trace/simple.c

   1 /*
   2  * Simple trace backend
   3  *
   4  * Copyright IBM, Corp. 2010
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2.  See
   7  * the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include <stdlib.h>
  12 #include <stdint.h>
  13 #include <stdio.h>
  14 #include <time.h>
  15 #ifndef _WIN32
  16 #include <signal.h>
  17 #include <pthread.h>
  18 #endif
  19 #include "qemu/timer.h"
  20 #include "trace.h"
  21 #include "trace/control.h"
  22
  23 /** Trace file header event ID */
  24 #define HEADER_EVENT_ID (~(uint64_t)0) /* avoids conflicting with TraceEventIDs */
  25
  26 /** Trace file magic number */
  27 #define HEADER_MAGIC 0xf2b177cb0aa429b4ULL
  28
  29 /** Trace file version number, bump if format changes */
  30 #define HEADER_VERSION 2
  31
  32 /** Records were dropped event ID */
  33 #define DROPPED_EVENT_ID (~(uint64_t)0 - 1)
  34
  35 /** Trace record is valid */
  36 #define TRACE_RECORD_VALID ((uint64_t)1 << 63)
  37
  38 /*
  39  * Trace records are written out by a dedicated thread.  The thread waits for
  40  * records to become available, writes them out, and then waits again.
  41  */
  42 static GStaticMutex trace_lock = G_STATIC_MUTEX_INIT;
  43
  44 /* g_cond_new() was deprecated in glib 2.31 but we still need to support it */
  45 #if GLIB_CHECK_VERSION(2, 31, 0)
  46 static GCond the_trace_available_cond;
  47 static GCond the_trace_empty_cond;
  48 static GCond *trace_available_cond = &the_trace_available_cond;
  49 static GCond *trace_empty_cond = &the_trace_empty_cond;
  50 #else
  51 static GCond *trace_available_cond;
  52 static GCond *trace_empty_cond;
  53 #endif
  54
  55 static bool trace_available;
  56 static bool trace_writeout_enabled;
  57
  58 enum {
  59     TRACE_BUF_LEN = 4096 * 64,
  60     TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
  61 };
  62
  63 uint8_t trace_buf[TRACE_BUF_LEN];
  64 static volatile gint trace_idx;
  65 static unsigned int writeout_idx;
  66 static volatile gint dropped_events;
  67 static FILE *trace_fp;
  68 static char *trace_file_name;
  69
  70 /* * Trace buffer entry */
  71 typedef struct {
  72     uint64_t event; /*   TraceEventID */
  73     uint64_t timestamp_ns;
  74     uint32_t length;   /*    in bytes */
  75     uint32_t reserved; /*    unused */
  76     uint64_t arguments[];
  77 } TraceRecord;
  78
  79 typedef struct {
  80     uint64_t header_event_id; /* HEADER_EVENT_ID */
  81     uint64_t header_magic;    /* HEADER_MAGIC    */
  82     uint64_t header_version;  /* HEADER_VERSION  */
  83 } TraceLogHeader;
  84
  85
  86 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
  87 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
  88
  89 static void clear_buffer_range(unsigned int idx, size_t len)
  90 {
  91     uint32_t num = 0;
  92     while (num < len) {
  93         if (idx >= TRACE_BUF_LEN) {
  94             idx = idx % TRACE_BUF_LEN;
  95         }
  96         trace_buf[idx++] = 0;
  97         num++;
  98     }
  99 }
 100 /**
 101  * Read a trace record from the trace buffer
 102  *
 103  * @idx         Trace buffer index
 104  * @record      Trace record to fill
 105  *
 106  * Returns false if the record is not valid.
 107  */
 108 static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
 109 {
 110     uint64_t event_flag = 0;
 111     TraceRecord record;
 112     /* read the event flag to see if its a valid record */
 113     read_from_buffer(idx, &record, sizeof(event_flag));
 114
 115     if (!(record.event & TRACE_RECORD_VALID)) {
 116         return false;
 117     }
 118
 119     smp_rmb(); /* read memory barrier before accessing record */
 120     /* read the record header to know record length */
 121     read_from_buffer(idx, &record, sizeof(TraceRecord));
 122     *recordptr = malloc(record.length); /* dont use g_malloc, can deadlock when traced */
 123     /* make a copy of record to avoid being overwritten */
 124     read_from_buffer(idx, *recordptr, record.length);
 125     smp_rmb(); /* memory barrier before clearing valid flag */
 126     (*recordptr)->event &= ~TRACE_RECORD_VALID;
 127     /* clear the trace buffer range for consumed record otherwise any byte
 128      * with its MSB set may be considered as a valid event id when the writer
 129      * thread crosses this range of buffer again.
 130      */
 131     clear_buffer_range(idx, record.length);
 132     return true;
 133 }
 134
 135 /**
 136  * Kick writeout thread
 137  *
 138  * @wait        Whether to wait for writeout thread to complete
 139  */
 140 static void flush_trace_file(bool wait)
 141 {
 142     g_static_mutex_lock(&trace_lock);
 143     trace_available = true;
 144     g_cond_signal(trace_available_cond);
 145
 146     if (wait) {
 147         g_cond_wait(trace_empty_cond, g_static_mutex_get_mutex(&trace_lock));
 148     }
 149
 150     g_static_mutex_unlock(&trace_lock);
 151 }
 152
 153 static void wait_for_trace_records_available(void)
 154 {
 155     g_static_mutex_lock(&trace_lock);
 156     while (!(trace_available && trace_writeout_enabled)) {
 157         g_cond_signal(trace_empty_cond);
 158         g_cond_wait(trace_available_cond,
 159                     g_static_mutex_get_mutex(&trace_lock));
 160     }
 161     trace_available = false;
 162     g_static_mutex_unlock(&trace_lock);
 163 }
 164
 165 static gpointer writeout_thread(gpointer opaque)
 166 {
 167     TraceRecord *recordptr;
 168     union {
 169         TraceRecord rec;
 170         uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
 171     } dropped;
 172     unsigned int idx = 0;
 173     int dropped_count;
 174     size_t unused __attribute__ ((unused));
 175
 176     for (;;) {
 177         wait_for_trace_records_available();
 178
 179         if (g_atomic_int_get(&dropped_events)) {
 180             dropped.rec.event = DROPPED_EVENT_ID,
 181             dropped.rec.timestamp_ns = get_clock();
 182             dropped.rec.length = sizeof(TraceRecord) + sizeof(uint64_t),
 183             dropped.rec.reserved = 0;
 184             do {
 185                 dropped_count = g_atomic_int_get(&dropped_events);
 186             } while (!g_atomic_int_compare_and_exchange(&dropped_events,
 187                                                         dropped_count, 0));
 188             dropped.rec.arguments[0] = dropped_count;
 189             unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
 190         }
 191
 192         while (get_trace_record(idx, &recordptr)) {
 193             unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
 194             writeout_idx += recordptr->length;
 195             free(recordptr); /* dont use g_free, can deadlock when traced */
 196             idx = writeout_idx % TRACE_BUF_LEN;
 197         }
 198
 199         fflush(trace_fp);
 200     }
 201     return NULL;
 202 }
 203
 204 void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
 205 {
 206     rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
 207 }
 208
 209 void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
 210 {
 211     /* Write string length first */
 212     rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
 213     /* Write actual string now */
 214     rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
 215 }
 216
 217 int trace_record_start(TraceBufferRecord *rec, TraceEventID event, size_t datasize)
 218 {
 219     unsigned int idx, rec_off, old_idx, new_idx;
 220     uint32_t rec_len = sizeof(TraceRecord) + datasize;
 221     uint64_t event_u64 = event;
 222     uint64_t timestamp_ns = get_clock();
 223
 224     do {
 225         old_idx = g_atomic_int_get(&trace_idx);
 226         smp_rmb();
 227         new_idx = old_idx + rec_len;
 228
 229         if (new_idx - writeout_idx > TRACE_BUF_LEN) {
 230             /* Trace Buffer Full, Event dropped ! */
 231             g_atomic_int_inc(&dropped_events);
 232             return -ENOSPC;
 233         }
 234     } while (!g_atomic_int_compare_and_exchange(&trace_idx, old_idx, new_idx));
 235
 236     idx = old_idx % TRACE_BUF_LEN;
 237
 238     rec_off = idx;
 239     rec_off = write_to_buffer(rec_off, &event_u64, sizeof(event_u64));
 240     rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
 241     rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
 242
 243     rec->tbuf_idx = idx;
 244     rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
 245     return 0;
 246 }
 247
 248 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
 249 {
 250     uint8_t *data_ptr = dataptr;
 251     uint32_t x = 0;
 252     while (x < size) {
 253         if (idx >= TRACE_BUF_LEN) {
 254             idx = idx % TRACE_BUF_LEN;
 255         }
 256         data_ptr[x++] = trace_buf[idx++];
 257     }
 258 }
 259
 260 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
 261 {
 262     uint8_t *data_ptr = dataptr;
 263     uint32_t x = 0;
 264     while (x < size) {
 265         if (idx >= TRACE_BUF_LEN) {
 266             idx = idx % TRACE_BUF_LEN;
 267         }
 268         trace_buf[idx++] = data_ptr[x++];
 269     }
 270     return idx; /* most callers wants to know where to write next */
 271 }
 272
 273 void trace_record_finish(TraceBufferRecord *rec)
 274 {
 275     TraceRecord record;
 276     read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
 277     smp_wmb(); /* write barrier before marking as valid */
 278     record.event |= TRACE_RECORD_VALID;
 279     write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
 280
 281     if (((unsigned int)g_atomic_int_get(&trace_idx) - writeout_idx)
 282         > TRACE_BUF_FLUSH_THRESHOLD) {
 283         flush_trace_file(false);
 284     }
 285 }
 286
 287 void st_set_trace_file_enabled(bool enable)
 288 {
 289     if (enable == !!trace_fp) {
 290         return; /* no change */
 291     }
 292
 293     /* Halt trace writeout */
 294     flush_trace_file(true);
 295     trace_writeout_enabled = false;
 296     flush_trace_file(true);
 297
 298     if (enable) {
 299         static const TraceLogHeader header = {
 300             .header_event_id = HEADER_EVENT_ID,
 301             .header_magic = HEADER_MAGIC,
 302             /* Older log readers will check for version at next location */
 303             .header_version = HEADER_VERSION,
 304         };
 305
 306         trace_fp = fopen(trace_file_name, "wb");
 307         if (!trace_fp) {
 308             return;
 309         }
 310
 311         if (fwrite(&header, sizeof header, 1, trace_fp) != 1) {
 312             fclose(trace_fp);
 313             trace_fp = NULL;
 314             return;
 315         }
 316
 317         /* Resume trace writeout */
 318         trace_writeout_enabled = true;
 319         flush_trace_file(false);
 320     } else {
 321         fclose(trace_fp);
 322         trace_fp = NULL;
 323     }
 324 }
 325
 326 /**
 327  * Set the name of a trace file
 328  *
 329  * @file        The trace file name or NULL for the default name-<pid> set at
 330  *              config time
 331  */
 332 bool st_set_trace_file(const char *file)
 333 {
 334     st_set_trace_file_enabled(false);
 335
 336     g_free(trace_file_name);
 337
 338     if (!file) {
 339         trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, getpid());
 340     } else {
 341         trace_file_name = g_strdup_printf("%s", file);
 342     }
 343
 344     st_set_trace_file_enabled(true);
 345     return true;
 346 }
 347
 348 void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
 349 {
 350     stream_printf(stream, "Trace file \"%s\" %s.\n",
 351                   trace_file_name, trace_fp ? "on" : "off");
 352 }
 353
 354 void st_flush_trace_buffer(void)
 355 {
 356     flush_trace_file(true);
 357 }
 358
 359 void trace_print_events(FILE *stream, fprintf_function stream_printf)
 360 {
 361     unsigned int i;
 362
 363     for (i = 0; i < trace_event_count(); i++) {
 364         TraceEvent *ev = trace_event_id(i);
 365         stream_printf(stream, "%s [Event ID %u] : state %u\n",
 366                       trace_event_get_name(ev), i, trace_event_get_state_dynamic(ev));
 367     }
 368 }
 369
 370 void trace_event_set_state_dynamic_backend(TraceEvent *ev, bool state)
 371 {
 372     ev->dstate = state;
 373 }
 374
 375 /* Helper function to create a thread with signals blocked.  Use glib's
 376  * portable threads since QEMU abstractions cannot be used due to reentrancy in
 377  * the tracer.  Also note the signal masking on POSIX hosts so that the thread
 378  * does not steal signals when the rest of the program wants them blocked.
 379  */
 380 static GThread *trace_thread_create(GThreadFunc fn)
 381 {
 382     GThread *thread;
 383 #ifndef _WIN32
 384     sigset_t set, oldset;
 385
 386     sigfillset(&set);
 387     pthread_sigmask(SIG_SETMASK, &set, &oldset);
 388 #endif
 389
 390 #if GLIB_CHECK_VERSION(2, 31, 0)
 391     thread = g_thread_new("trace-thread", fn, NULL);
 392 #else
 393     thread = g_thread_create(fn, NULL, FALSE, NULL);
 394 #endif
 395
 396 #ifndef _WIN32
 397     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 398 #endif
 399
 400     return thread;
 401 }
 402
 403 bool trace_backend_init(const char *events, const char *file)
 404 {
 405     GThread *thread;
 406
 407     if (!g_thread_supported()) {
 408 #if !GLIB_CHECK_VERSION(2, 31, 0)
 409         g_thread_init(NULL);
 410 #else
 411         fprintf(stderr, "glib threading failed to initialize.\n");
 412         exit(1);
 413 #endif
 414     }
 415
 416 #if !GLIB_CHECK_VERSION(2, 31, 0)
 417     trace_available_cond = g_cond_new();
 418     trace_empty_cond = g_cond_new();
 419 #endif
 420
 421     thread = trace_thread_create(writeout_thread);
 422     if (!thread) {
 423         fprintf(stderr, "warning: unable to initialize simple trace backend\n");
 424         return false;
 425     }
 426
 427     atexit(st_flush_trace_buffer);
 428     trace_backend_init_events(events);
 429     st_set_trace_file(file);
 430     return true;
 431 }