xref: /openbmc/qemu/trace/simple.c (revision c39f95dc)
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/control.h"
17 #include "trace/simple.h"
18 #include "qemu/error-report.h"
19 
20 /** Trace file header event ID, picked to avoid conflict with real event IDs */
21 #define HEADER_EVENT_ID (~(uint64_t)0)
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 4
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 #define TRACE_RECORD_TYPE_MAPPING 0
60 #define TRACE_RECORD_TYPE_EVENT   1
61 
62 /* * Trace buffer entry */
63 typedef struct {
64     uint64_t event; /* event ID value */
65     uint64_t timestamp_ns;
66     uint32_t length;   /*    in bytes */
67     uint32_t pid;
68     uint64_t arguments[];
69 } TraceRecord;
70 
71 typedef struct {
72     uint64_t header_event_id; /* HEADER_EVENT_ID */
73     uint64_t header_magic;    /* HEADER_MAGIC    */
74     uint64_t header_version;  /* HEADER_VERSION  */
75 } TraceLogHeader;
76 
77 
78 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
79 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
80 
81 static void clear_buffer_range(unsigned int idx, size_t len)
82 {
83     uint32_t num = 0;
84     while (num < len) {
85         if (idx >= TRACE_BUF_LEN) {
86             idx = idx % TRACE_BUF_LEN;
87         }
88         trace_buf[idx++] = 0;
89         num++;
90     }
91 }
92 /**
93  * Read a trace record from the trace buffer
94  *
95  * @idx         Trace buffer index
96  * @record      Trace record to fill
97  *
98  * Returns false if the record is not valid.
99  */
100 static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
101 {
102     uint64_t event_flag = 0;
103     TraceRecord record;
104     /* read the event flag to see if its a valid record */
105     read_from_buffer(idx, &record, sizeof(event_flag));
106 
107     if (!(record.event & TRACE_RECORD_VALID)) {
108         return false;
109     }
110 
111     smp_rmb(); /* read memory barrier before accessing record */
112     /* read the record header to know record length */
113     read_from_buffer(idx, &record, sizeof(TraceRecord));
114     *recordptr = malloc(record.length); /* don't use g_malloc, can deadlock when traced */
115     /* make a copy of record to avoid being overwritten */
116     read_from_buffer(idx, *recordptr, record.length);
117     smp_rmb(); /* memory barrier before clearing valid flag */
118     (*recordptr)->event &= ~TRACE_RECORD_VALID;
119     /* clear the trace buffer range for consumed record otherwise any byte
120      * with its MSB set may be considered as a valid event id when the writer
121      * thread crosses this range of buffer again.
122      */
123     clear_buffer_range(idx, record.length);
124     return true;
125 }
126 
127 /**
128  * Kick writeout thread
129  *
130  * @wait        Whether to wait for writeout thread to complete
131  */
132 static void flush_trace_file(bool wait)
133 {
134     g_mutex_lock(&trace_lock);
135     trace_available = true;
136     g_cond_signal(&trace_available_cond);
137 
138     if (wait) {
139         g_cond_wait(&trace_empty_cond, &trace_lock);
140     }
141 
142     g_mutex_unlock(&trace_lock);
143 }
144 
145 static void wait_for_trace_records_available(void)
146 {
147     g_mutex_lock(&trace_lock);
148     while (!(trace_available && trace_writeout_enabled)) {
149         g_cond_signal(&trace_empty_cond);
150         g_cond_wait(&trace_available_cond, &trace_lock);
151     }
152     trace_available = false;
153     g_mutex_unlock(&trace_lock);
154 }
155 
156 static gpointer writeout_thread(gpointer opaque)
157 {
158     TraceRecord *recordptr;
159     union {
160         TraceRecord rec;
161         uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
162     } dropped;
163     unsigned int idx = 0;
164     int dropped_count;
165     size_t unused __attribute__ ((unused));
166     uint64_t type = TRACE_RECORD_TYPE_EVENT;
167 
168     for (;;) {
169         wait_for_trace_records_available();
170 
171         if (g_atomic_int_get(&dropped_events)) {
172             dropped.rec.event = DROPPED_EVENT_ID,
173             dropped.rec.timestamp_ns = get_clock();
174             dropped.rec.length = sizeof(TraceRecord) + sizeof(uint64_t),
175             dropped.rec.pid = trace_pid;
176             do {
177                 dropped_count = g_atomic_int_get(&dropped_events);
178             } while (!g_atomic_int_compare_and_exchange(&dropped_events,
179                                                         dropped_count, 0));
180             dropped.rec.arguments[0] = dropped_count;
181             unused = fwrite(&type, sizeof(type), 1, trace_fp);
182             unused = fwrite(&dropped.rec, dropped.rec.length, 1, trace_fp);
183         }
184 
185         while (get_trace_record(idx, &recordptr)) {
186             unused = fwrite(&type, sizeof(type), 1, trace_fp);
187             unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
188             writeout_idx += recordptr->length;
189             free(recordptr); /* don't use g_free, can deadlock when traced */
190             idx = writeout_idx % TRACE_BUF_LEN;
191         }
192 
193         fflush(trace_fp);
194     }
195     return NULL;
196 }
197 
198 void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
199 {
200     rec->rec_off = write_to_buffer(rec->rec_off, &val, sizeof(uint64_t));
201 }
202 
203 void trace_record_write_str(TraceBufferRecord *rec, const char *s, uint32_t slen)
204 {
205     /* Write string length first */
206     rec->rec_off = write_to_buffer(rec->rec_off, &slen, sizeof(slen));
207     /* Write actual string now */
208     rec->rec_off = write_to_buffer(rec->rec_off, (void*)s, slen);
209 }
210 
211 int trace_record_start(TraceBufferRecord *rec, uint32_t event, size_t datasize)
212 {
213     unsigned int idx, rec_off, old_idx, new_idx;
214     uint32_t rec_len = sizeof(TraceRecord) + datasize;
215     uint64_t event_u64 = event;
216     uint64_t timestamp_ns = get_clock();
217 
218     do {
219         old_idx = g_atomic_int_get(&trace_idx);
220         smp_rmb();
221         new_idx = old_idx + rec_len;
222 
223         if (new_idx - writeout_idx > TRACE_BUF_LEN) {
224             /* Trace Buffer Full, Event dropped ! */
225             g_atomic_int_inc(&dropped_events);
226             return -ENOSPC;
227         }
228     } while (!g_atomic_int_compare_and_exchange(&trace_idx, old_idx, new_idx));
229 
230     idx = old_idx % TRACE_BUF_LEN;
231 
232     rec_off = idx;
233     rec_off = write_to_buffer(rec_off, &event_u64, sizeof(event_u64));
234     rec_off = write_to_buffer(rec_off, &timestamp_ns, sizeof(timestamp_ns));
235     rec_off = write_to_buffer(rec_off, &rec_len, sizeof(rec_len));
236     rec_off = write_to_buffer(rec_off, &trace_pid, sizeof(trace_pid));
237 
238     rec->tbuf_idx = idx;
239     rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
240     return 0;
241 }
242 
243 static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
244 {
245     uint8_t *data_ptr = dataptr;
246     uint32_t x = 0;
247     while (x < size) {
248         if (idx >= TRACE_BUF_LEN) {
249             idx = idx % TRACE_BUF_LEN;
250         }
251         data_ptr[x++] = trace_buf[idx++];
252     }
253 }
254 
255 static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
256 {
257     uint8_t *data_ptr = dataptr;
258     uint32_t x = 0;
259     while (x < size) {
260         if (idx >= TRACE_BUF_LEN) {
261             idx = idx % TRACE_BUF_LEN;
262         }
263         trace_buf[idx++] = data_ptr[x++];
264     }
265     return idx; /* most callers wants to know where to write next */
266 }
267 
268 void trace_record_finish(TraceBufferRecord *rec)
269 {
270     TraceRecord record;
271     read_from_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
272     smp_wmb(); /* write barrier before marking as valid */
273     record.event |= TRACE_RECORD_VALID;
274     write_to_buffer(rec->tbuf_idx, &record, sizeof(TraceRecord));
275 
276     if (((unsigned int)g_atomic_int_get(&trace_idx) - writeout_idx)
277         > TRACE_BUF_FLUSH_THRESHOLD) {
278         flush_trace_file(false);
279     }
280 }
281 
282 static int st_write_event_mapping(void)
283 {
284     uint64_t type = TRACE_RECORD_TYPE_MAPPING;
285     TraceEventIter iter;
286     TraceEvent *ev;
287 
288     trace_event_iter_init(&iter, NULL);
289     while ((ev = trace_event_iter_next(&iter)) != NULL) {
290         uint64_t id = trace_event_get_id(ev);
291         const char *name = trace_event_get_name(ev);
292         uint32_t len = strlen(name);
293         if (fwrite(&type, sizeof(type), 1, trace_fp) != 1 ||
294             fwrite(&id, sizeof(id), 1, trace_fp) != 1 ||
295             fwrite(&len, sizeof(len), 1, trace_fp) != 1 ||
296             fwrite(name, len, 1, trace_fp) != 1) {
297             return -1;
298         }
299     }
300 
301     return 0;
302 }
303 
304 void st_set_trace_file_enabled(bool enable)
305 {
306     if (enable == !!trace_fp) {
307         return; /* no change */
308     }
309 
310     /* Halt trace writeout */
311     flush_trace_file(true);
312     trace_writeout_enabled = false;
313     flush_trace_file(true);
314 
315     if (enable) {
316         static const TraceLogHeader header = {
317             .header_event_id = HEADER_EVENT_ID,
318             .header_magic = HEADER_MAGIC,
319             /* Older log readers will check for version at next location */
320             .header_version = HEADER_VERSION,
321         };
322 
323         trace_fp = fopen(trace_file_name, "wb");
324         if (!trace_fp) {
325             return;
326         }
327 
328         if (fwrite(&header, sizeof header, 1, trace_fp) != 1 ||
329             st_write_event_mapping() < 0) {
330             fclose(trace_fp);
331             trace_fp = NULL;
332             return;
333         }
334 
335         /* Resume trace writeout */
336         trace_writeout_enabled = true;
337         flush_trace_file(false);
338     } else {
339         fclose(trace_fp);
340         trace_fp = NULL;
341     }
342 }
343 
344 /**
345  * Set the name of a trace file
346  *
347  * @file        The trace file name or NULL for the default name-<pid> set at
348  *              config time
349  */
350 void st_set_trace_file(const char *file)
351 {
352     st_set_trace_file_enabled(false);
353 
354     g_free(trace_file_name);
355 
356     if (!file) {
357         /* Type cast needed for Windows where getpid() returns an int. */
358         trace_file_name = g_strdup_printf(CONFIG_TRACE_FILE, (pid_t)getpid());
359     } else {
360         trace_file_name = g_strdup_printf("%s", file);
361     }
362 
363     st_set_trace_file_enabled(true);
364 }
365 
366 void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
367 {
368     stream_printf(stream, "Trace file \"%s\" %s.\n",
369                   trace_file_name, trace_fp ? "on" : "off");
370 }
371 
372 void st_flush_trace_buffer(void)
373 {
374     flush_trace_file(true);
375 }
376 
377 /* Helper function to create a thread with signals blocked.  Use glib's
378  * portable threads since QEMU abstractions cannot be used due to reentrancy in
379  * the tracer.  Also note the signal masking on POSIX hosts so that the thread
380  * does not steal signals when the rest of the program wants them blocked.
381  */
382 static GThread *trace_thread_create(GThreadFunc fn)
383 {
384     GThread *thread;
385 #ifndef _WIN32
386     sigset_t set, oldset;
387 
388     sigfillset(&set);
389     pthread_sigmask(SIG_SETMASK, &set, &oldset);
390 #endif
391 
392     thread = g_thread_new("trace-thread", fn, NULL);
393 
394 #ifndef _WIN32
395     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
396 #endif
397 
398     return thread;
399 }
400 
401 bool st_init(void)
402 {
403     GThread *thread;
404 
405     trace_pid = getpid();
406 
407     thread = trace_thread_create(writeout_thread);
408     if (!thread) {
409         warn_report("unable to initialize simple trace backend");
410         return false;
411     }
412 
413     atexit(st_flush_trace_buffer);
414     return true;
415 }
416