ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev_iouring.c
(Generate patch)

Comparing libev/ev_iouring.c (file contents):
Revision 1.5 by root, Tue Nov 19 13:09:00 2019 UTC vs.
Revision 1.21 by root, Wed Jan 22 02:20:47 2020 UTC

1/* 1/*
2 * libev linux io_uring fd activity backend 2 * libev linux io_uring fd activity backend
3 * 3 *
4 * Copyright (c) 2019 Marc Alexander Lehmann <libev@schmorp.de> 4 * Copyright (c) 2019-2020 Marc Alexander Lehmann <libev@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
… …
44 * b) best is not necessarily very good. 44 * b) best is not necessarily very good.
45 * c) it's better than the aio mess, doesn't suffer from the fork problems 45 * c) it's better than the aio mess, doesn't suffer from the fork problems
46 * of linux aio or epoll and so on and so on. and you could do event stuff 46 * of linux aio or epoll and so on and so on. and you could do event stuff
47 * without any syscalls. what's not to like? 47 * without any syscalls. what's not to like?
48 * d) ok, it's vastly more complex, but that's ok, really. 48 * d) ok, it's vastly more complex, but that's ok, really.
49 * e) why 3 mmaps instead of one? one would be more space-efficient, 49 * e) why two mmaps instead of one? one would be more space-efficient,
50 * and I can't see what benefit three would have (other than being 50 * and I can't see what benefit two would have (other than being
51 * somehow resizable/relocatable, but that's apparently not possible). 51 * somehow resizable/relocatable, but that's apparently not possible).
52 * f) hmm, it's practiclaly undebuggable (gdb can't access the memory, and 52 * f) hmm, it's practically undebuggable (gdb can't access the memory, and
53 the bizarre way structure offsets are commuinicated makes it hard to 53 * the bizarre way structure offsets are communicated makes it hard to
54 * just print the ring buffer heads, even *iff* the memory were visible 54 * just print the ring buffer heads, even *iff* the memory were visible
55 * in gdb. but then, that's also ok, really. 55 * in gdb. but then, that's also ok, really.
56 * g) well, you cannot specify a timeout when waiting for events. no, 56 * g) well, you cannot specify a timeout when waiting for events. no,
57 * seriously, the interface doesn't support a timeout. never seen _that_ 57 * seriously, the interface doesn't support a timeout. never seen _that_
58 * before. sure, you can use a timerfd, but that's another syscall 58 * before. sure, you can use a timerfd, but that's another syscall
59 * you could have avoided. overall, this bizarre omission smells 59 * you could have avoided. overall, this bizarre omission smells
60 * like a µ-optimisation by the io_uring author for his personal 60 * like a µ-optimisation by the io_uring author for his personal
61 * applications, to the detriment of everybody else who just wants 61 * applications, to the detriment of everybody else who just wants
62 * an event loop. but, umm, ok, if that's all, it could be worse. 62 * an event loop. but, umm, ok, if that's all, it could be worse.
63 * (from what I gather from the author Jens Axboe, it simply didn't
64 * occur to him, and he made good on it by adding an unlimited nuber
65 * of timeouts later :).
63 * h) there is a hardcoded limit of 4096 outstanding events. okay, 66 * h) initially there was a hardcoded limit of 4096 outstanding events.
64 * at least there is no arbitrary low system-wide limit... 67 * later versions not only bump this to 32k, but also can handle
68 * an unlimited amount of events, so this only affects the batch size.
65 * i) unlike linux aio, you *can* register more then the limit 69 * i) unlike linux aio, you *can* register more then the limit
66 * of fd events, and the kernel will "gracefully" signal an 70 * of fd events. while early verisons of io_uring signalled an overflow
67 * overflow, after which you could destroy and recreate the kernel 71 * and you ended up getting wet. 5.5+ does not do this anymore.
68 * state, a bit bigger, or fall back to e.g. poll. thats not
69 * totally insane, but kind of questions the point a high
70 * performance I/O framework when it doesn't really work
71 * under stress.
72 * j) but, oh my! is has exactly the same bugs as the linux aio backend, 72 * j) but, oh my! it had exactly the same bugs as the linux aio backend,
73 * where some undocumented poll combinations just fail. 73 * where some undocumented poll combinations just fail. fortunately,
74 * so we need epoll AGAIN as a fallback. AGAIN! epoll!! and of course, 74 * after finally reaching the author, he was more than willing to fix
75 * this is completely undocumented, have I mantioned this already? 75 * this probably in 5.6+.
76 * k) overall, the *API* itself is, I dare to say, not a total trainwreck. 76 * k) overall, the *API* itself is, I dare to say, not a total trainwreck.
77 * the big isuess with it are the bugs requiring epoll, which might 77 * once the bugs ae fixed (probably in 5.6+), it will be without
78 * or might not get fixed (do I hold my breath?). 78 * competition.
79 */ 79 */
80
81/* TODO: use internal TIMEOUT */
82/* TODO: take advantage of single mmap, NODROP etc. */
83/* TODO: resize cq/sq size independently */
80 84
81#include <sys/timerfd.h> 85#include <sys/timerfd.h>
82#include <sys/mman.h> 86#include <sys/mman.h>
83#include <poll.h> 87#include <poll.h>
88#include <stdint.h>
84 89
85#define IOURING_INIT_ENTRIES 32 90#define IOURING_INIT_ENTRIES 32
86 91
87/*****************************************************************************/ 92/*****************************************************************************/
88/* syscall wrapdadoop - this section has the raw api/abi definitions */ 93/* syscall wrapdadoop - this section has the raw api/abi definitions */
… …
96{ 101{
97 __u8 opcode; 102 __u8 opcode;
98 __u8 flags; 103 __u8 flags;
99 __u16 ioprio; 104 __u16 ioprio;
100 __s32 fd; 105 __s32 fd;
106 union {
101 __u64 off; 107 __u64 off;
108 __u64 addr2;
109 };
102 __u64 addr; 110 __u64 addr;
103 __u32 len; 111 __u32 len;
104 union { 112 union {
105 __kernel_rwf_t rw_flags; 113 __kernel_rwf_t rw_flags;
106 __u32 fsync_flags; 114 __u32 fsync_flags;
107 __u16 poll_events; 115 __u16 poll_events;
108 __u32 sync_range_flags; 116 __u32 sync_range_flags;
109 __u32 msg_flags; 117 __u32 msg_flags;
118 __u32 timeout_flags;
119 __u32 accept_flags;
120 __u32 cancel_flags;
121 __u32 open_flags;
122 __u32 statx_flags;
110 }; 123 };
111 __u64 user_data; 124 __u64 user_data;
112 union { 125 union {
113 __u16 buf_index; 126 __u16 buf_index;
114 __u64 __pad2[3]; 127 __u64 __pad2[3];
… …
151 __u32 sq_entries; 164 __u32 sq_entries;
152 __u32 cq_entries; 165 __u32 cq_entries;
153 __u32 flags; 166 __u32 flags;
154 __u32 sq_thread_cpu; 167 __u32 sq_thread_cpu;
155 __u32 sq_thread_idle; 168 __u32 sq_thread_idle;
169 __u32 features;
156 __u32 resv[5]; 170 __u32 resv[4];
157 struct io_sqring_offsets sq_off; 171 struct io_sqring_offsets sq_off;
158 struct io_cqring_offsets cq_off; 172 struct io_cqring_offsets cq_off;
159}; 173};
160 174
175#define IORING_SETUP_CQSIZE 0x00000008
176
161#define IORING_OP_POLL_ADD 6 177#define IORING_OP_POLL_ADD 6
162#define IORING_OP_POLL_REMOVE 7 178#define IORING_OP_POLL_REMOVE 7
179#define IORING_OP_TIMEOUT 11
180#define IORING_OP_TIMEOUT_REMOVE 12
181
182/* relative or absolute, reference clock is CLOCK_MONOTONIC */
183struct iouring_kernel_timespec
184{
185 int64_t tv_sec;
186 long long tv_nsec;
187};
188
189#define IORING_TIMEOUT_ABS 0x00000001
163 190
164#define IORING_ENTER_GETEVENTS 0x01 191#define IORING_ENTER_GETEVENTS 0x01
165 192
166#define IORING_OFF_SQ_RING 0x00000000ULL 193#define IORING_OFF_SQ_RING 0x00000000ULL
167#define IORING_OFF_CQ_RING 0x08000000ULL 194#define IORING_OFF_CQ_RING 0x08000000ULL
168#define IORING_OFF_SQES 0x10000000ULL 195#define IORING_OFF_SQES 0x10000000ULL
169 196
197#define IORING_FEAT_SINGLE_MMAP 0x00000001
198#define IORING_FEAT_NODROP 0x00000002
199#define IORING_FEAT_SUBMIT_STABLE 0x00000004
200
170inline_size 201inline_size
171int 202int
172evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) 203evsys_io_uring_setup (unsigned entries, struct io_uring_params *params)
173{ 204{
174 return ev_syscall2 (SYS_io_uring_setup, entries, params); 205 return ev_syscall2 (SYS_io_uring_setup, entries, params);
… …
193 224
194/* the submit/completion queue entries */ 225/* the submit/completion queue entries */
195#define EV_SQES ((struct io_uring_sqe *) iouring_sqes) 226#define EV_SQES ((struct io_uring_sqe *) iouring_sqes)
196#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_cq_ring + iouring_cq_cqes)) 227#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_cq_ring + iouring_cq_cqes))
197 228
229inline_speed
230int
231iouring_enter (EV_P_ ev_tstamp timeout)
232{
233 int res;
234
235 EV_RELEASE_CB;
236
237 res = evsys_io_uring_enter (iouring_fd, iouring_to_submit, 1,
238 timeout > EV_TS_CONST (0.) ? IORING_ENTER_GETEVENTS : 0, 0, 0);
239
240 assert (("libev: io_uring_enter did not consume all sqes", (res < 0 || res == iouring_to_submit)));
241
242 iouring_to_submit = 0;
243
244 EV_ACQUIRE_CB;
245
246 return res;
247}
248
249/* TODO: can we move things around so we don't need this forward-reference? */
250static void
251iouring_poll (EV_P_ ev_tstamp timeout);
252
198static 253static
199struct io_uring_sqe * 254struct io_uring_sqe *
200iouring_sqe_get (EV_P) 255iouring_sqe_get (EV_P)
201{ 256{
257 unsigned tail;
258
259 for (;;)
260 {
202 unsigned tail = EV_SQ_VAR (tail); 261 tail = EV_SQ_VAR (tail);
203 262
204 if (tail + 1 - EV_SQ_VAR (head) > EV_SQ_VAR (ring_entries)) 263 if (ecb_expect_true (tail + 1 - EV_SQ_VAR (head) <= EV_SQ_VAR (ring_entries)))
264 break; /* whats the problem, we have free sqes */
265
266 /* queue full, need to flush and possibly handle some events */
267
268#if EV_FEATURE_CODE
269 /* first we ask the kernel nicely, most often this frees up some sqes */
270 int res = iouring_enter (EV_A_ EV_TS_CONST (0.));
271
272 ECB_MEMORY_FENCE_ACQUIRE; /* better safe than sorry */
273
274 if (res >= 0)
275 continue; /* yes, it worked, try again */
276#endif
277
278 /* some problem, possibly EBUSY - do the full poll and let it handle any issues */
279
280 iouring_poll (EV_A_ EV_TS_CONST (0.));
281 /* iouring_poll should have done ECB_MEMORY_FENCE_ACQUIRE for us */
205 { 282 }
206 /* queue full, flush */
207 evsys_io_uring_enter (iouring_fd, iouring_to_submit, 0, 0, 0, 0);
208 iouring_to_submit = 0;
209 }
210 283
211 assert (("libev: io_uring queue full after flush", tail + 1 - EV_SQ_VAR (head) <= EV_SQ_VAR (ring_entries))); 284 /*assert (("libev: io_uring queue full after flush", tail + 1 - EV_SQ_VAR (head) <= EV_SQ_VAR (ring_entries)));*/
212 285
213 return EV_SQES + (tail & EV_SQ_VAR (ring_mask)); 286 return EV_SQES + (tail & EV_SQ_VAR (ring_mask));
214} 287}
215 288
216inline_size 289inline_size
… …
236iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) 309iouring_tfd_cb (EV_P_ struct ev_io *w, int revents)
237{ 310{
238 iouring_tfd_to = EV_TSTAMP_HUGE; 311 iouring_tfd_to = EV_TSTAMP_HUGE;
239} 312}
240 313
241static void
242iouring_epoll_cb (EV_P_ struct ev_io *w, int revents)
243{
244 epoll_poll (EV_A_ 0);
245}
246
247/* called for full and partial cleanup */ 314/* called for full and partial cleanup */
248ecb_cold 315ecb_cold
249static int 316static int
250iouring_internal_destroy (EV_P) 317iouring_internal_destroy (EV_P)
251{ 318{
… …
254 321
255 if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size); 322 if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size);
256 if (iouring_cq_ring != MAP_FAILED) munmap (iouring_cq_ring, iouring_cq_ring_size); 323 if (iouring_cq_ring != MAP_FAILED) munmap (iouring_cq_ring, iouring_cq_ring_size);
257 if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size ); 324 if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size );
258 325
259 if (ev_is_active (&iouring_epoll_w)) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_epoll_w); 326 if (ev_is_active (&iouring_tfd_w))
260 if (ev_is_active (&iouring_tfd_w )) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_tfd_w ); 327 {
328 ev_ref (EV_A);
329 ev_io_stop (EV_A_ &iouring_tfd_w);
330 }
261} 331}
262 332
263ecb_cold 333ecb_cold
264static int 334static int
265iouring_internal_init (EV_P) 335iouring_internal_init (EV_P)
… …
271 iouring_tfd = -1; 341 iouring_tfd = -1;
272 iouring_sq_ring = MAP_FAILED; 342 iouring_sq_ring = MAP_FAILED;
273 iouring_cq_ring = MAP_FAILED; 343 iouring_cq_ring = MAP_FAILED;
274 iouring_sqes = MAP_FAILED; 344 iouring_sqes = MAP_FAILED;
275 345
346 if (!have_monotonic) /* cannot really happen, but what if11 */
347 return -1;
348
276 for (;;) 349 for (;;)
277 { 350 {
278 iouring_fd = evsys_io_uring_setup (iouring_entries, &params); 351 iouring_fd = evsys_io_uring_setup (iouring_entries, &params);
279 352
280 if (iouring_fd >= 0) 353 if (iouring_fd >= 0)
281 break; /* yippie */ 354 break; /* yippie */
282 355
283 if (errno != EINVAL) 356 if (errno != EINVAL)
284 return -1; /* we failed */ 357 return -1; /* we failed */
358
359#if TODO
360 if ((~params.features) & (IORING_FEAT_NODROP | IORING_FEATURE_SINGLE_MMAP | IORING_FEAT_SUBMIT_STABLE))
361 return -1; /* we require the above features */
362#endif
285 363
286 /* EINVAL: lots of possible reasons, but maybe 364 /* EINVAL: lots of possible reasons, but maybe
287 * it is because we hit the unqueryable hardcoded size limit 365 * it is because we hit the unqueryable hardcoded size limit
288 */ 366 */
289 367
… …
342 iouring_internal_destroy (EV_A); 420 iouring_internal_destroy (EV_A);
343 421
344 while (iouring_internal_init (EV_A) < 0) 422 while (iouring_internal_init (EV_A) < 0)
345 ev_syserr ("(libev) io_uring_setup"); 423 ev_syserr ("(libev) io_uring_setup");
346 424
347 /* forking epoll should also effectively unregister all fds from the backend */
348 epoll_fork (EV_A);
349 /* epoll_fork already did this. hopefully */
350 /*fd_rearm_all (EV_A);*/ 425 fd_rearm_all (EV_A);
351
352 ev_io_stop (EV_A_ &iouring_epoll_w);
353 ev_io_set (EV_A_ &iouring_epoll_w, backend_fd, EV_READ);
354 ev_io_start (EV_A_ &iouring_epoll_w);
355 426
356 ev_io_stop (EV_A_ &iouring_tfd_w); 427 ev_io_stop (EV_A_ &iouring_tfd_w);
357 ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); 428 ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ);
358 ev_io_start (EV_A_ &iouring_tfd_w); 429 ev_io_start (EV_A_ &iouring_tfd_w);
359} 430}
… …
361/*****************************************************************************/ 432/*****************************************************************************/
362 433
363static void 434static void
364iouring_modify (EV_P_ int fd, int oev, int nev) 435iouring_modify (EV_P_ int fd, int oev, int nev)
365{ 436{
366 if (ecb_expect_false (anfds [fd].eflags))
367 {
368 /* we handed this fd over to epoll, so undo this first */
369 /* we do it manually because the optimisations on epoll_modify won't do us any good */
370 epoll_ctl (iouring_fd, EPOLL_CTL_DEL, fd, 0);
371 anfds [fd].eflags = 0;
372 oev = 0;
373 }
374
375 if (oev) 437 if (oev)
376 { 438 {
377 /* we assume the sqe's are all "properly" initialised */ 439 /* we assume the sqe's are all "properly" initialised */
378 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); 440 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A);
379 sqe->opcode = IORING_OP_POLL_REMOVE; 441 sqe->opcode = IORING_OP_POLL_REMOVE;
380 sqe->fd = fd; 442 sqe->fd = fd;
443 /* Jens Axboe notified me that user_data is not what is documented, but is
444 * some kind of unique ID that has to match, otherwise the request cannot
445 * be removed. Since we don't *really* have that, we pass in the old
446 * generation counter - if that fails, too bad, it will hopefully be removed
447 * at close time and then be ignored. */
448 sqe->addr = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32);
381 sqe->user_data = -1; 449 sqe->user_data = (uint64_t)-1;
382 iouring_sqe_submit (EV_A_ sqe); 450 iouring_sqe_submit (EV_A_ sqe);
383 451
384 /* increment generation counter to avoid handling old events */ 452 /* increment generation counter to avoid handling old events */
385 ++anfds [fd].egen; 453 ++anfds [fd].egen;
386 } 454 }
… …
388 if (nev) 456 if (nev)
389 { 457 {
390 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); 458 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A);
391 sqe->opcode = IORING_OP_POLL_ADD; 459 sqe->opcode = IORING_OP_POLL_ADD;
392 sqe->fd = fd; 460 sqe->fd = fd;
461 sqe->addr = 0;
393 sqe->user_data = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32); 462 sqe->user_data = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32);
394 sqe->poll_events = 463 sqe->poll_events =
395 (nev & EV_READ ? POLLIN : 0) 464 (nev & EV_READ ? POLLIN : 0)
396 | (nev & EV_WRITE ? POLLOUT : 0); 465 | (nev & EV_WRITE ? POLLOUT : 0);
397 iouring_sqe_submit (EV_A_ sqe); 466 iouring_sqe_submit (EV_A_ sqe);
… …
427{ 496{
428 int fd = cqe->user_data & 0xffffffffU; 497 int fd = cqe->user_data & 0xffffffffU;
429 uint32_t gen = cqe->user_data >> 32; 498 uint32_t gen = cqe->user_data >> 32;
430 int res = cqe->res; 499 int res = cqe->res;
431 500
432 /* ignore fd removal events, if there are any. TODO: verify */ 501 /* user_data -1 is a remove that we are not atm. interested in */
433 if (cqe->user_data == (__u64)-1) 502 if (cqe->user_data == (uint64_t)-1)
434 abort ();//D 503 return;
435 504
436 assert (("libev: io_uring fd must be in-bounds", fd >= 0 && fd < anfdmax)); 505 assert (("libev: io_uring fd must be in-bounds", fd >= 0 && fd < anfdmax));
437 506
438 /* documentation lies, of course. the result value is NOT like 507 /* documentation lies, of course. the result value is NOT like
439 * normal syscalls, but like linux raw syscalls, i.e. negative 508 * normal syscalls, but like linux raw syscalls, i.e. negative
440 * error numbers. fortunate, as otherwise there would be no way 509 * error numbers. fortunate, as otherwise there would be no way
441 * to get error codes at all. still, why not document this? 510 * to get error codes at all. still, why not document this?
442 */ 511 */
443 512
444 /* ignore event if generation doesn't match */ 513 /* ignore event if generation doesn't match */
514 /* other than skipping removal events, */
445 /* this should actually be very rare */ 515 /* this should actually be very rare */
446 if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) 516 if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen))
447 return; 517 return;
448 518
449 if (ecb_expect_false (res < 0)) 519 if (ecb_expect_false (res < 0))
450 { 520 {
451 if (res == -EINVAL) 521 /*TODO: EINVAL handling (was something failed with this fd)*/
452 {
453 /* we assume this error code means the fd/poll combination is buggy
454 * and fall back to epoll.
455 * this error code might also indicate a bug, but the kernel doesn't
456 * distinguish between those two conditions, so... sigh...
457 */
458 522
459 epoll_modify (EV_A_ fd, 0, anfds [fd].events);
460 }
461 else if (res == -EBADF) 523 if (res == -EBADF)
462 { 524 {
463 assert (("libev: event loop rejected bad fd", res != -EBADF)); 525 assert (("libev: event loop rejected bad fd", res != -EBADF));
464 fd_kill (EV_A_ fd); 526 fd_kill (EV_A_ fd);
465 } 527 }
466 else 528 else
… …
492iouring_overflow (EV_P) 554iouring_overflow (EV_P)
493{ 555{
494 /* we have two options, resize the queue (by tearing down 556 /* we have two options, resize the queue (by tearing down
495 * everything and recreating it, or living with it 557 * everything and recreating it, or living with it
496 * and polling. 558 * and polling.
497 * we implement this by resizing tghe queue, and, if that fails, 559 * we implement this by resizing the queue, and, if that fails,
498 * we just recreate the state on every failure, which 560 * we just recreate the state on every failure, which
499 * kind of is a very inefficient poll. 561 * kind of is a very inefficient poll.
500 * one danger is, due to the bios toward lower fds, 562 * one danger is, due to the bios toward lower fds,
501 * we will only really get events for those, so 563 * we will only really get events for those, so
502 * maybe we need a poll() fallback, after all. 564 * maybe we need a poll() fallback, after all.
… …
514 else 576 else
515 { 577 {
516 /* we hit the kernel limit, we should fall back to something else. 578 /* we hit the kernel limit, we should fall back to something else.
517 * we can either poll() a few times and hope for the best, 579 * we can either poll() a few times and hope for the best,
518 * poll always, or switch to epoll. 580 * poll always, or switch to epoll.
519 * since we use epoll anyways, go epoll. 581 * TODO: is this necessary with newer kernels?
520 */ 582 */
521 583
522 iouring_internal_destroy (EV_A); 584 iouring_internal_destroy (EV_A);
523 585
524 /* this should make it so that on return, we don'T call any uring functions */ 586 /* this should make it so that on return, we don't call any uring functions */
525 iouring_to_submit = 0; 587 iouring_to_submit = 0;
526 588
527 for (;;) 589 for (;;)
528 { 590 {
529 backend = epoll_init (EV_A_ 0); 591 backend = epoll_init (EV_A_ 0);
… …
570 632
571static void 633static void
572iouring_poll (EV_P_ ev_tstamp timeout) 634iouring_poll (EV_P_ ev_tstamp timeout)
573{ 635{
574 /* if we have events, no need for extra syscalls, but we might have to queue events */ 636 /* if we have events, no need for extra syscalls, but we might have to queue events */
637 /* we also clar the timeout if there are outstanding fdchanges */
638 /* the latter should only happen if both the sq and cq are full, most likely */
639 /* because we have a lot of event sources that immediately complete */
640 /* TODO: fdchacngecnt is always 0 because fd_reify does not have two buffers yet */
575 if (iouring_handle_cq (EV_A)) 641 if (iouring_handle_cq (EV_A) || fdchangecnt)
576 timeout = EV_TS_CONST (0.); 642 timeout = EV_TS_CONST (0.);
577 else 643 else
578 /* no events, so maybe wait for some */ 644 /* no events, so maybe wait for some */
579 iouring_tfd_update (EV_A_ timeout); 645 iouring_tfd_update (EV_A_ timeout);
580 646
581 /* only enter the kernel if we have somethign to submit, or we need to wait */ 647 /* only enter the kernel if we have something to submit, or we need to wait */
582 if (timeout || iouring_to_submit) 648 if (timeout || iouring_to_submit)
583 { 649 {
584 int res; 650 int res = iouring_enter (EV_A_ timeout);
585
586 EV_RELEASE_CB;
587
588 res = evsys_io_uring_enter (iouring_fd, iouring_to_submit, 1,
589 timeout > EV_TS_CONST (0.) ? IORING_ENTER_GETEVENTS : 0, 0, 0);
590 iouring_to_submit = 0;
591
592 EV_ACQUIRE_CB;
593 651
594 if (ecb_expect_false (res < 0)) 652 if (ecb_expect_false (res < 0))
595 if (errno == EINTR) 653 if (errno == EINTR)
596 /* ignore */; 654 /* ignore */;
655 else if (errno == EBUSY)
656 /* cq full, cannot submit - should be rare because we flush the cq first, so simply ignore */;
597 else 657 else
598 ev_syserr ("(libev) iouring setup"); 658 ev_syserr ("(libev) iouring setup");
599 else 659 else
600 iouring_handle_cq (EV_A); 660 iouring_handle_cq (EV_A);
601 } 661 }
… …
603 663
604inline_size 664inline_size
605int 665int
606iouring_init (EV_P_ int flags) 666iouring_init (EV_P_ int flags)
607{ 667{
608 if (!epoll_init (EV_A_ 0))
609 return 0;
610
611 ev_io_init (&iouring_epoll_w, iouring_epoll_cb, backend_fd, EV_READ);
612 ev_set_priority (&iouring_epoll_w, EV_MAXPRI);
613
614 ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ);
615 ev_set_priority (&iouring_tfd_w, EV_MAXPRI);
616
617 iouring_entries = IOURING_INIT_ENTRIES; 668 iouring_entries = IOURING_INIT_ENTRIES;
618 iouring_max_entries = 0; 669 iouring_max_entries = 0;
619 670
620 if (iouring_internal_init (EV_A) < 0) 671 if (iouring_internal_init (EV_A) < 0)
621 { 672 {
622 iouring_internal_destroy (EV_A); 673 iouring_internal_destroy (EV_A);
623 return 0; 674 return 0;
624 } 675 }
625 676
626 ev_io_start (EV_A_ &iouring_epoll_w); 677 ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ);
627 ev_unref (EV_A); /* watcher should not keep loop alive */ 678 ev_set_priority (&iouring_tfd_w, EV_MINPRI);
628
629 ev_io_start (EV_A_ &iouring_tfd_w); 679 ev_io_start (EV_A_ &iouring_tfd_w);
630 ev_unref (EV_A); /* watcher should not keep loop alive */ 680 ev_unref (EV_A); /* watcher should not keep loop alive */
631 681
632 backend_modify = iouring_modify; 682 backend_modify = iouring_modify;
633 backend_poll = iouring_poll; 683 backend_poll = iouring_poll;
… …
638inline_size 688inline_size
639void 689void
640iouring_destroy (EV_P) 690iouring_destroy (EV_P)
641{ 691{
642 iouring_internal_destroy (EV_A); 692 iouring_internal_destroy (EV_A);
643 epoll_destroy (EV_A);
644} 693}
645 694

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines