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 1.7 by root, Sun Dec 22 15:05:32 2019 UTC vs.
Revision 1.14 by root, Sat Dec 28 05:20:17 2019 UTC

… …
80 * k) overall, the *API* itself is, I dare to say, not a total trainwreck. 80 * k) overall, the *API* itself is, I dare to say, not a total trainwreck.
81 * the big isuess with it are the bugs requiring epoll, which might 81 * the big isuess with it are the bugs requiring epoll, which might
82 * or might not get fixed (do I hold my breath?). 82 * or might not get fixed (do I hold my breath?).
83 */ 83 */
84 84
85/* TODO: use internal TIMEOUT */
86/* TODO: take advantage of single mmap, NODROP etc. */
87/* TODO: resize cq/sq size independently */
88
85#include <sys/timerfd.h> 89#include <sys/timerfd.h>
86#include <sys/mman.h> 90#include <sys/mman.h>
87#include <poll.h> 91#include <poll.h>
92#include <stdint.h>
88 93
89#define IOURING_INIT_ENTRIES 32 94#define IOURING_INIT_ENTRIES 32
90 95
91/*****************************************************************************/ 96/*****************************************************************************/
92/* syscall wrapdadoop - this section has the raw api/abi definitions */ 97/* syscall wrapdadoop - this section has the raw api/abi definitions */
… …
100{ 105{
101 __u8 opcode; 106 __u8 opcode;
102 __u8 flags; 107 __u8 flags;
103 __u16 ioprio; 108 __u16 ioprio;
104 __s32 fd; 109 __s32 fd;
110 union {
105 __u64 off; 111 __u64 off;
112 __u64 addr2;
113 };
106 __u64 addr; 114 __u64 addr;
107 __u32 len; 115 __u32 len;
108 union { 116 union {
109 __kernel_rwf_t rw_flags; 117 __kernel_rwf_t rw_flags;
110 __u32 fsync_flags; 118 __u32 fsync_flags;
111 __u16 poll_events; 119 __u16 poll_events;
112 __u32 sync_range_flags; 120 __u32 sync_range_flags;
113 __u32 msg_flags; 121 __u32 msg_flags;
122 __u32 timeout_flags;
123 __u32 accept_flags;
124 __u32 cancel_flags;
125 __u32 open_flags;
126 __u32 statx_flags;
114 }; 127 };
115 __u64 user_data; 128 __u64 user_data;
116 union { 129 union {
117 __u16 buf_index; 130 __u16 buf_index;
118 __u64 __pad2[3]; 131 __u64 __pad2[3];
… …
155 __u32 sq_entries; 168 __u32 sq_entries;
156 __u32 cq_entries; 169 __u32 cq_entries;
157 __u32 flags; 170 __u32 flags;
158 __u32 sq_thread_cpu; 171 __u32 sq_thread_cpu;
159 __u32 sq_thread_idle; 172 __u32 sq_thread_idle;
173 __u32 features;
160 __u32 resv[5]; 174 __u32 resv[4];
161 struct io_sqring_offsets sq_off; 175 struct io_sqring_offsets sq_off;
162 struct io_cqring_offsets cq_off; 176 struct io_cqring_offsets cq_off;
163}; 177};
164 178
179#define IORING_SETUP_CQSIZE 0x00000008
180
165#define IORING_OP_POLL_ADD 6 181#define IORING_OP_POLL_ADD 6
166#define IORING_OP_POLL_REMOVE 7 182#define IORING_OP_POLL_REMOVE 7
183#define IORING_OP_TIMEOUT 11
184#define IORING_OP_TIMEOUT_REMOVE 12
185
186/* relative or absolute, reference clock is CLOCK_MONOTONIC */
187struct iouring_kernel_timespec
188{
189 int64_t tv_sec;
190 long long tv_nsec;
191};
192
193#define IORING_TIMEOUT_ABS 0x00000001
167 194
168#define IORING_ENTER_GETEVENTS 0x01 195#define IORING_ENTER_GETEVENTS 0x01
169 196
170#define IORING_OFF_SQ_RING 0x00000000ULL 197#define IORING_OFF_SQ_RING 0x00000000ULL
171#define IORING_OFF_CQ_RING 0x08000000ULL 198#define IORING_OFF_CQ_RING 0x08000000ULL
172#define IORING_OFF_SQES 0x10000000ULL 199#define IORING_OFF_SQES 0x10000000ULL
173 200
201#define IORING_FEAT_SINGLE_MMAP 0x00000001
202#define IORING_FEAT_NODROP 0x00000002
203#define IORING_FEAT_SUBMIT_STABLE 0x00000004
204
174inline_size 205inline_size
175int 206int
176evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) 207evsys_io_uring_setup (unsigned entries, struct io_uring_params *params)
177{ 208{
178 return ev_syscall2 (SYS_io_uring_setup, entries, params); 209 return ev_syscall2 (SYS_io_uring_setup, entries, params);
… …
197 228
198/* the submit/completion queue entries */ 229/* the submit/completion queue entries */
199#define EV_SQES ((struct io_uring_sqe *) iouring_sqes) 230#define EV_SQES ((struct io_uring_sqe *) iouring_sqes)
200#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_cq_ring + iouring_cq_cqes)) 231#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_cq_ring + iouring_cq_cqes))
201 232
233/* TODO: this is not enough, we might have to reap events */
234/* TODO: but we can't, as that will re-arm events, causing */
235/* TODO: an endless loop in fd_reify */
236static int
237iouring_enter (EV_P_ ev_tstamp timeout)
238{
239 int res;
240
241 EV_RELEASE_CB;
242
243 res = evsys_io_uring_enter (iouring_fd, iouring_to_submit, 1,
244 timeout > EV_TS_CONST (0.) ? IORING_ENTER_GETEVENTS : 0, 0, 0);
245
246 assert (("libev: io_uring_enter did not consume all sqes", (res < 0 || res == iouring_to_submit)));
247
248 iouring_to_submit = 0;
249
250 EV_ACQUIRE_CB;
251
252 return res;
253}
254
202static 255static
203struct io_uring_sqe * 256struct io_uring_sqe *
204iouring_sqe_get (EV_P) 257iouring_sqe_get (EV_P)
205{ 258{
206 unsigned tail = EV_SQ_VAR (tail); 259 unsigned tail = EV_SQ_VAR (tail);
207 260
208 if (tail + 1 - EV_SQ_VAR (head) > EV_SQ_VAR (ring_entries)) 261 while (ecb_expect_false (tail + 1 - EV_SQ_VAR (head) > EV_SQ_VAR (ring_entries)))
209 { 262 {
210 /* queue full, flush */ 263 /* queue full, need to flush */
211 evsys_io_uring_enter (iouring_fd, iouring_to_submit, 0, 0, 0, 0);
212 iouring_to_submit = 0;
213 }
214 264
265 int res = iouring_enter (EV_A_ EV_TS_CONST (0.));
266
267 /* io_uring_enter might fail with EBUSY and won't submit anything */
268 /* unfortunately, we can't handle this at the moment */
269
270 if (res < 0 && errno == EBUSY)
271 //TODO
272 ev_syserr ("(libev) io_uring_enter could not clear sq");
273 else
274 break;
275
276 /* iouring_poll should have done ECB_MEMORY_FENCE_ACQUIRE */
277 }
278
215 assert (("libev: io_uring queue full after flush", tail + 1 - EV_SQ_VAR (head) <= EV_SQ_VAR (ring_entries))); 279 /*assert (("libev: io_uring queue full after flush", tail + 1 - EV_SQ_VAR (head) <= EV_SQ_VAR (ring_entries)));*/
216 280
217 return EV_SQES + (tail & EV_SQ_VAR (ring_mask)); 281 return EV_SQES + (tail & EV_SQ_VAR (ring_mask));
218} 282}
219 283
220inline_size 284inline_size
… …
240iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) 304iouring_tfd_cb (EV_P_ struct ev_io *w, int revents)
241{ 305{
242 iouring_tfd_to = EV_TSTAMP_HUGE; 306 iouring_tfd_to = EV_TSTAMP_HUGE;
243} 307}
244 308
245static void
246iouring_epoll_cb (EV_P_ struct ev_io *w, int revents)
247{
248 epoll_poll (EV_A_ 0);
249}
250
251/* called for full and partial cleanup */ 309/* called for full and partial cleanup */
252ecb_cold 310ecb_cold
253static int 311static int
254iouring_internal_destroy (EV_P) 312iouring_internal_destroy (EV_P)
255{ 313{
… …
258 316
259 if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size); 317 if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size);
260 if (iouring_cq_ring != MAP_FAILED) munmap (iouring_cq_ring, iouring_cq_ring_size); 318 if (iouring_cq_ring != MAP_FAILED) munmap (iouring_cq_ring, iouring_cq_ring_size);
261 if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size ); 319 if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size );
262 320
263 if (ev_is_active (&iouring_epoll_w)) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_epoll_w); 321 if (ev_is_active (&iouring_tfd_w))
264 if (ev_is_active (&iouring_tfd_w )) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_tfd_w ); 322 {
323 ev_ref (EV_A);
324 ev_io_stop (EV_A_ &iouring_tfd_w);
325 }
265} 326}
266 327
267ecb_cold 328ecb_cold
268static int 329static int
269iouring_internal_init (EV_P) 330iouring_internal_init (EV_P)
… …
275 iouring_tfd = -1; 336 iouring_tfd = -1;
276 iouring_sq_ring = MAP_FAILED; 337 iouring_sq_ring = MAP_FAILED;
277 iouring_cq_ring = MAP_FAILED; 338 iouring_cq_ring = MAP_FAILED;
278 iouring_sqes = MAP_FAILED; 339 iouring_sqes = MAP_FAILED;
279 340
341 if (!have_monotonic) /* cannot really happen, but what if11 */
342 return -1;
343
280 for (;;) 344 for (;;)
281 { 345 {
282 iouring_fd = evsys_io_uring_setup (iouring_entries, &params); 346 iouring_fd = evsys_io_uring_setup (iouring_entries, &params);
283 347
284 if (iouring_fd >= 0) 348 if (iouring_fd >= 0)
285 break; /* yippie */ 349 break; /* yippie */
286 350
287 if (errno != EINVAL) 351 if (errno != EINVAL)
288 return -1; /* we failed */ 352 return -1; /* we failed */
353
354#if TODO
355 if ((~params.features) & (IORING_FEAT_NODROP | IORING_FEATURE_SINGLE_MMAP))
356 return -1; /* we require the above features */
357#endif
289 358
290 /* EINVAL: lots of possible reasons, but maybe 359 /* EINVAL: lots of possible reasons, but maybe
291 * it is because we hit the unqueryable hardcoded size limit 360 * it is because we hit the unqueryable hardcoded size limit
292 */ 361 */
293 362
… …
346 iouring_internal_destroy (EV_A); 415 iouring_internal_destroy (EV_A);
347 416
348 while (iouring_internal_init (EV_A) < 0) 417 while (iouring_internal_init (EV_A) < 0)
349 ev_syserr ("(libev) io_uring_setup"); 418 ev_syserr ("(libev) io_uring_setup");
350 419
351 /* forking epoll should also effectively unregister all fds from the backend */
352 epoll_fork (EV_A);
353 /* epoll_fork already did this. hopefully */
354 /*fd_rearm_all (EV_A);*/ 420 fd_rearm_all (EV_A);
355
356 ev_io_stop (EV_A_ &iouring_epoll_w);
357 ev_io_set (EV_A_ &iouring_epoll_w, backend_fd, EV_READ);
358 ev_io_start (EV_A_ &iouring_epoll_w);
359 421
360 ev_io_stop (EV_A_ &iouring_tfd_w); 422 ev_io_stop (EV_A_ &iouring_tfd_w);
361 ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); 423 ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ);
362 ev_io_start (EV_A_ &iouring_tfd_w); 424 ev_io_start (EV_A_ &iouring_tfd_w);
363} 425}
… …
365/*****************************************************************************/ 427/*****************************************************************************/
366 428
367static void 429static void
368iouring_modify (EV_P_ int fd, int oev, int nev) 430iouring_modify (EV_P_ int fd, int oev, int nev)
369{ 431{
370 if (ecb_expect_false (anfds [fd].eflags))
371 {
372 /* we handed this fd over to epoll, so undo this first */
373 /* we do it manually because the optimisations on epoll_modify won't do us any good */
374 epoll_ctl (iouring_fd, EPOLL_CTL_DEL, fd, 0);
375 anfds [fd].eflags = 0;
376 oev = 0;
377 }
378
379 if (oev) 432 if (oev)
380 { 433 {
381 /* we assume the sqe's are all "properly" initialised */ 434 /* we assume the sqe's are all "properly" initialised */
382 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); 435 struct io_uring_sqe *sqe = iouring_sqe_get (EV_A);
383 sqe->opcode = IORING_OP_POLL_REMOVE; 436 sqe->opcode = IORING_OP_POLL_REMOVE;
384 sqe->fd = fd; 437 sqe->fd = fd;
385 sqe->user_data = -1; 438 /* Jens Axboe notified me that user_data is not what is documented, but is
439 * some kind of unique ID that has to match, otherwise the request cannot
440 * be removed. Since we don't *really* have that, we pass in the old
441 * generation counter - if that fails, too bad, it will hopefully be removed
442 * at close time and then be ignored. */
443 sqe->user_data = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32);
386 iouring_sqe_submit (EV_A_ sqe); 444 iouring_sqe_submit (EV_A_ sqe);
387 445
388 /* increment generation counter to avoid handling old events */ 446 /* increment generation counter to avoid handling old events */
389 ++anfds [fd].egen; 447 ++anfds [fd].egen;
390 } 448 }
… …
431{ 489{
432 int fd = cqe->user_data & 0xffffffffU; 490 int fd = cqe->user_data & 0xffffffffU;
433 uint32_t gen = cqe->user_data >> 32; 491 uint32_t gen = cqe->user_data >> 32;
434 int res = cqe->res; 492 int res = cqe->res;
435 493
436 /* ignore fd removal events, if there are any. TODO: verify */
437 if (cqe->user_data == (__u64)-1)
438 abort ();//D
439
440 assert (("libev: io_uring fd must be in-bounds", fd >= 0 && fd < anfdmax)); 494 assert (("libev: io_uring fd must be in-bounds", fd >= 0 && fd < anfdmax));
441 495
442 /* documentation lies, of course. the result value is NOT like 496 /* documentation lies, of course. the result value is NOT like
443 * normal syscalls, but like linux raw syscalls, i.e. negative 497 * normal syscalls, but like linux raw syscalls, i.e. negative
444 * error numbers. fortunate, as otherwise there would be no way 498 * error numbers. fortunate, as otherwise there would be no way
445 * to get error codes at all. still, why not document this? 499 * to get error codes at all. still, why not document this?
446 */ 500 */
447 501
448 /* ignore event if generation doesn't match */ 502 /* ignore event if generation doesn't match */
503 /* other than skipping removal events, */
449 /* this should actually be very rare */ 504 /* this should actually be very rare */
450 if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) 505 if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen))
451 return; 506 return;
452 507
453 if (ecb_expect_false (res < 0)) 508 if (ecb_expect_false (res < 0))
454 { 509 {
455 if (res == -EINVAL) 510 /*TODO: EINVAL handling (was something failed with this fd)*/
456 { 511 /*TODO: EBUSY happens when?*/
457 /* we assume this error code means the fd/poll combination is buggy
458 * and fall back to epoll.
459 * this error code might also indicate a bug, but the kernel doesn't
460 * distinguish between those two conditions, so... sigh...
461 */
462 512
463 epoll_modify (EV_A_ fd, 0, anfds [fd].events);
464 }
465 else if (res == -EBADF) 513 if (res == -EBADF)
466 { 514 {
467 assert (("libev: event loop rejected bad fd", res != -EBADF)); 515 assert (("libev: event loop rejected bad fd", res != -EBADF));
468 fd_kill (EV_A_ fd); 516 fd_kill (EV_A_ fd);
469 } 517 }
470 else 518 else
… …
496iouring_overflow (EV_P) 544iouring_overflow (EV_P)
497{ 545{
498 /* we have two options, resize the queue (by tearing down 546 /* we have two options, resize the queue (by tearing down
499 * everything and recreating it, or living with it 547 * everything and recreating it, or living with it
500 * and polling. 548 * and polling.
501 * we implement this by resizing tghe queue, and, if that fails, 549 * we implement this by resizing the queue, and, if that fails,
502 * we just recreate the state on every failure, which 550 * we just recreate the state on every failure, which
503 * kind of is a very inefficient poll. 551 * kind of is a very inefficient poll.
504 * one danger is, due to the bios toward lower fds, 552 * one danger is, due to the bios toward lower fds,
505 * we will only really get events for those, so 553 * we will only really get events for those, so
506 * maybe we need a poll() fallback, after all. 554 * maybe we need a poll() fallback, after all.
… …
518 else 566 else
519 { 567 {
520 /* we hit the kernel limit, we should fall back to something else. 568 /* we hit the kernel limit, we should fall back to something else.
521 * we can either poll() a few times and hope for the best, 569 * we can either poll() a few times and hope for the best,
522 * poll always, or switch to epoll. 570 * poll always, or switch to epoll.
523 * since we use epoll anyways, go epoll. 571 * TODO: is this necessary with newer kernels?
524 */ 572 */
525 573
526 iouring_internal_destroy (EV_A); 574 iouring_internal_destroy (EV_A);
527 575
528 /* this should make it so that on return, we don'T call any uring functions */ 576 /* this should make it so that on return, we don't call any uring functions */
529 iouring_to_submit = 0; 577 iouring_to_submit = 0;
530 578
531 for (;;) 579 for (;;)
532 { 580 {
533 backend = epoll_init (EV_A_ 0); 581 backend = epoll_init (EV_A_ 0);
… …
583 iouring_tfd_update (EV_A_ timeout); 631 iouring_tfd_update (EV_A_ timeout);
584 632
585 /* only enter the kernel if we have something to submit, or we need to wait */ 633 /* only enter the kernel if we have something to submit, or we need to wait */
586 if (timeout || iouring_to_submit) 634 if (timeout || iouring_to_submit)
587 { 635 {
588 int res; 636 int res = iouring_enter (EV_A_ timeout);
589
590 EV_RELEASE_CB;
591
592 res = evsys_io_uring_enter (iouring_fd, iouring_to_submit, 1,
593 timeout > EV_TS_CONST (0.) ? IORING_ENTER_GETEVENTS : 0, 0, 0);
594 iouring_to_submit = 0;
595
596 EV_ACQUIRE_CB;
597 637
598 if (ecb_expect_false (res < 0)) 638 if (ecb_expect_false (res < 0))
599 if (errno == EINTR) 639 if (errno == EINTR)
600 /* ignore */; 640 /* ignore */;
641 else if (errno == EBUSY)
642 /* cq full, cannot submit - should be rare because we flush the cq first, so simply ignore */;
601 else 643 else
602 ev_syserr ("(libev) iouring setup"); 644 ev_syserr ("(libev) iouring setup");
603 else 645 else
604 iouring_handle_cq (EV_A); 646 iouring_handle_cq (EV_A);
605 } 647 }
… …
607 649
608inline_size 650inline_size
609int 651int
610iouring_init (EV_P_ int flags) 652iouring_init (EV_P_ int flags)
611{ 653{
612 if (!epoll_init (EV_A_ 0))
613 return 0;
614
615 iouring_entries = IOURING_INIT_ENTRIES; 654 iouring_entries = IOURING_INIT_ENTRIES;
616 iouring_max_entries = 0; 655 iouring_max_entries = 0;
617 656
618 if (iouring_internal_init (EV_A) < 0) 657 if (iouring_internal_init (EV_A) < 0)
619 { 658 {
620 iouring_internal_destroy (EV_A); 659 iouring_internal_destroy (EV_A);
621 return 0; 660 return 0;
622 } 661 }
623 662
624 ev_io_init (&iouring_epoll_w, iouring_epoll_cb, backend_fd, EV_READ);
625 ev_set_priority (&iouring_epoll_w, EV_MAXPRI);
626
627 ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ); 663 ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ);
628 ev_set_priority (&iouring_tfd_w, EV_MAXPRI); 664 ev_set_priority (&iouring_tfd_w, EV_MINPRI);
629
630 ev_io_start (EV_A_ &iouring_epoll_w);
631 ev_unref (EV_A); /* watcher should not keep loop alive */
632
633 ev_io_start (EV_A_ &iouring_tfd_w); 665 ev_io_start (EV_A_ &iouring_tfd_w);
634 ev_unref (EV_A); /* watcher should not keep loop alive */ 666 ev_unref (EV_A); /* watcher should not keep loop alive */
635 667
636 backend_modify = iouring_modify; 668 backend_modify = iouring_modify;
637 backend_poll = iouring_poll; 669 backend_poll = iouring_poll;
… …
642inline_size 674inline_size
643void 675void
644iouring_destroy (EV_P) 676iouring_destroy (EV_P)
645{ 677{
646 iouring_internal_destroy (EV_A); 678 iouring_internal_destroy (EV_A);
647 epoll_destroy (EV_A);
648} 679}
649 680

Diff Legend

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