| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* Arm Statistical Profiling Extensions (SPE) support |
| 4 |
* Copyright (c) 2017-2018, Arm Ltd. |
| 5 |
*/ |
| 6 |
|
| 7 |
#include <linux/kernel.h> |
| 8 |
#include <linux/types.h> |
| 9 |
#include <linux/bitops.h> |
| 10 |
#include <linux/log2.h> |
| 11 |
#include <linux/string.h> |
| 12 |
#include <linux/zalloc.h> |
| 13 |
#include <errno.h> |
| 14 |
#include <time.h> |
| 15 |
|
| 16 |
#include "../../../util/cpumap.h" |
| 17 |
#include "../../../util/event.h" |
| 18 |
#include "../../../util/evsel.h" |
| 19 |
#include "../../../util/evsel_config.h" |
| 20 |
#include "../../../util/evlist.h" |
| 21 |
#include "../../../util/session.h" |
| 22 |
#include <internal/lib.h> // page_size |
| 23 |
#include "../../../util/pmu.h" |
| 24 |
#include "../../../util/debug.h" |
| 25 |
#include "../../../util/auxtrace.h" |
| 26 |
#include "../../../util/record.h" |
| 27 |
#include "../../../util/header.h" |
| 28 |
#include "../../../util/arm-spe.h" |
| 29 |
#include <tools/libc_compat.h> // reallocarray |
| 30 |
|
| 31 |
#define ARM_SPE_CPU_MAGIC 0x1010101010101010ULL |
| 32 |
|
| 33 |
#define KiB(x) ((x) * 1024) |
| 34 |
#define MiB(x) ((x) * 1024 * 1024) |
| 35 |
|
| 36 |
struct arm_spe_recording { |
| 37 |
struct auxtrace_record itr; |
| 38 |
struct perf_pmu *arm_spe_pmu; |
| 39 |
struct evlist *evlist; |
| 40 |
int wrapped_cnt; |
| 41 |
bool *wrapped; |
| 42 |
}; |
| 43 |
|
| 44 |
/* Iterate config list to detect if the "freq" parameter is set */ |
| 45 |
static bool arm_spe_is_set_freq(struct evsel *evsel) |
| 46 |
{ |
| 47 |
struct evsel_config_term *term; |
| 48 |
|
| 49 |
list_for_each_entry(term, &evsel->config_terms, list) { |
| 50 |
if (term->type == EVSEL__CONFIG_TERM_FREQ) |
| 51 |
return true; |
| 52 |
} |
| 53 |
|
| 54 |
return false; |
| 55 |
} |
| 56 |
|
| 57 |
/* |
| 58 |
* arm_spe_find_cpus() returns a new cpu map, and the caller should invoke |
| 59 |
* perf_cpu_map__put() to release the map after use. |
| 60 |
*/ |
| 61 |
static struct perf_cpu_map *arm_spe_find_cpus(struct evlist *evlist) |
| 62 |
{ |
| 63 |
struct perf_cpu_map *event_cpus = evlist->core.user_requested_cpus; |
| 64 |
struct perf_cpu_map *online_cpus = perf_cpu_map__new_online_cpus(); |
| 65 |
struct perf_cpu_map *intersect_cpus; |
| 66 |
|
| 67 |
/* cpu map is not "any" CPU , we have specific CPUs to work with */ |
| 68 |
if (!perf_cpu_map__has_any_cpu(event_cpus)) { |
| 69 |
intersect_cpus = perf_cpu_map__intersect(event_cpus, online_cpus); |
| 70 |
perf_cpu_map__put(online_cpus); |
| 71 |
/* Event can be "any" CPU so count all CPUs. */ |
| 72 |
} else { |
| 73 |
intersect_cpus = online_cpus; |
| 74 |
} |
| 75 |
|
| 76 |
return intersect_cpus; |
| 77 |
} |
| 78 |
|
| 79 |
static size_t |
| 80 |
arm_spe_info_priv_size(struct auxtrace_record *itr __maybe_unused, |
| 81 |
struct evlist *evlist) |
| 82 |
{ |
| 83 |
struct perf_cpu_map *cpu_map = arm_spe_find_cpus(evlist); |
| 84 |
size_t size; |
| 85 |
|
| 86 |
if (!cpu_map) |
| 87 |
return 0; |
| 88 |
|
| 89 |
size = ARM_SPE_AUXTRACE_PRIV_MAX + |
| 90 |
ARM_SPE_CPU_PRIV_MAX * perf_cpu_map__nr(cpu_map); |
| 91 |
size *= sizeof(u64); |
| 92 |
|
| 93 |
perf_cpu_map__put(cpu_map); |
| 94 |
return size; |
| 95 |
} |
| 96 |
|
| 97 |
static int arm_spe_save_cpu_header(struct auxtrace_record *itr, |
| 98 |
struct perf_cpu cpu, __u64 data[]) |
| 99 |
{ |
| 100 |
struct arm_spe_recording *sper = |
| 101 |
container_of(itr, struct arm_spe_recording, itr); |
| 102 |
struct perf_pmu *pmu = NULL; |
| 103 |
char *cpuid = NULL; |
| 104 |
u64 val; |
| 105 |
|
| 106 |
/* Read CPU MIDR */ |
| 107 |
cpuid = get_cpuid_allow_env_override(cpu); |
| 108 |
if (!cpuid) |
| 109 |
return -ENOMEM; |
| 110 |
val = strtol(cpuid, NULL, 16); |
| 111 |
|
| 112 |
data[ARM_SPE_MAGIC] = ARM_SPE_CPU_MAGIC; |
| 113 |
data[ARM_SPE_CPU] = cpu.cpu; |
| 114 |
data[ARM_SPE_CPU_NR_PARAMS] = ARM_SPE_CPU_PRIV_MAX - ARM_SPE_CPU_MIDR; |
| 115 |
data[ARM_SPE_CPU_MIDR] = val; |
| 116 |
|
| 117 |
/* Find the associate Arm SPE PMU for the CPU */ |
| 118 |
if (perf_cpu_map__has(sper->arm_spe_pmu->cpus, cpu)) |
| 119 |
pmu = sper->arm_spe_pmu; |
| 120 |
|
| 121 |
if (!pmu) { |
| 122 |
/* No Arm SPE PMU is found */ |
| 123 |
data[ARM_SPE_CPU_PMU_TYPE] = ULLONG_MAX; |
| 124 |
data[ARM_SPE_CAP_MIN_IVAL] = 0; |
| 125 |
data[ARM_SPE_CAP_EVENT_FILTER] = 0; |
| 126 |
} else { |
| 127 |
data[ARM_SPE_CPU_PMU_TYPE] = pmu->type; |
| 128 |
|
| 129 |
if (perf_pmu__scan_file(pmu, "caps/min_interval", "%lu", &val) != 1) |
| 130 |
val = 0; |
| 131 |
data[ARM_SPE_CAP_MIN_IVAL] = val; |
| 132 |
|
| 133 |
if (perf_pmu__scan_file(pmu, "caps/event_filter", "%lx", &val) != 1) |
| 134 |
val = 0; |
| 135 |
data[ARM_SPE_CAP_EVENT_FILTER] = val; |
| 136 |
} |
| 137 |
|
| 138 |
free(cpuid); |
| 139 |
return ARM_SPE_CPU_PRIV_MAX; |
| 140 |
} |
| 141 |
|
| 142 |
static int arm_spe_info_fill(struct auxtrace_record *itr, |
| 143 |
struct perf_session *session, |
| 144 |
struct perf_record_auxtrace_info *auxtrace_info, |
| 145 |
size_t priv_size) |
| 146 |
{ |
| 147 |
int i, ret; |
| 148 |
size_t offset; |
| 149 |
struct arm_spe_recording *sper = |
| 150 |
container_of(itr, struct arm_spe_recording, itr); |
| 151 |
struct perf_pmu *arm_spe_pmu = sper->arm_spe_pmu; |
| 152 |
struct perf_cpu_map *cpu_map; |
| 153 |
struct perf_cpu cpu; |
| 154 |
__u64 *data; |
| 155 |
|
| 156 |
if (priv_size != arm_spe_info_priv_size(itr, session->evlist)) |
| 157 |
return -EINVAL; |
| 158 |
|
| 159 |
if (!session->evlist->core.nr_mmaps) |
| 160 |
return -EINVAL; |
| 161 |
|
| 162 |
cpu_map = arm_spe_find_cpus(session->evlist); |
| 163 |
if (!cpu_map) |
| 164 |
return -EINVAL; |
| 165 |
|
| 166 |
auxtrace_info->type = PERF_AUXTRACE_ARM_SPE; |
| 167 |
auxtrace_info->priv[ARM_SPE_HEADER_VERSION] = ARM_SPE_HEADER_CURRENT_VERSION; |
| 168 |
auxtrace_info->priv[ARM_SPE_HEADER_SIZE] = |
| 169 |
ARM_SPE_AUXTRACE_PRIV_MAX - ARM_SPE_HEADER_VERSION; |
| 170 |
auxtrace_info->priv[ARM_SPE_PMU_TYPE_V2] = arm_spe_pmu->type; |
| 171 |
auxtrace_info->priv[ARM_SPE_CPUS_NUM] = perf_cpu_map__nr(cpu_map); |
| 172 |
|
| 173 |
offset = ARM_SPE_AUXTRACE_PRIV_MAX; |
| 174 |
perf_cpu_map__for_each_cpu(cpu, i, cpu_map) { |
| 175 |
assert(offset < priv_size); |
| 176 |
data = &auxtrace_info->priv[offset]; |
| 177 |
ret = arm_spe_save_cpu_header(itr, cpu, data); |
| 178 |
if (ret < 0) |
| 179 |
goto out; |
| 180 |
offset += ret; |
| 181 |
} |
| 182 |
|
| 183 |
ret = 0; |
| 184 |
out: |
| 185 |
perf_cpu_map__put(cpu_map); |
| 186 |
return ret; |
| 187 |
} |
| 188 |
|
| 189 |
static void |
| 190 |
arm_spe_snapshot_resolve_auxtrace_defaults(struct record_opts *opts, |
| 191 |
bool privileged) |
| 192 |
{ |
| 193 |
/* |
| 194 |
* The default snapshot size is the auxtrace mmap size. If neither auxtrace mmap size nor |
| 195 |
* snapshot size is specified, then the default is 4MiB for privileged users, 128KiB for |
| 196 |
* unprivileged users. |
| 197 |
* |
| 198 |
* The default auxtrace mmap size is 4MiB/page_size for privileged users, 128KiB for |
| 199 |
* unprivileged users. If an unprivileged user does not specify mmap pages, the mmap pages |
| 200 |
* will be reduced from the default 512KiB/page_size to 256KiB/page_size, otherwise the |
| 201 |
* user is likely to get an error as they exceed their mlock limmit. |
| 202 |
*/ |
| 203 |
|
| 204 |
/* |
| 205 |
* No size were given to '-S' or '-m,', so go with the default |
| 206 |
*/ |
| 207 |
if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) { |
| 208 |
if (privileged) { |
| 209 |
opts->auxtrace_mmap_pages = MiB(4) / page_size; |
| 210 |
} else { |
| 211 |
opts->auxtrace_mmap_pages = KiB(128) / page_size; |
| 212 |
if (opts->mmap_pages == UINT_MAX) |
| 213 |
opts->mmap_pages = KiB(256) / page_size; |
| 214 |
} |
| 215 |
} else if (!opts->auxtrace_mmap_pages && !privileged && opts->mmap_pages == UINT_MAX) { |
| 216 |
opts->mmap_pages = KiB(256) / page_size; |
| 217 |
} |
| 218 |
|
| 219 |
/* |
| 220 |
* '-m,xyz' was specified but no snapshot size, so make the snapshot size as big as the |
| 221 |
* auxtrace mmap area. |
| 222 |
*/ |
| 223 |
if (!opts->auxtrace_snapshot_size) |
| 224 |
opts->auxtrace_snapshot_size = opts->auxtrace_mmap_pages * (size_t)page_size; |
| 225 |
|
| 226 |
/* |
| 227 |
* '-Sxyz' was specified but no auxtrace mmap area, so make the auxtrace mmap area big |
| 228 |
* enough to fit the requested snapshot size. |
| 229 |
*/ |
| 230 |
if (!opts->auxtrace_mmap_pages) { |
| 231 |
size_t sz = opts->auxtrace_snapshot_size; |
| 232 |
|
| 233 |
sz = round_up(sz, page_size) / page_size; |
| 234 |
opts->auxtrace_mmap_pages = roundup_pow_of_two(sz); |
| 235 |
} |
| 236 |
} |
| 237 |
|
| 238 |
static __u64 arm_spe_pmu__sample_period(const struct perf_pmu *arm_spe_pmu) |
| 239 |
{ |
| 240 |
static __u64 sample_period; |
| 241 |
|
| 242 |
if (sample_period) |
| 243 |
return sample_period; |
| 244 |
|
| 245 |
/* |
| 246 |
* If kernel driver doesn't advertise a minimum, |
| 247 |
* use max allowable by PMSIDR_EL1.INTERVAL |
| 248 |
*/ |
| 249 |
if (perf_pmu__scan_file(arm_spe_pmu, "caps/min_interval", "%llu", |
| 250 |
&sample_period) != 1) { |
| 251 |
pr_debug("arm_spe driver doesn't advertise a min. interval. Using 4096\n"); |
| 252 |
sample_period = 4096; |
| 253 |
} |
| 254 |
return sample_period; |
| 255 |
} |
| 256 |
|
| 257 |
static void arm_spe_setup_evsel(struct evsel *evsel, struct perf_cpu_map *cpus) |
| 258 |
{ |
| 259 |
u64 pa_enable_bit; |
| 260 |
|
| 261 |
evsel->core.attr.freq = 0; |
| 262 |
evsel->core.attr.sample_period = arm_spe_pmu__sample_period(evsel->pmu); |
| 263 |
evsel->needs_auxtrace_mmap = true; |
| 264 |
|
| 265 |
/* |
| 266 |
* To obtain the auxtrace buffer file descriptor, the auxtrace event |
| 267 |
* must come first. |
| 268 |
*/ |
| 269 |
evlist__to_front(evsel->evlist, evsel); |
| 270 |
|
| 271 |
/* |
| 272 |
* In the case of per-cpu mmaps, sample CPU for AUX event; |
| 273 |
* also enable the timestamp tracing for samples correlation. |
| 274 |
*/ |
| 275 |
if (!perf_cpu_map__is_any_cpu_or_is_empty(cpus)) { |
| 276 |
evsel__set_sample_bit(evsel, CPU); |
| 277 |
evsel__set_config_if_unset(evsel, "ts_enable", 1); |
| 278 |
} |
| 279 |
|
| 280 |
/* |
| 281 |
* Set this only so that perf report knows that SPE generates memory info. It has no effect |
| 282 |
* on the opening of the event or the SPE data produced. |
| 283 |
*/ |
| 284 |
evsel__set_sample_bit(evsel, DATA_SRC); |
| 285 |
|
| 286 |
/* |
| 287 |
* The PHYS_ADDR flag does not affect the driver behaviour, it is used to |
| 288 |
* inform that the resulting output's SPE samples contain physical addresses |
| 289 |
* where applicable. |
| 290 |
*/ |
| 291 |
|
| 292 |
if (!evsel__get_config_val(evsel, "pa_enable", &pa_enable_bit)) |
| 293 |
if (pa_enable_bit) |
| 294 |
evsel__set_sample_bit(evsel, PHYS_ADDR); |
| 295 |
} |
| 296 |
|
| 297 |
static int arm_spe_setup_aux_buffer(struct record_opts *opts) |
| 298 |
{ |
| 299 |
bool privileged = perf_event_paranoid_check(-1); |
| 300 |
|
| 301 |
/* |
| 302 |
* we are in snapshot mode. |
| 303 |
*/ |
| 304 |
if (opts->auxtrace_snapshot_mode) { |
| 305 |
/* |
| 306 |
* Command arguments '-Sxyz' and/or '-m,xyz' are missing, so fill those in with |
| 307 |
* default values. |
| 308 |
*/ |
| 309 |
if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) |
| 310 |
arm_spe_snapshot_resolve_auxtrace_defaults(opts, privileged); |
| 311 |
|
| 312 |
/* |
| 313 |
* Snapshot size can't be bigger than the auxtrace area. |
| 314 |
*/ |
| 315 |
if (opts->auxtrace_snapshot_size > opts->auxtrace_mmap_pages * (size_t)page_size) { |
| 316 |
pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n", |
| 317 |
opts->auxtrace_snapshot_size, |
| 318 |
opts->auxtrace_mmap_pages * (size_t)page_size); |
| 319 |
return -EINVAL; |
| 320 |
} |
| 321 |
|
| 322 |
/* |
| 323 |
* Something went wrong somewhere - this shouldn't happen. |
| 324 |
*/ |
| 325 |
if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) { |
| 326 |
pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n"); |
| 327 |
return -EINVAL; |
| 328 |
} |
| 329 |
|
| 330 |
pr_debug2("%sx snapshot size: %zu\n", ARM_SPE_PMU_NAME, |
| 331 |
opts->auxtrace_snapshot_size); |
| 332 |
} |
| 333 |
|
| 334 |
/* We are in full trace mode but '-m,xyz' wasn't specified */ |
| 335 |
if (!opts->auxtrace_mmap_pages) { |
| 336 |
if (privileged) { |
| 337 |
opts->auxtrace_mmap_pages = MiB(4) / page_size; |
| 338 |
} else { |
| 339 |
opts->auxtrace_mmap_pages = KiB(128) / page_size; |
| 340 |
if (opts->mmap_pages == UINT_MAX) |
| 341 |
opts->mmap_pages = KiB(256) / page_size; |
| 342 |
} |
| 343 |
} |
| 344 |
|
| 345 |
/* Validate auxtrace_mmap_pages */ |
| 346 |
if (opts->auxtrace_mmap_pages) { |
| 347 |
size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size; |
| 348 |
size_t min_sz = KiB(8); |
| 349 |
|
| 350 |
if (sz < min_sz || !is_power_of_2(sz)) { |
| 351 |
pr_err("Invalid mmap size for ARM SPE: must be at least %zuKiB and a power of 2\n", |
| 352 |
min_sz / 1024); |
| 353 |
return -EINVAL; |
| 354 |
} |
| 355 |
} |
| 356 |
|
| 357 |
return 0; |
| 358 |
} |
| 359 |
|
| 360 |
static int arm_spe_setup_tracking_event(struct evlist *evlist, |
| 361 |
struct record_opts *opts) |
| 362 |
{ |
| 363 |
int err; |
| 364 |
struct evsel *tracking_evsel; |
| 365 |
struct perf_cpu_map *cpus = evlist->core.user_requested_cpus; |
| 366 |
|
| 367 |
/* Add dummy event to keep tracking */ |
| 368 |
err = parse_event(evlist, "dummy:u"); |
| 369 |
if (err) |
| 370 |
return err; |
| 371 |
|
| 372 |
tracking_evsel = evlist__last(evlist); |
| 373 |
evlist__set_tracking_event(evlist, tracking_evsel); |
| 374 |
|
| 375 |
tracking_evsel->core.attr.freq = 0; |
| 376 |
tracking_evsel->core.attr.sample_period = 1; |
| 377 |
|
| 378 |
/* In per-cpu case, always need the time of mmap events etc */ |
| 379 |
if (!perf_cpu_map__is_any_cpu_or_is_empty(cpus)) { |
| 380 |
evsel__set_sample_bit(tracking_evsel, TIME); |
| 381 |
evsel__set_sample_bit(tracking_evsel, CPU); |
| 382 |
|
| 383 |
/* also track task context switch */ |
| 384 |
if (!record_opts__no_switch_events(opts)) |
| 385 |
tracking_evsel->core.attr.context_switch = 1; |
| 386 |
} |
| 387 |
|
| 388 |
return 0; |
| 389 |
} |
| 390 |
|
| 391 |
static int arm_spe_recording_options(struct auxtrace_record *itr, |
| 392 |
struct evlist *evlist, |
| 393 |
struct record_opts *opts) |
| 394 |
{ |
| 395 |
struct arm_spe_recording *sper = |
| 396 |
container_of(itr, struct arm_spe_recording, itr); |
| 397 |
struct evsel *evsel, *tmp; |
| 398 |
struct perf_cpu_map *cpus = evlist->core.user_requested_cpus; |
| 399 |
bool discard = false; |
| 400 |
int err; |
| 401 |
u64 discard_bit; |
| 402 |
|
| 403 |
sper->evlist = evlist; |
| 404 |
|
| 405 |
evlist__for_each_entry(evlist, evsel) { |
| 406 |
if (evsel__is_aux_event(evsel)) { |
| 407 |
if (!strstarts(evsel->pmu->name, ARM_SPE_PMU_NAME)) { |
| 408 |
pr_err("Found unexpected auxtrace event: %s\n", |
| 409 |
evsel->pmu->name); |
| 410 |
return -EINVAL; |
| 411 |
} |
| 412 |
opts->full_auxtrace = true; |
| 413 |
|
| 414 |
if (opts->user_freq != UINT_MAX || |
| 415 |
arm_spe_is_set_freq(evsel)) { |
| 416 |
pr_err("Arm SPE: Frequency is not supported. " |
| 417 |
"Set period with -c option or PMU parameter (-e %s/period=NUM/).\n", |
| 418 |
evsel->pmu->name); |
| 419 |
return -EINVAL; |
| 420 |
} |
| 421 |
} |
| 422 |
} |
| 423 |
|
| 424 |
if (!opts->full_auxtrace) |
| 425 |
return 0; |
| 426 |
|
| 427 |
evlist__for_each_entry_safe(evlist, tmp, evsel) { |
| 428 |
if (evsel__is_aux_event(evsel)) { |
| 429 |
arm_spe_setup_evsel(evsel, cpus); |
| 430 |
if (!evsel__get_config_val(evsel, "discard", &discard_bit)) |
| 431 |
discard = !!discard_bit; |
| 432 |
} |
| 433 |
} |
| 434 |
|
| 435 |
if (discard) |
| 436 |
return 0; |
| 437 |
|
| 438 |
err = arm_spe_setup_aux_buffer(opts); |
| 439 |
if (err) |
| 440 |
return err; |
| 441 |
|
| 442 |
return arm_spe_setup_tracking_event(evlist, opts); |
| 443 |
} |
| 444 |
|
| 445 |
static int arm_spe_parse_snapshot_options(struct auxtrace_record *itr __maybe_unused, |
| 446 |
struct record_opts *opts, |
| 447 |
const char *str) |
| 448 |
{ |
| 449 |
unsigned long long snapshot_size = 0; |
| 450 |
char *endptr; |
| 451 |
|
| 452 |
if (str) { |
| 453 |
snapshot_size = strtoull(str, &endptr, 0); |
| 454 |
if (*endptr || snapshot_size > SIZE_MAX) |
| 455 |
return -1; |
| 456 |
} |
| 457 |
|
| 458 |
opts->auxtrace_snapshot_mode = true; |
| 459 |
opts->auxtrace_snapshot_size = snapshot_size; |
| 460 |
|
| 461 |
return 0; |
| 462 |
} |
| 463 |
|
| 464 |
static int arm_spe_snapshot_start(struct auxtrace_record *itr) |
| 465 |
{ |
| 466 |
struct arm_spe_recording *ptr = |
| 467 |
container_of(itr, struct arm_spe_recording, itr); |
| 468 |
struct evsel *evsel; |
| 469 |
int ret = -EINVAL; |
| 470 |
|
| 471 |
evlist__for_each_entry(ptr->evlist, evsel) { |
| 472 |
if (evsel__is_aux_event(evsel)) { |
| 473 |
ret = evsel__disable(evsel); |
| 474 |
if (ret < 0) |
| 475 |
return ret; |
| 476 |
} |
| 477 |
} |
| 478 |
return ret; |
| 479 |
} |
| 480 |
|
| 481 |
static int arm_spe_snapshot_finish(struct auxtrace_record *itr) |
| 482 |
{ |
| 483 |
struct arm_spe_recording *ptr = |
| 484 |
container_of(itr, struct arm_spe_recording, itr); |
| 485 |
struct evsel *evsel; |
| 486 |
int ret = -EINVAL; |
| 487 |
|
| 488 |
evlist__for_each_entry(ptr->evlist, evsel) { |
| 489 |
if (evsel__is_aux_event(evsel)) { |
| 490 |
ret = evsel__enable(evsel); |
| 491 |
if (ret < 0) |
| 492 |
return ret; |
| 493 |
} |
| 494 |
} |
| 495 |
return ret; |
| 496 |
} |
| 497 |
|
| 498 |
static int arm_spe_alloc_wrapped_array(struct arm_spe_recording *ptr, int idx) |
| 499 |
{ |
| 500 |
bool *wrapped; |
| 501 |
int cnt = ptr->wrapped_cnt, new_cnt, i; |
| 502 |
|
| 503 |
/* |
| 504 |
* No need to allocate, so return early. |
| 505 |
*/ |
| 506 |
if (idx < cnt) |
| 507 |
return 0; |
| 508 |
|
| 509 |
/* |
| 510 |
* Make ptr->wrapped as big as idx. |
| 511 |
*/ |
| 512 |
new_cnt = idx + 1; |
| 513 |
|
| 514 |
/* |
| 515 |
* Free'ed in arm_spe_recording_free(). |
| 516 |
*/ |
| 517 |
wrapped = reallocarray(ptr->wrapped, new_cnt, sizeof(bool)); |
| 518 |
if (!wrapped) |
| 519 |
return -ENOMEM; |
| 520 |
|
| 521 |
/* |
| 522 |
* init new allocated values. |
| 523 |
*/ |
| 524 |
for (i = cnt; i < new_cnt; i++) |
| 525 |
wrapped[i] = false; |
| 526 |
|
| 527 |
ptr->wrapped_cnt = new_cnt; |
| 528 |
ptr->wrapped = wrapped; |
| 529 |
|
| 530 |
return 0; |
| 531 |
} |
| 532 |
|
| 533 |
static bool arm_spe_buffer_has_wrapped(unsigned char *buffer, |
| 534 |
size_t buffer_size, u64 head) |
| 535 |
{ |
| 536 |
u64 i, watermark; |
| 537 |
u64 *buf = (u64 *)buffer; |
| 538 |
size_t buf_size = buffer_size; |
| 539 |
|
| 540 |
/* |
| 541 |
* Defensively handle the case where head might be continually increasing - if its value is |
| 542 |
* equal or greater than the size of the ring buffer, then we can safely determine it has |
| 543 |
* wrapped around. Otherwise, continue to detect if head might have wrapped. |
| 544 |
*/ |
| 545 |
if (head >= buffer_size) |
| 546 |
return true; |
| 547 |
|
| 548 |
/* |
| 549 |
* We want to look the very last 512 byte (chosen arbitrarily) in the ring buffer. |
| 550 |
*/ |
| 551 |
watermark = buf_size - 512; |
| 552 |
|
| 553 |
/* |
| 554 |
* The value of head is somewhere within the size of the ring buffer. This can be that there |
| 555 |
* hasn't been enough data to fill the ring buffer yet or the trace time was so long that |
| 556 |
* head has numerically wrapped around. To find we need to check if we have data at the |
| 557 |
* very end of the ring buffer. We can reliably do this because mmap'ed pages are zeroed |
| 558 |
* out and there is a fresh mapping with every new session. |
| 559 |
*/ |
| 560 |
|
| 561 |
/* |
| 562 |
* head is less than 512 byte from the end of the ring buffer. |
| 563 |
*/ |
| 564 |
if (head > watermark) |
| 565 |
watermark = head; |
| 566 |
|
| 567 |
/* |
| 568 |
* Speed things up by using 64 bit transactions (see "u64 *buf" above) |
| 569 |
*/ |
| 570 |
watermark /= sizeof(u64); |
| 571 |
buf_size /= sizeof(u64); |
| 572 |
|
| 573 |
/* |
| 574 |
* If we find trace data at the end of the ring buffer, head has been there and has |
| 575 |
* numerically wrapped around at least once. |
| 576 |
*/ |
| 577 |
for (i = watermark; i < buf_size; i++) |
| 578 |
if (buf[i]) |
| 579 |
return true; |
| 580 |
|
| 581 |
return false; |
| 582 |
} |
| 583 |
|
| 584 |
static int arm_spe_find_snapshot(struct auxtrace_record *itr, int idx, |
| 585 |
struct auxtrace_mmap *mm, unsigned char *data, |
| 586 |
u64 *head, u64 *old) |
| 587 |
{ |
| 588 |
int err; |
| 589 |
bool wrapped; |
| 590 |
struct arm_spe_recording *ptr = |
| 591 |
container_of(itr, struct arm_spe_recording, itr); |
| 592 |
|
| 593 |
/* |
| 594 |
* Allocate memory to keep track of wrapping if this is the first |
| 595 |
* time we deal with this *mm. |
| 596 |
*/ |
| 597 |
if (idx >= ptr->wrapped_cnt) { |
| 598 |
err = arm_spe_alloc_wrapped_array(ptr, idx); |
| 599 |
if (err) |
| 600 |
return err; |
| 601 |
} |
| 602 |
|
| 603 |
/* |
| 604 |
* Check to see if *head has wrapped around. If it hasn't only the |
| 605 |
* amount of data between *head and *old is snapshot'ed to avoid |
| 606 |
* bloating the perf.data file with zeros. But as soon as *head has |
| 607 |
* wrapped around the entire size of the AUX ring buffer it taken. |
| 608 |
*/ |
| 609 |
wrapped = ptr->wrapped[idx]; |
| 610 |
if (!wrapped && arm_spe_buffer_has_wrapped(data, mm->len, *head)) { |
| 611 |
wrapped = true; |
| 612 |
ptr->wrapped[idx] = true; |
| 613 |
} |
| 614 |
|
| 615 |
pr_debug3("%s: mmap index %d old head %zu new head %zu size %zu\n", |
| 616 |
__func__, idx, (size_t)*old, (size_t)*head, mm->len); |
| 617 |
|
| 618 |
/* |
| 619 |
* No wrap has occurred, we can just use *head and *old. |
| 620 |
*/ |
| 621 |
if (!wrapped) |
| 622 |
return 0; |
| 623 |
|
| 624 |
/* |
| 625 |
* *head has wrapped around - adjust *head and *old to pickup the |
| 626 |
* entire content of the AUX buffer. |
| 627 |
*/ |
| 628 |
if (*head >= mm->len) { |
| 629 |
*old = *head - mm->len; |
| 630 |
} else { |
| 631 |
*head += mm->len; |
| 632 |
*old = *head - mm->len; |
| 633 |
} |
| 634 |
|
| 635 |
return 0; |
| 636 |
} |
| 637 |
|
| 638 |
static u64 arm_spe_reference(struct auxtrace_record *itr __maybe_unused) |
| 639 |
{ |
| 640 |
struct timespec ts; |
| 641 |
|
| 642 |
clock_gettime(CLOCK_MONOTONIC_RAW, &ts); |
| 643 |
|
| 644 |
return ts.tv_sec ^ ts.tv_nsec; |
| 645 |
} |
| 646 |
|
| 647 |
static void arm_spe_recording_free(struct auxtrace_record *itr) |
| 648 |
{ |
| 649 |
struct arm_spe_recording *sper = |
| 650 |
container_of(itr, struct arm_spe_recording, itr); |
| 651 |
|
| 652 |
zfree(&sper->wrapped); |
| 653 |
free(sper); |
| 654 |
} |
| 655 |
|
| 656 |
struct auxtrace_record *arm_spe_recording_init(int *err, |
| 657 |
struct perf_pmu *arm_spe_pmu) |
| 658 |
{ |
| 659 |
struct arm_spe_recording *sper; |
| 660 |
|
| 661 |
if (!arm_spe_pmu) { |
| 662 |
*err = -ENODEV; |
| 663 |
return NULL; |
| 664 |
} |
| 665 |
|
| 666 |
sper = zalloc(sizeof(struct arm_spe_recording)); |
| 667 |
if (!sper) { |
| 668 |
*err = -ENOMEM; |
| 669 |
return NULL; |
| 670 |
} |
| 671 |
|
| 672 |
sper->arm_spe_pmu = arm_spe_pmu; |
| 673 |
sper->itr.snapshot_start = arm_spe_snapshot_start; |
| 674 |
sper->itr.snapshot_finish = arm_spe_snapshot_finish; |
| 675 |
sper->itr.find_snapshot = arm_spe_find_snapshot; |
| 676 |
sper->itr.parse_snapshot_options = arm_spe_parse_snapshot_options; |
| 677 |
sper->itr.recording_options = arm_spe_recording_options; |
| 678 |
sper->itr.info_priv_size = arm_spe_info_priv_size; |
| 679 |
sper->itr.info_fill = arm_spe_info_fill; |
| 680 |
sper->itr.free = arm_spe_recording_free; |
| 681 |
sper->itr.reference = arm_spe_reference; |
| 682 |
sper->itr.read_finish = auxtrace_record__read_finish; |
| 683 |
sper->itr.alignment = 0; |
| 684 |
|
| 685 |
*err = 0; |
| 686 |
return &sper->itr; |
| 687 |
} |
| 688 |
|
| 689 |
void |
| 690 |
arm_spe_pmu_default_config(const struct perf_pmu *arm_spe_pmu, struct perf_event_attr *attr) |
| 691 |
{ |
| 692 |
attr->sample_period = arm_spe_pmu__sample_period(arm_spe_pmu); |
| 693 |
} |