1 /*
2 * Copyright © 2020 Google, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 */
23
24 #include <inttypes.h>
25
26 #include "util/list.h"
27 #include "util/u_debug.h"
28 #include "util/u_inlines.h"
29 #include "util/u_fifo.h"
30 #include "util/u_vector.h"
31
32 #include "u_trace.h"
33
34 #define __NEEDS_TRACE_PRIV
35 #include "u_trace_priv.h"
36
37 #define PAYLOAD_BUFFER_SIZE 0x100
38 #define TIMESTAMP_BUF_SIZE 0x1000
39 #define TRACES_PER_CHUNK (TIMESTAMP_BUF_SIZE / sizeof(uint64_t))
40
41 #ifdef HAVE_PERFETTO
42 int ut_perfetto_enabled;
43
44 /**
45 * Global list of contexts, so we can defer starting the queue until
46 * perfetto tracing is started.
47 *
48 * TODO locking
49 */
50 struct list_head ctx_list = { &ctx_list, &ctx_list };
51 #endif
52
53 struct u_trace_payload_buf {
54 uint32_t refcount;
55
56 uint8_t *buf;
57 uint8_t *next;
58 uint8_t *end;
59 };
60
61 struct u_trace_event {
62 const struct u_tracepoint *tp;
63 const void *payload;
64 };
65
66 /**
67 * A "chunk" of trace-events and corresponding timestamp buffer. As
68 * trace events are emitted, additional trace chucks will be allocated
69 * as needed. When u_trace_flush() is called, they are transferred
70 * from the u_trace to the u_trace_context queue.
71 */
72 struct u_trace_chunk {
73 struct list_head node;
74
75 struct u_trace_context *utctx;
76
77 /* The number of traces this chunk contains so far: */
78 unsigned num_traces;
79
80 /* table of trace events: */
81 struct u_trace_event traces[TRACES_PER_CHUNK];
82
83 /* table of driver recorded 64b timestamps, index matches index
84 * into traces table
85 */
86 void *timestamps;
87
88 /* Array of u_trace_payload_buf referenced by traces[] elements.
89 */
90 struct u_vector payloads;
91
92 /* Current payload buffer being written. */
93 struct u_trace_payload_buf *payload;
94
95 struct util_queue_fence fence;
96
97 bool last; /* this chunk is last in batch */
98 bool eof; /* this chunk is last in frame */
99
100 void *flush_data; /* assigned by u_trace_flush */
101
102 /**
103 * Several chunks reference a single flush_data instance thus only
104 * one chunk should be designated to free the data.
105 */
106 bool free_flush_data;
107 };
108
109 static struct u_trace_payload_buf *
u_trace_payload_buf_create(void)110 u_trace_payload_buf_create(void)
111 {
112 struct u_trace_payload_buf *payload =
113 malloc(sizeof(*payload) + PAYLOAD_BUFFER_SIZE);
114
115 p_atomic_set(&payload->refcount, 1);
116
117 payload->buf = (uint8_t *) (payload + 1);
118 payload->end = payload->buf + PAYLOAD_BUFFER_SIZE;
119 payload->next = payload->buf;
120
121 return payload;
122 }
123
124 static struct u_trace_payload_buf *
u_trace_payload_buf_ref(struct u_trace_payload_buf * payload)125 u_trace_payload_buf_ref(struct u_trace_payload_buf *payload)
126 {
127 p_atomic_inc(&payload->refcount);
128 return payload;
129 }
130
131 static void
u_trace_payload_buf_unref(struct u_trace_payload_buf * payload)132 u_trace_payload_buf_unref(struct u_trace_payload_buf *payload)
133 {
134 if (p_atomic_dec_zero(&payload->refcount))
135 free(payload);
136 }
137
138 static void
free_chunk(void * ptr)139 free_chunk(void *ptr)
140 {
141 struct u_trace_chunk *chunk = ptr;
142
143 chunk->utctx->delete_timestamp_buffer(chunk->utctx, chunk->timestamps);
144
145 /* Unref payloads attached to this chunk. */
146 struct u_trace_payload_buf **payload;
147 u_vector_foreach(payload, &chunk->payloads)
148 u_trace_payload_buf_unref(*payload);
149 u_vector_finish(&chunk->payloads);
150
151 list_del(&chunk->node);
152 free(chunk);
153 }
154
155 static void
free_chunks(struct list_head * chunks)156 free_chunks(struct list_head *chunks)
157 {
158 while (!list_is_empty(chunks)) {
159 struct u_trace_chunk *chunk = list_first_entry(chunks,
160 struct u_trace_chunk, node);
161 free_chunk(chunk);
162 }
163 }
164
165 static struct u_trace_chunk *
get_chunk(struct u_trace * ut,size_t payload_size)166 get_chunk(struct u_trace *ut, size_t payload_size)
167 {
168 struct u_trace_chunk *chunk;
169
170 assert(payload_size <= PAYLOAD_BUFFER_SIZE);
171
172 /* do we currently have a non-full chunk to append msgs to? */
173 if (!list_is_empty(&ut->trace_chunks)) {
174 chunk = list_last_entry(&ut->trace_chunks,
175 struct u_trace_chunk, node);
176 /* Can we store a new trace in the chunk? */
177 if (chunk->num_traces < TRACES_PER_CHUNK) {
178 /* If no payload required, nothing else to check. */
179 if (payload_size <= 0)
180 return chunk;
181
182 /* If the payload buffer has space for the payload, we're good.
183 */
184 if (chunk->payload &&
185 (chunk->payload->end - chunk->payload->next) >= payload_size)
186 return chunk;
187
188 /* If we don't have enough space in the payload buffer, can we
189 * allocate a new one?
190 */
191 struct u_trace_payload_buf **buf = u_vector_add(&chunk->payloads);
192 *buf = u_trace_payload_buf_create();
193 chunk->payload = *buf;
194 return chunk;
195 }
196 /* we need to expand to add another chunk to the batch, so
197 * the current one is no longer the last one of the batch:
198 */
199 chunk->last = false;
200 }
201
202 /* .. if not, then create a new one: */
203 chunk = calloc(1, sizeof(*chunk));
204
205 chunk->utctx = ut->utctx;
206 chunk->timestamps = ut->utctx->create_timestamp_buffer(ut->utctx, TIMESTAMP_BUF_SIZE);
207 chunk->last = true;
208 u_vector_init(&chunk->payloads, 4, sizeof(struct u_trace_payload_buf *));
209 if (payload_size > 0) {
210 struct u_trace_payload_buf **buf = u_vector_add(&chunk->payloads);
211 *buf = u_trace_payload_buf_create();
212 chunk->payload = *buf;
213 }
214
215 list_addtail(&chunk->node, &ut->trace_chunks);
216
217 return chunk;
218 }
219
220 DEBUG_GET_ONCE_BOOL_OPTION(trace, "GPU_TRACE", false)
221 DEBUG_GET_ONCE_FILE_OPTION(trace_file, "GPU_TRACEFILE", NULL, "w")
222
223 static FILE *
get_tracefile(void)224 get_tracefile(void)
225 {
226 static FILE *tracefile = NULL;
227 static bool firsttime = true;
228
229 if (firsttime) {
230 tracefile = debug_get_option_trace_file();
231 if (!tracefile && debug_get_option_trace()) {
232 tracefile = stdout;
233 }
234
235 firsttime = false;
236 }
237
238 return tracefile;
239 }
240
241 static void
queue_init(struct u_trace_context * utctx)242 queue_init(struct u_trace_context *utctx)
243 {
244 if (utctx->queue.jobs)
245 return;
246
247 bool ret = util_queue_init(&utctx->queue, "traceq", 256, 1,
248 UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY |
249 UTIL_QUEUE_INIT_RESIZE_IF_FULL, NULL);
250 assert(ret);
251
252 if (!ret)
253 utctx->out = NULL;
254 }
255
256 void
u_trace_context_init(struct u_trace_context * utctx,void * pctx,u_trace_create_ts_buffer create_timestamp_buffer,u_trace_delete_ts_buffer delete_timestamp_buffer,u_trace_record_ts record_timestamp,u_trace_read_ts read_timestamp,u_trace_delete_flush_data delete_flush_data)257 u_trace_context_init(struct u_trace_context *utctx,
258 void *pctx,
259 u_trace_create_ts_buffer create_timestamp_buffer,
260 u_trace_delete_ts_buffer delete_timestamp_buffer,
261 u_trace_record_ts record_timestamp,
262 u_trace_read_ts read_timestamp,
263 u_trace_delete_flush_data delete_flush_data)
264 {
265 utctx->pctx = pctx;
266 utctx->create_timestamp_buffer = create_timestamp_buffer;
267 utctx->delete_timestamp_buffer = delete_timestamp_buffer;
268 utctx->record_timestamp = record_timestamp;
269 utctx->read_timestamp = read_timestamp;
270 utctx->delete_flush_data = delete_flush_data;
271
272 utctx->last_time_ns = 0;
273 utctx->first_time_ns = 0;
274 utctx->frame_nr = 0;
275
276 list_inithead(&utctx->flushed_trace_chunks);
277
278 utctx->out = get_tracefile();
279
280 #ifdef HAVE_PERFETTO
281 list_add(&utctx->node, &ctx_list);
282 #endif
283
284 if (!u_trace_context_tracing(utctx))
285 return;
286
287 queue_init(utctx);
288 }
289
290 void
u_trace_context_fini(struct u_trace_context * utctx)291 u_trace_context_fini(struct u_trace_context *utctx)
292 {
293 #ifdef HAVE_PERFETTO
294 list_del(&utctx->node);
295 #endif
296 if (!utctx->queue.jobs)
297 return;
298 util_queue_finish(&utctx->queue);
299 util_queue_destroy(&utctx->queue);
300 fflush(utctx->out);
301 free_chunks(&utctx->flushed_trace_chunks);
302 }
303
304 #ifdef HAVE_PERFETTO
305 void
u_trace_perfetto_start(void)306 u_trace_perfetto_start(void)
307 {
308 list_for_each_entry (struct u_trace_context, utctx, &ctx_list, node)
309 queue_init(utctx);
310 ut_perfetto_enabled++;
311 }
312
313 void
u_trace_perfetto_stop(void)314 u_trace_perfetto_stop(void)
315 {
316 assert(ut_perfetto_enabled > 0);
317 ut_perfetto_enabled--;
318 }
319 #endif
320
321 static void
process_chunk(void * job,void * gdata,int thread_index)322 process_chunk(void *job, void *gdata, int thread_index)
323 {
324 struct u_trace_chunk *chunk = job;
325 struct u_trace_context *utctx = chunk->utctx;
326
327 /* For first chunk of batch, accumulated times will be zerod: */
328 if (utctx->out && !utctx->last_time_ns) {
329 fprintf(utctx->out, "+----- NS -----+ +-- Δ --+ +----- MSG -----\n");
330 }
331
332 for (unsigned idx = 0; idx < chunk->num_traces; idx++) {
333 const struct u_trace_event *evt = &chunk->traces[idx];
334
335 if (!evt->tp)
336 continue;
337
338 uint64_t ns = utctx->read_timestamp(utctx, chunk->timestamps, idx, chunk->flush_data);
339 int32_t delta;
340
341 if (!utctx->first_time_ns)
342 utctx->first_time_ns = ns;
343
344 if (ns != U_TRACE_NO_TIMESTAMP) {
345 delta = utctx->last_time_ns ? ns - utctx->last_time_ns : 0;
346 utctx->last_time_ns = ns;
347 } else {
348 /* we skipped recording the timestamp, so it should be
349 * the same as last msg:
350 */
351 ns = utctx->last_time_ns;
352 delta = 0;
353 }
354
355 if (utctx->out) {
356 if (evt->tp->print) {
357 fprintf(utctx->out, "%016"PRIu64" %+9d: %s: ", ns, delta, evt->tp->name);
358 evt->tp->print(utctx->out, evt->payload);
359 } else {
360 fprintf(utctx->out, "%016"PRIu64" %+9d: %s\n", ns, delta, evt->tp->name);
361 }
362 }
363 #ifdef HAVE_PERFETTO
364 if (evt->tp->perfetto) {
365 evt->tp->perfetto(utctx->pctx, ns, chunk->flush_data, evt->payload);
366 }
367 #endif
368 }
369
370 if (chunk->last) {
371 if (utctx->out) {
372 uint64_t elapsed = utctx->last_time_ns - utctx->first_time_ns;
373 fprintf(utctx->out, "ELAPSED: %"PRIu64" ns\n", elapsed);
374 }
375
376 utctx->last_time_ns = 0;
377 utctx->first_time_ns = 0;
378 }
379
380 if (chunk->free_flush_data && utctx->delete_flush_data) {
381 utctx->delete_flush_data(utctx, chunk->flush_data);
382 }
383
384 if (utctx->out && chunk->eof) {
385 fprintf(utctx->out, "END OF FRAME %u\n", utctx->frame_nr++);
386 }
387 }
388
389 static void
cleanup_chunk(void * job,void * gdata,int thread_index)390 cleanup_chunk(void *job, void *gdata, int thread_index)
391 {
392 free_chunk(job);
393 }
394
395 void
u_trace_context_process(struct u_trace_context * utctx,bool eof)396 u_trace_context_process(struct u_trace_context *utctx, bool eof)
397 {
398 struct list_head *chunks = &utctx->flushed_trace_chunks;
399
400 if (list_is_empty(chunks))
401 return;
402
403 struct u_trace_chunk *last_chunk = list_last_entry(chunks,
404 struct u_trace_chunk, node);
405 last_chunk->eof = eof;
406
407 while (!list_is_empty(chunks)) {
408 struct u_trace_chunk *chunk = list_first_entry(chunks,
409 struct u_trace_chunk, node);
410
411 /* remove from list before enqueuing, because chunk is freed
412 * once it is processed by the queue:
413 */
414 list_delinit(&chunk->node);
415
416 util_queue_add_job(&utctx->queue, chunk, &chunk->fence,
417 process_chunk, cleanup_chunk,
418 TIMESTAMP_BUF_SIZE);
419 }
420 }
421
422
423 void
u_trace_init(struct u_trace * ut,struct u_trace_context * utctx)424 u_trace_init(struct u_trace *ut, struct u_trace_context *utctx)
425 {
426 ut->utctx = utctx;
427 list_inithead(&ut->trace_chunks);
428 ut->enabled = u_trace_context_tracing(utctx);
429 }
430
431 void
u_trace_fini(struct u_trace * ut)432 u_trace_fini(struct u_trace *ut)
433 {
434 /* Normally the list of trace-chunks would be empty, if they
435 * have been flushed to the trace-context.
436 */
437 free_chunks(&ut->trace_chunks);
438 }
439
440 bool
u_trace_has_points(struct u_trace * ut)441 u_trace_has_points(struct u_trace *ut)
442 {
443 return !list_is_empty(&ut->trace_chunks);
444 }
445
446 struct u_trace_iterator
u_trace_begin_iterator(struct u_trace * ut)447 u_trace_begin_iterator(struct u_trace *ut)
448 {
449 if (!ut->enabled)
450 return (struct u_trace_iterator) {NULL, NULL, 0};
451
452 struct u_trace_chunk *first_chunk =
453 list_first_entry(&ut->trace_chunks, struct u_trace_chunk, node);
454
455 return (struct u_trace_iterator) { ut, first_chunk, 0};
456 }
457
458 struct u_trace_iterator
u_trace_end_iterator(struct u_trace * ut)459 u_trace_end_iterator(struct u_trace *ut)
460 {
461 if (!ut->enabled)
462 return (struct u_trace_iterator) {NULL, NULL, 0};
463
464 struct u_trace_chunk *last_chunk =
465 list_last_entry(&ut->trace_chunks, struct u_trace_chunk, node);
466
467 return (struct u_trace_iterator) { ut, last_chunk, last_chunk->num_traces};
468 }
469
470 bool
u_trace_iterator_equal(struct u_trace_iterator a,struct u_trace_iterator b)471 u_trace_iterator_equal(struct u_trace_iterator a,
472 struct u_trace_iterator b)
473 {
474 return a.ut == b.ut &&
475 a.chunk == b.chunk &&
476 a.event_idx == b.event_idx;
477 }
478
479 void
u_trace_clone_append(struct u_trace_iterator begin_it,struct u_trace_iterator end_it,struct u_trace * into,void * cmdstream,u_trace_copy_ts_buffer copy_ts_buffer)480 u_trace_clone_append(struct u_trace_iterator begin_it,
481 struct u_trace_iterator end_it,
482 struct u_trace *into,
483 void *cmdstream,
484 u_trace_copy_ts_buffer copy_ts_buffer)
485 {
486 struct u_trace_chunk *from_chunk = begin_it.chunk;
487 uint32_t from_idx = begin_it.event_idx;
488
489 while (from_chunk != end_it.chunk || from_idx != end_it.event_idx) {
490 struct u_trace_chunk *to_chunk = get_chunk(into, 0 /* payload_size */);
491
492 unsigned to_copy = MIN2(TRACES_PER_CHUNK - to_chunk->num_traces,
493 from_chunk->num_traces - from_idx);
494 if (from_chunk == end_it.chunk)
495 to_copy = MIN2(to_copy, end_it.event_idx - from_idx);
496
497 copy_ts_buffer(begin_it.ut->utctx, cmdstream,
498 from_chunk->timestamps, from_idx,
499 to_chunk->timestamps, to_chunk->num_traces,
500 to_copy);
501
502 memcpy(&to_chunk->traces[to_chunk->num_traces],
503 &from_chunk->traces[from_idx],
504 to_copy * sizeof(struct u_trace_event));
505
506 /* Take a refcount on payloads from from_chunk if needed. */
507 if (begin_it.ut != into) {
508 struct u_trace_payload_buf **in_payload;
509 u_vector_foreach(in_payload, &from_chunk->payloads) {
510 struct u_trace_payload_buf **out_payload =
511 u_vector_add(&to_chunk->payloads);
512
513 *out_payload = u_trace_payload_buf_ref(*in_payload);
514 }
515 }
516
517 to_chunk->num_traces += to_copy;
518 from_idx += to_copy;
519
520 assert(from_idx <= from_chunk->num_traces);
521 if (from_idx == from_chunk->num_traces) {
522 if (from_chunk == end_it.chunk)
523 break;
524
525 from_idx = 0;
526 from_chunk = LIST_ENTRY(struct u_trace_chunk, from_chunk->node.next, node);
527 }
528 }
529 }
530
531 void
u_trace_disable_event_range(struct u_trace_iterator begin_it,struct u_trace_iterator end_it)532 u_trace_disable_event_range(struct u_trace_iterator begin_it,
533 struct u_trace_iterator end_it)
534 {
535 struct u_trace_chunk *current_chunk = begin_it.chunk;
536 uint32_t start_idx = begin_it.event_idx;
537
538 while(current_chunk != end_it.chunk) {
539 memset(¤t_chunk->traces[start_idx], 0,
540 (current_chunk->num_traces - start_idx) * sizeof(struct u_trace_event));
541 start_idx = 0;
542 current_chunk = LIST_ENTRY(struct u_trace_chunk, current_chunk->node.next, node);
543 }
544
545 memset(¤t_chunk->traces[start_idx], 0,
546 (end_it.event_idx - start_idx) * sizeof(struct u_trace_event));
547 }
548
549 /**
550 * Append a trace event, returning pointer to buffer of tp->payload_sz
551 * to be filled in with trace payload. Called by generated tracepoint
552 * functions.
553 */
554 void *
u_trace_append(struct u_trace * ut,void * cs,const struct u_tracepoint * tp)555 u_trace_append(struct u_trace *ut, void *cs, const struct u_tracepoint *tp)
556 {
557 struct u_trace_chunk *chunk = get_chunk(ut, tp->payload_sz);
558
559 assert(tp->payload_sz == ALIGN_NPOT(tp->payload_sz, 8));
560
561 /* sub-allocate storage for trace payload: */
562 void *payload = NULL;
563 if (tp->payload_sz > 0) {
564 payload = chunk->payload->next;
565 chunk->payload->next += tp->payload_sz;
566 }
567
568 /* record a timestamp for the trace: */
569 ut->utctx->record_timestamp(ut, cs, chunk->timestamps, chunk->num_traces);
570
571 chunk->traces[chunk->num_traces] = (struct u_trace_event) {
572 .tp = tp,
573 .payload = payload,
574 };
575
576 chunk->num_traces++;
577
578 return payload;
579 }
580
581 void
u_trace_flush(struct u_trace * ut,void * flush_data,bool free_data)582 u_trace_flush(struct u_trace *ut, void *flush_data, bool free_data)
583 {
584 list_for_each_entry(struct u_trace_chunk, chunk, &ut->trace_chunks, node) {
585 chunk->flush_data = flush_data;
586 chunk->free_flush_data = false;
587 }
588
589 if (free_data && !list_is_empty(&ut->trace_chunks)) {
590 struct u_trace_chunk *last_chunk =
591 list_last_entry(&ut->trace_chunks, struct u_trace_chunk, node);
592 last_chunk->free_flush_data = true;
593 }
594
595 /* transfer batch's log chunks to context: */
596 list_splicetail(&ut->trace_chunks, &ut->utctx->flushed_trace_chunks);
597 list_inithead(&ut->trace_chunks);
598 }
599