X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=t%2Fio_uring.c;h=d563638019ad87edff09d3d7944a553bd145add8;hp=78b92685316cb7bc751cd3b49295be388f8a5807;hb=6c5d3a1c08bda1bbf22187c7b80573400e1c1053;hpb=e4db5bd917962ce3c4ccf2d88c908059e4ab15cb

diff --git a/t/io_uring.c b/t/io_uring.c
index 78b92685..d5636380 100644
--- a/t/io_uring.c
+++ b/t/io_uring.c
@@ -24,8 +24,6 @@
 #include "../lib/types.h"
 #include "../os/linux/io_uring.h"
 
-#define barrier()	__asm__ __volatile__("": : :"memory")
-
 #define min(a, b)		((a < b) ? (a) : (b))
 
 struct io_sq_ring {
@@ -46,10 +44,8 @@ struct io_cq_ring {
 };
 
 #define DEPTH			128
-
 #define BATCH_SUBMIT		32
 #define BATCH_COMPLETE		32
-
 #define BS			4096
 
 #define MAX_FDS			16
@@ -66,16 +62,14 @@ struct file {
 struct submitter {
 	pthread_t thread;
 	int ring_fd;
-	struct drand48_data rand;
+	int index;
 	struct io_sq_ring sq_ring;
 	struct io_uring_sqe *sqes;
-	struct iovec iovecs[DEPTH];
 	struct io_cq_ring cq_ring;
 	int inflight;
 	unsigned long reaps;
 	unsigned long done;
 	unsigned long calls;
-	unsigned long cachehit, cachemiss;
 	volatile int finish;
 
 	__s32 *fds;
@@ -83,11 +77,16 @@ struct submitter {
 	struct file files[MAX_FDS];
 	unsigned nr_files;
 	unsigned cur_file;
+	struct iovec iovecs[];
 };
 
-static struct submitter submitters[1];
+static struct submitter *submitter;
 static volatile int finish;
 
+static int depth = DEPTH;
+static int batch_submit = BATCH_SUBMIT;
+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 register_files = 1;	/* use fixed files */
@@ -95,24 +94,21 @@ static int buffered = 0;	/* use buffered IO, not O_DIRECT */
 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 nthreads = 1;
+
+static int vectored = 1;
 
 static int io_uring_register_buffers(struct submitter *s)
 {
-	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, &reg);
+	return syscall(__NR_io_uring_register, s->ring_fd,
+			IORING_REGISTER_BUFFERS, s->iovecs, depth);
 }
 
 static int io_uring_register_files(struct submitter *s)
 {
-	struct io_uring_register_files reg;
 	int i;
 
 	if (do_nop)
@@ -123,33 +119,56 @@ static int io_uring_register_files(struct submitter *s)
 		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, &reg);
+	return syscall(__NR_io_uring_register, s->ring_fd,
+			IORING_REGISTER_FILES, s->fds, s->nr_files);
 }
 
 static int io_uring_setup(unsigned entries, struct io_uring_params *p)
 {
-	return syscall(__NR_sys_io_uring_setup, entries, p);
+	return syscall(__NR_io_uring_setup, entries, p);
+}
+
+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));
+	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;
+
+	if (IORING_OP_READ > p->ops_len)
+		goto out;
+
+	if ((p->ops[IORING_OP_READ].flags & IO_URING_OP_SUPPORTED))
+		vectored = 0;
+out:
+	free(p);
 }
 
 static int io_uring_enter(struct submitter *s, unsigned int to_submit,
 			  unsigned int min_complete, unsigned int flags)
 {
-	return syscall(__NR_sys_io_uring_enter, s->ring_fd, to_submit,
-			min_complete, flags);
+	return syscall(__NR_io_uring_enter, s->ring_fd, to_submit, min_complete,
+			flags, NULL, 0);
 }
 
+#ifndef CONFIG_HAVE_GETTID
 static int gettid(void)
 {
 	return syscall(__NR_gettid);
 }
+#endif
 
 static unsigned file_depth(struct submitter *s)
 {
-	return (DEPTH + s->nr_files - 1) / s->nr_files;
+	return (depth + s->nr_files - 1) / s->nr_files;
 }
 
 static void init_io(struct submitter *s, unsigned index)
@@ -177,8 +196,8 @@ static void init_io(struct submitter *s, unsigned index)
 	}
 	f->pending_ios++;
 
-	lrand48_r(&s->rand, &r);
-	offset = (r % (f->max_blocks - 1)) * BS;
+	r = lrand48();
+	offset = (r % (f->max_blocks - 1)) * bs;
 
 	if (register_files) {
 		sqe->flags = IOSQE_FIXED_FILE;
@@ -189,12 +208,17 @@ static void init_io(struct submitter *s, unsigned index)
 	}
 	if (fixedbufs) {
 		sqe->opcode = IORING_OP_READ_FIXED;
-		sqe->addr = s->iovecs[index].iov_base;
-		sqe->len = BS;
+		sqe->addr = (unsigned long) s->iovecs[index].iov_base;
+		sqe->len = bs;
 		sqe->buf_index = index;
+	} else if (!vectored) {
+		sqe->opcode = IORING_OP_READ;
+		sqe->addr = (unsigned long) s->iovecs[index].iov_base;
+		sqe->len = bs;
+		sqe->buf_index = 0;
 	} else {
 		sqe->opcode = IORING_OP_READV;
-		sqe->addr = &s->iovecs[index];
+		sqe->addr = (unsigned long) &s->iovecs[index];
 		sqe->len = 1;
 		sqe->buf_index = 0;
 	}
@@ -211,8 +235,7 @@ static int prep_more_ios(struct submitter *s, int max_ios)
 	next_tail = tail = *ring->tail;
 	do {
 		next_tail++;
-		barrier();
-		if (next_tail == *ring->head)
+		if (next_tail == atomic_load_acquire(ring->head))
 			break;
 
 		index = tail & sq_ring_mask;
@@ -222,12 +245,8 @@ static int prep_more_ios(struct submitter *s, int max_ios)
 		tail = next_tail;
 	} while (prepped < max_ios);
 
-	if (*ring->tail != tail) {
-		/* order tail store with writes to sqes above */
-		barrier();
-		*ring->tail = tail;
-		barrier();
-	}
+	if (prepped)
+		atomic_store_release(ring->tail, tail);
 	return prepped;
 }
 
@@ -243,10 +262,10 @@ static int get_file_size(struct file *f)
 		if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0)
 			return -1;
 
-		f->max_blocks = bytes / BS;
+		f->max_blocks = bytes / bs;
 		return 0;
 	} else if (S_ISREG(st.st_mode)) {
-		f->max_blocks = st.st_size / BS;
+		f->max_blocks = st.st_size / bs;
 		return 0;
 	}
 
@@ -263,29 +282,28 @@ static int reap_events(struct submitter *s)
 	do {
 		struct file *f;
 
-		barrier();
-		if (head == *ring->tail)
+		read_barrier();
+		if (head == atomic_load_acquire(ring->tail))
 			break;
 		cqe = &ring->cqes[head & cq_ring_mask];
 		if (!do_nop) {
 			f = (struct file *) (uintptr_t) cqe->user_data;
 			f->pending_ios--;
-			if (cqe->res != BS) {
+			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->flags & IOCQE_FLAG_CACHEHIT)
-			s->cachehit++;
-		else
-			s->cachemiss++;
 		reaped++;
 		head++;
 	} while (1);
 
-	s->inflight -= reaped;
-	*ring->head = head;
-	barrier();
+	if (reaped) {
+		s->inflight -= reaped;
+		atomic_store_release(ring->head, head);
+	}
 	return reaped;
 }
 
@@ -297,36 +315,44 @@ static void *submitter_fn(void *data)
 
 	printf("submitter=%d\n", gettid());
 
-	srand48_r(pthread_self(), &s->rand);
+	srand48(pthread_self());
 
 	prepped = 0;
 	do {
 		int to_wait, to_submit, this_reap, to_prep;
+		unsigned ring_flags = 0;
 
-		if (!prepped && s->inflight < DEPTH) {
-			to_prep = min(DEPTH - s->inflight, BATCH_SUBMIT);
+		if (!prepped && s->inflight < depth) {
+			to_prep = min(depth - s->inflight, batch_submit);
 			prepped = prep_more_ios(s, to_prep);
 		}
 		s->inflight += prepped;
 submit_more:
 		to_submit = prepped;
 submit:
-		if (to_submit && (s->inflight + to_submit <= DEPTH))
+		if (to_submit && (s->inflight + to_submit <= depth))
 			to_wait = 0;
 		else
-			to_wait = min(s->inflight + to_submit, BATCH_COMPLETE);
+			to_wait = min(s->inflight + to_submit, batch_complete);
 
 		/*
 		 * 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)) {
+		if (sq_thread_poll)
+			ring_flags = atomic_load_acquire(ring->flags);
+		if (!sq_thread_poll || ring_flags & IORING_SQ_NEED_WAKEUP) {
 			unsigned flags = 0;
 
 			if (to_wait)
 				flags = IORING_ENTER_GETEVENTS;
+			if (ring_flags & IORING_SQ_NEED_WAKEUP)
+				flags |= IORING_ENTER_SQ_WAKEUP;
 			ret = io_uring_enter(s, to_submit, to_wait, flags);
 			s->calls++;
+		} else {
+			/* for SQPOLL, we submitted it all effectively */
+			ret = to_submit;
 		}
 
 		/*
@@ -338,9 +364,10 @@ submit:
 		do {
 			int r;
 			r = reap_events(s);
-			if (r == -1)
+			if (r == -1) {
+				s->finish = 1;
 				break;
-			else if (r > 0)
+			} else if (r > 0)
 				this_reap += r;
 		} while (sq_thread_poll && this_reap < to_wait);
 		s->reaps += this_reap;
@@ -379,10 +406,25 @@ submit:
 	return NULL;
 }
 
+static struct submitter *get_submitter(int offset)
+{
+	void *ret;
+
+	ret = submitter;
+	if (offset)
+		ret += offset * (sizeof(*submitter) + depth * sizeof(struct iovec));
+	return ret;
+}
+
 static void sig_int(int sig)
 {
+	int j;
+
 	printf("Exiting on signal %d\n", sig);
-	submitters[0].finish = 1;
+	for (j = 0; j < nthreads; j++) {
+		struct submitter *s = get_submitter(j);
+		s->finish = 1;
+	}
 	finish = 1;
 }
 
@@ -406,7 +448,7 @@ static int setup_ring(struct submitter *s)
 
 	memset(&p, 0, sizeof(p));
 
-	if (polled)
+	if (polled && !do_nop)
 		p.flags |= IORING_SETUP_IOPOLL;
 	if (sq_thread_poll) {
 		p.flags |= IORING_SETUP_SQPOLL;
@@ -416,14 +458,23 @@ static int setup_ring(struct submitter *s)
 		}
 	}
 
-	fd = io_uring_setup(DEPTH, &p);
+	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");
@@ -471,121 +522,223 @@ static int setup_ring(struct submitter *s)
 
 static void file_depths(char *buf)
 {
-	struct submitter *s = &submitters[0];
+	bool prev = false;
 	char *p;
-	int i;
+	int i, j;
 
+	buf[0] = '\0';
 	p = buf;
-	for (i = 0; i < s->nr_files; i++) {
-		struct file *f = &s->files[i];
+	for (j = 0; j < nthreads; j++) {
+		struct submitter *s = get_submitter(j);
 
-		if (i + 1 == s->nr_files)
-			p += sprintf(p, "%d", f->pending_ios);
-		else
-			p += sprintf(p, "%d, ", f->pending_ios);
+		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)
+{
+	printf("%s [options] -- [filenames]\n"
+		" -d <int>  : IO Depth, default %d\n"
+		" -s <int>  : Batch submit, default %d\n"
+		" -c <int>  : Batch complete, default %d\n"
+		" -b <int>  : Block size, default %d\n"
+		" -p <bool> : Polled IO, default %d\n"
+		" -B <bool> : Fixed buffers, default %d\n"
+		" -F <bool> : Register files, default %d\n"
+		" -n <int>  : Number of threads, default %d\n"
+		" -O <bool> : Use O_DIRECT, default %d\n"
+		" -N <bool> : Perform just no-op requests, default %d\n",
+		argv, DEPTH, BATCH_SUBMIT, BATCH_COMPLETE, BS, polled,
+		fixedbufs, register_files, nthreads, !buffered, do_nop);
+	exit(status);
+}
+
 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 submitter *s;
+	unsigned long done, calls, reap;
+	int err, i, j, flags, fd, opt, threads_per_f, threads_rem = 0, nfiles;
+	struct file f;
 	char *fdepths;
 	void *ret;
 
-	if (!do_nop && argc < 2) {
-		printf("%s: filename\n", argv[0]);
-		return 1;
+	if (!do_nop && argc < 2)
+		usage(argv[0], 1);
+
+	while ((opt = getopt(argc, argv, "d:s:c:b:p:B:F:n:N:O:h?")) != -1) {
+		switch (opt) {
+		case 'd':
+			depth = atoi(optarg);
+			break;
+		case 's':
+			batch_submit = atoi(optarg);
+			break;
+		case 'c':
+			batch_complete = atoi(optarg);
+			break;
+		case 'b':
+			bs = atoi(optarg);
+			break;
+		case 'p':
+			polled = !!atoi(optarg);
+			break;
+		case 'B':
+			fixedbufs = !!atoi(optarg);
+			break;
+		case 'F':
+			register_files = !!atoi(optarg);
+			break;
+		case 'n':
+			nthreads = atoi(optarg);
+			if (!nthreads) {
+				printf("Threads must be non-zero\n");
+				usage(argv[0], 1);
+			}
+			break;
+		case 'N':
+			do_nop = !!atoi(optarg);
+			break;
+		case 'O':
+			buffered = !atoi(optarg);
+			break;
+		case 'h':
+		case '?':
+		default:
+			usage(argv[0], 0);
+			break;
+		}
+	}
+
+	if (batch_complete > depth)
+		batch_complete = depth;
+	if (batch_submit > depth)
+		batch_submit = depth;
+
+	submitter = calloc(nthreads, sizeof(*submitter) +
+				depth * sizeof(struct iovec));
+	for (j = 0; j < nthreads; j++) {
+		s = get_submitter(j);
+		s->index = j;
+		s->done = s->calls = s->reaps = 0;
 	}
 
 	flags = O_RDONLY | O_NOATIME;
 	if (!buffered)
 		flags |= O_DIRECT;
 
-	i = 1;
+	j = 0;
+	i = optind;
+	nfiles = argc - i;
+	if (!do_nop) {
+		if (!nfiles) {
+			printf("No files specified\n");
+			usage(argv[0], 1);
+		}
+		threads_per_f = nthreads / nfiles;
+		/* make sure each thread gets assigned files */
+		if (threads_per_f == 0) {
+			threads_per_f = 1;
+		} else {
+			threads_rem = nthreads - threads_per_f * nfiles;
+		}
+	}
 	while (!do_nop && i < argc) {
-		struct file *f = &s->files[s->nr_files];
+		int k, limit;
+
+		memset(&f, 0, sizeof(f));
 
 		fd = open(argv[i], flags);
 		if (fd < 0) {
 			perror("open");
 			return 1;
 		}
-		f->real_fd = fd;
-		if (get_file_size(f)) {
+		f.real_fd = fd;
+		if (get_file_size(&f)) {
 			printf("failed getting size of device/file\n");
 			return 1;
 		}
-		if (f->max_blocks <= 1) {
+		if (f.max_blocks <= 1) {
 			printf("Zero file/device size?\n");
 			return 1;
 		}
-		f->max_blocks--;
+		f.max_blocks--;
 
-		printf("Added file %s\n", argv[i]);
-		s->nr_files++;
-		i++;
-	}
+		limit = threads_per_f;
+		limit += threads_rem > 0 ? 1 : 0;
+		for (k = 0; k < limit; k++) {
+			s = get_submitter((j + k) % nthreads);
 
-	if (fixedbufs) {
-		struct rlimit rlim;
+			if (s->nr_files == MAX_FDS) {
+				printf("Max number of files (%d) reached\n", MAX_FDS);
+				break;
+			}
 
-		rlim.rlim_cur = RLIM_INFINITY;
-		rlim.rlim_max = RLIM_INFINITY;
-		if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
-			perror("setrlimit");
-			return 1;
+			memcpy(&s->files[s->nr_files], &f, sizeof(f));
+
+			printf("Added file %s (submitter %d)\n", argv[i], s->index);
+			s->nr_files++;
 		}
+		threads_rem--;
+		i++;
+		j += limit;
 	}
 
 	arm_sig_int();
 
-	for (i = 0; i < DEPTH; i++) {
-		void *buf;
+	for (j = 0; j < nthreads; j++) {
+		s = get_submitter(j);
+		for (i = 0; i < depth; i++) {
+			void *buf;
 
-		if (posix_memalign(&buf, BS, BS)) {
-			printf("failed alloc\n");
-			return 1;
+			if (posix_memalign(&buf, bs, bs)) {
+				printf("failed alloc\n");
+				return 1;
+			}
+			s->iovecs[i].iov_base = buf;
+			s->iovecs[i].iov_len = bs;
 		}
-		s->iovecs[i].iov_base = buf;
-		s->iovecs[i].iov_len = BS;
 	}
 
-	err = setup_ring(s);
-	if (err) {
-		printf("ring setup failed: %s, %d\n", strerror(errno), err);
-		return 1;
+	for (j = 0; j < nthreads; j++) {
+		s = get_submitter(j);
+
+		err = setup_ring(s);
+		if (err) {
+			printf("ring setup failed: %s, %d\n", strerror(errno), err);
+			return 1;
+		}
 	}
-	printf("polled=%d, fixedbufs=%d, buffered=%d", polled, fixedbufs, buffered);
-	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", DEPTH, *s->sq_ring.ring_entries, *s->cq_ring.ring_entries);
+	s = get_submitter(0);
+	printf("polled=%d, fixedbufs=%d, register_files=%d, buffered=%d", polled, fixedbufs, register_files, buffered);
+	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", depth, *s->sq_ring.ring_entries, *s->cq_ring.ring_entries);
 
-	pthread_create(&s->thread, NULL, submitter_fn, s);
+	for (j = 0; j < nthreads; j++) {
+		s = get_submitter(j);
+		pthread_create(&s->thread, NULL, submitter_fn, s);
+	}
 
-	fdepths = malloc(8 * s->nr_files);
-	cache_hit = cache_miss = reap = calls = done = 0;
+	fdepths = malloc(8 * s->nr_files * nthreads);
+	reap = calls = done = 0;
 	do {
 		unsigned long this_done = 0;
 		unsigned long this_reap = 0;
 		unsigned long this_call = 0;
-		unsigned long this_cache_hit = 0;
-		unsigned long this_cache_miss = 0;
 		unsigned long rpc = 0, ipc = 0;
-		double hit = 0.0;
+		unsigned long iops, bw;
 
 		sleep(1);
-		this_done += s->done;
-		this_call += s->calls;
-		this_reap += s->reaps;
-		this_cache_hit += s->cachehit;
-		this_cache_miss += s->cachemiss;
-		if (this_cache_hit && this_cache_miss) {
-			unsigned long hits, total;
-
-			hits = this_cache_hit - cache_hit;
-			total = hits + this_cache_miss - cache_miss;
-			hit = (double) hits / (double) total;
-			hit *= 100.0;
+		for (j = 0; j < nthreads; j++) {
+			this_done += s->done;
+			this_call += s->calls;
+			this_reap += s->reaps;
 		}
 		if (this_call - calls) {
 			rpc = (this_done - done) / (this_call - calls);
@@ -593,18 +746,26 @@ int main(int argc, char *argv[])
 		} 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,
-				fdepths, hit);
+		iops = this_done - done;
+		if (bs > 1048576)
+			bw = iops * (bs / 1048576);
+		else
+			bw = iops / (1048576 / bs);
+		printf("IOPS=%lu, ", iops);
+		if (!do_nop)
+			printf("BW=%luMiB/s, ", bw);
+		printf("IOS/call=%ld/%ld, inflight=(%s)\n", rpc, ipc, fdepths);
 		done = this_done;
 		calls = this_call;
 		reap = this_reap;
-		cache_hit = s->cachehit;
-		cache_miss = s->cachemiss;
 	} while (!finish);
 
-	pthread_join(s->thread, &ret);
-	close(s->ring_fd);
+	for (j = 0; j < nthreads; j++) {
+		s = get_submitter(j);
+		pthread_join(s->thread, &ret);
+		close(s->ring_fd);
+	}
 	free(fdepths);
+	free(submitter);
 	return 0;
 }