| 1 | /* |
| 2 | * ring buffer based function tracer |
| 3 | * |
| 4 | * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com> |
| 5 | * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com> |
| 6 | * |
| 7 | * Originally taken from the RT patch by: |
| 8 | * Arnaldo Carvalho de Melo <acme@redhat.com> |
| 9 | * |
| 10 | * Based on code from the latency_tracer, that is: |
| 11 | * Copyright (C) 2004-2006 Ingo Molnar |
| 12 | * Copyright (C) 2004 Nadia Yvette Chambers |
| 13 | */ |
| 14 | #include <linux/ring_buffer.h> |
| 15 | #include <generated/utsrelease.h> |
| 16 | #include <linux/stacktrace.h> |
| 17 | #include <linux/writeback.h> |
| 18 | #include <linux/kallsyms.h> |
| 19 | #include <linux/seq_file.h> |
| 20 | #include <linux/notifier.h> |
| 21 | #include <linux/irqflags.h> |
| 22 | #include <linux/debugfs.h> |
| 23 | #include <linux/tracefs.h> |
| 24 | #include <linux/pagemap.h> |
| 25 | #include <linux/hardirq.h> |
| 26 | #include <linux/linkage.h> |
| 27 | #include <linux/uaccess.h> |
| 28 | #include <linux/kprobes.h> |
| 29 | #include <linux/ftrace.h> |
| 30 | #include <linux/module.h> |
| 31 | #include <linux/percpu.h> |
| 32 | #include <linux/splice.h> |
| 33 | #include <linux/kdebug.h> |
| 34 | #include <linux/string.h> |
| 35 | #include <linux/mount.h> |
| 36 | #include <linux/rwsem.h> |
| 37 | #include <linux/slab.h> |
| 38 | #include <linux/ctype.h> |
| 39 | #include <linux/init.h> |
| 40 | #include <linux/poll.h> |
| 41 | #include <linux/nmi.h> |
| 42 | #include <linux/fs.h> |
| 43 | #include <linux/sched/rt.h> |
| 44 | |
| 45 | #include "trace.h" |
| 46 | #include "trace_output.h" |
| 47 | |
| 48 | /* |
| 49 | * On boot up, the ring buffer is set to the minimum size, so that |
| 50 | * we do not waste memory on systems that are not using tracing. |
| 51 | */ |
| 52 | bool ring_buffer_expanded; |
| 53 | |
| 54 | /* |
| 55 | * We need to change this state when a selftest is running. |
| 56 | * A selftest will lurk into the ring-buffer to count the |
| 57 | * entries inserted during the selftest although some concurrent |
| 58 | * insertions into the ring-buffer such as trace_printk could occurred |
| 59 | * at the same time, giving false positive or negative results. |
| 60 | */ |
| 61 | static bool __read_mostly tracing_selftest_running; |
| 62 | |
| 63 | /* |
| 64 | * If a tracer is running, we do not want to run SELFTEST. |
| 65 | */ |
| 66 | bool __read_mostly tracing_selftest_disabled; |
| 67 | |
| 68 | /* Pipe tracepoints to printk */ |
| 69 | struct trace_iterator *tracepoint_print_iter; |
| 70 | int tracepoint_printk; |
| 71 | |
| 72 | /* For tracers that don't implement custom flags */ |
| 73 | static struct tracer_opt dummy_tracer_opt[] = { |
| 74 | { } |
| 75 | }; |
| 76 | |
| 77 | static struct tracer_flags dummy_tracer_flags = { |
| 78 | .val = 0, |
| 79 | .opts = dummy_tracer_opt |
| 80 | }; |
| 81 | |
| 82 | static int |
| 83 | dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set) |
| 84 | { |
| 85 | return 0; |
| 86 | } |
| 87 | |
| 88 | /* |
| 89 | * To prevent the comm cache from being overwritten when no |
| 90 | * tracing is active, only save the comm when a trace event |
| 91 | * occurred. |
| 92 | */ |
| 93 | static DEFINE_PER_CPU(bool, trace_cmdline_save); |
| 94 | |
| 95 | /* |
| 96 | * Kill all tracing for good (never come back). |
| 97 | * It is initialized to 1 but will turn to zero if the initialization |
| 98 | * of the tracer is successful. But that is the only place that sets |
| 99 | * this back to zero. |
| 100 | */ |
| 101 | static int tracing_disabled = 1; |
| 102 | |
| 103 | DEFINE_PER_CPU(int, ftrace_cpu_disabled); |
| 104 | |
| 105 | cpumask_var_t __read_mostly tracing_buffer_mask; |
| 106 | |
| 107 | /* |
| 108 | * ftrace_dump_on_oops - variable to dump ftrace buffer on oops |
| 109 | * |
| 110 | * If there is an oops (or kernel panic) and the ftrace_dump_on_oops |
| 111 | * is set, then ftrace_dump is called. This will output the contents |
| 112 | * of the ftrace buffers to the console. This is very useful for |
| 113 | * capturing traces that lead to crashes and outputing it to a |
| 114 | * serial console. |
| 115 | * |
| 116 | * It is default off, but you can enable it with either specifying |
| 117 | * "ftrace_dump_on_oops" in the kernel command line, or setting |
| 118 | * /proc/sys/kernel/ftrace_dump_on_oops |
| 119 | * Set 1 if you want to dump buffers of all CPUs |
| 120 | * Set 2 if you want to dump the buffer of the CPU that triggered oops |
| 121 | */ |
| 122 | |
| 123 | enum ftrace_dump_mode ftrace_dump_on_oops; |
| 124 | |
| 125 | /* When set, tracing will stop when a WARN*() is hit */ |
| 126 | int __disable_trace_on_warning; |
| 127 | |
| 128 | #ifdef CONFIG_TRACE_ENUM_MAP_FILE |
| 129 | /* Map of enums to their values, for "enum_map" file */ |
| 130 | struct trace_enum_map_head { |
| 131 | struct module *mod; |
| 132 | unsigned long length; |
| 133 | }; |
| 134 | |
| 135 | union trace_enum_map_item; |
| 136 | |
| 137 | struct trace_enum_map_tail { |
| 138 | /* |
| 139 | * "end" is first and points to NULL as it must be different |
| 140 | * than "mod" or "enum_string" |
| 141 | */ |
| 142 | union trace_enum_map_item *next; |
| 143 | const char *end; /* points to NULL */ |
| 144 | }; |
| 145 | |
| 146 | static DEFINE_MUTEX(trace_enum_mutex); |
| 147 | |
| 148 | /* |
| 149 | * The trace_enum_maps are saved in an array with two extra elements, |
| 150 | * one at the beginning, and one at the end. The beginning item contains |
| 151 | * the count of the saved maps (head.length), and the module they |
| 152 | * belong to if not built in (head.mod). The ending item contains a |
| 153 | * pointer to the next array of saved enum_map items. |
| 154 | */ |
| 155 | union trace_enum_map_item { |
| 156 | struct trace_enum_map map; |
| 157 | struct trace_enum_map_head head; |
| 158 | struct trace_enum_map_tail tail; |
| 159 | }; |
| 160 | |
| 161 | static union trace_enum_map_item *trace_enum_maps; |
| 162 | #endif /* CONFIG_TRACE_ENUM_MAP_FILE */ |
| 163 | |
| 164 | static int tracing_set_tracer(struct trace_array *tr, const char *buf); |
| 165 | |
| 166 | #define MAX_TRACER_SIZE 100 |
| 167 | static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata; |
| 168 | static char *default_bootup_tracer; |
| 169 | |
| 170 | static bool allocate_snapshot; |
| 171 | |
| 172 | static int __init set_cmdline_ftrace(char *str) |
| 173 | { |
| 174 | strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE); |
| 175 | default_bootup_tracer = bootup_tracer_buf; |
| 176 | /* We are using ftrace early, expand it */ |
| 177 | ring_buffer_expanded = true; |
| 178 | return 1; |
| 179 | } |
| 180 | __setup("ftrace=", set_cmdline_ftrace); |
| 181 | |
| 182 | static int __init set_ftrace_dump_on_oops(char *str) |
| 183 | { |
| 184 | if (*str++ != '=' || !*str) { |
| 185 | ftrace_dump_on_oops = DUMP_ALL; |
| 186 | return 1; |
| 187 | } |
| 188 | |
| 189 | if (!strcmp("orig_cpu", str)) { |
| 190 | ftrace_dump_on_oops = DUMP_ORIG; |
| 191 | return 1; |
| 192 | } |
| 193 | |
| 194 | return 0; |
| 195 | } |
| 196 | __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops); |
| 197 | |
| 198 | static int __init stop_trace_on_warning(char *str) |
| 199 | { |
| 200 | if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0)) |
| 201 | __disable_trace_on_warning = 1; |
| 202 | return 1; |
| 203 | } |
| 204 | __setup("traceoff_on_warning", stop_trace_on_warning); |
| 205 | |
| 206 | static int __init boot_alloc_snapshot(char *str) |
| 207 | { |
| 208 | allocate_snapshot = true; |
| 209 | /* We also need the main ring buffer expanded */ |
| 210 | ring_buffer_expanded = true; |
| 211 | return 1; |
| 212 | } |
| 213 | __setup("alloc_snapshot", boot_alloc_snapshot); |
| 214 | |
| 215 | |
| 216 | static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata; |
| 217 | static char *trace_boot_options __initdata; |
| 218 | |
| 219 | static int __init set_trace_boot_options(char *str) |
| 220 | { |
| 221 | strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE); |
| 222 | trace_boot_options = trace_boot_options_buf; |
| 223 | return 0; |
| 224 | } |
| 225 | __setup("trace_options=", set_trace_boot_options); |
| 226 | |
| 227 | static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata; |
| 228 | static char *trace_boot_clock __initdata; |
| 229 | |
| 230 | static int __init set_trace_boot_clock(char *str) |
| 231 | { |
| 232 | strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE); |
| 233 | trace_boot_clock = trace_boot_clock_buf; |
| 234 | return 0; |
| 235 | } |
| 236 | __setup("trace_clock=", set_trace_boot_clock); |
| 237 | |
| 238 | static int __init set_tracepoint_printk(char *str) |
| 239 | { |
| 240 | if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0)) |
| 241 | tracepoint_printk = 1; |
| 242 | return 1; |
| 243 | } |
| 244 | __setup("tp_printk", set_tracepoint_printk); |
| 245 | |
| 246 | unsigned long long ns2usecs(cycle_t nsec) |
| 247 | { |
| 248 | nsec += 500; |
| 249 | do_div(nsec, 1000); |
| 250 | return nsec; |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * The global_trace is the descriptor that holds the tracing |
| 255 | * buffers for the live tracing. For each CPU, it contains |
| 256 | * a link list of pages that will store trace entries. The |
| 257 | * page descriptor of the pages in the memory is used to hold |
| 258 | * the link list by linking the lru item in the page descriptor |
| 259 | * to each of the pages in the buffer per CPU. |
| 260 | * |
| 261 | * For each active CPU there is a data field that holds the |
| 262 | * pages for the buffer for that CPU. Each CPU has the same number |
| 263 | * of pages allocated for its buffer. |
| 264 | */ |
| 265 | static struct trace_array global_trace; |
| 266 | |
| 267 | LIST_HEAD(ftrace_trace_arrays); |
| 268 | |
| 269 | int trace_array_get(struct trace_array *this_tr) |
| 270 | { |
| 271 | struct trace_array *tr; |
| 272 | int ret = -ENODEV; |
| 273 | |
| 274 | mutex_lock(&trace_types_lock); |
| 275 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { |
| 276 | if (tr == this_tr) { |
| 277 | tr->ref++; |
| 278 | ret = 0; |
| 279 | break; |
| 280 | } |
| 281 | } |
| 282 | mutex_unlock(&trace_types_lock); |
| 283 | |
| 284 | return ret; |
| 285 | } |
| 286 | |
| 287 | static void __trace_array_put(struct trace_array *this_tr) |
| 288 | { |
| 289 | WARN_ON(!this_tr->ref); |
| 290 | this_tr->ref--; |
| 291 | } |
| 292 | |
| 293 | void trace_array_put(struct trace_array *this_tr) |
| 294 | { |
| 295 | mutex_lock(&trace_types_lock); |
| 296 | __trace_array_put(this_tr); |
| 297 | mutex_unlock(&trace_types_lock); |
| 298 | } |
| 299 | |
| 300 | int filter_check_discard(struct trace_event_file *file, void *rec, |
| 301 | struct ring_buffer *buffer, |
| 302 | struct ring_buffer_event *event) |
| 303 | { |
| 304 | if (unlikely(file->flags & EVENT_FILE_FL_FILTERED) && |
| 305 | !filter_match_preds(file->filter, rec)) { |
| 306 | ring_buffer_discard_commit(buffer, event); |
| 307 | return 1; |
| 308 | } |
| 309 | |
| 310 | return 0; |
| 311 | } |
| 312 | EXPORT_SYMBOL_GPL(filter_check_discard); |
| 313 | |
| 314 | int call_filter_check_discard(struct trace_event_call *call, void *rec, |
| 315 | struct ring_buffer *buffer, |
| 316 | struct ring_buffer_event *event) |
| 317 | { |
| 318 | if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) && |
| 319 | !filter_match_preds(call->filter, rec)) { |
| 320 | ring_buffer_discard_commit(buffer, event); |
| 321 | return 1; |
| 322 | } |
| 323 | |
| 324 | return 0; |
| 325 | } |
| 326 | EXPORT_SYMBOL_GPL(call_filter_check_discard); |
| 327 | |
| 328 | static cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu) |
| 329 | { |
| 330 | u64 ts; |
| 331 | |
| 332 | /* Early boot up does not have a buffer yet */ |
| 333 | if (!buf->buffer) |
| 334 | return trace_clock_local(); |
| 335 | |
| 336 | ts = ring_buffer_time_stamp(buf->buffer, cpu); |
| 337 | ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts); |
| 338 | |
| 339 | return ts; |
| 340 | } |
| 341 | |
| 342 | cycle_t ftrace_now(int cpu) |
| 343 | { |
| 344 | return buffer_ftrace_now(&global_trace.trace_buffer, cpu); |
| 345 | } |
| 346 | |
| 347 | /** |
| 348 | * tracing_is_enabled - Show if global_trace has been disabled |
| 349 | * |
| 350 | * Shows if the global trace has been enabled or not. It uses the |
| 351 | * mirror flag "buffer_disabled" to be used in fast paths such as for |
| 352 | * the irqsoff tracer. But it may be inaccurate due to races. If you |
| 353 | * need to know the accurate state, use tracing_is_on() which is a little |
| 354 | * slower, but accurate. |
| 355 | */ |
| 356 | int tracing_is_enabled(void) |
| 357 | { |
| 358 | /* |
| 359 | * For quick access (irqsoff uses this in fast path), just |
| 360 | * return the mirror variable of the state of the ring buffer. |
| 361 | * It's a little racy, but we don't really care. |
| 362 | */ |
| 363 | smp_rmb(); |
| 364 | return !global_trace.buffer_disabled; |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * trace_buf_size is the size in bytes that is allocated |
| 369 | * for a buffer. Note, the number of bytes is always rounded |
| 370 | * to page size. |
| 371 | * |
| 372 | * This number is purposely set to a low number of 16384. |
| 373 | * If the dump on oops happens, it will be much appreciated |
| 374 | * to not have to wait for all that output. Anyway this can be |
| 375 | * boot time and run time configurable. |
| 376 | */ |
| 377 | #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */ |
| 378 | |
| 379 | static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT; |
| 380 | |
| 381 | /* trace_types holds a link list of available tracers. */ |
| 382 | static struct tracer *trace_types __read_mostly; |
| 383 | |
| 384 | /* |
| 385 | * trace_types_lock is used to protect the trace_types list. |
| 386 | */ |
| 387 | DEFINE_MUTEX(trace_types_lock); |
| 388 | |
| 389 | /* |
| 390 | * serialize the access of the ring buffer |
| 391 | * |
| 392 | * ring buffer serializes readers, but it is low level protection. |
| 393 | * The validity of the events (which returns by ring_buffer_peek() ..etc) |
| 394 | * are not protected by ring buffer. |
| 395 | * |
| 396 | * The content of events may become garbage if we allow other process consumes |
| 397 | * these events concurrently: |
| 398 | * A) the page of the consumed events may become a normal page |
| 399 | * (not reader page) in ring buffer, and this page will be rewrited |
| 400 | * by events producer. |
| 401 | * B) The page of the consumed events may become a page for splice_read, |
| 402 | * and this page will be returned to system. |
| 403 | * |
| 404 | * These primitives allow multi process access to different cpu ring buffer |
| 405 | * concurrently. |
| 406 | * |
| 407 | * These primitives don't distinguish read-only and read-consume access. |
| 408 | * Multi read-only access are also serialized. |
| 409 | */ |
| 410 | |
| 411 | #ifdef CONFIG_SMP |
| 412 | static DECLARE_RWSEM(all_cpu_access_lock); |
| 413 | static DEFINE_PER_CPU(struct mutex, cpu_access_lock); |
| 414 | |
| 415 | static inline void trace_access_lock(int cpu) |
| 416 | { |
| 417 | if (cpu == RING_BUFFER_ALL_CPUS) { |
| 418 | /* gain it for accessing the whole ring buffer. */ |
| 419 | down_write(&all_cpu_access_lock); |
| 420 | } else { |
| 421 | /* gain it for accessing a cpu ring buffer. */ |
| 422 | |
| 423 | /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */ |
| 424 | down_read(&all_cpu_access_lock); |
| 425 | |
| 426 | /* Secondly block other access to this @cpu ring buffer. */ |
| 427 | mutex_lock(&per_cpu(cpu_access_lock, cpu)); |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | static inline void trace_access_unlock(int cpu) |
| 432 | { |
| 433 | if (cpu == RING_BUFFER_ALL_CPUS) { |
| 434 | up_write(&all_cpu_access_lock); |
| 435 | } else { |
| 436 | mutex_unlock(&per_cpu(cpu_access_lock, cpu)); |
| 437 | up_read(&all_cpu_access_lock); |
| 438 | } |
| 439 | } |
| 440 | |
| 441 | static inline void trace_access_lock_init(void) |
| 442 | { |
| 443 | int cpu; |
| 444 | |
| 445 | for_each_possible_cpu(cpu) |
| 446 | mutex_init(&per_cpu(cpu_access_lock, cpu)); |
| 447 | } |
| 448 | |
| 449 | #else |
| 450 | |
| 451 | static DEFINE_MUTEX(access_lock); |
| 452 | |
| 453 | static inline void trace_access_lock(int cpu) |
| 454 | { |
| 455 | (void)cpu; |
| 456 | mutex_lock(&access_lock); |
| 457 | } |
| 458 | |
| 459 | static inline void trace_access_unlock(int cpu) |
| 460 | { |
| 461 | (void)cpu; |
| 462 | mutex_unlock(&access_lock); |
| 463 | } |
| 464 | |
| 465 | static inline void trace_access_lock_init(void) |
| 466 | { |
| 467 | } |
| 468 | |
| 469 | #endif |
| 470 | |
| 471 | #ifdef CONFIG_STACKTRACE |
| 472 | static void __ftrace_trace_stack(struct ring_buffer *buffer, |
| 473 | unsigned long flags, |
| 474 | int skip, int pc, struct pt_regs *regs); |
| 475 | static inline void ftrace_trace_stack(struct ring_buffer *buffer, |
| 476 | unsigned long flags, |
| 477 | int skip, int pc, struct pt_regs *regs); |
| 478 | |
| 479 | #else |
| 480 | static inline void __ftrace_trace_stack(struct ring_buffer *buffer, |
| 481 | unsigned long flags, |
| 482 | int skip, int pc, struct pt_regs *regs) |
| 483 | { |
| 484 | } |
| 485 | static inline void ftrace_trace_stack(struct ring_buffer *buffer, |
| 486 | unsigned long flags, |
| 487 | int skip, int pc, struct pt_regs *regs) |
| 488 | { |
| 489 | } |
| 490 | |
| 491 | #endif |
| 492 | |
| 493 | /* trace_flags holds trace_options default values */ |
| 494 | unsigned long trace_flags = |
| 495 | FUNCTION_DEFAULT_FLAGS | FUNCTION_GRAPH_DEFAULT_FLAGS | |
| 496 | TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | |
| 497 | TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | |
| 498 | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | |
| 499 | TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS |
| 500 | ; |
| 501 | |
| 502 | static void tracer_tracing_on(struct trace_array *tr) |
| 503 | { |
| 504 | if (tr->trace_buffer.buffer) |
| 505 | ring_buffer_record_on(tr->trace_buffer.buffer); |
| 506 | /* |
| 507 | * This flag is looked at when buffers haven't been allocated |
| 508 | * yet, or by some tracers (like irqsoff), that just want to |
| 509 | * know if the ring buffer has been disabled, but it can handle |
| 510 | * races of where it gets disabled but we still do a record. |
| 511 | * As the check is in the fast path of the tracers, it is more |
| 512 | * important to be fast than accurate. |
| 513 | */ |
| 514 | tr->buffer_disabled = 0; |
| 515 | /* Make the flag seen by readers */ |
| 516 | smp_wmb(); |
| 517 | } |
| 518 | |
| 519 | /** |
| 520 | * tracing_on - enable tracing buffers |
| 521 | * |
| 522 | * This function enables tracing buffers that may have been |
| 523 | * disabled with tracing_off. |
| 524 | */ |
| 525 | void tracing_on(void) |
| 526 | { |
| 527 | tracer_tracing_on(&global_trace); |
| 528 | } |
| 529 | EXPORT_SYMBOL_GPL(tracing_on); |
| 530 | |
| 531 | /** |
| 532 | * __trace_puts - write a constant string into the trace buffer. |
| 533 | * @ip: The address of the caller |
| 534 | * @str: The constant string to write |
| 535 | * @size: The size of the string. |
| 536 | */ |
| 537 | int __trace_puts(unsigned long ip, const char *str, int size) |
| 538 | { |
| 539 | struct ring_buffer_event *event; |
| 540 | struct ring_buffer *buffer; |
| 541 | struct print_entry *entry; |
| 542 | unsigned long irq_flags; |
| 543 | int alloc; |
| 544 | int pc; |
| 545 | |
| 546 | if (!(trace_flags & TRACE_ITER_PRINTK)) |
| 547 | return 0; |
| 548 | |
| 549 | pc = preempt_count(); |
| 550 | |
| 551 | if (unlikely(tracing_selftest_running || tracing_disabled)) |
| 552 | return 0; |
| 553 | |
| 554 | alloc = sizeof(*entry) + size + 2; /* possible \n added */ |
| 555 | |
| 556 | local_save_flags(irq_flags); |
| 557 | buffer = global_trace.trace_buffer.buffer; |
| 558 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc, |
| 559 | irq_flags, pc); |
| 560 | if (!event) |
| 561 | return 0; |
| 562 | |
| 563 | entry = ring_buffer_event_data(event); |
| 564 | entry->ip = ip; |
| 565 | |
| 566 | memcpy(&entry->buf, str, size); |
| 567 | |
| 568 | /* Add a newline if necessary */ |
| 569 | if (entry->buf[size - 1] != '\n') { |
| 570 | entry->buf[size] = '\n'; |
| 571 | entry->buf[size + 1] = '\0'; |
| 572 | } else |
| 573 | entry->buf[size] = '\0'; |
| 574 | |
| 575 | __buffer_unlock_commit(buffer, event); |
| 576 | ftrace_trace_stack(buffer, irq_flags, 4, pc, NULL); |
| 577 | |
| 578 | return size; |
| 579 | } |
| 580 | EXPORT_SYMBOL_GPL(__trace_puts); |
| 581 | |
| 582 | /** |
| 583 | * __trace_bputs - write the pointer to a constant string into trace buffer |
| 584 | * @ip: The address of the caller |
| 585 | * @str: The constant string to write to the buffer to |
| 586 | */ |
| 587 | int __trace_bputs(unsigned long ip, const char *str) |
| 588 | { |
| 589 | struct ring_buffer_event *event; |
| 590 | struct ring_buffer *buffer; |
| 591 | struct bputs_entry *entry; |
| 592 | unsigned long irq_flags; |
| 593 | int size = sizeof(struct bputs_entry); |
| 594 | int pc; |
| 595 | |
| 596 | if (!(trace_flags & TRACE_ITER_PRINTK)) |
| 597 | return 0; |
| 598 | |
| 599 | pc = preempt_count(); |
| 600 | |
| 601 | if (unlikely(tracing_selftest_running || tracing_disabled)) |
| 602 | return 0; |
| 603 | |
| 604 | local_save_flags(irq_flags); |
| 605 | buffer = global_trace.trace_buffer.buffer; |
| 606 | event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size, |
| 607 | irq_flags, pc); |
| 608 | if (!event) |
| 609 | return 0; |
| 610 | |
| 611 | entry = ring_buffer_event_data(event); |
| 612 | entry->ip = ip; |
| 613 | entry->str = str; |
| 614 | |
| 615 | __buffer_unlock_commit(buffer, event); |
| 616 | ftrace_trace_stack(buffer, irq_flags, 4, pc, NULL); |
| 617 | |
| 618 | return 1; |
| 619 | } |
| 620 | EXPORT_SYMBOL_GPL(__trace_bputs); |
| 621 | |
| 622 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 623 | /** |
| 624 | * trace_snapshot - take a snapshot of the current buffer. |
| 625 | * |
| 626 | * This causes a swap between the snapshot buffer and the current live |
| 627 | * tracing buffer. You can use this to take snapshots of the live |
| 628 | * trace when some condition is triggered, but continue to trace. |
| 629 | * |
| 630 | * Note, make sure to allocate the snapshot with either |
| 631 | * a tracing_snapshot_alloc(), or by doing it manually |
| 632 | * with: echo 1 > /sys/kernel/debug/tracing/snapshot |
| 633 | * |
| 634 | * If the snapshot buffer is not allocated, it will stop tracing. |
| 635 | * Basically making a permanent snapshot. |
| 636 | */ |
| 637 | void tracing_snapshot(void) |
| 638 | { |
| 639 | struct trace_array *tr = &global_trace; |
| 640 | struct tracer *tracer = tr->current_trace; |
| 641 | unsigned long flags; |
| 642 | |
| 643 | if (in_nmi()) { |
| 644 | internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n"); |
| 645 | internal_trace_puts("*** snapshot is being ignored ***\n"); |
| 646 | return; |
| 647 | } |
| 648 | |
| 649 | if (!tr->allocated_snapshot) { |
| 650 | internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n"); |
| 651 | internal_trace_puts("*** stopping trace here! ***\n"); |
| 652 | tracing_off(); |
| 653 | return; |
| 654 | } |
| 655 | |
| 656 | /* Note, snapshot can not be used when the tracer uses it */ |
| 657 | if (tracer->use_max_tr) { |
| 658 | internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n"); |
| 659 | internal_trace_puts("*** Can not use snapshot (sorry) ***\n"); |
| 660 | return; |
| 661 | } |
| 662 | |
| 663 | local_irq_save(flags); |
| 664 | update_max_tr(tr, current, smp_processor_id()); |
| 665 | local_irq_restore(flags); |
| 666 | } |
| 667 | EXPORT_SYMBOL_GPL(tracing_snapshot); |
| 668 | |
| 669 | static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf, |
| 670 | struct trace_buffer *size_buf, int cpu_id); |
| 671 | static void set_buffer_entries(struct trace_buffer *buf, unsigned long val); |
| 672 | |
| 673 | static int alloc_snapshot(struct trace_array *tr) |
| 674 | { |
| 675 | int ret; |
| 676 | |
| 677 | if (!tr->allocated_snapshot) { |
| 678 | |
| 679 | /* allocate spare buffer */ |
| 680 | ret = resize_buffer_duplicate_size(&tr->max_buffer, |
| 681 | &tr->trace_buffer, RING_BUFFER_ALL_CPUS); |
| 682 | if (ret < 0) |
| 683 | return ret; |
| 684 | |
| 685 | tr->allocated_snapshot = true; |
| 686 | } |
| 687 | |
| 688 | return 0; |
| 689 | } |
| 690 | |
| 691 | static void free_snapshot(struct trace_array *tr) |
| 692 | { |
| 693 | /* |
| 694 | * We don't free the ring buffer. instead, resize it because |
| 695 | * The max_tr ring buffer has some state (e.g. ring->clock) and |
| 696 | * we want preserve it. |
| 697 | */ |
| 698 | ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS); |
| 699 | set_buffer_entries(&tr->max_buffer, 1); |
| 700 | tracing_reset_online_cpus(&tr->max_buffer); |
| 701 | tr->allocated_snapshot = false; |
| 702 | } |
| 703 | |
| 704 | /** |
| 705 | * tracing_alloc_snapshot - allocate snapshot buffer. |
| 706 | * |
| 707 | * This only allocates the snapshot buffer if it isn't already |
| 708 | * allocated - it doesn't also take a snapshot. |
| 709 | * |
| 710 | * This is meant to be used in cases where the snapshot buffer needs |
| 711 | * to be set up for events that can't sleep but need to be able to |
| 712 | * trigger a snapshot. |
| 713 | */ |
| 714 | int tracing_alloc_snapshot(void) |
| 715 | { |
| 716 | struct trace_array *tr = &global_trace; |
| 717 | int ret; |
| 718 | |
| 719 | ret = alloc_snapshot(tr); |
| 720 | WARN_ON(ret < 0); |
| 721 | |
| 722 | return ret; |
| 723 | } |
| 724 | EXPORT_SYMBOL_GPL(tracing_alloc_snapshot); |
| 725 | |
| 726 | /** |
| 727 | * trace_snapshot_alloc - allocate and take a snapshot of the current buffer. |
| 728 | * |
| 729 | * This is similar to trace_snapshot(), but it will allocate the |
| 730 | * snapshot buffer if it isn't already allocated. Use this only |
| 731 | * where it is safe to sleep, as the allocation may sleep. |
| 732 | * |
| 733 | * This causes a swap between the snapshot buffer and the current live |
| 734 | * tracing buffer. You can use this to take snapshots of the live |
| 735 | * trace when some condition is triggered, but continue to trace. |
| 736 | */ |
| 737 | void tracing_snapshot_alloc(void) |
| 738 | { |
| 739 | int ret; |
| 740 | |
| 741 | ret = tracing_alloc_snapshot(); |
| 742 | if (ret < 0) |
| 743 | return; |
| 744 | |
| 745 | tracing_snapshot(); |
| 746 | } |
| 747 | EXPORT_SYMBOL_GPL(tracing_snapshot_alloc); |
| 748 | #else |
| 749 | void tracing_snapshot(void) |
| 750 | { |
| 751 | WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used"); |
| 752 | } |
| 753 | EXPORT_SYMBOL_GPL(tracing_snapshot); |
| 754 | int tracing_alloc_snapshot(void) |
| 755 | { |
| 756 | WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used"); |
| 757 | return -ENODEV; |
| 758 | } |
| 759 | EXPORT_SYMBOL_GPL(tracing_alloc_snapshot); |
| 760 | void tracing_snapshot_alloc(void) |
| 761 | { |
| 762 | /* Give warning */ |
| 763 | tracing_snapshot(); |
| 764 | } |
| 765 | EXPORT_SYMBOL_GPL(tracing_snapshot_alloc); |
| 766 | #endif /* CONFIG_TRACER_SNAPSHOT */ |
| 767 | |
| 768 | static void tracer_tracing_off(struct trace_array *tr) |
| 769 | { |
| 770 | if (tr->trace_buffer.buffer) |
| 771 | ring_buffer_record_off(tr->trace_buffer.buffer); |
| 772 | /* |
| 773 | * This flag is looked at when buffers haven't been allocated |
| 774 | * yet, or by some tracers (like irqsoff), that just want to |
| 775 | * know if the ring buffer has been disabled, but it can handle |
| 776 | * races of where it gets disabled but we still do a record. |
| 777 | * As the check is in the fast path of the tracers, it is more |
| 778 | * important to be fast than accurate. |
| 779 | */ |
| 780 | tr->buffer_disabled = 1; |
| 781 | /* Make the flag seen by readers */ |
| 782 | smp_wmb(); |
| 783 | } |
| 784 | |
| 785 | /** |
| 786 | * tracing_off - turn off tracing buffers |
| 787 | * |
| 788 | * This function stops the tracing buffers from recording data. |
| 789 | * It does not disable any overhead the tracers themselves may |
| 790 | * be causing. This function simply causes all recording to |
| 791 | * the ring buffers to fail. |
| 792 | */ |
| 793 | void tracing_off(void) |
| 794 | { |
| 795 | tracer_tracing_off(&global_trace); |
| 796 | } |
| 797 | EXPORT_SYMBOL_GPL(tracing_off); |
| 798 | |
| 799 | void disable_trace_on_warning(void) |
| 800 | { |
| 801 | if (__disable_trace_on_warning) |
| 802 | tracing_off(); |
| 803 | } |
| 804 | |
| 805 | /** |
| 806 | * tracer_tracing_is_on - show real state of ring buffer enabled |
| 807 | * @tr : the trace array to know if ring buffer is enabled |
| 808 | * |
| 809 | * Shows real state of the ring buffer if it is enabled or not. |
| 810 | */ |
| 811 | static int tracer_tracing_is_on(struct trace_array *tr) |
| 812 | { |
| 813 | if (tr->trace_buffer.buffer) |
| 814 | return ring_buffer_record_is_on(tr->trace_buffer.buffer); |
| 815 | return !tr->buffer_disabled; |
| 816 | } |
| 817 | |
| 818 | /** |
| 819 | * tracing_is_on - show state of ring buffers enabled |
| 820 | */ |
| 821 | int tracing_is_on(void) |
| 822 | { |
| 823 | return tracer_tracing_is_on(&global_trace); |
| 824 | } |
| 825 | EXPORT_SYMBOL_GPL(tracing_is_on); |
| 826 | |
| 827 | static int __init set_buf_size(char *str) |
| 828 | { |
| 829 | unsigned long buf_size; |
| 830 | |
| 831 | if (!str) |
| 832 | return 0; |
| 833 | buf_size = memparse(str, &str); |
| 834 | /* nr_entries can not be zero */ |
| 835 | if (buf_size == 0) |
| 836 | return 0; |
| 837 | trace_buf_size = buf_size; |
| 838 | return 1; |
| 839 | } |
| 840 | __setup("trace_buf_size=", set_buf_size); |
| 841 | |
| 842 | static int __init set_tracing_thresh(char *str) |
| 843 | { |
| 844 | unsigned long threshold; |
| 845 | int ret; |
| 846 | |
| 847 | if (!str) |
| 848 | return 0; |
| 849 | ret = kstrtoul(str, 0, &threshold); |
| 850 | if (ret < 0) |
| 851 | return 0; |
| 852 | tracing_thresh = threshold * 1000; |
| 853 | return 1; |
| 854 | } |
| 855 | __setup("tracing_thresh=", set_tracing_thresh); |
| 856 | |
| 857 | unsigned long nsecs_to_usecs(unsigned long nsecs) |
| 858 | { |
| 859 | return nsecs / 1000; |
| 860 | } |
| 861 | |
| 862 | /* |
| 863 | * TRACE_FLAGS is defined as a tuple matching bit masks with strings. |
| 864 | * It uses C(a, b) where 'a' is the enum name and 'b' is the string that |
| 865 | * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list |
| 866 | * of strings in the order that the enums were defined. |
| 867 | */ |
| 868 | #undef C |
| 869 | #define C(a, b) b |
| 870 | |
| 871 | /* These must match the bit postions in trace_iterator_flags */ |
| 872 | static const char *trace_options[] = { |
| 873 | TRACE_FLAGS |
| 874 | NULL |
| 875 | }; |
| 876 | |
| 877 | static struct { |
| 878 | u64 (*func)(void); |
| 879 | const char *name; |
| 880 | int in_ns; /* is this clock in nanoseconds? */ |
| 881 | } trace_clocks[] = { |
| 882 | { trace_clock_local, "local", 1 }, |
| 883 | { trace_clock_global, "global", 1 }, |
| 884 | { trace_clock_counter, "counter", 0 }, |
| 885 | { trace_clock_jiffies, "uptime", 0 }, |
| 886 | { trace_clock, "perf", 1 }, |
| 887 | { ktime_get_mono_fast_ns, "mono", 1 }, |
| 888 | { ktime_get_raw_fast_ns, "mono_raw", 1 }, |
| 889 | ARCH_TRACE_CLOCKS |
| 890 | }; |
| 891 | |
| 892 | /* |
| 893 | * trace_parser_get_init - gets the buffer for trace parser |
| 894 | */ |
| 895 | int trace_parser_get_init(struct trace_parser *parser, int size) |
| 896 | { |
| 897 | memset(parser, 0, sizeof(*parser)); |
| 898 | |
| 899 | parser->buffer = kmalloc(size, GFP_KERNEL); |
| 900 | if (!parser->buffer) |
| 901 | return 1; |
| 902 | |
| 903 | parser->size = size; |
| 904 | return 0; |
| 905 | } |
| 906 | |
| 907 | /* |
| 908 | * trace_parser_put - frees the buffer for trace parser |
| 909 | */ |
| 910 | void trace_parser_put(struct trace_parser *parser) |
| 911 | { |
| 912 | kfree(parser->buffer); |
| 913 | } |
| 914 | |
| 915 | /* |
| 916 | * trace_get_user - reads the user input string separated by space |
| 917 | * (matched by isspace(ch)) |
| 918 | * |
| 919 | * For each string found the 'struct trace_parser' is updated, |
| 920 | * and the function returns. |
| 921 | * |
| 922 | * Returns number of bytes read. |
| 923 | * |
| 924 | * See kernel/trace/trace.h for 'struct trace_parser' details. |
| 925 | */ |
| 926 | int trace_get_user(struct trace_parser *parser, const char __user *ubuf, |
| 927 | size_t cnt, loff_t *ppos) |
| 928 | { |
| 929 | char ch; |
| 930 | size_t read = 0; |
| 931 | ssize_t ret; |
| 932 | |
| 933 | if (!*ppos) |
| 934 | trace_parser_clear(parser); |
| 935 | |
| 936 | ret = get_user(ch, ubuf++); |
| 937 | if (ret) |
| 938 | goto out; |
| 939 | |
| 940 | read++; |
| 941 | cnt--; |
| 942 | |
| 943 | /* |
| 944 | * The parser is not finished with the last write, |
| 945 | * continue reading the user input without skipping spaces. |
| 946 | */ |
| 947 | if (!parser->cont) { |
| 948 | /* skip white space */ |
| 949 | while (cnt && isspace(ch)) { |
| 950 | ret = get_user(ch, ubuf++); |
| 951 | if (ret) |
| 952 | goto out; |
| 953 | read++; |
| 954 | cnt--; |
| 955 | } |
| 956 | |
| 957 | /* only spaces were written */ |
| 958 | if (isspace(ch)) { |
| 959 | *ppos += read; |
| 960 | ret = read; |
| 961 | goto out; |
| 962 | } |
| 963 | |
| 964 | parser->idx = 0; |
| 965 | } |
| 966 | |
| 967 | /* read the non-space input */ |
| 968 | while (cnt && !isspace(ch)) { |
| 969 | if (parser->idx < parser->size - 1) |
| 970 | parser->buffer[parser->idx++] = ch; |
| 971 | else { |
| 972 | ret = -EINVAL; |
| 973 | goto out; |
| 974 | } |
| 975 | ret = get_user(ch, ubuf++); |
| 976 | if (ret) |
| 977 | goto out; |
| 978 | read++; |
| 979 | cnt--; |
| 980 | } |
| 981 | |
| 982 | /* We either got finished input or we have to wait for another call. */ |
| 983 | if (isspace(ch)) { |
| 984 | parser->buffer[parser->idx] = 0; |
| 985 | parser->cont = false; |
| 986 | } else if (parser->idx < parser->size - 1) { |
| 987 | parser->cont = true; |
| 988 | parser->buffer[parser->idx++] = ch; |
| 989 | } else { |
| 990 | ret = -EINVAL; |
| 991 | goto out; |
| 992 | } |
| 993 | |
| 994 | *ppos += read; |
| 995 | ret = read; |
| 996 | |
| 997 | out: |
| 998 | return ret; |
| 999 | } |
| 1000 | |
| 1001 | /* TODO add a seq_buf_to_buffer() */ |
| 1002 | static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt) |
| 1003 | { |
| 1004 | int len; |
| 1005 | |
| 1006 | if (trace_seq_used(s) <= s->seq.readpos) |
| 1007 | return -EBUSY; |
| 1008 | |
| 1009 | len = trace_seq_used(s) - s->seq.readpos; |
| 1010 | if (cnt > len) |
| 1011 | cnt = len; |
| 1012 | memcpy(buf, s->buffer + s->seq.readpos, cnt); |
| 1013 | |
| 1014 | s->seq.readpos += cnt; |
| 1015 | return cnt; |
| 1016 | } |
| 1017 | |
| 1018 | unsigned long __read_mostly tracing_thresh; |
| 1019 | |
| 1020 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1021 | /* |
| 1022 | * Copy the new maximum trace into the separate maximum-trace |
| 1023 | * structure. (this way the maximum trace is permanently saved, |
| 1024 | * for later retrieval via /sys/kernel/debug/tracing/latency_trace) |
| 1025 | */ |
| 1026 | static void |
| 1027 | __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) |
| 1028 | { |
| 1029 | struct trace_buffer *trace_buf = &tr->trace_buffer; |
| 1030 | struct trace_buffer *max_buf = &tr->max_buffer; |
| 1031 | struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu); |
| 1032 | struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu); |
| 1033 | |
| 1034 | max_buf->cpu = cpu; |
| 1035 | max_buf->time_start = data->preempt_timestamp; |
| 1036 | |
| 1037 | max_data->saved_latency = tr->max_latency; |
| 1038 | max_data->critical_start = data->critical_start; |
| 1039 | max_data->critical_end = data->critical_end; |
| 1040 | |
| 1041 | memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN); |
| 1042 | max_data->pid = tsk->pid; |
| 1043 | /* |
| 1044 | * If tsk == current, then use current_uid(), as that does not use |
| 1045 | * RCU. The irq tracer can be called out of RCU scope. |
| 1046 | */ |
| 1047 | if (tsk == current) |
| 1048 | max_data->uid = current_uid(); |
| 1049 | else |
| 1050 | max_data->uid = task_uid(tsk); |
| 1051 | |
| 1052 | max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO; |
| 1053 | max_data->policy = tsk->policy; |
| 1054 | max_data->rt_priority = tsk->rt_priority; |
| 1055 | |
| 1056 | /* record this tasks comm */ |
| 1057 | tracing_record_cmdline(tsk); |
| 1058 | } |
| 1059 | |
| 1060 | /** |
| 1061 | * update_max_tr - snapshot all trace buffers from global_trace to max_tr |
| 1062 | * @tr: tracer |
| 1063 | * @tsk: the task with the latency |
| 1064 | * @cpu: The cpu that initiated the trace. |
| 1065 | * |
| 1066 | * Flip the buffers between the @tr and the max_tr and record information |
| 1067 | * about which task was the cause of this latency. |
| 1068 | */ |
| 1069 | void |
| 1070 | update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) |
| 1071 | { |
| 1072 | struct ring_buffer *buf; |
| 1073 | |
| 1074 | if (tr->stop_count) |
| 1075 | return; |
| 1076 | |
| 1077 | WARN_ON_ONCE(!irqs_disabled()); |
| 1078 | |
| 1079 | if (!tr->allocated_snapshot) { |
| 1080 | /* Only the nop tracer should hit this when disabling */ |
| 1081 | WARN_ON_ONCE(tr->current_trace != &nop_trace); |
| 1082 | return; |
| 1083 | } |
| 1084 | |
| 1085 | arch_spin_lock(&tr->max_lock); |
| 1086 | |
| 1087 | buf = tr->trace_buffer.buffer; |
| 1088 | tr->trace_buffer.buffer = tr->max_buffer.buffer; |
| 1089 | tr->max_buffer.buffer = buf; |
| 1090 | |
| 1091 | __update_max_tr(tr, tsk, cpu); |
| 1092 | arch_spin_unlock(&tr->max_lock); |
| 1093 | } |
| 1094 | |
| 1095 | /** |
| 1096 | * update_max_tr_single - only copy one trace over, and reset the rest |
| 1097 | * @tr - tracer |
| 1098 | * @tsk - task with the latency |
| 1099 | * @cpu - the cpu of the buffer to copy. |
| 1100 | * |
| 1101 | * Flip the trace of a single CPU buffer between the @tr and the max_tr. |
| 1102 | */ |
| 1103 | void |
| 1104 | update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu) |
| 1105 | { |
| 1106 | int ret; |
| 1107 | |
| 1108 | if (tr->stop_count) |
| 1109 | return; |
| 1110 | |
| 1111 | WARN_ON_ONCE(!irqs_disabled()); |
| 1112 | if (!tr->allocated_snapshot) { |
| 1113 | /* Only the nop tracer should hit this when disabling */ |
| 1114 | WARN_ON_ONCE(tr->current_trace != &nop_trace); |
| 1115 | return; |
| 1116 | } |
| 1117 | |
| 1118 | arch_spin_lock(&tr->max_lock); |
| 1119 | |
| 1120 | ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu); |
| 1121 | |
| 1122 | if (ret == -EBUSY) { |
| 1123 | /* |
| 1124 | * We failed to swap the buffer due to a commit taking |
| 1125 | * place on this CPU. We fail to record, but we reset |
| 1126 | * the max trace buffer (no one writes directly to it) |
| 1127 | * and flag that it failed. |
| 1128 | */ |
| 1129 | trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_, |
| 1130 | "Failed to swap buffers due to commit in progress\n"); |
| 1131 | } |
| 1132 | |
| 1133 | WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY); |
| 1134 | |
| 1135 | __update_max_tr(tr, tsk, cpu); |
| 1136 | arch_spin_unlock(&tr->max_lock); |
| 1137 | } |
| 1138 | #endif /* CONFIG_TRACER_MAX_TRACE */ |
| 1139 | |
| 1140 | static int wait_on_pipe(struct trace_iterator *iter, bool full) |
| 1141 | { |
| 1142 | /* Iterators are static, they should be filled or empty */ |
| 1143 | if (trace_buffer_iter(iter, iter->cpu_file)) |
| 1144 | return 0; |
| 1145 | |
| 1146 | return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file, |
| 1147 | full); |
| 1148 | } |
| 1149 | |
| 1150 | #ifdef CONFIG_FTRACE_STARTUP_TEST |
| 1151 | static int run_tracer_selftest(struct tracer *type) |
| 1152 | { |
| 1153 | struct trace_array *tr = &global_trace; |
| 1154 | struct tracer *saved_tracer = tr->current_trace; |
| 1155 | int ret; |
| 1156 | |
| 1157 | if (!type->selftest || tracing_selftest_disabled) |
| 1158 | return 0; |
| 1159 | |
| 1160 | /* |
| 1161 | * Run a selftest on this tracer. |
| 1162 | * Here we reset the trace buffer, and set the current |
| 1163 | * tracer to be this tracer. The tracer can then run some |
| 1164 | * internal tracing to verify that everything is in order. |
| 1165 | * If we fail, we do not register this tracer. |
| 1166 | */ |
| 1167 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 1168 | |
| 1169 | tr->current_trace = type; |
| 1170 | |
| 1171 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1172 | if (type->use_max_tr) { |
| 1173 | /* If we expanded the buffers, make sure the max is expanded too */ |
| 1174 | if (ring_buffer_expanded) |
| 1175 | ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size, |
| 1176 | RING_BUFFER_ALL_CPUS); |
| 1177 | tr->allocated_snapshot = true; |
| 1178 | } |
| 1179 | #endif |
| 1180 | |
| 1181 | /* the test is responsible for initializing and enabling */ |
| 1182 | pr_info("Testing tracer %s: ", type->name); |
| 1183 | ret = type->selftest(type, tr); |
| 1184 | /* the test is responsible for resetting too */ |
| 1185 | tr->current_trace = saved_tracer; |
| 1186 | if (ret) { |
| 1187 | printk(KERN_CONT "FAILED!\n"); |
| 1188 | /* Add the warning after printing 'FAILED' */ |
| 1189 | WARN_ON(1); |
| 1190 | return -1; |
| 1191 | } |
| 1192 | /* Only reset on passing, to avoid touching corrupted buffers */ |
| 1193 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 1194 | |
| 1195 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1196 | if (type->use_max_tr) { |
| 1197 | tr->allocated_snapshot = false; |
| 1198 | |
| 1199 | /* Shrink the max buffer again */ |
| 1200 | if (ring_buffer_expanded) |
| 1201 | ring_buffer_resize(tr->max_buffer.buffer, 1, |
| 1202 | RING_BUFFER_ALL_CPUS); |
| 1203 | } |
| 1204 | #endif |
| 1205 | |
| 1206 | printk(KERN_CONT "PASSED\n"); |
| 1207 | return 0; |
| 1208 | } |
| 1209 | #else |
| 1210 | static inline int run_tracer_selftest(struct tracer *type) |
| 1211 | { |
| 1212 | return 0; |
| 1213 | } |
| 1214 | #endif /* CONFIG_FTRACE_STARTUP_TEST */ |
| 1215 | |
| 1216 | static void add_tracer_options(struct trace_array *tr, struct tracer *t); |
| 1217 | |
| 1218 | /** |
| 1219 | * register_tracer - register a tracer with the ftrace system. |
| 1220 | * @type - the plugin for the tracer |
| 1221 | * |
| 1222 | * Register a new plugin tracer. |
| 1223 | */ |
| 1224 | int register_tracer(struct tracer *type) |
| 1225 | { |
| 1226 | struct tracer *t; |
| 1227 | int ret = 0; |
| 1228 | |
| 1229 | if (!type->name) { |
| 1230 | pr_info("Tracer must have a name\n"); |
| 1231 | return -1; |
| 1232 | } |
| 1233 | |
| 1234 | if (strlen(type->name) >= MAX_TRACER_SIZE) { |
| 1235 | pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE); |
| 1236 | return -1; |
| 1237 | } |
| 1238 | |
| 1239 | mutex_lock(&trace_types_lock); |
| 1240 | |
| 1241 | tracing_selftest_running = true; |
| 1242 | |
| 1243 | for (t = trace_types; t; t = t->next) { |
| 1244 | if (strcmp(type->name, t->name) == 0) { |
| 1245 | /* already found */ |
| 1246 | pr_info("Tracer %s already registered\n", |
| 1247 | type->name); |
| 1248 | ret = -1; |
| 1249 | goto out; |
| 1250 | } |
| 1251 | } |
| 1252 | |
| 1253 | if (!type->set_flag) |
| 1254 | type->set_flag = &dummy_set_flag; |
| 1255 | if (!type->flags) |
| 1256 | type->flags = &dummy_tracer_flags; |
| 1257 | else |
| 1258 | if (!type->flags->opts) |
| 1259 | type->flags->opts = dummy_tracer_opt; |
| 1260 | |
| 1261 | ret = run_tracer_selftest(type); |
| 1262 | if (ret < 0) |
| 1263 | goto out; |
| 1264 | |
| 1265 | type->next = trace_types; |
| 1266 | trace_types = type; |
| 1267 | add_tracer_options(&global_trace, type); |
| 1268 | |
| 1269 | out: |
| 1270 | tracing_selftest_running = false; |
| 1271 | mutex_unlock(&trace_types_lock); |
| 1272 | |
| 1273 | if (ret || !default_bootup_tracer) |
| 1274 | goto out_unlock; |
| 1275 | |
| 1276 | if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE)) |
| 1277 | goto out_unlock; |
| 1278 | |
| 1279 | printk(KERN_INFO "Starting tracer '%s'\n", type->name); |
| 1280 | /* Do we want this tracer to start on bootup? */ |
| 1281 | tracing_set_tracer(&global_trace, type->name); |
| 1282 | default_bootup_tracer = NULL; |
| 1283 | /* disable other selftests, since this will break it. */ |
| 1284 | tracing_selftest_disabled = true; |
| 1285 | #ifdef CONFIG_FTRACE_STARTUP_TEST |
| 1286 | printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n", |
| 1287 | type->name); |
| 1288 | #endif |
| 1289 | |
| 1290 | out_unlock: |
| 1291 | return ret; |
| 1292 | } |
| 1293 | |
| 1294 | void tracing_reset(struct trace_buffer *buf, int cpu) |
| 1295 | { |
| 1296 | struct ring_buffer *buffer = buf->buffer; |
| 1297 | |
| 1298 | if (!buffer) |
| 1299 | return; |
| 1300 | |
| 1301 | ring_buffer_record_disable(buffer); |
| 1302 | |
| 1303 | /* Make sure all commits have finished */ |
| 1304 | synchronize_sched(); |
| 1305 | ring_buffer_reset_cpu(buffer, cpu); |
| 1306 | |
| 1307 | ring_buffer_record_enable(buffer); |
| 1308 | } |
| 1309 | |
| 1310 | void tracing_reset_online_cpus(struct trace_buffer *buf) |
| 1311 | { |
| 1312 | struct ring_buffer *buffer = buf->buffer; |
| 1313 | int cpu; |
| 1314 | |
| 1315 | if (!buffer) |
| 1316 | return; |
| 1317 | |
| 1318 | ring_buffer_record_disable(buffer); |
| 1319 | |
| 1320 | /* Make sure all commits have finished */ |
| 1321 | synchronize_sched(); |
| 1322 | |
| 1323 | buf->time_start = buffer_ftrace_now(buf, buf->cpu); |
| 1324 | |
| 1325 | for_each_online_cpu(cpu) |
| 1326 | ring_buffer_reset_cpu(buffer, cpu); |
| 1327 | |
| 1328 | ring_buffer_record_enable(buffer); |
| 1329 | } |
| 1330 | |
| 1331 | /* Must have trace_types_lock held */ |
| 1332 | void tracing_reset_all_online_cpus(void) |
| 1333 | { |
| 1334 | struct trace_array *tr; |
| 1335 | |
| 1336 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { |
| 1337 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 1338 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1339 | tracing_reset_online_cpus(&tr->max_buffer); |
| 1340 | #endif |
| 1341 | } |
| 1342 | } |
| 1343 | |
| 1344 | #define SAVED_CMDLINES_DEFAULT 128 |
| 1345 | #define NO_CMDLINE_MAP UINT_MAX |
| 1346 | static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED; |
| 1347 | struct saved_cmdlines_buffer { |
| 1348 | unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1]; |
| 1349 | unsigned *map_cmdline_to_pid; |
| 1350 | unsigned cmdline_num; |
| 1351 | int cmdline_idx; |
| 1352 | char *saved_cmdlines; |
| 1353 | }; |
| 1354 | static struct saved_cmdlines_buffer *savedcmd; |
| 1355 | |
| 1356 | /* temporary disable recording */ |
| 1357 | static atomic_t trace_record_cmdline_disabled __read_mostly; |
| 1358 | |
| 1359 | static inline char *get_saved_cmdlines(int idx) |
| 1360 | { |
| 1361 | return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN]; |
| 1362 | } |
| 1363 | |
| 1364 | static inline void set_cmdline(int idx, const char *cmdline) |
| 1365 | { |
| 1366 | memcpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN); |
| 1367 | } |
| 1368 | |
| 1369 | static int allocate_cmdlines_buffer(unsigned int val, |
| 1370 | struct saved_cmdlines_buffer *s) |
| 1371 | { |
| 1372 | s->map_cmdline_to_pid = kmalloc(val * sizeof(*s->map_cmdline_to_pid), |
| 1373 | GFP_KERNEL); |
| 1374 | if (!s->map_cmdline_to_pid) |
| 1375 | return -ENOMEM; |
| 1376 | |
| 1377 | s->saved_cmdlines = kmalloc(val * TASK_COMM_LEN, GFP_KERNEL); |
| 1378 | if (!s->saved_cmdlines) { |
| 1379 | kfree(s->map_cmdline_to_pid); |
| 1380 | return -ENOMEM; |
| 1381 | } |
| 1382 | |
| 1383 | s->cmdline_idx = 0; |
| 1384 | s->cmdline_num = val; |
| 1385 | memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP, |
| 1386 | sizeof(s->map_pid_to_cmdline)); |
| 1387 | memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP, |
| 1388 | val * sizeof(*s->map_cmdline_to_pid)); |
| 1389 | |
| 1390 | return 0; |
| 1391 | } |
| 1392 | |
| 1393 | static int trace_create_savedcmd(void) |
| 1394 | { |
| 1395 | int ret; |
| 1396 | |
| 1397 | savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL); |
| 1398 | if (!savedcmd) |
| 1399 | return -ENOMEM; |
| 1400 | |
| 1401 | ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd); |
| 1402 | if (ret < 0) { |
| 1403 | kfree(savedcmd); |
| 1404 | savedcmd = NULL; |
| 1405 | return -ENOMEM; |
| 1406 | } |
| 1407 | |
| 1408 | return 0; |
| 1409 | } |
| 1410 | |
| 1411 | int is_tracing_stopped(void) |
| 1412 | { |
| 1413 | return global_trace.stop_count; |
| 1414 | } |
| 1415 | |
| 1416 | /** |
| 1417 | * tracing_start - quick start of the tracer |
| 1418 | * |
| 1419 | * If tracing is enabled but was stopped by tracing_stop, |
| 1420 | * this will start the tracer back up. |
| 1421 | */ |
| 1422 | void tracing_start(void) |
| 1423 | { |
| 1424 | struct ring_buffer *buffer; |
| 1425 | unsigned long flags; |
| 1426 | |
| 1427 | if (tracing_disabled) |
| 1428 | return; |
| 1429 | |
| 1430 | raw_spin_lock_irqsave(&global_trace.start_lock, flags); |
| 1431 | if (--global_trace.stop_count) { |
| 1432 | if (global_trace.stop_count < 0) { |
| 1433 | /* Someone screwed up their debugging */ |
| 1434 | WARN_ON_ONCE(1); |
| 1435 | global_trace.stop_count = 0; |
| 1436 | } |
| 1437 | goto out; |
| 1438 | } |
| 1439 | |
| 1440 | /* Prevent the buffers from switching */ |
| 1441 | arch_spin_lock(&global_trace.max_lock); |
| 1442 | |
| 1443 | buffer = global_trace.trace_buffer.buffer; |
| 1444 | if (buffer) |
| 1445 | ring_buffer_record_enable(buffer); |
| 1446 | |
| 1447 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1448 | buffer = global_trace.max_buffer.buffer; |
| 1449 | if (buffer) |
| 1450 | ring_buffer_record_enable(buffer); |
| 1451 | #endif |
| 1452 | |
| 1453 | arch_spin_unlock(&global_trace.max_lock); |
| 1454 | |
| 1455 | out: |
| 1456 | raw_spin_unlock_irqrestore(&global_trace.start_lock, flags); |
| 1457 | } |
| 1458 | |
| 1459 | static void tracing_start_tr(struct trace_array *tr) |
| 1460 | { |
| 1461 | struct ring_buffer *buffer; |
| 1462 | unsigned long flags; |
| 1463 | |
| 1464 | if (tracing_disabled) |
| 1465 | return; |
| 1466 | |
| 1467 | /* If global, we need to also start the max tracer */ |
| 1468 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) |
| 1469 | return tracing_start(); |
| 1470 | |
| 1471 | raw_spin_lock_irqsave(&tr->start_lock, flags); |
| 1472 | |
| 1473 | if (--tr->stop_count) { |
| 1474 | if (tr->stop_count < 0) { |
| 1475 | /* Someone screwed up their debugging */ |
| 1476 | WARN_ON_ONCE(1); |
| 1477 | tr->stop_count = 0; |
| 1478 | } |
| 1479 | goto out; |
| 1480 | } |
| 1481 | |
| 1482 | buffer = tr->trace_buffer.buffer; |
| 1483 | if (buffer) |
| 1484 | ring_buffer_record_enable(buffer); |
| 1485 | |
| 1486 | out: |
| 1487 | raw_spin_unlock_irqrestore(&tr->start_lock, flags); |
| 1488 | } |
| 1489 | |
| 1490 | /** |
| 1491 | * tracing_stop - quick stop of the tracer |
| 1492 | * |
| 1493 | * Light weight way to stop tracing. Use in conjunction with |
| 1494 | * tracing_start. |
| 1495 | */ |
| 1496 | void tracing_stop(void) |
| 1497 | { |
| 1498 | struct ring_buffer *buffer; |
| 1499 | unsigned long flags; |
| 1500 | |
| 1501 | raw_spin_lock_irqsave(&global_trace.start_lock, flags); |
| 1502 | if (global_trace.stop_count++) |
| 1503 | goto out; |
| 1504 | |
| 1505 | /* Prevent the buffers from switching */ |
| 1506 | arch_spin_lock(&global_trace.max_lock); |
| 1507 | |
| 1508 | buffer = global_trace.trace_buffer.buffer; |
| 1509 | if (buffer) |
| 1510 | ring_buffer_record_disable(buffer); |
| 1511 | |
| 1512 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 1513 | buffer = global_trace.max_buffer.buffer; |
| 1514 | if (buffer) |
| 1515 | ring_buffer_record_disable(buffer); |
| 1516 | #endif |
| 1517 | |
| 1518 | arch_spin_unlock(&global_trace.max_lock); |
| 1519 | |
| 1520 | out: |
| 1521 | raw_spin_unlock_irqrestore(&global_trace.start_lock, flags); |
| 1522 | } |
| 1523 | |
| 1524 | static void tracing_stop_tr(struct trace_array *tr) |
| 1525 | { |
| 1526 | struct ring_buffer *buffer; |
| 1527 | unsigned long flags; |
| 1528 | |
| 1529 | /* If global, we need to also stop the max tracer */ |
| 1530 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) |
| 1531 | return tracing_stop(); |
| 1532 | |
| 1533 | raw_spin_lock_irqsave(&tr->start_lock, flags); |
| 1534 | if (tr->stop_count++) |
| 1535 | goto out; |
| 1536 | |
| 1537 | buffer = tr->trace_buffer.buffer; |
| 1538 | if (buffer) |
| 1539 | ring_buffer_record_disable(buffer); |
| 1540 | |
| 1541 | out: |
| 1542 | raw_spin_unlock_irqrestore(&tr->start_lock, flags); |
| 1543 | } |
| 1544 | |
| 1545 | void trace_stop_cmdline_recording(void); |
| 1546 | |
| 1547 | static int trace_save_cmdline(struct task_struct *tsk) |
| 1548 | { |
| 1549 | unsigned pid, idx; |
| 1550 | |
| 1551 | if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT)) |
| 1552 | return 0; |
| 1553 | |
| 1554 | /* |
| 1555 | * It's not the end of the world if we don't get |
| 1556 | * the lock, but we also don't want to spin |
| 1557 | * nor do we want to disable interrupts, |
| 1558 | * so if we miss here, then better luck next time. |
| 1559 | */ |
| 1560 | if (!arch_spin_trylock(&trace_cmdline_lock)) |
| 1561 | return 0; |
| 1562 | |
| 1563 | idx = savedcmd->map_pid_to_cmdline[tsk->pid]; |
| 1564 | if (idx == NO_CMDLINE_MAP) { |
| 1565 | idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num; |
| 1566 | |
| 1567 | /* |
| 1568 | * Check whether the cmdline buffer at idx has a pid |
| 1569 | * mapped. We are going to overwrite that entry so we |
| 1570 | * need to clear the map_pid_to_cmdline. Otherwise we |
| 1571 | * would read the new comm for the old pid. |
| 1572 | */ |
| 1573 | pid = savedcmd->map_cmdline_to_pid[idx]; |
| 1574 | if (pid != NO_CMDLINE_MAP) |
| 1575 | savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP; |
| 1576 | |
| 1577 | savedcmd->map_cmdline_to_pid[idx] = tsk->pid; |
| 1578 | savedcmd->map_pid_to_cmdline[tsk->pid] = idx; |
| 1579 | |
| 1580 | savedcmd->cmdline_idx = idx; |
| 1581 | } |
| 1582 | |
| 1583 | set_cmdline(idx, tsk->comm); |
| 1584 | |
| 1585 | arch_spin_unlock(&trace_cmdline_lock); |
| 1586 | |
| 1587 | return 1; |
| 1588 | } |
| 1589 | |
| 1590 | static void __trace_find_cmdline(int pid, char comm[]) |
| 1591 | { |
| 1592 | unsigned map; |
| 1593 | |
| 1594 | if (!pid) { |
| 1595 | strcpy(comm, "<idle>"); |
| 1596 | return; |
| 1597 | } |
| 1598 | |
| 1599 | if (WARN_ON_ONCE(pid < 0)) { |
| 1600 | strcpy(comm, "<XXX>"); |
| 1601 | return; |
| 1602 | } |
| 1603 | |
| 1604 | if (pid > PID_MAX_DEFAULT) { |
| 1605 | strcpy(comm, "<...>"); |
| 1606 | return; |
| 1607 | } |
| 1608 | |
| 1609 | map = savedcmd->map_pid_to_cmdline[pid]; |
| 1610 | if (map != NO_CMDLINE_MAP) |
| 1611 | strcpy(comm, get_saved_cmdlines(map)); |
| 1612 | else |
| 1613 | strcpy(comm, "<...>"); |
| 1614 | } |
| 1615 | |
| 1616 | void trace_find_cmdline(int pid, char comm[]) |
| 1617 | { |
| 1618 | preempt_disable(); |
| 1619 | arch_spin_lock(&trace_cmdline_lock); |
| 1620 | |
| 1621 | __trace_find_cmdline(pid, comm); |
| 1622 | |
| 1623 | arch_spin_unlock(&trace_cmdline_lock); |
| 1624 | preempt_enable(); |
| 1625 | } |
| 1626 | |
| 1627 | void tracing_record_cmdline(struct task_struct *tsk) |
| 1628 | { |
| 1629 | if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on()) |
| 1630 | return; |
| 1631 | |
| 1632 | if (!__this_cpu_read(trace_cmdline_save)) |
| 1633 | return; |
| 1634 | |
| 1635 | if (trace_save_cmdline(tsk)) |
| 1636 | __this_cpu_write(trace_cmdline_save, false); |
| 1637 | } |
| 1638 | |
| 1639 | void |
| 1640 | tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, |
| 1641 | int pc) |
| 1642 | { |
| 1643 | struct task_struct *tsk = current; |
| 1644 | |
| 1645 | entry->preempt_count = pc & 0xff; |
| 1646 | entry->pid = (tsk) ? tsk->pid : 0; |
| 1647 | entry->flags = |
| 1648 | #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT |
| 1649 | (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | |
| 1650 | #else |
| 1651 | TRACE_FLAG_IRQS_NOSUPPORT | |
| 1652 | #endif |
| 1653 | ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) | |
| 1654 | ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) | |
| 1655 | (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) | |
| 1656 | (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0); |
| 1657 | } |
| 1658 | EXPORT_SYMBOL_GPL(tracing_generic_entry_update); |
| 1659 | |
| 1660 | struct ring_buffer_event * |
| 1661 | trace_buffer_lock_reserve(struct ring_buffer *buffer, |
| 1662 | int type, |
| 1663 | unsigned long len, |
| 1664 | unsigned long flags, int pc) |
| 1665 | { |
| 1666 | struct ring_buffer_event *event; |
| 1667 | |
| 1668 | event = ring_buffer_lock_reserve(buffer, len); |
| 1669 | if (event != NULL) { |
| 1670 | struct trace_entry *ent = ring_buffer_event_data(event); |
| 1671 | |
| 1672 | tracing_generic_entry_update(ent, flags, pc); |
| 1673 | ent->type = type; |
| 1674 | } |
| 1675 | |
| 1676 | return event; |
| 1677 | } |
| 1678 | |
| 1679 | void |
| 1680 | __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event) |
| 1681 | { |
| 1682 | __this_cpu_write(trace_cmdline_save, true); |
| 1683 | ring_buffer_unlock_commit(buffer, event); |
| 1684 | } |
| 1685 | |
| 1686 | void trace_buffer_unlock_commit(struct trace_array *tr, |
| 1687 | struct ring_buffer *buffer, |
| 1688 | struct ring_buffer_event *event, |
| 1689 | unsigned long flags, int pc) |
| 1690 | { |
| 1691 | __buffer_unlock_commit(buffer, event); |
| 1692 | |
| 1693 | ftrace_trace_stack(buffer, flags, 6, pc, NULL); |
| 1694 | ftrace_trace_userstack(buffer, flags, pc); |
| 1695 | } |
| 1696 | EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit); |
| 1697 | |
| 1698 | static struct ring_buffer *temp_buffer; |
| 1699 | |
| 1700 | struct ring_buffer_event * |
| 1701 | trace_event_buffer_lock_reserve(struct ring_buffer **current_rb, |
| 1702 | struct trace_event_file *trace_file, |
| 1703 | int type, unsigned long len, |
| 1704 | unsigned long flags, int pc) |
| 1705 | { |
| 1706 | struct ring_buffer_event *entry; |
| 1707 | |
| 1708 | *current_rb = trace_file->tr->trace_buffer.buffer; |
| 1709 | entry = trace_buffer_lock_reserve(*current_rb, |
| 1710 | type, len, flags, pc); |
| 1711 | /* |
| 1712 | * If tracing is off, but we have triggers enabled |
| 1713 | * we still need to look at the event data. Use the temp_buffer |
| 1714 | * to store the trace event for the tigger to use. It's recusive |
| 1715 | * safe and will not be recorded anywhere. |
| 1716 | */ |
| 1717 | if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) { |
| 1718 | *current_rb = temp_buffer; |
| 1719 | entry = trace_buffer_lock_reserve(*current_rb, |
| 1720 | type, len, flags, pc); |
| 1721 | } |
| 1722 | return entry; |
| 1723 | } |
| 1724 | EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve); |
| 1725 | |
| 1726 | struct ring_buffer_event * |
| 1727 | trace_current_buffer_lock_reserve(struct ring_buffer **current_rb, |
| 1728 | int type, unsigned long len, |
| 1729 | unsigned long flags, int pc) |
| 1730 | { |
| 1731 | *current_rb = global_trace.trace_buffer.buffer; |
| 1732 | return trace_buffer_lock_reserve(*current_rb, |
| 1733 | type, len, flags, pc); |
| 1734 | } |
| 1735 | EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve); |
| 1736 | |
| 1737 | void trace_buffer_unlock_commit_regs(struct trace_array *tr, |
| 1738 | struct ring_buffer *buffer, |
| 1739 | struct ring_buffer_event *event, |
| 1740 | unsigned long flags, int pc, |
| 1741 | struct pt_regs *regs) |
| 1742 | { |
| 1743 | __buffer_unlock_commit(buffer, event); |
| 1744 | |
| 1745 | ftrace_trace_stack(buffer, flags, 6, pc, regs); |
| 1746 | ftrace_trace_userstack(buffer, flags, pc); |
| 1747 | } |
| 1748 | EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs); |
| 1749 | |
| 1750 | void trace_current_buffer_discard_commit(struct ring_buffer *buffer, |
| 1751 | struct ring_buffer_event *event) |
| 1752 | { |
| 1753 | ring_buffer_discard_commit(buffer, event); |
| 1754 | } |
| 1755 | EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit); |
| 1756 | |
| 1757 | void |
| 1758 | trace_function(struct trace_array *tr, |
| 1759 | unsigned long ip, unsigned long parent_ip, unsigned long flags, |
| 1760 | int pc) |
| 1761 | { |
| 1762 | struct trace_event_call *call = &event_function; |
| 1763 | struct ring_buffer *buffer = tr->trace_buffer.buffer; |
| 1764 | struct ring_buffer_event *event; |
| 1765 | struct ftrace_entry *entry; |
| 1766 | |
| 1767 | /* If we are reading the ring buffer, don't trace */ |
| 1768 | if (unlikely(__this_cpu_read(ftrace_cpu_disabled))) |
| 1769 | return; |
| 1770 | |
| 1771 | event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry), |
| 1772 | flags, pc); |
| 1773 | if (!event) |
| 1774 | return; |
| 1775 | entry = ring_buffer_event_data(event); |
| 1776 | entry->ip = ip; |
| 1777 | entry->parent_ip = parent_ip; |
| 1778 | |
| 1779 | if (!call_filter_check_discard(call, entry, buffer, event)) |
| 1780 | __buffer_unlock_commit(buffer, event); |
| 1781 | } |
| 1782 | |
| 1783 | #ifdef CONFIG_STACKTRACE |
| 1784 | |
| 1785 | #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long)) |
| 1786 | struct ftrace_stack { |
| 1787 | unsigned long calls[FTRACE_STACK_MAX_ENTRIES]; |
| 1788 | }; |
| 1789 | |
| 1790 | static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack); |
| 1791 | static DEFINE_PER_CPU(int, ftrace_stack_reserve); |
| 1792 | |
| 1793 | static void __ftrace_trace_stack(struct ring_buffer *buffer, |
| 1794 | unsigned long flags, |
| 1795 | int skip, int pc, struct pt_regs *regs) |
| 1796 | { |
| 1797 | struct trace_event_call *call = &event_kernel_stack; |
| 1798 | struct ring_buffer_event *event; |
| 1799 | struct stack_entry *entry; |
| 1800 | struct stack_trace trace; |
| 1801 | int use_stack; |
| 1802 | int size = FTRACE_STACK_ENTRIES; |
| 1803 | |
| 1804 | trace.nr_entries = 0; |
| 1805 | trace.skip = skip; |
| 1806 | |
| 1807 | /* |
| 1808 | * Since events can happen in NMIs there's no safe way to |
| 1809 | * use the per cpu ftrace_stacks. We reserve it and if an interrupt |
| 1810 | * or NMI comes in, it will just have to use the default |
| 1811 | * FTRACE_STACK_SIZE. |
| 1812 | */ |
| 1813 | preempt_disable_notrace(); |
| 1814 | |
| 1815 | use_stack = __this_cpu_inc_return(ftrace_stack_reserve); |
| 1816 | /* |
| 1817 | * We don't need any atomic variables, just a barrier. |
| 1818 | * If an interrupt comes in, we don't care, because it would |
| 1819 | * have exited and put the counter back to what we want. |
| 1820 | * We just need a barrier to keep gcc from moving things |
| 1821 | * around. |
| 1822 | */ |
| 1823 | barrier(); |
| 1824 | if (use_stack == 1) { |
| 1825 | trace.entries = this_cpu_ptr(ftrace_stack.calls); |
| 1826 | trace.max_entries = FTRACE_STACK_MAX_ENTRIES; |
| 1827 | |
| 1828 | if (regs) |
| 1829 | save_stack_trace_regs(regs, &trace); |
| 1830 | else |
| 1831 | save_stack_trace(&trace); |
| 1832 | |
| 1833 | if (trace.nr_entries > size) |
| 1834 | size = trace.nr_entries; |
| 1835 | } else |
| 1836 | /* From now on, use_stack is a boolean */ |
| 1837 | use_stack = 0; |
| 1838 | |
| 1839 | size *= sizeof(unsigned long); |
| 1840 | |
| 1841 | event = trace_buffer_lock_reserve(buffer, TRACE_STACK, |
| 1842 | sizeof(*entry) + size, flags, pc); |
| 1843 | if (!event) |
| 1844 | goto out; |
| 1845 | entry = ring_buffer_event_data(event); |
| 1846 | |
| 1847 | memset(&entry->caller, 0, size); |
| 1848 | |
| 1849 | if (use_stack) |
| 1850 | memcpy(&entry->caller, trace.entries, |
| 1851 | trace.nr_entries * sizeof(unsigned long)); |
| 1852 | else { |
| 1853 | trace.max_entries = FTRACE_STACK_ENTRIES; |
| 1854 | trace.entries = entry->caller; |
| 1855 | if (regs) |
| 1856 | save_stack_trace_regs(regs, &trace); |
| 1857 | else |
| 1858 | save_stack_trace(&trace); |
| 1859 | } |
| 1860 | |
| 1861 | entry->size = trace.nr_entries; |
| 1862 | |
| 1863 | if (!call_filter_check_discard(call, entry, buffer, event)) |
| 1864 | __buffer_unlock_commit(buffer, event); |
| 1865 | |
| 1866 | out: |
| 1867 | /* Again, don't let gcc optimize things here */ |
| 1868 | barrier(); |
| 1869 | __this_cpu_dec(ftrace_stack_reserve); |
| 1870 | preempt_enable_notrace(); |
| 1871 | |
| 1872 | } |
| 1873 | |
| 1874 | static inline void ftrace_trace_stack(struct ring_buffer *buffer, |
| 1875 | unsigned long flags, |
| 1876 | int skip, int pc, struct pt_regs *regs) |
| 1877 | { |
| 1878 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) |
| 1879 | return; |
| 1880 | |
| 1881 | __ftrace_trace_stack(buffer, flags, skip, pc, regs); |
| 1882 | } |
| 1883 | |
| 1884 | void __trace_stack(struct trace_array *tr, unsigned long flags, int skip, |
| 1885 | int pc) |
| 1886 | { |
| 1887 | __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL); |
| 1888 | } |
| 1889 | |
| 1890 | /** |
| 1891 | * trace_dump_stack - record a stack back trace in the trace buffer |
| 1892 | * @skip: Number of functions to skip (helper handlers) |
| 1893 | */ |
| 1894 | void trace_dump_stack(int skip) |
| 1895 | { |
| 1896 | unsigned long flags; |
| 1897 | |
| 1898 | if (tracing_disabled || tracing_selftest_running) |
| 1899 | return; |
| 1900 | |
| 1901 | local_save_flags(flags); |
| 1902 | |
| 1903 | /* |
| 1904 | * Skip 3 more, seems to get us at the caller of |
| 1905 | * this function. |
| 1906 | */ |
| 1907 | skip += 3; |
| 1908 | __ftrace_trace_stack(global_trace.trace_buffer.buffer, |
| 1909 | flags, skip, preempt_count(), NULL); |
| 1910 | } |
| 1911 | |
| 1912 | static DEFINE_PER_CPU(int, user_stack_count); |
| 1913 | |
| 1914 | void |
| 1915 | ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc) |
| 1916 | { |
| 1917 | struct trace_event_call *call = &event_user_stack; |
| 1918 | struct ring_buffer_event *event; |
| 1919 | struct userstack_entry *entry; |
| 1920 | struct stack_trace trace; |
| 1921 | |
| 1922 | if (!(trace_flags & TRACE_ITER_USERSTACKTRACE)) |
| 1923 | return; |
| 1924 | |
| 1925 | /* |
| 1926 | * NMIs can not handle page faults, even with fix ups. |
| 1927 | * The save user stack can (and often does) fault. |
| 1928 | */ |
| 1929 | if (unlikely(in_nmi())) |
| 1930 | return; |
| 1931 | |
| 1932 | /* |
| 1933 | * prevent recursion, since the user stack tracing may |
| 1934 | * trigger other kernel events. |
| 1935 | */ |
| 1936 | preempt_disable(); |
| 1937 | if (__this_cpu_read(user_stack_count)) |
| 1938 | goto out; |
| 1939 | |
| 1940 | __this_cpu_inc(user_stack_count); |
| 1941 | |
| 1942 | event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK, |
| 1943 | sizeof(*entry), flags, pc); |
| 1944 | if (!event) |
| 1945 | goto out_drop_count; |
| 1946 | entry = ring_buffer_event_data(event); |
| 1947 | |
| 1948 | entry->tgid = current->tgid; |
| 1949 | memset(&entry->caller, 0, sizeof(entry->caller)); |
| 1950 | |
| 1951 | trace.nr_entries = 0; |
| 1952 | trace.max_entries = FTRACE_STACK_ENTRIES; |
| 1953 | trace.skip = 0; |
| 1954 | trace.entries = entry->caller; |
| 1955 | |
| 1956 | save_stack_trace_user(&trace); |
| 1957 | if (!call_filter_check_discard(call, entry, buffer, event)) |
| 1958 | __buffer_unlock_commit(buffer, event); |
| 1959 | |
| 1960 | out_drop_count: |
| 1961 | __this_cpu_dec(user_stack_count); |
| 1962 | out: |
| 1963 | preempt_enable(); |
| 1964 | } |
| 1965 | |
| 1966 | #ifdef UNUSED |
| 1967 | static void __trace_userstack(struct trace_array *tr, unsigned long flags) |
| 1968 | { |
| 1969 | ftrace_trace_userstack(tr, flags, preempt_count()); |
| 1970 | } |
| 1971 | #endif /* UNUSED */ |
| 1972 | |
| 1973 | #endif /* CONFIG_STACKTRACE */ |
| 1974 | |
| 1975 | /* created for use with alloc_percpu */ |
| 1976 | struct trace_buffer_struct { |
| 1977 | char buffer[TRACE_BUF_SIZE]; |
| 1978 | }; |
| 1979 | |
| 1980 | static struct trace_buffer_struct *trace_percpu_buffer; |
| 1981 | static struct trace_buffer_struct *trace_percpu_sirq_buffer; |
| 1982 | static struct trace_buffer_struct *trace_percpu_irq_buffer; |
| 1983 | static struct trace_buffer_struct *trace_percpu_nmi_buffer; |
| 1984 | |
| 1985 | /* |
| 1986 | * The buffer used is dependent on the context. There is a per cpu |
| 1987 | * buffer for normal context, softirq contex, hard irq context and |
| 1988 | * for NMI context. Thise allows for lockless recording. |
| 1989 | * |
| 1990 | * Note, if the buffers failed to be allocated, then this returns NULL |
| 1991 | */ |
| 1992 | static char *get_trace_buf(void) |
| 1993 | { |
| 1994 | struct trace_buffer_struct *percpu_buffer; |
| 1995 | |
| 1996 | /* |
| 1997 | * If we have allocated per cpu buffers, then we do not |
| 1998 | * need to do any locking. |
| 1999 | */ |
| 2000 | if (in_nmi()) |
| 2001 | percpu_buffer = trace_percpu_nmi_buffer; |
| 2002 | else if (in_irq()) |
| 2003 | percpu_buffer = trace_percpu_irq_buffer; |
| 2004 | else if (in_softirq()) |
| 2005 | percpu_buffer = trace_percpu_sirq_buffer; |
| 2006 | else |
| 2007 | percpu_buffer = trace_percpu_buffer; |
| 2008 | |
| 2009 | if (!percpu_buffer) |
| 2010 | return NULL; |
| 2011 | |
| 2012 | return this_cpu_ptr(&percpu_buffer->buffer[0]); |
| 2013 | } |
| 2014 | |
| 2015 | static int alloc_percpu_trace_buffer(void) |
| 2016 | { |
| 2017 | struct trace_buffer_struct *buffers; |
| 2018 | struct trace_buffer_struct *sirq_buffers; |
| 2019 | struct trace_buffer_struct *irq_buffers; |
| 2020 | struct trace_buffer_struct *nmi_buffers; |
| 2021 | |
| 2022 | buffers = alloc_percpu(struct trace_buffer_struct); |
| 2023 | if (!buffers) |
| 2024 | goto err_warn; |
| 2025 | |
| 2026 | sirq_buffers = alloc_percpu(struct trace_buffer_struct); |
| 2027 | if (!sirq_buffers) |
| 2028 | goto err_sirq; |
| 2029 | |
| 2030 | irq_buffers = alloc_percpu(struct trace_buffer_struct); |
| 2031 | if (!irq_buffers) |
| 2032 | goto err_irq; |
| 2033 | |
| 2034 | nmi_buffers = alloc_percpu(struct trace_buffer_struct); |
| 2035 | if (!nmi_buffers) |
| 2036 | goto err_nmi; |
| 2037 | |
| 2038 | trace_percpu_buffer = buffers; |
| 2039 | trace_percpu_sirq_buffer = sirq_buffers; |
| 2040 | trace_percpu_irq_buffer = irq_buffers; |
| 2041 | trace_percpu_nmi_buffer = nmi_buffers; |
| 2042 | |
| 2043 | return 0; |
| 2044 | |
| 2045 | err_nmi: |
| 2046 | free_percpu(irq_buffers); |
| 2047 | err_irq: |
| 2048 | free_percpu(sirq_buffers); |
| 2049 | err_sirq: |
| 2050 | free_percpu(buffers); |
| 2051 | err_warn: |
| 2052 | WARN(1, "Could not allocate percpu trace_printk buffer"); |
| 2053 | return -ENOMEM; |
| 2054 | } |
| 2055 | |
| 2056 | static int buffers_allocated; |
| 2057 | |
| 2058 | void trace_printk_init_buffers(void) |
| 2059 | { |
| 2060 | if (buffers_allocated) |
| 2061 | return; |
| 2062 | |
| 2063 | if (alloc_percpu_trace_buffer()) |
| 2064 | return; |
| 2065 | |
| 2066 | /* trace_printk() is for debug use only. Don't use it in production. */ |
| 2067 | |
| 2068 | pr_warning("\n"); |
| 2069 | pr_warning("**********************************************************\n"); |
| 2070 | pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); |
| 2071 | pr_warning("** **\n"); |
| 2072 | pr_warning("** trace_printk() being used. Allocating extra memory. **\n"); |
| 2073 | pr_warning("** **\n"); |
| 2074 | pr_warning("** This means that this is a DEBUG kernel and it is **\n"); |
| 2075 | pr_warning("** unsafe for production use. **\n"); |
| 2076 | pr_warning("** **\n"); |
| 2077 | pr_warning("** If you see this message and you are not debugging **\n"); |
| 2078 | pr_warning("** the kernel, report this immediately to your vendor! **\n"); |
| 2079 | pr_warning("** **\n"); |
| 2080 | pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); |
| 2081 | pr_warning("**********************************************************\n"); |
| 2082 | |
| 2083 | /* Expand the buffers to set size */ |
| 2084 | tracing_update_buffers(); |
| 2085 | |
| 2086 | buffers_allocated = 1; |
| 2087 | |
| 2088 | /* |
| 2089 | * trace_printk_init_buffers() can be called by modules. |
| 2090 | * If that happens, then we need to start cmdline recording |
| 2091 | * directly here. If the global_trace.buffer is already |
| 2092 | * allocated here, then this was called by module code. |
| 2093 | */ |
| 2094 | if (global_trace.trace_buffer.buffer) |
| 2095 | tracing_start_cmdline_record(); |
| 2096 | } |
| 2097 | |
| 2098 | void trace_printk_start_comm(void) |
| 2099 | { |
| 2100 | /* Start tracing comms if trace printk is set */ |
| 2101 | if (!buffers_allocated) |
| 2102 | return; |
| 2103 | tracing_start_cmdline_record(); |
| 2104 | } |
| 2105 | |
| 2106 | static void trace_printk_start_stop_comm(int enabled) |
| 2107 | { |
| 2108 | if (!buffers_allocated) |
| 2109 | return; |
| 2110 | |
| 2111 | if (enabled) |
| 2112 | tracing_start_cmdline_record(); |
| 2113 | else |
| 2114 | tracing_stop_cmdline_record(); |
| 2115 | } |
| 2116 | |
| 2117 | /** |
| 2118 | * trace_vbprintk - write binary msg to tracing buffer |
| 2119 | * |
| 2120 | */ |
| 2121 | int trace_vbprintk(unsigned long ip, const char *fmt, va_list args) |
| 2122 | { |
| 2123 | struct trace_event_call *call = &event_bprint; |
| 2124 | struct ring_buffer_event *event; |
| 2125 | struct ring_buffer *buffer; |
| 2126 | struct trace_array *tr = &global_trace; |
| 2127 | struct bprint_entry *entry; |
| 2128 | unsigned long flags; |
| 2129 | char *tbuffer; |
| 2130 | int len = 0, size, pc; |
| 2131 | |
| 2132 | if (unlikely(tracing_selftest_running || tracing_disabled)) |
| 2133 | return 0; |
| 2134 | |
| 2135 | /* Don't pollute graph traces with trace_vprintk internals */ |
| 2136 | pause_graph_tracing(); |
| 2137 | |
| 2138 | pc = preempt_count(); |
| 2139 | preempt_disable_notrace(); |
| 2140 | |
| 2141 | tbuffer = get_trace_buf(); |
| 2142 | if (!tbuffer) { |
| 2143 | len = 0; |
| 2144 | goto out; |
| 2145 | } |
| 2146 | |
| 2147 | len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args); |
| 2148 | |
| 2149 | if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0) |
| 2150 | goto out; |
| 2151 | |
| 2152 | local_save_flags(flags); |
| 2153 | size = sizeof(*entry) + sizeof(u32) * len; |
| 2154 | buffer = tr->trace_buffer.buffer; |
| 2155 | event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size, |
| 2156 | flags, pc); |
| 2157 | if (!event) |
| 2158 | goto out; |
| 2159 | entry = ring_buffer_event_data(event); |
| 2160 | entry->ip = ip; |
| 2161 | entry->fmt = fmt; |
| 2162 | |
| 2163 | memcpy(entry->buf, tbuffer, sizeof(u32) * len); |
| 2164 | if (!call_filter_check_discard(call, entry, buffer, event)) { |
| 2165 | __buffer_unlock_commit(buffer, event); |
| 2166 | ftrace_trace_stack(buffer, flags, 6, pc, NULL); |
| 2167 | } |
| 2168 | |
| 2169 | out: |
| 2170 | preempt_enable_notrace(); |
| 2171 | unpause_graph_tracing(); |
| 2172 | |
| 2173 | return len; |
| 2174 | } |
| 2175 | EXPORT_SYMBOL_GPL(trace_vbprintk); |
| 2176 | |
| 2177 | static int |
| 2178 | __trace_array_vprintk(struct ring_buffer *buffer, |
| 2179 | unsigned long ip, const char *fmt, va_list args) |
| 2180 | { |
| 2181 | struct trace_event_call *call = &event_print; |
| 2182 | struct ring_buffer_event *event; |
| 2183 | int len = 0, size, pc; |
| 2184 | struct print_entry *entry; |
| 2185 | unsigned long flags; |
| 2186 | char *tbuffer; |
| 2187 | |
| 2188 | if (tracing_disabled || tracing_selftest_running) |
| 2189 | return 0; |
| 2190 | |
| 2191 | /* Don't pollute graph traces with trace_vprintk internals */ |
| 2192 | pause_graph_tracing(); |
| 2193 | |
| 2194 | pc = preempt_count(); |
| 2195 | preempt_disable_notrace(); |
| 2196 | |
| 2197 | |
| 2198 | tbuffer = get_trace_buf(); |
| 2199 | if (!tbuffer) { |
| 2200 | len = 0; |
| 2201 | goto out; |
| 2202 | } |
| 2203 | |
| 2204 | len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args); |
| 2205 | |
| 2206 | local_save_flags(flags); |
| 2207 | size = sizeof(*entry) + len + 1; |
| 2208 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, |
| 2209 | flags, pc); |
| 2210 | if (!event) |
| 2211 | goto out; |
| 2212 | entry = ring_buffer_event_data(event); |
| 2213 | entry->ip = ip; |
| 2214 | |
| 2215 | memcpy(&entry->buf, tbuffer, len + 1); |
| 2216 | if (!call_filter_check_discard(call, entry, buffer, event)) { |
| 2217 | __buffer_unlock_commit(buffer, event); |
| 2218 | ftrace_trace_stack(buffer, flags, 6, pc, NULL); |
| 2219 | } |
| 2220 | out: |
| 2221 | preempt_enable_notrace(); |
| 2222 | unpause_graph_tracing(); |
| 2223 | |
| 2224 | return len; |
| 2225 | } |
| 2226 | |
| 2227 | int trace_array_vprintk(struct trace_array *tr, |
| 2228 | unsigned long ip, const char *fmt, va_list args) |
| 2229 | { |
| 2230 | return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args); |
| 2231 | } |
| 2232 | |
| 2233 | int trace_array_printk(struct trace_array *tr, |
| 2234 | unsigned long ip, const char *fmt, ...) |
| 2235 | { |
| 2236 | int ret; |
| 2237 | va_list ap; |
| 2238 | |
| 2239 | if (!(trace_flags & TRACE_ITER_PRINTK)) |
| 2240 | return 0; |
| 2241 | |
| 2242 | va_start(ap, fmt); |
| 2243 | ret = trace_array_vprintk(tr, ip, fmt, ap); |
| 2244 | va_end(ap); |
| 2245 | return ret; |
| 2246 | } |
| 2247 | |
| 2248 | int trace_array_printk_buf(struct ring_buffer *buffer, |
| 2249 | unsigned long ip, const char *fmt, ...) |
| 2250 | { |
| 2251 | int ret; |
| 2252 | va_list ap; |
| 2253 | |
| 2254 | if (!(trace_flags & TRACE_ITER_PRINTK)) |
| 2255 | return 0; |
| 2256 | |
| 2257 | va_start(ap, fmt); |
| 2258 | ret = __trace_array_vprintk(buffer, ip, fmt, ap); |
| 2259 | va_end(ap); |
| 2260 | return ret; |
| 2261 | } |
| 2262 | |
| 2263 | int trace_vprintk(unsigned long ip, const char *fmt, va_list args) |
| 2264 | { |
| 2265 | return trace_array_vprintk(&global_trace, ip, fmt, args); |
| 2266 | } |
| 2267 | EXPORT_SYMBOL_GPL(trace_vprintk); |
| 2268 | |
| 2269 | static void trace_iterator_increment(struct trace_iterator *iter) |
| 2270 | { |
| 2271 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu); |
| 2272 | |
| 2273 | iter->idx++; |
| 2274 | if (buf_iter) |
| 2275 | ring_buffer_read(buf_iter, NULL); |
| 2276 | } |
| 2277 | |
| 2278 | static struct trace_entry * |
| 2279 | peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts, |
| 2280 | unsigned long *lost_events) |
| 2281 | { |
| 2282 | struct ring_buffer_event *event; |
| 2283 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu); |
| 2284 | |
| 2285 | if (buf_iter) |
| 2286 | event = ring_buffer_iter_peek(buf_iter, ts); |
| 2287 | else |
| 2288 | event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts, |
| 2289 | lost_events); |
| 2290 | |
| 2291 | if (event) { |
| 2292 | iter->ent_size = ring_buffer_event_length(event); |
| 2293 | return ring_buffer_event_data(event); |
| 2294 | } |
| 2295 | iter->ent_size = 0; |
| 2296 | return NULL; |
| 2297 | } |
| 2298 | |
| 2299 | static struct trace_entry * |
| 2300 | __find_next_entry(struct trace_iterator *iter, int *ent_cpu, |
| 2301 | unsigned long *missing_events, u64 *ent_ts) |
| 2302 | { |
| 2303 | struct ring_buffer *buffer = iter->trace_buffer->buffer; |
| 2304 | struct trace_entry *ent, *next = NULL; |
| 2305 | unsigned long lost_events = 0, next_lost = 0; |
| 2306 | int cpu_file = iter->cpu_file; |
| 2307 | u64 next_ts = 0, ts; |
| 2308 | int next_cpu = -1; |
| 2309 | int next_size = 0; |
| 2310 | int cpu; |
| 2311 | |
| 2312 | /* |
| 2313 | * If we are in a per_cpu trace file, don't bother by iterating over |
| 2314 | * all cpu and peek directly. |
| 2315 | */ |
| 2316 | if (cpu_file > RING_BUFFER_ALL_CPUS) { |
| 2317 | if (ring_buffer_empty_cpu(buffer, cpu_file)) |
| 2318 | return NULL; |
| 2319 | ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events); |
| 2320 | if (ent_cpu) |
| 2321 | *ent_cpu = cpu_file; |
| 2322 | |
| 2323 | return ent; |
| 2324 | } |
| 2325 | |
| 2326 | for_each_tracing_cpu(cpu) { |
| 2327 | |
| 2328 | if (ring_buffer_empty_cpu(buffer, cpu)) |
| 2329 | continue; |
| 2330 | |
| 2331 | ent = peek_next_entry(iter, cpu, &ts, &lost_events); |
| 2332 | |
| 2333 | /* |
| 2334 | * Pick the entry with the smallest timestamp: |
| 2335 | */ |
| 2336 | if (ent && (!next || ts < next_ts)) { |
| 2337 | next = ent; |
| 2338 | next_cpu = cpu; |
| 2339 | next_ts = ts; |
| 2340 | next_lost = lost_events; |
| 2341 | next_size = iter->ent_size; |
| 2342 | } |
| 2343 | } |
| 2344 | |
| 2345 | iter->ent_size = next_size; |
| 2346 | |
| 2347 | if (ent_cpu) |
| 2348 | *ent_cpu = next_cpu; |
| 2349 | |
| 2350 | if (ent_ts) |
| 2351 | *ent_ts = next_ts; |
| 2352 | |
| 2353 | if (missing_events) |
| 2354 | *missing_events = next_lost; |
| 2355 | |
| 2356 | return next; |
| 2357 | } |
| 2358 | |
| 2359 | /* Find the next real entry, without updating the iterator itself */ |
| 2360 | struct trace_entry *trace_find_next_entry(struct trace_iterator *iter, |
| 2361 | int *ent_cpu, u64 *ent_ts) |
| 2362 | { |
| 2363 | return __find_next_entry(iter, ent_cpu, NULL, ent_ts); |
| 2364 | } |
| 2365 | |
| 2366 | /* Find the next real entry, and increment the iterator to the next entry */ |
| 2367 | void *trace_find_next_entry_inc(struct trace_iterator *iter) |
| 2368 | { |
| 2369 | iter->ent = __find_next_entry(iter, &iter->cpu, |
| 2370 | &iter->lost_events, &iter->ts); |
| 2371 | |
| 2372 | if (iter->ent) |
| 2373 | trace_iterator_increment(iter); |
| 2374 | |
| 2375 | return iter->ent ? iter : NULL; |
| 2376 | } |
| 2377 | |
| 2378 | static void trace_consume(struct trace_iterator *iter) |
| 2379 | { |
| 2380 | ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts, |
| 2381 | &iter->lost_events); |
| 2382 | } |
| 2383 | |
| 2384 | static void *s_next(struct seq_file *m, void *v, loff_t *pos) |
| 2385 | { |
| 2386 | struct trace_iterator *iter = m->private; |
| 2387 | int i = (int)*pos; |
| 2388 | void *ent; |
| 2389 | |
| 2390 | WARN_ON_ONCE(iter->leftover); |
| 2391 | |
| 2392 | (*pos)++; |
| 2393 | |
| 2394 | /* can't go backwards */ |
| 2395 | if (iter->idx > i) |
| 2396 | return NULL; |
| 2397 | |
| 2398 | if (iter->idx < 0) |
| 2399 | ent = trace_find_next_entry_inc(iter); |
| 2400 | else |
| 2401 | ent = iter; |
| 2402 | |
| 2403 | while (ent && iter->idx < i) |
| 2404 | ent = trace_find_next_entry_inc(iter); |
| 2405 | |
| 2406 | iter->pos = *pos; |
| 2407 | |
| 2408 | return ent; |
| 2409 | } |
| 2410 | |
| 2411 | void tracing_iter_reset(struct trace_iterator *iter, int cpu) |
| 2412 | { |
| 2413 | struct ring_buffer_event *event; |
| 2414 | struct ring_buffer_iter *buf_iter; |
| 2415 | unsigned long entries = 0; |
| 2416 | u64 ts; |
| 2417 | |
| 2418 | per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0; |
| 2419 | |
| 2420 | buf_iter = trace_buffer_iter(iter, cpu); |
| 2421 | if (!buf_iter) |
| 2422 | return; |
| 2423 | |
| 2424 | ring_buffer_iter_reset(buf_iter); |
| 2425 | |
| 2426 | /* |
| 2427 | * We could have the case with the max latency tracers |
| 2428 | * that a reset never took place on a cpu. This is evident |
| 2429 | * by the timestamp being before the start of the buffer. |
| 2430 | */ |
| 2431 | while ((event = ring_buffer_iter_peek(buf_iter, &ts))) { |
| 2432 | if (ts >= iter->trace_buffer->time_start) |
| 2433 | break; |
| 2434 | entries++; |
| 2435 | ring_buffer_read(buf_iter, NULL); |
| 2436 | } |
| 2437 | |
| 2438 | per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries; |
| 2439 | } |
| 2440 | |
| 2441 | /* |
| 2442 | * The current tracer is copied to avoid a global locking |
| 2443 | * all around. |
| 2444 | */ |
| 2445 | static void *s_start(struct seq_file *m, loff_t *pos) |
| 2446 | { |
| 2447 | struct trace_iterator *iter = m->private; |
| 2448 | struct trace_array *tr = iter->tr; |
| 2449 | int cpu_file = iter->cpu_file; |
| 2450 | void *p = NULL; |
| 2451 | loff_t l = 0; |
| 2452 | int cpu; |
| 2453 | |
| 2454 | /* |
| 2455 | * copy the tracer to avoid using a global lock all around. |
| 2456 | * iter->trace is a copy of current_trace, the pointer to the |
| 2457 | * name may be used instead of a strcmp(), as iter->trace->name |
| 2458 | * will point to the same string as current_trace->name. |
| 2459 | */ |
| 2460 | mutex_lock(&trace_types_lock); |
| 2461 | if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name)) |
| 2462 | *iter->trace = *tr->current_trace; |
| 2463 | mutex_unlock(&trace_types_lock); |
| 2464 | |
| 2465 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 2466 | if (iter->snapshot && iter->trace->use_max_tr) |
| 2467 | return ERR_PTR(-EBUSY); |
| 2468 | #endif |
| 2469 | |
| 2470 | if (!iter->snapshot) |
| 2471 | atomic_inc(&trace_record_cmdline_disabled); |
| 2472 | |
| 2473 | if (*pos != iter->pos) { |
| 2474 | iter->ent = NULL; |
| 2475 | iter->cpu = 0; |
| 2476 | iter->idx = -1; |
| 2477 | |
| 2478 | if (cpu_file == RING_BUFFER_ALL_CPUS) { |
| 2479 | for_each_tracing_cpu(cpu) |
| 2480 | tracing_iter_reset(iter, cpu); |
| 2481 | } else |
| 2482 | tracing_iter_reset(iter, cpu_file); |
| 2483 | |
| 2484 | iter->leftover = 0; |
| 2485 | for (p = iter; p && l < *pos; p = s_next(m, p, &l)) |
| 2486 | ; |
| 2487 | |
| 2488 | } else { |
| 2489 | /* |
| 2490 | * If we overflowed the seq_file before, then we want |
| 2491 | * to just reuse the trace_seq buffer again. |
| 2492 | */ |
| 2493 | if (iter->leftover) |
| 2494 | p = iter; |
| 2495 | else { |
| 2496 | l = *pos - 1; |
| 2497 | p = s_next(m, p, &l); |
| 2498 | } |
| 2499 | } |
| 2500 | |
| 2501 | trace_event_read_lock(); |
| 2502 | trace_access_lock(cpu_file); |
| 2503 | return p; |
| 2504 | } |
| 2505 | |
| 2506 | static void s_stop(struct seq_file *m, void *p) |
| 2507 | { |
| 2508 | struct trace_iterator *iter = m->private; |
| 2509 | |
| 2510 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 2511 | if (iter->snapshot && iter->trace->use_max_tr) |
| 2512 | return; |
| 2513 | #endif |
| 2514 | |
| 2515 | if (!iter->snapshot) |
| 2516 | atomic_dec(&trace_record_cmdline_disabled); |
| 2517 | |
| 2518 | trace_access_unlock(iter->cpu_file); |
| 2519 | trace_event_read_unlock(); |
| 2520 | } |
| 2521 | |
| 2522 | static void |
| 2523 | get_total_entries(struct trace_buffer *buf, |
| 2524 | unsigned long *total, unsigned long *entries) |
| 2525 | { |
| 2526 | unsigned long count; |
| 2527 | int cpu; |
| 2528 | |
| 2529 | *total = 0; |
| 2530 | *entries = 0; |
| 2531 | |
| 2532 | for_each_tracing_cpu(cpu) { |
| 2533 | count = ring_buffer_entries_cpu(buf->buffer, cpu); |
| 2534 | /* |
| 2535 | * If this buffer has skipped entries, then we hold all |
| 2536 | * entries for the trace and we need to ignore the |
| 2537 | * ones before the time stamp. |
| 2538 | */ |
| 2539 | if (per_cpu_ptr(buf->data, cpu)->skipped_entries) { |
| 2540 | count -= per_cpu_ptr(buf->data, cpu)->skipped_entries; |
| 2541 | /* total is the same as the entries */ |
| 2542 | *total += count; |
| 2543 | } else |
| 2544 | *total += count + |
| 2545 | ring_buffer_overrun_cpu(buf->buffer, cpu); |
| 2546 | *entries += count; |
| 2547 | } |
| 2548 | } |
| 2549 | |
| 2550 | static void print_lat_help_header(struct seq_file *m) |
| 2551 | { |
| 2552 | seq_puts(m, "# _------=> CPU# \n" |
| 2553 | "# / _-----=> irqs-off \n" |
| 2554 | "# | / _----=> need-resched \n" |
| 2555 | "# || / _---=> hardirq/softirq \n" |
| 2556 | "# ||| / _--=> preempt-depth \n" |
| 2557 | "# |||| / delay \n" |
| 2558 | "# cmd pid ||||| time | caller \n" |
| 2559 | "# \\ / ||||| \\ | / \n"); |
| 2560 | } |
| 2561 | |
| 2562 | static void print_event_info(struct trace_buffer *buf, struct seq_file *m) |
| 2563 | { |
| 2564 | unsigned long total; |
| 2565 | unsigned long entries; |
| 2566 | |
| 2567 | get_total_entries(buf, &total, &entries); |
| 2568 | seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n", |
| 2569 | entries, total, num_online_cpus()); |
| 2570 | seq_puts(m, "#\n"); |
| 2571 | } |
| 2572 | |
| 2573 | static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m) |
| 2574 | { |
| 2575 | print_event_info(buf, m); |
| 2576 | seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n" |
| 2577 | "# | | | | |\n"); |
| 2578 | } |
| 2579 | |
| 2580 | static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m) |
| 2581 | { |
| 2582 | print_event_info(buf, m); |
| 2583 | seq_puts(m, "# _-----=> irqs-off\n" |
| 2584 | "# / _----=> need-resched\n" |
| 2585 | "# | / _---=> hardirq/softirq\n" |
| 2586 | "# || / _--=> preempt-depth\n" |
| 2587 | "# ||| / delay\n" |
| 2588 | "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n" |
| 2589 | "# | | | |||| | |\n"); |
| 2590 | } |
| 2591 | |
| 2592 | void |
| 2593 | print_trace_header(struct seq_file *m, struct trace_iterator *iter) |
| 2594 | { |
| 2595 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); |
| 2596 | struct trace_buffer *buf = iter->trace_buffer; |
| 2597 | struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu); |
| 2598 | struct tracer *type = iter->trace; |
| 2599 | unsigned long entries; |
| 2600 | unsigned long total; |
| 2601 | const char *name = "preemption"; |
| 2602 | |
| 2603 | name = type->name; |
| 2604 | |
| 2605 | get_total_entries(buf, &total, &entries); |
| 2606 | |
| 2607 | seq_printf(m, "# %s latency trace v1.1.5 on %s\n", |
| 2608 | name, UTS_RELEASE); |
| 2609 | seq_puts(m, "# -----------------------------------" |
| 2610 | "---------------------------------\n"); |
| 2611 | seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |" |
| 2612 | " (M:%s VP:%d, KP:%d, SP:%d HP:%d", |
| 2613 | nsecs_to_usecs(data->saved_latency), |
| 2614 | entries, |
| 2615 | total, |
| 2616 | buf->cpu, |
| 2617 | #if defined(CONFIG_PREEMPT_NONE) |
| 2618 | "server", |
| 2619 | #elif defined(CONFIG_PREEMPT_VOLUNTARY) |
| 2620 | "desktop", |
| 2621 | #elif defined(CONFIG_PREEMPT) |
| 2622 | "preempt", |
| 2623 | #else |
| 2624 | "unknown", |
| 2625 | #endif |
| 2626 | /* These are reserved for later use */ |
| 2627 | 0, 0, 0, 0); |
| 2628 | #ifdef CONFIG_SMP |
| 2629 | seq_printf(m, " #P:%d)\n", num_online_cpus()); |
| 2630 | #else |
| 2631 | seq_puts(m, ")\n"); |
| 2632 | #endif |
| 2633 | seq_puts(m, "# -----------------\n"); |
| 2634 | seq_printf(m, "# | task: %.16s-%d " |
| 2635 | "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", |
| 2636 | data->comm, data->pid, |
| 2637 | from_kuid_munged(seq_user_ns(m), data->uid), data->nice, |
| 2638 | data->policy, data->rt_priority); |
| 2639 | seq_puts(m, "# -----------------\n"); |
| 2640 | |
| 2641 | if (data->critical_start) { |
| 2642 | seq_puts(m, "# => started at: "); |
| 2643 | seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags); |
| 2644 | trace_print_seq(m, &iter->seq); |
| 2645 | seq_puts(m, "\n# => ended at: "); |
| 2646 | seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags); |
| 2647 | trace_print_seq(m, &iter->seq); |
| 2648 | seq_puts(m, "\n#\n"); |
| 2649 | } |
| 2650 | |
| 2651 | seq_puts(m, "#\n"); |
| 2652 | } |
| 2653 | |
| 2654 | static void test_cpu_buff_start(struct trace_iterator *iter) |
| 2655 | { |
| 2656 | struct trace_seq *s = &iter->seq; |
| 2657 | |
| 2658 | if (!(trace_flags & TRACE_ITER_ANNOTATE)) |
| 2659 | return; |
| 2660 | |
| 2661 | if (!(iter->iter_flags & TRACE_FILE_ANNOTATE)) |
| 2662 | return; |
| 2663 | |
| 2664 | if (cpumask_test_cpu(iter->cpu, iter->started)) |
| 2665 | return; |
| 2666 | |
| 2667 | if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries) |
| 2668 | return; |
| 2669 | |
| 2670 | cpumask_set_cpu(iter->cpu, iter->started); |
| 2671 | |
| 2672 | /* Don't print started cpu buffer for the first entry of the trace */ |
| 2673 | if (iter->idx > 1) |
| 2674 | trace_seq_printf(s, "##### CPU %u buffer started ####\n", |
| 2675 | iter->cpu); |
| 2676 | } |
| 2677 | |
| 2678 | static enum print_line_t print_trace_fmt(struct trace_iterator *iter) |
| 2679 | { |
| 2680 | struct trace_seq *s = &iter->seq; |
| 2681 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); |
| 2682 | struct trace_entry *entry; |
| 2683 | struct trace_event *event; |
| 2684 | |
| 2685 | entry = iter->ent; |
| 2686 | |
| 2687 | test_cpu_buff_start(iter); |
| 2688 | |
| 2689 | event = ftrace_find_event(entry->type); |
| 2690 | |
| 2691 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { |
| 2692 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) |
| 2693 | trace_print_lat_context(iter); |
| 2694 | else |
| 2695 | trace_print_context(iter); |
| 2696 | } |
| 2697 | |
| 2698 | if (trace_seq_has_overflowed(s)) |
| 2699 | return TRACE_TYPE_PARTIAL_LINE; |
| 2700 | |
| 2701 | if (event) |
| 2702 | return event->funcs->trace(iter, sym_flags, event); |
| 2703 | |
| 2704 | trace_seq_printf(s, "Unknown type %d\n", entry->type); |
| 2705 | |
| 2706 | return trace_handle_return(s); |
| 2707 | } |
| 2708 | |
| 2709 | static enum print_line_t print_raw_fmt(struct trace_iterator *iter) |
| 2710 | { |
| 2711 | struct trace_seq *s = &iter->seq; |
| 2712 | struct trace_entry *entry; |
| 2713 | struct trace_event *event; |
| 2714 | |
| 2715 | entry = iter->ent; |
| 2716 | |
| 2717 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) |
| 2718 | trace_seq_printf(s, "%d %d %llu ", |
| 2719 | entry->pid, iter->cpu, iter->ts); |
| 2720 | |
| 2721 | if (trace_seq_has_overflowed(s)) |
| 2722 | return TRACE_TYPE_PARTIAL_LINE; |
| 2723 | |
| 2724 | event = ftrace_find_event(entry->type); |
| 2725 | if (event) |
| 2726 | return event->funcs->raw(iter, 0, event); |
| 2727 | |
| 2728 | trace_seq_printf(s, "%d ?\n", entry->type); |
| 2729 | |
| 2730 | return trace_handle_return(s); |
| 2731 | } |
| 2732 | |
| 2733 | static enum print_line_t print_hex_fmt(struct trace_iterator *iter) |
| 2734 | { |
| 2735 | struct trace_seq *s = &iter->seq; |
| 2736 | unsigned char newline = '\n'; |
| 2737 | struct trace_entry *entry; |
| 2738 | struct trace_event *event; |
| 2739 | |
| 2740 | entry = iter->ent; |
| 2741 | |
| 2742 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { |
| 2743 | SEQ_PUT_HEX_FIELD(s, entry->pid); |
| 2744 | SEQ_PUT_HEX_FIELD(s, iter->cpu); |
| 2745 | SEQ_PUT_HEX_FIELD(s, iter->ts); |
| 2746 | if (trace_seq_has_overflowed(s)) |
| 2747 | return TRACE_TYPE_PARTIAL_LINE; |
| 2748 | } |
| 2749 | |
| 2750 | event = ftrace_find_event(entry->type); |
| 2751 | if (event) { |
| 2752 | enum print_line_t ret = event->funcs->hex(iter, 0, event); |
| 2753 | if (ret != TRACE_TYPE_HANDLED) |
| 2754 | return ret; |
| 2755 | } |
| 2756 | |
| 2757 | SEQ_PUT_FIELD(s, newline); |
| 2758 | |
| 2759 | return trace_handle_return(s); |
| 2760 | } |
| 2761 | |
| 2762 | static enum print_line_t print_bin_fmt(struct trace_iterator *iter) |
| 2763 | { |
| 2764 | struct trace_seq *s = &iter->seq; |
| 2765 | struct trace_entry *entry; |
| 2766 | struct trace_event *event; |
| 2767 | |
| 2768 | entry = iter->ent; |
| 2769 | |
| 2770 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { |
| 2771 | SEQ_PUT_FIELD(s, entry->pid); |
| 2772 | SEQ_PUT_FIELD(s, iter->cpu); |
| 2773 | SEQ_PUT_FIELD(s, iter->ts); |
| 2774 | if (trace_seq_has_overflowed(s)) |
| 2775 | return TRACE_TYPE_PARTIAL_LINE; |
| 2776 | } |
| 2777 | |
| 2778 | event = ftrace_find_event(entry->type); |
| 2779 | return event ? event->funcs->binary(iter, 0, event) : |
| 2780 | TRACE_TYPE_HANDLED; |
| 2781 | } |
| 2782 | |
| 2783 | int trace_empty(struct trace_iterator *iter) |
| 2784 | { |
| 2785 | struct ring_buffer_iter *buf_iter; |
| 2786 | int cpu; |
| 2787 | |
| 2788 | /* If we are looking at one CPU buffer, only check that one */ |
| 2789 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { |
| 2790 | cpu = iter->cpu_file; |
| 2791 | buf_iter = trace_buffer_iter(iter, cpu); |
| 2792 | if (buf_iter) { |
| 2793 | if (!ring_buffer_iter_empty(buf_iter)) |
| 2794 | return 0; |
| 2795 | } else { |
| 2796 | if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu)) |
| 2797 | return 0; |
| 2798 | } |
| 2799 | return 1; |
| 2800 | } |
| 2801 | |
| 2802 | for_each_tracing_cpu(cpu) { |
| 2803 | buf_iter = trace_buffer_iter(iter, cpu); |
| 2804 | if (buf_iter) { |
| 2805 | if (!ring_buffer_iter_empty(buf_iter)) |
| 2806 | return 0; |
| 2807 | } else { |
| 2808 | if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu)) |
| 2809 | return 0; |
| 2810 | } |
| 2811 | } |
| 2812 | |
| 2813 | return 1; |
| 2814 | } |
| 2815 | |
| 2816 | /* Called with trace_event_read_lock() held. */ |
| 2817 | enum print_line_t print_trace_line(struct trace_iterator *iter) |
| 2818 | { |
| 2819 | enum print_line_t ret; |
| 2820 | |
| 2821 | if (iter->lost_events) { |
| 2822 | trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n", |
| 2823 | iter->cpu, iter->lost_events); |
| 2824 | if (trace_seq_has_overflowed(&iter->seq)) |
| 2825 | return TRACE_TYPE_PARTIAL_LINE; |
| 2826 | } |
| 2827 | |
| 2828 | if (iter->trace && iter->trace->print_line) { |
| 2829 | ret = iter->trace->print_line(iter); |
| 2830 | if (ret != TRACE_TYPE_UNHANDLED) |
| 2831 | return ret; |
| 2832 | } |
| 2833 | |
| 2834 | if (iter->ent->type == TRACE_BPUTS && |
| 2835 | trace_flags & TRACE_ITER_PRINTK && |
| 2836 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) |
| 2837 | return trace_print_bputs_msg_only(iter); |
| 2838 | |
| 2839 | if (iter->ent->type == TRACE_BPRINT && |
| 2840 | trace_flags & TRACE_ITER_PRINTK && |
| 2841 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) |
| 2842 | return trace_print_bprintk_msg_only(iter); |
| 2843 | |
| 2844 | if (iter->ent->type == TRACE_PRINT && |
| 2845 | trace_flags & TRACE_ITER_PRINTK && |
| 2846 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) |
| 2847 | return trace_print_printk_msg_only(iter); |
| 2848 | |
| 2849 | if (trace_flags & TRACE_ITER_BIN) |
| 2850 | return print_bin_fmt(iter); |
| 2851 | |
| 2852 | if (trace_flags & TRACE_ITER_HEX) |
| 2853 | return print_hex_fmt(iter); |
| 2854 | |
| 2855 | if (trace_flags & TRACE_ITER_RAW) |
| 2856 | return print_raw_fmt(iter); |
| 2857 | |
| 2858 | return print_trace_fmt(iter); |
| 2859 | } |
| 2860 | |
| 2861 | void trace_latency_header(struct seq_file *m) |
| 2862 | { |
| 2863 | struct trace_iterator *iter = m->private; |
| 2864 | |
| 2865 | /* print nothing if the buffers are empty */ |
| 2866 | if (trace_empty(iter)) |
| 2867 | return; |
| 2868 | |
| 2869 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) |
| 2870 | print_trace_header(m, iter); |
| 2871 | |
| 2872 | if (!(trace_flags & TRACE_ITER_VERBOSE)) |
| 2873 | print_lat_help_header(m); |
| 2874 | } |
| 2875 | |
| 2876 | void trace_default_header(struct seq_file *m) |
| 2877 | { |
| 2878 | struct trace_iterator *iter = m->private; |
| 2879 | |
| 2880 | if (!(trace_flags & TRACE_ITER_CONTEXT_INFO)) |
| 2881 | return; |
| 2882 | |
| 2883 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { |
| 2884 | /* print nothing if the buffers are empty */ |
| 2885 | if (trace_empty(iter)) |
| 2886 | return; |
| 2887 | print_trace_header(m, iter); |
| 2888 | if (!(trace_flags & TRACE_ITER_VERBOSE)) |
| 2889 | print_lat_help_header(m); |
| 2890 | } else { |
| 2891 | if (!(trace_flags & TRACE_ITER_VERBOSE)) { |
| 2892 | if (trace_flags & TRACE_ITER_IRQ_INFO) |
| 2893 | print_func_help_header_irq(iter->trace_buffer, m); |
| 2894 | else |
| 2895 | print_func_help_header(iter->trace_buffer, m); |
| 2896 | } |
| 2897 | } |
| 2898 | } |
| 2899 | |
| 2900 | static void test_ftrace_alive(struct seq_file *m) |
| 2901 | { |
| 2902 | if (!ftrace_is_dead()) |
| 2903 | return; |
| 2904 | seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n" |
| 2905 | "# MAY BE MISSING FUNCTION EVENTS\n"); |
| 2906 | } |
| 2907 | |
| 2908 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 2909 | static void show_snapshot_main_help(struct seq_file *m) |
| 2910 | { |
| 2911 | seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n" |
| 2912 | "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n" |
| 2913 | "# Takes a snapshot of the main buffer.\n" |
| 2914 | "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n" |
| 2915 | "# (Doesn't have to be '2' works with any number that\n" |
| 2916 | "# is not a '0' or '1')\n"); |
| 2917 | } |
| 2918 | |
| 2919 | static void show_snapshot_percpu_help(struct seq_file *m) |
| 2920 | { |
| 2921 | seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n"); |
| 2922 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
| 2923 | seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n" |
| 2924 | "# Takes a snapshot of the main buffer for this cpu.\n"); |
| 2925 | #else |
| 2926 | seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n" |
| 2927 | "# Must use main snapshot file to allocate.\n"); |
| 2928 | #endif |
| 2929 | seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n" |
| 2930 | "# (Doesn't have to be '2' works with any number that\n" |
| 2931 | "# is not a '0' or '1')\n"); |
| 2932 | } |
| 2933 | |
| 2934 | static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) |
| 2935 | { |
| 2936 | if (iter->tr->allocated_snapshot) |
| 2937 | seq_puts(m, "#\n# * Snapshot is allocated *\n#\n"); |
| 2938 | else |
| 2939 | seq_puts(m, "#\n# * Snapshot is freed *\n#\n"); |
| 2940 | |
| 2941 | seq_puts(m, "# Snapshot commands:\n"); |
| 2942 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) |
| 2943 | show_snapshot_main_help(m); |
| 2944 | else |
| 2945 | show_snapshot_percpu_help(m); |
| 2946 | } |
| 2947 | #else |
| 2948 | /* Should never be called */ |
| 2949 | static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { } |
| 2950 | #endif |
| 2951 | |
| 2952 | static int s_show(struct seq_file *m, void *v) |
| 2953 | { |
| 2954 | struct trace_iterator *iter = v; |
| 2955 | int ret; |
| 2956 | |
| 2957 | if (iter->ent == NULL) { |
| 2958 | if (iter->tr) { |
| 2959 | seq_printf(m, "# tracer: %s\n", iter->trace->name); |
| 2960 | seq_puts(m, "#\n"); |
| 2961 | test_ftrace_alive(m); |
| 2962 | } |
| 2963 | if (iter->snapshot && trace_empty(iter)) |
| 2964 | print_snapshot_help(m, iter); |
| 2965 | else if (iter->trace && iter->trace->print_header) |
| 2966 | iter->trace->print_header(m); |
| 2967 | else |
| 2968 | trace_default_header(m); |
| 2969 | |
| 2970 | } else if (iter->leftover) { |
| 2971 | /* |
| 2972 | * If we filled the seq_file buffer earlier, we |
| 2973 | * want to just show it now. |
| 2974 | */ |
| 2975 | ret = trace_print_seq(m, &iter->seq); |
| 2976 | |
| 2977 | /* ret should this time be zero, but you never know */ |
| 2978 | iter->leftover = ret; |
| 2979 | |
| 2980 | } else { |
| 2981 | print_trace_line(iter); |
| 2982 | ret = trace_print_seq(m, &iter->seq); |
| 2983 | /* |
| 2984 | * If we overflow the seq_file buffer, then it will |
| 2985 | * ask us for this data again at start up. |
| 2986 | * Use that instead. |
| 2987 | * ret is 0 if seq_file write succeeded. |
| 2988 | * -1 otherwise. |
| 2989 | */ |
| 2990 | iter->leftover = ret; |
| 2991 | } |
| 2992 | |
| 2993 | return 0; |
| 2994 | } |
| 2995 | |
| 2996 | /* |
| 2997 | * Should be used after trace_array_get(), trace_types_lock |
| 2998 | * ensures that i_cdev was already initialized. |
| 2999 | */ |
| 3000 | static inline int tracing_get_cpu(struct inode *inode) |
| 3001 | { |
| 3002 | if (inode->i_cdev) /* See trace_create_cpu_file() */ |
| 3003 | return (long)inode->i_cdev - 1; |
| 3004 | return RING_BUFFER_ALL_CPUS; |
| 3005 | } |
| 3006 | |
| 3007 | static const struct seq_operations tracer_seq_ops = { |
| 3008 | .start = s_start, |
| 3009 | .next = s_next, |
| 3010 | .stop = s_stop, |
| 3011 | .show = s_show, |
| 3012 | }; |
| 3013 | |
| 3014 | static struct trace_iterator * |
| 3015 | __tracing_open(struct inode *inode, struct file *file, bool snapshot) |
| 3016 | { |
| 3017 | struct trace_array *tr = inode->i_private; |
| 3018 | struct trace_iterator *iter; |
| 3019 | int cpu; |
| 3020 | |
| 3021 | if (tracing_disabled) |
| 3022 | return ERR_PTR(-ENODEV); |
| 3023 | |
| 3024 | iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter)); |
| 3025 | if (!iter) |
| 3026 | return ERR_PTR(-ENOMEM); |
| 3027 | |
| 3028 | iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter), |
| 3029 | GFP_KERNEL); |
| 3030 | if (!iter->buffer_iter) |
| 3031 | goto release; |
| 3032 | |
| 3033 | /* |
| 3034 | * We make a copy of the current tracer to avoid concurrent |
| 3035 | * changes on it while we are reading. |
| 3036 | */ |
| 3037 | mutex_lock(&trace_types_lock); |
| 3038 | iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL); |
| 3039 | if (!iter->trace) |
| 3040 | goto fail; |
| 3041 | |
| 3042 | *iter->trace = *tr->current_trace; |
| 3043 | |
| 3044 | if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL)) |
| 3045 | goto fail; |
| 3046 | |
| 3047 | iter->tr = tr; |
| 3048 | |
| 3049 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 3050 | /* Currently only the top directory has a snapshot */ |
| 3051 | if (tr->current_trace->print_max || snapshot) |
| 3052 | iter->trace_buffer = &tr->max_buffer; |
| 3053 | else |
| 3054 | #endif |
| 3055 | iter->trace_buffer = &tr->trace_buffer; |
| 3056 | iter->snapshot = snapshot; |
| 3057 | iter->pos = -1; |
| 3058 | iter->cpu_file = tracing_get_cpu(inode); |
| 3059 | mutex_init(&iter->mutex); |
| 3060 | |
| 3061 | /* Notify the tracer early; before we stop tracing. */ |
| 3062 | if (iter->trace && iter->trace->open) |
| 3063 | iter->trace->open(iter); |
| 3064 | |
| 3065 | /* Annotate start of buffers if we had overruns */ |
| 3066 | if (ring_buffer_overruns(iter->trace_buffer->buffer)) |
| 3067 | iter->iter_flags |= TRACE_FILE_ANNOTATE; |
| 3068 | |
| 3069 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ |
| 3070 | if (trace_clocks[tr->clock_id].in_ns) |
| 3071 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; |
| 3072 | |
| 3073 | /* stop the trace while dumping if we are not opening "snapshot" */ |
| 3074 | if (!iter->snapshot) |
| 3075 | tracing_stop_tr(tr); |
| 3076 | |
| 3077 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) { |
| 3078 | for_each_tracing_cpu(cpu) { |
| 3079 | iter->buffer_iter[cpu] = |
| 3080 | ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu); |
| 3081 | } |
| 3082 | ring_buffer_read_prepare_sync(); |
| 3083 | for_each_tracing_cpu(cpu) { |
| 3084 | ring_buffer_read_start(iter->buffer_iter[cpu]); |
| 3085 | tracing_iter_reset(iter, cpu); |
| 3086 | } |
| 3087 | } else { |
| 3088 | cpu = iter->cpu_file; |
| 3089 | iter->buffer_iter[cpu] = |
| 3090 | ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu); |
| 3091 | ring_buffer_read_prepare_sync(); |
| 3092 | ring_buffer_read_start(iter->buffer_iter[cpu]); |
| 3093 | tracing_iter_reset(iter, cpu); |
| 3094 | } |
| 3095 | |
| 3096 | mutex_unlock(&trace_types_lock); |
| 3097 | |
| 3098 | return iter; |
| 3099 | |
| 3100 | fail: |
| 3101 | mutex_unlock(&trace_types_lock); |
| 3102 | kfree(iter->trace); |
| 3103 | kfree(iter->buffer_iter); |
| 3104 | release: |
| 3105 | seq_release_private(inode, file); |
| 3106 | return ERR_PTR(-ENOMEM); |
| 3107 | } |
| 3108 | |
| 3109 | int tracing_open_generic(struct inode *inode, struct file *filp) |
| 3110 | { |
| 3111 | if (tracing_disabled) |
| 3112 | return -ENODEV; |
| 3113 | |
| 3114 | filp->private_data = inode->i_private; |
| 3115 | return 0; |
| 3116 | } |
| 3117 | |
| 3118 | bool tracing_is_disabled(void) |
| 3119 | { |
| 3120 | return (tracing_disabled) ? true: false; |
| 3121 | } |
| 3122 | |
| 3123 | /* |
| 3124 | * Open and update trace_array ref count. |
| 3125 | * Must have the current trace_array passed to it. |
| 3126 | */ |
| 3127 | static int tracing_open_generic_tr(struct inode *inode, struct file *filp) |
| 3128 | { |
| 3129 | struct trace_array *tr = inode->i_private; |
| 3130 | |
| 3131 | if (tracing_disabled) |
| 3132 | return -ENODEV; |
| 3133 | |
| 3134 | if (trace_array_get(tr) < 0) |
| 3135 | return -ENODEV; |
| 3136 | |
| 3137 | filp->private_data = inode->i_private; |
| 3138 | |
| 3139 | return 0; |
| 3140 | } |
| 3141 | |
| 3142 | static int tracing_release(struct inode *inode, struct file *file) |
| 3143 | { |
| 3144 | struct trace_array *tr = inode->i_private; |
| 3145 | struct seq_file *m = file->private_data; |
| 3146 | struct trace_iterator *iter; |
| 3147 | int cpu; |
| 3148 | |
| 3149 | if (!(file->f_mode & FMODE_READ)) { |
| 3150 | trace_array_put(tr); |
| 3151 | return 0; |
| 3152 | } |
| 3153 | |
| 3154 | /* Writes do not use seq_file */ |
| 3155 | iter = m->private; |
| 3156 | mutex_lock(&trace_types_lock); |
| 3157 | |
| 3158 | for_each_tracing_cpu(cpu) { |
| 3159 | if (iter->buffer_iter[cpu]) |
| 3160 | ring_buffer_read_finish(iter->buffer_iter[cpu]); |
| 3161 | } |
| 3162 | |
| 3163 | if (iter->trace && iter->trace->close) |
| 3164 | iter->trace->close(iter); |
| 3165 | |
| 3166 | if (!iter->snapshot) |
| 3167 | /* reenable tracing if it was previously enabled */ |
| 3168 | tracing_start_tr(tr); |
| 3169 | |
| 3170 | __trace_array_put(tr); |
| 3171 | |
| 3172 | mutex_unlock(&trace_types_lock); |
| 3173 | |
| 3174 | mutex_destroy(&iter->mutex); |
| 3175 | free_cpumask_var(iter->started); |
| 3176 | kfree(iter->trace); |
| 3177 | kfree(iter->buffer_iter); |
| 3178 | seq_release_private(inode, file); |
| 3179 | |
| 3180 | return 0; |
| 3181 | } |
| 3182 | |
| 3183 | static int tracing_release_generic_tr(struct inode *inode, struct file *file) |
| 3184 | { |
| 3185 | struct trace_array *tr = inode->i_private; |
| 3186 | |
| 3187 | trace_array_put(tr); |
| 3188 | return 0; |
| 3189 | } |
| 3190 | |
| 3191 | static int tracing_single_release_tr(struct inode *inode, struct file *file) |
| 3192 | { |
| 3193 | struct trace_array *tr = inode->i_private; |
| 3194 | |
| 3195 | trace_array_put(tr); |
| 3196 | |
| 3197 | return single_release(inode, file); |
| 3198 | } |
| 3199 | |
| 3200 | static int tracing_open(struct inode *inode, struct file *file) |
| 3201 | { |
| 3202 | struct trace_array *tr = inode->i_private; |
| 3203 | struct trace_iterator *iter; |
| 3204 | int ret = 0; |
| 3205 | |
| 3206 | if (trace_array_get(tr) < 0) |
| 3207 | return -ENODEV; |
| 3208 | |
| 3209 | /* If this file was open for write, then erase contents */ |
| 3210 | if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) { |
| 3211 | int cpu = tracing_get_cpu(inode); |
| 3212 | |
| 3213 | if (cpu == RING_BUFFER_ALL_CPUS) |
| 3214 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 3215 | else |
| 3216 | tracing_reset(&tr->trace_buffer, cpu); |
| 3217 | } |
| 3218 | |
| 3219 | if (file->f_mode & FMODE_READ) { |
| 3220 | iter = __tracing_open(inode, file, false); |
| 3221 | if (IS_ERR(iter)) |
| 3222 | ret = PTR_ERR(iter); |
| 3223 | else if (trace_flags & TRACE_ITER_LATENCY_FMT) |
| 3224 | iter->iter_flags |= TRACE_FILE_LAT_FMT; |
| 3225 | } |
| 3226 | |
| 3227 | if (ret < 0) |
| 3228 | trace_array_put(tr); |
| 3229 | |
| 3230 | return ret; |
| 3231 | } |
| 3232 | |
| 3233 | /* |
| 3234 | * Some tracers are not suitable for instance buffers. |
| 3235 | * A tracer is always available for the global array (toplevel) |
| 3236 | * or if it explicitly states that it is. |
| 3237 | */ |
| 3238 | static bool |
| 3239 | trace_ok_for_array(struct tracer *t, struct trace_array *tr) |
| 3240 | { |
| 3241 | return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances; |
| 3242 | } |
| 3243 | |
| 3244 | /* Find the next tracer that this trace array may use */ |
| 3245 | static struct tracer * |
| 3246 | get_tracer_for_array(struct trace_array *tr, struct tracer *t) |
| 3247 | { |
| 3248 | while (t && !trace_ok_for_array(t, tr)) |
| 3249 | t = t->next; |
| 3250 | |
| 3251 | return t; |
| 3252 | } |
| 3253 | |
| 3254 | static void * |
| 3255 | t_next(struct seq_file *m, void *v, loff_t *pos) |
| 3256 | { |
| 3257 | struct trace_array *tr = m->private; |
| 3258 | struct tracer *t = v; |
| 3259 | |
| 3260 | (*pos)++; |
| 3261 | |
| 3262 | if (t) |
| 3263 | t = get_tracer_for_array(tr, t->next); |
| 3264 | |
| 3265 | return t; |
| 3266 | } |
| 3267 | |
| 3268 | static void *t_start(struct seq_file *m, loff_t *pos) |
| 3269 | { |
| 3270 | struct trace_array *tr = m->private; |
| 3271 | struct tracer *t; |
| 3272 | loff_t l = 0; |
| 3273 | |
| 3274 | mutex_lock(&trace_types_lock); |
| 3275 | |
| 3276 | t = get_tracer_for_array(tr, trace_types); |
| 3277 | for (; t && l < *pos; t = t_next(m, t, &l)) |
| 3278 | ; |
| 3279 | |
| 3280 | return t; |
| 3281 | } |
| 3282 | |
| 3283 | static void t_stop(struct seq_file *m, void *p) |
| 3284 | { |
| 3285 | mutex_unlock(&trace_types_lock); |
| 3286 | } |
| 3287 | |
| 3288 | static int t_show(struct seq_file *m, void *v) |
| 3289 | { |
| 3290 | struct tracer *t = v; |
| 3291 | |
| 3292 | if (!t) |
| 3293 | return 0; |
| 3294 | |
| 3295 | seq_puts(m, t->name); |
| 3296 | if (t->next) |
| 3297 | seq_putc(m, ' '); |
| 3298 | else |
| 3299 | seq_putc(m, '\n'); |
| 3300 | |
| 3301 | return 0; |
| 3302 | } |
| 3303 | |
| 3304 | static const struct seq_operations show_traces_seq_ops = { |
| 3305 | .start = t_start, |
| 3306 | .next = t_next, |
| 3307 | .stop = t_stop, |
| 3308 | .show = t_show, |
| 3309 | }; |
| 3310 | |
| 3311 | static int show_traces_open(struct inode *inode, struct file *file) |
| 3312 | { |
| 3313 | struct trace_array *tr = inode->i_private; |
| 3314 | struct seq_file *m; |
| 3315 | int ret; |
| 3316 | |
| 3317 | if (tracing_disabled) |
| 3318 | return -ENODEV; |
| 3319 | |
| 3320 | ret = seq_open(file, &show_traces_seq_ops); |
| 3321 | if (ret) |
| 3322 | return ret; |
| 3323 | |
| 3324 | m = file->private_data; |
| 3325 | m->private = tr; |
| 3326 | |
| 3327 | return 0; |
| 3328 | } |
| 3329 | |
| 3330 | static ssize_t |
| 3331 | tracing_write_stub(struct file *filp, const char __user *ubuf, |
| 3332 | size_t count, loff_t *ppos) |
| 3333 | { |
| 3334 | return count; |
| 3335 | } |
| 3336 | |
| 3337 | loff_t tracing_lseek(struct file *file, loff_t offset, int whence) |
| 3338 | { |
| 3339 | int ret; |
| 3340 | |
| 3341 | if (file->f_mode & FMODE_READ) |
| 3342 | ret = seq_lseek(file, offset, whence); |
| 3343 | else |
| 3344 | file->f_pos = ret = 0; |
| 3345 | |
| 3346 | return ret; |
| 3347 | } |
| 3348 | |
| 3349 | static const struct file_operations tracing_fops = { |
| 3350 | .open = tracing_open, |
| 3351 | .read = seq_read, |
| 3352 | .write = tracing_write_stub, |
| 3353 | .llseek = tracing_lseek, |
| 3354 | .release = tracing_release, |
| 3355 | }; |
| 3356 | |
| 3357 | static const struct file_operations show_traces_fops = { |
| 3358 | .open = show_traces_open, |
| 3359 | .read = seq_read, |
| 3360 | .release = seq_release, |
| 3361 | .llseek = seq_lseek, |
| 3362 | }; |
| 3363 | |
| 3364 | /* |
| 3365 | * The tracer itself will not take this lock, but still we want |
| 3366 | * to provide a consistent cpumask to user-space: |
| 3367 | */ |
| 3368 | static DEFINE_MUTEX(tracing_cpumask_update_lock); |
| 3369 | |
| 3370 | /* |
| 3371 | * Temporary storage for the character representation of the |
| 3372 | * CPU bitmask (and one more byte for the newline): |
| 3373 | */ |
| 3374 | static char mask_str[NR_CPUS + 1]; |
| 3375 | |
| 3376 | static ssize_t |
| 3377 | tracing_cpumask_read(struct file *filp, char __user *ubuf, |
| 3378 | size_t count, loff_t *ppos) |
| 3379 | { |
| 3380 | struct trace_array *tr = file_inode(filp)->i_private; |
| 3381 | int len; |
| 3382 | |
| 3383 | mutex_lock(&tracing_cpumask_update_lock); |
| 3384 | |
| 3385 | len = snprintf(mask_str, count, "%*pb\n", |
| 3386 | cpumask_pr_args(tr->tracing_cpumask)); |
| 3387 | if (len >= count) { |
| 3388 | count = -EINVAL; |
| 3389 | goto out_err; |
| 3390 | } |
| 3391 | count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1); |
| 3392 | |
| 3393 | out_err: |
| 3394 | mutex_unlock(&tracing_cpumask_update_lock); |
| 3395 | |
| 3396 | return count; |
| 3397 | } |
| 3398 | |
| 3399 | static ssize_t |
| 3400 | tracing_cpumask_write(struct file *filp, const char __user *ubuf, |
| 3401 | size_t count, loff_t *ppos) |
| 3402 | { |
| 3403 | struct trace_array *tr = file_inode(filp)->i_private; |
| 3404 | cpumask_var_t tracing_cpumask_new; |
| 3405 | int err, cpu; |
| 3406 | |
| 3407 | if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL)) |
| 3408 | return -ENOMEM; |
| 3409 | |
| 3410 | err = cpumask_parse_user(ubuf, count, tracing_cpumask_new); |
| 3411 | if (err) |
| 3412 | goto err_unlock; |
| 3413 | |
| 3414 | mutex_lock(&tracing_cpumask_update_lock); |
| 3415 | |
| 3416 | local_irq_disable(); |
| 3417 | arch_spin_lock(&tr->max_lock); |
| 3418 | for_each_tracing_cpu(cpu) { |
| 3419 | /* |
| 3420 | * Increase/decrease the disabled counter if we are |
| 3421 | * about to flip a bit in the cpumask: |
| 3422 | */ |
| 3423 | if (cpumask_test_cpu(cpu, tr->tracing_cpumask) && |
| 3424 | !cpumask_test_cpu(cpu, tracing_cpumask_new)) { |
| 3425 | atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled); |
| 3426 | ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu); |
| 3427 | } |
| 3428 | if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) && |
| 3429 | cpumask_test_cpu(cpu, tracing_cpumask_new)) { |
| 3430 | atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled); |
| 3431 | ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu); |
| 3432 | } |
| 3433 | } |
| 3434 | arch_spin_unlock(&tr->max_lock); |
| 3435 | local_irq_enable(); |
| 3436 | |
| 3437 | cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new); |
| 3438 | |
| 3439 | mutex_unlock(&tracing_cpumask_update_lock); |
| 3440 | free_cpumask_var(tracing_cpumask_new); |
| 3441 | |
| 3442 | return count; |
| 3443 | |
| 3444 | err_unlock: |
| 3445 | free_cpumask_var(tracing_cpumask_new); |
| 3446 | |
| 3447 | return err; |
| 3448 | } |
| 3449 | |
| 3450 | static const struct file_operations tracing_cpumask_fops = { |
| 3451 | .open = tracing_open_generic_tr, |
| 3452 | .read = tracing_cpumask_read, |
| 3453 | .write = tracing_cpumask_write, |
| 3454 | .release = tracing_release_generic_tr, |
| 3455 | .llseek = generic_file_llseek, |
| 3456 | }; |
| 3457 | |
| 3458 | static int tracing_trace_options_show(struct seq_file *m, void *v) |
| 3459 | { |
| 3460 | struct tracer_opt *trace_opts; |
| 3461 | struct trace_array *tr = m->private; |
| 3462 | u32 tracer_flags; |
| 3463 | int i; |
| 3464 | |
| 3465 | mutex_lock(&trace_types_lock); |
| 3466 | tracer_flags = tr->current_trace->flags->val; |
| 3467 | trace_opts = tr->current_trace->flags->opts; |
| 3468 | |
| 3469 | for (i = 0; trace_options[i]; i++) { |
| 3470 | if (trace_flags & (1 << i)) |
| 3471 | seq_printf(m, "%s\n", trace_options[i]); |
| 3472 | else |
| 3473 | seq_printf(m, "no%s\n", trace_options[i]); |
| 3474 | } |
| 3475 | |
| 3476 | for (i = 0; trace_opts[i].name; i++) { |
| 3477 | if (tracer_flags & trace_opts[i].bit) |
| 3478 | seq_printf(m, "%s\n", trace_opts[i].name); |
| 3479 | else |
| 3480 | seq_printf(m, "no%s\n", trace_opts[i].name); |
| 3481 | } |
| 3482 | mutex_unlock(&trace_types_lock); |
| 3483 | |
| 3484 | return 0; |
| 3485 | } |
| 3486 | |
| 3487 | static int __set_tracer_option(struct trace_array *tr, |
| 3488 | struct tracer_flags *tracer_flags, |
| 3489 | struct tracer_opt *opts, int neg) |
| 3490 | { |
| 3491 | struct tracer *trace = tr->current_trace; |
| 3492 | int ret; |
| 3493 | |
| 3494 | ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg); |
| 3495 | if (ret) |
| 3496 | return ret; |
| 3497 | |
| 3498 | if (neg) |
| 3499 | tracer_flags->val &= ~opts->bit; |
| 3500 | else |
| 3501 | tracer_flags->val |= opts->bit; |
| 3502 | return 0; |
| 3503 | } |
| 3504 | |
| 3505 | /* Try to assign a tracer specific option */ |
| 3506 | static int set_tracer_option(struct trace_array *tr, char *cmp, int neg) |
| 3507 | { |
| 3508 | struct tracer *trace = tr->current_trace; |
| 3509 | struct tracer_flags *tracer_flags = trace->flags; |
| 3510 | struct tracer_opt *opts = NULL; |
| 3511 | int i; |
| 3512 | |
| 3513 | for (i = 0; tracer_flags->opts[i].name; i++) { |
| 3514 | opts = &tracer_flags->opts[i]; |
| 3515 | |
| 3516 | if (strcmp(cmp, opts->name) == 0) |
| 3517 | return __set_tracer_option(tr, trace->flags, opts, neg); |
| 3518 | } |
| 3519 | |
| 3520 | return -EINVAL; |
| 3521 | } |
| 3522 | |
| 3523 | /* Some tracers require overwrite to stay enabled */ |
| 3524 | int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set) |
| 3525 | { |
| 3526 | if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set) |
| 3527 | return -1; |
| 3528 | |
| 3529 | return 0; |
| 3530 | } |
| 3531 | |
| 3532 | int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled) |
| 3533 | { |
| 3534 | /* do nothing if flag is already set */ |
| 3535 | if (!!(trace_flags & mask) == !!enabled) |
| 3536 | return 0; |
| 3537 | |
| 3538 | /* Give the tracer a chance to approve the change */ |
| 3539 | if (tr->current_trace->flag_changed) |
| 3540 | if (tr->current_trace->flag_changed(tr, mask, !!enabled)) |
| 3541 | return -EINVAL; |
| 3542 | |
| 3543 | if (enabled) |
| 3544 | trace_flags |= mask; |
| 3545 | else |
| 3546 | trace_flags &= ~mask; |
| 3547 | |
| 3548 | if (mask == TRACE_ITER_RECORD_CMD) |
| 3549 | trace_event_enable_cmd_record(enabled); |
| 3550 | |
| 3551 | if (mask == TRACE_ITER_OVERWRITE) { |
| 3552 | ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled); |
| 3553 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 3554 | ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled); |
| 3555 | #endif |
| 3556 | } |
| 3557 | |
| 3558 | if (mask == TRACE_ITER_PRINTK) |
| 3559 | trace_printk_start_stop_comm(enabled); |
| 3560 | |
| 3561 | return 0; |
| 3562 | } |
| 3563 | |
| 3564 | static int trace_set_options(struct trace_array *tr, char *option) |
| 3565 | { |
| 3566 | char *cmp; |
| 3567 | int neg = 0; |
| 3568 | int ret = -ENODEV; |
| 3569 | int i; |
| 3570 | |
| 3571 | cmp = strstrip(option); |
| 3572 | |
| 3573 | if (strncmp(cmp, "no", 2) == 0) { |
| 3574 | neg = 1; |
| 3575 | cmp += 2; |
| 3576 | } |
| 3577 | |
| 3578 | mutex_lock(&trace_types_lock); |
| 3579 | |
| 3580 | for (i = 0; trace_options[i]; i++) { |
| 3581 | if (strcmp(cmp, trace_options[i]) == 0) { |
| 3582 | ret = set_tracer_flag(tr, 1 << i, !neg); |
| 3583 | break; |
| 3584 | } |
| 3585 | } |
| 3586 | |
| 3587 | /* If no option could be set, test the specific tracer options */ |
| 3588 | if (!trace_options[i]) |
| 3589 | ret = set_tracer_option(tr, cmp, neg); |
| 3590 | |
| 3591 | mutex_unlock(&trace_types_lock); |
| 3592 | |
| 3593 | return ret; |
| 3594 | } |
| 3595 | |
| 3596 | static ssize_t |
| 3597 | tracing_trace_options_write(struct file *filp, const char __user *ubuf, |
| 3598 | size_t cnt, loff_t *ppos) |
| 3599 | { |
| 3600 | struct seq_file *m = filp->private_data; |
| 3601 | struct trace_array *tr = m->private; |
| 3602 | char buf[64]; |
| 3603 | int ret; |
| 3604 | |
| 3605 | if (cnt >= sizeof(buf)) |
| 3606 | return -EINVAL; |
| 3607 | |
| 3608 | if (copy_from_user(&buf, ubuf, cnt)) |
| 3609 | return -EFAULT; |
| 3610 | |
| 3611 | buf[cnt] = 0; |
| 3612 | |
| 3613 | ret = trace_set_options(tr, buf); |
| 3614 | if (ret < 0) |
| 3615 | return ret; |
| 3616 | |
| 3617 | *ppos += cnt; |
| 3618 | |
| 3619 | return cnt; |
| 3620 | } |
| 3621 | |
| 3622 | static int tracing_trace_options_open(struct inode *inode, struct file *file) |
| 3623 | { |
| 3624 | struct trace_array *tr = inode->i_private; |
| 3625 | int ret; |
| 3626 | |
| 3627 | if (tracing_disabled) |
| 3628 | return -ENODEV; |
| 3629 | |
| 3630 | if (trace_array_get(tr) < 0) |
| 3631 | return -ENODEV; |
| 3632 | |
| 3633 | ret = single_open(file, tracing_trace_options_show, inode->i_private); |
| 3634 | if (ret < 0) |
| 3635 | trace_array_put(tr); |
| 3636 | |
| 3637 | return ret; |
| 3638 | } |
| 3639 | |
| 3640 | static const struct file_operations tracing_iter_fops = { |
| 3641 | .open = tracing_trace_options_open, |
| 3642 | .read = seq_read, |
| 3643 | .llseek = seq_lseek, |
| 3644 | .release = tracing_single_release_tr, |
| 3645 | .write = tracing_trace_options_write, |
| 3646 | }; |
| 3647 | |
| 3648 | static const char readme_msg[] = |
| 3649 | "tracing mini-HOWTO:\n\n" |
| 3650 | "# echo 0 > tracing_on : quick way to disable tracing\n" |
| 3651 | "# echo 1 > tracing_on : quick way to re-enable tracing\n\n" |
| 3652 | " Important files:\n" |
| 3653 | " trace\t\t\t- The static contents of the buffer\n" |
| 3654 | "\t\t\t To clear the buffer write into this file: echo > trace\n" |
| 3655 | " trace_pipe\t\t- A consuming read to see the contents of the buffer\n" |
| 3656 | " current_tracer\t- function and latency tracers\n" |
| 3657 | " available_tracers\t- list of configured tracers for current_tracer\n" |
| 3658 | " buffer_size_kb\t- view and modify size of per cpu buffer\n" |
| 3659 | " buffer_total_size_kb - view total size of all cpu buffers\n\n" |
| 3660 | " trace_clock\t\t-change the clock used to order events\n" |
| 3661 | " local: Per cpu clock but may not be synced across CPUs\n" |
| 3662 | " global: Synced across CPUs but slows tracing down.\n" |
| 3663 | " counter: Not a clock, but just an increment\n" |
| 3664 | " uptime: Jiffy counter from time of boot\n" |
| 3665 | " perf: Same clock that perf events use\n" |
| 3666 | #ifdef CONFIG_X86_64 |
| 3667 | " x86-tsc: TSC cycle counter\n" |
| 3668 | #endif |
| 3669 | "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n" |
| 3670 | " tracing_cpumask\t- Limit which CPUs to trace\n" |
| 3671 | " instances\t\t- Make sub-buffers with: mkdir instances/foo\n" |
| 3672 | "\t\t\t Remove sub-buffer with rmdir\n" |
| 3673 | " trace_options\t\t- Set format or modify how tracing happens\n" |
| 3674 | "\t\t\t Disable an option by adding a suffix 'no' to the\n" |
| 3675 | "\t\t\t option name\n" |
| 3676 | " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n" |
| 3677 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 3678 | "\n available_filter_functions - list of functions that can be filtered on\n" |
| 3679 | " set_ftrace_filter\t- echo function name in here to only trace these\n" |
| 3680 | "\t\t\t functions\n" |
| 3681 | "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n" |
| 3682 | "\t modules: Can select a group via module\n" |
| 3683 | "\t Format: :mod:<module-name>\n" |
| 3684 | "\t example: echo :mod:ext3 > set_ftrace_filter\n" |
| 3685 | "\t triggers: a command to perform when function is hit\n" |
| 3686 | "\t Format: <function>:<trigger>[:count]\n" |
| 3687 | "\t trigger: traceon, traceoff\n" |
| 3688 | "\t\t enable_event:<system>:<event>\n" |
| 3689 | "\t\t disable_event:<system>:<event>\n" |
| 3690 | #ifdef CONFIG_STACKTRACE |
| 3691 | "\t\t stacktrace\n" |
| 3692 | #endif |
| 3693 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 3694 | "\t\t snapshot\n" |
| 3695 | #endif |
| 3696 | "\t\t dump\n" |
| 3697 | "\t\t cpudump\n" |
| 3698 | "\t example: echo do_fault:traceoff > set_ftrace_filter\n" |
| 3699 | "\t echo do_trap:traceoff:3 > set_ftrace_filter\n" |
| 3700 | "\t The first one will disable tracing every time do_fault is hit\n" |
| 3701 | "\t The second will disable tracing at most 3 times when do_trap is hit\n" |
| 3702 | "\t The first time do trap is hit and it disables tracing, the\n" |
| 3703 | "\t counter will decrement to 2. If tracing is already disabled,\n" |
| 3704 | "\t the counter will not decrement. It only decrements when the\n" |
| 3705 | "\t trigger did work\n" |
| 3706 | "\t To remove trigger without count:\n" |
| 3707 | "\t echo '!<function>:<trigger> > set_ftrace_filter\n" |
| 3708 | "\t To remove trigger with a count:\n" |
| 3709 | "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n" |
| 3710 | " set_ftrace_notrace\t- echo function name in here to never trace.\n" |
| 3711 | "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n" |
| 3712 | "\t modules: Can select a group via module command :mod:\n" |
| 3713 | "\t Does not accept triggers\n" |
| 3714 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
| 3715 | #ifdef CONFIG_FUNCTION_TRACER |
| 3716 | " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n" |
| 3717 | "\t\t (function)\n" |
| 3718 | #endif |
| 3719 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
| 3720 | " set_graph_function\t- Trace the nested calls of a function (function_graph)\n" |
| 3721 | " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n" |
| 3722 | " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n" |
| 3723 | #endif |
| 3724 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 3725 | "\n snapshot\t\t- Like 'trace' but shows the content of the static\n" |
| 3726 | "\t\t\t snapshot buffer. Read the contents for more\n" |
| 3727 | "\t\t\t information\n" |
| 3728 | #endif |
| 3729 | #ifdef CONFIG_STACK_TRACER |
| 3730 | " stack_trace\t\t- Shows the max stack trace when active\n" |
| 3731 | " stack_max_size\t- Shows current max stack size that was traced\n" |
| 3732 | "\t\t\t Write into this file to reset the max size (trigger a\n" |
| 3733 | "\t\t\t new trace)\n" |
| 3734 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 3735 | " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n" |
| 3736 | "\t\t\t traces\n" |
| 3737 | #endif |
| 3738 | #endif /* CONFIG_STACK_TRACER */ |
| 3739 | " events/\t\t- Directory containing all trace event subsystems:\n" |
| 3740 | " enable\t\t- Write 0/1 to enable/disable tracing of all events\n" |
| 3741 | " events/<system>/\t- Directory containing all trace events for <system>:\n" |
| 3742 | " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n" |
| 3743 | "\t\t\t events\n" |
| 3744 | " filter\t\t- If set, only events passing filter are traced\n" |
| 3745 | " events/<system>/<event>/\t- Directory containing control files for\n" |
| 3746 | "\t\t\t <event>:\n" |
| 3747 | " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n" |
| 3748 | " filter\t\t- If set, only events passing filter are traced\n" |
| 3749 | " trigger\t\t- If set, a command to perform when event is hit\n" |
| 3750 | "\t Format: <trigger>[:count][if <filter>]\n" |
| 3751 | "\t trigger: traceon, traceoff\n" |
| 3752 | "\t enable_event:<system>:<event>\n" |
| 3753 | "\t disable_event:<system>:<event>\n" |
| 3754 | #ifdef CONFIG_STACKTRACE |
| 3755 | "\t\t stacktrace\n" |
| 3756 | #endif |
| 3757 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 3758 | "\t\t snapshot\n" |
| 3759 | #endif |
| 3760 | "\t example: echo traceoff > events/block/block_unplug/trigger\n" |
| 3761 | "\t echo traceoff:3 > events/block/block_unplug/trigger\n" |
| 3762 | "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n" |
| 3763 | "\t events/block/block_unplug/trigger\n" |
| 3764 | "\t The first disables tracing every time block_unplug is hit.\n" |
| 3765 | "\t The second disables tracing the first 3 times block_unplug is hit.\n" |
| 3766 | "\t The third enables the kmalloc event the first 3 times block_unplug\n" |
| 3767 | "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n" |
| 3768 | "\t Like function triggers, the counter is only decremented if it\n" |
| 3769 | "\t enabled or disabled tracing.\n" |
| 3770 | "\t To remove a trigger without a count:\n" |
| 3771 | "\t echo '!<trigger> > <system>/<event>/trigger\n" |
| 3772 | "\t To remove a trigger with a count:\n" |
| 3773 | "\t echo '!<trigger>:0 > <system>/<event>/trigger\n" |
| 3774 | "\t Filters can be ignored when removing a trigger.\n" |
| 3775 | ; |
| 3776 | |
| 3777 | static ssize_t |
| 3778 | tracing_readme_read(struct file *filp, char __user *ubuf, |
| 3779 | size_t cnt, loff_t *ppos) |
| 3780 | { |
| 3781 | return simple_read_from_buffer(ubuf, cnt, ppos, |
| 3782 | readme_msg, strlen(readme_msg)); |
| 3783 | } |
| 3784 | |
| 3785 | static const struct file_operations tracing_readme_fops = { |
| 3786 | .open = tracing_open_generic, |
| 3787 | .read = tracing_readme_read, |
| 3788 | .llseek = generic_file_llseek, |
| 3789 | }; |
| 3790 | |
| 3791 | static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos) |
| 3792 | { |
| 3793 | unsigned int *ptr = v; |
| 3794 | |
| 3795 | if (*pos || m->count) |
| 3796 | ptr++; |
| 3797 | |
| 3798 | (*pos)++; |
| 3799 | |
| 3800 | for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num]; |
| 3801 | ptr++) { |
| 3802 | if (*ptr == -1 || *ptr == NO_CMDLINE_MAP) |
| 3803 | continue; |
| 3804 | |
| 3805 | return ptr; |
| 3806 | } |
| 3807 | |
| 3808 | return NULL; |
| 3809 | } |
| 3810 | |
| 3811 | static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos) |
| 3812 | { |
| 3813 | void *v; |
| 3814 | loff_t l = 0; |
| 3815 | |
| 3816 | preempt_disable(); |
| 3817 | arch_spin_lock(&trace_cmdline_lock); |
| 3818 | |
| 3819 | v = &savedcmd->map_cmdline_to_pid[0]; |
| 3820 | while (l <= *pos) { |
| 3821 | v = saved_cmdlines_next(m, v, &l); |
| 3822 | if (!v) |
| 3823 | return NULL; |
| 3824 | } |
| 3825 | |
| 3826 | return v; |
| 3827 | } |
| 3828 | |
| 3829 | static void saved_cmdlines_stop(struct seq_file *m, void *v) |
| 3830 | { |
| 3831 | arch_spin_unlock(&trace_cmdline_lock); |
| 3832 | preempt_enable(); |
| 3833 | } |
| 3834 | |
| 3835 | static int saved_cmdlines_show(struct seq_file *m, void *v) |
| 3836 | { |
| 3837 | char buf[TASK_COMM_LEN]; |
| 3838 | unsigned int *pid = v; |
| 3839 | |
| 3840 | __trace_find_cmdline(*pid, buf); |
| 3841 | seq_printf(m, "%d %s\n", *pid, buf); |
| 3842 | return 0; |
| 3843 | } |
| 3844 | |
| 3845 | static const struct seq_operations tracing_saved_cmdlines_seq_ops = { |
| 3846 | .start = saved_cmdlines_start, |
| 3847 | .next = saved_cmdlines_next, |
| 3848 | .stop = saved_cmdlines_stop, |
| 3849 | .show = saved_cmdlines_show, |
| 3850 | }; |
| 3851 | |
| 3852 | static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp) |
| 3853 | { |
| 3854 | if (tracing_disabled) |
| 3855 | return -ENODEV; |
| 3856 | |
| 3857 | return seq_open(filp, &tracing_saved_cmdlines_seq_ops); |
| 3858 | } |
| 3859 | |
| 3860 | static const struct file_operations tracing_saved_cmdlines_fops = { |
| 3861 | .open = tracing_saved_cmdlines_open, |
| 3862 | .read = seq_read, |
| 3863 | .llseek = seq_lseek, |
| 3864 | .release = seq_release, |
| 3865 | }; |
| 3866 | |
| 3867 | static ssize_t |
| 3868 | tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf, |
| 3869 | size_t cnt, loff_t *ppos) |
| 3870 | { |
| 3871 | char buf[64]; |
| 3872 | int r; |
| 3873 | |
| 3874 | arch_spin_lock(&trace_cmdline_lock); |
| 3875 | r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num); |
| 3876 | arch_spin_unlock(&trace_cmdline_lock); |
| 3877 | |
| 3878 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 3879 | } |
| 3880 | |
| 3881 | static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s) |
| 3882 | { |
| 3883 | kfree(s->saved_cmdlines); |
| 3884 | kfree(s->map_cmdline_to_pid); |
| 3885 | kfree(s); |
| 3886 | } |
| 3887 | |
| 3888 | static int tracing_resize_saved_cmdlines(unsigned int val) |
| 3889 | { |
| 3890 | struct saved_cmdlines_buffer *s, *savedcmd_temp; |
| 3891 | |
| 3892 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
| 3893 | if (!s) |
| 3894 | return -ENOMEM; |
| 3895 | |
| 3896 | if (allocate_cmdlines_buffer(val, s) < 0) { |
| 3897 | kfree(s); |
| 3898 | return -ENOMEM; |
| 3899 | } |
| 3900 | |
| 3901 | arch_spin_lock(&trace_cmdline_lock); |
| 3902 | savedcmd_temp = savedcmd; |
| 3903 | savedcmd = s; |
| 3904 | arch_spin_unlock(&trace_cmdline_lock); |
| 3905 | free_saved_cmdlines_buffer(savedcmd_temp); |
| 3906 | |
| 3907 | return 0; |
| 3908 | } |
| 3909 | |
| 3910 | static ssize_t |
| 3911 | tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf, |
| 3912 | size_t cnt, loff_t *ppos) |
| 3913 | { |
| 3914 | unsigned long val; |
| 3915 | int ret; |
| 3916 | |
| 3917 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 3918 | if (ret) |
| 3919 | return ret; |
| 3920 | |
| 3921 | /* must have at least 1 entry or less than PID_MAX_DEFAULT */ |
| 3922 | if (!val || val > PID_MAX_DEFAULT) |
| 3923 | return -EINVAL; |
| 3924 | |
| 3925 | ret = tracing_resize_saved_cmdlines((unsigned int)val); |
| 3926 | if (ret < 0) |
| 3927 | return ret; |
| 3928 | |
| 3929 | *ppos += cnt; |
| 3930 | |
| 3931 | return cnt; |
| 3932 | } |
| 3933 | |
| 3934 | static const struct file_operations tracing_saved_cmdlines_size_fops = { |
| 3935 | .open = tracing_open_generic, |
| 3936 | .read = tracing_saved_cmdlines_size_read, |
| 3937 | .write = tracing_saved_cmdlines_size_write, |
| 3938 | }; |
| 3939 | |
| 3940 | #ifdef CONFIG_TRACE_ENUM_MAP_FILE |
| 3941 | static union trace_enum_map_item * |
| 3942 | update_enum_map(union trace_enum_map_item *ptr) |
| 3943 | { |
| 3944 | if (!ptr->map.enum_string) { |
| 3945 | if (ptr->tail.next) { |
| 3946 | ptr = ptr->tail.next; |
| 3947 | /* Set ptr to the next real item (skip head) */ |
| 3948 | ptr++; |
| 3949 | } else |
| 3950 | return NULL; |
| 3951 | } |
| 3952 | return ptr; |
| 3953 | } |
| 3954 | |
| 3955 | static void *enum_map_next(struct seq_file *m, void *v, loff_t *pos) |
| 3956 | { |
| 3957 | union trace_enum_map_item *ptr = v; |
| 3958 | |
| 3959 | /* |
| 3960 | * Paranoid! If ptr points to end, we don't want to increment past it. |
| 3961 | * This really should never happen. |
| 3962 | */ |
| 3963 | ptr = update_enum_map(ptr); |
| 3964 | if (WARN_ON_ONCE(!ptr)) |
| 3965 | return NULL; |
| 3966 | |
| 3967 | ptr++; |
| 3968 | |
| 3969 | (*pos)++; |
| 3970 | |
| 3971 | ptr = update_enum_map(ptr); |
| 3972 | |
| 3973 | return ptr; |
| 3974 | } |
| 3975 | |
| 3976 | static void *enum_map_start(struct seq_file *m, loff_t *pos) |
| 3977 | { |
| 3978 | union trace_enum_map_item *v; |
| 3979 | loff_t l = 0; |
| 3980 | |
| 3981 | mutex_lock(&trace_enum_mutex); |
| 3982 | |
| 3983 | v = trace_enum_maps; |
| 3984 | if (v) |
| 3985 | v++; |
| 3986 | |
| 3987 | while (v && l < *pos) { |
| 3988 | v = enum_map_next(m, v, &l); |
| 3989 | } |
| 3990 | |
| 3991 | return v; |
| 3992 | } |
| 3993 | |
| 3994 | static void enum_map_stop(struct seq_file *m, void *v) |
| 3995 | { |
| 3996 | mutex_unlock(&trace_enum_mutex); |
| 3997 | } |
| 3998 | |
| 3999 | static int enum_map_show(struct seq_file *m, void *v) |
| 4000 | { |
| 4001 | union trace_enum_map_item *ptr = v; |
| 4002 | |
| 4003 | seq_printf(m, "%s %ld (%s)\n", |
| 4004 | ptr->map.enum_string, ptr->map.enum_value, |
| 4005 | ptr->map.system); |
| 4006 | |
| 4007 | return 0; |
| 4008 | } |
| 4009 | |
| 4010 | static const struct seq_operations tracing_enum_map_seq_ops = { |
| 4011 | .start = enum_map_start, |
| 4012 | .next = enum_map_next, |
| 4013 | .stop = enum_map_stop, |
| 4014 | .show = enum_map_show, |
| 4015 | }; |
| 4016 | |
| 4017 | static int tracing_enum_map_open(struct inode *inode, struct file *filp) |
| 4018 | { |
| 4019 | if (tracing_disabled) |
| 4020 | return -ENODEV; |
| 4021 | |
| 4022 | return seq_open(filp, &tracing_enum_map_seq_ops); |
| 4023 | } |
| 4024 | |
| 4025 | static const struct file_operations tracing_enum_map_fops = { |
| 4026 | .open = tracing_enum_map_open, |
| 4027 | .read = seq_read, |
| 4028 | .llseek = seq_lseek, |
| 4029 | .release = seq_release, |
| 4030 | }; |
| 4031 | |
| 4032 | static inline union trace_enum_map_item * |
| 4033 | trace_enum_jmp_to_tail(union trace_enum_map_item *ptr) |
| 4034 | { |
| 4035 | /* Return tail of array given the head */ |
| 4036 | return ptr + ptr->head.length + 1; |
| 4037 | } |
| 4038 | |
| 4039 | static void |
| 4040 | trace_insert_enum_map_file(struct module *mod, struct trace_enum_map **start, |
| 4041 | int len) |
| 4042 | { |
| 4043 | struct trace_enum_map **stop; |
| 4044 | struct trace_enum_map **map; |
| 4045 | union trace_enum_map_item *map_array; |
| 4046 | union trace_enum_map_item *ptr; |
| 4047 | |
| 4048 | stop = start + len; |
| 4049 | |
| 4050 | /* |
| 4051 | * The trace_enum_maps contains the map plus a head and tail item, |
| 4052 | * where the head holds the module and length of array, and the |
| 4053 | * tail holds a pointer to the next list. |
| 4054 | */ |
| 4055 | map_array = kmalloc(sizeof(*map_array) * (len + 2), GFP_KERNEL); |
| 4056 | if (!map_array) { |
| 4057 | pr_warning("Unable to allocate trace enum mapping\n"); |
| 4058 | return; |
| 4059 | } |
| 4060 | |
| 4061 | mutex_lock(&trace_enum_mutex); |
| 4062 | |
| 4063 | if (!trace_enum_maps) |
| 4064 | trace_enum_maps = map_array; |
| 4065 | else { |
| 4066 | ptr = trace_enum_maps; |
| 4067 | for (;;) { |
| 4068 | ptr = trace_enum_jmp_to_tail(ptr); |
| 4069 | if (!ptr->tail.next) |
| 4070 | break; |
| 4071 | ptr = ptr->tail.next; |
| 4072 | |
| 4073 | } |
| 4074 | ptr->tail.next = map_array; |
| 4075 | } |
| 4076 | map_array->head.mod = mod; |
| 4077 | map_array->head.length = len; |
| 4078 | map_array++; |
| 4079 | |
| 4080 | for (map = start; (unsigned long)map < (unsigned long)stop; map++) { |
| 4081 | map_array->map = **map; |
| 4082 | map_array++; |
| 4083 | } |
| 4084 | memset(map_array, 0, sizeof(*map_array)); |
| 4085 | |
| 4086 | mutex_unlock(&trace_enum_mutex); |
| 4087 | } |
| 4088 | |
| 4089 | static void trace_create_enum_file(struct dentry *d_tracer) |
| 4090 | { |
| 4091 | trace_create_file("enum_map", 0444, d_tracer, |
| 4092 | NULL, &tracing_enum_map_fops); |
| 4093 | } |
| 4094 | |
| 4095 | #else /* CONFIG_TRACE_ENUM_MAP_FILE */ |
| 4096 | static inline void trace_create_enum_file(struct dentry *d_tracer) { } |
| 4097 | static inline void trace_insert_enum_map_file(struct module *mod, |
| 4098 | struct trace_enum_map **start, int len) { } |
| 4099 | #endif /* !CONFIG_TRACE_ENUM_MAP_FILE */ |
| 4100 | |
| 4101 | static void trace_insert_enum_map(struct module *mod, |
| 4102 | struct trace_enum_map **start, int len) |
| 4103 | { |
| 4104 | struct trace_enum_map **map; |
| 4105 | |
| 4106 | if (len <= 0) |
| 4107 | return; |
| 4108 | |
| 4109 | map = start; |
| 4110 | |
| 4111 | trace_event_enum_update(map, len); |
| 4112 | |
| 4113 | trace_insert_enum_map_file(mod, start, len); |
| 4114 | } |
| 4115 | |
| 4116 | static ssize_t |
| 4117 | tracing_set_trace_read(struct file *filp, char __user *ubuf, |
| 4118 | size_t cnt, loff_t *ppos) |
| 4119 | { |
| 4120 | struct trace_array *tr = filp->private_data; |
| 4121 | char buf[MAX_TRACER_SIZE+2]; |
| 4122 | int r; |
| 4123 | |
| 4124 | mutex_lock(&trace_types_lock); |
| 4125 | r = sprintf(buf, "%s\n", tr->current_trace->name); |
| 4126 | mutex_unlock(&trace_types_lock); |
| 4127 | |
| 4128 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 4129 | } |
| 4130 | |
| 4131 | int tracer_init(struct tracer *t, struct trace_array *tr) |
| 4132 | { |
| 4133 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 4134 | return t->init(tr); |
| 4135 | } |
| 4136 | |
| 4137 | static void set_buffer_entries(struct trace_buffer *buf, unsigned long val) |
| 4138 | { |
| 4139 | int cpu; |
| 4140 | |
| 4141 | for_each_tracing_cpu(cpu) |
| 4142 | per_cpu_ptr(buf->data, cpu)->entries = val; |
| 4143 | } |
| 4144 | |
| 4145 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 4146 | /* resize @tr's buffer to the size of @size_tr's entries */ |
| 4147 | static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf, |
| 4148 | struct trace_buffer *size_buf, int cpu_id) |
| 4149 | { |
| 4150 | int cpu, ret = 0; |
| 4151 | |
| 4152 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
| 4153 | for_each_tracing_cpu(cpu) { |
| 4154 | ret = ring_buffer_resize(trace_buf->buffer, |
| 4155 | per_cpu_ptr(size_buf->data, cpu)->entries, cpu); |
| 4156 | if (ret < 0) |
| 4157 | break; |
| 4158 | per_cpu_ptr(trace_buf->data, cpu)->entries = |
| 4159 | per_cpu_ptr(size_buf->data, cpu)->entries; |
| 4160 | } |
| 4161 | } else { |
| 4162 | ret = ring_buffer_resize(trace_buf->buffer, |
| 4163 | per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id); |
| 4164 | if (ret == 0) |
| 4165 | per_cpu_ptr(trace_buf->data, cpu_id)->entries = |
| 4166 | per_cpu_ptr(size_buf->data, cpu_id)->entries; |
| 4167 | } |
| 4168 | |
| 4169 | return ret; |
| 4170 | } |
| 4171 | #endif /* CONFIG_TRACER_MAX_TRACE */ |
| 4172 | |
| 4173 | static int __tracing_resize_ring_buffer(struct trace_array *tr, |
| 4174 | unsigned long size, int cpu) |
| 4175 | { |
| 4176 | int ret; |
| 4177 | |
| 4178 | /* |
| 4179 | * If kernel or user changes the size of the ring buffer |
| 4180 | * we use the size that was given, and we can forget about |
| 4181 | * expanding it later. |
| 4182 | */ |
| 4183 | ring_buffer_expanded = true; |
| 4184 | |
| 4185 | /* May be called before buffers are initialized */ |
| 4186 | if (!tr->trace_buffer.buffer) |
| 4187 | return 0; |
| 4188 | |
| 4189 | ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu); |
| 4190 | if (ret < 0) |
| 4191 | return ret; |
| 4192 | |
| 4193 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 4194 | if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) || |
| 4195 | !tr->current_trace->use_max_tr) |
| 4196 | goto out; |
| 4197 | |
| 4198 | ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu); |
| 4199 | if (ret < 0) { |
| 4200 | int r = resize_buffer_duplicate_size(&tr->trace_buffer, |
| 4201 | &tr->trace_buffer, cpu); |
| 4202 | if (r < 0) { |
| 4203 | /* |
| 4204 | * AARGH! We are left with different |
| 4205 | * size max buffer!!!! |
| 4206 | * The max buffer is our "snapshot" buffer. |
| 4207 | * When a tracer needs a snapshot (one of the |
| 4208 | * latency tracers), it swaps the max buffer |
| 4209 | * with the saved snap shot. We succeeded to |
| 4210 | * update the size of the main buffer, but failed to |
| 4211 | * update the size of the max buffer. But when we tried |
| 4212 | * to reset the main buffer to the original size, we |
| 4213 | * failed there too. This is very unlikely to |
| 4214 | * happen, but if it does, warn and kill all |
| 4215 | * tracing. |
| 4216 | */ |
| 4217 | WARN_ON(1); |
| 4218 | tracing_disabled = 1; |
| 4219 | } |
| 4220 | return ret; |
| 4221 | } |
| 4222 | |
| 4223 | if (cpu == RING_BUFFER_ALL_CPUS) |
| 4224 | set_buffer_entries(&tr->max_buffer, size); |
| 4225 | else |
| 4226 | per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size; |
| 4227 | |
| 4228 | out: |
| 4229 | #endif /* CONFIG_TRACER_MAX_TRACE */ |
| 4230 | |
| 4231 | if (cpu == RING_BUFFER_ALL_CPUS) |
| 4232 | set_buffer_entries(&tr->trace_buffer, size); |
| 4233 | else |
| 4234 | per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size; |
| 4235 | |
| 4236 | return ret; |
| 4237 | } |
| 4238 | |
| 4239 | static ssize_t tracing_resize_ring_buffer(struct trace_array *tr, |
| 4240 | unsigned long size, int cpu_id) |
| 4241 | { |
| 4242 | int ret = size; |
| 4243 | |
| 4244 | mutex_lock(&trace_types_lock); |
| 4245 | |
| 4246 | if (cpu_id != RING_BUFFER_ALL_CPUS) { |
| 4247 | /* make sure, this cpu is enabled in the mask */ |
| 4248 | if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) { |
| 4249 | ret = -EINVAL; |
| 4250 | goto out; |
| 4251 | } |
| 4252 | } |
| 4253 | |
| 4254 | ret = __tracing_resize_ring_buffer(tr, size, cpu_id); |
| 4255 | if (ret < 0) |
| 4256 | ret = -ENOMEM; |
| 4257 | |
| 4258 | out: |
| 4259 | mutex_unlock(&trace_types_lock); |
| 4260 | |
| 4261 | return ret; |
| 4262 | } |
| 4263 | |
| 4264 | |
| 4265 | /** |
| 4266 | * tracing_update_buffers - used by tracing facility to expand ring buffers |
| 4267 | * |
| 4268 | * To save on memory when the tracing is never used on a system with it |
| 4269 | * configured in. The ring buffers are set to a minimum size. But once |
| 4270 | * a user starts to use the tracing facility, then they need to grow |
| 4271 | * to their default size. |
| 4272 | * |
| 4273 | * This function is to be called when a tracer is about to be used. |
| 4274 | */ |
| 4275 | int tracing_update_buffers(void) |
| 4276 | { |
| 4277 | int ret = 0; |
| 4278 | |
| 4279 | mutex_lock(&trace_types_lock); |
| 4280 | if (!ring_buffer_expanded) |
| 4281 | ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size, |
| 4282 | RING_BUFFER_ALL_CPUS); |
| 4283 | mutex_unlock(&trace_types_lock); |
| 4284 | |
| 4285 | return ret; |
| 4286 | } |
| 4287 | |
| 4288 | struct trace_option_dentry; |
| 4289 | |
| 4290 | static struct trace_option_dentry * |
| 4291 | create_trace_option_files(struct trace_array *tr, struct tracer *tracer); |
| 4292 | |
| 4293 | /* |
| 4294 | * Used to clear out the tracer before deletion of an instance. |
| 4295 | * Must have trace_types_lock held. |
| 4296 | */ |
| 4297 | static void tracing_set_nop(struct trace_array *tr) |
| 4298 | { |
| 4299 | if (tr->current_trace == &nop_trace) |
| 4300 | return; |
| 4301 | |
| 4302 | tr->current_trace->enabled--; |
| 4303 | |
| 4304 | if (tr->current_trace->reset) |
| 4305 | tr->current_trace->reset(tr); |
| 4306 | |
| 4307 | tr->current_trace = &nop_trace; |
| 4308 | } |
| 4309 | |
| 4310 | static void add_tracer_options(struct trace_array *tr, struct tracer *t) |
| 4311 | { |
| 4312 | /* Only enable if the directory has been created already. */ |
| 4313 | if (!tr->dir) |
| 4314 | return; |
| 4315 | |
| 4316 | /* Currently, only the top instance has options */ |
| 4317 | if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL)) |
| 4318 | return; |
| 4319 | |
| 4320 | /* Ignore if they were already created */ |
| 4321 | if (t->topts) |
| 4322 | return; |
| 4323 | |
| 4324 | t->topts = create_trace_option_files(tr, t); |
| 4325 | } |
| 4326 | |
| 4327 | static int tracing_set_tracer(struct trace_array *tr, const char *buf) |
| 4328 | { |
| 4329 | struct tracer *t; |
| 4330 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 4331 | bool had_max_tr; |
| 4332 | #endif |
| 4333 | int ret = 0; |
| 4334 | |
| 4335 | mutex_lock(&trace_types_lock); |
| 4336 | |
| 4337 | if (!ring_buffer_expanded) { |
| 4338 | ret = __tracing_resize_ring_buffer(tr, trace_buf_size, |
| 4339 | RING_BUFFER_ALL_CPUS); |
| 4340 | if (ret < 0) |
| 4341 | goto out; |
| 4342 | ret = 0; |
| 4343 | } |
| 4344 | |
| 4345 | for (t = trace_types; t; t = t->next) { |
| 4346 | if (strcmp(t->name, buf) == 0) |
| 4347 | break; |
| 4348 | } |
| 4349 | if (!t) { |
| 4350 | ret = -EINVAL; |
| 4351 | goto out; |
| 4352 | } |
| 4353 | if (t == tr->current_trace) |
| 4354 | goto out; |
| 4355 | |
| 4356 | /* Some tracers are only allowed for the top level buffer */ |
| 4357 | if (!trace_ok_for_array(t, tr)) { |
| 4358 | ret = -EINVAL; |
| 4359 | goto out; |
| 4360 | } |
| 4361 | |
| 4362 | /* If trace pipe files are being read, we can't change the tracer */ |
| 4363 | if (tr->current_trace->ref) { |
| 4364 | ret = -EBUSY; |
| 4365 | goto out; |
| 4366 | } |
| 4367 | |
| 4368 | trace_branch_disable(); |
| 4369 | |
| 4370 | tr->current_trace->enabled--; |
| 4371 | |
| 4372 | if (tr->current_trace->reset) |
| 4373 | tr->current_trace->reset(tr); |
| 4374 | |
| 4375 | /* Current trace needs to be nop_trace before synchronize_sched */ |
| 4376 | tr->current_trace = &nop_trace; |
| 4377 | |
| 4378 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 4379 | had_max_tr = tr->allocated_snapshot; |
| 4380 | |
| 4381 | if (had_max_tr && !t->use_max_tr) { |
| 4382 | /* |
| 4383 | * We need to make sure that the update_max_tr sees that |
| 4384 | * current_trace changed to nop_trace to keep it from |
| 4385 | * swapping the buffers after we resize it. |
| 4386 | * The update_max_tr is called from interrupts disabled |
| 4387 | * so a synchronized_sched() is sufficient. |
| 4388 | */ |
| 4389 | synchronize_sched(); |
| 4390 | free_snapshot(tr); |
| 4391 | } |
| 4392 | #endif |
| 4393 | |
| 4394 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 4395 | if (t->use_max_tr && !had_max_tr) { |
| 4396 | ret = alloc_snapshot(tr); |
| 4397 | if (ret < 0) |
| 4398 | goto out; |
| 4399 | } |
| 4400 | #endif |
| 4401 | |
| 4402 | if (t->init) { |
| 4403 | ret = tracer_init(t, tr); |
| 4404 | if (ret) |
| 4405 | goto out; |
| 4406 | } |
| 4407 | |
| 4408 | tr->current_trace = t; |
| 4409 | tr->current_trace->enabled++; |
| 4410 | trace_branch_enable(tr); |
| 4411 | out: |
| 4412 | mutex_unlock(&trace_types_lock); |
| 4413 | |
| 4414 | return ret; |
| 4415 | } |
| 4416 | |
| 4417 | static ssize_t |
| 4418 | tracing_set_trace_write(struct file *filp, const char __user *ubuf, |
| 4419 | size_t cnt, loff_t *ppos) |
| 4420 | { |
| 4421 | struct trace_array *tr = filp->private_data; |
| 4422 | char buf[MAX_TRACER_SIZE+1]; |
| 4423 | int i; |
| 4424 | size_t ret; |
| 4425 | int err; |
| 4426 | |
| 4427 | ret = cnt; |
| 4428 | |
| 4429 | if (cnt > MAX_TRACER_SIZE) |
| 4430 | cnt = MAX_TRACER_SIZE; |
| 4431 | |
| 4432 | if (copy_from_user(&buf, ubuf, cnt)) |
| 4433 | return -EFAULT; |
| 4434 | |
| 4435 | buf[cnt] = 0; |
| 4436 | |
| 4437 | /* strip ending whitespace. */ |
| 4438 | for (i = cnt - 1; i > 0 && isspace(buf[i]); i--) |
| 4439 | buf[i] = 0; |
| 4440 | |
| 4441 | err = tracing_set_tracer(tr, buf); |
| 4442 | if (err) |
| 4443 | return err; |
| 4444 | |
| 4445 | *ppos += ret; |
| 4446 | |
| 4447 | return ret; |
| 4448 | } |
| 4449 | |
| 4450 | static ssize_t |
| 4451 | tracing_nsecs_read(unsigned long *ptr, char __user *ubuf, |
| 4452 | size_t cnt, loff_t *ppos) |
| 4453 | { |
| 4454 | char buf[64]; |
| 4455 | int r; |
| 4456 | |
| 4457 | r = snprintf(buf, sizeof(buf), "%ld\n", |
| 4458 | *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr)); |
| 4459 | if (r > sizeof(buf)) |
| 4460 | r = sizeof(buf); |
| 4461 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 4462 | } |
| 4463 | |
| 4464 | static ssize_t |
| 4465 | tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf, |
| 4466 | size_t cnt, loff_t *ppos) |
| 4467 | { |
| 4468 | unsigned long val; |
| 4469 | int ret; |
| 4470 | |
| 4471 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 4472 | if (ret) |
| 4473 | return ret; |
| 4474 | |
| 4475 | *ptr = val * 1000; |
| 4476 | |
| 4477 | return cnt; |
| 4478 | } |
| 4479 | |
| 4480 | static ssize_t |
| 4481 | tracing_thresh_read(struct file *filp, char __user *ubuf, |
| 4482 | size_t cnt, loff_t *ppos) |
| 4483 | { |
| 4484 | return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos); |
| 4485 | } |
| 4486 | |
| 4487 | static ssize_t |
| 4488 | tracing_thresh_write(struct file *filp, const char __user *ubuf, |
| 4489 | size_t cnt, loff_t *ppos) |
| 4490 | { |
| 4491 | struct trace_array *tr = filp->private_data; |
| 4492 | int ret; |
| 4493 | |
| 4494 | mutex_lock(&trace_types_lock); |
| 4495 | ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos); |
| 4496 | if (ret < 0) |
| 4497 | goto out; |
| 4498 | |
| 4499 | if (tr->current_trace->update_thresh) { |
| 4500 | ret = tr->current_trace->update_thresh(tr); |
| 4501 | if (ret < 0) |
| 4502 | goto out; |
| 4503 | } |
| 4504 | |
| 4505 | ret = cnt; |
| 4506 | out: |
| 4507 | mutex_unlock(&trace_types_lock); |
| 4508 | |
| 4509 | return ret; |
| 4510 | } |
| 4511 | |
| 4512 | static ssize_t |
| 4513 | tracing_max_lat_read(struct file *filp, char __user *ubuf, |
| 4514 | size_t cnt, loff_t *ppos) |
| 4515 | { |
| 4516 | return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos); |
| 4517 | } |
| 4518 | |
| 4519 | static ssize_t |
| 4520 | tracing_max_lat_write(struct file *filp, const char __user *ubuf, |
| 4521 | size_t cnt, loff_t *ppos) |
| 4522 | { |
| 4523 | return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos); |
| 4524 | } |
| 4525 | |
| 4526 | static int tracing_open_pipe(struct inode *inode, struct file *filp) |
| 4527 | { |
| 4528 | struct trace_array *tr = inode->i_private; |
| 4529 | struct trace_iterator *iter; |
| 4530 | int ret = 0; |
| 4531 | |
| 4532 | if (tracing_disabled) |
| 4533 | return -ENODEV; |
| 4534 | |
| 4535 | if (trace_array_get(tr) < 0) |
| 4536 | return -ENODEV; |
| 4537 | |
| 4538 | mutex_lock(&trace_types_lock); |
| 4539 | |
| 4540 | /* create a buffer to store the information to pass to userspace */ |
| 4541 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); |
| 4542 | if (!iter) { |
| 4543 | ret = -ENOMEM; |
| 4544 | __trace_array_put(tr); |
| 4545 | goto out; |
| 4546 | } |
| 4547 | |
| 4548 | trace_seq_init(&iter->seq); |
| 4549 | iter->trace = tr->current_trace; |
| 4550 | |
| 4551 | if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) { |
| 4552 | ret = -ENOMEM; |
| 4553 | goto fail; |
| 4554 | } |
| 4555 | |
| 4556 | /* trace pipe does not show start of buffer */ |
| 4557 | cpumask_setall(iter->started); |
| 4558 | |
| 4559 | if (trace_flags & TRACE_ITER_LATENCY_FMT) |
| 4560 | iter->iter_flags |= TRACE_FILE_LAT_FMT; |
| 4561 | |
| 4562 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ |
| 4563 | if (trace_clocks[tr->clock_id].in_ns) |
| 4564 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; |
| 4565 | |
| 4566 | iter->tr = tr; |
| 4567 | iter->trace_buffer = &tr->trace_buffer; |
| 4568 | iter->cpu_file = tracing_get_cpu(inode); |
| 4569 | mutex_init(&iter->mutex); |
| 4570 | filp->private_data = iter; |
| 4571 | |
| 4572 | if (iter->trace->pipe_open) |
| 4573 | iter->trace->pipe_open(iter); |
| 4574 | |
| 4575 | nonseekable_open(inode, filp); |
| 4576 | |
| 4577 | tr->current_trace->ref++; |
| 4578 | out: |
| 4579 | mutex_unlock(&trace_types_lock); |
| 4580 | return ret; |
| 4581 | |
| 4582 | fail: |
| 4583 | kfree(iter->trace); |
| 4584 | kfree(iter); |
| 4585 | __trace_array_put(tr); |
| 4586 | mutex_unlock(&trace_types_lock); |
| 4587 | return ret; |
| 4588 | } |
| 4589 | |
| 4590 | static int tracing_release_pipe(struct inode *inode, struct file *file) |
| 4591 | { |
| 4592 | struct trace_iterator *iter = file->private_data; |
| 4593 | struct trace_array *tr = inode->i_private; |
| 4594 | |
| 4595 | mutex_lock(&trace_types_lock); |
| 4596 | |
| 4597 | tr->current_trace->ref--; |
| 4598 | |
| 4599 | if (iter->trace->pipe_close) |
| 4600 | iter->trace->pipe_close(iter); |
| 4601 | |
| 4602 | mutex_unlock(&trace_types_lock); |
| 4603 | |
| 4604 | free_cpumask_var(iter->started); |
| 4605 | mutex_destroy(&iter->mutex); |
| 4606 | kfree(iter); |
| 4607 | |
| 4608 | trace_array_put(tr); |
| 4609 | |
| 4610 | return 0; |
| 4611 | } |
| 4612 | |
| 4613 | static unsigned int |
| 4614 | trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table) |
| 4615 | { |
| 4616 | /* Iterators are static, they should be filled or empty */ |
| 4617 | if (trace_buffer_iter(iter, iter->cpu_file)) |
| 4618 | return POLLIN | POLLRDNORM; |
| 4619 | |
| 4620 | if (trace_flags & TRACE_ITER_BLOCK) |
| 4621 | /* |
| 4622 | * Always select as readable when in blocking mode |
| 4623 | */ |
| 4624 | return POLLIN | POLLRDNORM; |
| 4625 | else |
| 4626 | return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file, |
| 4627 | filp, poll_table); |
| 4628 | } |
| 4629 | |
| 4630 | static unsigned int |
| 4631 | tracing_poll_pipe(struct file *filp, poll_table *poll_table) |
| 4632 | { |
| 4633 | struct trace_iterator *iter = filp->private_data; |
| 4634 | |
| 4635 | return trace_poll(iter, filp, poll_table); |
| 4636 | } |
| 4637 | |
| 4638 | /* Must be called with iter->mutex held. */ |
| 4639 | static int tracing_wait_pipe(struct file *filp) |
| 4640 | { |
| 4641 | struct trace_iterator *iter = filp->private_data; |
| 4642 | int ret; |
| 4643 | |
| 4644 | while (trace_empty(iter)) { |
| 4645 | |
| 4646 | if ((filp->f_flags & O_NONBLOCK)) { |
| 4647 | return -EAGAIN; |
| 4648 | } |
| 4649 | |
| 4650 | /* |
| 4651 | * We block until we read something and tracing is disabled. |
| 4652 | * We still block if tracing is disabled, but we have never |
| 4653 | * read anything. This allows a user to cat this file, and |
| 4654 | * then enable tracing. But after we have read something, |
| 4655 | * we give an EOF when tracing is again disabled. |
| 4656 | * |
| 4657 | * iter->pos will be 0 if we haven't read anything. |
| 4658 | */ |
| 4659 | if (!tracing_is_on() && iter->pos) |
| 4660 | break; |
| 4661 | |
| 4662 | mutex_unlock(&iter->mutex); |
| 4663 | |
| 4664 | ret = wait_on_pipe(iter, false); |
| 4665 | |
| 4666 | mutex_lock(&iter->mutex); |
| 4667 | |
| 4668 | if (ret) |
| 4669 | return ret; |
| 4670 | } |
| 4671 | |
| 4672 | return 1; |
| 4673 | } |
| 4674 | |
| 4675 | /* |
| 4676 | * Consumer reader. |
| 4677 | */ |
| 4678 | static ssize_t |
| 4679 | tracing_read_pipe(struct file *filp, char __user *ubuf, |
| 4680 | size_t cnt, loff_t *ppos) |
| 4681 | { |
| 4682 | struct trace_iterator *iter = filp->private_data; |
| 4683 | ssize_t sret; |
| 4684 | |
| 4685 | /* return any leftover data */ |
| 4686 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); |
| 4687 | if (sret != -EBUSY) |
| 4688 | return sret; |
| 4689 | |
| 4690 | trace_seq_init(&iter->seq); |
| 4691 | |
| 4692 | /* |
| 4693 | * Avoid more than one consumer on a single file descriptor |
| 4694 | * This is just a matter of traces coherency, the ring buffer itself |
| 4695 | * is protected. |
| 4696 | */ |
| 4697 | mutex_lock(&iter->mutex); |
| 4698 | if (iter->trace->read) { |
| 4699 | sret = iter->trace->read(iter, filp, ubuf, cnt, ppos); |
| 4700 | if (sret) |
| 4701 | goto out; |
| 4702 | } |
| 4703 | |
| 4704 | waitagain: |
| 4705 | sret = tracing_wait_pipe(filp); |
| 4706 | if (sret <= 0) |
| 4707 | goto out; |
| 4708 | |
| 4709 | /* stop when tracing is finished */ |
| 4710 | if (trace_empty(iter)) { |
| 4711 | sret = 0; |
| 4712 | goto out; |
| 4713 | } |
| 4714 | |
| 4715 | if (cnt >= PAGE_SIZE) |
| 4716 | cnt = PAGE_SIZE - 1; |
| 4717 | |
| 4718 | /* reset all but tr, trace, and overruns */ |
| 4719 | memset(&iter->seq, 0, |
| 4720 | sizeof(struct trace_iterator) - |
| 4721 | offsetof(struct trace_iterator, seq)); |
| 4722 | cpumask_clear(iter->started); |
| 4723 | iter->pos = -1; |
| 4724 | |
| 4725 | trace_event_read_lock(); |
| 4726 | trace_access_lock(iter->cpu_file); |
| 4727 | while (trace_find_next_entry_inc(iter) != NULL) { |
| 4728 | enum print_line_t ret; |
| 4729 | int save_len = iter->seq.seq.len; |
| 4730 | |
| 4731 | ret = print_trace_line(iter); |
| 4732 | if (ret == TRACE_TYPE_PARTIAL_LINE) { |
| 4733 | /* don't print partial lines */ |
| 4734 | iter->seq.seq.len = save_len; |
| 4735 | break; |
| 4736 | } |
| 4737 | if (ret != TRACE_TYPE_NO_CONSUME) |
| 4738 | trace_consume(iter); |
| 4739 | |
| 4740 | if (trace_seq_used(&iter->seq) >= cnt) |
| 4741 | break; |
| 4742 | |
| 4743 | /* |
| 4744 | * Setting the full flag means we reached the trace_seq buffer |
| 4745 | * size and we should leave by partial output condition above. |
| 4746 | * One of the trace_seq_* functions is not used properly. |
| 4747 | */ |
| 4748 | WARN_ONCE(iter->seq.full, "full flag set for trace type %d", |
| 4749 | iter->ent->type); |
| 4750 | } |
| 4751 | trace_access_unlock(iter->cpu_file); |
| 4752 | trace_event_read_unlock(); |
| 4753 | |
| 4754 | /* Now copy what we have to the user */ |
| 4755 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); |
| 4756 | if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq)) |
| 4757 | trace_seq_init(&iter->seq); |
| 4758 | |
| 4759 | /* |
| 4760 | * If there was nothing to send to user, in spite of consuming trace |
| 4761 | * entries, go back to wait for more entries. |
| 4762 | */ |
| 4763 | if (sret == -EBUSY) |
| 4764 | goto waitagain; |
| 4765 | |
| 4766 | out: |
| 4767 | mutex_unlock(&iter->mutex); |
| 4768 | |
| 4769 | return sret; |
| 4770 | } |
| 4771 | |
| 4772 | static void tracing_spd_release_pipe(struct splice_pipe_desc *spd, |
| 4773 | unsigned int idx) |
| 4774 | { |
| 4775 | __free_page(spd->pages[idx]); |
| 4776 | } |
| 4777 | |
| 4778 | static const struct pipe_buf_operations tracing_pipe_buf_ops = { |
| 4779 | .can_merge = 0, |
| 4780 | .confirm = generic_pipe_buf_confirm, |
| 4781 | .release = generic_pipe_buf_release, |
| 4782 | .steal = generic_pipe_buf_steal, |
| 4783 | .get = generic_pipe_buf_get, |
| 4784 | }; |
| 4785 | |
| 4786 | static size_t |
| 4787 | tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter) |
| 4788 | { |
| 4789 | size_t count; |
| 4790 | int save_len; |
| 4791 | int ret; |
| 4792 | |
| 4793 | /* Seq buffer is page-sized, exactly what we need. */ |
| 4794 | for (;;) { |
| 4795 | save_len = iter->seq.seq.len; |
| 4796 | ret = print_trace_line(iter); |
| 4797 | |
| 4798 | if (trace_seq_has_overflowed(&iter->seq)) { |
| 4799 | iter->seq.seq.len = save_len; |
| 4800 | break; |
| 4801 | } |
| 4802 | |
| 4803 | /* |
| 4804 | * This should not be hit, because it should only |
| 4805 | * be set if the iter->seq overflowed. But check it |
| 4806 | * anyway to be safe. |
| 4807 | */ |
| 4808 | if (ret == TRACE_TYPE_PARTIAL_LINE) { |
| 4809 | iter->seq.seq.len = save_len; |
| 4810 | break; |
| 4811 | } |
| 4812 | |
| 4813 | count = trace_seq_used(&iter->seq) - save_len; |
| 4814 | if (rem < count) { |
| 4815 | rem = 0; |
| 4816 | iter->seq.seq.len = save_len; |
| 4817 | break; |
| 4818 | } |
| 4819 | |
| 4820 | if (ret != TRACE_TYPE_NO_CONSUME) |
| 4821 | trace_consume(iter); |
| 4822 | rem -= count; |
| 4823 | if (!trace_find_next_entry_inc(iter)) { |
| 4824 | rem = 0; |
| 4825 | iter->ent = NULL; |
| 4826 | break; |
| 4827 | } |
| 4828 | } |
| 4829 | |
| 4830 | return rem; |
| 4831 | } |
| 4832 | |
| 4833 | static ssize_t tracing_splice_read_pipe(struct file *filp, |
| 4834 | loff_t *ppos, |
| 4835 | struct pipe_inode_info *pipe, |
| 4836 | size_t len, |
| 4837 | unsigned int flags) |
| 4838 | { |
| 4839 | struct page *pages_def[PIPE_DEF_BUFFERS]; |
| 4840 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; |
| 4841 | struct trace_iterator *iter = filp->private_data; |
| 4842 | struct splice_pipe_desc spd = { |
| 4843 | .pages = pages_def, |
| 4844 | .partial = partial_def, |
| 4845 | .nr_pages = 0, /* This gets updated below. */ |
| 4846 | .nr_pages_max = PIPE_DEF_BUFFERS, |
| 4847 | .flags = flags, |
| 4848 | .ops = &tracing_pipe_buf_ops, |
| 4849 | .spd_release = tracing_spd_release_pipe, |
| 4850 | }; |
| 4851 | ssize_t ret; |
| 4852 | size_t rem; |
| 4853 | unsigned int i; |
| 4854 | |
| 4855 | if (splice_grow_spd(pipe, &spd)) |
| 4856 | return -ENOMEM; |
| 4857 | |
| 4858 | mutex_lock(&iter->mutex); |
| 4859 | |
| 4860 | if (iter->trace->splice_read) { |
| 4861 | ret = iter->trace->splice_read(iter, filp, |
| 4862 | ppos, pipe, len, flags); |
| 4863 | if (ret) |
| 4864 | goto out_err; |
| 4865 | } |
| 4866 | |
| 4867 | ret = tracing_wait_pipe(filp); |
| 4868 | if (ret <= 0) |
| 4869 | goto out_err; |
| 4870 | |
| 4871 | if (!iter->ent && !trace_find_next_entry_inc(iter)) { |
| 4872 | ret = -EFAULT; |
| 4873 | goto out_err; |
| 4874 | } |
| 4875 | |
| 4876 | trace_event_read_lock(); |
| 4877 | trace_access_lock(iter->cpu_file); |
| 4878 | |
| 4879 | /* Fill as many pages as possible. */ |
| 4880 | for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) { |
| 4881 | spd.pages[i] = alloc_page(GFP_KERNEL); |
| 4882 | if (!spd.pages[i]) |
| 4883 | break; |
| 4884 | |
| 4885 | rem = tracing_fill_pipe_page(rem, iter); |
| 4886 | |
| 4887 | /* Copy the data into the page, so we can start over. */ |
| 4888 | ret = trace_seq_to_buffer(&iter->seq, |
| 4889 | page_address(spd.pages[i]), |
| 4890 | trace_seq_used(&iter->seq)); |
| 4891 | if (ret < 0) { |
| 4892 | __free_page(spd.pages[i]); |
| 4893 | break; |
| 4894 | } |
| 4895 | spd.partial[i].offset = 0; |
| 4896 | spd.partial[i].len = trace_seq_used(&iter->seq); |
| 4897 | |
| 4898 | trace_seq_init(&iter->seq); |
| 4899 | } |
| 4900 | |
| 4901 | trace_access_unlock(iter->cpu_file); |
| 4902 | trace_event_read_unlock(); |
| 4903 | mutex_unlock(&iter->mutex); |
| 4904 | |
| 4905 | spd.nr_pages = i; |
| 4906 | |
| 4907 | ret = splice_to_pipe(pipe, &spd); |
| 4908 | out: |
| 4909 | splice_shrink_spd(&spd); |
| 4910 | return ret; |
| 4911 | |
| 4912 | out_err: |
| 4913 | mutex_unlock(&iter->mutex); |
| 4914 | goto out; |
| 4915 | } |
| 4916 | |
| 4917 | static ssize_t |
| 4918 | tracing_entries_read(struct file *filp, char __user *ubuf, |
| 4919 | size_t cnt, loff_t *ppos) |
| 4920 | { |
| 4921 | struct inode *inode = file_inode(filp); |
| 4922 | struct trace_array *tr = inode->i_private; |
| 4923 | int cpu = tracing_get_cpu(inode); |
| 4924 | char buf[64]; |
| 4925 | int r = 0; |
| 4926 | ssize_t ret; |
| 4927 | |
| 4928 | mutex_lock(&trace_types_lock); |
| 4929 | |
| 4930 | if (cpu == RING_BUFFER_ALL_CPUS) { |
| 4931 | int cpu, buf_size_same; |
| 4932 | unsigned long size; |
| 4933 | |
| 4934 | size = 0; |
| 4935 | buf_size_same = 1; |
| 4936 | /* check if all cpu sizes are same */ |
| 4937 | for_each_tracing_cpu(cpu) { |
| 4938 | /* fill in the size from first enabled cpu */ |
| 4939 | if (size == 0) |
| 4940 | size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries; |
| 4941 | if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) { |
| 4942 | buf_size_same = 0; |
| 4943 | break; |
| 4944 | } |
| 4945 | } |
| 4946 | |
| 4947 | if (buf_size_same) { |
| 4948 | if (!ring_buffer_expanded) |
| 4949 | r = sprintf(buf, "%lu (expanded: %lu)\n", |
| 4950 | size >> 10, |
| 4951 | trace_buf_size >> 10); |
| 4952 | else |
| 4953 | r = sprintf(buf, "%lu\n", size >> 10); |
| 4954 | } else |
| 4955 | r = sprintf(buf, "X\n"); |
| 4956 | } else |
| 4957 | r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10); |
| 4958 | |
| 4959 | mutex_unlock(&trace_types_lock); |
| 4960 | |
| 4961 | ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 4962 | return ret; |
| 4963 | } |
| 4964 | |
| 4965 | static ssize_t |
| 4966 | tracing_entries_write(struct file *filp, const char __user *ubuf, |
| 4967 | size_t cnt, loff_t *ppos) |
| 4968 | { |
| 4969 | struct inode *inode = file_inode(filp); |
| 4970 | struct trace_array *tr = inode->i_private; |
| 4971 | unsigned long val; |
| 4972 | int ret; |
| 4973 | |
| 4974 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 4975 | if (ret) |
| 4976 | return ret; |
| 4977 | |
| 4978 | /* must have at least 1 entry */ |
| 4979 | if (!val) |
| 4980 | return -EINVAL; |
| 4981 | |
| 4982 | /* value is in KB */ |
| 4983 | val <<= 10; |
| 4984 | ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode)); |
| 4985 | if (ret < 0) |
| 4986 | return ret; |
| 4987 | |
| 4988 | *ppos += cnt; |
| 4989 | |
| 4990 | return cnt; |
| 4991 | } |
| 4992 | |
| 4993 | static ssize_t |
| 4994 | tracing_total_entries_read(struct file *filp, char __user *ubuf, |
| 4995 | size_t cnt, loff_t *ppos) |
| 4996 | { |
| 4997 | struct trace_array *tr = filp->private_data; |
| 4998 | char buf[64]; |
| 4999 | int r, cpu; |
| 5000 | unsigned long size = 0, expanded_size = 0; |
| 5001 | |
| 5002 | mutex_lock(&trace_types_lock); |
| 5003 | for_each_tracing_cpu(cpu) { |
| 5004 | size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10; |
| 5005 | if (!ring_buffer_expanded) |
| 5006 | expanded_size += trace_buf_size >> 10; |
| 5007 | } |
| 5008 | if (ring_buffer_expanded) |
| 5009 | r = sprintf(buf, "%lu\n", size); |
| 5010 | else |
| 5011 | r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size); |
| 5012 | mutex_unlock(&trace_types_lock); |
| 5013 | |
| 5014 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 5015 | } |
| 5016 | |
| 5017 | static ssize_t |
| 5018 | tracing_free_buffer_write(struct file *filp, const char __user *ubuf, |
| 5019 | size_t cnt, loff_t *ppos) |
| 5020 | { |
| 5021 | /* |
| 5022 | * There is no need to read what the user has written, this function |
| 5023 | * is just to make sure that there is no error when "echo" is used |
| 5024 | */ |
| 5025 | |
| 5026 | *ppos += cnt; |
| 5027 | |
| 5028 | return cnt; |
| 5029 | } |
| 5030 | |
| 5031 | static int |
| 5032 | tracing_free_buffer_release(struct inode *inode, struct file *filp) |
| 5033 | { |
| 5034 | struct trace_array *tr = inode->i_private; |
| 5035 | |
| 5036 | /* disable tracing ? */ |
| 5037 | if (trace_flags & TRACE_ITER_STOP_ON_FREE) |
| 5038 | tracer_tracing_off(tr); |
| 5039 | /* resize the ring buffer to 0 */ |
| 5040 | tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS); |
| 5041 | |
| 5042 | trace_array_put(tr); |
| 5043 | |
| 5044 | return 0; |
| 5045 | } |
| 5046 | |
| 5047 | static ssize_t |
| 5048 | tracing_mark_write(struct file *filp, const char __user *ubuf, |
| 5049 | size_t cnt, loff_t *fpos) |
| 5050 | { |
| 5051 | unsigned long addr = (unsigned long)ubuf; |
| 5052 | struct trace_array *tr = filp->private_data; |
| 5053 | struct ring_buffer_event *event; |
| 5054 | struct ring_buffer *buffer; |
| 5055 | struct print_entry *entry; |
| 5056 | unsigned long irq_flags; |
| 5057 | struct page *pages[2]; |
| 5058 | void *map_page[2]; |
| 5059 | int nr_pages = 1; |
| 5060 | ssize_t written; |
| 5061 | int offset; |
| 5062 | int size; |
| 5063 | int len; |
| 5064 | int ret; |
| 5065 | int i; |
| 5066 | |
| 5067 | if (tracing_disabled) |
| 5068 | return -EINVAL; |
| 5069 | |
| 5070 | if (!(trace_flags & TRACE_ITER_MARKERS)) |
| 5071 | return -EINVAL; |
| 5072 | |
| 5073 | if (cnt > TRACE_BUF_SIZE) |
| 5074 | cnt = TRACE_BUF_SIZE; |
| 5075 | |
| 5076 | /* |
| 5077 | * Userspace is injecting traces into the kernel trace buffer. |
| 5078 | * We want to be as non intrusive as possible. |
| 5079 | * To do so, we do not want to allocate any special buffers |
| 5080 | * or take any locks, but instead write the userspace data |
| 5081 | * straight into the ring buffer. |
| 5082 | * |
| 5083 | * First we need to pin the userspace buffer into memory, |
| 5084 | * which, most likely it is, because it just referenced it. |
| 5085 | * But there's no guarantee that it is. By using get_user_pages_fast() |
| 5086 | * and kmap_atomic/kunmap_atomic() we can get access to the |
| 5087 | * pages directly. We then write the data directly into the |
| 5088 | * ring buffer. |
| 5089 | */ |
| 5090 | BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE); |
| 5091 | |
| 5092 | /* check if we cross pages */ |
| 5093 | if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK)) |
| 5094 | nr_pages = 2; |
| 5095 | |
| 5096 | offset = addr & (PAGE_SIZE - 1); |
| 5097 | addr &= PAGE_MASK; |
| 5098 | |
| 5099 | ret = get_user_pages_fast(addr, nr_pages, 0, pages); |
| 5100 | if (ret < nr_pages) { |
| 5101 | while (--ret >= 0) |
| 5102 | put_page(pages[ret]); |
| 5103 | written = -EFAULT; |
| 5104 | goto out; |
| 5105 | } |
| 5106 | |
| 5107 | for (i = 0; i < nr_pages; i++) |
| 5108 | map_page[i] = kmap_atomic(pages[i]); |
| 5109 | |
| 5110 | local_save_flags(irq_flags); |
| 5111 | size = sizeof(*entry) + cnt + 2; /* possible \n added */ |
| 5112 | buffer = tr->trace_buffer.buffer; |
| 5113 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, |
| 5114 | irq_flags, preempt_count()); |
| 5115 | if (!event) { |
| 5116 | /* Ring buffer disabled, return as if not open for write */ |
| 5117 | written = -EBADF; |
| 5118 | goto out_unlock; |
| 5119 | } |
| 5120 | |
| 5121 | entry = ring_buffer_event_data(event); |
| 5122 | entry->ip = _THIS_IP_; |
| 5123 | |
| 5124 | if (nr_pages == 2) { |
| 5125 | len = PAGE_SIZE - offset; |
| 5126 | memcpy(&entry->buf, map_page[0] + offset, len); |
| 5127 | memcpy(&entry->buf[len], map_page[1], cnt - len); |
| 5128 | } else |
| 5129 | memcpy(&entry->buf, map_page[0] + offset, cnt); |
| 5130 | |
| 5131 | if (entry->buf[cnt - 1] != '\n') { |
| 5132 | entry->buf[cnt] = '\n'; |
| 5133 | entry->buf[cnt + 1] = '\0'; |
| 5134 | } else |
| 5135 | entry->buf[cnt] = '\0'; |
| 5136 | |
| 5137 | __buffer_unlock_commit(buffer, event); |
| 5138 | |
| 5139 | written = cnt; |
| 5140 | |
| 5141 | *fpos += written; |
| 5142 | |
| 5143 | out_unlock: |
| 5144 | for (i = nr_pages - 1; i >= 0; i--) { |
| 5145 | kunmap_atomic(map_page[i]); |
| 5146 | put_page(pages[i]); |
| 5147 | } |
| 5148 | out: |
| 5149 | return written; |
| 5150 | } |
| 5151 | |
| 5152 | static int tracing_clock_show(struct seq_file *m, void *v) |
| 5153 | { |
| 5154 | struct trace_array *tr = m->private; |
| 5155 | int i; |
| 5156 | |
| 5157 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) |
| 5158 | seq_printf(m, |
| 5159 | "%s%s%s%s", i ? " " : "", |
| 5160 | i == tr->clock_id ? "[" : "", trace_clocks[i].name, |
| 5161 | i == tr->clock_id ? "]" : ""); |
| 5162 | seq_putc(m, '\n'); |
| 5163 | |
| 5164 | return 0; |
| 5165 | } |
| 5166 | |
| 5167 | static int tracing_set_clock(struct trace_array *tr, const char *clockstr) |
| 5168 | { |
| 5169 | int i; |
| 5170 | |
| 5171 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) { |
| 5172 | if (strcmp(trace_clocks[i].name, clockstr) == 0) |
| 5173 | break; |
| 5174 | } |
| 5175 | if (i == ARRAY_SIZE(trace_clocks)) |
| 5176 | return -EINVAL; |
| 5177 | |
| 5178 | mutex_lock(&trace_types_lock); |
| 5179 | |
| 5180 | tr->clock_id = i; |
| 5181 | |
| 5182 | ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func); |
| 5183 | |
| 5184 | /* |
| 5185 | * New clock may not be consistent with the previous clock. |
| 5186 | * Reset the buffer so that it doesn't have incomparable timestamps. |
| 5187 | */ |
| 5188 | tracing_reset_online_cpus(&tr->trace_buffer); |
| 5189 | |
| 5190 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 5191 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer) |
| 5192 | ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func); |
| 5193 | tracing_reset_online_cpus(&tr->max_buffer); |
| 5194 | #endif |
| 5195 | |
| 5196 | mutex_unlock(&trace_types_lock); |
| 5197 | |
| 5198 | return 0; |
| 5199 | } |
| 5200 | |
| 5201 | static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf, |
| 5202 | size_t cnt, loff_t *fpos) |
| 5203 | { |
| 5204 | struct seq_file *m = filp->private_data; |
| 5205 | struct trace_array *tr = m->private; |
| 5206 | char buf[64]; |
| 5207 | const char *clockstr; |
| 5208 | int ret; |
| 5209 | |
| 5210 | if (cnt >= sizeof(buf)) |
| 5211 | return -EINVAL; |
| 5212 | |
| 5213 | if (copy_from_user(&buf, ubuf, cnt)) |
| 5214 | return -EFAULT; |
| 5215 | |
| 5216 | buf[cnt] = 0; |
| 5217 | |
| 5218 | clockstr = strstrip(buf); |
| 5219 | |
| 5220 | ret = tracing_set_clock(tr, clockstr); |
| 5221 | if (ret) |
| 5222 | return ret; |
| 5223 | |
| 5224 | *fpos += cnt; |
| 5225 | |
| 5226 | return cnt; |
| 5227 | } |
| 5228 | |
| 5229 | static int tracing_clock_open(struct inode *inode, struct file *file) |
| 5230 | { |
| 5231 | struct trace_array *tr = inode->i_private; |
| 5232 | int ret; |
| 5233 | |
| 5234 | if (tracing_disabled) |
| 5235 | return -ENODEV; |
| 5236 | |
| 5237 | if (trace_array_get(tr)) |
| 5238 | return -ENODEV; |
| 5239 | |
| 5240 | ret = single_open(file, tracing_clock_show, inode->i_private); |
| 5241 | if (ret < 0) |
| 5242 | trace_array_put(tr); |
| 5243 | |
| 5244 | return ret; |
| 5245 | } |
| 5246 | |
| 5247 | struct ftrace_buffer_info { |
| 5248 | struct trace_iterator iter; |
| 5249 | void *spare; |
| 5250 | unsigned int read; |
| 5251 | }; |
| 5252 | |
| 5253 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 5254 | static int tracing_snapshot_open(struct inode *inode, struct file *file) |
| 5255 | { |
| 5256 | struct trace_array *tr = inode->i_private; |
| 5257 | struct trace_iterator *iter; |
| 5258 | struct seq_file *m; |
| 5259 | int ret = 0; |
| 5260 | |
| 5261 | if (trace_array_get(tr) < 0) |
| 5262 | return -ENODEV; |
| 5263 | |
| 5264 | if (file->f_mode & FMODE_READ) { |
| 5265 | iter = __tracing_open(inode, file, true); |
| 5266 | if (IS_ERR(iter)) |
| 5267 | ret = PTR_ERR(iter); |
| 5268 | } else { |
| 5269 | /* Writes still need the seq_file to hold the private data */ |
| 5270 | ret = -ENOMEM; |
| 5271 | m = kzalloc(sizeof(*m), GFP_KERNEL); |
| 5272 | if (!m) |
| 5273 | goto out; |
| 5274 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); |
| 5275 | if (!iter) { |
| 5276 | kfree(m); |
| 5277 | goto out; |
| 5278 | } |
| 5279 | ret = 0; |
| 5280 | |
| 5281 | iter->tr = tr; |
| 5282 | iter->trace_buffer = &tr->max_buffer; |
| 5283 | iter->cpu_file = tracing_get_cpu(inode); |
| 5284 | m->private = iter; |
| 5285 | file->private_data = m; |
| 5286 | } |
| 5287 | out: |
| 5288 | if (ret < 0) |
| 5289 | trace_array_put(tr); |
| 5290 | |
| 5291 | return ret; |
| 5292 | } |
| 5293 | |
| 5294 | static ssize_t |
| 5295 | tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt, |
| 5296 | loff_t *ppos) |
| 5297 | { |
| 5298 | struct seq_file *m = filp->private_data; |
| 5299 | struct trace_iterator *iter = m->private; |
| 5300 | struct trace_array *tr = iter->tr; |
| 5301 | unsigned long val; |
| 5302 | int ret; |
| 5303 | |
| 5304 | ret = tracing_update_buffers(); |
| 5305 | if (ret < 0) |
| 5306 | return ret; |
| 5307 | |
| 5308 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 5309 | if (ret) |
| 5310 | return ret; |
| 5311 | |
| 5312 | mutex_lock(&trace_types_lock); |
| 5313 | |
| 5314 | if (tr->current_trace->use_max_tr) { |
| 5315 | ret = -EBUSY; |
| 5316 | goto out; |
| 5317 | } |
| 5318 | |
| 5319 | switch (val) { |
| 5320 | case 0: |
| 5321 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { |
| 5322 | ret = -EINVAL; |
| 5323 | break; |
| 5324 | } |
| 5325 | if (tr->allocated_snapshot) |
| 5326 | free_snapshot(tr); |
| 5327 | break; |
| 5328 | case 1: |
| 5329 | /* Only allow per-cpu swap if the ring buffer supports it */ |
| 5330 | #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP |
| 5331 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { |
| 5332 | ret = -EINVAL; |
| 5333 | break; |
| 5334 | } |
| 5335 | #endif |
| 5336 | if (!tr->allocated_snapshot) { |
| 5337 | ret = alloc_snapshot(tr); |
| 5338 | if (ret < 0) |
| 5339 | break; |
| 5340 | } |
| 5341 | local_irq_disable(); |
| 5342 | /* Now, we're going to swap */ |
| 5343 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) |
| 5344 | update_max_tr(tr, current, smp_processor_id()); |
| 5345 | else |
| 5346 | update_max_tr_single(tr, current, iter->cpu_file); |
| 5347 | local_irq_enable(); |
| 5348 | break; |
| 5349 | default: |
| 5350 | if (tr->allocated_snapshot) { |
| 5351 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) |
| 5352 | tracing_reset_online_cpus(&tr->max_buffer); |
| 5353 | else |
| 5354 | tracing_reset(&tr->max_buffer, iter->cpu_file); |
| 5355 | } |
| 5356 | break; |
| 5357 | } |
| 5358 | |
| 5359 | if (ret >= 0) { |
| 5360 | *ppos += cnt; |
| 5361 | ret = cnt; |
| 5362 | } |
| 5363 | out: |
| 5364 | mutex_unlock(&trace_types_lock); |
| 5365 | return ret; |
| 5366 | } |
| 5367 | |
| 5368 | static int tracing_snapshot_release(struct inode *inode, struct file *file) |
| 5369 | { |
| 5370 | struct seq_file *m = file->private_data; |
| 5371 | int ret; |
| 5372 | |
| 5373 | ret = tracing_release(inode, file); |
| 5374 | |
| 5375 | if (file->f_mode & FMODE_READ) |
| 5376 | return ret; |
| 5377 | |
| 5378 | /* If write only, the seq_file is just a stub */ |
| 5379 | if (m) |
| 5380 | kfree(m->private); |
| 5381 | kfree(m); |
| 5382 | |
| 5383 | return 0; |
| 5384 | } |
| 5385 | |
| 5386 | static int tracing_buffers_open(struct inode *inode, struct file *filp); |
| 5387 | static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf, |
| 5388 | size_t count, loff_t *ppos); |
| 5389 | static int tracing_buffers_release(struct inode *inode, struct file *file); |
| 5390 | static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos, |
| 5391 | struct pipe_inode_info *pipe, size_t len, unsigned int flags); |
| 5392 | |
| 5393 | static int snapshot_raw_open(struct inode *inode, struct file *filp) |
| 5394 | { |
| 5395 | struct ftrace_buffer_info *info; |
| 5396 | int ret; |
| 5397 | |
| 5398 | ret = tracing_buffers_open(inode, filp); |
| 5399 | if (ret < 0) |
| 5400 | return ret; |
| 5401 | |
| 5402 | info = filp->private_data; |
| 5403 | |
| 5404 | if (info->iter.trace->use_max_tr) { |
| 5405 | tracing_buffers_release(inode, filp); |
| 5406 | return -EBUSY; |
| 5407 | } |
| 5408 | |
| 5409 | info->iter.snapshot = true; |
| 5410 | info->iter.trace_buffer = &info->iter.tr->max_buffer; |
| 5411 | |
| 5412 | return ret; |
| 5413 | } |
| 5414 | |
| 5415 | #endif /* CONFIG_TRACER_SNAPSHOT */ |
| 5416 | |
| 5417 | |
| 5418 | static const struct file_operations tracing_thresh_fops = { |
| 5419 | .open = tracing_open_generic, |
| 5420 | .read = tracing_thresh_read, |
| 5421 | .write = tracing_thresh_write, |
| 5422 | .llseek = generic_file_llseek, |
| 5423 | }; |
| 5424 | |
| 5425 | static const struct file_operations tracing_max_lat_fops = { |
| 5426 | .open = tracing_open_generic, |
| 5427 | .read = tracing_max_lat_read, |
| 5428 | .write = tracing_max_lat_write, |
| 5429 | .llseek = generic_file_llseek, |
| 5430 | }; |
| 5431 | |
| 5432 | static const struct file_operations set_tracer_fops = { |
| 5433 | .open = tracing_open_generic, |
| 5434 | .read = tracing_set_trace_read, |
| 5435 | .write = tracing_set_trace_write, |
| 5436 | .llseek = generic_file_llseek, |
| 5437 | }; |
| 5438 | |
| 5439 | static const struct file_operations tracing_pipe_fops = { |
| 5440 | .open = tracing_open_pipe, |
| 5441 | .poll = tracing_poll_pipe, |
| 5442 | .read = tracing_read_pipe, |
| 5443 | .splice_read = tracing_splice_read_pipe, |
| 5444 | .release = tracing_release_pipe, |
| 5445 | .llseek = no_llseek, |
| 5446 | }; |
| 5447 | |
| 5448 | static const struct file_operations tracing_entries_fops = { |
| 5449 | .open = tracing_open_generic_tr, |
| 5450 | .read = tracing_entries_read, |
| 5451 | .write = tracing_entries_write, |
| 5452 | .llseek = generic_file_llseek, |
| 5453 | .release = tracing_release_generic_tr, |
| 5454 | }; |
| 5455 | |
| 5456 | static const struct file_operations tracing_total_entries_fops = { |
| 5457 | .open = tracing_open_generic_tr, |
| 5458 | .read = tracing_total_entries_read, |
| 5459 | .llseek = generic_file_llseek, |
| 5460 | .release = tracing_release_generic_tr, |
| 5461 | }; |
| 5462 | |
| 5463 | static const struct file_operations tracing_free_buffer_fops = { |
| 5464 | .open = tracing_open_generic_tr, |
| 5465 | .write = tracing_free_buffer_write, |
| 5466 | .release = tracing_free_buffer_release, |
| 5467 | }; |
| 5468 | |
| 5469 | static const struct file_operations tracing_mark_fops = { |
| 5470 | .open = tracing_open_generic_tr, |
| 5471 | .write = tracing_mark_write, |
| 5472 | .llseek = generic_file_llseek, |
| 5473 | .release = tracing_release_generic_tr, |
| 5474 | }; |
| 5475 | |
| 5476 | static const struct file_operations trace_clock_fops = { |
| 5477 | .open = tracing_clock_open, |
| 5478 | .read = seq_read, |
| 5479 | .llseek = seq_lseek, |
| 5480 | .release = tracing_single_release_tr, |
| 5481 | .write = tracing_clock_write, |
| 5482 | }; |
| 5483 | |
| 5484 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 5485 | static const struct file_operations snapshot_fops = { |
| 5486 | .open = tracing_snapshot_open, |
| 5487 | .read = seq_read, |
| 5488 | .write = tracing_snapshot_write, |
| 5489 | .llseek = tracing_lseek, |
| 5490 | .release = tracing_snapshot_release, |
| 5491 | }; |
| 5492 | |
| 5493 | static const struct file_operations snapshot_raw_fops = { |
| 5494 | .open = snapshot_raw_open, |
| 5495 | .read = tracing_buffers_read, |
| 5496 | .release = tracing_buffers_release, |
| 5497 | .splice_read = tracing_buffers_splice_read, |
| 5498 | .llseek = no_llseek, |
| 5499 | }; |
| 5500 | |
| 5501 | #endif /* CONFIG_TRACER_SNAPSHOT */ |
| 5502 | |
| 5503 | static int tracing_buffers_open(struct inode *inode, struct file *filp) |
| 5504 | { |
| 5505 | struct trace_array *tr = inode->i_private; |
| 5506 | struct ftrace_buffer_info *info; |
| 5507 | int ret; |
| 5508 | |
| 5509 | if (tracing_disabled) |
| 5510 | return -ENODEV; |
| 5511 | |
| 5512 | if (trace_array_get(tr) < 0) |
| 5513 | return -ENODEV; |
| 5514 | |
| 5515 | info = kzalloc(sizeof(*info), GFP_KERNEL); |
| 5516 | if (!info) { |
| 5517 | trace_array_put(tr); |
| 5518 | return -ENOMEM; |
| 5519 | } |
| 5520 | |
| 5521 | mutex_lock(&trace_types_lock); |
| 5522 | |
| 5523 | info->iter.tr = tr; |
| 5524 | info->iter.cpu_file = tracing_get_cpu(inode); |
| 5525 | info->iter.trace = tr->current_trace; |
| 5526 | info->iter.trace_buffer = &tr->trace_buffer; |
| 5527 | info->spare = NULL; |
| 5528 | /* Force reading ring buffer for first read */ |
| 5529 | info->read = (unsigned int)-1; |
| 5530 | |
| 5531 | filp->private_data = info; |
| 5532 | |
| 5533 | tr->current_trace->ref++; |
| 5534 | |
| 5535 | mutex_unlock(&trace_types_lock); |
| 5536 | |
| 5537 | ret = nonseekable_open(inode, filp); |
| 5538 | if (ret < 0) |
| 5539 | trace_array_put(tr); |
| 5540 | |
| 5541 | return ret; |
| 5542 | } |
| 5543 | |
| 5544 | static unsigned int |
| 5545 | tracing_buffers_poll(struct file *filp, poll_table *poll_table) |
| 5546 | { |
| 5547 | struct ftrace_buffer_info *info = filp->private_data; |
| 5548 | struct trace_iterator *iter = &info->iter; |
| 5549 | |
| 5550 | return trace_poll(iter, filp, poll_table); |
| 5551 | } |
| 5552 | |
| 5553 | static ssize_t |
| 5554 | tracing_buffers_read(struct file *filp, char __user *ubuf, |
| 5555 | size_t count, loff_t *ppos) |
| 5556 | { |
| 5557 | struct ftrace_buffer_info *info = filp->private_data; |
| 5558 | struct trace_iterator *iter = &info->iter; |
| 5559 | ssize_t ret; |
| 5560 | ssize_t size; |
| 5561 | |
| 5562 | if (!count) |
| 5563 | return 0; |
| 5564 | |
| 5565 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 5566 | if (iter->snapshot && iter->tr->current_trace->use_max_tr) |
| 5567 | return -EBUSY; |
| 5568 | #endif |
| 5569 | |
| 5570 | if (!info->spare) |
| 5571 | info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer, |
| 5572 | iter->cpu_file); |
| 5573 | if (!info->spare) |
| 5574 | return -ENOMEM; |
| 5575 | |
| 5576 | /* Do we have previous read data to read? */ |
| 5577 | if (info->read < PAGE_SIZE) |
| 5578 | goto read; |
| 5579 | |
| 5580 | again: |
| 5581 | trace_access_lock(iter->cpu_file); |
| 5582 | ret = ring_buffer_read_page(iter->trace_buffer->buffer, |
| 5583 | &info->spare, |
| 5584 | count, |
| 5585 | iter->cpu_file, 0); |
| 5586 | trace_access_unlock(iter->cpu_file); |
| 5587 | |
| 5588 | if (ret < 0) { |
| 5589 | if (trace_empty(iter)) { |
| 5590 | if ((filp->f_flags & O_NONBLOCK)) |
| 5591 | return -EAGAIN; |
| 5592 | |
| 5593 | ret = wait_on_pipe(iter, false); |
| 5594 | if (ret) |
| 5595 | return ret; |
| 5596 | |
| 5597 | goto again; |
| 5598 | } |
| 5599 | return 0; |
| 5600 | } |
| 5601 | |
| 5602 | info->read = 0; |
| 5603 | read: |
| 5604 | size = PAGE_SIZE - info->read; |
| 5605 | if (size > count) |
| 5606 | size = count; |
| 5607 | |
| 5608 | ret = copy_to_user(ubuf, info->spare + info->read, size); |
| 5609 | if (ret == size) |
| 5610 | return -EFAULT; |
| 5611 | |
| 5612 | size -= ret; |
| 5613 | |
| 5614 | *ppos += size; |
| 5615 | info->read += size; |
| 5616 | |
| 5617 | return size; |
| 5618 | } |
| 5619 | |
| 5620 | static int tracing_buffers_release(struct inode *inode, struct file *file) |
| 5621 | { |
| 5622 | struct ftrace_buffer_info *info = file->private_data; |
| 5623 | struct trace_iterator *iter = &info->iter; |
| 5624 | |
| 5625 | mutex_lock(&trace_types_lock); |
| 5626 | |
| 5627 | iter->tr->current_trace->ref--; |
| 5628 | |
| 5629 | __trace_array_put(iter->tr); |
| 5630 | |
| 5631 | if (info->spare) |
| 5632 | ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare); |
| 5633 | kfree(info); |
| 5634 | |
| 5635 | mutex_unlock(&trace_types_lock); |
| 5636 | |
| 5637 | return 0; |
| 5638 | } |
| 5639 | |
| 5640 | struct buffer_ref { |
| 5641 | struct ring_buffer *buffer; |
| 5642 | void *page; |
| 5643 | int ref; |
| 5644 | }; |
| 5645 | |
| 5646 | static void buffer_pipe_buf_release(struct pipe_inode_info *pipe, |
| 5647 | struct pipe_buffer *buf) |
| 5648 | { |
| 5649 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; |
| 5650 | |
| 5651 | if (--ref->ref) |
| 5652 | return; |
| 5653 | |
| 5654 | ring_buffer_free_read_page(ref->buffer, ref->page); |
| 5655 | kfree(ref); |
| 5656 | buf->private = 0; |
| 5657 | } |
| 5658 | |
| 5659 | static void buffer_pipe_buf_get(struct pipe_inode_info *pipe, |
| 5660 | struct pipe_buffer *buf) |
| 5661 | { |
| 5662 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; |
| 5663 | |
| 5664 | ref->ref++; |
| 5665 | } |
| 5666 | |
| 5667 | /* Pipe buffer operations for a buffer. */ |
| 5668 | static const struct pipe_buf_operations buffer_pipe_buf_ops = { |
| 5669 | .can_merge = 0, |
| 5670 | .confirm = generic_pipe_buf_confirm, |
| 5671 | .release = buffer_pipe_buf_release, |
| 5672 | .steal = generic_pipe_buf_steal, |
| 5673 | .get = buffer_pipe_buf_get, |
| 5674 | }; |
| 5675 | |
| 5676 | /* |
| 5677 | * Callback from splice_to_pipe(), if we need to release some pages |
| 5678 | * at the end of the spd in case we error'ed out in filling the pipe. |
| 5679 | */ |
| 5680 | static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i) |
| 5681 | { |
| 5682 | struct buffer_ref *ref = |
| 5683 | (struct buffer_ref *)spd->partial[i].private; |
| 5684 | |
| 5685 | if (--ref->ref) |
| 5686 | return; |
| 5687 | |
| 5688 | ring_buffer_free_read_page(ref->buffer, ref->page); |
| 5689 | kfree(ref); |
| 5690 | spd->partial[i].private = 0; |
| 5691 | } |
| 5692 | |
| 5693 | static ssize_t |
| 5694 | tracing_buffers_splice_read(struct file *file, loff_t *ppos, |
| 5695 | struct pipe_inode_info *pipe, size_t len, |
| 5696 | unsigned int flags) |
| 5697 | { |
| 5698 | struct ftrace_buffer_info *info = file->private_data; |
| 5699 | struct trace_iterator *iter = &info->iter; |
| 5700 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; |
| 5701 | struct page *pages_def[PIPE_DEF_BUFFERS]; |
| 5702 | struct splice_pipe_desc spd = { |
| 5703 | .pages = pages_def, |
| 5704 | .partial = partial_def, |
| 5705 | .nr_pages_max = PIPE_DEF_BUFFERS, |
| 5706 | .flags = flags, |
| 5707 | .ops = &buffer_pipe_buf_ops, |
| 5708 | .spd_release = buffer_spd_release, |
| 5709 | }; |
| 5710 | struct buffer_ref *ref; |
| 5711 | int entries, size, i; |
| 5712 | ssize_t ret = 0; |
| 5713 | |
| 5714 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 5715 | if (iter->snapshot && iter->tr->current_trace->use_max_tr) |
| 5716 | return -EBUSY; |
| 5717 | #endif |
| 5718 | |
| 5719 | if (splice_grow_spd(pipe, &spd)) |
| 5720 | return -ENOMEM; |
| 5721 | |
| 5722 | if (*ppos & (PAGE_SIZE - 1)) |
| 5723 | return -EINVAL; |
| 5724 | |
| 5725 | if (len & (PAGE_SIZE - 1)) { |
| 5726 | if (len < PAGE_SIZE) |
| 5727 | return -EINVAL; |
| 5728 | len &= PAGE_MASK; |
| 5729 | } |
| 5730 | |
| 5731 | again: |
| 5732 | trace_access_lock(iter->cpu_file); |
| 5733 | entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file); |
| 5734 | |
| 5735 | for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) { |
| 5736 | struct page *page; |
| 5737 | int r; |
| 5738 | |
| 5739 | ref = kzalloc(sizeof(*ref), GFP_KERNEL); |
| 5740 | if (!ref) { |
| 5741 | ret = -ENOMEM; |
| 5742 | break; |
| 5743 | } |
| 5744 | |
| 5745 | ref->ref = 1; |
| 5746 | ref->buffer = iter->trace_buffer->buffer; |
| 5747 | ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file); |
| 5748 | if (!ref->page) { |
| 5749 | ret = -ENOMEM; |
| 5750 | kfree(ref); |
| 5751 | break; |
| 5752 | } |
| 5753 | |
| 5754 | r = ring_buffer_read_page(ref->buffer, &ref->page, |
| 5755 | len, iter->cpu_file, 1); |
| 5756 | if (r < 0) { |
| 5757 | ring_buffer_free_read_page(ref->buffer, ref->page); |
| 5758 | kfree(ref); |
| 5759 | break; |
| 5760 | } |
| 5761 | |
| 5762 | /* |
| 5763 | * zero out any left over data, this is going to |
| 5764 | * user land. |
| 5765 | */ |
| 5766 | size = ring_buffer_page_len(ref->page); |
| 5767 | if (size < PAGE_SIZE) |
| 5768 | memset(ref->page + size, 0, PAGE_SIZE - size); |
| 5769 | |
| 5770 | page = virt_to_page(ref->page); |
| 5771 | |
| 5772 | spd.pages[i] = page; |
| 5773 | spd.partial[i].len = PAGE_SIZE; |
| 5774 | spd.partial[i].offset = 0; |
| 5775 | spd.partial[i].private = (unsigned long)ref; |
| 5776 | spd.nr_pages++; |
| 5777 | *ppos += PAGE_SIZE; |
| 5778 | |
| 5779 | entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file); |
| 5780 | } |
| 5781 | |
| 5782 | trace_access_unlock(iter->cpu_file); |
| 5783 | spd.nr_pages = i; |
| 5784 | |
| 5785 | /* did we read anything? */ |
| 5786 | if (!spd.nr_pages) { |
| 5787 | if (ret) |
| 5788 | return ret; |
| 5789 | |
| 5790 | if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) |
| 5791 | return -EAGAIN; |
| 5792 | |
| 5793 | ret = wait_on_pipe(iter, true); |
| 5794 | if (ret) |
| 5795 | return ret; |
| 5796 | |
| 5797 | goto again; |
| 5798 | } |
| 5799 | |
| 5800 | ret = splice_to_pipe(pipe, &spd); |
| 5801 | splice_shrink_spd(&spd); |
| 5802 | |
| 5803 | return ret; |
| 5804 | } |
| 5805 | |
| 5806 | static const struct file_operations tracing_buffers_fops = { |
| 5807 | .open = tracing_buffers_open, |
| 5808 | .read = tracing_buffers_read, |
| 5809 | .poll = tracing_buffers_poll, |
| 5810 | .release = tracing_buffers_release, |
| 5811 | .splice_read = tracing_buffers_splice_read, |
| 5812 | .llseek = no_llseek, |
| 5813 | }; |
| 5814 | |
| 5815 | static ssize_t |
| 5816 | tracing_stats_read(struct file *filp, char __user *ubuf, |
| 5817 | size_t count, loff_t *ppos) |
| 5818 | { |
| 5819 | struct inode *inode = file_inode(filp); |
| 5820 | struct trace_array *tr = inode->i_private; |
| 5821 | struct trace_buffer *trace_buf = &tr->trace_buffer; |
| 5822 | int cpu = tracing_get_cpu(inode); |
| 5823 | struct trace_seq *s; |
| 5824 | unsigned long cnt; |
| 5825 | unsigned long long t; |
| 5826 | unsigned long usec_rem; |
| 5827 | |
| 5828 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
| 5829 | if (!s) |
| 5830 | return -ENOMEM; |
| 5831 | |
| 5832 | trace_seq_init(s); |
| 5833 | |
| 5834 | cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu); |
| 5835 | trace_seq_printf(s, "entries: %ld\n", cnt); |
| 5836 | |
| 5837 | cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu); |
| 5838 | trace_seq_printf(s, "overrun: %ld\n", cnt); |
| 5839 | |
| 5840 | cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu); |
| 5841 | trace_seq_printf(s, "commit overrun: %ld\n", cnt); |
| 5842 | |
| 5843 | cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu); |
| 5844 | trace_seq_printf(s, "bytes: %ld\n", cnt); |
| 5845 | |
| 5846 | if (trace_clocks[tr->clock_id].in_ns) { |
| 5847 | /* local or global for trace_clock */ |
| 5848 | t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu)); |
| 5849 | usec_rem = do_div(t, USEC_PER_SEC); |
| 5850 | trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", |
| 5851 | t, usec_rem); |
| 5852 | |
| 5853 | t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu)); |
| 5854 | usec_rem = do_div(t, USEC_PER_SEC); |
| 5855 | trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem); |
| 5856 | } else { |
| 5857 | /* counter or tsc mode for trace_clock */ |
| 5858 | trace_seq_printf(s, "oldest event ts: %llu\n", |
| 5859 | ring_buffer_oldest_event_ts(trace_buf->buffer, cpu)); |
| 5860 | |
| 5861 | trace_seq_printf(s, "now ts: %llu\n", |
| 5862 | ring_buffer_time_stamp(trace_buf->buffer, cpu)); |
| 5863 | } |
| 5864 | |
| 5865 | cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu); |
| 5866 | trace_seq_printf(s, "dropped events: %ld\n", cnt); |
| 5867 | |
| 5868 | cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu); |
| 5869 | trace_seq_printf(s, "read events: %ld\n", cnt); |
| 5870 | |
| 5871 | count = simple_read_from_buffer(ubuf, count, ppos, |
| 5872 | s->buffer, trace_seq_used(s)); |
| 5873 | |
| 5874 | kfree(s); |
| 5875 | |
| 5876 | return count; |
| 5877 | } |
| 5878 | |
| 5879 | static const struct file_operations tracing_stats_fops = { |
| 5880 | .open = tracing_open_generic_tr, |
| 5881 | .read = tracing_stats_read, |
| 5882 | .llseek = generic_file_llseek, |
| 5883 | .release = tracing_release_generic_tr, |
| 5884 | }; |
| 5885 | |
| 5886 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 5887 | |
| 5888 | int __weak ftrace_arch_read_dyn_info(char *buf, int size) |
| 5889 | { |
| 5890 | return 0; |
| 5891 | } |
| 5892 | |
| 5893 | static ssize_t |
| 5894 | tracing_read_dyn_info(struct file *filp, char __user *ubuf, |
| 5895 | size_t cnt, loff_t *ppos) |
| 5896 | { |
| 5897 | static char ftrace_dyn_info_buffer[1024]; |
| 5898 | static DEFINE_MUTEX(dyn_info_mutex); |
| 5899 | unsigned long *p = filp->private_data; |
| 5900 | char *buf = ftrace_dyn_info_buffer; |
| 5901 | int size = ARRAY_SIZE(ftrace_dyn_info_buffer); |
| 5902 | int r; |
| 5903 | |
| 5904 | mutex_lock(&dyn_info_mutex); |
| 5905 | r = sprintf(buf, "%ld ", *p); |
| 5906 | |
| 5907 | r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r); |
| 5908 | buf[r++] = '\n'; |
| 5909 | |
| 5910 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 5911 | |
| 5912 | mutex_unlock(&dyn_info_mutex); |
| 5913 | |
| 5914 | return r; |
| 5915 | } |
| 5916 | |
| 5917 | static const struct file_operations tracing_dyn_info_fops = { |
| 5918 | .open = tracing_open_generic, |
| 5919 | .read = tracing_read_dyn_info, |
| 5920 | .llseek = generic_file_llseek, |
| 5921 | }; |
| 5922 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
| 5923 | |
| 5924 | #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) |
| 5925 | static void |
| 5926 | ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data) |
| 5927 | { |
| 5928 | tracing_snapshot(); |
| 5929 | } |
| 5930 | |
| 5931 | static void |
| 5932 | ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data) |
| 5933 | { |
| 5934 | unsigned long *count = (long *)data; |
| 5935 | |
| 5936 | if (!*count) |
| 5937 | return; |
| 5938 | |
| 5939 | if (*count != -1) |
| 5940 | (*count)--; |
| 5941 | |
| 5942 | tracing_snapshot(); |
| 5943 | } |
| 5944 | |
| 5945 | static int |
| 5946 | ftrace_snapshot_print(struct seq_file *m, unsigned long ip, |
| 5947 | struct ftrace_probe_ops *ops, void *data) |
| 5948 | { |
| 5949 | long count = (long)data; |
| 5950 | |
| 5951 | seq_printf(m, "%ps:", (void *)ip); |
| 5952 | |
| 5953 | seq_puts(m, "snapshot"); |
| 5954 | |
| 5955 | if (count == -1) |
| 5956 | seq_puts(m, ":unlimited\n"); |
| 5957 | else |
| 5958 | seq_printf(m, ":count=%ld\n", count); |
| 5959 | |
| 5960 | return 0; |
| 5961 | } |
| 5962 | |
| 5963 | static struct ftrace_probe_ops snapshot_probe_ops = { |
| 5964 | .func = ftrace_snapshot, |
| 5965 | .print = ftrace_snapshot_print, |
| 5966 | }; |
| 5967 | |
| 5968 | static struct ftrace_probe_ops snapshot_count_probe_ops = { |
| 5969 | .func = ftrace_count_snapshot, |
| 5970 | .print = ftrace_snapshot_print, |
| 5971 | }; |
| 5972 | |
| 5973 | static int |
| 5974 | ftrace_trace_snapshot_callback(struct ftrace_hash *hash, |
| 5975 | char *glob, char *cmd, char *param, int enable) |
| 5976 | { |
| 5977 | struct ftrace_probe_ops *ops; |
| 5978 | void *count = (void *)-1; |
| 5979 | char *number; |
| 5980 | int ret; |
| 5981 | |
| 5982 | /* hash funcs only work with set_ftrace_filter */ |
| 5983 | if (!enable) |
| 5984 | return -EINVAL; |
| 5985 | |
| 5986 | ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops; |
| 5987 | |
| 5988 | if (glob[0] == '!') { |
| 5989 | unregister_ftrace_function_probe_func(glob+1, ops); |
| 5990 | return 0; |
| 5991 | } |
| 5992 | |
| 5993 | if (!param) |
| 5994 | goto out_reg; |
| 5995 | |
| 5996 | number = strsep(¶m, ":"); |
| 5997 | |
| 5998 | if (!strlen(number)) |
| 5999 | goto out_reg; |
| 6000 | |
| 6001 | /* |
| 6002 | * We use the callback data field (which is a pointer) |
| 6003 | * as our counter. |
| 6004 | */ |
| 6005 | ret = kstrtoul(number, 0, (unsigned long *)&count); |
| 6006 | if (ret) |
| 6007 | return ret; |
| 6008 | |
| 6009 | out_reg: |
| 6010 | ret = register_ftrace_function_probe(glob, ops, count); |
| 6011 | |
| 6012 | if (ret >= 0) |
| 6013 | alloc_snapshot(&global_trace); |
| 6014 | |
| 6015 | return ret < 0 ? ret : 0; |
| 6016 | } |
| 6017 | |
| 6018 | static struct ftrace_func_command ftrace_snapshot_cmd = { |
| 6019 | .name = "snapshot", |
| 6020 | .func = ftrace_trace_snapshot_callback, |
| 6021 | }; |
| 6022 | |
| 6023 | static __init int register_snapshot_cmd(void) |
| 6024 | { |
| 6025 | return register_ftrace_command(&ftrace_snapshot_cmd); |
| 6026 | } |
| 6027 | #else |
| 6028 | static inline __init int register_snapshot_cmd(void) { return 0; } |
| 6029 | #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */ |
| 6030 | |
| 6031 | static struct dentry *tracing_get_dentry(struct trace_array *tr) |
| 6032 | { |
| 6033 | if (WARN_ON(!tr->dir)) |
| 6034 | return ERR_PTR(-ENODEV); |
| 6035 | |
| 6036 | /* Top directory uses NULL as the parent */ |
| 6037 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) |
| 6038 | return NULL; |
| 6039 | |
| 6040 | /* All sub buffers have a descriptor */ |
| 6041 | return tr->dir; |
| 6042 | } |
| 6043 | |
| 6044 | static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu) |
| 6045 | { |
| 6046 | struct dentry *d_tracer; |
| 6047 | |
| 6048 | if (tr->percpu_dir) |
| 6049 | return tr->percpu_dir; |
| 6050 | |
| 6051 | d_tracer = tracing_get_dentry(tr); |
| 6052 | if (IS_ERR(d_tracer)) |
| 6053 | return NULL; |
| 6054 | |
| 6055 | tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer); |
| 6056 | |
| 6057 | WARN_ONCE(!tr->percpu_dir, |
| 6058 | "Could not create tracefs directory 'per_cpu/%d'\n", cpu); |
| 6059 | |
| 6060 | return tr->percpu_dir; |
| 6061 | } |
| 6062 | |
| 6063 | static struct dentry * |
| 6064 | trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent, |
| 6065 | void *data, long cpu, const struct file_operations *fops) |
| 6066 | { |
| 6067 | struct dentry *ret = trace_create_file(name, mode, parent, data, fops); |
| 6068 | |
| 6069 | if (ret) /* See tracing_get_cpu() */ |
| 6070 | d_inode(ret)->i_cdev = (void *)(cpu + 1); |
| 6071 | return ret; |
| 6072 | } |
| 6073 | |
| 6074 | static void |
| 6075 | tracing_init_tracefs_percpu(struct trace_array *tr, long cpu) |
| 6076 | { |
| 6077 | struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu); |
| 6078 | struct dentry *d_cpu; |
| 6079 | char cpu_dir[30]; /* 30 characters should be more than enough */ |
| 6080 | |
| 6081 | if (!d_percpu) |
| 6082 | return; |
| 6083 | |
| 6084 | snprintf(cpu_dir, 30, "cpu%ld", cpu); |
| 6085 | d_cpu = tracefs_create_dir(cpu_dir, d_percpu); |
| 6086 | if (!d_cpu) { |
| 6087 | pr_warning("Could not create tracefs '%s' entry\n", cpu_dir); |
| 6088 | return; |
| 6089 | } |
| 6090 | |
| 6091 | /* per cpu trace_pipe */ |
| 6092 | trace_create_cpu_file("trace_pipe", 0444, d_cpu, |
| 6093 | tr, cpu, &tracing_pipe_fops); |
| 6094 | |
| 6095 | /* per cpu trace */ |
| 6096 | trace_create_cpu_file("trace", 0644, d_cpu, |
| 6097 | tr, cpu, &tracing_fops); |
| 6098 | |
| 6099 | trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu, |
| 6100 | tr, cpu, &tracing_buffers_fops); |
| 6101 | |
| 6102 | trace_create_cpu_file("stats", 0444, d_cpu, |
| 6103 | tr, cpu, &tracing_stats_fops); |
| 6104 | |
| 6105 | trace_create_cpu_file("buffer_size_kb", 0444, d_cpu, |
| 6106 | tr, cpu, &tracing_entries_fops); |
| 6107 | |
| 6108 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 6109 | trace_create_cpu_file("snapshot", 0644, d_cpu, |
| 6110 | tr, cpu, &snapshot_fops); |
| 6111 | |
| 6112 | trace_create_cpu_file("snapshot_raw", 0444, d_cpu, |
| 6113 | tr, cpu, &snapshot_raw_fops); |
| 6114 | #endif |
| 6115 | } |
| 6116 | |
| 6117 | #ifdef CONFIG_FTRACE_SELFTEST |
| 6118 | /* Let selftest have access to static functions in this file */ |
| 6119 | #include "trace_selftest.c" |
| 6120 | #endif |
| 6121 | |
| 6122 | static ssize_t |
| 6123 | trace_options_read(struct file *filp, char __user *ubuf, size_t cnt, |
| 6124 | loff_t *ppos) |
| 6125 | { |
| 6126 | struct trace_option_dentry *topt = filp->private_data; |
| 6127 | char *buf; |
| 6128 | |
| 6129 | if (topt->flags->val & topt->opt->bit) |
| 6130 | buf = "1\n"; |
| 6131 | else |
| 6132 | buf = "0\n"; |
| 6133 | |
| 6134 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); |
| 6135 | } |
| 6136 | |
| 6137 | static ssize_t |
| 6138 | trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt, |
| 6139 | loff_t *ppos) |
| 6140 | { |
| 6141 | struct trace_option_dentry *topt = filp->private_data; |
| 6142 | unsigned long val; |
| 6143 | int ret; |
| 6144 | |
| 6145 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 6146 | if (ret) |
| 6147 | return ret; |
| 6148 | |
| 6149 | if (val != 0 && val != 1) |
| 6150 | return -EINVAL; |
| 6151 | |
| 6152 | if (!!(topt->flags->val & topt->opt->bit) != val) { |
| 6153 | mutex_lock(&trace_types_lock); |
| 6154 | ret = __set_tracer_option(topt->tr, topt->flags, |
| 6155 | topt->opt, !val); |
| 6156 | mutex_unlock(&trace_types_lock); |
| 6157 | if (ret) |
| 6158 | return ret; |
| 6159 | } |
| 6160 | |
| 6161 | *ppos += cnt; |
| 6162 | |
| 6163 | return cnt; |
| 6164 | } |
| 6165 | |
| 6166 | |
| 6167 | static const struct file_operations trace_options_fops = { |
| 6168 | .open = tracing_open_generic, |
| 6169 | .read = trace_options_read, |
| 6170 | .write = trace_options_write, |
| 6171 | .llseek = generic_file_llseek, |
| 6172 | }; |
| 6173 | |
| 6174 | static ssize_t |
| 6175 | trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt, |
| 6176 | loff_t *ppos) |
| 6177 | { |
| 6178 | long index = (long)filp->private_data; |
| 6179 | char *buf; |
| 6180 | |
| 6181 | if (trace_flags & (1 << index)) |
| 6182 | buf = "1\n"; |
| 6183 | else |
| 6184 | buf = "0\n"; |
| 6185 | |
| 6186 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); |
| 6187 | } |
| 6188 | |
| 6189 | static ssize_t |
| 6190 | trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, |
| 6191 | loff_t *ppos) |
| 6192 | { |
| 6193 | struct trace_array *tr = &global_trace; |
| 6194 | long index = (long)filp->private_data; |
| 6195 | unsigned long val; |
| 6196 | int ret; |
| 6197 | |
| 6198 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 6199 | if (ret) |
| 6200 | return ret; |
| 6201 | |
| 6202 | if (val != 0 && val != 1) |
| 6203 | return -EINVAL; |
| 6204 | |
| 6205 | mutex_lock(&trace_types_lock); |
| 6206 | ret = set_tracer_flag(tr, 1 << index, val); |
| 6207 | mutex_unlock(&trace_types_lock); |
| 6208 | |
| 6209 | if (ret < 0) |
| 6210 | return ret; |
| 6211 | |
| 6212 | *ppos += cnt; |
| 6213 | |
| 6214 | return cnt; |
| 6215 | } |
| 6216 | |
| 6217 | static const struct file_operations trace_options_core_fops = { |
| 6218 | .open = tracing_open_generic, |
| 6219 | .read = trace_options_core_read, |
| 6220 | .write = trace_options_core_write, |
| 6221 | .llseek = generic_file_llseek, |
| 6222 | }; |
| 6223 | |
| 6224 | struct dentry *trace_create_file(const char *name, |
| 6225 | umode_t mode, |
| 6226 | struct dentry *parent, |
| 6227 | void *data, |
| 6228 | const struct file_operations *fops) |
| 6229 | { |
| 6230 | struct dentry *ret; |
| 6231 | |
| 6232 | ret = tracefs_create_file(name, mode, parent, data, fops); |
| 6233 | if (!ret) |
| 6234 | pr_warning("Could not create tracefs '%s' entry\n", name); |
| 6235 | |
| 6236 | return ret; |
| 6237 | } |
| 6238 | |
| 6239 | |
| 6240 | static struct dentry *trace_options_init_dentry(struct trace_array *tr) |
| 6241 | { |
| 6242 | struct dentry *d_tracer; |
| 6243 | |
| 6244 | if (tr->options) |
| 6245 | return tr->options; |
| 6246 | |
| 6247 | d_tracer = tracing_get_dentry(tr); |
| 6248 | if (IS_ERR(d_tracer)) |
| 6249 | return NULL; |
| 6250 | |
| 6251 | tr->options = tracefs_create_dir("options", d_tracer); |
| 6252 | if (!tr->options) { |
| 6253 | pr_warning("Could not create tracefs directory 'options'\n"); |
| 6254 | return NULL; |
| 6255 | } |
| 6256 | |
| 6257 | return tr->options; |
| 6258 | } |
| 6259 | |
| 6260 | static void |
| 6261 | create_trace_option_file(struct trace_array *tr, |
| 6262 | struct trace_option_dentry *topt, |
| 6263 | struct tracer_flags *flags, |
| 6264 | struct tracer_opt *opt) |
| 6265 | { |
| 6266 | struct dentry *t_options; |
| 6267 | |
| 6268 | t_options = trace_options_init_dentry(tr); |
| 6269 | if (!t_options) |
| 6270 | return; |
| 6271 | |
| 6272 | topt->flags = flags; |
| 6273 | topt->opt = opt; |
| 6274 | topt->tr = tr; |
| 6275 | |
| 6276 | topt->entry = trace_create_file(opt->name, 0644, t_options, topt, |
| 6277 | &trace_options_fops); |
| 6278 | |
| 6279 | } |
| 6280 | |
| 6281 | static struct trace_option_dentry * |
| 6282 | create_trace_option_files(struct trace_array *tr, struct tracer *tracer) |
| 6283 | { |
| 6284 | struct trace_option_dentry *topts; |
| 6285 | struct tracer_flags *flags; |
| 6286 | struct tracer_opt *opts; |
| 6287 | int cnt; |
| 6288 | |
| 6289 | if (!tracer) |
| 6290 | return NULL; |
| 6291 | |
| 6292 | flags = tracer->flags; |
| 6293 | |
| 6294 | if (!flags || !flags->opts) |
| 6295 | return NULL; |
| 6296 | |
| 6297 | opts = flags->opts; |
| 6298 | |
| 6299 | for (cnt = 0; opts[cnt].name; cnt++) |
| 6300 | ; |
| 6301 | |
| 6302 | topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL); |
| 6303 | if (!topts) |
| 6304 | return NULL; |
| 6305 | |
| 6306 | for (cnt = 0; opts[cnt].name; cnt++) { |
| 6307 | create_trace_option_file(tr, &topts[cnt], flags, |
| 6308 | &opts[cnt]); |
| 6309 | WARN_ONCE(topts[cnt].entry == NULL, |
| 6310 | "Failed to create trace option: %s", |
| 6311 | opts[cnt].name); |
| 6312 | } |
| 6313 | |
| 6314 | return topts; |
| 6315 | } |
| 6316 | |
| 6317 | static struct dentry * |
| 6318 | create_trace_option_core_file(struct trace_array *tr, |
| 6319 | const char *option, long index) |
| 6320 | { |
| 6321 | struct dentry *t_options; |
| 6322 | |
| 6323 | t_options = trace_options_init_dentry(tr); |
| 6324 | if (!t_options) |
| 6325 | return NULL; |
| 6326 | |
| 6327 | return trace_create_file(option, 0644, t_options, (void *)index, |
| 6328 | &trace_options_core_fops); |
| 6329 | } |
| 6330 | |
| 6331 | static __init void create_trace_options_dir(struct trace_array *tr) |
| 6332 | { |
| 6333 | struct dentry *t_options; |
| 6334 | int i; |
| 6335 | |
| 6336 | t_options = trace_options_init_dentry(tr); |
| 6337 | if (!t_options) |
| 6338 | return; |
| 6339 | |
| 6340 | for (i = 0; trace_options[i]; i++) |
| 6341 | create_trace_option_core_file(tr, trace_options[i], i); |
| 6342 | } |
| 6343 | |
| 6344 | static ssize_t |
| 6345 | rb_simple_read(struct file *filp, char __user *ubuf, |
| 6346 | size_t cnt, loff_t *ppos) |
| 6347 | { |
| 6348 | struct trace_array *tr = filp->private_data; |
| 6349 | char buf[64]; |
| 6350 | int r; |
| 6351 | |
| 6352 | r = tracer_tracing_is_on(tr); |
| 6353 | r = sprintf(buf, "%d\n", r); |
| 6354 | |
| 6355 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); |
| 6356 | } |
| 6357 | |
| 6358 | static ssize_t |
| 6359 | rb_simple_write(struct file *filp, const char __user *ubuf, |
| 6360 | size_t cnt, loff_t *ppos) |
| 6361 | { |
| 6362 | struct trace_array *tr = filp->private_data; |
| 6363 | struct ring_buffer *buffer = tr->trace_buffer.buffer; |
| 6364 | unsigned long val; |
| 6365 | int ret; |
| 6366 | |
| 6367 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); |
| 6368 | if (ret) |
| 6369 | return ret; |
| 6370 | |
| 6371 | if (buffer) { |
| 6372 | mutex_lock(&trace_types_lock); |
| 6373 | if (val) { |
| 6374 | tracer_tracing_on(tr); |
| 6375 | if (tr->current_trace->start) |
| 6376 | tr->current_trace->start(tr); |
| 6377 | } else { |
| 6378 | tracer_tracing_off(tr); |
| 6379 | if (tr->current_trace->stop) |
| 6380 | tr->current_trace->stop(tr); |
| 6381 | } |
| 6382 | mutex_unlock(&trace_types_lock); |
| 6383 | } |
| 6384 | |
| 6385 | (*ppos)++; |
| 6386 | |
| 6387 | return cnt; |
| 6388 | } |
| 6389 | |
| 6390 | static const struct file_operations rb_simple_fops = { |
| 6391 | .open = tracing_open_generic_tr, |
| 6392 | .read = rb_simple_read, |
| 6393 | .write = rb_simple_write, |
| 6394 | .release = tracing_release_generic_tr, |
| 6395 | .llseek = default_llseek, |
| 6396 | }; |
| 6397 | |
| 6398 | struct dentry *trace_instance_dir; |
| 6399 | |
| 6400 | static void |
| 6401 | init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer); |
| 6402 | |
| 6403 | static int |
| 6404 | allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size) |
| 6405 | { |
| 6406 | enum ring_buffer_flags rb_flags; |
| 6407 | |
| 6408 | rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0; |
| 6409 | |
| 6410 | buf->tr = tr; |
| 6411 | |
| 6412 | buf->buffer = ring_buffer_alloc(size, rb_flags); |
| 6413 | if (!buf->buffer) |
| 6414 | return -ENOMEM; |
| 6415 | |
| 6416 | buf->data = alloc_percpu(struct trace_array_cpu); |
| 6417 | if (!buf->data) { |
| 6418 | ring_buffer_free(buf->buffer); |
| 6419 | return -ENOMEM; |
| 6420 | } |
| 6421 | |
| 6422 | /* Allocate the first page for all buffers */ |
| 6423 | set_buffer_entries(&tr->trace_buffer, |
| 6424 | ring_buffer_size(tr->trace_buffer.buffer, 0)); |
| 6425 | |
| 6426 | return 0; |
| 6427 | } |
| 6428 | |
| 6429 | static int allocate_trace_buffers(struct trace_array *tr, int size) |
| 6430 | { |
| 6431 | int ret; |
| 6432 | |
| 6433 | ret = allocate_trace_buffer(tr, &tr->trace_buffer, size); |
| 6434 | if (ret) |
| 6435 | return ret; |
| 6436 | |
| 6437 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 6438 | ret = allocate_trace_buffer(tr, &tr->max_buffer, |
| 6439 | allocate_snapshot ? size : 1); |
| 6440 | if (WARN_ON(ret)) { |
| 6441 | ring_buffer_free(tr->trace_buffer.buffer); |
| 6442 | free_percpu(tr->trace_buffer.data); |
| 6443 | return -ENOMEM; |
| 6444 | } |
| 6445 | tr->allocated_snapshot = allocate_snapshot; |
| 6446 | |
| 6447 | /* |
| 6448 | * Only the top level trace array gets its snapshot allocated |
| 6449 | * from the kernel command line. |
| 6450 | */ |
| 6451 | allocate_snapshot = false; |
| 6452 | #endif |
| 6453 | return 0; |
| 6454 | } |
| 6455 | |
| 6456 | static void free_trace_buffer(struct trace_buffer *buf) |
| 6457 | { |
| 6458 | if (buf->buffer) { |
| 6459 | ring_buffer_free(buf->buffer); |
| 6460 | buf->buffer = NULL; |
| 6461 | free_percpu(buf->data); |
| 6462 | buf->data = NULL; |
| 6463 | } |
| 6464 | } |
| 6465 | |
| 6466 | static void free_trace_buffers(struct trace_array *tr) |
| 6467 | { |
| 6468 | if (!tr) |
| 6469 | return; |
| 6470 | |
| 6471 | free_trace_buffer(&tr->trace_buffer); |
| 6472 | |
| 6473 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 6474 | free_trace_buffer(&tr->max_buffer); |
| 6475 | #endif |
| 6476 | } |
| 6477 | |
| 6478 | static int instance_mkdir(const char *name) |
| 6479 | { |
| 6480 | struct trace_array *tr; |
| 6481 | int ret; |
| 6482 | |
| 6483 | mutex_lock(&trace_types_lock); |
| 6484 | |
| 6485 | ret = -EEXIST; |
| 6486 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { |
| 6487 | if (tr->name && strcmp(tr->name, name) == 0) |
| 6488 | goto out_unlock; |
| 6489 | } |
| 6490 | |
| 6491 | ret = -ENOMEM; |
| 6492 | tr = kzalloc(sizeof(*tr), GFP_KERNEL); |
| 6493 | if (!tr) |
| 6494 | goto out_unlock; |
| 6495 | |
| 6496 | tr->name = kstrdup(name, GFP_KERNEL); |
| 6497 | if (!tr->name) |
| 6498 | goto out_free_tr; |
| 6499 | |
| 6500 | if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL)) |
| 6501 | goto out_free_tr; |
| 6502 | |
| 6503 | cpumask_copy(tr->tracing_cpumask, cpu_all_mask); |
| 6504 | |
| 6505 | raw_spin_lock_init(&tr->start_lock); |
| 6506 | |
| 6507 | tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
| 6508 | |
| 6509 | tr->current_trace = &nop_trace; |
| 6510 | |
| 6511 | INIT_LIST_HEAD(&tr->systems); |
| 6512 | INIT_LIST_HEAD(&tr->events); |
| 6513 | |
| 6514 | if (allocate_trace_buffers(tr, trace_buf_size) < 0) |
| 6515 | goto out_free_tr; |
| 6516 | |
| 6517 | tr->dir = tracefs_create_dir(name, trace_instance_dir); |
| 6518 | if (!tr->dir) |
| 6519 | goto out_free_tr; |
| 6520 | |
| 6521 | ret = event_trace_add_tracer(tr->dir, tr); |
| 6522 | if (ret) { |
| 6523 | tracefs_remove_recursive(tr->dir); |
| 6524 | goto out_free_tr; |
| 6525 | } |
| 6526 | |
| 6527 | init_tracer_tracefs(tr, tr->dir); |
| 6528 | |
| 6529 | list_add(&tr->list, &ftrace_trace_arrays); |
| 6530 | |
| 6531 | mutex_unlock(&trace_types_lock); |
| 6532 | |
| 6533 | return 0; |
| 6534 | |
| 6535 | out_free_tr: |
| 6536 | free_trace_buffers(tr); |
| 6537 | free_cpumask_var(tr->tracing_cpumask); |
| 6538 | kfree(tr->name); |
| 6539 | kfree(tr); |
| 6540 | |
| 6541 | out_unlock: |
| 6542 | mutex_unlock(&trace_types_lock); |
| 6543 | |
| 6544 | return ret; |
| 6545 | |
| 6546 | } |
| 6547 | |
| 6548 | static int instance_rmdir(const char *name) |
| 6549 | { |
| 6550 | struct trace_array *tr; |
| 6551 | int found = 0; |
| 6552 | int ret; |
| 6553 | |
| 6554 | mutex_lock(&trace_types_lock); |
| 6555 | |
| 6556 | ret = -ENODEV; |
| 6557 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { |
| 6558 | if (tr->name && strcmp(tr->name, name) == 0) { |
| 6559 | found = 1; |
| 6560 | break; |
| 6561 | } |
| 6562 | } |
| 6563 | if (!found) |
| 6564 | goto out_unlock; |
| 6565 | |
| 6566 | ret = -EBUSY; |
| 6567 | if (tr->ref || (tr->current_trace && tr->current_trace->ref)) |
| 6568 | goto out_unlock; |
| 6569 | |
| 6570 | list_del(&tr->list); |
| 6571 | |
| 6572 | tracing_set_nop(tr); |
| 6573 | event_trace_del_tracer(tr); |
| 6574 | ftrace_destroy_function_files(tr); |
| 6575 | debugfs_remove_recursive(tr->dir); |
| 6576 | free_trace_buffers(tr); |
| 6577 | |
| 6578 | kfree(tr->name); |
| 6579 | kfree(tr); |
| 6580 | |
| 6581 | ret = 0; |
| 6582 | |
| 6583 | out_unlock: |
| 6584 | mutex_unlock(&trace_types_lock); |
| 6585 | |
| 6586 | return ret; |
| 6587 | } |
| 6588 | |
| 6589 | static __init void create_trace_instances(struct dentry *d_tracer) |
| 6590 | { |
| 6591 | trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer, |
| 6592 | instance_mkdir, |
| 6593 | instance_rmdir); |
| 6594 | if (WARN_ON(!trace_instance_dir)) |
| 6595 | return; |
| 6596 | } |
| 6597 | |
| 6598 | static void |
| 6599 | init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer) |
| 6600 | { |
| 6601 | int cpu; |
| 6602 | |
| 6603 | trace_create_file("available_tracers", 0444, d_tracer, |
| 6604 | tr, &show_traces_fops); |
| 6605 | |
| 6606 | trace_create_file("current_tracer", 0644, d_tracer, |
| 6607 | tr, &set_tracer_fops); |
| 6608 | |
| 6609 | trace_create_file("tracing_cpumask", 0644, d_tracer, |
| 6610 | tr, &tracing_cpumask_fops); |
| 6611 | |
| 6612 | trace_create_file("trace_options", 0644, d_tracer, |
| 6613 | tr, &tracing_iter_fops); |
| 6614 | |
| 6615 | trace_create_file("trace", 0644, d_tracer, |
| 6616 | tr, &tracing_fops); |
| 6617 | |
| 6618 | trace_create_file("trace_pipe", 0444, d_tracer, |
| 6619 | tr, &tracing_pipe_fops); |
| 6620 | |
| 6621 | trace_create_file("buffer_size_kb", 0644, d_tracer, |
| 6622 | tr, &tracing_entries_fops); |
| 6623 | |
| 6624 | trace_create_file("buffer_total_size_kb", 0444, d_tracer, |
| 6625 | tr, &tracing_total_entries_fops); |
| 6626 | |
| 6627 | trace_create_file("free_buffer", 0200, d_tracer, |
| 6628 | tr, &tracing_free_buffer_fops); |
| 6629 | |
| 6630 | trace_create_file("trace_marker", 0220, d_tracer, |
| 6631 | tr, &tracing_mark_fops); |
| 6632 | |
| 6633 | trace_create_file("trace_clock", 0644, d_tracer, tr, |
| 6634 | &trace_clock_fops); |
| 6635 | |
| 6636 | trace_create_file("tracing_on", 0644, d_tracer, |
| 6637 | tr, &rb_simple_fops); |
| 6638 | |
| 6639 | #ifdef CONFIG_TRACER_MAX_TRACE |
| 6640 | trace_create_file("tracing_max_latency", 0644, d_tracer, |
| 6641 | &tr->max_latency, &tracing_max_lat_fops); |
| 6642 | #endif |
| 6643 | |
| 6644 | if (ftrace_create_function_files(tr, d_tracer)) |
| 6645 | WARN(1, "Could not allocate function filter files"); |
| 6646 | |
| 6647 | #ifdef CONFIG_TRACER_SNAPSHOT |
| 6648 | trace_create_file("snapshot", 0644, d_tracer, |
| 6649 | tr, &snapshot_fops); |
| 6650 | #endif |
| 6651 | |
| 6652 | for_each_tracing_cpu(cpu) |
| 6653 | tracing_init_tracefs_percpu(tr, cpu); |
| 6654 | |
| 6655 | } |
| 6656 | |
| 6657 | static struct vfsmount *trace_automount(void *ingore) |
| 6658 | { |
| 6659 | struct vfsmount *mnt; |
| 6660 | struct file_system_type *type; |
| 6661 | |
| 6662 | /* |
| 6663 | * To maintain backward compatibility for tools that mount |
| 6664 | * debugfs to get to the tracing facility, tracefs is automatically |
| 6665 | * mounted to the debugfs/tracing directory. |
| 6666 | */ |
| 6667 | type = get_fs_type("tracefs"); |
| 6668 | if (!type) |
| 6669 | return NULL; |
| 6670 | mnt = vfs_kern_mount(type, 0, "tracefs", NULL); |
| 6671 | put_filesystem(type); |
| 6672 | if (IS_ERR(mnt)) |
| 6673 | return NULL; |
| 6674 | mntget(mnt); |
| 6675 | |
| 6676 | return mnt; |
| 6677 | } |
| 6678 | |
| 6679 | /** |
| 6680 | * tracing_init_dentry - initialize top level trace array |
| 6681 | * |
| 6682 | * This is called when creating files or directories in the tracing |
| 6683 | * directory. It is called via fs_initcall() by any of the boot up code |
| 6684 | * and expects to return the dentry of the top level tracing directory. |
| 6685 | */ |
| 6686 | struct dentry *tracing_init_dentry(void) |
| 6687 | { |
| 6688 | struct trace_array *tr = &global_trace; |
| 6689 | |
| 6690 | /* The top level trace array uses NULL as parent */ |
| 6691 | if (tr->dir) |
| 6692 | return NULL; |
| 6693 | |
| 6694 | if (WARN_ON(!debugfs_initialized())) |
| 6695 | return ERR_PTR(-ENODEV); |
| 6696 | |
| 6697 | /* |
| 6698 | * As there may still be users that expect the tracing |
| 6699 | * files to exist in debugfs/tracing, we must automount |
| 6700 | * the tracefs file system there, so older tools still |
| 6701 | * work with the newer kerenl. |
| 6702 | */ |
| 6703 | tr->dir = debugfs_create_automount("tracing", NULL, |
| 6704 | trace_automount, NULL); |
| 6705 | if (!tr->dir) { |
| 6706 | pr_warn_once("Could not create debugfs directory 'tracing'\n"); |
| 6707 | return ERR_PTR(-ENOMEM); |
| 6708 | } |
| 6709 | |
| 6710 | return NULL; |
| 6711 | } |
| 6712 | |
| 6713 | extern struct trace_enum_map *__start_ftrace_enum_maps[]; |
| 6714 | extern struct trace_enum_map *__stop_ftrace_enum_maps[]; |
| 6715 | |
| 6716 | static void __init trace_enum_init(void) |
| 6717 | { |
| 6718 | int len; |
| 6719 | |
| 6720 | len = __stop_ftrace_enum_maps - __start_ftrace_enum_maps; |
| 6721 | trace_insert_enum_map(NULL, __start_ftrace_enum_maps, len); |
| 6722 | } |
| 6723 | |
| 6724 | #ifdef CONFIG_MODULES |
| 6725 | static void trace_module_add_enums(struct module *mod) |
| 6726 | { |
| 6727 | if (!mod->num_trace_enums) |
| 6728 | return; |
| 6729 | |
| 6730 | /* |
| 6731 | * Modules with bad taint do not have events created, do |
| 6732 | * not bother with enums either. |
| 6733 | */ |
| 6734 | if (trace_module_has_bad_taint(mod)) |
| 6735 | return; |
| 6736 | |
| 6737 | trace_insert_enum_map(mod, mod->trace_enums, mod->num_trace_enums); |
| 6738 | } |
| 6739 | |
| 6740 | #ifdef CONFIG_TRACE_ENUM_MAP_FILE |
| 6741 | static void trace_module_remove_enums(struct module *mod) |
| 6742 | { |
| 6743 | union trace_enum_map_item *map; |
| 6744 | union trace_enum_map_item **last = &trace_enum_maps; |
| 6745 | |
| 6746 | if (!mod->num_trace_enums) |
| 6747 | return; |
| 6748 | |
| 6749 | mutex_lock(&trace_enum_mutex); |
| 6750 | |
| 6751 | map = trace_enum_maps; |
| 6752 | |
| 6753 | while (map) { |
| 6754 | if (map->head.mod == mod) |
| 6755 | break; |
| 6756 | map = trace_enum_jmp_to_tail(map); |
| 6757 | last = &map->tail.next; |
| 6758 | map = map->tail.next; |
| 6759 | } |
| 6760 | if (!map) |
| 6761 | goto out; |
| 6762 | |
| 6763 | *last = trace_enum_jmp_to_tail(map)->tail.next; |
| 6764 | kfree(map); |
| 6765 | out: |
| 6766 | mutex_unlock(&trace_enum_mutex); |
| 6767 | } |
| 6768 | #else |
| 6769 | static inline void trace_module_remove_enums(struct module *mod) { } |
| 6770 | #endif /* CONFIG_TRACE_ENUM_MAP_FILE */ |
| 6771 | |
| 6772 | static int trace_module_notify(struct notifier_block *self, |
| 6773 | unsigned long val, void *data) |
| 6774 | { |
| 6775 | struct module *mod = data; |
| 6776 | |
| 6777 | switch (val) { |
| 6778 | case MODULE_STATE_COMING: |
| 6779 | trace_module_add_enums(mod); |
| 6780 | break; |
| 6781 | case MODULE_STATE_GOING: |
| 6782 | trace_module_remove_enums(mod); |
| 6783 | break; |
| 6784 | } |
| 6785 | |
| 6786 | return 0; |
| 6787 | } |
| 6788 | |
| 6789 | static struct notifier_block trace_module_nb = { |
| 6790 | .notifier_call = trace_module_notify, |
| 6791 | .priority = 0, |
| 6792 | }; |
| 6793 | #endif /* CONFIG_MODULES */ |
| 6794 | |
| 6795 | static __init int tracer_init_tracefs(void) |
| 6796 | { |
| 6797 | struct dentry *d_tracer; |
| 6798 | struct tracer *t; |
| 6799 | |
| 6800 | trace_access_lock_init(); |
| 6801 | |
| 6802 | d_tracer = tracing_init_dentry(); |
| 6803 | if (IS_ERR(d_tracer)) |
| 6804 | return 0; |
| 6805 | |
| 6806 | init_tracer_tracefs(&global_trace, d_tracer); |
| 6807 | |
| 6808 | trace_create_file("tracing_thresh", 0644, d_tracer, |
| 6809 | &global_trace, &tracing_thresh_fops); |
| 6810 | |
| 6811 | trace_create_file("README", 0444, d_tracer, |
| 6812 | NULL, &tracing_readme_fops); |
| 6813 | |
| 6814 | trace_create_file("saved_cmdlines", 0444, d_tracer, |
| 6815 | NULL, &tracing_saved_cmdlines_fops); |
| 6816 | |
| 6817 | trace_create_file("saved_cmdlines_size", 0644, d_tracer, |
| 6818 | NULL, &tracing_saved_cmdlines_size_fops); |
| 6819 | |
| 6820 | trace_enum_init(); |
| 6821 | |
| 6822 | trace_create_enum_file(d_tracer); |
| 6823 | |
| 6824 | #ifdef CONFIG_MODULES |
| 6825 | register_module_notifier(&trace_module_nb); |
| 6826 | #endif |
| 6827 | |
| 6828 | #ifdef CONFIG_DYNAMIC_FTRACE |
| 6829 | trace_create_file("dyn_ftrace_total_info", 0444, d_tracer, |
| 6830 | &ftrace_update_tot_cnt, &tracing_dyn_info_fops); |
| 6831 | #endif |
| 6832 | |
| 6833 | create_trace_instances(d_tracer); |
| 6834 | |
| 6835 | create_trace_options_dir(&global_trace); |
| 6836 | |
| 6837 | mutex_lock(&trace_types_lock); |
| 6838 | for (t = trace_types; t; t = t->next) |
| 6839 | add_tracer_options(&global_trace, t); |
| 6840 | mutex_unlock(&trace_types_lock); |
| 6841 | |
| 6842 | return 0; |
| 6843 | } |
| 6844 | |
| 6845 | static int trace_panic_handler(struct notifier_block *this, |
| 6846 | unsigned long event, void *unused) |
| 6847 | { |
| 6848 | if (ftrace_dump_on_oops) |
| 6849 | ftrace_dump(ftrace_dump_on_oops); |
| 6850 | return NOTIFY_OK; |
| 6851 | } |
| 6852 | |
| 6853 | static struct notifier_block trace_panic_notifier = { |
| 6854 | .notifier_call = trace_panic_handler, |
| 6855 | .next = NULL, |
| 6856 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ |
| 6857 | }; |
| 6858 | |
| 6859 | static int trace_die_handler(struct notifier_block *self, |
| 6860 | unsigned long val, |
| 6861 | void *data) |
| 6862 | { |
| 6863 | switch (val) { |
| 6864 | case DIE_OOPS: |
| 6865 | if (ftrace_dump_on_oops) |
| 6866 | ftrace_dump(ftrace_dump_on_oops); |
| 6867 | break; |
| 6868 | default: |
| 6869 | break; |
| 6870 | } |
| 6871 | return NOTIFY_OK; |
| 6872 | } |
| 6873 | |
| 6874 | static struct notifier_block trace_die_notifier = { |
| 6875 | .notifier_call = trace_die_handler, |
| 6876 | .priority = 200 |
| 6877 | }; |
| 6878 | |
| 6879 | /* |
| 6880 | * printk is set to max of 1024, we really don't need it that big. |
| 6881 | * Nothing should be printing 1000 characters anyway. |
| 6882 | */ |
| 6883 | #define TRACE_MAX_PRINT 1000 |
| 6884 | |
| 6885 | /* |
| 6886 | * Define here KERN_TRACE so that we have one place to modify |
| 6887 | * it if we decide to change what log level the ftrace dump |
| 6888 | * should be at. |
| 6889 | */ |
| 6890 | #define KERN_TRACE KERN_EMERG |
| 6891 | |
| 6892 | void |
| 6893 | trace_printk_seq(struct trace_seq *s) |
| 6894 | { |
| 6895 | /* Probably should print a warning here. */ |
| 6896 | if (s->seq.len >= TRACE_MAX_PRINT) |
| 6897 | s->seq.len = TRACE_MAX_PRINT; |
| 6898 | |
| 6899 | /* |
| 6900 | * More paranoid code. Although the buffer size is set to |
| 6901 | * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just |
| 6902 | * an extra layer of protection. |
| 6903 | */ |
| 6904 | if (WARN_ON_ONCE(s->seq.len >= s->seq.size)) |
| 6905 | s->seq.len = s->seq.size - 1; |
| 6906 | |
| 6907 | /* should be zero ended, but we are paranoid. */ |
| 6908 | s->buffer[s->seq.len] = 0; |
| 6909 | |
| 6910 | printk(KERN_TRACE "%s", s->buffer); |
| 6911 | |
| 6912 | trace_seq_init(s); |
| 6913 | } |
| 6914 | |
| 6915 | void trace_init_global_iter(struct trace_iterator *iter) |
| 6916 | { |
| 6917 | iter->tr = &global_trace; |
| 6918 | iter->trace = iter->tr->current_trace; |
| 6919 | iter->cpu_file = RING_BUFFER_ALL_CPUS; |
| 6920 | iter->trace_buffer = &global_trace.trace_buffer; |
| 6921 | |
| 6922 | if (iter->trace && iter->trace->open) |
| 6923 | iter->trace->open(iter); |
| 6924 | |
| 6925 | /* Annotate start of buffers if we had overruns */ |
| 6926 | if (ring_buffer_overruns(iter->trace_buffer->buffer)) |
| 6927 | iter->iter_flags |= TRACE_FILE_ANNOTATE; |
| 6928 | |
| 6929 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ |
| 6930 | if (trace_clocks[iter->tr->clock_id].in_ns) |
| 6931 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; |
| 6932 | } |
| 6933 | |
| 6934 | void ftrace_dump(enum ftrace_dump_mode oops_dump_mode) |
| 6935 | { |
| 6936 | /* use static because iter can be a bit big for the stack */ |
| 6937 | static struct trace_iterator iter; |
| 6938 | static atomic_t dump_running; |
| 6939 | unsigned int old_userobj; |
| 6940 | unsigned long flags; |
| 6941 | int cnt = 0, cpu; |
| 6942 | |
| 6943 | /* Only allow one dump user at a time. */ |
| 6944 | if (atomic_inc_return(&dump_running) != 1) { |
| 6945 | atomic_dec(&dump_running); |
| 6946 | return; |
| 6947 | } |
| 6948 | |
| 6949 | /* |
| 6950 | * Always turn off tracing when we dump. |
| 6951 | * We don't need to show trace output of what happens |
| 6952 | * between multiple crashes. |
| 6953 | * |
| 6954 | * If the user does a sysrq-z, then they can re-enable |
| 6955 | * tracing with echo 1 > tracing_on. |
| 6956 | */ |
| 6957 | tracing_off(); |
| 6958 | |
| 6959 | local_irq_save(flags); |
| 6960 | |
| 6961 | /* Simulate the iterator */ |
| 6962 | trace_init_global_iter(&iter); |
| 6963 | |
| 6964 | for_each_tracing_cpu(cpu) { |
| 6965 | atomic_inc(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled); |
| 6966 | } |
| 6967 | |
| 6968 | old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ; |
| 6969 | |
| 6970 | /* don't look at user memory in panic mode */ |
| 6971 | trace_flags &= ~TRACE_ITER_SYM_USEROBJ; |
| 6972 | |
| 6973 | switch (oops_dump_mode) { |
| 6974 | case DUMP_ALL: |
| 6975 | iter.cpu_file = RING_BUFFER_ALL_CPUS; |
| 6976 | break; |
| 6977 | case DUMP_ORIG: |
| 6978 | iter.cpu_file = raw_smp_processor_id(); |
| 6979 | break; |
| 6980 | case DUMP_NONE: |
| 6981 | goto out_enable; |
| 6982 | default: |
| 6983 | printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n"); |
| 6984 | iter.cpu_file = RING_BUFFER_ALL_CPUS; |
| 6985 | } |
| 6986 | |
| 6987 | printk(KERN_TRACE "Dumping ftrace buffer:\n"); |
| 6988 | |
| 6989 | /* Did function tracer already get disabled? */ |
| 6990 | if (ftrace_is_dead()) { |
| 6991 | printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n"); |
| 6992 | printk("# MAY BE MISSING FUNCTION EVENTS\n"); |
| 6993 | } |
| 6994 | |
| 6995 | /* |
| 6996 | * We need to stop all tracing on all CPUS to read the |
| 6997 | * the next buffer. This is a bit expensive, but is |
| 6998 | * not done often. We fill all what we can read, |
| 6999 | * and then release the locks again. |
| 7000 | */ |
| 7001 | |
| 7002 | while (!trace_empty(&iter)) { |
| 7003 | |
| 7004 | if (!cnt) |
| 7005 | printk(KERN_TRACE "---------------------------------\n"); |
| 7006 | |
| 7007 | cnt++; |
| 7008 | |
| 7009 | /* reset all but tr, trace, and overruns */ |
| 7010 | memset(&iter.seq, 0, |
| 7011 | sizeof(struct trace_iterator) - |
| 7012 | offsetof(struct trace_iterator, seq)); |
| 7013 | iter.iter_flags |= TRACE_FILE_LAT_FMT; |
| 7014 | iter.pos = -1; |
| 7015 | |
| 7016 | if (trace_find_next_entry_inc(&iter) != NULL) { |
| 7017 | int ret; |
| 7018 | |
| 7019 | ret = print_trace_line(&iter); |
| 7020 | if (ret != TRACE_TYPE_NO_CONSUME) |
| 7021 | trace_consume(&iter); |
| 7022 | } |
| 7023 | touch_nmi_watchdog(); |
| 7024 | |
| 7025 | trace_printk_seq(&iter.seq); |
| 7026 | } |
| 7027 | |
| 7028 | if (!cnt) |
| 7029 | printk(KERN_TRACE " (ftrace buffer empty)\n"); |
| 7030 | else |
| 7031 | printk(KERN_TRACE "---------------------------------\n"); |
| 7032 | |
| 7033 | out_enable: |
| 7034 | trace_flags |= old_userobj; |
| 7035 | |
| 7036 | for_each_tracing_cpu(cpu) { |
| 7037 | atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled); |
| 7038 | } |
| 7039 | atomic_dec(&dump_running); |
| 7040 | local_irq_restore(flags); |
| 7041 | } |
| 7042 | EXPORT_SYMBOL_GPL(ftrace_dump); |
| 7043 | |
| 7044 | __init static int tracer_alloc_buffers(void) |
| 7045 | { |
| 7046 | int ring_buf_size; |
| 7047 | int ret = -ENOMEM; |
| 7048 | |
| 7049 | /* |
| 7050 | * Make sure we don't accidently add more trace options |
| 7051 | * than we have bits for. |
| 7052 | */ |
| 7053 | BUILD_BUG_ON(TRACE_ITER_LAST_BIT > 32); |
| 7054 | |
| 7055 | if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL)) |
| 7056 | goto out; |
| 7057 | |
| 7058 | if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL)) |
| 7059 | goto out_free_buffer_mask; |
| 7060 | |
| 7061 | /* Only allocate trace_printk buffers if a trace_printk exists */ |
| 7062 | if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt) |
| 7063 | /* Must be called before global_trace.buffer is allocated */ |
| 7064 | trace_printk_init_buffers(); |
| 7065 | |
| 7066 | /* To save memory, keep the ring buffer size to its minimum */ |
| 7067 | if (ring_buffer_expanded) |
| 7068 | ring_buf_size = trace_buf_size; |
| 7069 | else |
| 7070 | ring_buf_size = 1; |
| 7071 | |
| 7072 | cpumask_copy(tracing_buffer_mask, cpu_possible_mask); |
| 7073 | cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask); |
| 7074 | |
| 7075 | raw_spin_lock_init(&global_trace.start_lock); |
| 7076 | |
| 7077 | /* Used for event triggers */ |
| 7078 | temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE); |
| 7079 | if (!temp_buffer) |
| 7080 | goto out_free_cpumask; |
| 7081 | |
| 7082 | if (trace_create_savedcmd() < 0) |
| 7083 | goto out_free_temp_buffer; |
| 7084 | |
| 7085 | /* TODO: make the number of buffers hot pluggable with CPUS */ |
| 7086 | if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) { |
| 7087 | printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); |
| 7088 | WARN_ON(1); |
| 7089 | goto out_free_savedcmd; |
| 7090 | } |
| 7091 | |
| 7092 | if (global_trace.buffer_disabled) |
| 7093 | tracing_off(); |
| 7094 | |
| 7095 | if (trace_boot_clock) { |
| 7096 | ret = tracing_set_clock(&global_trace, trace_boot_clock); |
| 7097 | if (ret < 0) |
| 7098 | pr_warning("Trace clock %s not defined, going back to default\n", |
| 7099 | trace_boot_clock); |
| 7100 | } |
| 7101 | |
| 7102 | /* |
| 7103 | * register_tracer() might reference current_trace, so it |
| 7104 | * needs to be set before we register anything. This is |
| 7105 | * just a bootstrap of current_trace anyway. |
| 7106 | */ |
| 7107 | global_trace.current_trace = &nop_trace; |
| 7108 | |
| 7109 | global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
| 7110 | |
| 7111 | ftrace_init_global_array_ops(&global_trace); |
| 7112 | |
| 7113 | register_tracer(&nop_trace); |
| 7114 | |
| 7115 | /* All seems OK, enable tracing */ |
| 7116 | tracing_disabled = 0; |
| 7117 | |
| 7118 | atomic_notifier_chain_register(&panic_notifier_list, |
| 7119 | &trace_panic_notifier); |
| 7120 | |
| 7121 | register_die_notifier(&trace_die_notifier); |
| 7122 | |
| 7123 | global_trace.flags = TRACE_ARRAY_FL_GLOBAL; |
| 7124 | |
| 7125 | INIT_LIST_HEAD(&global_trace.systems); |
| 7126 | INIT_LIST_HEAD(&global_trace.events); |
| 7127 | list_add(&global_trace.list, &ftrace_trace_arrays); |
| 7128 | |
| 7129 | while (trace_boot_options) { |
| 7130 | char *option; |
| 7131 | |
| 7132 | option = strsep(&trace_boot_options, ","); |
| 7133 | trace_set_options(&global_trace, option); |
| 7134 | } |
| 7135 | |
| 7136 | register_snapshot_cmd(); |
| 7137 | |
| 7138 | return 0; |
| 7139 | |
| 7140 | out_free_savedcmd: |
| 7141 | free_saved_cmdlines_buffer(savedcmd); |
| 7142 | out_free_temp_buffer: |
| 7143 | ring_buffer_free(temp_buffer); |
| 7144 | out_free_cpumask: |
| 7145 | free_cpumask_var(global_trace.tracing_cpumask); |
| 7146 | out_free_buffer_mask: |
| 7147 | free_cpumask_var(tracing_buffer_mask); |
| 7148 | out: |
| 7149 | return ret; |
| 7150 | } |
| 7151 | |
| 7152 | void __init trace_init(void) |
| 7153 | { |
| 7154 | if (tracepoint_printk) { |
| 7155 | tracepoint_print_iter = |
| 7156 | kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL); |
| 7157 | if (WARN_ON(!tracepoint_print_iter)) |
| 7158 | tracepoint_printk = 0; |
| 7159 | } |
| 7160 | tracer_alloc_buffers(); |
| 7161 | trace_event_init(); |
| 7162 | } |
| 7163 | |
| 7164 | __init static int clear_boot_tracer(void) |
| 7165 | { |
| 7166 | /* |
| 7167 | * The default tracer at boot buffer is an init section. |
| 7168 | * This function is called in lateinit. If we did not |
| 7169 | * find the boot tracer, then clear it out, to prevent |
| 7170 | * later registration from accessing the buffer that is |
| 7171 | * about to be freed. |
| 7172 | */ |
| 7173 | if (!default_bootup_tracer) |
| 7174 | return 0; |
| 7175 | |
| 7176 | printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n", |
| 7177 | default_bootup_tracer); |
| 7178 | default_bootup_tracer = NULL; |
| 7179 | |
| 7180 | return 0; |
| 7181 | } |
| 7182 | |
| 7183 | fs_initcall(tracer_init_tracefs); |
| 7184 | late_initcall(clear_boot_tracer); |