simple.c 12.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Simple trace backend
 *
 * Copyright IBM, Corp. 2010
 *
 * This work is licensed under the terms of the GNU GPL, version 2.  See
 * the COPYING file in the top-level directory.
 *
 */

#include <stdlib.h>
#include <stdint.h>
#include <stdio.h>
#include <time.h>
15
#ifndef _WIN32
16 17
#include <signal.h>
#include <pthread.h>
18
#endif
19
#include "qemu-timer.h"
20
#include "trace.h"
21
#include "trace/control.h"
22 23 24 25 26 27 28 29

/** Trace file header event ID */
#define HEADER_EVENT_ID (~(uint64_t)0) /* avoids conflicting with TraceEventIDs */

/** Trace file magic number */
#define HEADER_MAGIC 0xf2b177cb0aa429b4ULL

/** Trace file version number, bump if format changes */
30
#define HEADER_VERSION 2
31

32 33 34 35 36 37 38 39 40 41
/** Records were dropped event ID */
#define DROPPED_EVENT_ID (~(uint64_t)0 - 1)

/** Trace record is valid */
#define TRACE_RECORD_VALID ((uint64_t)1 << 63)

/*
 * Trace records are written out by a dedicated thread.  The thread waits for
 * records to become available, writes them out, and then waits again.
 */
42 43 44
static GStaticMutex trace_lock = G_STATIC_MUTEX_INIT;
static GCond *trace_available_cond;
static GCond *trace_empty_cond;
45 46 47
static bool trace_available;
static bool trace_writeout_enabled;

48 49 50 51 52 53
enum {
    TRACE_BUF_LEN = 4096 * 64,
    TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
};

uint8_t trace_buf[TRACE_BUF_LEN];
54
static unsigned int trace_idx;
55 56
static unsigned int writeout_idx;
static uint64_t dropped_events;
57
static FILE *trace_fp;
58
static char *trace_file_name;
59

60 61 62 63 64 65 66 67 68 69 70 71 72
/* * Trace buffer entry */
typedef struct {
    uint64_t event; /*   TraceEventID */
    uint64_t timestamp_ns;
    uint32_t length;   /*    in bytes */
    uint32_t reserved; /*    unused */
    uint8_t arguments[];
} TraceRecord;

typedef struct {
    uint64_t header_event_id; /* HEADER_EVENT_ID */
    uint64_t header_magic;    /* HEADER_MAGIC    */
    uint64_t header_version;  /* HEADER_VERSION  */
73
} TraceLogHeader;
74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89


static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);

static void clear_buffer_range(unsigned int idx, size_t len)
{
    uint32_t num = 0;
    while (num < len) {
        if (idx >= TRACE_BUF_LEN) {
            idx = idx % TRACE_BUF_LEN;
        }
        trace_buf[idx++] = 0;
        num++;
    }
}
90
/**
91 92 93 94 95 96
 * Read a trace record from the trace buffer
 *
 * @idx         Trace buffer index
 * @record      Trace record to fill
 *
 * Returns false if the record is not valid.
97
 */
98
static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
P
Prerna Saxena 已提交
99
{
100 101 102 103 104 105
    uint64_t event_flag = 0;
    TraceRecord record;
    /* read the event flag to see if its a valid record */
    read_from_buffer(idx, &record, sizeof(event_flag));

    if (!(record.event & TRACE_RECORD_VALID)) {
106
        return false;
P
Prerna Saxena 已提交
107 108
    }

109 110 111 112 113 114 115 116 117 118 119 120 121
    smp_rmb(); /* read memory barrier before accessing record */
    /* read the record header to know record length */
    read_from_buffer(idx, &record, sizeof(TraceRecord));
    *recordptr = malloc(record.length); /* dont use g_malloc, can deadlock when traced */
    /* make a copy of record to avoid being overwritten */
    read_from_buffer(idx, *recordptr, record.length);
    smp_rmb(); /* memory barrier before clearing valid flag */
    (*recordptr)->event &= ~TRACE_RECORD_VALID;
    /* clear the trace buffer range for consumed record otherwise any byte
     * with its MSB set may be considered as a valid event id when the writer
     * thread crosses this range of buffer again.
     */
    clear_buffer_range(idx, record.length);
122
    return true;
P
Prerna Saxena 已提交
123 124
}

125 126 127 128 129 130
/**
 * Kick writeout thread
 *
 * @wait        Whether to wait for writeout thread to complete
 */
static void flush_trace_file(bool wait)
131
{
132
    g_static_mutex_lock(&trace_lock);
133
    trace_available = true;
134
    g_cond_signal(trace_available_cond);
135

136
    if (wait) {
137
        g_cond_wait(trace_empty_cond, g_static_mutex_get_mutex(&trace_lock));
138
    }
139

140
    g_static_mutex_unlock(&trace_lock);
141 142
}

143
static void wait_for_trace_records_available(void)
144
{
145
    g_static_mutex_lock(&trace_lock);
146
    while (!(trace_available && trace_writeout_enabled)) {
147 148 149
        g_cond_signal(trace_empty_cond);
        g_cond_wait(trace_available_cond,
                    g_static_mutex_get_mutex(&trace_lock));
150
    }
151
    trace_available = false;
152
    g_static_mutex_unlock(&trace_lock);
153 154
}

155
static gpointer writeout_thread(gpointer opaque)
156
{
157 158 159 160 161 162 163
    TraceRecord *recordptr;
    union {
        TraceRecord rec;
        uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
    } dropped;
    unsigned int idx = 0;
    uint64_t dropped_count;
164
    size_t unused __attribute__ ((unused));
165 166 167 168

    for (;;) {
        wait_for_trace_records_available();

169 170 171 172 173 174 175 176 177 178 179 180 181 182
        if (dropped_events) {
            dropped.rec.event = DROPPED_EVENT_ID,
            dropped.rec.timestamp_ns = get_clock();
            dropped.rec.length = sizeof(TraceRecord) + sizeof(dropped_events),
            dropped.rec.reserved = 0;
            while (1) {
                dropped_count = dropped_events;
                if (g_atomic_int_compare_and_exchange((gint *)&dropped_events,
                                                      dropped_count, 0)) {
                    break;
                }
            }
            memcpy(dropped.rec.arguments, &dropped_count, sizeof(uint64_t));
            unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
183
        }
184

185 186 187 188 189
        while (get_trace_record(idx, &recordptr)) {
            unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
            writeout_idx += recordptr->length;
            free(recordptr); /* dont use g_free, can deadlock when traced */
            idx = writeout_idx % TRACE_BUF_LEN;
190
        }
191

192
        fflush(trace_fp);
193
    }
194
    return NULL;
195 196
}

197
void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
198
{
199
    rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
200 201
}

202
void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
203
{
204 205 206 207
    /* Write string length first */
    rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
    /* Write actual string now */
    rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
208 209
}

210
int trace_record_start(TraceBufferRecord *rec, TraceEventID event, size_t datasize)
211
{
212 213 214 215 216 217 218 219 220 221 222 223 224 225
    unsigned int idx, rec_off, old_idx, new_idx;
    uint32_t rec_len = sizeof(TraceRecord) + datasize;
    uint64_t timestamp_ns = get_clock();

    while (1) {
        old_idx = trace_idx;
        smp_rmb();
        new_idx = old_idx + rec_len;

        if (new_idx - writeout_idx > TRACE_BUF_LEN) {
            /* Trace Buffer Full, Event dropped ! */
            g_atomic_int_inc((gint *)&dropped_events);
            return -ENOSPC;
        }
226

227 228 229 230 231
        if (g_atomic_int_compare_and_exchange((gint *)&trace_idx,
                                              old_idx, new_idx)) {
            break;
        }
    }
232

233 234 235
    idx = old_idx % TRACE_BUF_LEN;

    rec_off = idx;
236 237 238
    rec_off = write_to_buffer(rec_off, &event, sizeof(event));
    rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
    rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
239 240 241 242

    rec->tbuf_idx = idx;
    rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
    return 0;
243 244
}

245
static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
246
{
247 248 249 250 251 252 253 254
    uint8_t *data_ptr = dataptr;
    uint32_t x = 0;
    while (x < size) {
        if (idx >= TRACE_BUF_LEN) {
            idx = idx % TRACE_BUF_LEN;
        }
        data_ptr[x++] = trace_buf[idx++];
    }
255 256
}

257
static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
258
{
259 260 261 262 263 264 265 266 267
    uint8_t *data_ptr = dataptr;
    uint32_t x = 0;
    while (x < size) {
        if (idx >= TRACE_BUF_LEN) {
            idx = idx % TRACE_BUF_LEN;
        }
        trace_buf[idx++] = data_ptr[x++];
    }
    return idx; /* most callers wants to know where to write next */
268 269
}

270
void trace_record_finish(TraceBufferRecord *rec)
271
{
272 273
    TraceRecord record;
    read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
274
    smp_wmb(); /* write barrier before marking as valid */
275 276
    record.event |= TRACE_RECORD_VALID;
    write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
277 278 279 280

    if ((trace_idx - writeout_idx) > TRACE_BUF_FLUSH_THRESHOLD) {
        flush_trace_file(false);
    }
281 282
}

283 284 285 286 287 288 289 290 291 292 293 294
void st_set_trace_file_enabled(bool enable)
{
    if (enable == !!trace_fp) {
        return; /* no change */
    }

    /* Halt trace writeout */
    flush_trace_file(true);
    trace_writeout_enabled = false;
    flush_trace_file(true);

    if (enable) {
295
        static const TraceLogHeader header = {
296 297 298 299
            .header_event_id = HEADER_EVENT_ID,
            .header_magic = HEADER_MAGIC,
            /* Older log readers will check for version at next location */
            .header_version = HEADER_VERSION,
300 301
        };

302
        trace_fp = fopen(trace_file_name, "wb");
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
        if (!trace_fp) {
            return;
        }

        if (fwrite(&header, sizeof header, 1, trace_fp) != 1) {
            fclose(trace_fp);
            trace_fp = NULL;
            return;
        }

        /* Resume trace writeout */
        trace_writeout_enabled = true;
        flush_trace_file(false);
    } else {
        fclose(trace_fp);
        trace_fp = NULL;
    }
}

322
/**
323 324 325 326
 * Set the name of a trace file
 *
 * @file        The trace file name or NULL for the default name-<pid> set at
 *              config time
327
 */
328
bool st_set_trace_file(const char *file)
329
{
330 331
    st_set_trace_file_enabled(false);

332
    g_free(trace_file_name);
333 334

    if (!file) {
335
        trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, getpid());
336
    } else {
337
        trace_file_name = g_strdup_printf("%s", file);
338 339 340 341 342 343 344 345 346 347
    }

    st_set_trace_file_enabled(true);
    return true;
}

void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
{
    stream_printf(stream, "Trace file \"%s\" %s.\n",
                  trace_file_name, trace_fp ? "on" : "off");
348
}
349

350 351 352 353 354 355
void st_flush_trace_buffer(void)
{
    flush_trace_file(true);
}

void trace_print_events(FILE *stream, fprintf_function stream_printf)
356 357 358 359 360 361 362 363 364
{
    unsigned int i;

    for (i = 0; i < NR_TRACE_EVENTS; i++) {
        stream_printf(stream, "%s [Event ID %u] : state %u\n",
                      trace_list[i].tp_name, i, trace_list[i].state);
    }
}

365
bool trace_event_set_state(const char *name, bool state)
366 367
{
    unsigned int i;
M
Mark Wu 已提交
368 369 370 371 372 373 374 375 376
    unsigned int len;
    bool wildcard = false;
    bool matched = false;

    len = strlen(name);
    if (len > 0 && name[len - 1] == '*') {
        wildcard = true;
        len -= 1;
    }
377
    for (i = 0; i < NR_TRACE_EVENTS; i++) {
M
Mark Wu 已提交
378 379 380 381 382 383 384
        if (wildcard) {
            if (!strncmp(trace_list[i].tp_name, name, len)) {
                trace_list[i].state = state;
                matched = true;
            }
            continue;
        }
385
        if (!strcmp(trace_list[i].tp_name, name)) {
386
            trace_list[i].state = state;
387
            return true;
388 389
        }
    }
M
Mark Wu 已提交
390
    return matched;
391 392
}

393 394 395 396 397 398
/* Helper function to create a thread with signals blocked.  Use glib's
 * portable threads since QEMU abstractions cannot be used due to reentrancy in
 * the tracer.  Also note the signal masking on POSIX hosts so that the thread
 * does not steal signals when the rest of the program wants them blocked.
 */
static GThread *trace_thread_create(GThreadFunc fn)
399
{
400 401
    GThread *thread;
#ifndef _WIN32
402
    sigset_t set, oldset;
403

404 405
    sigfillset(&set);
    pthread_sigmask(SIG_SETMASK, &set, &oldset);
406
#endif
407
    thread = g_thread_create(fn, NULL, FALSE, NULL);
408
#ifndef _WIN32
409
    pthread_sigmask(SIG_SETMASK, &oldset, NULL);
410
#endif
411

412 413 414 415 416 417 418 419
    return thread;
}

bool trace_backend_init(const char *events, const char *file)
{
    GThread *thread;

    if (!g_thread_supported()) {
420
#if !GLIB_CHECK_VERSION(2, 31, 0)
421
        g_thread_init(NULL);
422 423 424 425
#else
        fprintf(stderr, "glib threading failed to initialize.\n");
        exit(1);
#endif
426 427 428 429 430 431 432
    }

    trace_available_cond = g_cond_new();
    trace_empty_cond = g_cond_new();

    thread = trace_thread_create(writeout_thread);
    if (!thread) {
433
        fprintf(stderr, "warning: unable to initialize simple trace backend\n");
434
        return false;
435
    }
436

437 438 439
    atexit(st_flush_trace_buffer);
    trace_backend_init_events(events);
    st_set_trace_file(file);
440
    return true;
441
}