liblttng-ust/lttng-context-perf-counters.c

   1 /*
   2  * lttng-context-perf-counters.c
   3  *
   4  * LTTng UST performance monitoring counters (perf-counters) integration.
   5  *
   6  * Copyright (C) 2009-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  *
   8  * This library is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; only
  11  * version 2.1 of the License.
  12  *
  13  * This library is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with this library; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include <sys/types.h>
  24 #include <unistd.h>
  25 #include <string.h>
  26 #include <stdlib.h>
  27 #include <stdio.h>
  28 #include <sys/mman.h>
  29 #include <sys/syscall.h>
  30 #include <linux/perf_event.h>
  31 #include <lttng/ust-events.h>
  32 #include <lttng/ust-tracer.h>
  33 #include <lttng/ringbuffer-config.h>
  34 #include <urcu/system.h>
  35 #include <urcu/arch.h>
  36 #include <urcu/rculist.h>
  37 #include <helper.h>
  38 #include <urcu/ref.h>
  39 #include <usterr-signal-safe.h>
  40 #include <signal.h>
  41 #include "lttng-tracer-core.h"
  42
  43 /*
  44  * We use a global perf counter key and iterate on per-thread RCU lists
  45  * of fields in the fast path, even though this is not strictly speaking
  46  * what would provide the best fast-path complexity, to ensure teardown
  47  * of sessions vs thread exit is handled racelessly.
  48  *
  49  * Updates and traversals of thread_list are protected by UST lock.
  50  * Updates to rcu_field_list are protected by UST lock.
  51  */
  52
  53 struct lttng_perf_counter_thread_field {
  54         struct lttng_perf_counter_field *field; /* Back reference */
  55         struct perf_event_mmap_page *pc;
  56         struct cds_list_head thread_field_node; /* Per-field list of thread fields (node) */
  57         struct cds_list_head rcu_field_node;    /* RCU per-thread list of fields (node) */
  58 };
  59
  60 struct lttng_perf_counter_thread {
  61         struct cds_list_head rcu_field_list;    /* RCU per-thread list of fields */
  62 };
  63
  64 struct lttng_perf_counter_field {
  65         struct perf_event_attr attr;
  66         struct cds_list_head thread_field_list; /* Per-field list of thread fields */
  67 };
  68
  69 static pthread_key_t perf_counter_key;
  70
  71 static
  72 size_t perf_counter_get_size(size_t offset)
  73 {
  74         size_t size = 0;
  75
  76         size += lib_ring_buffer_align(offset, lttng_alignof(uint64_t));
  77         size += sizeof(uint64_t);
  78         return size;
  79 }
  80
  81 #if defined(__x86_64__) || defined(__i386__)
  82
  83 static
  84 uint64_t rdpmc(unsigned int counter)
  85 {
  86         unsigned int low, high;
  87
  88         asm volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (counter));
  89
  90         return low | ((uint64_t) high) << 32;
  91 }
  92
  93 #else /* defined(__x86_64__) || defined(__i386__) */
  94
  95 #error "Perf event counters are only supported on x86 so far."
  96
  97 #endif /* #else defined(__x86_64__) || defined(__i386__) */
  98
  99 static
 100 uint64_t read_perf_counter(struct perf_event_mmap_page *pc)
 101 {
 102         uint32_t seq, idx;
 103         uint64_t count;
 104
 105         if (caa_unlikely(!pc))
 106                 return 0;
 107
 108         do {
 109                 seq = CMM_LOAD_SHARED(pc->lock);
 110                 cmm_barrier();
 111
 112                 idx = pc->index;
 113                 if (idx)
 114                         count = pc->offset + rdpmc(idx - 1);
 115                 else
 116                         count = 0;
 117
 118                 cmm_barrier();
 119         } while (CMM_LOAD_SHARED(pc->lock) != seq);
 120
 121         return count;
 122 }
 123
 124 static
 125 int sys_perf_event_open(struct perf_event_attr *attr,
 126                 pid_t pid, int cpu, int group_fd,
 127                 unsigned long flags)
 128 {
 129         return syscall(SYS_perf_event_open, attr, pid, cpu,
 130                         group_fd, flags);
 131 }
 132
 133 static
 134 struct perf_event_mmap_page *setup_perf(struct perf_event_attr *attr)
 135 {
 136         void *perf_addr;
 137         int fd, ret;
 138
 139         fd = sys_perf_event_open(attr, 0, -1, -1, 0);
 140         if (fd < 0)
 141                 return NULL;
 142
 143         perf_addr = mmap(NULL, sizeof(struct perf_event_mmap_page),
 144                         PROT_READ, MAP_SHARED, fd, 0);
 145         if (perf_addr == MAP_FAILED)
 146                 return NULL;
 147         ret = close(fd);
 148         if (ret) {
 149                 perror("Error closing LTTng-UST perf memory mapping FD");
 150         }
 151         return perf_addr;
 152 }
 153
 154 static
 155 void unmap_perf_page(struct perf_event_mmap_page *pc)
 156 {
 157         int ret;
 158
 159         if (!pc)
 160                 return;
 161         ret = munmap(pc, sizeof(struct perf_event_mmap_page));
 162         if (ret < 0) {
 163                 PERROR("Error in munmap");
 164                 abort();
 165         }
 166 }
 167
 168 static
 169 struct lttng_perf_counter_thread *alloc_perf_counter_thread(void)
 170 {
 171         struct lttng_perf_counter_thread *perf_thread;
 172         sigset_t newmask, oldmask;
 173         int ret;
 174
 175         ret = sigfillset(&newmask);
 176         if (ret)
 177                 abort();
 178         ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask);
 179         if (ret)
 180                 abort();
 181         /* Check again with signals disabled */
 182         perf_thread = pthread_getspecific(perf_counter_key);
 183         if (perf_thread)
 184                 goto skip;
 185         perf_thread = zmalloc(sizeof(*perf_thread));
 186         if (!perf_thread)
 187                 abort();
 188         CDS_INIT_LIST_HEAD(&perf_thread->rcu_field_list);
 189         ret = pthread_setspecific(perf_counter_key, perf_thread);
 190         if (ret)
 191                 abort();
 192 skip:
 193         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 194         if (ret)
 195                 abort();
 196         return perf_thread;
 197 }
 198
 199 static
 200 struct lttng_perf_counter_thread_field *
 201         add_thread_field(struct lttng_perf_counter_field *perf_field,
 202                 struct lttng_perf_counter_thread *perf_thread)
 203 {
 204         struct lttng_perf_counter_thread_field *thread_field;
 205         sigset_t newmask, oldmask;
 206         int ret;
 207
 208         ret = sigfillset(&newmask);
 209         if (ret)
 210                 abort();
 211         ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask);
 212         if (ret)
 213                 abort();
 214         /* Check again with signals disabled */
 215         cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list,
 216                         rcu_field_node) {
 217                 if (thread_field->field == perf_field)
 218                         goto skip;
 219         }
 220         thread_field = zmalloc(sizeof(*thread_field));
 221         if (!thread_field)
 222                 abort();
 223         thread_field->field = perf_field;
 224         thread_field->pc = setup_perf(&perf_field->attr);
 225         /* Note: thread_field->pc can be NULL if setup_perf() fails. */
 226         ust_lock_nocheck();
 227         cds_list_add_rcu(&thread_field->rcu_field_node,
 228                         &perf_thread->rcu_field_list);
 229         cds_list_add(&thread_field->thread_field_node,
 230                         &perf_field->thread_field_list);
 231         ust_unlock();
 232 skip:
 233         ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 234         if (ret)
 235                 abort();
 236         return thread_field;
 237 }
 238
 239 static
 240 struct lttng_perf_counter_thread_field *
 241                 get_thread_field(struct lttng_perf_counter_field *field)
 242 {
 243         struct lttng_perf_counter_thread *perf_thread;
 244         struct lttng_perf_counter_thread_field *thread_field;
 245
 246         perf_thread = pthread_getspecific(perf_counter_key);
 247         if (!perf_thread)
 248                 perf_thread = alloc_perf_counter_thread();
 249         cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list,
 250                         rcu_field_node) {
 251                 if (thread_field->field == field)
 252                         return thread_field;
 253         }
 254         /* perf_counter_thread_field not found, need to add one */
 255         return add_thread_field(field, perf_thread);
 256 }
 257
 258 static
 259 uint64_t wrapper_perf_counter_read(struct lttng_ctx_field *field)
 260 {
 261         struct lttng_perf_counter_field *perf_field;
 262         struct lttng_perf_counter_thread_field *perf_thread_field;
 263
 264         perf_field = field->u.perf_counter;
 265         perf_thread_field = get_thread_field(perf_field);
 266         return read_perf_counter(perf_thread_field->pc);
 267 }
 268
 269 static
 270 void perf_counter_record(struct lttng_ctx_field *field,
 271                  struct lttng_ust_lib_ring_buffer_ctx *ctx,
 272                  struct lttng_channel *chan)
 273 {
 274         uint64_t value;
 275
 276         value = wrapper_perf_counter_read(field);
 277         lib_ring_buffer_align_ctx(ctx, lttng_alignof(value));
 278         chan->ops->event_write(ctx, &value, sizeof(value));
 279 }
 280
 281 static
 282 void perf_counter_get_value(struct lttng_ctx_field *field,
 283                 union lttng_ctx_value *value)
 284 {
 285         uint64_t v;
 286
 287         v = wrapper_perf_counter_read(field);
 288         value->s64 = v;
 289 }
 290
 291 /* Called with UST lock held */
 292 static
 293 void lttng_destroy_perf_thread_field(
 294                 struct lttng_perf_counter_thread_field *thread_field)
 295 {
 296         unmap_perf_page(thread_field->pc);
 297         cds_list_del_rcu(&thread_field->rcu_field_node);
 298         cds_list_del(&thread_field->thread_field_node);
 299         free(thread_field);
 300 }
 301
 302 static
 303 void lttng_destroy_perf_thread_key(void *_key)
 304 {
 305         struct lttng_perf_counter_thread *perf_thread = _key;
 306         struct lttng_perf_counter_thread_field *pos, *p;
 307
 308         ust_lock_nocheck();
 309         cds_list_for_each_entry_safe(pos, p, &perf_thread->rcu_field_list,
 310                         rcu_field_node)
 311                 lttng_destroy_perf_thread_field(pos);
 312         ust_unlock();
 313         free(perf_thread);
 314 }
 315
 316 /* Called with UST lock held */
 317 static
 318 void lttng_destroy_perf_counter_field(struct lttng_ctx_field *field)
 319 {
 320         struct lttng_perf_counter_field *perf_field;
 321         struct lttng_perf_counter_thread_field *pos, *p;
 322
 323         free((char *) field->event_field.name);
 324         perf_field = field->u.perf_counter;
 325         /*
 326          * This put is performed when no threads can concurrently
 327          * perform a "get" concurrently, thanks to urcu-bp grace
 328          * period.
 329          */
 330         cds_list_for_each_entry_safe(pos, p, &perf_field->thread_field_list,
 331                         thread_field_node)
 332                 lttng_destroy_perf_thread_field(pos);
 333         free(perf_field);
 334 }
 335
 336 /* Called with UST lock held */
 337 int lttng_add_perf_counter_to_ctx(uint32_t type,
 338                                 uint64_t config,
 339                                 const char *name,
 340                                 struct lttng_ctx **ctx)
 341 {
 342         struct lttng_ctx_field *field;
 343         struct lttng_perf_counter_field *perf_field;
 344         struct perf_event_mmap_page *tmp_pc;
 345         char *name_alloc;
 346         int ret;
 347
 348         name_alloc = strdup(name);
 349         if (!name_alloc) {
 350                 ret = -ENOMEM;
 351                 goto name_alloc_error;
 352         }
 353         perf_field = zmalloc(sizeof(*perf_field));
 354         if (!perf_field) {
 355                 ret = -ENOMEM;
 356                 goto perf_field_alloc_error;
 357         }
 358         field = lttng_append_context(ctx);
 359         if (!field) {
 360                 ret = -ENOMEM;
 361                 goto append_context_error;
 362         }
 363         if (lttng_find_context(*ctx, name_alloc)) {
 364                 ret = -EEXIST;
 365                 goto find_error;
 366         }
 367
 368         field->destroy = lttng_destroy_perf_counter_field;
 369
 370         field->event_field.name = name_alloc;
 371         field->event_field.type.atype = atype_integer;
 372         field->event_field.type.u.basic.integer.size =
 373                         sizeof(uint64_t) * CHAR_BIT;
 374         field->event_field.type.u.basic.integer.alignment =
 375                         lttng_alignof(uint64_t) * CHAR_BIT;
 376         field->event_field.type.u.basic.integer.signedness =
 377                         lttng_is_signed_type(uint64_t);
 378         field->event_field.type.u.basic.integer.reverse_byte_order = 0;
 379         field->event_field.type.u.basic.integer.base = 10;
 380         field->event_field.type.u.basic.integer.encoding = lttng_encode_none;
 381         field->get_size = perf_counter_get_size;
 382         field->record = perf_counter_record;
 383         field->get_value = perf_counter_get_value;
 384
 385         perf_field->attr.type = type;
 386         perf_field->attr.config = config;
 387         perf_field->attr.exclude_kernel = 1;
 388         CDS_INIT_LIST_HEAD(&perf_field->thread_field_list);
 389         field->u.perf_counter = perf_field;
 390
 391         /* Ensure that this perf counter can be used in this process. */
 392         tmp_pc = setup_perf(&perf_field->attr);
 393         if (!tmp_pc) {
 394                 ret = -ENODEV;
 395                 goto setup_error;
 396         }
 397         unmap_perf_page(tmp_pc);
 398
 399         /*
 400          * Contexts can only be added before tracing is started, so we
 401          * don't have to synchronize against concurrent threads using
 402          * the field here.
 403          */
 404
 405         lttng_context_update(*ctx);
 406         return 0;
 407
 408 setup_error:
 409 find_error:
 410         lttng_remove_context_field(ctx, field);
 411 append_context_error:
 412         free(perf_field);
 413 perf_field_alloc_error:
 414         free(name_alloc);
 415 name_alloc_error:
 416         return ret;
 417 }
 418
 419 int lttng_perf_counter_init(void)
 420 {
 421         int ret;
 422
 423         ret = pthread_key_create(&perf_counter_key,
 424                         lttng_destroy_perf_thread_key);
 425         if (ret)
 426                 ret = -ret;
 427         return ret;
 428 }
 429
 430 void lttng_perf_counter_exit(void)
 431 {
 432         int ret;
 433
 434         ret = pthread_key_delete(perf_counter_key);
 435         if (ret) {
 436                 errno = ret;
 437                 PERROR("Error in pthread_key_delete");
 438         }
 439 }