xref: /openbmc/qemu/trace/simple.c (revision 1fd6bb44)
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 }
432