X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=t%2Fio_uring.c;h=dd510be5611a6365992fe228cb69932ae8e53844;hp=af20bbf3dad3800d7645ce208aa9f75d4db5a6d9;hb=ac21bfabdedd0a5e734b8816d6b09ea5c0c1452d;hpb=2ea53ca36174ee16e331ecab33cb413799168e26 diff --git a/t/io_uring.c b/t/io_uring.c index af20bbf3..dd510be5 100644 --- a/t/io_uring.c +++ b/t/io_uring.c @@ -22,7 +22,7 @@ #include "../arch/arch.h" #include "../lib/types.h" -#include "../os/io_uring.h" +#include "../os/linux/io_uring.h" #define barrier() __asm__ __volatile__("": : :"memory") @@ -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 32 +#define BATCH_COMPLETE 32 #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; @@ -83,10 +89,12 @@ 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 register_files = 1; /* use fixed files */ static int buffered = 0; /* use buffered IO, not O_DIRECT */ -static int sq_thread = 0; /* use kernel submission/poller 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_register_buffers(struct submitter *s) { @@ -95,10 +103,33 @@ static int io_uring_register_buffers(struct submitter *s) .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); @@ -116,6 +147,11 @@ 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]; @@ -123,29 +159,48 @@ 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 = &s->files[s->cur_file]; + } + } + f->pending_ios++; lrand48_r(&s->rand, &r); offset = (r % (f->max_blocks - 1)) * BS; - sqe->flags = 0; - sqe->opcode = IORING_OP_READV; + if (register_files) { + sqe->flags = IOSQE_FIXED_FILE; + sqe->fd = f->fixed_fd; + } else { + sqe->flags = 0; + sqe->fd = f->real_fd; + } if (fixedbufs) { + sqe->opcode = IORING_OP_READ_FIXED; sqe->addr = s->iovecs[index].iov_base; sqe->len = BS; sqe->buf_index = index; - sqe->flags |= IOSQE_FIXED_BUFFER; } else { + sqe->opcode = IORING_OP_READV; sqe->addr = &s->iovecs[index]; sqe->len = 1; sqe->buf_index = 0; } sqe->ioprio = 0; - sqe->fd = f->fd; sqe->off = offset; + sqe->user_data = (unsigned long) f; } static int prep_more_ios(struct submitter *s, int max_ios) @@ -180,12 +235,12 @@ 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; @@ -206,13 +261,19 @@ 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 *) (uintptr_t) 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++; @@ -231,6 +292,7 @@ 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()); @@ -249,18 +311,38 @@ static void *submitter_fn(void *data) submit_more: to_submit = prepped; submit: - if (s->inflight + BATCH_SUBMIT < DEPTH) + if (to_submit && (s->inflight + to_submit <= DEPTH)) to_wait = 0; 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) { @@ -326,9 +408,12 @@ static int setup_ring(struct submitter *s) if (polled) p.flags |= IORING_SETUP_IOPOLL; - if (sq_thread) { + if (sq_thread_poll) { p.flags |= IORING_SETUP_SQPOLL; - p.sq_thread_cpu = sq_thread_cpu; + if (sq_thread_cpu != -1) { + p.flags |= IORING_SETUP_SQ_AFF; + p.sq_thread_cpu = sq_thread_cpu; + } } fd = io_uring_setup(DEPTH, &p); @@ -341,7 +426,15 @@ static int setup_ring(struct submitter *s) if (fixedbufs) { ret = io_uring_register_buffers(s); if (ret < 0) { - perror("io_uring_register"); + perror("io_uring_register_buffers"); + return 1; + } + } + + if (register_files) { + ret = io_uring_register_files(s); + if (ret < 0) { + perror("io_uring_register_files"); return 1; } } @@ -354,6 +447,7 @@ 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; @@ -375,25 +469,43 @@ static int setup_ring(struct submitter *s) return 0; } +static void file_depths(char *buf) +{ + struct submitter *s = &submitters[0]; + char *p; + int i; + + buf[0] = '\0'; + p = buf; + for (i = 0; i < s->nr_files; i++) { + struct file *f = &s->files[i]; + + if (i + 1 == s->nr_files) + p += sprintf(p, "%d", f->pending_ios); + else + p += sprintf(p, "%d, ", f->pending_ios); + } +} + int main(int argc, char *argv[]) { struct submitter *s = &submitters[0]; unsigned long done, calls, reap, cache_hit, cache_miss; int err, i, flags, fd; - struct rlimit rlim; + char *fdepths; 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); @@ -401,7 +513,7 @@ 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; @@ -417,11 +529,15 @@ int main(int argc, char *argv[]) i++; } - rlim.rlim_cur = RLIM_INFINITY; - rlim.rlim_max = RLIM_INFINITY; - if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) { - perror("setrlimit"); - return 1; + if (fixedbufs) { + struct rlimit rlim; + + rlim.rlim_cur = RLIM_INFINITY; + rlim.rlim_max = RLIM_INFINITY; + if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) { + perror("setrlimit"); + return 1; + } } arm_sig_int(); @@ -447,6 +563,7 @@ int main(int argc, char *argv[]) pthread_create(&s->thread, NULL, submitter_fn, s); + fdepths = malloc(8 * s->nr_files); cache_hit = cache_miss = reap = calls = done = 0; do { unsigned long this_done = 0; @@ -474,10 +591,12 @@ 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; + file_depths(fdepths); + printf("IOPS=%lu, IOS/call=%ld/%ld, inflight=%u (%s), Cachehit=%0.2f%%\n", this_done - done, rpc, ipc, s->inflight, - *s->cq_ring.head, *s->cq_ring.tail, hit); + fdepths, hit); done = this_done; calls = this_call; reap = this_reap; @@ -487,5 +606,6 @@ int main(int argc, char *argv[]) pthread_join(s->thread, &ret); close(s->ring_fd); + free(fdepths); return 0; }