/src/qemu/block/accounting.c
Line | Count | Source |
1 | | /* |
2 | | * QEMU System Emulator block accounting |
3 | | * |
4 | | * Copyright (c) 2011 Christoph Hellwig |
5 | | * Copyright (c) 2015 Igalia, S.L. |
6 | | * |
7 | | * Permission is hereby granted, free of charge, to any person obtaining a copy |
8 | | * of this software and associated documentation files (the "Software"), to deal |
9 | | * in the Software without restriction, including without limitation the rights |
10 | | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
11 | | * copies of the Software, and to permit persons to whom the Software is |
12 | | * furnished to do so, subject to the following conditions: |
13 | | * |
14 | | * The above copyright notice and this permission notice shall be included in |
15 | | * all copies or substantial portions of the Software. |
16 | | * |
17 | | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
18 | | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
19 | | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
20 | | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
21 | | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
22 | | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
23 | | * THE SOFTWARE. |
24 | | */ |
25 | | |
26 | | #include "qemu/osdep.h" |
27 | | #include "block/accounting.h" |
28 | | #include "block/block_int.h" |
29 | | #include "qemu/timer.h" |
30 | | #include "system/qtest.h" |
31 | | #include "qapi/error.h" |
32 | | |
33 | | static QEMUClockType clock_type = QEMU_CLOCK_REALTIME; |
34 | | static const int qtest_latency_ns = NANOSECONDS_PER_SECOND / 1000; |
35 | | |
36 | | void block_acct_init(BlockAcctStats *stats) |
37 | 39 | { |
38 | 39 | qemu_mutex_init(&stats->lock); |
39 | 39 | if (qtest_enabled()) { |
40 | 3 | clock_type = QEMU_CLOCK_VIRTUAL; |
41 | 3 | } |
42 | 39 | stats->account_invalid = true; |
43 | 39 | stats->account_failed = true; |
44 | 39 | } |
45 | | |
46 | | static bool bool_from_onoffauto(OnOffAuto val, bool def) |
47 | 76 | { |
48 | 76 | switch (val) { |
49 | 76 | case ON_OFF_AUTO_AUTO: |
50 | 76 | return def; |
51 | 0 | case ON_OFF_AUTO_ON: |
52 | 0 | return true; |
53 | 0 | case ON_OFF_AUTO_OFF: |
54 | 0 | return false; |
55 | 0 | default: |
56 | 0 | abort(); |
57 | 76 | } |
58 | 76 | } |
59 | | |
60 | | bool block_acct_setup(BlockAcctStats *stats, enum OnOffAuto account_invalid, |
61 | | enum OnOffAuto account_failed, uint32_t *stats_intervals, |
62 | | uint32_t num_stats_intervals, Error **errp) |
63 | 38 | { |
64 | 38 | stats->account_invalid = bool_from_onoffauto(account_invalid, |
65 | 38 | stats->account_invalid); |
66 | 38 | stats->account_failed = bool_from_onoffauto(account_failed, |
67 | 38 | stats->account_failed); |
68 | 38 | if (stats_intervals) { |
69 | 0 | for (int i = 0; i < num_stats_intervals; i++) { |
70 | 0 | if (stats_intervals[i] <= 0) { |
71 | 0 | error_setg(errp, "Invalid interval length: %u", stats_intervals[i]); |
72 | 0 | return false; |
73 | 0 | } |
74 | 0 | block_acct_add_interval(stats, stats_intervals[i]); |
75 | 0 | } |
76 | 0 | } |
77 | 38 | return true; |
78 | 38 | } |
79 | | |
80 | | void block_acct_cleanup(BlockAcctStats *stats) |
81 | 15 | { |
82 | 15 | BlockAcctTimedStats *s, *next; |
83 | 15 | block_latency_histograms_clear(stats); |
84 | | |
85 | 15 | QSLIST_FOREACH_SAFE(s, &stats->intervals, entries, next) { |
86 | 0 | g_free(s); |
87 | 0 | } |
88 | 15 | qemu_mutex_destroy(&stats->lock); |
89 | 15 | } |
90 | | |
91 | | void block_acct_add_interval(BlockAcctStats *stats, unsigned interval_length) |
92 | 0 | { |
93 | 0 | BlockAcctTimedStats *s; |
94 | 0 | unsigned i; |
95 | |
|
96 | 0 | s = g_new0(BlockAcctTimedStats, 1); |
97 | 0 | s->interval_length = interval_length; |
98 | 0 | s->stats = stats; |
99 | 0 | qemu_mutex_lock(&stats->lock); |
100 | 0 | QSLIST_INSERT_HEAD(&stats->intervals, s, entries); |
101 | |
|
102 | 0 | for (i = 0; i < BLOCK_MAX_IOTYPE; i++) { |
103 | 0 | timed_average_init(&s->latency[i], clock_type, |
104 | 0 | (uint64_t) interval_length * NANOSECONDS_PER_SECOND); |
105 | 0 | } |
106 | 0 | qemu_mutex_unlock(&stats->lock); |
107 | 0 | } |
108 | | |
109 | | BlockAcctTimedStats *block_acct_interval_next(BlockAcctStats *stats, |
110 | | BlockAcctTimedStats *s) |
111 | 0 | { |
112 | 0 | if (s == NULL) { |
113 | 0 | return QSLIST_FIRST(&stats->intervals); |
114 | 0 | } else { |
115 | 0 | return QSLIST_NEXT(s, entries); |
116 | 0 | } |
117 | 0 | } |
118 | | |
119 | | void block_acct_start(BlockAcctStats *stats, BlockAcctCookie *cookie, |
120 | | int64_t bytes, enum BlockAcctType type) |
121 | 22.8M | { |
122 | 22.8M | assert(type < BLOCK_MAX_IOTYPE); |
123 | | |
124 | 22.8M | cookie->bytes = bytes; |
125 | 22.8M | cookie->start_time_ns = qemu_clock_get_ns(clock_type); |
126 | 22.8M | cookie->type = type; |
127 | 22.8M | } |
128 | | |
129 | | /* block_latency_histogram_compare_func: |
130 | | * Compare @key with interval [@it[0], @it[1]). |
131 | | * Return: -1 if @key < @it[0] |
132 | | * 0 if @key in [@it[0], @it[1]) |
133 | | * +1 if @key >= @it[1] |
134 | | */ |
135 | | static int block_latency_histogram_compare_func(const void *key, const void *it) |
136 | 0 | { |
137 | 0 | uint64_t k = *(uint64_t *)key; |
138 | 0 | uint64_t a = ((uint64_t *)it)[0]; |
139 | 0 | uint64_t b = ((uint64_t *)it)[1]; |
140 | |
|
141 | 0 | return k < a ? -1 : (k < b ? 0 : 1); |
142 | 0 | } |
143 | | |
144 | | static void block_latency_histogram_account(BlockLatencyHistogram *hist, |
145 | | int64_t latency_ns) |
146 | 22.8M | { |
147 | 22.8M | uint64_t *pos; |
148 | | |
149 | 22.8M | if (hist->bins == NULL) { |
150 | | /* histogram disabled */ |
151 | 22.8M | return; |
152 | 22.8M | } |
153 | | |
154 | | |
155 | 0 | if (latency_ns < hist->boundaries[0]) { |
156 | 0 | hist->bins[0]++; |
157 | 0 | return; |
158 | 0 | } |
159 | | |
160 | 0 | if (latency_ns >= hist->boundaries[hist->nbins - 2]) { |
161 | 0 | hist->bins[hist->nbins - 1]++; |
162 | 0 | return; |
163 | 0 | } |
164 | | |
165 | 0 | pos = bsearch(&latency_ns, hist->boundaries, hist->nbins - 2, |
166 | 0 | sizeof(hist->boundaries[0]), |
167 | 0 | block_latency_histogram_compare_func); |
168 | 0 | assert(pos != NULL); |
169 | |
|
170 | 0 | hist->bins[pos - hist->boundaries + 1]++; |
171 | 0 | } |
172 | | |
173 | | int block_latency_histogram_set(BlockAcctStats *stats, enum BlockAcctType type, |
174 | | uint64List *boundaries) |
175 | 0 | { |
176 | 0 | BlockLatencyHistogram *hist = &stats->latency_histogram[type]; |
177 | 0 | uint64List *entry; |
178 | 0 | uint64_t *ptr; |
179 | 0 | uint64_t prev = 0; |
180 | 0 | int new_nbins = 1; |
181 | |
|
182 | 0 | for (entry = boundaries; entry; entry = entry->next) { |
183 | 0 | if (entry->value <= prev) { |
184 | 0 | return -EINVAL; |
185 | 0 | } |
186 | 0 | new_nbins++; |
187 | 0 | prev = entry->value; |
188 | 0 | } |
189 | | |
190 | | /* |
191 | | * block_latency_histogram_account() assumes that it can always access |
192 | | * hist->boundaries[0], so require at least one boundary. A histogram with |
193 | | * a single bin is useless anyway. |
194 | | */ |
195 | 0 | if (new_nbins <= 1) { |
196 | 0 | return -EINVAL; |
197 | 0 | } |
198 | | |
199 | 0 | qemu_mutex_lock(&stats->lock); |
200 | |
|
201 | 0 | hist->nbins = new_nbins; |
202 | 0 | g_free(hist->boundaries); |
203 | 0 | hist->boundaries = g_new(uint64_t, hist->nbins - 1); |
204 | 0 | for (entry = boundaries, ptr = hist->boundaries; entry; |
205 | 0 | entry = entry->next, ptr++) |
206 | 0 | { |
207 | 0 | *ptr = entry->value; |
208 | 0 | } |
209 | |
|
210 | 0 | g_free(hist->bins); |
211 | 0 | hist->bins = g_new0(uint64_t, hist->nbins); |
212 | |
|
213 | 0 | qemu_mutex_unlock(&stats->lock); |
214 | |
|
215 | 0 | return 0; |
216 | 0 | } |
217 | | |
218 | | void block_latency_histograms_clear(BlockAcctStats *stats) |
219 | 15 | { |
220 | 15 | int i; |
221 | | |
222 | 15 | qemu_mutex_lock(&stats->lock); |
223 | | |
224 | 105 | for (i = 0; i < BLOCK_MAX_IOTYPE; i++) { |
225 | 90 | BlockLatencyHistogram *hist = &stats->latency_histogram[i]; |
226 | 90 | g_free(hist->bins); |
227 | 90 | g_free(hist->boundaries); |
228 | 90 | memset(hist, 0, sizeof(*hist)); |
229 | 90 | } |
230 | | |
231 | 15 | qemu_mutex_unlock(&stats->lock); |
232 | 15 | } |
233 | | |
234 | | static void block_account_one_io(BlockAcctStats *stats, BlockAcctCookie *cookie, |
235 | | bool failed) |
236 | 22.8M | { |
237 | 22.8M | BlockAcctTimedStats *s; |
238 | 22.8M | int64_t time_ns = qemu_clock_get_ns(clock_type); |
239 | 22.8M | int64_t latency_ns = time_ns - cookie->start_time_ns; |
240 | | |
241 | 22.8M | if (qtest_enabled()) { |
242 | 22.8M | latency_ns = qtest_latency_ns; |
243 | 22.8M | } |
244 | | |
245 | 22.8M | assert(cookie->type < BLOCK_MAX_IOTYPE); |
246 | | |
247 | 22.8M | if (cookie->type == BLOCK_ACCT_NONE) { |
248 | 0 | return; |
249 | 0 | } |
250 | | |
251 | 22.8M | WITH_QEMU_LOCK_GUARD(&stats->lock) { |
252 | 22.8M | if (failed) { |
253 | 2.57M | stats->failed_ops[cookie->type]++; |
254 | 20.2M | } else { |
255 | 20.2M | stats->nr_bytes[cookie->type] += cookie->bytes; |
256 | 20.2M | stats->nr_ops[cookie->type]++; |
257 | 20.2M | } |
258 | | |
259 | 22.8M | block_latency_histogram_account(&stats->latency_histogram[cookie->type], |
260 | 22.8M | latency_ns); |
261 | | |
262 | 22.8M | if (!failed || stats->account_failed) { |
263 | 22.8M | stats->total_time_ns[cookie->type] += latency_ns; |
264 | 22.8M | stats->last_access_time_ns = time_ns; |
265 | | |
266 | 22.8M | QSLIST_FOREACH(s, &stats->intervals, entries) { |
267 | 0 | timed_average_account(&s->latency[cookie->type], latency_ns); |
268 | 0 | } |
269 | 22.8M | } |
270 | 22.8M | } |
271 | | |
272 | 22.8M | cookie->type = BLOCK_ACCT_NONE; |
273 | 22.8M | } |
274 | | |
275 | | void block_acct_done(BlockAcctStats *stats, BlockAcctCookie *cookie) |
276 | 20.2M | { |
277 | 20.2M | block_account_one_io(stats, cookie, false); |
278 | 20.2M | } |
279 | | |
280 | | void block_acct_failed(BlockAcctStats *stats, BlockAcctCookie *cookie) |
281 | 2.57M | { |
282 | 2.57M | block_account_one_io(stats, cookie, true); |
283 | 2.57M | } |
284 | | |
285 | | void block_acct_invalid(BlockAcctStats *stats, enum BlockAcctType type) |
286 | 9.02k | { |
287 | 9.02k | assert(type < BLOCK_MAX_IOTYPE); |
288 | | |
289 | | /* block_account_one_io() updates total_time_ns[], but this one does |
290 | | * not. The reason is that invalid requests are accounted during their |
291 | | * submission, therefore there's no actual I/O involved. |
292 | | */ |
293 | 18.0k | qemu_mutex_lock(&stats->lock); |
294 | 18.0k | stats->invalid_ops[type]++; |
295 | | |
296 | 18.0k | if (stats->account_invalid) { |
297 | 9.02k | stats->last_access_time_ns = qemu_clock_get_ns(clock_type); |
298 | 9.02k | } |
299 | 18.0k | qemu_mutex_unlock(&stats->lock); |
300 | 18.0k | } |
301 | | |
302 | | void block_acct_merge_done(BlockAcctStats *stats, enum BlockAcctType type, |
303 | | int num_requests) |
304 | 0 | { |
305 | 0 | assert(type < BLOCK_MAX_IOTYPE); |
306 | |
|
307 | 0 | qemu_mutex_lock(&stats->lock); |
308 | 0 | stats->merged[type] += num_requests; |
309 | 0 | qemu_mutex_unlock(&stats->lock); |
310 | 0 | } |
311 | | |
312 | | int64_t block_acct_idle_time_ns(BlockAcctStats *stats) |
313 | 0 | { |
314 | 0 | return qemu_clock_get_ns(clock_type) - stats->last_access_time_ns; |
315 | 0 | } |
316 | | |
317 | | double block_acct_queue_depth(BlockAcctTimedStats *stats, |
318 | | enum BlockAcctType type) |
319 | 0 | { |
320 | 0 | uint64_t sum, elapsed; |
321 | |
|
322 | 0 | assert(type < BLOCK_MAX_IOTYPE); |
323 | 0 | assert(qemu_mutex_trylock(&stats->stats->lock) == -EBUSY); |
324 | |
|
325 | 0 | sum = timed_average_sum(&stats->latency[type], &elapsed); |
326 | |
|
327 | 0 | return (double) sum / elapsed; |
328 | 0 | } |