X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=t%2Fio_uring.c;h=55b75f6ed9664858a7cc6155ce0c159a146a4479;hb=HEAD;hp=f513d7dcd8aa81cdb4c46bf841889b8a6846d583;hpb=a0639afe121870a2a9b69cadc07619ba82959e3e;p=fio.git diff --git a/t/io_uring.c b/t/io_uring.c index f513d7dc..aa6e09e9 100644 --- a/t/io_uring.c +++ b/t/io_uring.c @@ -11,6 +11,10 @@ #include #endif +#ifdef CONFIG_LIBNUMA +#include +#endif + #include #include #include @@ -24,13 +28,16 @@ #include #include #include +#include #include "../arch/arch.h" +#include "../os/os.h" #include "../lib/types.h" #include "../lib/roundup.h" #include "../lib/rand.h" #include "../minmax.h" #include "../os/linux/io_uring.h" +#include "../engines/nvme.h" struct io_sq_ring { unsigned *head; @@ -63,6 +70,8 @@ struct file { unsigned long max_size; unsigned long cur_off; unsigned pending_ios; + unsigned int nsid; /* nsid field required for nvme-passthrough */ + unsigned int lba_shift; /* lba_shift field required for nvme-passthrough */ int real_fd; int fixed_fd; int fileno; @@ -76,6 +85,7 @@ struct file { struct submitter { pthread_t thread; int ring_fd; + int enter_ring_fd; int index; struct io_sq_ring sq_ring; struct io_uring_sqe *sqes; @@ -85,6 +95,7 @@ struct submitter { unsigned long reaps; unsigned long done; unsigned long calls; + unsigned long io_errors; volatile int finish; __s32 *fds; @@ -99,6 +110,10 @@ struct submitter { io_context_t aio_ctx; #endif + int numa_node; + int per_file_depth; + const char *filename; + struct file files[MAX_FDS]; unsigned nr_files; unsigned cur_file; @@ -109,6 +124,7 @@ static struct submitter *submitter; static volatile int finish; static int stats_running; static unsigned long max_iops; +static long t_io_uring_page_size; static int depth = DEPTH; static int batch_submit = BATCH_SUBMIT; @@ -116,7 +132,6 @@ static int batch_complete = BATCH_COMPLETE; static int bs = BS; static int polled = 1; /* use IO polling */ static int fixedbufs = 1; /* use fixed user buffers */ -static int dma_map; /* pre-map DMA buffers */ static int register_files = 1; /* use fixed files */ static int buffered = 0; /* use buffered IO, not O_DIRECT */ static int sq_thread_poll = 0; /* use kernel submission/poller thread */ @@ -127,6 +142,10 @@ static int stats = 0; /* generate IO stats */ static int aio = 0; /* use libaio */ static int runtime = 0; /* runtime */ static int random_io = 1; /* random or sequential IO */ +static int register_ring = 1; /* register ring */ +static int use_sync = 0; /* use preadv2 */ +static int numa_placement = 0; /* set to node of device */ +static int pt = 0; /* passthrough I/O or not */ static unsigned long tsc_rate; @@ -138,16 +157,53 @@ static float plist[] = { 1.0, 5.0, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 95.0, 99.0, 99.5, 99.9, 99.95, 99.99 }; static int plist_len = 17; -#ifndef IORING_REGISTER_MAP_BUFFERS -#define IORING_REGISTER_MAP_BUFFERS 20 -struct io_uring_map_buffers { - __s32 fd; - __u32 buf_start; - __u32 buf_end; - __u32 flags; - __u64 rsvd[2]; -}; -#endif +static int nvme_identify(int fd, __u32 nsid, enum nvme_identify_cns cns, + enum nvme_csi csi, void *data) +{ + struct nvme_passthru_cmd cmd = { + .opcode = nvme_admin_identify, + .nsid = nsid, + .addr = (__u64)(uintptr_t)data, + .data_len = NVME_IDENTIFY_DATA_SIZE, + .cdw10 = cns, + .cdw11 = csi << NVME_IDENTIFY_CSI_SHIFT, + .timeout_ms = NVME_DEFAULT_IOCTL_TIMEOUT, + }; + + return ioctl(fd, NVME_IOCTL_ADMIN_CMD, &cmd); +} + +static int nvme_get_info(int fd, __u32 *nsid, __u32 *lba_sz, __u64 *nlba) +{ + struct nvme_id_ns ns; + int namespace_id; + int err; + + namespace_id = ioctl(fd, NVME_IOCTL_ID); + if (namespace_id < 0) { + fprintf(stderr, "error failed to fetch namespace-id\n"); + close(fd); + return -errno; + } + + /* + * Identify namespace to get namespace-id, namespace size in LBA's + * and LBA data size. + */ + err = nvme_identify(fd, namespace_id, NVME_IDENTIFY_CNS_NS, + NVME_CSI_NVM, &ns); + if (err) { + fprintf(stderr, "error failed to fetch identify namespace\n"); + close(fd); + return err; + } + + *nsid = namespace_id; + *lba_sz = 1 << ns.lbaf[(ns.flbas & 0x0f)].ds; + *nlba = ns.nsze; + + return 0; +} static unsigned long cycles_to_nsec(unsigned long cycles) { @@ -183,9 +239,9 @@ static unsigned long plat_idx_to_val(unsigned int idx) return cycles_to_nsec(base + ((k + 0.5) * (1 << error_bits))); } -unsigned int calc_clat_percentiles(unsigned long *io_u_plat, unsigned long nr, - unsigned long **output, - unsigned long *maxv, unsigned long *minv) +unsigned int calculate_clat_percentiles(unsigned long *io_u_plat, + unsigned long nr, unsigned long **output, + unsigned long *maxv, unsigned long *minv) { unsigned long sum = 0; unsigned int len = plist_len, i, j = 0; @@ -239,7 +295,7 @@ static void show_clat_percentiles(unsigned long *io_u_plat, unsigned long nr, bool is_last; char fmt[32]; - len = calc_clat_percentiles(io_u_plat, nr, &ovals, &maxv, &minv); + len = calculate_clat_percentiles(io_u_plat, nr, &ovals, &maxv, &minv); if (!len || !ovals) goto out; @@ -340,24 +396,6 @@ static void add_stat(struct submitter *s, int clock_index, int nr) #endif } -static int io_uring_map_buffers(struct submitter *s) -{ - struct io_uring_map_buffers map = { - .fd = s->files[0].real_fd, - .buf_end = depth, - }; - - if (do_nop) - return 0; - if (s->nr_files > 1) { - fprintf(stderr, "Can't map buffers with multiple files\n"); - return -1; - } - - return syscall(__NR_io_uring_register, s->ring_fd, - IORING_REGISTER_MAP_BUFFERS, &map, 1); -} - static int io_uring_register_buffers(struct submitter *s) { if (do_nop) @@ -386,6 +424,8 @@ static int io_uring_register_files(struct submitter *s) static int io_uring_setup(unsigned entries, struct io_uring_params *p) { + int ret; + /* * Clamp CQ ring size at our SQ ring size, we don't need more entries * than that. @@ -393,7 +433,28 @@ static int io_uring_setup(unsigned entries, struct io_uring_params *p) p->flags |= IORING_SETUP_CQSIZE; p->cq_entries = entries; - return syscall(__NR_io_uring_setup, entries, p); + p->flags |= IORING_SETUP_COOP_TASKRUN; + p->flags |= IORING_SETUP_SINGLE_ISSUER; + p->flags |= IORING_SETUP_DEFER_TASKRUN; +retry: + ret = syscall(__NR_io_uring_setup, entries, p); + if (!ret) + return 0; + + if (errno == EINVAL && p->flags & IORING_SETUP_COOP_TASKRUN) { + p->flags &= ~IORING_SETUP_COOP_TASKRUN; + goto retry; + } + if (errno == EINVAL && p->flags & IORING_SETUP_SINGLE_ISSUER) { + p->flags &= ~IORING_SETUP_SINGLE_ISSUER; + goto retry; + } + if (errno == EINVAL && p->flags & IORING_SETUP_DEFER_TASKRUN) { + p->flags &= ~IORING_SETUP_DEFER_TASKRUN; + goto retry; + } + + return ret; } static void io_uring_probe(int fd) @@ -401,11 +462,10 @@ static void io_uring_probe(int fd) struct io_uring_probe *p; int ret; - p = malloc(sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); + p = calloc(1, sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); if (!p) return; - memset(p, 0, sizeof(*p) + 256 * sizeof(struct io_uring_probe_op)); ret = syscall(__NR_io_uring_register, fd, IORING_REGISTER_PROBE, p, 256); if (ret < 0) goto out; @@ -422,57 +482,70 @@ out: static int io_uring_enter(struct submitter *s, unsigned int to_submit, unsigned int min_complete, unsigned int flags) { - return syscall(__NR_io_uring_enter, s->ring_fd, to_submit, min_complete, - flags, NULL, 0); + if (register_ring) + flags |= IORING_ENTER_REGISTERED_RING; +#ifdef FIO_ARCH_HAS_SYSCALL + return __do_syscall6(__NR_io_uring_enter, s->enter_ring_fd, to_submit, + min_complete, flags, NULL, 0); +#else + return syscall(__NR_io_uring_enter, s->enter_ring_fd, to_submit, + min_complete, flags, NULL, 0); +#endif } -#ifndef CONFIG_HAVE_GETTID -static int gettid(void) +static unsigned long long get_offset(struct submitter *s, struct file *f) { - return syscall(__NR_gettid); -} -#endif + unsigned long long offset; + long r; -static unsigned file_depth(struct submitter *s) -{ - return (depth + s->nr_files - 1) / s->nr_files; + if (random_io) { + unsigned long long block; + + r = __rand64(&s->rand_state); + block = r % f->max_blocks; + offset = block * (unsigned long long) bs; + } else { + offset = f->cur_off; + f->cur_off += bs; + if (f->cur_off + bs > f->max_size) + f->cur_off = 0; + } + + return offset; } -static void init_io(struct submitter *s, unsigned index) +static struct file *get_next_file(struct submitter *s) { - struct io_uring_sqe *sqe = &s->sqes[index]; - unsigned long offset; struct file *f; - long r; - - 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)) { + if (f->pending_ios >= s->per_file_depth) { s->cur_file++; if (s->cur_file == s->nr_files) s->cur_file = 0; f = &s->files[s->cur_file]; } } + f->pending_ios++; + return f; +} - if (random_io) { - r = __rand64(&s->rand_state); - offset = (r % (f->max_blocks - 1)) * bs; - } else { - offset = f->cur_off; - f->cur_off += bs; - if (f->cur_off + bs > f->max_size) - f->cur_off = 0; +static void init_io(struct submitter *s, unsigned index) +{ + struct io_uring_sqe *sqe = &s->sqes[index]; + struct file *f; + + if (do_nop) { + sqe->opcode = IORING_OP_NOP; + return; } + f = get_next_file(s); + if (register_files) { sqe->flags = IOSQE_FIXED_FILE; sqe->fd = f->fixed_fd; @@ -497,26 +570,76 @@ static void init_io(struct submitter *s, unsigned index) sqe->buf_index = 0; } sqe->ioprio = 0; - sqe->off = offset; + sqe->off = get_offset(s, f); sqe->user_data = (unsigned long) f->fileno; if (stats && stats_running) sqe->user_data |= ((uint64_t)s->clock_index << 32); } +static void init_io_pt(struct submitter *s, unsigned index) +{ + struct io_uring_sqe *sqe = &s->sqes[index << 1]; + unsigned long offset; + struct file *f; + struct nvme_uring_cmd *cmd; + unsigned long long slba; + unsigned long long nlb; + + f = get_next_file(s); + + offset = get_offset(s, f); + + if (register_files) { + sqe->fd = f->fixed_fd; + sqe->flags = IOSQE_FIXED_FILE; + } else { + sqe->fd = f->real_fd; + sqe->flags = 0; + } + sqe->opcode = IORING_OP_URING_CMD; + sqe->user_data = (unsigned long) f->fileno; + if (stats) + sqe->user_data |= ((__u64) s->clock_index << 32ULL); + sqe->cmd_op = NVME_URING_CMD_IO; + slba = offset >> f->lba_shift; + nlb = (bs >> f->lba_shift) - 1; + cmd = (struct nvme_uring_cmd *)&sqe->cmd; + /* cdw10 and cdw11 represent starting slba*/ + cmd->cdw10 = slba & 0xffffffff; + cmd->cdw11 = slba >> 32; + /* cdw12 represent number of lba to be read*/ + cmd->cdw12 = nlb; + cmd->addr = (unsigned long) s->iovecs[index].iov_base; + cmd->data_len = bs; + if (fixedbufs) { + sqe->uring_cmd_flags = IORING_URING_CMD_FIXED; + sqe->buf_index = index; + } + cmd->nsid = f->nsid; + cmd->opcode = 2; +} + static int prep_more_ios_uring(struct submitter *s, int max_ios) { struct io_sq_ring *ring = &s->sq_ring; - unsigned index, tail, next_tail, prepped = 0; + unsigned head, index, tail, next_tail, prepped = 0; + + if (sq_thread_poll) + head = atomic_load_acquire(ring->head); + else + head = *ring->head; next_tail = tail = *ring->tail; do { next_tail++; - if (next_tail == atomic_load_acquire(ring->head)) + if (next_tail == head) break; index = tail & sq_ring_mask; - init_io(s, index); - ring->array[index] = index; + if (pt) + init_io_pt(s, index); + else + init_io(s, index); prepped++; tail = next_tail; } while (prepped < max_ios); @@ -532,7 +655,29 @@ static int get_file_size(struct file *f) if (fstat(f->real_fd, &st) < 0) return -1; - if (S_ISBLK(st.st_mode)) { + if (pt) { + __u64 nlba; + __u32 lbs; + int ret; + + if (!S_ISCHR(st.st_mode)) { + fprintf(stderr, "passthrough works with only nvme-ns " + "generic devices (/dev/ngXnY)\n"); + return -1; + } + ret = nvme_get_info(f->real_fd, &f->nsid, &lbs, &nlba); + if (ret) + return -1; + if ((bs % lbs) != 0) { + printf("error: bs:%d should be a multiple logical_block_size:%d\n", + bs, lbs); + return -1; + } + f->max_blocks = nlba; + f->max_size = nlba; + f->lba_shift = ilog2(lbs); + return 0; + } else if (S_ISBLK(st.st_mode)) { unsigned long long bytes; if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0) @@ -561,7 +706,6 @@ static int reap_events_uring(struct submitter *s) do { struct file *f; - read_barrier(); if (head == atomic_load_acquire(ring->tail)) break; cqe = &ring->cqes[head & cq_ring_mask]; @@ -571,10 +715,14 @@ static int reap_events_uring(struct submitter *s) f = &s->files[fileno]; f->pending_ios--; if (cqe->res != bs) { - printf("io: unexpected ret=%d\n", cqe->res); - if (polled && cqe->res == -EOPNOTSUPP) - printf("Your filesystem/driver/kernel doesn't support polled IO\n"); - return -1; + if (cqe->res == -ENODATA || cqe->res == -EIO) { + s->io_errors++; + } else { + printf("io: unexpected ret=%d\n", cqe->res); + if (polled && cqe->res == -EOPNOTSUPP) + printf("Your filesystem/driver/kernel doesn't support polled IO\n"); + return -1; + } } } if (stats) { @@ -593,29 +741,311 @@ static int reap_events_uring(struct submitter *s) head++; } while (1); - if (stat_nr) - add_stat(s, last_idx, stat_nr); + if (stat_nr) + add_stat(s, last_idx, stat_nr); + + if (reaped) { + s->inflight -= reaped; + atomic_store_release(ring->head, head); + } + return reaped; +} + +static int reap_events_uring_pt(struct submitter *s) +{ + struct io_cq_ring *ring = &s->cq_ring; + struct io_uring_cqe *cqe; + unsigned head, reaped = 0; + int last_idx = -1, stat_nr = 0; + unsigned index; + int fileno; + + head = *ring->head; + do { + struct file *f; + + if (head == atomic_load_acquire(ring->tail)) + break; + index = head & cq_ring_mask; + cqe = &ring->cqes[index << 1]; + fileno = cqe->user_data & 0xffffffff; + f = &s->files[fileno]; + f->pending_ios--; + + if (cqe->res != 0) { + printf("io: unexpected ret=%d\n", cqe->res); + if (polled && cqe->res == -EINVAL) + printf("passthrough doesn't support polled IO\n"); + return -1; + } + if (stats) { + int clock_index = cqe->user_data >> 32; + + if (last_idx != clock_index) { + if (last_idx != -1) { + add_stat(s, last_idx, stat_nr); + stat_nr = 0; + } + last_idx = clock_index; + } + stat_nr++; + } + reaped++; + head++; + } while (1); + + if (stat_nr) + add_stat(s, last_idx, stat_nr); + + if (reaped) { + s->inflight -= reaped; + atomic_store_release(ring->head, head); + } + return reaped; +} + +static void set_affinity(struct submitter *s) +{ +#ifdef CONFIG_LIBNUMA + struct bitmask *mask; + + if (s->numa_node == -1) + return; + + numa_set_preferred(s->numa_node); + + mask = numa_allocate_cpumask(); + numa_node_to_cpus(s->numa_node, mask); + numa_sched_setaffinity(s->tid, mask); +#endif +} + +static int detect_node(struct submitter *s, char *name) +{ +#ifdef CONFIG_LIBNUMA + const char *base = basename(name); + char str[128]; + int ret, fd, node; + + if (pt) + sprintf(str, "/sys/class/nvme-generic/%s/device/numa_node", base); + else + sprintf(str, "/sys/block/%s/device/numa_node", base); + fd = open(str, O_RDONLY); + if (fd < 0) + return -1; + + ret = read(fd, str, sizeof(str)); + if (ret < 0) { + close(fd); + return -1; + } + node = atoi(str); + s->numa_node = node; + close(fd); +#else + s->numa_node = -1; +#endif + return 0; +} + +static int setup_aio(struct submitter *s) +{ +#ifdef CONFIG_LIBAIO + if (polled) { + fprintf(stderr, "aio does not support polled IO\n"); + polled = 0; + } + if (sq_thread_poll) { + fprintf(stderr, "aio does not support SQPOLL IO\n"); + sq_thread_poll = 0; + } + if (do_nop) { + fprintf(stderr, "aio does not support polled IO\n"); + do_nop = 0; + } + if (fixedbufs || register_files) { + fprintf(stderr, "aio does not support registered files or buffers\n"); + fixedbufs = register_files = 0; + } + + s->per_file_depth = (depth + s->nr_files - 1) / s->nr_files; + return io_queue_init(roundup_pow2(depth), &s->aio_ctx); +#else + fprintf(stderr, "Legacy AIO not available on this system/build\n"); + errno = EINVAL; + return -1; +#endif +} + +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, i; + void *ptr; + size_t len; + + memset(&p, 0, sizeof(p)); + + if (polled && !do_nop) + p.flags |= IORING_SETUP_IOPOLL; + 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 (pt) { + p.flags |= IORING_SETUP_SQE128; + p.flags |= IORING_SETUP_CQE32; + } + + fd = io_uring_setup(depth, &p); + if (fd < 0) { + perror("io_uring_setup"); + return 1; + } + s->ring_fd = s->enter_ring_fd = fd; + + io_uring_probe(fd); + + if (fixedbufs) { + struct rlimit rlim; + + rlim.rlim_cur = RLIM_INFINITY; + rlim.rlim_max = RLIM_INFINITY; + /* ignore potential error, not needed on newer kernels */ + setrlimit(RLIMIT_MEMLOCK, &rlim); + + ret = io_uring_register_buffers(s); + if (ret < 0) { + 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; + } + } + + 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); + sring->head = ptr + p.sq_off.head; + 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; + + if (p.flags & IORING_SETUP_SQE128) + len = 2 * p.sq_entries * sizeof(struct io_uring_sqe); + else + len = p.sq_entries * sizeof(struct io_uring_sqe); + s->sqes = mmap(0, len, + PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, + IORING_OFF_SQES); + + if (p.flags & IORING_SETUP_CQE32) { + len = p.cq_off.cqes + + 2 * p.cq_entries * sizeof(struct io_uring_cqe); + } else { + len = p.cq_off.cqes + + p.cq_entries * sizeof(struct io_uring_cqe); + } + ptr = mmap(0, len, + PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, + IORING_OFF_CQ_RING); + cring->head = ptr + p.cq_off.head; + cring->tail = ptr + p.cq_off.tail; + cring->ring_mask = ptr + p.cq_off.ring_mask; + cring->ring_entries = ptr + p.cq_off.ring_entries; + cring->cqes = ptr + p.cq_off.cqes; + cq_ring_mask = *cring->ring_mask; + + for (i = 0; i < p.sq_entries; i++) + sring->array[i] = i; + + s->per_file_depth = INT_MAX; + if (s->nr_files) + s->per_file_depth = (depth + s->nr_files - 1) / s->nr_files; + return 0; +} + +static void *allocate_mem(struct submitter *s, int size) +{ + void *buf; - if (reaped) { - s->inflight -= reaped; - atomic_store_release(ring->head, head); +#ifdef CONFIG_LIBNUMA + if (s->numa_node != -1) + return numa_alloc_onnode(size, s->numa_node); +#endif + + if (posix_memalign(&buf, t_io_uring_page_size, bs)) { + printf("failed alloc\n"); + return NULL; } - return reaped; + + return buf; } static int submitter_init(struct submitter *s) { - int i, nr_batch; - + int i, nr_batch, err; + static int init_printed; + char buf[80]; s->tid = gettid(); - printf("submitter=%d, tid=%d\n", s->index, s->tid); + printf("submitter=%d, tid=%d, file=%s, nfiles=%d, node=%d\n", s->index, s->tid, + s->filename, s->nr_files, s->numa_node); - __init_rand64(&s->rand_state, pthread_self()); - srand48(pthread_self()); + set_affinity(s); + + __init_rand64(&s->rand_state, s->tid); + srand48(s->tid); for (i = 0; i < MAX_FDS; i++) s->files[i].fileno = i; + for (i = 0; i < roundup_pow2(depth); i++) { + void *buf; + + buf = allocate_mem(s, bs); + if (!buf) + return -1; + s->iovecs[i].iov_base = buf; + s->iovecs[i].iov_len = bs; + } + + if (use_sync) { + sprintf(buf, "Engine=preadv2\n"); + err = 0; + } else if (!aio) { + err = setup_ring(s); + if (!err) + sprintf(buf, "Engine=io_uring, sq_ring=%d, cq_ring=%d\n", *s->sq_ring.ring_entries, *s->cq_ring.ring_entries); + } else { + sprintf(buf, "Engine=aio\n"); + err = setup_aio(s); + } + if (err) { + printf("queue setup failed: %s, %d\n", strerror(errno), err); + return -1; + } + + if (!init_printed) { + printf("polled=%d, fixedbufs=%d, register_files=%d, buffered=%d, QD=%d\n", polled, fixedbufs, register_files, buffered, depth); + printf("%s", buf); + init_printed = 1; + } + if (stats) { nr_batch = roundup_pow2(depth / batch_submit); if (nr_batch < 2) @@ -629,7 +1059,16 @@ static int submitter_init(struct submitter *s) s->plat = NULL; nr_batch = 0; } + /* perform the expensive command initialization part for passthrough here + * rather than in the fast path + */ + if (pt) { + for (i = 0; i < roundup_pow2(depth); i++) { + struct io_uring_sqe *sqe = &s->sqes[i << 1]; + memset(&sqe->cmd, 0, sizeof(struct nvme_uring_cmd)); + } + } return nr_batch; } @@ -637,32 +1076,17 @@ static int submitter_init(struct submitter *s) static int prep_more_ios_aio(struct submitter *s, int max_ios, struct iocb *iocbs) { uint64_t data; - long long offset; struct file *f; unsigned index; - long r; index = 0; while (index < max_ios) { struct iocb *iocb = &iocbs[index]; - 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++; + f = get_next_file(s); - r = lrand48(); - offset = (r % (f->max_blocks - 1)) * bs; io_prep_pread(iocb, f->real_fd, s->iovecs[index].iov_base, - s->iovecs[index].iov_len, offset); + s->iovecs[index].iov_len, get_offset(s, f)); data = f->fileno; if (stats && stats_running) @@ -684,10 +1108,14 @@ static int reap_events_aio(struct submitter *s, struct io_event *events, int evs f->pending_ios--; if (events[reaped].res != bs) { - printf("io: unexpected ret=%ld\n", events[reaped].res); - return -1; - } - if (stats) { + if (events[reaped].res == -ENODATA || + events[reaped].res == -EIO) { + s->io_errors++; + } else { + printf("io: unexpected ret=%ld\n", events[reaped].res); + return -1; + } + } else if (stats) { int clock_index = data >> 32; if (last_idx != clock_index) { @@ -719,9 +1147,15 @@ static void *submitter_aio_fn(void *data) struct iocb *iocbs; struct io_event *events; #ifdef ARCH_HAVE_CPU_CLOCK - int nr_batch = submitter_init(s); -#else - submitter_init(s); + int nr_batch; +#endif + + ret = submitter_init(s); + if (ret < 0) + goto done; + +#ifdef ARCH_HAVE_CPU_CLOCK + nr_batch = ret; #endif iocbsptr = calloc(depth, sizeof(struct iocb *)); @@ -785,22 +1219,60 @@ static void *submitter_aio_fn(void *data) free(iocbsptr); free(iocbs); free(events); +done: finish = 1; return NULL; } #endif +static void io_uring_unregister_ring(struct submitter *s) +{ + struct io_uring_rsrc_update up = { + .offset = s->enter_ring_fd, + }; + + syscall(__NR_io_uring_register, s->ring_fd, IORING_UNREGISTER_RING_FDS, + &up, 1); +} + +static int io_uring_register_ring(struct submitter *s) +{ + struct io_uring_rsrc_update up = { + .data = s->ring_fd, + .offset = -1U, + }; + int ret; + + ret = syscall(__NR_io_uring_register, s->ring_fd, + IORING_REGISTER_RING_FDS, &up, 1); + if (ret == 1) { + s->enter_ring_fd = up.offset; + return 0; + } + register_ring = 0; + return -1; +} + static void *submitter_uring_fn(void *data) { struct submitter *s = data; struct io_sq_ring *ring = &s->sq_ring; int ret, prepped; #ifdef ARCH_HAVE_CPU_CLOCK - int nr_batch = submitter_init(s); -#else - submitter_init(s); + int nr_batch; +#endif + + ret = submitter_init(s); + if (ret < 0) + goto done; + +#ifdef ARCH_HAVE_CPU_CLOCK + nr_batch = ret; #endif + if (register_ring) + io_uring_register_ring(s); + prepped = 0; do { int to_wait, to_submit, this_reap, to_prep; @@ -854,7 +1326,10 @@ submit: do { int r; - r = reap_events_uring(s); + if (pt) + r = reap_events_uring_pt(s); + else + r = reap_events_uring(s); if (r == -1) { s->finish = 1; break; @@ -893,9 +1368,68 @@ submit: } } while (!s->finish); + if (register_ring) + io_uring_unregister_ring(s); + +done: + finish = 1; + return NULL; +} + +#ifdef CONFIG_PWRITEV2 +static void *submitter_sync_fn(void *data) +{ + struct submitter *s = data; + int ret; + + if (submitter_init(s) < 0) + goto done; + + do { + uint64_t offset; + struct file *f; + + f = get_next_file(s); + +#ifdef ARCH_HAVE_CPU_CLOCK + if (stats) + s->clock_batch[s->clock_index] = get_cpu_clock(); +#endif + + s->inflight++; + s->calls++; + + offset = get_offset(s, f); + if (polled) + ret = preadv2(f->real_fd, &s->iovecs[0], 1, offset, RWF_HIPRI); + else + ret = preadv2(f->real_fd, &s->iovecs[0], 1, offset, 0); + + if (ret < 0) { + perror("preadv2"); + break; + } else if (ret != bs) { + break; + } + + s->done++; + s->inflight--; + f->pending_ios--; + if (stats) + add_stat(s, s->clock_index, 1); + } while (!s->finish); + +done: + finish = 1; + return NULL; +} +#else +static void *submitter_sync_fn(void *data) +{ finish = 1; return NULL; } +#endif static struct submitter *get_submitter(int offset) { @@ -910,15 +1444,21 @@ static struct submitter *get_submitter(int offset) static void do_finish(const char *reason) { int j; + printf("Exiting on %s\n", reason); for (j = 0; j < nthreads; j++) { struct submitter *s = get_submitter(j); s->finish = 1; } - if (max_iops > 100000) - printf("Maximum IOPS=%luK\n", max_iops / 1000); - else if (max_iops) + if (max_iops > 1000000) { + double miops = (double) max_iops / 1000000.0; + printf("Maximum IOPS=%.2fM\n", miops); + } else if (max_iops > 100000) { + double kiops = (double) max_iops / 1000.0; + printf("Maximum IOPS=%.2fK\n", kiops); + } else { printf("Maximum IOPS=%lu\n", max_iops); + } finish = 1; } @@ -942,144 +1482,6 @@ static void arm_sig_int(void) #endif } -static int setup_aio(struct submitter *s) -{ -#ifdef CONFIG_LIBAIO - if (polled) { - fprintf(stderr, "aio does not support polled IO\n"); - polled = 0; - } - if (sq_thread_poll) { - fprintf(stderr, "aio does not support SQPOLL IO\n"); - sq_thread_poll = 0; - } - if (do_nop) { - fprintf(stderr, "aio does not support polled IO\n"); - do_nop = 0; - } - if (fixedbufs || register_files) { - fprintf(stderr, "aio does not support registered files or buffers\n"); - fixedbufs = register_files = 0; - } - - return io_queue_init(roundup_pow2(depth), &s->aio_ctx); -#else - fprintf(stderr, "Legacy AIO not available on this system/build\n"); - errno = EINVAL; - return -1; -#endif -} - -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; - - memset(&p, 0, sizeof(p)); - - if (polled && !do_nop) - p.flags |= IORING_SETUP_IOPOLL; - 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; - } - } - - fd = io_uring_setup(depth, &p); - if (fd < 0) { - perror("io_uring_setup"); - return 1; - } - s->ring_fd = fd; - - io_uring_probe(fd); - - if (fixedbufs) { - struct rlimit rlim; - - rlim.rlim_cur = RLIM_INFINITY; - rlim.rlim_max = RLIM_INFINITY; - /* ignore potential error, not needed on newer kernels */ - setrlimit(RLIMIT_MEMLOCK, &rlim); - - ret = io_uring_register_buffers(s); - if (ret < 0) { - perror("io_uring_register_buffers"); - return 1; - } - - if (dma_map) { - ret = io_uring_map_buffers(s); - if (ret < 0) { - perror("io_uring_map_buffers"); - return 1; - } - } - } - - if (register_files) { - 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); - sring->head = ptr + p.sq_off.head; - 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; - - s->sqes = mmap(0, p.sq_entries * sizeof(struct io_uring_sqe), - PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, - IORING_OFF_SQES); - - ptr = mmap(0, p.cq_off.cqes + p.cq_entries * sizeof(struct io_uring_cqe), - PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, - IORING_OFF_CQ_RING); - cring->head = ptr + p.cq_off.head; - cring->tail = ptr + p.cq_off.tail; - cring->ring_mask = ptr + p.cq_off.ring_mask; - cring->ring_entries = ptr + p.cq_off.ring_entries; - cring->cqes = ptr + p.cq_off.cqes; - cq_ring_mask = *cring->ring_mask; - return 0; -} - -static void file_depths(char *buf) -{ - bool prev = false; - char *p; - int i, j; - - buf[0] = '\0'; - p = buf; - for (j = 0; j < nthreads; j++) { - struct submitter *s = get_submitter(j); - - for (i = 0; i < s->nr_files; i++) { - struct file *f = &s->files[i]; - - if (prev) - p += sprintf(p, " %d", f->pending_ios); - else - p += sprintf(p, "%d", f->pending_ios); - prev = true; - } - } -} - static void usage(char *argv, int status) { char runtime_str[16]; @@ -1091,7 +1493,6 @@ static void usage(char *argv, int status) " -b : Block size, default %d\n" " -p : Polled IO, default %d\n" " -B : Fixed buffers, default %d\n" - " -D : DMA map fixed buffers, default %d\n" " -F : Register files, default %d\n" " -n : Number of threads, default %d\n" " -O : Use O_DIRECT, default %d\n" @@ -1100,10 +1501,15 @@ static void usage(char *argv, int status) " -T : TSC rate in HZ\n" " -r : Runtime in seconds, default %s\n" " -R : Use random IO, default %d\n" - " -a : Use legacy aio, default %d\n", + " -a : Use legacy aio, default %d\n" + " -S : Use sync IO (preadv2), default %d\n" + " -X : Use registered ring %d\n" + " -P : Automatically place on device home node %d\n" + " -u : Use nvme-passthrough I/O, default %d\n", argv, DEPTH, BATCH_SUBMIT, BATCH_COMPLETE, BS, polled, - fixedbufs, dma_map, register_files, nthreads, !buffered, do_nop, - stats, runtime == 0 ? "unlimited" : runtime_str, random_io, aio); + fixedbufs, register_files, nthreads, !buffered, do_nop, + stats, runtime == 0 ? "unlimited" : runtime_str, random_io, aio, + use_sync, register_ring, numa_placement, pt); exit(status); } @@ -1154,17 +1560,15 @@ static void write_tsc_rate(void) int main(int argc, char *argv[]) { struct submitter *s; - unsigned long done, calls, reap; - int err, i, j, flags, fd, opt, threads_per_f, threads_rem = 0, nfiles; - long page_size; + unsigned long done, calls, reap, io_errors; + int i, j, flags, fd, opt, threads_per_f, threads_rem = 0, nfiles; struct file f; - char *fdepths; void *ret; if (!do_nop && argc < 2) usage(argv[0], 1); - while ((opt = getopt(argc, argv, "d:s:c:b:p:B:F:n:N:O:t:T:a:r:D:R:h?")) != -1) { + while ((opt = getopt(argc, argv, "d:s:c:b:p:B:F:n:N:O:t:T:a:r:D:R:X:S:P:u:h?")) != -1) { switch (opt) { case 'a': aio = !!atoi(optarg); @@ -1225,12 +1629,26 @@ int main(int argc, char *argv[]) case 'r': runtime = atoi(optarg); break; - case 'D': - dma_map = !!atoi(optarg); - break; case 'R': random_io = !!atoi(optarg); break; + case 'X': + register_ring = !!atoi(optarg); + break; + case 'S': +#ifdef CONFIG_PWRITEV2 + use_sync = !!atoi(optarg); +#else + fprintf(stderr, "preadv2 not supported\n"); + exit(1); +#endif + break; + case 'P': + numa_placement = !!atoi(optarg); + break; + case 'u': + pt = !!atoi(optarg); + break; case 'h': case '?': default: @@ -1246,15 +1664,14 @@ int main(int argc, char *argv[]) batch_complete = depth; if (batch_submit > depth) batch_submit = depth; - if (!fixedbufs && dma_map) - dma_map = 0; submitter = calloc(nthreads, sizeof(*submitter) + roundup_pow2(depth) * sizeof(struct iovec)); for (j = 0; j < nthreads; j++) { s = get_submitter(j); + s->numa_node = -1; s->index = j; - s->done = s->calls = s->reaps = 0; + s->done = s->calls = s->reaps = s->io_errors = 0; } flags = O_RDONLY | O_NOATIME; @@ -1310,7 +1727,10 @@ int main(int argc, char *argv[]) memcpy(&s->files[s->nr_files], &f, sizeof(f)); - printf("Added file %s (submitter %d)\n", argv[i], s->index); + if (numa_placement) + detect_node(s, argv[i]); + + s->filename = argv[i]; s->nr_files++; } threads_rem--; @@ -1320,46 +1740,15 @@ int main(int argc, char *argv[]) arm_sig_int(); - page_size = sysconf(_SC_PAGESIZE); - if (page_size < 0) - page_size = 4096; - - for (j = 0; j < nthreads; j++) { - s = get_submitter(j); - for (i = 0; i < roundup_pow2(depth); i++) { - void *buf; - - if (posix_memalign(&buf, page_size, bs)) { - printf("failed alloc\n"); - return 1; - } - s->iovecs[i].iov_base = buf; - s->iovecs[i].iov_len = bs; - } - } - - for (j = 0; j < nthreads; j++) { - s = get_submitter(j); - - if (!aio) - err = setup_ring(s); - else - err = setup_aio(s); - if (err) { - printf("ring setup failed: %s, %d\n", strerror(errno), err); - return 1; - } - } - s = get_submitter(0); - printf("polled=%d, fixedbufs=%d/%d, register_files=%d, buffered=%d, QD=%d\n", polled, fixedbufs, dma_map, register_files, buffered, depth); - if (!aio) - printf("Engine=io_uring, sq_ring=%d, cq_ring=%d\n", *s->sq_ring.ring_entries, *s->cq_ring.ring_entries); - else - printf("Engine=aio\n"); + t_io_uring_page_size = sysconf(_SC_PAGESIZE); + if (t_io_uring_page_size < 0) + t_io_uring_page_size = 4096; for (j = 0; j < nthreads; j++) { s = get_submitter(j); - if (!aio) + if (use_sync) + pthread_create(&s->thread, NULL, submitter_sync_fn, s); + else if (!aio) pthread_create(&s->thread, NULL, submitter_uring_fn, s); #ifdef CONFIG_LIBAIO else @@ -1367,12 +1756,12 @@ int main(int argc, char *argv[]) #endif } - fdepths = malloc(8 * s->nr_files * nthreads); - reap = calls = done = 0; + reap = calls = done = io_errors = 0; do { unsigned long this_done = 0; unsigned long this_reap = 0; unsigned long this_call = 0; + unsigned long this_io_errors = 0; unsigned long rpc = 0, ipc = 0; unsigned long iops, bw; @@ -1393,29 +1782,43 @@ int main(int argc, char *argv[]) this_done += s->done; this_call += s->calls; this_reap += s->reaps; + this_io_errors += s->io_errors; } if (this_call - calls) { rpc = (this_done - done) / (this_call - calls); ipc = (this_reap - reap) / (this_call - calls); } else rpc = ipc = -1; - file_depths(fdepths); iops = this_done - done; + iops -= this_io_errors - io_errors; if (bs > 1048576) bw = iops * (bs / 1048576); else bw = iops / (1048576 / bs); - if (iops > 100000) - printf("IOPS=%luK, ", iops / 1000); - else + if (iops > 1000000) { + double miops = (double) iops / 1000000.0; + printf("IOPS=%.2fM, ", miops); + } else if (iops > 100000) { + double kiops = (double) iops / 1000.0; + printf("IOPS=%.2fK, ", kiops); + } else { printf("IOPS=%lu, ", iops); + } max_iops = max(max_iops, iops); - if (!do_nop) - printf("BW=%luMiB/s, ", bw); - printf("IOS/call=%ld/%ld, inflight=(%s)\n", rpc, ipc, fdepths); + if (!do_nop) { + if (bw > 2000) { + double bw_g = (double) bw / 1000.0; + + printf("BW=%.2fGiB/s, ", bw_g); + } else { + printf("BW=%luMiB/s, ", bw); + } + } + printf("IOS/call=%ld/%ld\n", rpc, ipc); done = this_done; calls = this_call; reap = this_reap; + io_errors = this_io_errors; } while (!finish); for (j = 0; j < nthreads; j++) { @@ -1423,6 +1826,9 @@ int main(int argc, char *argv[]) pthread_join(s->thread, &ret); close(s->ring_fd); + if (s->io_errors) + printf("%d: %lu IO errors\n", s->tid, s->io_errors); + if (stats) { unsigned long nr; @@ -1435,7 +1841,6 @@ int main(int argc, char *argv[]) } } - free(fdepths); free(submitter); return 0; }