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