZD |
751 | if (has_sample_pages && mode != DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
752 | error_setg(errp, "sample-pages is used only in page-sampling mode"); | |
0e21bf24 HH |
753 | return; |
754 | } | |
755 | ||
7afa08cd HH |
756 | if (has_sample_pages) { |
757 | if (!is_sample_pages_valid(sample_pages)) { | |
758 | error_setg(errp, "sample-pages is out of range[%d, %d].", | |
759 | MIN_SAMPLE_PAGE_COUNT, | |
760 | MAX_SAMPLE_PAGE_COUNT); | |
761 | return; | |
762 | } | |
763 | } else { | |
764 | sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES; | |
765 | } | |
766 | ||
0e21bf24 HH |
767 | /* |
768 | * dirty ring mode only works when kvm dirty ring is enabled. | |
826b8bc8 | 769 | * on the contrary, dirty bitmap mode is not. |
0e21bf24 | 770 | */ |
826b8bc8 HH |
771 | if (((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) && |
772 | !kvm_dirty_ring_enabled()) || | |
773 | ((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) && | |
774 | kvm_dirty_ring_enabled())) { | |
775 | error_setg(errp, "mode %s is not enabled, use other method instead.", | |
776 | DirtyRateMeasureMode_str(mode)); | |
777 | return; | |
0e21bf24 HH |
778 | } |
779 | ||
4c437254 CZ |
780 | /* |
781 | * Init calculation state as unstarted. | |
782 | */ | |
783 | ret = dirtyrate_set_state(&CalculatingState, CalculatingState, | |
784 | DIRTY_RATE_STATUS_UNSTARTED); | |
785 | if (ret == -1) { | |
786 | error_setg(errp, "init dirty rate calculation state failed."); | |
787 | return; | |
788 | } | |
789 | ||
790 | config.sample_period_seconds = calc_time; | |
7afa08cd | 791 | config.sample_pages_per_gigabytes = sample_pages; |
0e21bf24 HH |
792 | config.mode = mode; |
793 | ||
794 | cleanup_dirtyrate_stat(config); | |
795 | ||
796 | /* | |
797 | * update dirty rate mode so that we can figure out what mode has | |
798 | * been used in last calculation | |
799 | **/ | |
800 | dirtyrate_mode = mode; | |
9865d0f6 HH |
801 | |
802 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; | |
803 | init_dirtyrate_stat(start_time, config); | |
804 | ||
4c437254 CZ |
805 | qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread, |
806 | (void *)&config, QEMU_THREAD_DETACHED); | |
807 | } | |
808 | ||
809 | struct DirtyRateInfo *qmp_query_dirty_rate(Error **errp) | |
810 | { | |
811 | return query_dirty_rate_info(); | |
812 | } | |
a4a571d9 PX |
813 | |
814 | void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict) | |
815 | { | |
816 | DirtyRateInfo *info = query_dirty_rate_info(); | |
817 | ||
818 | monitor_printf(mon, "Status: %s\n", | |
819 | DirtyRateStatus_str(info->status)); | |
820 | monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n", | |
821 | info->start_time); | |
bd9510d3 ZD |
822 | if (info->mode == DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
823 | monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n", | |
824 | info->sample_pages); | |
825 | } | |
a4a571d9 PX |
826 | monitor_printf(mon, "Period: %"PRIi64" (sec)\n", |
827 | info->calc_time); | |
0e21bf24 HH |
828 | monitor_printf(mon, "Mode: %s\n", |
829 | DirtyRateMeasureMode_str(info->mode)); | |
a4a571d9 PX |
830 | monitor_printf(mon, "Dirty rate: "); |
831 | if (info->has_dirty_rate) { | |
832 | monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate); | |
0e21bf24 HH |
833 | if (info->has_vcpu_dirty_rate) { |
834 | DirtyRateVcpuList *rate, *head = info->vcpu_dirty_rate; | |
835 | for (rate = head; rate != NULL; rate = rate->next) { | |
836 | monitor_printf(mon, "vcpu[%"PRIi64"], Dirty rate: %"PRIi64 | |
837 | " (MB/s)\n", rate->value->id, | |
838 | rate->value->dirty_rate); | |
839 | } | |
840 | } | |
a4a571d9 PX |
841 | } else { |
842 | monitor_printf(mon, "(not ready)\n"); | |
843 | } | |
0e21bf24 HH |
844 | |
845 | qapi_free_DirtyRateVcpuList(info->vcpu_dirty_rate); | |
a4a571d9 PX |
846 | g_free(info); |
847 | } | |
848 | ||
849 | void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict) | |
850 | { | |
851 | int64_t sec = qdict_get_try_int(qdict, "second", 0); | |
852 | int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1); | |
853 | bool has_sample_pages = (sample_pages != -1); | |
0e21bf24 | 854 | bool dirty_ring = qdict_get_try_bool(qdict, "dirty_ring", false); |
826b8bc8 HH |
855 | bool dirty_bitmap = qdict_get_try_bool(qdict, "dirty_bitmap", false); |
856 | DirtyRateMeasureMode mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
a4a571d9 PX |
857 | Error *err = NULL; |
858 | ||
859 | if (!sec) { | |
860 | monitor_printf(mon, "Incorrect period length specified!\n"); | |
861 | return; | |
862 | } | |
863 | ||
826b8bc8 HH |
864 | if (dirty_ring && dirty_bitmap) { |
865 | monitor_printf(mon, "Either dirty ring or dirty bitmap " | |
866 | "can be specified!\n"); | |
867 | return; | |
868 | } | |
869 | ||
870 | if (dirty_bitmap) { | |
871 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP; | |
872 | } else if (dirty_ring) { | |
873 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_RING; | |
874 | } | |
875 | ||
0e21bf24 HH |
876 | qmp_calc_dirty_rate(sec, has_sample_pages, sample_pages, true, |
877 | mode, &err); | |
a4a571d9 PX |
878 | if (err) { |
879 | hmp_handle_error(mon, err); | |
880 | return; | |
881 | } | |
882 | ||
883 | monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64 | |
884 | " seconds\n", sec); | |
885 | monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n"); | |
886 | } |