--- libev/ev_iouring.c 2019/12/28 07:39:18 1.17 +++ libev/ev_iouring.c 2023/05/14 19:02:31 1.27 @@ -1,7 +1,7 @@ /* * libev linux io_uring fd activity backend * - * Copyright (c) 2019 Marc Alexander Lehmann + * Copyright (c) 2019-2020 Marc Alexander Lehmann * All rights reserved. * * Redistribution and use in source and binary forms, with or without modifica- @@ -46,11 +46,10 @@ * of linux aio or epoll and so on and so on. and you could do event stuff * without any syscalls. what's not to like? * d) ok, it's vastly more complex, but that's ok, really. - * e) why 3 mmaps instead of one? one would be more space-efficient, - * and I can't see what benefit three would have (other than being + * e) why two mmaps instead of one? one would be more space-efficient, + * and I can't see what benefit two would have (other than being * somehow resizable/relocatable, but that's apparently not possible). - * (FIXME: newer kernels can use 2 mmaps only, need to look into this). - * f) hmm, it's practiclaly undebuggable (gdb can't access the memory, and + * f) hmm, it's practically undebuggable (gdb can't access the memory, and * the bizarre way structure offsets are communicated makes it hard to * just print the ring buffer heads, even *iff* the memory were visible * in gdb. but then, that's also ok, really. @@ -61,25 +60,22 @@ * like a ยต-optimisation by the io_uring author for his personal * applications, to the detriment of everybody else who just wants * an event loop. but, umm, ok, if that's all, it could be worse. - * (FIXME: jens mentioned timeout commands, need to investigate) - * h) there is a hardcoded limit of 4096 outstanding events. okay, - * at least there is no arbitrary low system-wide limit... - * (FIXME: apparently, this was increased to 32768 in later kernels( + * (from what I gather from the author Jens Axboe, it simply didn't + * occur to him, and he made good on it by adding an unlimited number + * of timeouts later :). + * h) initially there was a hardcoded limit of 4096 outstanding events. + * later versions not only bump this to 32k, but also can handle + * an unlimited amount of events, so this only affects the batch size. * i) unlike linux aio, you *can* register more then the limit - * of fd events, and the kernel will "gracefully" signal an - * overflow, after which you could destroy and recreate the kernel - * state, a bit bigger, or fall back to e.g. poll. thats not - * totally insane, but kind of questions the point a high - * performance I/O framework when it doesn't really work - * under stress. - * (FIXME: iouring should no longer drop events, need to investigate) - * j) but, oh my! is has exactly the same bugs as the linux aio backend, - * where some undocumented poll combinations just fail. - * so we need epoll AGAIN as a fallback. AGAIN! epoll!! and of course, - * this is completely undocumented, have I mantioned this already? + * of fd events. while early versions of io_uring signalled an overflow + * and you ended up getting wet. 5.5+ does not do this anymore. + * j) but, oh my! it had exactly the same bugs as the linux aio backend, + * where some undocumented poll combinations just fail. fortunately, + * after finally reaching the author, he was more than willing to fix + * this probably in 5.6+. * k) overall, the *API* itself is, I dare to say, not a total trainwreck. - * the big isuess with it are the bugs requiring epoll, which might - * or might not get fixed (do I hold my breath?). + * once the bugs ae fixed (probably in 5.6+), it will be without + * competition. */ /* TODO: use internal TIMEOUT */ @@ -124,10 +120,12 @@ __u32 cancel_flags; __u32 open_flags; __u32 statx_flags; + __u32 fadvise_advice; }; __u64 user_data; union { __u16 buf_index; + __u16 personality; __u64 __pad2[3]; }; }; @@ -176,13 +174,40 @@ struct io_cqring_offsets cq_off; }; +#define IORING_FEAT_SINGLE_MMAP 0x00000001 +#define IORING_FEAT_NODROP 0x00000002 +#define IORING_FEAT_SUBMIT_STABLE 0x00000004 + #define IORING_SETUP_CQSIZE 0x00000008 +#define IORING_SETUP_CLAMP 0x00000010 #define IORING_OP_POLL_ADD 6 #define IORING_OP_POLL_REMOVE 7 #define IORING_OP_TIMEOUT 11 #define IORING_OP_TIMEOUT_REMOVE 12 +#define IORING_REGISTER_EVENTFD 4 +#define IORING_REGISTER_EVENTFD_ASYNC 7 +#define IORING_REGISTER_PROBE 8 + +#define IO_URING_OP_SUPPORTED 1 + +struct io_uring_probe_op { + __u8 op; + __u8 resv; + __u16 flags; + __u32 resv2; +}; + +struct io_uring_probe +{ + __u8 last_op; + __u8 ops_len; + __u16 resv; + __u32 resv2[3]; + struct io_uring_probe_op ops[0]; +}; + /* relative or absolute, reference clock is CLOCK_MONOTONIC */ struct iouring_kernel_timespec { @@ -195,7 +220,6 @@ #define IORING_ENTER_GETEVENTS 0x01 #define IORING_OFF_SQ_RING 0x00000000ULL -#define IORING_OFF_CQ_RING 0x08000000ULL #define IORING_OFF_SQES 0x10000000ULL #define IORING_FEAT_SINGLE_MMAP 0x00000001 @@ -216,19 +240,26 @@ return ev_syscall6 (SYS_io_uring_enter, fd, to_submit, min_complete, flags, sig, sigsz); } +inline_size +int +evsys_io_uring_register (unsigned int fd, unsigned int opcode, void *arg, unsigned int nr_args) +{ + return ev_syscall4 (SYS_io_uring_register, fd, opcode, arg, nr_args); +} + /*****************************************************************************/ -/* actual backed implementation */ +/* actual backend implementation */ /* we hope that volatile will make the compiler access this variables only once */ -#define EV_SQ_VAR(name) *(volatile unsigned *)((char *)iouring_sq_ring + iouring_sq_ ## name) -#define EV_CQ_VAR(name) *(volatile unsigned *)((char *)iouring_cq_ring + iouring_cq_ ## name) +#define EV_SQ_VAR(name) *(volatile unsigned *)((char *)iouring_ring + iouring_sq_ ## name) +#define EV_CQ_VAR(name) *(volatile unsigned *)((char *)iouring_ring + iouring_cq_ ## name) /* the index array */ -#define EV_SQ_ARRAY ((unsigned *)((char *)iouring_sq_ring + iouring_sq_array)) +#define EV_SQ_ARRAY ((unsigned *)((char *)iouring_ring + iouring_sq_array)) /* the submit/completion queue entries */ #define EV_SQES ((struct io_uring_sqe *) iouring_sqes) -#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_cq_ring + iouring_cq_cqes)) +#define EV_CQES ((struct io_uring_cqe *)((char *)iouring_ring + iouring_cq_cqes)) inline_speed int @@ -291,7 +322,7 @@ } inline_size -struct io_uring_sqe * +void iouring_sqe_submit (EV_P_ struct io_uring_sqe *sqe) { unsigned idx = sqe - EV_SQES; @@ -317,15 +348,14 @@ /* called for full and partial cleanup */ ecb_cold -static int +static void iouring_internal_destroy (EV_P) { close (iouring_tfd); close (iouring_fd); - if (iouring_sq_ring != MAP_FAILED) munmap (iouring_sq_ring, iouring_sq_ring_size); - if (iouring_cq_ring != MAP_FAILED) munmap (iouring_cq_ring, iouring_cq_ring_size); - if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes , iouring_sqes_size ); + if (iouring_ring != MAP_FAILED) munmap (iouring_ring, iouring_ring_size); + if (iouring_sqes != MAP_FAILED) munmap (iouring_sqes, iouring_sqes_size); if (ev_is_active (&iouring_tfd_w)) { @@ -339,57 +369,41 @@ iouring_internal_init (EV_P) { struct io_uring_params params = { 0 }; + uint32_t sq_size, cq_size; + + params.flags = IORING_SETUP_CLAMP; iouring_to_submit = 0; - iouring_tfd = -1; - iouring_sq_ring = MAP_FAILED; - iouring_cq_ring = MAP_FAILED; - iouring_sqes = MAP_FAILED; + iouring_tfd = -1; + iouring_ring = MAP_FAILED; + iouring_sqes = MAP_FAILED; if (!have_monotonic) /* cannot really happen, but what if11 */ return -1; - for (;;) - { - iouring_fd = evsys_io_uring_setup (iouring_entries, ¶ms); + iouring_fd = evsys_io_uring_setup (iouring_entries, ¶ms); - if (iouring_fd >= 0) - break; /* yippie */ + if (iouring_fd < 0) + return -1; - if (errno != EINVAL) - return -1; /* we failed */ + if ((~params.features) & (IORING_FEAT_NODROP | IORING_FEAT_SINGLE_MMAP | IORING_FEAT_SUBMIT_STABLE)) + return -1; /* we require the above features */ -#if TODO - if ((~params.features) & (IORING_FEAT_NODROP | IORING_FEATURE_SINGLE_MMAP | IORING_FEAT_SUBMIT_STABLE)) - return -1; /* we require the above features */ -#endif + /* TODO: remember somehow whether our queue size has been clamped */ - /* EINVAL: lots of possible reasons, but maybe - * it is because we hit the unqueryable hardcoded size limit - */ - - /* we hit the limit already, give up */ - if (iouring_max_entries) - return -1; - - /* first time we hit EINVAL? assume we hit the limit, so go back and retry */ - iouring_entries >>= 1; - iouring_max_entries = iouring_entries; - } + sq_size = params.sq_off.array + params.sq_entries * sizeof (unsigned); + cq_size = params.cq_off.cqes + params.cq_entries * sizeof (struct io_uring_cqe); + + iouring_ring_size = sq_size > cq_size ? sq_size : cq_size; + iouring_sqes_size = params.sq_entries * sizeof (struct io_uring_sqe); - iouring_sq_ring_size = params.sq_off.array + params.sq_entries * sizeof (unsigned); - iouring_cq_ring_size = params.cq_off.cqes + params.cq_entries * sizeof (struct io_uring_cqe); - iouring_sqes_size = params.sq_entries * sizeof (struct io_uring_sqe); - - iouring_sq_ring = mmap (0, iouring_sq_ring_size, PROT_READ | PROT_WRITE, - MAP_SHARED | MAP_POPULATE, iouring_fd, IORING_OFF_SQ_RING); - iouring_cq_ring = mmap (0, iouring_cq_ring_size, PROT_READ | PROT_WRITE, - MAP_SHARED | MAP_POPULATE, iouring_fd, IORING_OFF_CQ_RING); - iouring_sqes = mmap (0, iouring_sqes_size, PROT_READ | PROT_WRITE, - MAP_SHARED | MAP_POPULATE, iouring_fd, IORING_OFF_SQES); + iouring_ring = mmap (0, iouring_ring_size, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_POPULATE, iouring_fd, IORING_OFF_SQ_RING); + iouring_sqes = mmap (0, iouring_sqes_size, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_POPULATE, iouring_fd, IORING_OFF_SQES); - if (iouring_sq_ring == MAP_FAILED || iouring_cq_ring == MAP_FAILED || iouring_sqes == MAP_FAILED) + if (iouring_ring == MAP_FAILED || iouring_sqes == MAP_FAILED) return -1; iouring_sq_head = params.sq_off.head; @@ -407,12 +421,12 @@ iouring_cq_overflow = params.cq_off.overflow; iouring_cq_cqes = params.cq_off.cqes; + iouring_tfd_to = EV_TSTAMP_HUGE; + iouring_tfd = timerfd_create (CLOCK_MONOTONIC, TFD_CLOEXEC); if (iouring_tfd < 0) - return iouring_tfd; - - iouring_tfd_to = EV_TSTAMP_HUGE; + return -1; return 0; } @@ -462,6 +476,7 @@ struct io_uring_sqe *sqe = iouring_sqe_get (EV_A); sqe->opcode = IORING_OP_POLL_ADD; sqe->fd = fd; + sqe->addr = 0; sqe->user_data = (uint32_t)fd | ((__u64)(uint32_t)anfds [fd].egen << 32); sqe->poll_events = (nev & EV_READ ? POLLIN : 0) @@ -522,7 +537,6 @@ if (ecb_expect_false (res < 0)) { /*TODO: EINVAL handling (was something failed with this fd)*/ - /*TODO: EBUSY happens when?*/ if (res == -EBADF) {