… | |
… | |
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 */ |
|
|
187 | struct 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 |
|
|
200 | |
|
|
201 | #define IORING_FEAT_SINGLE_MMAP 0x00000001 |
|
|
202 | #define IORING_FEAT_NODROP 0x00000002 |
|
|
203 | #define IORING_FEAT_SUBMIT_STABLE 0x00000004 |
173 | |
204 | |
174 | inline_size |
205 | inline_size |
175 | int |
206 | int |
176 | evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) |
207 | evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) |
177 | { |
208 | { |
… | |
… | |
240 | iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) |
271 | iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) |
241 | { |
272 | { |
242 | iouring_tfd_to = EV_TSTAMP_HUGE; |
273 | iouring_tfd_to = EV_TSTAMP_HUGE; |
243 | } |
274 | } |
244 | |
275 | |
245 | static void |
|
|
246 | iouring_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 */ |
276 | /* called for full and partial cleanup */ |
252 | ecb_cold |
277 | ecb_cold |
253 | static int |
278 | static int |
254 | iouring_internal_destroy (EV_P) |
279 | iouring_internal_destroy (EV_P) |
255 | { |
280 | { |
… | |
… | |
258 | |
283 | |
259 | if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size); |
284 | 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); |
285 | 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 ); |
286 | if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size ); |
262 | |
287 | |
263 | if (ev_is_active (&iouring_epoll_w)) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_epoll_w); |
288 | 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 ); |
289 | { |
|
|
290 | ev_ref (EV_A); |
|
|
291 | ev_io_stop (EV_A_ &iouring_tfd_w); |
|
|
292 | } |
265 | } |
293 | } |
266 | |
294 | |
267 | ecb_cold |
295 | ecb_cold |
268 | static int |
296 | static int |
269 | iouring_internal_init (EV_P) |
297 | iouring_internal_init (EV_P) |
… | |
… | |
275 | iouring_tfd = -1; |
303 | iouring_tfd = -1; |
276 | iouring_sq_ring = MAP_FAILED; |
304 | iouring_sq_ring = MAP_FAILED; |
277 | iouring_cq_ring = MAP_FAILED; |
305 | iouring_cq_ring = MAP_FAILED; |
278 | iouring_sqes = MAP_FAILED; |
306 | iouring_sqes = MAP_FAILED; |
279 | |
307 | |
|
|
308 | if (!have_monotonic) /* cannot really happen, but what if11 */ |
|
|
309 | return -1; |
|
|
310 | |
280 | for (;;) |
311 | for (;;) |
281 | { |
312 | { |
282 | iouring_fd = evsys_io_uring_setup (iouring_entries, ¶ms); |
313 | iouring_fd = evsys_io_uring_setup (iouring_entries, ¶ms); |
283 | |
314 | |
284 | if (iouring_fd >= 0) |
315 | if (iouring_fd >= 0) |
285 | break; /* yippie */ |
316 | break; /* yippie */ |
286 | |
317 | |
287 | if (errno != EINVAL) |
318 | if (errno != EINVAL) |
288 | return -1; /* we failed */ |
319 | return -1; /* we failed */ |
|
|
320 | |
|
|
321 | #if TODO |
|
|
322 | if ((~params.features) & (IORING_FEAT_NODROP | IORING_FEATURE_SINGLE_MMAP)) |
|
|
323 | return -1; /* we require the above features */ |
|
|
324 | #endif |
289 | |
325 | |
290 | /* EINVAL: lots of possible reasons, but maybe |
326 | /* EINVAL: lots of possible reasons, but maybe |
291 | * it is because we hit the unqueryable hardcoded size limit |
327 | * it is because we hit the unqueryable hardcoded size limit |
292 | */ |
328 | */ |
293 | |
329 | |
… | |
… | |
346 | iouring_internal_destroy (EV_A); |
382 | iouring_internal_destroy (EV_A); |
347 | |
383 | |
348 | while (iouring_internal_init (EV_A) < 0) |
384 | while (iouring_internal_init (EV_A) < 0) |
349 | ev_syserr ("(libev) io_uring_setup"); |
385 | ev_syserr ("(libev) io_uring_setup"); |
350 | |
386 | |
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);*/ |
387 | 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 | |
388 | |
360 | ev_io_stop (EV_A_ &iouring_tfd_w); |
389 | ev_io_stop (EV_A_ &iouring_tfd_w); |
361 | ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); |
390 | ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); |
362 | ev_io_start (EV_A_ &iouring_tfd_w); |
391 | ev_io_start (EV_A_ &iouring_tfd_w); |
363 | } |
392 | } |
… | |
… | |
365 | /*****************************************************************************/ |
394 | /*****************************************************************************/ |
366 | |
395 | |
367 | static void |
396 | static void |
368 | iouring_modify (EV_P_ int fd, int oev, int nev) |
397 | iouring_modify (EV_P_ int fd, int oev, int nev) |
369 | { |
398 | { |
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) |
399 | if (oev) |
380 | { |
400 | { |
381 | /* we assume the sqe's are all "properly" initialised */ |
401 | /* we assume the sqe's are all "properly" initialised */ |
382 | struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); |
402 | struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); |
383 | sqe->opcode = IORING_OP_POLL_REMOVE; |
403 | sqe->opcode = IORING_OP_POLL_REMOVE; |
384 | sqe->fd = fd; |
404 | sqe->fd = fd; |
385 | sqe->user_data = -1; |
405 | /* Jens Axboe notified me that user_data is not what is documented, but is |
|
|
406 | * some kind of unique ID that has to match, otherwise the request cannot |
|
|
407 | * be removed. Since we don't *really* have that, we pass in the old |
|
|
408 | * generation counter - if that fails, too bad, it will hopefully be removed |
|
|
409 | * at close time and then be ignored. */ |
|
|
410 | sqe->user_data = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32); |
386 | iouring_sqe_submit (EV_A_ sqe); |
411 | iouring_sqe_submit (EV_A_ sqe); |
387 | |
412 | |
388 | /* increment generation counter to avoid handling old events */ |
413 | /* increment generation counter to avoid handling old events */ |
389 | ++anfds [fd].egen; |
414 | ++anfds [fd].egen; |
390 | } |
415 | } |
… | |
… | |
431 | { |
456 | { |
432 | int fd = cqe->user_data & 0xffffffffU; |
457 | int fd = cqe->user_data & 0xffffffffU; |
433 | uint32_t gen = cqe->user_data >> 32; |
458 | uint32_t gen = cqe->user_data >> 32; |
434 | int res = cqe->res; |
459 | int res = cqe->res; |
435 | |
460 | |
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)); |
461 | assert (("libev: io_uring fd must be in-bounds", fd >= 0 && fd < anfdmax)); |
441 | |
462 | |
442 | /* documentation lies, of course. the result value is NOT like |
463 | /* documentation lies, of course. the result value is NOT like |
443 | * normal syscalls, but like linux raw syscalls, i.e. negative |
464 | * normal syscalls, but like linux raw syscalls, i.e. negative |
444 | * error numbers. fortunate, as otherwise there would be no way |
465 | * error numbers. fortunate, as otherwise there would be no way |
445 | * to get error codes at all. still, why not document this? |
466 | * to get error codes at all. still, why not document this? |
446 | */ |
467 | */ |
447 | |
468 | |
448 | /* ignore event if generation doesn't match */ |
469 | /* ignore event if generation doesn't match */ |
|
|
470 | /* other than skipping removal events, */ |
449 | /* this should actually be very rare */ |
471 | /* this should actually be very rare */ |
450 | if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) |
472 | if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) |
451 | return; |
473 | return; |
452 | |
474 | |
453 | if (ecb_expect_false (res < 0)) |
475 | if (ecb_expect_false (res < 0)) |
454 | { |
476 | { |
455 | if (res == -EINVAL) |
477 | /*TODO: EINVAL handling (was something failed with this fd)*/ |
456 | { |
478 | /*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 | |
479 | |
463 | epoll_modify (EV_A_ fd, 0, anfds [fd].events); |
|
|
464 | } |
|
|
465 | else if (res == -EBADF) |
480 | if (res == -EBADF) |
466 | { |
481 | { |
467 | assert (("libev: event loop rejected bad fd", res != -EBADF)); |
482 | assert (("libev: event loop rejected bad fd", res != -EBADF)); |
468 | fd_kill (EV_A_ fd); |
483 | fd_kill (EV_A_ fd); |
469 | } |
484 | } |
470 | else |
485 | else |
… | |
… | |
496 | iouring_overflow (EV_P) |
511 | iouring_overflow (EV_P) |
497 | { |
512 | { |
498 | /* we have two options, resize the queue (by tearing down |
513 | /* we have two options, resize the queue (by tearing down |
499 | * everything and recreating it, or living with it |
514 | * everything and recreating it, or living with it |
500 | * and polling. |
515 | * and polling. |
501 | * we implement this by resizing tghe queue, and, if that fails, |
516 | * we implement this by resizing the queue, and, if that fails, |
502 | * we just recreate the state on every failure, which |
517 | * we just recreate the state on every failure, which |
503 | * kind of is a very inefficient poll. |
518 | * kind of is a very inefficient poll. |
504 | * one danger is, due to the bios toward lower fds, |
519 | * one danger is, due to the bios toward lower fds, |
505 | * we will only really get events for those, so |
520 | * we will only really get events for those, so |
506 | * maybe we need a poll() fallback, after all. |
521 | * maybe we need a poll() fallback, after all. |
… | |
… | |
518 | else |
533 | else |
519 | { |
534 | { |
520 | /* we hit the kernel limit, we should fall back to something else. |
535 | /* 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, |
536 | * we can either poll() a few times and hope for the best, |
522 | * poll always, or switch to epoll. |
537 | * poll always, or switch to epoll. |
523 | * since we use epoll anyways, go epoll. |
538 | * TODO: is this necessary with newer kernels? |
524 | */ |
539 | */ |
525 | |
540 | |
526 | iouring_internal_destroy (EV_A); |
541 | iouring_internal_destroy (EV_A); |
527 | |
542 | |
528 | /* this should make it so that on return, we don'T call any uring functions */ |
543 | /* this should make it so that on return, we don't call any uring functions */ |
529 | iouring_to_submit = 0; |
544 | iouring_to_submit = 0; |
530 | |
545 | |
531 | for (;;) |
546 | for (;;) |
532 | { |
547 | { |
533 | backend = epoll_init (EV_A_ 0); |
548 | backend = epoll_init (EV_A_ 0); |
… | |
… | |
607 | |
622 | |
608 | inline_size |
623 | inline_size |
609 | int |
624 | int |
610 | iouring_init (EV_P_ int flags) |
625 | iouring_init (EV_P_ int flags) |
611 | { |
626 | { |
612 | if (!epoll_init (EV_A_ 0)) |
|
|
613 | return 0; |
|
|
614 | |
|
|
615 | iouring_entries = IOURING_INIT_ENTRIES; |
627 | iouring_entries = IOURING_INIT_ENTRIES; |
616 | iouring_max_entries = 0; |
628 | iouring_max_entries = 0; |
617 | |
629 | |
618 | if (iouring_internal_init (EV_A) < 0) |
630 | if (iouring_internal_init (EV_A) < 0) |
619 | { |
631 | { |
620 | iouring_internal_destroy (EV_A); |
632 | iouring_internal_destroy (EV_A); |
621 | return 0; |
633 | return 0; |
622 | } |
634 | } |
623 | |
635 | |
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); |
636 | ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ); |
628 | ev_set_priority (&iouring_tfd_w, EV_MAXPRI); |
637 | 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); |
638 | ev_io_start (EV_A_ &iouring_tfd_w); |
634 | ev_unref (EV_A); /* watcher should not keep loop alive */ |
639 | ev_unref (EV_A); /* watcher should not keep loop alive */ |
635 | |
640 | |
636 | backend_modify = iouring_modify; |
641 | backend_modify = iouring_modify; |
637 | backend_poll = iouring_poll; |
642 | backend_poll = iouring_poll; |
… | |
… | |
642 | inline_size |
647 | inline_size |
643 | void |
648 | void |
644 | iouring_destroy (EV_P) |
649 | iouring_destroy (EV_P) |
645 | { |
650 | { |
646 | iouring_internal_destroy (EV_A); |
651 | iouring_internal_destroy (EV_A); |
647 | epoll_destroy (EV_A); |
|
|
648 | } |
652 | } |
649 | |
653 | |