]>
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> |
4240dcee CZ |
16 | #include "qapi/error.h" |
17 | #include "cpu.h" | |
4240dcee | 18 | #include "exec/ramblock.h" |
826b8bc8 | 19 | #include "exec/ram_addr.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" | |
32 | ||
4998a37e HH |
33 | /* |
34 | * total_dirty_pages is procted by BQL and is used | |
35 | * to stat dirty pages during the period of two | |
36 | * memory_global_dirty_log_sync | |
37 | */ | |
38 | uint64_t total_dirty_pages; | |
39 | ||
0e21bf24 HH |
40 | typedef struct DirtyPageRecord { |
41 | uint64_t start_pages; | |
42 | uint64_t end_pages; | |
43 | } DirtyPageRecord; | |
4240dcee | 44 | |
7df3aa30 | 45 | static int CalculatingState = DIRTY_RATE_STATUS_UNSTARTED; |
c9a58d71 | 46 | static struct DirtyRateStat DirtyStat; |
0e21bf24 HH |
47 | static DirtyRateMeasureMode dirtyrate_mode = |
48 | DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
7df3aa30 | 49 | |
8244166d | 50 | static int64_t dirty_stat_wait(int64_t msec, int64_t initial_time) |
eca58224 CZ |
51 | { |
52 | int64_t current_time; | |
53 | ||
54 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
55 | if ((current_time - initial_time) >= msec) { | |
56 | msec = current_time - initial_time; | |
57 | } else { | |
58 | g_usleep((msec + initial_time - current_time) * 1000); | |
59 | } | |
60 | ||
61 | return msec; | |
62 | } | |
63 | ||
8244166d HH |
64 | static inline void record_dirtypages(DirtyPageRecord *dirty_pages, |
65 | CPUState *cpu, bool start) | |
66 | { | |
67 | if (start) { | |
68 | dirty_pages[cpu->cpu_index].start_pages = cpu->dirty_pages; | |
69 | } else { | |
70 | dirty_pages[cpu->cpu_index].end_pages = cpu->dirty_pages; | |
71 | } | |
72 | } | |
73 | ||
74 | static int64_t do_calculate_dirtyrate(DirtyPageRecord dirty_pages, | |
75 | int64_t calc_time_ms) | |
76 | { | |
77 | uint64_t memory_size_MB; | |
78 | uint64_t increased_dirty_pages = | |
79 | dirty_pages.end_pages - dirty_pages.start_pages; | |
80 | ||
81 | memory_size_MB = (increased_dirty_pages * TARGET_PAGE_SIZE) >> 20; | |
82 | ||
83 | return memory_size_MB * 1000 / calc_time_ms; | |
84 | } | |
85 | ||
86 | void global_dirty_log_change(unsigned int flag, bool start) | |
87 | { | |
88 | qemu_mutex_lock_iothread(); | |
89 | if (start) { | |
90 | memory_global_dirty_log_start(flag); | |
91 | } else { | |
92 | memory_global_dirty_log_stop(flag); | |
93 | } | |
94 | qemu_mutex_unlock_iothread(); | |
95 | } | |
96 | ||
97 | /* | |
98 | * global_dirty_log_sync | |
99 | * 1. sync dirty log from kvm | |
100 | * 2. stop dirty tracking if needed. | |
101 | */ | |
102 | static void global_dirty_log_sync(unsigned int flag, bool one_shot) | |
103 | { | |
104 | qemu_mutex_lock_iothread(); | |
105 | memory_global_dirty_log_sync(); | |
106 | if (one_shot) { | |
107 | memory_global_dirty_log_stop(flag); | |
108 | } | |
109 | qemu_mutex_unlock_iothread(); | |
110 | } | |
111 | ||
112 | static DirtyPageRecord *vcpu_dirty_stat_alloc(VcpuStat *stat) | |
113 | { | |
114 | CPUState *cpu; | |
8244166d HH |
115 | int nvcpu = 0; |
116 | ||
117 | CPU_FOREACH(cpu) { | |
118 | nvcpu++; | |
119 | } | |
120 | ||
121 | stat->nvcpu = nvcpu; | |
c5e8d518 | 122 | stat->rates = g_new0(DirtyRateVcpu, nvcpu); |
8244166d | 123 | |
66997c42 | 124 | return g_new0(DirtyPageRecord, nvcpu); |
8244166d HH |
125 | } |
126 | ||
127 | static void vcpu_dirty_stat_collect(VcpuStat *stat, | |
128 | DirtyPageRecord *records, | |
129 | bool start) | |
130 | { | |
131 | CPUState *cpu; | |
132 | ||
133 | CPU_FOREACH(cpu) { | |
134 | record_dirtypages(records, cpu, start); | |
135 | } | |
136 | } | |
137 | ||
138 | int64_t vcpu_calculate_dirtyrate(int64_t calc_time_ms, | |
139 | VcpuStat *stat, | |
140 | unsigned int flag, | |
141 | bool one_shot) | |
142 | { | |
143 | DirtyPageRecord *records; | |
144 | int64_t init_time_ms; | |
145 | int64_t duration; | |
146 | int64_t dirtyrate; | |
147 | int i = 0; | |
148 | unsigned int gen_id; | |
149 | ||
150 | retry: | |
151 | init_time_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
152 | ||
370ed600 JI |
153 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
154 | gen_id = cpu_list_generation_id_get(); | |
155 | records = vcpu_dirty_stat_alloc(stat); | |
156 | vcpu_dirty_stat_collect(stat, records, true); | |
157 | } | |
8244166d HH |
158 | |
159 | duration = dirty_stat_wait(calc_time_ms, init_time_ms); | |
160 | ||
161 | global_dirty_log_sync(flag, one_shot); | |
162 | ||
370ed600 JI |
163 | WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { |
164 | if (gen_id != cpu_list_generation_id_get()) { | |
165 | g_free(records); | |
166 | g_free(stat->rates); | |
167 | cpu_list_unlock(); | |
168 | goto retry; | |
169 | } | |
170 | vcpu_dirty_stat_collect(stat, records, false); | |
8244166d | 171 | } |
8244166d HH |
172 | |
173 | for (i = 0; i < stat->nvcpu; i++) { | |
174 | dirtyrate = do_calculate_dirtyrate(records[i], duration); | |
175 | ||
176 | stat->rates[i].id = i; | |
177 | stat->rates[i].dirty_rate = dirtyrate; | |
178 | ||
179 | trace_dirtyrate_do_calculate_vcpu(i, dirtyrate); | |
180 | } | |
181 | ||
182 | g_free(records); | |
183 | ||
184 | return duration; | |
185 | } | |
186 | ||
eca58224 CZ |
187 | static bool is_sample_period_valid(int64_t sec) |
188 | { | |
189 | if (sec < MIN_FETCH_DIRTYRATE_TIME_SEC || | |
190 | sec > MAX_FETCH_DIRTYRATE_TIME_SEC) { | |
191 | return false; | |
192 | } | |
193 | ||
194 | return true; | |
195 | } | |
196 | ||
7afa08cd HH |
197 | static bool is_sample_pages_valid(int64_t pages) |
198 | { | |
199 | return pages >= MIN_SAMPLE_PAGE_COUNT && | |
200 | pages <= MAX_SAMPLE_PAGE_COUNT; | |
201 | } | |
202 | ||
7df3aa30 CZ |
203 | static int dirtyrate_set_state(int *state, int old_state, int new_state) |
204 | { | |
205 | assert(new_state < DIRTY_RATE_STATUS__MAX); | |
3c0b5dff | 206 | trace_dirtyrate_set_state(DirtyRateStatus_str(new_state)); |
7df3aa30 CZ |
207 | if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { |
208 | return 0; | |
209 | } else { | |
210 | return -1; | |
211 | } | |
212 | } | |
213 | ||
4c437254 CZ |
214 | static struct DirtyRateInfo *query_dirty_rate_info(void) |
215 | { | |
0e21bf24 | 216 | int i; |
4c437254 | 217 | int64_t dirty_rate = DirtyStat.dirty_rate; |
b21e2380 | 218 | struct DirtyRateInfo *info = g_new0(DirtyRateInfo, 1); |
0e21bf24 | 219 | DirtyRateVcpuList *head = NULL, **tail = &head; |
4c437254 CZ |
220 | |
221 | info->status = CalculatingState; | |
222 | info->start_time = DirtyStat.start_time; | |
223 | info->calc_time = DirtyStat.calc_time; | |
7afa08cd | 224 | info->sample_pages = DirtyStat.sample_pages; |
0e21bf24 HH |
225 | info->mode = dirtyrate_mode; |
226 | ||
227 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) { | |
228 | info->has_dirty_rate = true; | |
229 | info->dirty_rate = dirty_rate; | |
230 | ||
231 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
232 | /* | |
233 | * set sample_pages with 0 to indicate page sampling | |
234 | * isn't enabled | |
235 | **/ | |
236 | info->sample_pages = 0; | |
237 | info->has_vcpu_dirty_rate = true; | |
238 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { | |
b21e2380 | 239 | DirtyRateVcpu *rate = g_new0(DirtyRateVcpu, 1); |
0e21bf24 HH |
240 | rate->id = DirtyStat.dirty_ring.rates[i].id; |
241 | rate->dirty_rate = DirtyStat.dirty_ring.rates[i].dirty_rate; | |
242 | QAPI_LIST_APPEND(tail, rate); | |
243 | } | |
244 | info->vcpu_dirty_rate = head; | |
245 | } | |
826b8bc8 HH |
246 | |
247 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { | |
248 | info->sample_pages = 0; | |
249 | } | |
0e21bf24 | 250 | } |
4c437254 | 251 | |
3c0b5dff CZ |
252 | trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState)); |
253 | ||
4c437254 CZ |
254 | return info; |
255 | } | |
256 | ||
71864ead HH |
257 | static void init_dirtyrate_stat(int64_t start_time, |
258 | struct DirtyRateConfig config) | |
c9a58d71 | 259 | { |
c9a58d71 | 260 | DirtyStat.dirty_rate = -1; |
aa84b506 | 261 | DirtyStat.start_time = start_time; |
71864ead HH |
262 | DirtyStat.calc_time = config.sample_period_seconds; |
263 | DirtyStat.sample_pages = config.sample_pages_per_gigabytes; | |
264 | ||
265 | switch (config.mode) { | |
266 | case DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING: | |
267 | DirtyStat.page_sampling.total_dirty_samples = 0; | |
268 | DirtyStat.page_sampling.total_sample_count = 0; | |
269 | DirtyStat.page_sampling.total_block_mem_MB = 0; | |
270 | break; | |
271 | case DIRTY_RATE_MEASURE_MODE_DIRTY_RING: | |
272 | DirtyStat.dirty_ring.nvcpu = -1; | |
273 | DirtyStat.dirty_ring.rates = NULL; | |
274 | break; | |
275 | default: | |
276 | break; | |
277 | } | |
c9a58d71 CZ |
278 | } |
279 | ||
0e21bf24 HH |
280 | static void cleanup_dirtyrate_stat(struct DirtyRateConfig config) |
281 | { | |
282 | /* last calc-dirty-rate qmp use dirty ring mode */ | |
283 | if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
284 | free(DirtyStat.dirty_ring.rates); | |
285 | DirtyStat.dirty_ring.rates = NULL; | |
286 | } | |
287 | } | |
288 | ||
c9a58d71 CZ |
289 | static void update_dirtyrate_stat(struct RamblockDirtyInfo *info) |
290 | { | |
71864ead HH |
291 | DirtyStat.page_sampling.total_dirty_samples += info->sample_dirty_count; |
292 | DirtyStat.page_sampling.total_sample_count += info->sample_pages_count; | |
c9a58d71 | 293 | /* size of total pages in MB */ |
71864ead HH |
294 | DirtyStat.page_sampling.total_block_mem_MB += (info->ramblock_pages * |
295 | TARGET_PAGE_SIZE) >> 20; | |
c9a58d71 CZ |
296 | } |
297 | ||
298 | static void update_dirtyrate(uint64_t msec) | |
299 | { | |
300 | uint64_t dirtyrate; | |
71864ead HH |
301 | uint64_t total_dirty_samples = DirtyStat.page_sampling.total_dirty_samples; |
302 | uint64_t total_sample_count = DirtyStat.page_sampling.total_sample_count; | |
303 | uint64_t total_block_mem_MB = DirtyStat.page_sampling.total_block_mem_MB; | |
c9a58d71 CZ |
304 | |
305 | dirtyrate = total_dirty_samples * total_block_mem_MB * | |
306 | 1000 / (total_sample_count * msec); | |
307 | ||
308 | DirtyStat.dirty_rate = dirtyrate; | |
309 | } | |
7df3aa30 | 310 | |
ba0e519f CZ |
311 | /* |
312 | * get hash result for the sampled memory with length of TARGET_PAGE_SIZE | |
313 | * in ramblock, which starts from ramblock base address. | |
314 | */ | |
315 | static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info, | |
316 | uint64_t vfn) | |
317 | { | |
318 | uint32_t crc; | |
319 | ||
320 | crc = crc32(0, (info->ramblock_addr + | |
321 | vfn * TARGET_PAGE_SIZE), TARGET_PAGE_SIZE); | |
322 | ||
3c0b5dff | 323 | trace_get_ramblock_vfn_hash(info->idstr, vfn, crc); |
ba0e519f CZ |
324 | return crc; |
325 | } | |
326 | ||
327 | static bool save_ramblock_hash(struct RamblockDirtyInfo *info) | |
328 | { | |
329 | unsigned int sample_pages_count; | |
330 | int i; | |
331 | GRand *rand; | |
332 | ||
333 | sample_pages_count = info->sample_pages_count; | |
334 | ||
335 | /* ramblock size less than one page, return success to skip this ramblock */ | |
336 | if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) { | |
337 | return true; | |
338 | } | |
339 | ||
340 | info->hash_result = g_try_malloc0_n(sample_pages_count, | |
341 | sizeof(uint32_t)); | |
342 | if (!info->hash_result) { | |
343 | return false; | |
344 | } | |
345 | ||
346 | info->sample_page_vfn = g_try_malloc0_n(sample_pages_count, | |
347 | sizeof(uint64_t)); | |
348 | if (!info->sample_page_vfn) { | |
349 | g_free(info->hash_result); | |
350 | return false; | |
351 | } | |
352 | ||
353 | rand = g_rand_new(); | |
354 | for (i = 0; i < sample_pages_count; i++) { | |
355 | info->sample_page_vfn[i] = g_rand_int_range(rand, 0, | |
356 | info->ramblock_pages - 1); | |
357 | info->hash_result[i] = get_ramblock_vfn_hash(info, | |
358 | info->sample_page_vfn[i]); | |
359 | } | |
360 | g_rand_free(rand); | |
361 | ||
362 | return true; | |
363 | } | |
364 | ||
365 | static void get_ramblock_dirty_info(RAMBlock *block, | |
366 | struct RamblockDirtyInfo *info, | |
367 | struct DirtyRateConfig *config) | |
368 | { | |
369 | uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes; | |
370 | ||
371 | /* Right shift 30 bits to calc ramblock size in GB */ | |
372 | info->sample_pages_count = (qemu_ram_get_used_length(block) * | |
373 | sample_pages_per_gigabytes) >> 30; | |
374 | /* Right shift TARGET_PAGE_BITS to calc page count */ | |
375 | info->ramblock_pages = qemu_ram_get_used_length(block) >> | |
376 | TARGET_PAGE_BITS; | |
377 | info->ramblock_addr = qemu_ram_get_host_addr(block); | |
378 | strcpy(info->idstr, qemu_ram_get_idstr(block)); | |
379 | } | |
380 | ||
cf0bbb49 CZ |
381 | static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count) |
382 | { | |
383 | int i; | |
384 | ||
385 | if (!infos) { | |
386 | return; | |
387 | } | |
388 | ||
389 | for (i = 0; i < count; i++) { | |
390 | g_free(infos[i].sample_page_vfn); | |
391 | g_free(infos[i].hash_result); | |
392 | } | |
393 | g_free(infos); | |
394 | } | |
395 | ||
f82583cd CZ |
396 | static bool skip_sample_ramblock(RAMBlock *block) |
397 | { | |
398 | /* | |
399 | * Sample only blocks larger than MIN_RAMBLOCK_SIZE. | |
400 | */ | |
401 | if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) { | |
3c0b5dff CZ |
402 | trace_skip_sample_ramblock(block->idstr, |
403 | qemu_ram_get_used_length(block)); | |
f82583cd CZ |
404 | return true; |
405 | } | |
406 | ||
407 | return false; | |
408 | } | |
409 | ||
ba0e519f CZ |
410 | static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo, |
411 | struct DirtyRateConfig config, | |
412 | int *block_count) | |
413 | { | |
414 | struct RamblockDirtyInfo *info = NULL; | |
415 | struct RamblockDirtyInfo *dinfo = NULL; | |
416 | RAMBlock *block = NULL; | |
417 | int total_count = 0; | |
418 | int index = 0; | |
419 | bool ret = false; | |
420 | ||
421 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
422 | if (skip_sample_ramblock(block)) { |
423 | continue; | |
424 | } | |
ba0e519f CZ |
425 | total_count++; |
426 | } | |
427 | ||
428 | dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo)); | |
429 | if (dinfo == NULL) { | |
430 | goto out; | |
431 | } | |
432 | ||
433 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
434 | if (skip_sample_ramblock(block)) { |
435 | continue; | |
436 | } | |
ba0e519f CZ |
437 | if (index >= total_count) { |
438 | break; | |
439 | } | |
440 | info = &dinfo[index]; | |
441 | get_ramblock_dirty_info(block, info, &config); | |
442 | if (!save_ramblock_hash(info)) { | |
443 | goto out; | |
444 | } | |
445 | index++; | |
446 | } | |
447 | ret = true; | |
448 | ||
449 | out: | |
450 | *block_count = index; | |
451 | *block_dinfo = dinfo; | |
452 | return ret; | |
453 | } | |
454 | ||
9c04387b CZ |
455 | static void calc_page_dirty_rate(struct RamblockDirtyInfo *info) |
456 | { | |
457 | uint32_t crc; | |
458 | int i; | |
459 | ||
460 | for (i = 0; i < info->sample_pages_count; i++) { | |
461 | crc = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]); | |
462 | if (crc != info->hash_result[i]) { | |
3c0b5dff | 463 | trace_calc_page_dirty_rate(info->idstr, crc, info->hash_result[i]); |
9c04387b CZ |
464 | info->sample_dirty_count++; |
465 | } | |
466 | } | |
467 | } | |
468 | ||
469 | static struct RamblockDirtyInfo * | |
470 | find_block_matched(RAMBlock *block, int count, | |
471 | struct RamblockDirtyInfo *infos) | |
472 | { | |
473 | int i; | |
9c04387b CZ |
474 | |
475 | for (i = 0; i < count; i++) { | |
476 | if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) { | |
477 | break; | |
478 | } | |
479 | } | |
480 | ||
481 | if (i == count) { | |
482 | return NULL; | |
483 | } | |
484 | ||
485 | if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) || | |
486 | infos[i].ramblock_pages != | |
487 | (qemu_ram_get_used_length(block) >> TARGET_PAGE_BITS)) { | |
3c0b5dff | 488 | trace_find_page_matched(block->idstr); |
9c04387b CZ |
489 | return NULL; |
490 | } | |
491 | ||
66997c42 | 492 | return &infos[i]; |
9c04387b CZ |
493 | } |
494 | ||
495 | static bool compare_page_hash_info(struct RamblockDirtyInfo *info, | |
496 | int block_count) | |
497 | { | |
498 | struct RamblockDirtyInfo *block_dinfo = NULL; | |
499 | RAMBlock *block = NULL; | |
500 | ||
501 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
f82583cd CZ |
502 | if (skip_sample_ramblock(block)) { |
503 | continue; | |
504 | } | |
9c04387b CZ |
505 | block_dinfo = find_block_matched(block, block_count, info); |
506 | if (block_dinfo == NULL) { | |
507 | continue; | |
508 | } | |
509 | calc_page_dirty_rate(block_dinfo); | |
510 | update_dirtyrate_stat(block_dinfo); | |
511 | } | |
512 | ||
71864ead | 513 | if (DirtyStat.page_sampling.total_sample_count == 0) { |
9c04387b CZ |
514 | return false; |
515 | } | |
516 | ||
517 | return true; | |
518 | } | |
519 | ||
826b8bc8 HH |
520 | static inline void record_dirtypages_bitmap(DirtyPageRecord *dirty_pages, |
521 | bool start) | |
522 | { | |
523 | if (start) { | |
524 | dirty_pages->start_pages = total_dirty_pages; | |
525 | } else { | |
526 | dirty_pages->end_pages = total_dirty_pages; | |
527 | } | |
528 | } | |
529 | ||
826b8bc8 HH |
530 | static inline void dirtyrate_manual_reset_protect(void) |
531 | { | |
532 | RAMBlock *block = NULL; | |
533 | ||
534 | WITH_RCU_READ_LOCK_GUARD() { | |
535 | RAMBLOCK_FOREACH_MIGRATABLE(block) { | |
536 | memory_region_clear_dirty_bitmap(block->mr, 0, | |
537 | block->used_length); | |
538 | } | |
539 | } | |
540 | } | |
541 | ||
542 | static void calculate_dirtyrate_dirty_bitmap(struct DirtyRateConfig config) | |
543 | { | |
544 | int64_t msec = 0; | |
545 | int64_t start_time; | |
546 | DirtyPageRecord dirty_pages; | |
547 | ||
548 | qemu_mutex_lock_iothread(); | |
549 | memory_global_dirty_log_start(GLOBAL_DIRTY_DIRTY_RATE); | |
550 | ||
551 | /* | |
552 | * 1'round of log sync may return all 1 bits with | |
553 | * KVM_DIRTY_LOG_INITIALLY_SET enable | |
554 | * skip it unconditionally and start dirty tracking | |
555 | * from 2'round of log sync | |
556 | */ | |
557 | memory_global_dirty_log_sync(); | |
558 | ||
559 | /* | |
560 | * reset page protect manually and unconditionally. | |
561 | * this make sure kvm dirty log be cleared if | |
562 | * KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE cap is enabled. | |
563 | */ | |
564 | dirtyrate_manual_reset_protect(); | |
565 | qemu_mutex_unlock_iothread(); | |
566 | ||
567 | record_dirtypages_bitmap(&dirty_pages, true); | |
568 | ||
569 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
570 | DirtyStat.start_time = start_time / 1000; | |
571 | ||
572 | msec = config.sample_period_seconds * 1000; | |
8244166d | 573 | msec = dirty_stat_wait(msec, start_time); |
826b8bc8 HH |
574 | DirtyStat.calc_time = msec / 1000; |
575 | ||
576 | /* | |
8244166d | 577 | * do two things. |
826b8bc8 HH |
578 | * 1. fetch dirty bitmap from kvm |
579 | * 2. stop dirty tracking | |
580 | */ | |
8244166d | 581 | global_dirty_log_sync(GLOBAL_DIRTY_DIRTY_RATE, true); |
826b8bc8 HH |
582 | |
583 | record_dirtypages_bitmap(&dirty_pages, false); | |
584 | ||
8244166d | 585 | DirtyStat.dirty_rate = do_calculate_dirtyrate(dirty_pages, msec); |
826b8bc8 HH |
586 | } |
587 | ||
0e21bf24 HH |
588 | static void calculate_dirtyrate_dirty_ring(struct DirtyRateConfig config) |
589 | { | |
8244166d | 590 | int64_t duration; |
0e21bf24 HH |
591 | uint64_t dirtyrate = 0; |
592 | uint64_t dirtyrate_sum = 0; | |
0e21bf24 HH |
593 | int i = 0; |
594 | ||
8244166d HH |
595 | /* start log sync */ |
596 | global_dirty_log_change(GLOBAL_DIRTY_DIRTY_RATE, true); | |
0e21bf24 | 597 | |
8244166d | 598 | DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; |
0e21bf24 | 599 | |
8244166d HH |
600 | /* calculate vcpu dirtyrate */ |
601 | duration = vcpu_calculate_dirtyrate(config.sample_period_seconds * 1000, | |
602 | &DirtyStat.dirty_ring, | |
603 | GLOBAL_DIRTY_DIRTY_RATE, | |
604 | true); | |
0e21bf24 | 605 | |
8244166d | 606 | DirtyStat.calc_time = duration / 1000; |
0e21bf24 | 607 | |
8244166d | 608 | /* calculate vm dirtyrate */ |
0e21bf24 | 609 | for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { |
8244166d | 610 | dirtyrate = DirtyStat.dirty_ring.rates[i].dirty_rate; |
0e21bf24 HH |
611 | DirtyStat.dirty_ring.rates[i].dirty_rate = dirtyrate; |
612 | dirtyrate_sum += dirtyrate; | |
613 | } | |
614 | ||
615 | DirtyStat.dirty_rate = dirtyrate_sum; | |
0e21bf24 HH |
616 | } |
617 | ||
618 | static void calculate_dirtyrate_sample_vm(struct DirtyRateConfig config) | |
4240dcee | 619 | { |
cf0bbb49 CZ |
620 | struct RamblockDirtyInfo *block_dinfo = NULL; |
621 | int block_count = 0; | |
622 | int64_t msec = 0; | |
623 | int64_t initial_time; | |
624 | ||
cf0bbb49 CZ |
625 | rcu_read_lock(); |
626 | initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
627 | if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) { | |
628 | goto out; | |
629 | } | |
630 | rcu_read_unlock(); | |
631 | ||
632 | msec = config.sample_period_seconds * 1000; | |
8244166d | 633 | msec = dirty_stat_wait(msec, initial_time); |
4c437254 CZ |
634 | DirtyStat.start_time = initial_time / 1000; |
635 | DirtyStat.calc_time = msec / 1000; | |
cf0bbb49 CZ |
636 | |
637 | rcu_read_lock(); | |
638 | if (!compare_page_hash_info(block_dinfo, block_count)) { | |
639 | goto out; | |
640 | } | |
641 | ||
642 | update_dirtyrate(msec); | |
643 | ||
644 | out: | |
645 | rcu_read_unlock(); | |
646 | free_ramblock_dirty_info(block_dinfo, block_count); | |
4240dcee CZ |
647 | } |
648 | ||
0e21bf24 HH |
649 | static void calculate_dirtyrate(struct DirtyRateConfig config) |
650 | { | |
826b8bc8 HH |
651 | if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { |
652 | calculate_dirtyrate_dirty_bitmap(config); | |
653 | } else if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { | |
0e21bf24 HH |
654 | calculate_dirtyrate_dirty_ring(config); |
655 | } else { | |
656 | calculate_dirtyrate_sample_vm(config); | |
657 | } | |
658 | ||
659 | trace_dirtyrate_calculate(DirtyStat.dirty_rate); | |
660 | } | |
661 | ||
4240dcee CZ |
662 | void *get_dirtyrate_thread(void *arg) |
663 | { | |
664 | struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg; | |
7df3aa30 | 665 | int ret; |
15eb2d64 | 666 | rcu_register_thread(); |
7df3aa30 CZ |
667 | |
668 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED, | |
669 | DIRTY_RATE_STATUS_MEASURING); | |
670 | if (ret == -1) { | |
671 | error_report("change dirtyrate state failed."); | |
672 | return NULL; | |
673 | } | |
4240dcee CZ |
674 | |
675 | calculate_dirtyrate(config); | |
676 | ||
7df3aa30 CZ |
677 | ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING, |
678 | DIRTY_RATE_STATUS_MEASURED); | |
679 | if (ret == -1) { | |
680 | error_report("change dirtyrate state failed."); | |
681 | } | |
15eb2d64 HH |
682 | |
683 | rcu_unregister_thread(); | |
4240dcee CZ |
684 | return NULL; |
685 | } | |
4c437254 | 686 | |
0e21bf24 HH |
687 | void qmp_calc_dirty_rate(int64_t calc_time, |
688 | bool has_sample_pages, | |
689 | int64_t sample_pages, | |
690 | bool has_mode, | |
691 | DirtyRateMeasureMode mode, | |
692 | Error **errp) | |
4c437254 CZ |
693 | { |
694 | static struct DirtyRateConfig config; | |
695 | QemuThread thread; | |
696 | int ret; | |
9865d0f6 | 697 | int64_t start_time; |
4c437254 CZ |
698 | |
699 | /* | |
700 | * If the dirty rate is already being measured, don't attempt to start. | |
701 | */ | |
702 | if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) { | |
703 | error_setg(errp, "the dirty rate is already being measured."); | |
704 | return; | |
705 | } | |
706 | ||
707 | if (!is_sample_period_valid(calc_time)) { | |
708 | error_setg(errp, "calc-time is out of range[%d, %d].", | |
709 | MIN_FETCH_DIRTYRATE_TIME_SEC, | |
710 | MAX_FETCH_DIRTYRATE_TIME_SEC); | |
711 | return; | |
712 | } | |
713 | ||
0e21bf24 HH |
714 | if (!has_mode) { |
715 | mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
716 | } | |
717 | ||
bd9510d3 ZD |
718 | if (has_sample_pages && mode != DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
719 | error_setg(errp, "sample-pages is used only in page-sampling mode"); | |
0e21bf24 HH |
720 | return; |
721 | } | |
722 | ||
7afa08cd HH |
723 | if (has_sample_pages) { |
724 | if (!is_sample_pages_valid(sample_pages)) { | |
725 | error_setg(errp, "sample-pages is out of range[%d, %d].", | |
726 | MIN_SAMPLE_PAGE_COUNT, | |
727 | MAX_SAMPLE_PAGE_COUNT); | |
728 | return; | |
729 | } | |
730 | } else { | |
731 | sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES; | |
732 | } | |
733 | ||
0e21bf24 HH |
734 | /* |
735 | * dirty ring mode only works when kvm dirty ring is enabled. | |
826b8bc8 | 736 | * on the contrary, dirty bitmap mode is not. |
0e21bf24 | 737 | */ |
826b8bc8 HH |
738 | if (((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) && |
739 | !kvm_dirty_ring_enabled()) || | |
740 | ((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) && | |
741 | kvm_dirty_ring_enabled())) { | |
742 | error_setg(errp, "mode %s is not enabled, use other method instead.", | |
743 | DirtyRateMeasureMode_str(mode)); | |
744 | return; | |
0e21bf24 HH |
745 | } |
746 | ||
4c437254 CZ |
747 | /* |
748 | * Init calculation state as unstarted. | |
749 | */ | |
750 | ret = dirtyrate_set_state(&CalculatingState, CalculatingState, | |
751 | DIRTY_RATE_STATUS_UNSTARTED); | |
752 | if (ret == -1) { | |
753 | error_setg(errp, "init dirty rate calculation state failed."); | |
754 | return; | |
755 | } | |
756 | ||
757 | config.sample_period_seconds = calc_time; | |
7afa08cd | 758 | config.sample_pages_per_gigabytes = sample_pages; |
0e21bf24 HH |
759 | config.mode = mode; |
760 | ||
761 | cleanup_dirtyrate_stat(config); | |
762 | ||
763 | /* | |
764 | * update dirty rate mode so that we can figure out what mode has | |
765 | * been used in last calculation | |
766 | **/ | |
767 | dirtyrate_mode = mode; | |
9865d0f6 HH |
768 | |
769 | start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) / 1000; | |
770 | init_dirtyrate_stat(start_time, config); | |
771 | ||
4c437254 CZ |
772 | qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread, |
773 | (void *)&config, QEMU_THREAD_DETACHED); | |
774 | } | |
775 | ||
776 | struct DirtyRateInfo *qmp_query_dirty_rate(Error **errp) | |
777 | { | |
778 | return query_dirty_rate_info(); | |
779 | } | |
a4a571d9 PX |
780 | |
781 | void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict) | |
782 | { | |
783 | DirtyRateInfo *info = query_dirty_rate_info(); | |
784 | ||
785 | monitor_printf(mon, "Status: %s\n", | |
786 | DirtyRateStatus_str(info->status)); | |
787 | monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n", | |
788 | info->start_time); | |
bd9510d3 ZD |
789 | if (info->mode == DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { |
790 | monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n", | |
791 | info->sample_pages); | |
792 | } | |
a4a571d9 PX |
793 | monitor_printf(mon, "Period: %"PRIi64" (sec)\n", |
794 | info->calc_time); | |
0e21bf24 HH |
795 | monitor_printf(mon, "Mode: %s\n", |
796 | DirtyRateMeasureMode_str(info->mode)); | |
a4a571d9 PX |
797 | monitor_printf(mon, "Dirty rate: "); |
798 | if (info->has_dirty_rate) { | |
799 | monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate); | |
0e21bf24 HH |
800 | if (info->has_vcpu_dirty_rate) { |
801 | DirtyRateVcpuList *rate, *head = info->vcpu_dirty_rate; | |
802 | for (rate = head; rate != NULL; rate = rate->next) { | |
803 | monitor_printf(mon, "vcpu[%"PRIi64"], Dirty rate: %"PRIi64 | |
804 | " (MB/s)\n", rate->value->id, | |
805 | rate->value->dirty_rate); | |
806 | } | |
807 | } | |
a4a571d9 PX |
808 | } else { |
809 | monitor_printf(mon, "(not ready)\n"); | |
810 | } | |
0e21bf24 HH |
811 | |
812 | qapi_free_DirtyRateVcpuList(info->vcpu_dirty_rate); | |
a4a571d9 PX |
813 | g_free(info); |
814 | } | |
815 | ||
816 | void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict) | |
817 | { | |
818 | int64_t sec = qdict_get_try_int(qdict, "second", 0); | |
819 | int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1); | |
820 | bool has_sample_pages = (sample_pages != -1); | |
0e21bf24 | 821 | bool dirty_ring = qdict_get_try_bool(qdict, "dirty_ring", false); |
826b8bc8 HH |
822 | bool dirty_bitmap = qdict_get_try_bool(qdict, "dirty_bitmap", false); |
823 | DirtyRateMeasureMode mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; | |
a4a571d9 PX |
824 | Error *err = NULL; |
825 | ||
826 | if (!sec) { | |
827 | monitor_printf(mon, "Incorrect period length specified!\n"); | |
828 | return; | |
829 | } | |
830 | ||
826b8bc8 HH |
831 | if (dirty_ring && dirty_bitmap) { |
832 | monitor_printf(mon, "Either dirty ring or dirty bitmap " | |
833 | "can be specified!\n"); | |
834 | return; | |
835 | } | |
836 | ||
837 | if (dirty_bitmap) { | |
838 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP; | |
839 | } else if (dirty_ring) { | |
840 | mode = DIRTY_RATE_MEASURE_MODE_DIRTY_RING; | |
841 | } | |
842 | ||
0e21bf24 HH |
843 | qmp_calc_dirty_rate(sec, has_sample_pages, sample_pages, true, |
844 | mode, &err); | |
a4a571d9 PX |
845 | if (err) { |
846 | hmp_handle_error(mon, err); | |
847 | return; | |
848 | } | |
849 | ||
850 | monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64 | |
851 | " seconds\n", sec); | |
852 | monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n"); | |
853 | } |