xref: /openbmc/qemu/trace/simple.c (revision 35d08458)
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 3
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 #if GLIB_CHECK_VERSION(2, 32, 0)
44 static GMutex trace_lock;
45 #define lock_trace_lock() g_mutex_lock(&trace_lock)
46 #define unlock_trace_lock() g_mutex_unlock(&trace_lock)
47 #define get_trace_lock_mutex() (&trace_lock)
48 #else
49 static GStaticMutex trace_lock = G_STATIC_MUTEX_INIT;
50 #define lock_trace_lock() g_static_mutex_lock(&trace_lock)
51 #define unlock_trace_lock() g_static_mutex_unlock(&trace_lock)
52 #define get_trace_lock_mutex() g_static_mutex_get_mutex(&trace_lock)
53 #endif
54 
55 /* g_cond_new() was deprecated in glib 2.31 but we still need to support it */
56 #if GLIB_CHECK_VERSION(2, 31, 0)
57 static GCond the_trace_available_cond;
58 static GCond the_trace_empty_cond;
59 static GCond *trace_available_cond = &the_trace_available_cond;
60 static GCond *trace_empty_cond = &the_trace_empty_cond;
61 #else
62 static GCond *trace_available_cond;
63 static GCond *trace_empty_cond;
64 #endif
65 
66 static bool trace_available;
67 static bool trace_writeout_enabled;
68 
69 enum {
70     TRACE_BUF_LEN = 4096 * 64,
71     TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
72 };
73 
74 uint8_t trace_buf[TRACE_BUF_LEN];
75 static volatile gint trace_idx;
76 static unsigned int writeout_idx;
77 static volatile gint dropped_events;
78 static FILE *trace_fp;
79 static char *trace_file_name;
80 
81 /* * Trace buffer entry */
82 typedef struct {
83     uint64_t event; /*   TraceEventID */
84     uint64_t timestamp_ns;
85     uint32_t length;   /*    in bytes */
86     uint32_t reserved; /*    unused */
87     uint64_t arguments[];
88 } TraceRecord;
89 
90 typedef struct {
91     uint64_t header_event_id; /* HEADER_EVENT_ID */
92     uint64_t header_magic;    /* HEADER_MAGIC    */
93     uint64_t header_version;  /* HEADER_VERSION  */
94 } TraceLogHeader;
95 
96 
97 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
98 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
99 
100 static void clear_buffer_range(unsigned int idx, size_t len)
101 {
102     uint32_t num = 0;
103     while (num < len) {
104         if (idx >= TRACE_BUF_LEN) {
105             idx = idx % TRACE_BUF_LEN;
106         }
107         trace_buf[idx++] = 0;
108         num++;
109     }
110 }
111 /**
112  * Read a trace record from the trace buffer
113  *
114  * @idx         Trace buffer index
115  * @record      Trace record to fill
116  *
117  * Returns false if the record is not valid.
118  */
119 static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
120 {
121     uint64_t event_flag = 0;
122     TraceRecord record;
123     /* read the event flag to see if its a valid record */
124     read_from_buffer(idx, &record, sizeof(event_flag));
125 
126     if (!(record.event & TRACE_RECORD_VALID)) {
127         return false;
128     }
129 
130     smp_rmb(); /* read memory barrier before accessing record */
131     /* read the record header to know record length */
132     read_from_buffer(idx, &record, sizeof(TraceRecord));
133     *recordptr = malloc(record.length); /* dont use g_malloc, can deadlock when traced */
134     /* make a copy of record to avoid being overwritten */
135     read_from_buffer(idx, *recordptr, record.length);
136     smp_rmb(); /* memory barrier before clearing valid flag */
137     (*recordptr)->event &= ~TRACE_RECORD_VALID;
138     /* clear the trace buffer range for consumed record otherwise any byte
139      * with its MSB set may be considered as a valid event id when the writer
140      * thread crosses this range of buffer again.
141      */
142     clear_buffer_range(idx, record.length);
143     return true;
144 }
145 
146 /**
147  * Kick writeout thread
148  *
149  * @wait        Whether to wait for writeout thread to complete
150  */
151 static void flush_trace_file(bool wait)
152 {
153     lock_trace_lock();
154     trace_available = true;
155     g_cond_signal(trace_available_cond);
156 
157     if (wait) {
158         g_cond_wait(trace_empty_cond, get_trace_lock_mutex());
159     }
160 
161     unlock_trace_lock();
162 }
163 
164 static void wait_for_trace_records_available(void)
165 {
166     lock_trace_lock();
167     while (!(trace_available && trace_writeout_enabled)) {
168         g_cond_signal(trace_empty_cond);
169         g_cond_wait(trace_available_cond, get_trace_lock_mutex());
170     }
171     trace_available = false;
172     unlock_trace_lock();
173 }
174 
175 static gpointer writeout_thread(gpointer opaque)
176 {
177     TraceRecord *recordptr;
178     union {
179         TraceRecord rec;
180         uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
181     } dropped;
182     unsigned int idx = 0;
183     int dropped_count;
184     size_t unused __attribute__ ((unused));
185 
186     for (;;) {
187         wait_for_trace_records_available();
188 
189         if (g_atomic_int_get(&dropped_events)) {
190             dropped.rec.event = DROPPED_EVENT_ID,
191             dropped.rec.timestamp_ns = get_clock();
192             dropped.rec.length = sizeof(TraceRecord) + sizeof(uint64_t),
193             dropped.rec.reserved = 0;
194             do {
195                 dropped_count = g_atomic_int_get(&dropped_events);
196             } while (!g_atomic_int_compare_and_exchange(&dropped_events,
197                                                         dropped_count, 0));
198             dropped.rec.arguments[0] = dropped_count;
199             unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
200         }
201 
202         while (get_trace_record(idx, &recordptr)) {
203             unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
204             writeout_idx += recordptr->length;
205             free(recordptr); /* dont use g_free, can deadlock when traced */
206             idx = writeout_idx % TRACE_BUF_LEN;
207         }
208 
209         fflush(trace_fp);
210     }
211     return NULL;
212 }
213 
214 void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
215 {
216     rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
217 }
218 
219 void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
220 {
221     /* Write string length first */
222     rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
223     /* Write actual string now */
224     rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
225 }
226 
227 int trace_record_start(TraceBufferRecord *rec, TraceEventID event, size_t datasize)
228 {
229     unsigned int idx, rec_off, old_idx, new_idx;
230     uint32_t rec_len = sizeof(TraceRecord) + datasize;
231     uint64_t event_u64 = event;
232     uint64_t timestamp_ns = get_clock();
233 
234     do {
235         old_idx = g_atomic_int_get(&trace_idx);
236         smp_rmb();
237         new_idx = old_idx + rec_len;
238 
239         if (new_idx - writeout_idx > TRACE_BUF_LEN) {
240             /* Trace Buffer Full, Event dropped ! */
241             g_atomic_int_inc(&dropped_events);
242             return -ENOSPC;
243         }
244     } while (!g_atomic_int_compare_and_exchange(&trace_idx, old_idx, new_idx));
245 
246     idx = old_idx % TRACE_BUF_LEN;
247 
248     rec_off = idx;
249     rec_off = write_to_buffer(rec_off, &event_u64, sizeof(event_u64));
250     rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
251     rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
252 
253     rec->tbuf_idx = idx;
254     rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
255     return 0;
256 }
257 
258 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
259 {
260     uint8_t *data_ptr = dataptr;
261     uint32_t x = 0;
262     while (x < size) {
263         if (idx >= TRACE_BUF_LEN) {
264             idx = idx % TRACE_BUF_LEN;
265         }
266         data_ptr[x++] = trace_buf[idx++];
267     }
268 }
269 
270 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
271 {
272     uint8_t *data_ptr = dataptr;
273     uint32_t x = 0;
274     while (x < size) {
275         if (idx >= TRACE_BUF_LEN) {
276             idx = idx % TRACE_BUF_LEN;
277         }
278         trace_buf[idx++] = data_ptr[x++];
279     }
280     return idx; /* most callers wants to know where to write next */
281 }
282 
283 void trace_record_finish(TraceBufferRecord *rec)
284 {
285     TraceRecord record;
286     read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
287     smp_wmb(); /* write barrier before marking as valid */
288     record.event |= TRACE_RECORD_VALID;
289     write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
290 
291     if (((unsigned int)g_atomic_int_get(&trace_idx) - writeout_idx)
292         > TRACE_BUF_FLUSH_THRESHOLD) {
293         flush_trace_file(false);
294     }
295 }
296 
297 void st_set_trace_file_enabled(bool enable)
298 {
299     if (enable == !!trace_fp) {
300         return; /* no change */
301     }
302 
303     /* Halt trace writeout */
304     flush_trace_file(true);
305     trace_writeout_enabled = false;
306     flush_trace_file(true);
307 
308     if (enable) {
309         static const TraceLogHeader header = {
310             .header_event_id = HEADER_EVENT_ID,
311             .header_magic = HEADER_MAGIC,
312             /* Older log readers will check for version at next location */
313             .header_version = HEADER_VERSION,
314         };
315 
316         trace_fp = fopen(trace_file_name, "wb");
317         if (!trace_fp) {
318             return;
319         }
320 
321         if (fwrite(&header, sizeof header, 1, trace_fp) != 1) {
322             fclose(trace_fp);
323             trace_fp = NULL;
324             return;
325         }
326 
327         /* Resume trace writeout */
328         trace_writeout_enabled = true;
329         flush_trace_file(false);
330     } else {
331         fclose(trace_fp);
332         trace_fp = NULL;
333     }
334 }
335 
336 /**
337  * Set the name of a trace file
338  *
339  * @file        The trace file name or NULL for the default name-<pid> set at
340  *              config time
341  */
342 bool st_set_trace_file(const char *file)
343 {
344     st_set_trace_file_enabled(false);
345 
346     g_free(trace_file_name);
347 
348     if (!file) {
349         trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, getpid());
350     } else {
351         trace_file_name = g_strdup_printf("%s", file);
352     }
353 
354     st_set_trace_file_enabled(true);
355     return true;
356 }
357 
358 void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
359 {
360     stream_printf(stream, "Trace file \"%s\" %s.\n",
361                   trace_file_name, trace_fp ? "on" : "off");
362 }
363 
364 void st_flush_trace_buffer(void)
365 {
366     flush_trace_file(true);
367 }
368 
369 void trace_print_events(FILE *stream, fprintf_function stream_printf)
370 {
371     unsigned int i;
372 
373     for (i = 0; i < trace_event_count(); i++) {
374         TraceEvent *ev = trace_event_id(i);
375         stream_printf(stream, "%s [Event ID %u] : state %u\n",
376                       trace_event_get_name(ev), i, trace_event_get_state_dynamic(ev));
377     }
378 }
379 
380 void trace_event_set_state_dynamic_backend(TraceEvent *ev, bool state)
381 {
382     ev->dstate = state;
383 }
384 
385 /* Helper function to create a thread with signals blocked.  Use glib's
386  * portable threads since QEMU abstractions cannot be used due to reentrancy in
387  * the tracer.  Also note the signal masking on POSIX hosts so that the thread
388  * does not steal signals when the rest of the program wants them blocked.
389  */
390 static GThread *trace_thread_create(GThreadFunc fn)
391 {
392     GThread *thread;
393 #ifndef _WIN32
394     sigset_t set, oldset;
395 
396     sigfillset(&set);
397     pthread_sigmask(SIG_SETMASK, &set, &oldset);
398 #endif
399 
400 #if GLIB_CHECK_VERSION(2, 31, 0)
401     thread = g_thread_new("trace-thread", fn, NULL);
402 #else
403     thread = g_thread_create(fn, NULL, FALSE, NULL);
404 #endif
405 
406 #ifndef _WIN32
407     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
408 #endif
409 
410     return thread;
411 }
412 
413 bool trace_backend_init(const char *events, const char *file)
414 {
415     GThread *thread;
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