X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=t%2Fio_uring.c;h=8d3f3a9ba198086dd1145d48ef01185c9849054f;hp=76da6b292f03e2d21623b1d2f48f4bb88dfdb149;hb=8025517dfa599be4bc795e4af7c9012d10b81bc5;hpb=411a08212830ad9f7df297676f8f337efdb4bfc2 diff --git a/t/io_uring.c b/t/io_uring.c index 76da6b29..8d3f3a9b 100644 --- a/t/io_uring.c +++ b/t/io_uring.c @@ -33,6 +33,7 @@ struct io_sq_ring { unsigned *tail; unsigned *ring_mask; unsigned *ring_entries; + unsigned *flags; unsigned *array; }; @@ -44,10 +45,10 @@ struct io_cq_ring { struct io_uring_cqe *cqes; }; -#define DEPTH 32 +#define DEPTH 128 -#define BATCH_SUBMIT 8 -#define BATCH_COMPLETE 8 +#define BATCH_SUBMIT 64 +#define BATCH_COMPLETE 64 #define BS 4096 @@ -57,7 +58,9 @@ static unsigned sq_ring_mask, cq_ring_mask; struct file { unsigned long max_blocks; - int fd; + unsigned pending_ios; + int real_fd; + int fixed_fd; }; struct submitter { @@ -74,6 +77,9 @@ struct submitter { unsigned long calls; unsigned long cachehit, cachemiss; volatile int finish; + + __s32 *fds; + struct file files[MAX_FDS]; unsigned nr_files; unsigned cur_file; @@ -83,15 +89,49 @@ static struct submitter submitters[1]; static volatile int finish; static int polled = 1; /* use IO polling */ -static int fixedbufs = 0; /* use fixed user buffers */ +static int fixedbufs = 1; /* use fixed user buffers */ static int buffered = 0; /* use buffered IO, not O_DIRECT */ -static int sq_thread = 0; /* use kernel submission thread */ -static int sq_thread_cpu = 0; /* pin above thread to this CPU */ +static int sq_thread_poll = 0; /* use kernel submission/poller thread */ +static int sq_thread_cpu = -1; /* pin above thread to this CPU */ +static int do_nop = 0; /* no-op SQ ring commands */ -static int io_uring_setup(unsigned entries, struct iovec *iovecs, - unsigned nr_iovecs, struct io_uring_params *p) +static int io_uring_register_buffers(struct submitter *s) { - return syscall(__NR_sys_io_uring_setup, entries, iovecs, nr_iovecs, p); + struct io_uring_register_buffers reg = { + .iovecs = s->iovecs, + .nr_iovecs = DEPTH + }; + + if (do_nop) + return 0; + + return syscall(__NR_sys_io_uring_register, s->ring_fd, + IORING_REGISTER_BUFFERS, ®); +} + +static int io_uring_register_files(struct submitter *s) +{ + struct io_uring_register_files reg; + int i; + + if (do_nop) + return 0; + + s->fds = calloc(s->nr_files, sizeof(__s32)); + for (i = 0; i < s->nr_files; i++) { + s->fds[i] = s->files[i].real_fd; + s->files[i].fixed_fd = i; + } + reg.fds = s->fds; + reg.nr_fds = s->nr_files; + + return syscall(__NR_sys_io_uring_register, s->ring_fd, + IORING_REGISTER_FILES, ®); +} + +static int io_uring_setup(unsigned entries, struct io_uring_params *p) +{ + return syscall(__NR_sys_io_uring_setup, entries, p); } static int io_uring_enter(struct submitter *s, unsigned int to_submit, @@ -106,6 +146,11 @@ static int gettid(void) return syscall(__NR_gettid); } +static unsigned file_depth(struct submitter *s) +{ + return (DEPTH + s->nr_files - 1) / s->nr_files; +} + static void init_io(struct submitter *s, unsigned index) { struct io_uring_sqe *sqe = &s->sqes[index]; @@ -113,28 +158,42 @@ static void init_io(struct submitter *s, unsigned index) struct file *f; long r; - f = &s->files[s->cur_file]; - s->cur_file++; - if (s->cur_file == s->nr_files) - s->cur_file = 0; + if (do_nop) { + sqe->opcode = IORING_OP_NOP; + return; + } + + if (s->nr_files == 1) { + f = &s->files[0]; + } else { + f = &s->files[s->cur_file]; + if (f->pending_ios >= file_depth(s)) { + s->cur_file++; + if (s->cur_file == s->nr_files) + s->cur_file = 0; + } + } + f->pending_ios++; lrand48_r(&s->rand, &r); offset = (r % (f->max_blocks - 1)) * BS; + sqe->flags = IOSQE_FIXED_FILE; if (fixedbufs) { sqe->opcode = IORING_OP_READ_FIXED; sqe->addr = s->iovecs[index].iov_base; sqe->len = BS; - sqe->index = index; + sqe->buf_index = index; } else { sqe->opcode = IORING_OP_READV; sqe->addr = &s->iovecs[index]; sqe->len = 1; + sqe->buf_index = 0; } - sqe->flags = 0; sqe->ioprio = 0; - sqe->fd = f->fd; + sqe->fd = f->fixed_fd; sqe->off = offset; + sqe->user_data = (unsigned long) f; } static int prep_more_ios(struct submitter *s, int max_ios) @@ -169,12 +228,12 @@ static int get_file_size(struct file *f) { struct stat st; - if (fstat(f->fd, &st) < 0) + if (fstat(f->real_fd, &st) < 0) return -1; if (S_ISBLK(st.st_mode)) { unsigned long long bytes; - if (ioctl(f->fd, BLKGETSIZE64, &bytes) != 0) + if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0) return -1; f->max_blocks = bytes / BS; @@ -195,13 +254,19 @@ static int reap_events(struct submitter *s) head = *ring->head; do { + struct file *f; + barrier(); if (head == *ring->tail) break; cqe = &ring->cqes[head & cq_ring_mask]; - if (cqe->res != BS) { - printf("io: unexpected ret=%d\n", cqe->res); - return -1; + if (!do_nop) { + f = (struct file *) cqe->user_data; + f->pending_ios--; + if (cqe->res != BS) { + printf("io: unexpected ret=%d\n", cqe->res); + return -1; + } } if (cqe->flags & IOCQE_FLAG_CACHEHIT) s->cachehit++; @@ -220,6 +285,7 @@ static int reap_events(struct submitter *s) static void *submitter_fn(void *data) { struct submitter *s = data; + struct io_sq_ring *ring = &s->sq_ring; int ret, prepped; printf("submitter=%d\n", gettid()); @@ -243,13 +309,33 @@ submit: else to_wait = min(s->inflight + to_submit, BATCH_COMPLETE); - ret = io_uring_enter(s, to_submit, to_wait, - IORING_ENTER_GETEVENTS); - s->calls++; + /* + * Only need to call io_uring_enter if we're not using SQ thread + * poll, or if IORING_SQ_NEED_WAKEUP is set. + */ + if (!sq_thread_poll || (*ring->flags & IORING_SQ_NEED_WAKEUP)) { + unsigned flags = 0; + + if (to_wait) + flags = IORING_ENTER_GETEVENTS; + ret = io_uring_enter(s, to_submit, to_wait, flags); + s->calls++; + } - this_reap = reap_events(s); - if (this_reap == -1) - break; + /* + * For non SQ thread poll, we already got the events we needed + * through the io_uring_enter() above. For SQ thread poll, we + * need to loop here until we find enough events. + */ + this_reap = 0; + do { + int r; + r = reap_events(s); + if (r == -1) + break; + else if (r > 0) + this_reap += r; + } while (sq_thread_poll && this_reap < to_wait); s->reaps += this_reap; if (ret >= 0) { @@ -308,30 +394,42 @@ static int setup_ring(struct submitter *s) struct io_sq_ring *sring = &s->sq_ring; struct io_cq_ring *cring = &s->cq_ring; struct io_uring_params p; + int ret, fd; void *ptr; - int fd; memset(&p, 0, sizeof(p)); if (polled) p.flags |= IORING_SETUP_IOPOLL; - if (buffered) - p.flags |= IORING_SETUP_SQWQ; - else if (sq_thread) { - p.flags |= IORING_SETUP_SQTHREAD; - p.sq_thread_cpu = sq_thread_cpu; + if (sq_thread_poll) { + p.flags |= IORING_SETUP_SQPOLL; + if (sq_thread_cpu != -1) { + p.flags |= IORING_SETUP_SQ_AFF; + p.sq_thread_cpu = sq_thread_cpu; + } } - if (fixedbufs) - fd = io_uring_setup(DEPTH, s->iovecs, DEPTH, &p); - else - fd = io_uring_setup(DEPTH, NULL, 0, &p); + fd = io_uring_setup(DEPTH, &p); if (fd < 0) { perror("io_uring_setup"); return 1; } - s->ring_fd = fd; + + if (fixedbufs) { + ret = io_uring_register_buffers(s); + if (ret < 0) { + perror("io_uring_register_buffers"); + return 1; + } + } + + ret = io_uring_register_files(s); + if (ret < 0) { + perror("io_uring_register_files"); + return 1; + } + ptr = mmap(0, p.sq_off.array + p.sq_entries * sizeof(__u32), PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_SQ_RING); @@ -340,6 +438,7 @@ static int setup_ring(struct submitter *s) sring->tail = ptr + p.sq_off.tail; sring->ring_mask = ptr + p.sq_off.ring_mask; sring->ring_entries = ptr + p.sq_off.ring_entries; + sring->flags = ptr + p.sq_off.flags; sring->array = ptr + p.sq_off.array; sq_ring_mask = *sring->ring_mask; @@ -369,17 +468,17 @@ int main(int argc, char *argv[]) struct rlimit rlim; void *ret; - if (argc < 2) { + if (!do_nop && argc < 2) { printf("%s: filename\n", argv[0]); return 1; } - flags = O_RDONLY; + flags = O_RDONLY | O_NOATIME; if (!buffered) flags |= O_DIRECT; i = 1; - while (i < argc) { + while (!do_nop && i < argc) { struct file *f = &s->files[s->nr_files]; fd = open(argv[i], flags); @@ -387,7 +486,7 @@ int main(int argc, char *argv[]) perror("open"); return 1; } - f->fd = fd; + f->real_fd = fd; if (get_file_size(f)) { printf("failed getting size of device/file\n"); return 1; @@ -460,8 +559,9 @@ int main(int argc, char *argv[]) if (this_call - calls) { rpc = (this_done - done) / (this_call - calls); ipc = (this_reap - reap) / (this_call - calls); - } - printf("IOPS=%lu, IOS/call=%lu/%lu, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n", + } else + rpc = ipc = -1; + printf("IOPS=%lu, IOS/call=%ld/%ld, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n", this_done - done, rpc, ipc, s->inflight, *s->cq_ring.head, *s->cq_ring.tail, hit); done = this_done;