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