… | |
… | |
100 | { |
100 | { |
101 | __u8 opcode; |
101 | __u8 opcode; |
102 | __u8 flags; |
102 | __u8 flags; |
103 | __u16 ioprio; |
103 | __u16 ioprio; |
104 | __s32 fd; |
104 | __s32 fd; |
|
|
105 | union { |
105 | __u64 off; |
106 | __u64 off; |
|
|
107 | __u64 addr2; |
|
|
108 | }; |
106 | __u64 addr; |
109 | __u64 addr; |
107 | __u32 len; |
110 | __u32 len; |
108 | union { |
111 | union { |
109 | __kernel_rwf_t rw_flags; |
112 | __kernel_rwf_t rw_flags; |
110 | __u32 fsync_flags; |
113 | __u32 fsync_flags; |
111 | __u16 poll_events; |
114 | __u16 poll_events; |
112 | __u32 sync_range_flags; |
115 | __u32 sync_range_flags; |
113 | __u32 msg_flags; |
116 | __u32 msg_flags; |
|
|
117 | __u32 timeout_flags; |
|
|
118 | __u32 accept_flags; |
|
|
119 | __u32 cancel_flags; |
|
|
120 | __u32 open_flags; |
|
|
121 | __u32 statx_flags; |
114 | }; |
122 | }; |
115 | __u64 user_data; |
123 | __u64 user_data; |
116 | union { |
124 | union { |
117 | __u16 buf_index; |
125 | __u16 buf_index; |
118 | __u64 __pad2[3]; |
126 | __u64 __pad2[3]; |
… | |
… | |
155 | __u32 sq_entries; |
163 | __u32 sq_entries; |
156 | __u32 cq_entries; |
164 | __u32 cq_entries; |
157 | __u32 flags; |
165 | __u32 flags; |
158 | __u32 sq_thread_cpu; |
166 | __u32 sq_thread_cpu; |
159 | __u32 sq_thread_idle; |
167 | __u32 sq_thread_idle; |
|
|
168 | __u32 features; |
160 | __u32 resv[5]; |
169 | __u32 resv[4]; |
161 | struct io_sqring_offsets sq_off; |
170 | struct io_sqring_offsets sq_off; |
162 | struct io_cqring_offsets cq_off; |
171 | struct io_cqring_offsets cq_off; |
163 | }; |
172 | }; |
164 | |
173 | |
165 | #define IORING_OP_POLL_ADD 6 |
174 | #define IORING_OP_POLL_ADD 6 |
… | |
… | |
168 | #define IORING_ENTER_GETEVENTS 0x01 |
177 | #define IORING_ENTER_GETEVENTS 0x01 |
169 | |
178 | |
170 | #define IORING_OFF_SQ_RING 0x00000000ULL |
179 | #define IORING_OFF_SQ_RING 0x00000000ULL |
171 | #define IORING_OFF_CQ_RING 0x08000000ULL |
180 | #define IORING_OFF_CQ_RING 0x08000000ULL |
172 | #define IORING_OFF_SQES 0x10000000ULL |
181 | #define IORING_OFF_SQES 0x10000000ULL |
|
|
182 | |
|
|
183 | #define IORING_FEAT_SINGLE_MMAP 0x1 |
|
|
184 | #define IORING_FEAT_NODROP 0x2 |
|
|
185 | #define IORING_FEAT_SUBMIT_STABLE 0x4 |
173 | |
186 | |
174 | inline_size |
187 | inline_size |
175 | int |
188 | int |
176 | evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) |
189 | evsys_io_uring_setup (unsigned entries, struct io_uring_params *params) |
177 | { |
190 | { |
… | |
… | |
240 | iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) |
253 | iouring_tfd_cb (EV_P_ struct ev_io *w, int revents) |
241 | { |
254 | { |
242 | iouring_tfd_to = EV_TSTAMP_HUGE; |
255 | iouring_tfd_to = EV_TSTAMP_HUGE; |
243 | } |
256 | } |
244 | |
257 | |
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 */ |
258 | /* called for full and partial cleanup */ |
252 | ecb_cold |
259 | ecb_cold |
253 | static int |
260 | static int |
254 | iouring_internal_destroy (EV_P) |
261 | iouring_internal_destroy (EV_P) |
255 | { |
262 | { |
… | |
… | |
258 | |
265 | |
259 | if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size); |
266 | 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); |
267 | 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 ); |
268 | if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size ); |
262 | |
269 | |
263 | if (ev_is_active (&iouring_epoll_w)) ev_ref (EV_A); ev_io_stop (EV_A_ &iouring_epoll_w); |
270 | 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 ); |
271 | { |
|
|
272 | ev_ref (EV_A); |
|
|
273 | ev_io_stop (EV_A_ &iouring_tfd_w); |
|
|
274 | } |
265 | } |
275 | } |
266 | |
276 | |
267 | ecb_cold |
277 | ecb_cold |
268 | static int |
278 | static int |
269 | iouring_internal_init (EV_P) |
279 | iouring_internal_init (EV_P) |
… | |
… | |
346 | iouring_internal_destroy (EV_A); |
356 | iouring_internal_destroy (EV_A); |
347 | |
357 | |
348 | while (iouring_internal_init (EV_A) < 0) |
358 | while (iouring_internal_init (EV_A) < 0) |
349 | ev_syserr ("(libev) io_uring_setup"); |
359 | ev_syserr ("(libev) io_uring_setup"); |
350 | |
360 | |
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);*/ |
361 | 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 | |
362 | |
360 | ev_io_stop (EV_A_ &iouring_tfd_w); |
363 | ev_io_stop (EV_A_ &iouring_tfd_w); |
361 | ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); |
364 | ev_io_set (EV_A_ &iouring_tfd_w, iouring_tfd, EV_READ); |
362 | ev_io_start (EV_A_ &iouring_tfd_w); |
365 | ev_io_start (EV_A_ &iouring_tfd_w); |
363 | } |
366 | } |
… | |
… | |
365 | /*****************************************************************************/ |
368 | /*****************************************************************************/ |
366 | |
369 | |
367 | static void |
370 | static void |
368 | iouring_modify (EV_P_ int fd, int oev, int nev) |
371 | iouring_modify (EV_P_ int fd, int oev, int nev) |
369 | { |
372 | { |
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) |
373 | if (oev) |
380 | { |
374 | { |
381 | /* we assume the sqe's are all "properly" initialised */ |
375 | /* we assume the sqe's are all "properly" initialised */ |
382 | struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); |
376 | struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); |
383 | sqe->opcode = IORING_OP_POLL_REMOVE; |
377 | sqe->opcode = IORING_OP_POLL_REMOVE; |
… | |
… | |
450 | if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) |
444 | if (ecb_expect_false (gen != (uint32_t)anfds [fd].egen)) |
451 | return; |
445 | return; |
452 | |
446 | |
453 | if (ecb_expect_false (res < 0)) |
447 | if (ecb_expect_false (res < 0)) |
454 | { |
448 | { |
455 | if (res == -EINVAL) |
449 | //TODO: EINVAL handling (was something failed with this fd) |
456 | { |
450 | //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 | |
451 | |
463 | epoll_modify (EV_A_ fd, 0, anfds [fd].events); |
|
|
464 | } |
|
|
465 | else if (res == -EBADF) |
452 | if (res == -EBADF) |
466 | { |
453 | { |
467 | assert (("libev: event loop rejected bad fd", res != -EBADF)); |
454 | assert (("libev: event loop rejected bad fd", res != -EBADF)); |
468 | fd_kill (EV_A_ fd); |
455 | fd_kill (EV_A_ fd); |
469 | } |
456 | } |
470 | else |
457 | else |
… | |
… | |
607 | |
594 | |
608 | inline_size |
595 | inline_size |
609 | int |
596 | int |
610 | iouring_init (EV_P_ int flags) |
597 | iouring_init (EV_P_ int flags) |
611 | { |
598 | { |
612 | if (!epoll_init (EV_A_ 0)) |
|
|
613 | return 0; |
|
|
614 | |
|
|
615 | iouring_entries = IOURING_INIT_ENTRIES; |
599 | iouring_entries = IOURING_INIT_ENTRIES; |
616 | iouring_max_entries = 0; |
600 | iouring_max_entries = 0; |
617 | |
601 | |
618 | if (iouring_internal_init (EV_A) < 0) |
602 | if (iouring_internal_init (EV_A) < 0) |
619 | { |
603 | { |
620 | iouring_internal_destroy (EV_A); |
604 | iouring_internal_destroy (EV_A); |
621 | return 0; |
605 | return 0; |
622 | } |
606 | } |
623 | |
607 | |
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); |
608 | ev_io_init (&iouring_tfd_w, iouring_tfd_cb, iouring_tfd, EV_READ); |
628 | ev_set_priority (&iouring_tfd_w, EV_MAXPRI); |
609 | 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); |
610 | ev_io_start (EV_A_ &iouring_tfd_w); |
634 | ev_unref (EV_A); /* watcher should not keep loop alive */ |
611 | ev_unref (EV_A); /* watcher should not keep loop alive */ |
635 | |
612 | |
636 | backend_modify = iouring_modify; |
613 | backend_modify = iouring_modify; |
637 | backend_poll = iouring_poll; |
614 | backend_poll = iouring_poll; |
… | |
… | |
642 | inline_size |
619 | inline_size |
643 | void |
620 | void |
644 | iouring_destroy (EV_P) |
621 | iouring_destroy (EV_P) |
645 | { |
622 | { |
646 | iouring_internal_destroy (EV_A); |
623 | iouring_internal_destroy (EV_A); |
647 | epoll_destroy (EV_A); |
|
|
648 | } |
624 | } |
649 | |
625 | |