xref: /openbmc/linux/include/trace/events/io_uring.h (revision 8dda2eac)
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 
10 struct io_wq_work;
11 
12 /**
13  * io_uring_create - called after a new io_uring context was prepared
14  *
15  * @fd:		corresponding file descriptor
16  * @ctx:	pointer to a ring context structure
17  * @sq_entries:	actual SQ size
18  * @cq_entries:	actual CQ size
19  * @flags:	SQ ring flags, provided to io_uring_setup(2)
20  *
21  * Allows to trace io_uring creation and provide pointer to a context, that can
22  * be used later to find correlated events.
23  */
24 TRACE_EVENT(io_uring_create,
25 
26 	TP_PROTO(int fd, void *ctx, u32 sq_entries, u32 cq_entries, u32 flags),
27 
28 	TP_ARGS(fd, ctx, sq_entries, cq_entries, flags),
29 
30 	TP_STRUCT__entry (
31 		__field(  int,		fd			)
32 		__field(  void *,	ctx			)
33 		__field(  u32,		sq_entries	)
34 		__field(  u32,		cq_entries	)
35 		__field(  u32,		flags		)
36 	),
37 
38 	TP_fast_assign(
39 		__entry->fd			= fd;
40 		__entry->ctx		= ctx;
41 		__entry->sq_entries	= sq_entries;
42 		__entry->cq_entries	= cq_entries;
43 		__entry->flags		= flags;
44 	),
45 
46 	TP_printk("ring %p, fd %d sq size %d, cq size %d, flags %d",
47 			  __entry->ctx, __entry->fd, __entry->sq_entries,
48 			  __entry->cq_entries, __entry->flags)
49 );
50 
51 /**
52  * io_uring_register - called after a buffer/file/eventfd was successfully
53  * 					   registered for a ring
54  *
55  * @ctx:		pointer to a ring context structure
56  * @opcode:		describes which operation to perform
57  * @nr_user_files:	number of registered files
58  * @nr_user_bufs:	number of registered buffers
59  * @cq_ev_fd:		whether eventfs registered or not
60  * @ret:		return code
61  *
62  * Allows to trace fixed files/buffers/eventfds, 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, bool eventfd, long ret),
71 
72 	TP_ARGS(ctx, opcode, nr_files, nr_bufs, eventfd, 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(  bool,		eventfd		)
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->eventfd	= eventfd;
89 		__entry->ret		= ret;
90 	),
91 
92 	TP_printk("ring %p, opcode %d, nr_user_files %d, nr_user_bufs %d, "
93 			  "eventfd %d, ret %ld",
94 			  __entry->ctx, __entry->opcode, __entry->nr_files,
95 			  __entry->nr_bufs, __entry->eventfd, __entry->ret)
96 );
97 
98 /**
99  * io_uring_file_get - called before getting references to an SQE file
100  *
101  * @ctx:	pointer to a ring context structure
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, int fd),
111 
112 	TP_ARGS(ctx, fd),
113 
114 	TP_STRUCT__entry (
115 		__field(  void *,	ctx	)
116 		__field(  int,		fd	)
117 	),
118 
119 	TP_fast_assign(
120 		__entry->ctx	= ctx;
121 		__entry->fd		= fd;
122 	),
123 
124 	TP_printk("ring %p, fd %d", __entry->ctx, __entry->fd)
125 );
126 
127 /**
128  * io_uring_queue_async_work - called before submitting a new async work
129  *
130  * @ctx:	pointer to a ring context structure
131  * @hashed:	type of workqueue, hashed or normal
132  * @req:	pointer to a submitted request
133  * @work:	pointer to a submitted io_wq_work
134  *
135  * Allows to trace asynchronous work submission.
136  */
137 TRACE_EVENT(io_uring_queue_async_work,
138 
139 	TP_PROTO(void *ctx, int rw, void * req, struct io_wq_work *work,
140 			 unsigned int flags),
141 
142 	TP_ARGS(ctx, rw, req, work, flags),
143 
144 	TP_STRUCT__entry (
145 		__field(  void *,			ctx	)
146 		__field(  int,				rw	)
147 		__field(  void *,			req	)
148 		__field(  struct io_wq_work *,		work	)
149 		__field(  unsigned int,			flags	)
150 	),
151 
152 	TP_fast_assign(
153 		__entry->ctx	= ctx;
154 		__entry->rw	= rw;
155 		__entry->req	= req;
156 		__entry->work	= work;
157 		__entry->flags	= flags;
158 	),
159 
160 	TP_printk("ring %p, request %p, flags %d, %s queue, work %p",
161 			  __entry->ctx, __entry->req, __entry->flags,
162 			  __entry->rw ? "hashed" : "normal", __entry->work)
163 );
164 
165 /**
166  * io_uring_defer - called when an io_uring request is deferred
167  *
168  * @ctx:	pointer to a ring context structure
169  * @req:	pointer to a deferred request
170  * @user_data:	user data associated with the request
171  *
172  * Allows to track deferred requests, to get an insight about what requests are
173  * not started immediately.
174  */
175 TRACE_EVENT(io_uring_defer,
176 
177 	TP_PROTO(void *ctx, void *req, unsigned long long user_data),
178 
179 	TP_ARGS(ctx, req, user_data),
180 
181 	TP_STRUCT__entry (
182 		__field(  void *,	ctx		)
183 		__field(  void *,	req		)
184 		__field(  unsigned long long, data	)
185 	),
186 
187 	TP_fast_assign(
188 		__entry->ctx	= ctx;
189 		__entry->req	= req;
190 		__entry->data	= user_data;
191 	),
192 
193 	TP_printk("ring %p, request %p user_data %llu", __entry->ctx,
194 			__entry->req, __entry->data)
195 );
196 
197 /**
198  * io_uring_link - called before the io_uring request added into link_list of
199  * 		   another request
200  *
201  * @ctx:		pointer to a ring context structure
202  * @req:		pointer to a linked request
203  * @target_req:		pointer to a previous request, that would contain @req
204  *
205  * Allows to track linked requests, to understand dependencies between requests
206  * and how does it influence their execution flow.
207  */
208 TRACE_EVENT(io_uring_link,
209 
210 	TP_PROTO(void *ctx, void *req, void *target_req),
211 
212 	TP_ARGS(ctx, req, target_req),
213 
214 	TP_STRUCT__entry (
215 		__field(  void *,	ctx		)
216 		__field(  void *,	req		)
217 		__field(  void *,	target_req	)
218 	),
219 
220 	TP_fast_assign(
221 		__entry->ctx		= ctx;
222 		__entry->req		= req;
223 		__entry->target_req	= target_req;
224 	),
225 
226 	TP_printk("ring %p, request %p linked after %p",
227 			  __entry->ctx, __entry->req, __entry->target_req)
228 );
229 
230 /**
231  * io_uring_cqring_wait - called before start waiting for an available CQE
232  *
233  * @ctx:		pointer to a ring context structure
234  * @min_events:	minimal number of events to wait for
235  *
236  * Allows to track waiting for CQE, so that we can e.g. troubleshoot
237  * situations, when an application wants to wait for an event, that never
238  * comes.
239  */
240 TRACE_EVENT(io_uring_cqring_wait,
241 
242 	TP_PROTO(void *ctx, int min_events),
243 
244 	TP_ARGS(ctx, min_events),
245 
246 	TP_STRUCT__entry (
247 		__field(  void *,	ctx		)
248 		__field(  int,		min_events	)
249 	),
250 
251 	TP_fast_assign(
252 		__entry->ctx	= ctx;
253 		__entry->min_events	= min_events;
254 	),
255 
256 	TP_printk("ring %p, min_events %d", __entry->ctx, __entry->min_events)
257 );
258 
259 /**
260  * io_uring_fail_link - called before failing a linked request
261  *
262  * @req:	request, which links were cancelled
263  * @link:	cancelled link
264  *
265  * Allows to track linked requests cancellation, to see not only that some work
266  * was cancelled, but also which request was the reason.
267  */
268 TRACE_EVENT(io_uring_fail_link,
269 
270 	TP_PROTO(void *req, void *link),
271 
272 	TP_ARGS(req, link),
273 
274 	TP_STRUCT__entry (
275 		__field(  void *,	req	)
276 		__field(  void *,	link	)
277 	),
278 
279 	TP_fast_assign(
280 		__entry->req	= req;
281 		__entry->link	= link;
282 	),
283 
284 	TP_printk("request %p, link %p", __entry->req, __entry->link)
285 );
286 
287 /**
288  * io_uring_complete - called when completing an SQE
289  *
290  * @ctx:		pointer to a ring context structure
291  * @user_data:		user data associated with the request
292  * @res:		result of the request
293  * @cflags:		completion flags
294  *
295  */
296 TRACE_EVENT(io_uring_complete,
297 
298 	TP_PROTO(void *ctx, u64 user_data, long res, unsigned cflags),
299 
300 	TP_ARGS(ctx, user_data, res, cflags),
301 
302 	TP_STRUCT__entry (
303 		__field(  void *,	ctx		)
304 		__field(  u64,		user_data	)
305 		__field(  long,		res		)
306 		__field(  unsigned,	cflags		)
307 	),
308 
309 	TP_fast_assign(
310 		__entry->ctx		= ctx;
311 		__entry->user_data	= user_data;
312 		__entry->res		= res;
313 		__entry->cflags		= cflags;
314 	),
315 
316 	TP_printk("ring %p, user_data 0x%llx, result %ld, cflags %x",
317 			  __entry->ctx, (unsigned long long)__entry->user_data,
318 			  __entry->res, __entry->cflags)
319 );
320 
321 /**
322  * io_uring_submit_sqe - called before submitting one SQE
323  *
324  * @ctx:		pointer to a ring context structure
325  * @req:		pointer to a submitted request
326  * @opcode:		opcode of request
327  * @user_data:		user data associated with the request
328  * @flags		request flags
329  * @force_nonblock:	whether a context blocking or not
330  * @sq_thread:		true if sq_thread has submitted this SQE
331  *
332  * Allows to track SQE submitting, to understand what was the source of it, SQ
333  * thread or io_uring_enter call.
334  */
335 TRACE_EVENT(io_uring_submit_sqe,
336 
337 	TP_PROTO(void *ctx, void *req, u8 opcode, u64 user_data, u32 flags,
338 		 bool force_nonblock, bool sq_thread),
339 
340 	TP_ARGS(ctx, req, opcode, user_data, flags, force_nonblock, sq_thread),
341 
342 	TP_STRUCT__entry (
343 		__field(  void *,	ctx		)
344 		__field(  void *,	req		)
345 		__field(  u8,		opcode		)
346 		__field(  u64,		user_data	)
347 		__field(  u32,		flags		)
348 		__field(  bool,		force_nonblock	)
349 		__field(  bool,		sq_thread	)
350 	),
351 
352 	TP_fast_assign(
353 		__entry->ctx		= ctx;
354 		__entry->req		= req;
355 		__entry->opcode		= opcode;
356 		__entry->user_data	= user_data;
357 		__entry->flags		= flags;
358 		__entry->force_nonblock	= force_nonblock;
359 		__entry->sq_thread	= sq_thread;
360 	),
361 
362 	TP_printk("ring %p, req %p, op %d, data 0x%llx, flags %u, "
363 		  "non block %d, sq_thread %d", __entry->ctx, __entry->req,
364 		  __entry->opcode, (unsigned long long)__entry->user_data,
365 		  __entry->flags, __entry->force_nonblock, __entry->sq_thread)
366 );
367 
368 /*
369  * io_uring_poll_arm - called after arming a poll wait if successful
370  *
371  * @ctx:		pointer to a ring context structure
372  * @req:		pointer to the armed request
373  * @opcode:		opcode of request
374  * @user_data:		user data associated with the request
375  * @mask:		request poll events mask
376  * @events:		registered events of interest
377  *
378  * Allows to track which fds are waiting for and what are the events of
379  * interest.
380  */
381 TRACE_EVENT(io_uring_poll_arm,
382 
383 	TP_PROTO(void *ctx, void *req, u8 opcode, u64 user_data,
384 		 int mask, int events),
385 
386 	TP_ARGS(ctx, req, opcode, user_data, mask, events),
387 
388 	TP_STRUCT__entry (
389 		__field(  void *,	ctx		)
390 		__field(  void *,	req		)
391 		__field(  u8,		opcode		)
392 		__field(  u64,		user_data	)
393 		__field(  int,		mask		)
394 		__field(  int,		events		)
395 	),
396 
397 	TP_fast_assign(
398 		__entry->ctx		= ctx;
399 		__entry->req		= req;
400 		__entry->opcode		= opcode;
401 		__entry->user_data	= user_data;
402 		__entry->mask		= mask;
403 		__entry->events		= events;
404 	),
405 
406 	TP_printk("ring %p, req %p, op %d, data 0x%llx, mask 0x%x, events 0x%x",
407 		  __entry->ctx, __entry->req, __entry->opcode,
408 		  (unsigned long long) __entry->user_data,
409 		  __entry->mask, __entry->events)
410 );
411 
412 TRACE_EVENT(io_uring_poll_wake,
413 
414 	TP_PROTO(void *ctx, u8 opcode, u64 user_data, int mask),
415 
416 	TP_ARGS(ctx, opcode, user_data, mask),
417 
418 	TP_STRUCT__entry (
419 		__field(  void *,	ctx		)
420 		__field(  u8,		opcode		)
421 		__field(  u64,		user_data	)
422 		__field(  int,		mask		)
423 	),
424 
425 	TP_fast_assign(
426 		__entry->ctx		= ctx;
427 		__entry->opcode		= opcode;
428 		__entry->user_data	= user_data;
429 		__entry->mask		= mask;
430 	),
431 
432 	TP_printk("ring %p, op %d, data 0x%llx, mask 0x%x",
433 			  __entry->ctx, __entry->opcode,
434 			  (unsigned long long) __entry->user_data,
435 			  __entry->mask)
436 );
437 
438 TRACE_EVENT(io_uring_task_add,
439 
440 	TP_PROTO(void *ctx, u8 opcode, u64 user_data, int mask),
441 
442 	TP_ARGS(ctx, opcode, user_data, mask),
443 
444 	TP_STRUCT__entry (
445 		__field(  void *,	ctx		)
446 		__field(  u8,		opcode		)
447 		__field(  u64,		user_data	)
448 		__field(  int,		mask		)
449 	),
450 
451 	TP_fast_assign(
452 		__entry->ctx		= ctx;
453 		__entry->opcode		= opcode;
454 		__entry->user_data	= user_data;
455 		__entry->mask		= mask;
456 	),
457 
458 	TP_printk("ring %p, op %d, data 0x%llx, mask %x",
459 			  __entry->ctx, __entry->opcode,
460 			  (unsigned long long) __entry->user_data,
461 			  __entry->mask)
462 );
463 
464 /*
465  * io_uring_task_run - called when task_work_run() executes the poll events
466  *                     notification callbacks
467  *
468  * @ctx:		pointer to a ring context structure
469  * @req:		pointer to the armed request
470  * @opcode:		opcode of request
471  * @user_data:		user data associated with the request
472  *
473  * Allows to track when notified poll events are processed
474  */
475 TRACE_EVENT(io_uring_task_run,
476 
477 	TP_PROTO(void *ctx, void *req, u8 opcode, u64 user_data),
478 
479 	TP_ARGS(ctx, req, opcode, user_data),
480 
481 	TP_STRUCT__entry (
482 		__field(  void *,	ctx		)
483 		__field(  void *,	req		)
484 		__field(  u8,		opcode		)
485 		__field(  u64,		user_data	)
486 	),
487 
488 	TP_fast_assign(
489 		__entry->ctx		= ctx;
490 		__entry->req		= req;
491 		__entry->opcode		= opcode;
492 		__entry->user_data	= user_data;
493 	),
494 
495 	TP_printk("ring %p, req %p, op %d, data 0x%llx",
496 		  __entry->ctx, __entry->req, __entry->opcode,
497 		  (unsigned long long) __entry->user_data)
498 );
499 
500 #endif /* _TRACE_IO_URING_H */
501 
502 /* This part must be outside protection */
503 #include <trace/define_trace.h>
504