xref: /openbmc/linux/include/trace/events/io_uring.h (revision 0a94608f)
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 
11 struct io_wq_work;
12 
13 /**
14  * io_uring_create - called after a new io_uring context was prepared
15  *
16  * @fd:		corresponding file descriptor
17  * @ctx:	pointer to a ring context structure
18  * @sq_entries:	actual SQ size
19  * @cq_entries:	actual CQ size
20  * @flags:	SQ ring flags, provided to io_uring_setup(2)
21  *
22  * Allows to trace io_uring creation and provide pointer to a context, that can
23  * be used later to find correlated events.
24  */
25 TRACE_EVENT(io_uring_create,
26 
27 	TP_PROTO(int fd, void *ctx, u32 sq_entries, u32 cq_entries, u32 flags),
28 
29 	TP_ARGS(fd, ctx, sq_entries, cq_entries, flags),
30 
31 	TP_STRUCT__entry (
32 		__field(  int,		fd		)
33 		__field(  void *,	ctx		)
34 		__field(  u32,		sq_entries	)
35 		__field(  u32,		cq_entries	)
36 		__field(  u32,		flags		)
37 	),
38 
39 	TP_fast_assign(
40 		__entry->fd		= fd;
41 		__entry->ctx		= ctx;
42 		__entry->sq_entries	= sq_entries;
43 		__entry->cq_entries	= cq_entries;
44 		__entry->flags		= flags;
45 	),
46 
47 	TP_printk("ring %p, fd %d sq size %d, cq size %d, flags 0x%x",
48 			  __entry->ctx, __entry->fd, __entry->sq_entries,
49 			  __entry->cq_entries, __entry->flags)
50 );
51 
52 /**
53  * io_uring_register - called after a buffer/file/eventfd was successfully
54  * 					   registered for a ring
55  *
56  * @ctx:		pointer to a ring context structure
57  * @opcode:		describes which operation to perform
58  * @nr_user_files:	number of registered files
59  * @nr_user_bufs:	number of registered buffers
60  * @ret:		return code
61  *
62  * Allows to trace fixed files/buffers, that could be registered to
63  * avoid an overhead of getting references to them for every operation. This
64  * event, together with io_uring_file_get, can provide a full picture of how
65  * much overhead one can reduce via fixing.
66  */
67 TRACE_EVENT(io_uring_register,
68 
69 	TP_PROTO(void *ctx, unsigned opcode, unsigned nr_files,
70 			 unsigned nr_bufs, long ret),
71 
72 	TP_ARGS(ctx, opcode, nr_files, nr_bufs, ret),
73 
74 	TP_STRUCT__entry (
75 		__field(  void *,	ctx	)
76 		__field(  unsigned,	opcode	)
77 		__field(  unsigned,	nr_files)
78 		__field(  unsigned,	nr_bufs	)
79 		__field(  long,		ret	)
80 	),
81 
82 	TP_fast_assign(
83 		__entry->ctx		= ctx;
84 		__entry->opcode		= opcode;
85 		__entry->nr_files	= nr_files;
86 		__entry->nr_bufs	= nr_bufs;
87 		__entry->ret		= ret;
88 	),
89 
90 	TP_printk("ring %p, opcode %d, nr_user_files %d, nr_user_bufs %d, "
91 			  "ret %ld",
92 			  __entry->ctx, __entry->opcode, __entry->nr_files,
93 			  __entry->nr_bufs, __entry->ret)
94 );
95 
96 /**
97  * io_uring_file_get - called before getting references to an SQE file
98  *
99  * @ctx:	pointer to a ring context structure
100  * @req:	pointer to a submitted request
101  * @user_data:	user data associated with the 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(void *ctx, void *req, unsigned long long user_data, int fd),
111 
112 	TP_ARGS(ctx, req, user_data, 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		= ctx;
123 		__entry->req		= req;
124 		__entry->user_data	= 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  * @ctx:	pointer to a ring context structure
136  * @req:	pointer to a submitted request
137  * @user_data:	user data associated with the request
138  * @opcode:	opcode of request
139  * @flags	request flags
140  * @work:	pointer to a submitted io_wq_work
141  * @rw:		type of workqueue, hashed or normal
142  *
143  * Allows to trace asynchronous work submission.
144  */
145 TRACE_EVENT(io_uring_queue_async_work,
146 
147 	TP_PROTO(void *ctx, void * req, unsigned long long user_data, u8 opcode,
148 		unsigned int flags, struct io_wq_work *work, int rw),
149 
150 	TP_ARGS(ctx, req, user_data, flags, opcode, work, rw),
151 
152 	TP_STRUCT__entry (
153 		__field(  void *,			ctx		)
154 		__field(  void *,			req		)
155 		__field(  u64,				user_data	)
156 		__field(  u8,				opcode		)
157 		__field(  unsigned int,			flags		)
158 		__field(  struct io_wq_work *,		work		)
159 		__field(  int,				rw		)
160 	),
161 
162 	TP_fast_assign(
163 		__entry->ctx		= ctx;
164 		__entry->req		= req;
165 		__entry->user_data	= user_data;
166 		__entry->flags		= flags;
167 		__entry->opcode		= opcode;
168 		__entry->work		= work;
169 		__entry->rw		= rw;
170 	),
171 
172 	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %d, flags 0x%x, %s queue, work %p",
173 		__entry->ctx, __entry->req, __entry->user_data, __entry->opcode,
174 		__entry->flags, __entry->rw ? "hashed" : "normal", __entry->work)
175 );
176 
177 /**
178  * io_uring_defer - called when an io_uring request is deferred
179  *
180  * @ctx:	pointer to a ring context structure
181  * @req:	pointer to a deferred request
182  * @user_data:	user data associated with the request
183  * @opcode:	opcode of request
184  *
185  * Allows to track deferred requests, to get an insight about what requests are
186  * not started immediately.
187  */
188 TRACE_EVENT(io_uring_defer,
189 
190 	TP_PROTO(void *ctx, void *req, unsigned long long user_data, u8 opcode),
191 
192 	TP_ARGS(ctx, req, user_data, opcode),
193 
194 	TP_STRUCT__entry (
195 		__field(  void *,		ctx	)
196 		__field(  void *,		req	)
197 		__field(  unsigned long long,	data	)
198 		__field(  u8,			opcode	)
199 	),
200 
201 	TP_fast_assign(
202 		__entry->ctx	= ctx;
203 		__entry->req	= req;
204 		__entry->data	= user_data;
205 		__entry->opcode	= opcode;
206 	),
207 
208 	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %d",
209 		__entry->ctx, __entry->req, __entry->data, __entry->opcode)
210 );
211 
212 /**
213  * io_uring_link - called before the io_uring request added into link_list of
214  * 		   another request
215  *
216  * @ctx:		pointer to a ring context structure
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(void *ctx, void *req, void *target_req),
226 
227 	TP_ARGS(ctx, 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		= 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  * @ctx:	pointer to a ring context structure
278  * @req:	request, which links were cancelled
279  * @user_data:	user data associated with the request
280  * @opcode:	opcode of request
281  * @link:	cancelled link
282  *
283  * Allows to track linked requests cancellation, to see not only that some work
284  * was cancelled, but also which request was the reason.
285  */
286 TRACE_EVENT(io_uring_fail_link,
287 
288 	TP_PROTO(void *ctx, void *req, unsigned long long user_data, u8 opcode, void *link),
289 
290 	TP_ARGS(ctx, req, user_data, opcode, link),
291 
292 	TP_STRUCT__entry (
293 		__field(  void *,		ctx		)
294 		__field(  void *,		req		)
295 		__field(  unsigned long long,	user_data	)
296 		__field(  u8,			opcode		)
297 		__field(  void *,		link		)
298 	),
299 
300 	TP_fast_assign(
301 		__entry->ctx		= ctx;
302 		__entry->req		= req;
303 		__entry->user_data	= user_data;
304 		__entry->opcode		= opcode;
305 		__entry->link		= link;
306 	),
307 
308 	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %d, link %p",
309 		__entry->ctx, __entry->req, __entry->user_data, __entry->opcode,
310 		__entry->link)
311 );
312 
313 /**
314  * io_uring_complete - called when completing an SQE
315  *
316  * @ctx:		pointer to a ring context structure
317  * @req:		pointer to a submitted request
318  * @user_data:		user data associated with the request
319  * @res:		result of the request
320  * @cflags:		completion flags
321  *
322  */
323 TRACE_EVENT(io_uring_complete,
324 
325 	TP_PROTO(void *ctx, void *req, u64 user_data, int res, unsigned cflags),
326 
327 	TP_ARGS(ctx, req, user_data, res, cflags),
328 
329 	TP_STRUCT__entry (
330 		__field(  void *,	ctx		)
331 		__field(  void *,	req		)
332 		__field(  u64,		user_data	)
333 		__field(  int,		res		)
334 		__field(  unsigned,	cflags		)
335 	),
336 
337 	TP_fast_assign(
338 		__entry->ctx		= ctx;
339 		__entry->req		= req;
340 		__entry->user_data	= user_data;
341 		__entry->res		= res;
342 		__entry->cflags		= cflags;
343 	),
344 
345 	TP_printk("ring %p, req %p, user_data 0x%llx, result %d, cflags 0x%x",
346 		__entry->ctx, __entry->req,
347 		__entry->user_data,
348 		__entry->res, __entry->cflags)
349 );
350 
351 /**
352  * io_uring_submit_sqe - called before submitting one SQE
353  *
354  * @ctx:		pointer to a ring context structure
355  * @req:		pointer to a submitted request
356  * @user_data:		user data associated with the request
357  * @opcode:		opcode of request
358  * @flags		request flags
359  * @force_nonblock:	whether a context blocking or not
360  * @sq_thread:		true if sq_thread has submitted this SQE
361  *
362  * Allows to track SQE submitting, to understand what was the source of it, SQ
363  * thread or io_uring_enter call.
364  */
365 TRACE_EVENT(io_uring_submit_sqe,
366 
367 	TP_PROTO(void *ctx, void *req, unsigned long long user_data, u8 opcode, u32 flags,
368 		 bool force_nonblock, bool sq_thread),
369 
370 	TP_ARGS(ctx, req, user_data, opcode, flags, force_nonblock, sq_thread),
371 
372 	TP_STRUCT__entry (
373 		__field(  void *,		ctx		)
374 		__field(  void *,		req		)
375 		__field(  unsigned long long,	user_data	)
376 		__field(  u8,			opcode		)
377 		__field(  u32,			flags		)
378 		__field(  bool,			force_nonblock	)
379 		__field(  bool,			sq_thread	)
380 	),
381 
382 	TP_fast_assign(
383 		__entry->ctx		= ctx;
384 		__entry->req		= req;
385 		__entry->user_data	= user_data;
386 		__entry->opcode		= opcode;
387 		__entry->flags		= flags;
388 		__entry->force_nonblock	= force_nonblock;
389 		__entry->sq_thread	= sq_thread;
390 	),
391 
392 	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %d, flags 0x%x, "
393 		  "non block %d, sq_thread %d", __entry->ctx, __entry->req,
394 		  __entry->user_data, __entry->opcode,
395 		  __entry->flags, __entry->force_nonblock, __entry->sq_thread)
396 );
397 
398 /*
399  * io_uring_poll_arm - called after arming a poll wait if successful
400  *
401  * @ctx:		pointer to a ring context structure
402  * @req:		pointer to the armed request
403  * @user_data:		user data associated with the request
404  * @opcode:		opcode of request
405  * @mask:		request poll events mask
406  * @events:		registered events of interest
407  *
408  * Allows to track which fds are waiting for and what are the events of
409  * interest.
410  */
411 TRACE_EVENT(io_uring_poll_arm,
412 
413 	TP_PROTO(void *ctx, void *req, u64 user_data, u8 opcode,
414 		 int mask, int events),
415 
416 	TP_ARGS(ctx, req, user_data, opcode, mask, events),
417 
418 	TP_STRUCT__entry (
419 		__field(  void *,		ctx		)
420 		__field(  void *,		req		)
421 		__field(  unsigned long long,	user_data	)
422 		__field(  u8,			opcode		)
423 		__field(  int,			mask		)
424 		__field(  int,			events		)
425 	),
426 
427 	TP_fast_assign(
428 		__entry->ctx		= ctx;
429 		__entry->req		= req;
430 		__entry->user_data	= user_data;
431 		__entry->opcode		= opcode;
432 		__entry->mask		= mask;
433 		__entry->events		= events;
434 	),
435 
436 	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %d, mask 0x%x, events 0x%x",
437 		  __entry->ctx, __entry->req, __entry->user_data, __entry->opcode,
438 		  __entry->mask, __entry->events)
439 );
440 
441 /*
442  * io_uring_task_add - called after adding a task
443  *
444  * @ctx:		pointer to a ring context structure
445  * @req:		pointer to request
446  * @user_data:		user data associated with the request
447  * @opcode:		opcode of request
448  * @mask:		request poll events mask
449  *
450  */
451 TRACE_EVENT(io_uring_task_add,
452 
453 	TP_PROTO(void *ctx, void *req, unsigned long long user_data, u8 opcode, int mask),
454 
455 	TP_ARGS(ctx, req, user_data, opcode, mask),
456 
457 	TP_STRUCT__entry (
458 		__field(  void *,		ctx		)
459 		__field(  void *,		req		)
460 		__field(  unsigned long long,	user_data	)
461 		__field(  u8,			opcode		)
462 		__field(  int,			mask		)
463 	),
464 
465 	TP_fast_assign(
466 		__entry->ctx		= ctx;
467 		__entry->req		= req;
468 		__entry->user_data	= user_data;
469 		__entry->opcode		= opcode;
470 		__entry->mask		= mask;
471 	),
472 
473 	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %d, mask %x",
474 		__entry->ctx, __entry->req, __entry->user_data, __entry->opcode,
475 		__entry->mask)
476 );
477 
478 /*
479  * io_uring_req_failed - called when an sqe is errored dring submission
480  *
481  * @sqe:		pointer to the io_uring_sqe that failed
482  * @ctx:		pointer to a ring context structure
483  * @req:		pointer to request
484  * @error:		error it failed with
485  *
486  * Allows easier diagnosing of malformed requests in production systems.
487  */
488 TRACE_EVENT(io_uring_req_failed,
489 
490 	TP_PROTO(const struct io_uring_sqe *sqe, void *ctx, void *req, int error),
491 
492 	TP_ARGS(sqe, ctx, req, error),
493 
494 	TP_STRUCT__entry (
495 		__field(  void *,		ctx		)
496 		__field(  void *,		req		)
497 		__field(  unsigned long long,	user_data	)
498 		__field(  u8,			opcode		)
499 		__field(  u8,			flags		)
500 		__field(  u8,			ioprio		)
501 		__field( u64,			off		)
502 		__field( u64,			addr		)
503 		__field( u32,			len		)
504 		__field( u32,			op_flags	)
505 		__field( u16,			buf_index	)
506 		__field( u16,			personality	)
507 		__field( u32,			file_index	)
508 		__field( u64,			pad1		)
509 		__field( u64,			pad2		)
510 		__field( int,			error		)
511 	),
512 
513 	TP_fast_assign(
514 		__entry->ctx		= ctx;
515 		__entry->req		= req;
516 		__entry->user_data	= sqe->user_data;
517 		__entry->opcode		= sqe->opcode;
518 		__entry->flags		= sqe->flags;
519 		__entry->ioprio		= sqe->ioprio;
520 		__entry->off		= sqe->off;
521 		__entry->addr		= sqe->addr;
522 		__entry->len		= sqe->len;
523 		__entry->op_flags	= sqe->rw_flags;
524 		__entry->buf_index	= sqe->buf_index;
525 		__entry->personality	= sqe->personality;
526 		__entry->file_index	= sqe->file_index;
527 		__entry->pad1		= sqe->__pad2[0];
528 		__entry->pad2		= sqe->__pad2[1];
529 		__entry->error		= error;
530 	),
531 
532 	TP_printk("ring %p, req %p, user_data 0x%llx, "
533 		"op %d, flags 0x%x, prio=%d, off=%llu, addr=%llu, "
534 		  "len=%u, rw_flags=0x%x, buf_index=%d, "
535 		  "personality=%d, file_index=%d, pad=0x%llx/%llx, error=%d",
536 		  __entry->ctx, __entry->req, __entry->user_data,
537 		  __entry->opcode, __entry->flags, __entry->ioprio,
538 		  (unsigned long long)__entry->off,
539 		  (unsigned long long) __entry->addr, __entry->len,
540 		  __entry->op_flags,
541 		  __entry->buf_index, __entry->personality, __entry->file_index,
542 		  (unsigned long long) __entry->pad1,
543 		  (unsigned long long) __entry->pad2, __entry->error)
544 );
545 
546 #endif /* _TRACE_IO_URING_H */
547 
548 /* This part must be outside protection */
549 #include <trace/define_trace.h>
550