1 /* SPDX-License-Identifier: GPL-2.0 */
2 #undef TRACE_SYSTEM
3 #define TRACE_SYSTEM io_uring
4
5 #if !defined(_TRACE_IO_URING_H) || defined(TRACE_HEADER_MULTI_READ)
6 #define _TRACE_IO_URING_H
7
8 #include <linux/tracepoint.h>
9 #include <uapi/linux/io_uring.h>
10 #include <linux/io_uring_types.h>
11 #include <linux/io_uring.h>
12
13 struct io_wq_work;
14
15 /**
16 * io_uring_create - called after a new io_uring context was prepared
17 *
18 * @fd: corresponding file descriptor
19 * @ctx: pointer to a ring context structure
20 * @sq_entries: actual SQ size
21 * @cq_entries: actual CQ size
22 * @flags: SQ ring flags, provided to io_uring_setup(2)
23 *
24 * Allows to trace io_uring creation and provide pointer to a context, that can
25 * be used later to find correlated events.
26 */
27 TRACE_EVENT(io_uring_create,
28
29 TP_PROTO(int fd, void *ctx, u32 sq_entries, u32 cq_entries, u32 flags),
30
31 TP_ARGS(fd, ctx, sq_entries, cq_entries, flags),
32
33 TP_STRUCT__entry (
34 __field( int, fd )
35 __field( void *, ctx )
36 __field( u32, sq_entries )
37 __field( u32, cq_entries )
38 __field( u32, flags )
39 ),
40
41 TP_fast_assign(
42 __entry->fd = fd;
43 __entry->ctx = ctx;
44 __entry->sq_entries = sq_entries;
45 __entry->cq_entries = cq_entries;
46 __entry->flags = flags;
47 ),
48
49 TP_printk("ring %p, fd %d sq size %d, cq size %d, flags 0x%x",
50 __entry->ctx, __entry->fd, __entry->sq_entries,
51 __entry->cq_entries, __entry->flags)
52 );
53
54 /**
55 * io_uring_register - called after a buffer/file/eventfd was successfully
56 * registered for a ring
57 *
58 * @ctx: pointer to a ring context structure
59 * @opcode: describes which operation to perform
60 * @nr_user_files: number of registered files
61 * @nr_user_bufs: number of registered buffers
62 * @ret: return code
63 *
64 * Allows to trace fixed files/buffers, that could be registered to
65 * avoid an overhead of getting references to them for every operation. This
66 * event, together with io_uring_file_get, can provide a full picture of how
67 * much overhead one can reduce via fixing.
68 */
69 TRACE_EVENT(io_uring_register,
70
71 TP_PROTO(void *ctx, unsigned opcode, unsigned nr_files,
72 unsigned nr_bufs, long ret),
73
74 TP_ARGS(ctx, opcode, nr_files, nr_bufs, ret),
75
76 TP_STRUCT__entry (
77 __field( void *, ctx )
78 __field( unsigned, opcode )
79 __field( unsigned, nr_files)
80 __field( unsigned, nr_bufs )
81 __field( long, ret )
82 ),
83
84 TP_fast_assign(
85 __entry->ctx = ctx;
86 __entry->opcode = opcode;
87 __entry->nr_files = nr_files;
88 __entry->nr_bufs = nr_bufs;
89 __entry->ret = ret;
90 ),
91
92 TP_printk("ring %p, opcode %d, nr_user_files %d, nr_user_bufs %d, "
93 "ret %ld",
94 __entry->ctx, __entry->opcode, __entry->nr_files,
95 __entry->nr_bufs, __entry->ret)
96 );
97
98 /**
99 * io_uring_file_get - called before getting references to an SQE file
100 *
101 * @req: pointer to a submitted request
102 * @fd: SQE file descriptor
103 *
104 * Allows to trace out how often an SQE file reference is obtained, which can
105 * help figuring out if it makes sense to use fixed files, or check that fixed
106 * files are used correctly.
107 */
108 TRACE_EVENT(io_uring_file_get,
109
110 TP_PROTO(struct io_kiocb *req, int fd),
111
112 TP_ARGS(req, fd),
113
114 TP_STRUCT__entry (
115 __field( void *, ctx )
116 __field( void *, req )
117 __field( u64, user_data )
118 __field( int, fd )
119 ),
120
121 TP_fast_assign(
122 __entry->ctx = req->ctx;
123 __entry->req = req;
124 __entry->user_data = req->cqe.user_data;
125 __entry->fd = fd;
126 ),
127
128 TP_printk("ring %p, req %p, user_data 0x%llx, fd %d",
129 __entry->ctx, __entry->req, __entry->user_data, __entry->fd)
130 );
131
132 /**
133 * io_uring_queue_async_work - called before submitting a new async work
134 *
135 * @req: pointer to a submitted request
136 * @rw: type of workqueue, hashed or normal
137 *
138 * Allows to trace asynchronous work submission.
139 */
140 TRACE_EVENT(io_uring_queue_async_work,
141
142 TP_PROTO(struct io_kiocb *req, int rw),
143
144 TP_ARGS(req, rw),
145
146 TP_STRUCT__entry (
147 __field( void *, ctx )
148 __field( void *, req )
149 __field( u64, user_data )
150 __field( u8, opcode )
151 __field( unsigned long long, flags )
152 __field( struct io_wq_work *, work )
153 __field( int, rw )
154
155 __string( op_str, io_uring_get_opcode(req->opcode) )
156 ),
157
158 TP_fast_assign(
159 __entry->ctx = req->ctx;
160 __entry->req = req;
161 __entry->user_data = req->cqe.user_data;
162 __entry->flags = (__force unsigned long long) req->flags;
163 __entry->opcode = req->opcode;
164 __entry->work = &req->work;
165 __entry->rw = rw;
166
167 __assign_str(op_str);
168 ),
169
170 TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%llx, %s queue, work %p",
171 __entry->ctx, __entry->req, __entry->user_data,
172 __get_str(op_str), __entry->flags,
173 __entry->rw ? "hashed" : "normal", __entry->work)
174 );
175
176 /**
177 * io_uring_defer - called when an io_uring request is deferred
178 *
179 * @req: pointer to a deferred request
180 *
181 * Allows to track deferred requests, to get an insight about what requests are
182 * not started immediately.
183 */
184 TRACE_EVENT(io_uring_defer,
185
186 TP_PROTO(struct io_kiocb *req),
187
188 TP_ARGS(req),
189
190 TP_STRUCT__entry (
191 __field( void *, ctx )
192 __field( void *, req )
193 __field( unsigned long long, data )
194 __field( u8, opcode )
195
196 __string( op_str, io_uring_get_opcode(req->opcode) )
197 ),
198
199 TP_fast_assign(
200 __entry->ctx = req->ctx;
201 __entry->req = req;
202 __entry->data = req->cqe.user_data;
203 __entry->opcode = req->opcode;
204
205 __assign_str(op_str);
206 ),
207
208 TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s",
209 __entry->ctx, __entry->req, __entry->data,
210 __get_str(op_str))
211 );
212
213 /**
214 * io_uring_link - called before the io_uring request added into link_list of
215 * another request
216 *
217 * @req: pointer to a linked request
218 * @target_req: pointer to a previous request, that would contain @req
219 *
220 * Allows to track linked requests, to understand dependencies between requests
221 * and how does it influence their execution flow.
222 */
223 TRACE_EVENT(io_uring_link,
224
225 TP_PROTO(struct io_kiocb *req, struct io_kiocb *target_req),
226
227 TP_ARGS(req, target_req),
228
229 TP_STRUCT__entry (
230 __field( void *, ctx )
231 __field( void *, req )
232 __field( void *, target_req )
233 ),
234
235 TP_fast_assign(
236 __entry->ctx = req->ctx;
237 __entry->req = req;
238 __entry->target_req = target_req;
239 ),
240
241 TP_printk("ring %p, request %p linked after %p",
242 __entry->ctx, __entry->req, __entry->target_req)
243 );
244
245 /**
246 * io_uring_cqring_wait - called before start waiting for an available CQE
247 *
248 * @ctx: pointer to a ring context structure
249 * @min_events: minimal number of events to wait for
250 *
251 * Allows to track waiting for CQE, so that we can e.g. troubleshoot
252 * situations, when an application wants to wait for an event, that never
253 * comes.
254 */
255 TRACE_EVENT(io_uring_cqring_wait,
256
257 TP_PROTO(void *ctx, int min_events),
258
259 TP_ARGS(ctx, min_events),
260
261 TP_STRUCT__entry (
262 __field( void *, ctx )
263 __field( int, min_events )
264 ),
265
266 TP_fast_assign(
267 __entry->ctx = ctx;
268 __entry->min_events = min_events;
269 ),
270
271 TP_printk("ring %p, min_events %d", __entry->ctx, __entry->min_events)
272 );
273
274 /**
275 * io_uring_fail_link - called before failing a linked request
276 *
277 * @req: request, which links were cancelled
278 * @link: cancelled link
279 *
280 * Allows to track linked requests cancellation, to see not only that some work
281 * was cancelled, but also which request was the reason.
282 */
283 TRACE_EVENT(io_uring_fail_link,
284
285 TP_PROTO(struct io_kiocb *req, struct io_kiocb *link),
286
287 TP_ARGS(req, link),
288
289 TP_STRUCT__entry (
290 __field( void *, ctx )
291 __field( void *, req )
292 __field( unsigned long long, user_data )
293 __field( u8, opcode )
294 __field( void *, link )
295
296 __string( op_str, io_uring_get_opcode(req->opcode) )
297 ),
298
299 TP_fast_assign(
300 __entry->ctx = req->ctx;
301 __entry->req = req;
302 __entry->user_data = req->cqe.user_data;
303 __entry->opcode = req->opcode;
304 __entry->link = link;
305
306 __assign_str(op_str);
307 ),
308
309 TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, link %p",
310 __entry->ctx, __entry->req, __entry->user_data,
311 __get_str(op_str), __entry->link)
312 );
313
314 /**
315 * io_uring_complete - called when completing an SQE
316 *
317 * @ctx: pointer to a ring context structure
318 * @req: pointer to a submitted request
319 * @user_data: user data associated with the request
320 * @res: result of the request
321 * @cflags: completion flags
322 * @extra1: extra 64-bit data for CQE32
323 * @extra2: extra 64-bit data for CQE32
324 *
325 */
326 TRACE_EVENT(io_uring_complete,
327
328 TP_PROTO(void *ctx, void *req, u64 user_data, int res, unsigned cflags,
329 u64 extra1, u64 extra2),
330
331 TP_ARGS(ctx, req, user_data, res, cflags, extra1, extra2),
332
333 TP_STRUCT__entry (
334 __field( void *, ctx )
335 __field( void *, req )
336 __field( u64, user_data )
337 __field( int, res )
338 __field( unsigned, cflags )
339 __field( u64, extra1 )
340 __field( u64, extra2 )
341 ),
342
343 TP_fast_assign(
344 __entry->ctx = ctx;
345 __entry->req = req;
346 __entry->user_data = user_data;
347 __entry->res = res;
348 __entry->cflags = cflags;
349 __entry->extra1 = extra1;
350 __entry->extra2 = extra2;
351 ),
352
353 TP_printk("ring %p, req %p, user_data 0x%llx, result %d, cflags 0x%x "
354 "extra1 %llu extra2 %llu ",
355 __entry->ctx, __entry->req,
356 __entry->user_data,
357 __entry->res, __entry->cflags,
358 (unsigned long long) __entry->extra1,
359 (unsigned long long) __entry->extra2)
360 );
361
362 /**
363 * io_uring_submit_req - called before submitting a request
364 *
365 * @req: pointer to a submitted request
366 *
367 * Allows to track SQE submitting, to understand what was the source of it, SQ
368 * thread or io_uring_enter call.
369 */
370 TRACE_EVENT(io_uring_submit_req,
371
372 TP_PROTO(struct io_kiocb *req),
373
374 TP_ARGS(req),
375
376 TP_STRUCT__entry (
377 __field( void *, ctx )
378 __field( void *, req )
379 __field( unsigned long long, user_data )
380 __field( u8, opcode )
381 __field( unsigned long long, flags )
382 __field( bool, sq_thread )
383
384 __string( op_str, io_uring_get_opcode(req->opcode) )
385 ),
386
387 TP_fast_assign(
388 __entry->ctx = req->ctx;
389 __entry->req = req;
390 __entry->user_data = req->cqe.user_data;
391 __entry->opcode = req->opcode;
392 __entry->flags = (__force unsigned long long) req->flags;
393 __entry->sq_thread = req->ctx->flags & IORING_SETUP_SQPOLL;
394
395 __assign_str(op_str);
396 ),
397
398 TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%llx, "
399 "sq_thread %d", __entry->ctx, __entry->req,
400 __entry->user_data, __get_str(op_str), __entry->flags,
401 __entry->sq_thread)
402 );
403
404 /*
405 * io_uring_poll_arm - called after arming a poll wait if successful
406 *
407 * @req: pointer to the armed request
408 * @mask: request poll events mask
409 * @events: registered events of interest
410 *
411 * Allows to track which fds are waiting for and what are the events of
412 * interest.
413 */
414 TRACE_EVENT(io_uring_poll_arm,
415
416 TP_PROTO(struct io_kiocb *req, int mask, int events),
417
418 TP_ARGS(req, mask, events),
419
420 TP_STRUCT__entry (
421 __field( void *, ctx )
422 __field( void *, req )
423 __field( unsigned long long, user_data )
424 __field( u8, opcode )
425 __field( int, mask )
426 __field( int, events )
427
428 __string( op_str, io_uring_get_opcode(req->opcode) )
429 ),
430
431 TP_fast_assign(
432 __entry->ctx = req->ctx;
433 __entry->req = req;
434 __entry->user_data = req->cqe.user_data;
435 __entry->opcode = req->opcode;
436 __entry->mask = mask;
437 __entry->events = events;
438
439 __assign_str(op_str);
440 ),
441
442 TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, mask 0x%x, events 0x%x",
443 __entry->ctx, __entry->req, __entry->user_data,
444 __get_str(op_str),
445 __entry->mask, __entry->events)
446 );
447
448 /*
449 * io_uring_task_add - called after adding a task
450 *
451 * @req: pointer to request
452 * @mask: request poll events mask
453 *
454 */
455 TRACE_EVENT(io_uring_task_add,
456
457 TP_PROTO(struct io_kiocb *req, int mask),
458
459 TP_ARGS(req, mask),
460
461 TP_STRUCT__entry (
462 __field( void *, ctx )
463 __field( void *, req )
464 __field( unsigned long long, user_data )
465 __field( u8, opcode )
466 __field( int, mask )
467
468 __string( op_str, io_uring_get_opcode(req->opcode) )
469 ),
470
471 TP_fast_assign(
472 __entry->ctx = req->ctx;
473 __entry->req = req;
474 __entry->user_data = req->cqe.user_data;
475 __entry->opcode = req->opcode;
476 __entry->mask = mask;
477
478 __assign_str(op_str);
479 ),
480
481 TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, mask %x",
482 __entry->ctx, __entry->req, __entry->user_data,
483 __get_str(op_str),
484 __entry->mask)
485 );
486
487 /*
488 * io_uring_req_failed - called when an sqe is errored dring submission
489 *
490 * @sqe: pointer to the io_uring_sqe that failed
491 * @req: pointer to request
492 * @error: error it failed with
493 *
494 * Allows easier diagnosing of malformed requests in production systems.
495 */
496 TRACE_EVENT(io_uring_req_failed,
497
498 TP_PROTO(const struct io_uring_sqe *sqe, struct io_kiocb *req, int error),
499
500 TP_ARGS(sqe, req, error),
501
502 TP_STRUCT__entry (
503 __field( void *, ctx )
504 __field( void *, req )
505 __field( unsigned long long, user_data )
506 __field( u8, opcode )
507 __field( u8, flags )
508 __field( u8, ioprio )
509 __field( u64, off )
510 __field( u64, addr )
511 __field( u32, len )
512 __field( u32, op_flags )
513 __field( u16, buf_index )
514 __field( u16, personality )
515 __field( u32, file_index )
516 __field( u64, pad1 )
517 __field( u64, addr3 )
518 __field( int, error )
519
520 __string( op_str, io_uring_get_opcode(sqe->opcode) )
521 ),
522
523 TP_fast_assign(
524 __entry->ctx = req->ctx;
525 __entry->req = req;
526 __entry->user_data = sqe->user_data;
527 __entry->opcode = sqe->opcode;
528 __entry->flags = sqe->flags;
529 __entry->ioprio = sqe->ioprio;
530 __entry->off = sqe->off;
531 __entry->addr = sqe->addr;
532 __entry->len = sqe->len;
533 __entry->op_flags = sqe->poll32_events;
534 __entry->buf_index = sqe->buf_index;
535 __entry->personality = sqe->personality;
536 __entry->file_index = sqe->file_index;
537 __entry->pad1 = sqe->__pad2[0];
538 __entry->addr3 = sqe->addr3;
539 __entry->error = error;
540
541 __assign_str(op_str);
542 ),
543
544 TP_printk("ring %p, req %p, user_data 0x%llx, "
545 "opcode %s, flags 0x%x, prio=%d, off=%llu, addr=%llu, "
546 "len=%u, rw_flags=0x%x, buf_index=%d, "
547 "personality=%d, file_index=%d, pad=0x%llx, addr3=%llx, "
548 "error=%d",
549 __entry->ctx, __entry->req, __entry->user_data,
550 __get_str(op_str),
551 __entry->flags, __entry->ioprio,
552 (unsigned long long)__entry->off,
553 (unsigned long long) __entry->addr, __entry->len,
554 __entry->op_flags,
555 __entry->buf_index, __entry->personality, __entry->file_index,
556 (unsigned long long) __entry->pad1,
557 (unsigned long long) __entry->addr3, __entry->error)
558 );
559
560
561 /*
562 * io_uring_cqe_overflow - a CQE overflowed
563 *
564 * @ctx: pointer to a ring context structure
565 * @user_data: user data associated with the request
566 * @res: CQE result
567 * @cflags: CQE flags
568 * @ocqe: pointer to the overflow cqe (if available)
569 *
570 */
571 TRACE_EVENT(io_uring_cqe_overflow,
572
573 TP_PROTO(void *ctx, unsigned long long user_data, s32 res, u32 cflags,
574 void *ocqe),
575
576 TP_ARGS(ctx, user_data, res, cflags, ocqe),
577
578 TP_STRUCT__entry (
579 __field( void *, ctx )
580 __field( unsigned long long, user_data )
581 __field( s32, res )
582 __field( u32, cflags )
583 __field( void *, ocqe )
584 ),
585
586 TP_fast_assign(
587 __entry->ctx = ctx;
588 __entry->user_data = user_data;
589 __entry->res = res;
590 __entry->cflags = cflags;
591 __entry->ocqe = ocqe;
592 ),
593
594 TP_printk("ring %p, user_data 0x%llx, res %d, cflags 0x%x, "
595 "overflow_cqe %p",
596 __entry->ctx, __entry->user_data, __entry->res,
597 __entry->cflags, __entry->ocqe)
598 );
599
600 /*
601 * io_uring_task_work_run - ran task work
602 *
603 * @tctx: pointer to a io_uring_task
604 * @count: how many functions it ran
605 *
606 */
607 TRACE_EVENT(io_uring_task_work_run,
608
609 TP_PROTO(void *tctx, unsigned int count),
610
611 TP_ARGS(tctx, count),
612
613 TP_STRUCT__entry (
614 __field( void *, tctx )
615 __field( unsigned int, count )
616 ),
617
618 TP_fast_assign(
619 __entry->tctx = tctx;
620 __entry->count = count;
621 ),
622
623 TP_printk("tctx %p, count %u", __entry->tctx, __entry->count)
624 );
625
626 TRACE_EVENT(io_uring_short_write,
627
628 TP_PROTO(void *ctx, u64 fpos, u64 wanted, u64 got),
629
630 TP_ARGS(ctx, fpos, wanted, got),
631
632 TP_STRUCT__entry(
633 __field(void *, ctx)
634 __field(u64, fpos)
635 __field(u64, wanted)
636 __field(u64, got)
637 ),
638
639 TP_fast_assign(
640 __entry->ctx = ctx;
641 __entry->fpos = fpos;
642 __entry->wanted = wanted;
643 __entry->got = got;
644 ),
645
646 TP_printk("ring %p, fpos %lld, wanted %lld, got %lld",
647 __entry->ctx, __entry->fpos,
648 __entry->wanted, __entry->got)
649 );
650
651 /*
652 * io_uring_local_work_run - ran ring local task work
653 *
654 * @tctx: pointer to a io_uring_ctx
655 * @count: how many functions it ran
656 * @loops: how many loops it ran
657 *
658 */
659 TRACE_EVENT(io_uring_local_work_run,
660
661 TP_PROTO(void *ctx, int count, unsigned int loops),
662
663 TP_ARGS(ctx, count, loops),
664
665 TP_STRUCT__entry (
666 __field(void *, ctx )
667 __field(int, count )
668 __field(unsigned int, loops )
669 ),
670
671 TP_fast_assign(
672 __entry->ctx = ctx;
673 __entry->count = count;
674 __entry->loops = loops;
675 ),
676
677 TP_printk("ring %p, count %d, loops %u", __entry->ctx, __entry->count, __entry->loops)
678 );
679
680 #endif /* _TRACE_IO_URING_H */
681
682 /* This part must be outside protection */
683 #include <trace/define_trace.h>
684
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.