xref: /openbmc/qemu/trace/simple.c (revision b618c288)
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 #include "trace/simple.h"
23 
24 /** Trace file header event ID */
25 #define HEADER_EVENT_ID (~(uint64_t)0) /* avoids conflicting with TraceEventIDs */
26 
27 /** Trace file magic number */
28 #define HEADER_MAGIC 0xf2b177cb0aa429b4ULL
29 
30 /** Trace file version number, bump if format changes */
31 #define HEADER_VERSION 2
32 
33 /** Records were dropped event ID */
34 #define DROPPED_EVENT_ID (~(uint64_t)0 - 1)
35 
36 /** Trace record is valid */
37 #define TRACE_RECORD_VALID ((uint64_t)1 << 63)
38 
39 /*
40  * Trace records are written out by a dedicated thread.  The thread waits for
41  * records to become available, writes them out, and then waits again.
42  */
43 static GStaticMutex trace_lock = G_STATIC_MUTEX_INIT;
44 
45 /* g_cond_new() was deprecated in glib 2.31 but we still need to support it */
46 #if GLIB_CHECK_VERSION(2, 31, 0)
47 static GCond the_trace_available_cond;
48 static GCond the_trace_empty_cond;
49 static GCond *trace_available_cond = &the_trace_available_cond;
50 static GCond *trace_empty_cond = &the_trace_empty_cond;
51 #else
52 static GCond *trace_available_cond;
53 static GCond *trace_empty_cond;
54 #endif
55 
56 static bool trace_available;
57 static bool trace_writeout_enabled;
58 
59 enum {
60     TRACE_BUF_LEN = 4096 * 64,
61     TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
62 };
63 
64 uint8_t trace_buf[TRACE_BUF_LEN];
65 static volatile gint trace_idx;
66 static unsigned int writeout_idx;
67 static volatile gint dropped_events;
68 static FILE *trace_fp;
69 static char *trace_file_name;
70 
71 /* * Trace buffer entry */
72 typedef struct {
73     uint64_t event; /*   TraceEventID */
74     uint64_t timestamp_ns;
75     uint32_t length;   /*    in bytes */
76     uint32_t reserved; /*    unused */
77     uint64_t arguments[];
78 } TraceRecord;
79 
80 typedef struct {
81     uint64_t header_event_id; /* HEADER_EVENT_ID */
82     uint64_t header_magic;    /* HEADER_MAGIC    */
83     uint64_t header_version;  /* HEADER_VERSION  */
84 } TraceLogHeader;
85 
86 
87 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
88 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
89 
90 static void clear_buffer_range(unsigned int idx, size_t len)
91 {
92     uint32_t num = 0;
93     while (num < len) {
94         if (idx >= TRACE_BUF_LEN) {
95             idx = idx % TRACE_BUF_LEN;
96         }
97         trace_buf[idx++] = 0;
98         num++;
99     }
100 }
101 /**
102  * Read a trace record from the trace buffer
103  *
104  * @idx         Trace buffer index
105  * @record      Trace record to fill
106  *
107  * Returns false if the record is not valid.
108  */
109 static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
110 {
111     uint64_t event_flag = 0;
112     TraceRecord record;
113     /* read the event flag to see if its a valid record */
114     read_from_buffer(idx, &record, sizeof(event_flag));
115 
116     if (!(record.event & TRACE_RECORD_VALID)) {
117         return false;
118     }
119 
120     smp_rmb(); /* read memory barrier before accessing record */
121     /* read the record header to know record length */
122     read_from_buffer(idx, &record, sizeof(TraceRecord));
123     *recordptr = malloc(record.length); /* dont use g_malloc, can deadlock when traced */
124     /* make a copy of record to avoid being overwritten */
125     read_from_buffer(idx, *recordptr, record.length);
126     smp_rmb(); /* memory barrier before clearing valid flag */
127     (*recordptr)->event &= ~TRACE_RECORD_VALID;
128     /* clear the trace buffer range for consumed record otherwise any byte
129      * with its MSB set may be considered as a valid event id when the writer
130      * thread crosses this range of buffer again.
131      */
132     clear_buffer_range(idx, record.length);
133     return true;
134 }
135 
136 /**
137  * Kick writeout thread
138  *
139  * @wait        Whether to wait for writeout thread to complete
140  */
141 static void flush_trace_file(bool wait)
142 {
143     g_static_mutex_lock(&trace_lock);
144     trace_available = true;
145     g_cond_signal(trace_available_cond);
146 
147     if (wait) {
148         g_cond_wait(trace_empty_cond, g_static_mutex_get_mutex(&trace_lock));
149     }
150 
151     g_static_mutex_unlock(&trace_lock);
152 }
153 
154 static void wait_for_trace_records_available(void)
155 {
156     g_static_mutex_lock(&trace_lock);
157     while (!(trace_available && trace_writeout_enabled)) {
158         g_cond_signal(trace_empty_cond);
159         g_cond_wait(trace_available_cond,
160                     g_static_mutex_get_mutex(&trace_lock));
161     }
162     trace_available = false;
163     g_static_mutex_unlock(&trace_lock);
164 }
165 
166 static gpointer writeout_thread(gpointer opaque)
167 {
168     TraceRecord *recordptr;
169     union {
170         TraceRecord rec;
171         uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
172     } dropped;
173     unsigned int idx = 0;
174     int dropped_count;
175     size_t unused __attribute__ ((unused));
176 
177     for (;;) {
178         wait_for_trace_records_available();
179 
180         if (g_atomic_int_get(&dropped_events)) {
181             dropped.rec.event = DROPPED_EVENT_ID,
182             dropped.rec.timestamp_ns = get_clock();
183             dropped.rec.length = sizeof(TraceRecord) + sizeof(uint64_t),
184             dropped.rec.reserved = 0;
185             do {
186                 dropped_count = g_atomic_int_get(&dropped_events);
187             } while (!g_atomic_int_compare_and_exchange(&dropped_events,
188                                                         dropped_count, 0));
189             dropped.rec.arguments[0] = dropped_count;
190             unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
191         }
192 
193         while (get_trace_record(idx, &recordptr)) {
194             unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
195             writeout_idx += recordptr->length;
196             free(recordptr); /* dont use g_free, can deadlock when traced */
197             idx = writeout_idx % TRACE_BUF_LEN;
198         }
199 
200         fflush(trace_fp);
201     }
202     return NULL;
203 }
204 
205 void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
206 {
207     rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
208 }
209 
210 void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
211 {
212     /* Write string length first */
213     rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
214     /* Write actual string now */
215     rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
216 }
217 
218 int trace_record_start(TraceBufferRecord *rec, TraceEventID event, size_t datasize)
219 {
220     unsigned int idx, rec_off, old_idx, new_idx;
221     uint32_t rec_len = sizeof(TraceRecord) + datasize;
222     uint64_t event_u64 = event;
223     uint64_t timestamp_ns = get_clock();
224 
225     do {
226         old_idx = g_atomic_int_get(&trace_idx);
227         smp_rmb();
228         new_idx = old_idx + rec_len;
229 
230         if (new_idx - writeout_idx > TRACE_BUF_LEN) {
231             /* Trace Buffer Full, Event dropped ! */
232             g_atomic_int_inc(&dropped_events);
233             return -ENOSPC;
234         }
235     } while (!g_atomic_int_compare_and_exchange(&trace_idx, old_idx, new_idx));
236 
237     idx = old_idx % TRACE_BUF_LEN;
238 
239     rec_off = idx;
240     rec_off = write_to_buffer(rec_off, &event_u64, sizeof(event_u64));
241     rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
242     rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
243 
244     rec->tbuf_idx = idx;
245     rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
246     return 0;
247 }
248 
249 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
250 {
251     uint8_t *data_ptr = dataptr;
252     uint32_t x = 0;
253     while (x < size) {
254         if (idx >= TRACE_BUF_LEN) {
255             idx = idx % TRACE_BUF_LEN;
256         }
257         data_ptr[x++] = trace_buf[idx++];
258     }
259 }
260 
261 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
262 {
263     uint8_t *data_ptr = dataptr;
264     uint32_t x = 0;
265     while (x < size) {
266         if (idx >= TRACE_BUF_LEN) {
267             idx = idx % TRACE_BUF_LEN;
268         }
269         trace_buf[idx++] = data_ptr[x++];
270     }
271     return idx; /* most callers wants to know where to write next */
272 }
273 
274 void trace_record_finish(TraceBufferRecord *rec)
275 {
276     TraceRecord record;
277     read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
278     smp_wmb(); /* write barrier before marking as valid */
279     record.event |= TRACE_RECORD_VALID;
280     write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
281 
282     if (((unsigned int)g_atomic_int_get(&trace_idx) - writeout_idx)
283         > TRACE_BUF_FLUSH_THRESHOLD) {
284         flush_trace_file(false);
285     }
286 }
287 
288 void st_set_trace_file_enabled(bool enable)
289 {
290     if (enable == !!trace_fp) {
291         return; /* no change */
292     }
293 
294     /* Halt trace writeout */
295     flush_trace_file(true);
296     trace_writeout_enabled = false;
297     flush_trace_file(true);
298 
299     if (enable) {
300         static const TraceLogHeader header = {
301             .header_event_id = HEADER_EVENT_ID,
302             .header_magic = HEADER_MAGIC,
303             /* Older log readers will check for version at next location */
304             .header_version = HEADER_VERSION,
305         };
306 
307         trace_fp = fopen(trace_file_name, "wb");
308         if (!trace_fp) {
309             return;
310         }
311 
312         if (fwrite(&header, sizeof header, 1, trace_fp) != 1) {
313             fclose(trace_fp);
314             trace_fp = NULL;
315             return;
316         }
317 
318         /* Resume trace writeout */
319         trace_writeout_enabled = true;
320         flush_trace_file(false);
321     } else {
322         fclose(trace_fp);
323         trace_fp = NULL;
324     }
325 }
326 
327 /**
328  * Set the name of a trace file
329  *
330  * @file        The trace file name or NULL for the default name-<pid> set at
331  *              config time
332  */
333 bool st_set_trace_file(const char *file)
334 {
335     st_set_trace_file_enabled(false);
336 
337     g_free(trace_file_name);
338 
339     if (!file) {
340         trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, getpid());
341     } else {
342         trace_file_name = g_strdup_printf("%s", file);
343     }
344 
345     st_set_trace_file_enabled(true);
346     return true;
347 }
348 
349 void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
350 {
351     stream_printf(stream, "Trace file \"%s\" %s.\n",
352                   trace_file_name, trace_fp ? "on" : "off");
353 }
354 
355 void st_flush_trace_buffer(void)
356 {
357     flush_trace_file(true);
358 }
359 
360 void trace_print_events(FILE *stream, fprintf_function stream_printf)
361 {
362     unsigned int i;
363 
364     for (i = 0; i < trace_event_count(); i++) {
365         TraceEvent *ev = trace_event_id(i);
366         stream_printf(stream, "%s [Event ID %u] : state %u\n",
367                       trace_event_get_name(ev), i, trace_event_get_state_dynamic(ev));
368     }
369 }
370 
371 void trace_event_set_state_dynamic_backend(TraceEvent *ev, bool state)
372 {
373     ev->dstate = state;
374 }
375 
376 /* Helper function to create a thread with signals blocked.  Use glib's
377  * portable threads since QEMU abstractions cannot be used due to reentrancy in
378  * the tracer.  Also note the signal masking on POSIX hosts so that the thread
379  * does not steal signals when the rest of the program wants them blocked.
380  */
381 static GThread *trace_thread_create(GThreadFunc fn)
382 {
383     GThread *thread;
384 #ifndef _WIN32
385     sigset_t set, oldset;
386 
387     sigfillset(&set);
388     pthread_sigmask(SIG_SETMASK, &set, &oldset);
389 #endif
390 
391 #if GLIB_CHECK_VERSION(2, 31, 0)
392     thread = g_thread_new("trace-thread", fn, NULL);
393 #else
394     thread = g_thread_create(fn, NULL, FALSE, NULL);
395 #endif
396 
397 #ifndef _WIN32
398     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
399 #endif
400 
401     return thread;
402 }
403 
404 bool trace_backend_init(const char *events, const char *file)
405 {
406     GThread *thread;
407 
408     if (!g_thread_supported()) {
409 #if !GLIB_CHECK_VERSION(2, 31, 0)
410         g_thread_init(NULL);
411 #else
412         fprintf(stderr, "glib threading failed to initialize.\n");
413         exit(1);
414 #endif
415     }
416 
417 #if !GLIB_CHECK_VERSION(2, 31, 0)
418     trace_available_cond = g_cond_new();
419     trace_empty_cond = g_cond_new();
420 #endif
421 
422     thread = trace_thread_create(writeout_thread);
423     if (!thread) {
424         fprintf(stderr, "warning: unable to initialize simple trace backend\n");
425         return false;
426     }
427 
428     atexit(st_flush_trace_buffer);
429     trace_backend_init_events(events);
430     st_set_trace_file(file);
431     return true;
432 }
433