xref: /qemu/trace/simple.c (revision 7a4e543d)
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 */
21 #define HEADER_EVENT_ID (~(uint64_t)0) /* avoids conflicting with TraceEventIDs */
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 3
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 /* * Trace buffer entry */
60 typedef struct {
61     uint64_t event; /*   TraceEventID */
62     uint64_t timestamp_ns;
63     uint32_t length;   /*    in bytes */
64     uint32_t pid;
65     uint64_t arguments[];
66 } TraceRecord;
67 
68 typedef struct {
69     uint64_t header_event_id; /* HEADER_EVENT_ID */
70     uint64_t header_magic;    /* HEADER_MAGIC    */
71     uint64_t header_version;  /* HEADER_VERSION  */
72 } TraceLogHeader;
73 
74 
75 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
76 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
77 
78 static void clear_buffer_range(unsigned int idx, size_t len)
79 {
80     uint32_t num = 0;
81     while (num < len) {
82         if (idx >= TRACE_BUF_LEN) {
83             idx = idx % TRACE_BUF_LEN;
84         }
85         trace_buf[idx++] = 0;
86         num++;
87     }
88 }
89 /**
90  * Read a trace record from the trace buffer
91  *
92  * @idx         Trace buffer index
93  * @record      Trace record to fill
94  *
95  * Returns false if the record is not valid.
96  */
97 static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
98 {
99     uint64_t event_flag = 0;
100     TraceRecord record;
101     /* read the event flag to see if its a valid record */
102     read_from_buffer(idx, &record, sizeof(event_flag));
103 
104     if (!(record.event & TRACE_RECORD_VALID)) {
105         return false;
106     }
107 
108     smp_rmb(); /* read memory barrier before accessing record */
109     /* read the record header to know record length */
110     read_from_buffer(idx, &record, sizeof(TraceRecord));
111     *recordptr = malloc(record.length); /* dont use g_malloc, can deadlock when traced */
112     /* make a copy of record to avoid being overwritten */
113     read_from_buffer(idx, *recordptr, record.length);
114     smp_rmb(); /* memory barrier before clearing valid flag */
115     (*recordptr)->event &= ~TRACE_RECORD_VALID;
116     /* clear the trace buffer range for consumed record otherwise any byte
117      * with its MSB set may be considered as a valid event id when the writer
118      * thread crosses this range of buffer again.
119      */
120     clear_buffer_range(idx, record.length);
121     return true;
122 }
123 
124 /**
125  * Kick writeout thread
126  *
127  * @wait        Whether to wait for writeout thread to complete
128  */
129 static void flush_trace_file(bool wait)
130 {
131     g_mutex_lock(&trace_lock);
132     trace_available = true;
133     g_cond_signal(&trace_available_cond);
134 
135     if (wait) {
136         g_cond_wait(&trace_empty_cond, &trace_lock);
137     }
138 
139     g_mutex_unlock(&trace_lock);
140 }
141 
142 static void wait_for_trace_records_available(void)
143 {
144     g_mutex_lock(&trace_lock);
145     while (!(trace_available && trace_writeout_enabled)) {
146         g_cond_signal(&trace_empty_cond);
147         g_cond_wait(&trace_available_cond, &trace_lock);
148     }
149     trace_available = false;
150     g_mutex_unlock(&trace_lock);
151 }
152 
153 static gpointer writeout_thread(gpointer opaque)
154 {
155     TraceRecord *recordptr;
156     union {
157         TraceRecord rec;
158         uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
159     } dropped;
160     unsigned int idx = 0;
161     int dropped_count;
162     size_t unused __attribute__ ((unused));
163 
164     for (;;) {
165         wait_for_trace_records_available();
166 
167         if (g_atomic_int_get(&dropped_events)) {
168             dropped.rec.event = DROPPED_EVENT_ID,
169             dropped.rec.timestamp_ns = get_clock();
170             dropped.rec.length = sizeof(TraceRecord) + sizeof(uint64_t),
171             dropped.rec.pid = trace_pid;
172             do {
173                 dropped_count = g_atomic_int_get(&dropped_events);
174             } while (!g_atomic_int_compare_and_exchange(&dropped_events,
175                                                         dropped_count, 0));
176             dropped.rec.arguments[0] = dropped_count;
177             unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
178         }
179 
180         while (get_trace_record(idx, &recordptr)) {
181             unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
182             writeout_idx += recordptr->length;
183             free(recordptr); /* dont use g_free, can deadlock when traced */
184             idx = writeout_idx % TRACE_BUF_LEN;
185         }
186 
187         fflush(trace_fp);
188     }
189     return NULL;
190 }
191 
192 void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
193 {
194     rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
195 }
196 
197 void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
198 {
199     /* Write string length first */
200     rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
201     /* Write actual string now */
202     rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
203 }
204 
205 int trace_record_start(TraceBufferRecord *rec, TraceEventID event, size_t datasize)
206 {
207     unsigned int idx, rec_off, old_idx, new_idx;
208     uint32_t rec_len = sizeof(TraceRecord) + datasize;
209     uint64_t event_u64 = event;
210     uint64_t timestamp_ns = get_clock();
211 
212     do {
213         old_idx = g_atomic_int_get(&trace_idx);
214         smp_rmb();
215         new_idx = old_idx + rec_len;
216 
217         if (new_idx - writeout_idx > TRACE_BUF_LEN) {
218             /* Trace Buffer Full, Event dropped ! */
219             g_atomic_int_inc(&dropped_events);
220             return -ENOSPC;
221         }
222     } while (!g_atomic_int_compare_and_exchange(&trace_idx, old_idx, new_idx));
223 
224     idx = old_idx % TRACE_BUF_LEN;
225 
226     rec_off = idx;
227     rec_off = write_to_buffer(rec_off, &event_u64, sizeof(event_u64));
228     rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
229     rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
230     rec_off = write_to_buffer(rec_off, &trace_pid, sizeof(trace_pid));
231 
232     rec->tbuf_idx = idx;
233     rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
234     return 0;
235 }
236 
237 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
238 {
239     uint8_t *data_ptr = dataptr;
240     uint32_t x = 0;
241     while (x < size) {
242         if (idx >= TRACE_BUF_LEN) {
243             idx = idx % TRACE_BUF_LEN;
244         }
245         data_ptr[x++] = trace_buf[idx++];
246     }
247 }
248 
249 static unsigned int write_to_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         trace_buf[idx++] = data_ptr[x++];
258     }
259     return idx; /* most callers wants to know where to write next */
260 }
261 
262 void trace_record_finish(TraceBufferRecord *rec)
263 {
264     TraceRecord record;
265     read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
266     smp_wmb(); /* write barrier before marking as valid */
267     record.event |= TRACE_RECORD_VALID;
268     write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
269 
270     if (((unsigned int)g_atomic_int_get(&trace_idx) - writeout_idx)
271         > TRACE_BUF_FLUSH_THRESHOLD) {
272         flush_trace_file(false);
273     }
274 }
275 
276 void st_set_trace_file_enabled(bool enable)
277 {
278     if (enable == !!trace_fp) {
279         return; /* no change */
280     }
281 
282     /* Halt trace writeout */
283     flush_trace_file(true);
284     trace_writeout_enabled = false;
285     flush_trace_file(true);
286 
287     if (enable) {
288         static const TraceLogHeader header = {
289             .header_event_id = HEADER_EVENT_ID,
290             .header_magic = HEADER_MAGIC,
291             /* Older log readers will check for version at next location */
292             .header_version = HEADER_VERSION,
293         };
294 
295         trace_fp = fopen(trace_file_name, "wb");
296         if (!trace_fp) {
297             return;
298         }
299 
300         if (fwrite(&header, sizeof header, 1, trace_fp) != 1) {
301             fclose(trace_fp);
302             trace_fp = NULL;
303             return;
304         }
305 
306         /* Resume trace writeout */
307         trace_writeout_enabled = true;
308         flush_trace_file(false);
309     } else {
310         fclose(trace_fp);
311         trace_fp = NULL;
312     }
313 }
314 
315 /**
316  * Set the name of a trace file
317  *
318  * @file        The trace file name or NULL for the default name-<pid> set at
319  *              config time
320  */
321 void st_set_trace_file(const char *file)
322 {
323     st_set_trace_file_enabled(false);
324 
325     g_free(trace_file_name);
326 
327     if (!file) {
328         /* Type cast needed for Windows where getpid() returns an int. */
329         trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, (pid_t)getpid());
330     } else {
331         trace_file_name = g_strdup_printf("%s", file);
332     }
333 
334     st_set_trace_file_enabled(true);
335 }
336 
337 void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
338 {
339     stream_printf(stream, "Trace file \"%s\" %s.\n",
340                   trace_file_name, trace_fp ? "on" : "off");
341 }
342 
343 void st_flush_trace_buffer(void)
344 {
345     flush_trace_file(true);
346 }
347 
348 /* Helper function to create a thread with signals blocked.  Use glib's
349  * portable threads since QEMU abstractions cannot be used due to reentrancy in
350  * the tracer.  Also note the signal masking on POSIX hosts so that the thread
351  * does not steal signals when the rest of the program wants them blocked.
352  */
353 static GThread *trace_thread_create(GThreadFunc fn)
354 {
355     GThread *thread;
356 #ifndef _WIN32
357     sigset_t set, oldset;
358 
359     sigfillset(&set);
360     pthread_sigmask(SIG_SETMASK, &set, &oldset);
361 #endif
362 
363     thread = g_thread_new("trace-thread", fn, NULL);
364 
365 #ifndef _WIN32
366     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
367 #endif
368 
369     return thread;
370 }
371 
372 bool st_init(void)
373 {
374     GThread *thread;
375 
376     trace_pid = getpid();
377 
378     thread = trace_thread_create(writeout_thread);
379     if (!thread) {
380         fprintf(stderr, "warning: unable to initialize simple trace backend\n");
381         return false;
382     }
383 
384     atexit(st_flush_trace_buffer);
385     return true;
386 }
387