[fio.git] / t / io_uring.c

#include <stdio.h>
#include <errno.h>
#include <assert.h>
#include <stdlib.h>
#include <stddef.h>
#include <signal.h>
#include <inttypes.h>

#include <sys/types.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
#include <sys/syscall.h>
#include <sys/resource.h>
#include <sys/mman.h>
#include <sys/uio.h>
#include <linux/fs.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <pthread.h>
#include <sched.h>

#include "../arch/arch.h"
#include "../lib/types.h"
#include "../os/io_uring.h"

#define barrier()	__asm__ __volatile__("": : :"memory")

#define min(a, b)		((a < b) ? (a) : (b))

struct io_sq_ring {
	unsigned *head;
	unsigned *tail;
	unsigned *ring_mask;
	unsigned *ring_entries;
	unsigned *flags;
	unsigned *array;
};

struct io_cq_ring {
	unsigned *head;
	unsigned *tail;
	unsigned *ring_mask;
	unsigned *ring_entries;
	struct io_uring_cqe *cqes;
};

#define DEPTH			128

#define BATCH_SUBMIT		64
#define BATCH_COMPLETE		64

#define BS			4096

#define MAX_FDS			16

static unsigned sq_ring_mask, cq_ring_mask;

struct file {
	unsigned long max_blocks;
	unsigned pending_ios;
	int real_fd;
	int fixed_fd;
};

struct submitter {
	pthread_t thread;
	int ring_fd;
	struct drand48_data rand;
	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;

	struct file files[MAX_FDS];
	unsigned nr_files;
	unsigned cur_file;
};

static struct submitter submitters[1];
static volatile int finish;

static int polled = 1;		/* use IO polling */
static int fixedbufs = 1;	/* use fixed user buffers */
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 io_uring_register_buffers(struct submitter *s)
{
	struct io_uring_register_buffers reg = {
		.iovecs = s->iovecs,
		.nr_iovecs = DEPTH
	};

	return syscall(__NR_sys_io_uring_register, s->ring_fd,
			IORING_REGISTER_BUFFERS, &reg);
}

static int io_uring_register_files(struct submitter *s)
{
	struct io_uring_register_files reg;
	int i;

	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, &reg);
}

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,
			  unsigned int min_complete, unsigned int flags)
{
	return syscall(__NR_sys_io_uring_enter, s->ring_fd, to_submit,
			min_complete, flags);
}

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];
	unsigned long offset;
	struct file *f;
	long r;

	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->buf_index = index;
	} else {
		sqe->opcode = IORING_OP_READV;
		sqe->addr = &s->iovecs[index];
		sqe->len = 1;
		sqe->buf_index = 0;
	}
	sqe->ioprio = 0;
	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)
{
	struct io_sq_ring *ring = &s->sq_ring;
	unsigned index, tail, next_tail, prepped = 0;

	next_tail = tail = *ring->tail;
	do {
		next_tail++;
		barrier();
		if (next_tail == *ring->head)
			break;

		index = tail & sq_ring_mask;
		init_io(s, index);
		ring->array[index] = index;
		prepped++;
		tail = next_tail;
	} while (prepped < max_ios);

	if (*ring->tail != tail) {
		/* order tail store with writes to sqes above */
		barrier();
		*ring->tail = tail;
		barrier();
	}
	return prepped;
}

static int get_file_size(struct file *f)
{
	struct stat st;

	if (fstat(f->real_fd, &st) < 0)
		return -1;
	if (S_ISBLK(st.st_mode)) {
		unsigned long long bytes;

		if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0)
			return -1;

		f->max_blocks = bytes / BS;
		return 0;
	} else if (S_ISREG(st.st_mode)) {
		f->max_blocks = st.st_size / BS;
		return 0;
	}

	return -1;
}

static int reap_events(struct submitter *s)
{
	struct io_cq_ring *ring = &s->cq_ring;
	struct io_uring_cqe *cqe;
	unsigned head, reaped = 0;

	head = *ring->head;
	do {
		struct file *f;

		barrier();
		if (head == *ring->tail)
			break;
		cqe = &ring->cqes[head & cq_ring_mask];
		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++;
		else
			s->cachemiss++;
		reaped++;
		head++;
	} while (1);

	s->inflight -= reaped;
	*ring->head = head;
	barrier();
	return reaped;
}

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());

	srand48_r(pthread_self(), &s->rand);

	prepped = 0;
	do {
		int to_wait, to_submit, this_reap, to_prep;

		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 (s->inflight + BATCH_SUBMIT < DEPTH)
			to_wait = 0;
		else
			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)) {
			ret = io_uring_enter(s, to_submit, to_wait,
						IORING_ENTER_GETEVENTS);
			s->calls++;
		}

		/*
		 * 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) {
			if (!ret) {
				to_submit = 0;
				if (s->inflight)
					goto submit;
				continue;
			} else if (ret < to_submit) {
				int diff = to_submit - ret;

				s->done += ret;
				prepped -= diff;
				goto submit_more;
			}
			s->done += ret;
			prepped = 0;
			continue;
		} else if (ret < 0) {
			if (errno == EAGAIN) {
				if (s->finish)
					break;
				if (this_reap)
					goto submit;
				to_submit = 0;
				goto submit;
			}
			printf("io_submit: %s\n", strerror(errno));
			break;
		}
	} while (!s->finish);

	finish = 1;
	return NULL;
}

static void sig_int(int sig)
{
	printf("Exiting on signal %d\n", sig);
	submitters[0].finish = 1;
	finish = 1;
}

static void arm_sig_int(void)
{
	struct sigaction act;

	memset(&act, 0, sizeof(act));
	act.sa_handler = sig_int;
	act.sa_flags = SA_RESTART;
	sigaction(SIGINT, &act, NULL);
}

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)
		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;
	}

	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);
	printf("sq_ring ptr = 0x%p\n", ptr);
	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);
	printf("sqes ptr    = 0x%p\n", s->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);
	printf("cq_ring ptr = 0x%p\n", ptr);
	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;
}

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;
	void *ret;

	if (argc < 2) {
		printf("%s: filename\n", argv[0]);
		return 1;
	}

	flags = O_RDONLY | O_NOATIME;
	if (!buffered)
		flags |= O_DIRECT;

	i = 1;
	while (i < argc) {
		struct file *f = &s->files[s->nr_files];

		fd = open(argv[i], flags);
		if (fd < 0) {
			perror("open");
			return 1;
		}
		f->real_fd = fd;
		if (get_file_size(f)) {
			printf("failed getting size of device/file\n");
			return 1;
		}
		if (f->max_blocks <= 1) {
			printf("Zero file/device size?\n");
			return 1;
		}
		f->max_blocks--;

		printf("Added file %s\n", argv[i]);
		s->nr_files++;
		i++;
	}

	rlim.rlim_cur = RLIM_INFINITY;
	rlim.rlim_max = RLIM_INFINITY;
	if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
		perror("setrlimit");
		return 1;
	}

	arm_sig_int();

	for (i = 0; i < DEPTH; i++) {
		void *buf;

		if (posix_memalign(&buf, BS, BS)) {
			printf("failed alloc\n");
			return 1;
		}
		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;
	}
	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);

	pthread_create(&s->thread, NULL, submitter_fn, s);

	cache_hit = cache_miss = 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;

		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;
		}
		if (this_call - calls) {
			rpc = (this_done - done) / (this_call - calls);
			ipc = (this_reap - reap) / (this_call - calls);
		} 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;
		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);
	return 0;
}
Commit	Line	Data
	1	#include <stdio.h>
	2	#include <errno.h>
	3	#include <assert.h>
	4	#include <stdlib.h>
	5	#include <stddef.h>
	6	#include <signal.h>
	7	#include <inttypes.h>
	8
	9	#include <sys/types.h>
	10	#include <sys/stat.h>
	11	#include <sys/ioctl.h>
	12	#include <sys/syscall.h>
	13	#include <sys/resource.h>
	14	#include <sys/mman.h>
	15	#include <sys/uio.h>
	16	#include <linux/fs.h>
	17	#include <fcntl.h>
	18	#include <unistd.h>
	19	#include <string.h>
	20	#include <pthread.h>
	21	#include <sched.h>
	22
	23	#include "../arch/arch.h"
	24	#include "../lib/types.h"
	25	#include "../os/io_uring.h"
	26
	27	#define barrier() __asm__ __volatile__("": : :"memory")
	28
	29	#define min(a, b) ((a < b) ? (a) : (b))
	30
	31	struct io_sq_ring {
	32	unsigned *head;
	33	unsigned *tail;
	34	unsigned *ring_mask;
	35	unsigned *ring_entries;
	36	unsigned *flags;
	37	unsigned *array;
	38	};
	39
	40	struct io_cq_ring {
	41	unsigned *head;
	42	unsigned *tail;
	43	unsigned *ring_mask;
	44	unsigned *ring_entries;
	45	struct io_uring_cqe *cqes;
	46	};
	47
	48	#define DEPTH 128
	49
	50	#define BATCH_SUBMIT 64
	51	#define BATCH_COMPLETE 64
	52
	53	#define BS 4096
	54
	55	#define MAX_FDS 16
	56
	57	static unsigned sq_ring_mask, cq_ring_mask;
	58
	59	struct file {
	60	unsigned long max_blocks;
	61	unsigned pending_ios;
	62	int real_fd;
	63	int fixed_fd;
	64	};
	65
	66	struct submitter {
	67	pthread_t thread;
	68	int ring_fd;
	69	struct drand48_data rand;
	70	struct io_sq_ring sq_ring;
	71	struct io_uring_sqe *sqes;
	72	struct iovec iovecs[DEPTH];
	73	struct io_cq_ring cq_ring;
	74	int inflight;
	75	unsigned long reaps;
	76	unsigned long done;
	77	unsigned long calls;
	78	unsigned long cachehit, cachemiss;
	79	volatile int finish;
	80
	81	__s32 *fds;
	82
	83	struct file files[MAX_FDS];
	84	unsigned nr_files;
	85	unsigned cur_file;
	86	};
	87
	88	static struct submitter submitters[1];
	89	static volatile int finish;
	90
	91	static int polled = 1; /* use IO polling */
	92	static int fixedbufs = 1; /* use fixed user buffers */
	93	static int buffered = 0; /* use buffered IO, not O_DIRECT */
	94	static int sq_thread_poll = 0; /* use kernel submission/poller thread */
	95	static int sq_thread_cpu = -1; /* pin above thread to this CPU */
	96
	97	static int io_uring_register_buffers(struct submitter *s)
	98	{
	99	struct io_uring_register_buffers reg = {
	100	.iovecs = s->iovecs,
	101	.nr_iovecs = DEPTH
	102	};
	103
	104	return syscall(__NR_sys_io_uring_register, s->ring_fd,
	105	IORING_REGISTER_BUFFERS, &reg);
	106	}
	107
	108	static int io_uring_register_files(struct submitter *s)
	109	{
	110	struct io_uring_register_files reg;
	111	int i;
	112
	113	s->fds = calloc(s->nr_files, sizeof(__s32));
	114	for (i = 0; i < s->nr_files; i++) {
	115	s->fds[i] = s->files[i].real_fd;
	116	s->files[i].fixed_fd = i;
	117	}
	118	reg.fds = s->fds;
	119	reg.nr_fds = s->nr_files;
	120
	121	return syscall(__NR_sys_io_uring_register, s->ring_fd,
	122	IORING_REGISTER_FILES, &reg);
	123	}
	124
	125	static int io_uring_setup(unsigned entries, struct io_uring_params *p)
	126	{
	127	return syscall(__NR_sys_io_uring_setup, entries, p);
	128	}
	129
	130	static int io_uring_enter(struct submitter *s, unsigned int to_submit,
	131	unsigned int min_complete, unsigned int flags)
	132	{
	133	return syscall(__NR_sys_io_uring_enter, s->ring_fd, to_submit,
	134	min_complete, flags);
	135	}
	136
	137	static int gettid(void)
	138	{
	139	return syscall(__NR_gettid);
	140	}
	141
	142	static unsigned file_depth(struct submitter *s)
	143	{
	144	return (DEPTH + s->nr_files - 1) / s->nr_files;
	145	}
	146
	147	static void init_io(struct submitter *s, unsigned index)
	148	{
	149	struct io_uring_sqe *sqe = &s->sqes[index];
	150	unsigned long offset;
	151	struct file *f;
	152	long r;
	153
	154	if (s->nr_files == 1) {
	155	f = &s->files[0];
	156	} else {
	157	f = &s->files[s->cur_file];
	158	if (f->pending_ios >= file_depth(s)) {
	159	s->cur_file++;
	160	if (s->cur_file == s->nr_files)
	161	s->cur_file = 0;
	162	}
	163	}
	164	f->pending_ios++;
	165
	166	lrand48_r(&s->rand, &r);
	167	offset = (r % (f->max_blocks - 1)) * BS;
	168
	169	sqe->flags = IOSQE_FIXED_FILE;
	170	if (fixedbufs) {
	171	sqe->opcode = IORING_OP_READ_FIXED;
	172	sqe->addr = s->iovecs[index].iov_base;
	173	sqe->len = BS;
	174	sqe->buf_index = index;
	175	} else {
	176	sqe->opcode = IORING_OP_READV;
	177	sqe->addr = &s->iovecs[index];
	178	sqe->len = 1;
	179	sqe->buf_index = 0;
	180	}
	181	sqe->ioprio = 0;
	182	sqe->fd = f->fixed_fd;
	183	sqe->off = offset;
	184	sqe->user_data = (unsigned long) f;
	185	}
	186
	187	static int prep_more_ios(struct submitter *s, int max_ios)
	188	{
	189	struct io_sq_ring *ring = &s->sq_ring;
	190	unsigned index, tail, next_tail, prepped = 0;
	191
	192	next_tail = tail = *ring->tail;
	193	do {
	194	next_tail++;
	195	barrier();
	196	if (next_tail == *ring->head)
	197	break;
	198
	199	index = tail & sq_ring_mask;
	200	init_io(s, index);
	201	ring->array[index] = index;
	202	prepped++;
	203	tail = next_tail;
	204	} while (prepped < max_ios);
	205
	206	if (*ring->tail != tail) {
	207	/* order tail store with writes to sqes above */
	208	barrier();
	209	*ring->tail = tail;
	210	barrier();
	211	}
	212	return prepped;
	213	}
	214
	215	static int get_file_size(struct file *f)
	216	{
	217	struct stat st;
	218
	219	if (fstat(f->real_fd, &st) < 0)
	220	return -1;
	221	if (S_ISBLK(st.st_mode)) {
	222	unsigned long long bytes;
	223
	224	if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0)
	225	return -1;
	226
	227	f->max_blocks = bytes / BS;
	228	return 0;
	229	} else if (S_ISREG(st.st_mode)) {
	230	f->max_blocks = st.st_size / BS;
	231	return 0;
	232	}
	233
	234	return -1;
	235	}
	236
	237	static int reap_events(struct submitter *s)
	238	{
	239	struct io_cq_ring *ring = &s->cq_ring;
	240	struct io_uring_cqe *cqe;
	241	unsigned head, reaped = 0;
	242
	243	head = *ring->head;
	244	do {
	245	struct file *f;
	246
	247	barrier();
	248	if (head == *ring->tail)
	249	break;
	250	cqe = &ring->cqes[head & cq_ring_mask];
	251	f = (struct file *) cqe->user_data;
	252	f->pending_ios--;
	253	if (cqe->res != BS) {
	254	printf("io: unexpected ret=%d\n", cqe->res);
	255	return -1;
	256	}
	257	if (cqe->flags & IOCQE_FLAG_CACHEHIT)
	258	s->cachehit++;
	259	else
	260	s->cachemiss++;
	261	reaped++;
	262	head++;
	263	} while (1);
	264
	265	s->inflight -= reaped;
	266	*ring->head = head;
	267	barrier();
	268	return reaped;
	269	}
	270
	271	static void submitter_fn(void data)
	272	{
	273	struct submitter *s = data;
	274	struct io_sq_ring *ring = &s->sq_ring;
	275	int ret, prepped;
	276
	277	printf("submitter=%d\n", gettid());
	278
	279	srand48_r(pthread_self(), &s->rand);
	280
	281	prepped = 0;
	282	do {
	283	int to_wait, to_submit, this_reap, to_prep;
	284
	285	if (!prepped && s->inflight < DEPTH) {
	286	to_prep = min(DEPTH - s->inflight, BATCH_SUBMIT);
	287	prepped = prep_more_ios(s, to_prep);
	288	}
	289	s->inflight += prepped;
	290	submit_more:
	291	to_submit = prepped;
	292	submit:
	293	if (s->inflight + BATCH_SUBMIT < DEPTH)
	294	to_wait = 0;
	295	else
	296	to_wait = min(s->inflight + to_submit, BATCH_COMPLETE);
	297
	298	/*
	299	* Only need to call io_uring_enter if we're not using SQ thread
	300	* poll, or if IORING_SQ_NEED_WAKEUP is set.
	301	*/
	302	if (!sq_thread_poll \|\| (*ring->flags & IORING_SQ_NEED_WAKEUP)) {
	303	ret = io_uring_enter(s, to_submit, to_wait,
	304	IORING_ENTER_GETEVENTS);
	305	s->calls++;
	306	}
	307
	308	/*
	309	* For non SQ thread poll, we already got the events we needed
	310	* through the io_uring_enter() above. For SQ thread poll, we
	311	* need to loop here until we find enough events.
	312	*/
	313	this_reap = 0;
	314	do {
	315	int r;
	316	r = reap_events(s);
	317	if (r == -1)
	318	break;
	319	else if (r > 0)
	320	this_reap += r;
	321	} while (sq_thread_poll && this_reap < to_wait);
	322	s->reaps += this_reap;
	323
	324	if (ret >= 0) {
	325	if (!ret) {
	326	to_submit = 0;
	327	if (s->inflight)
	328	goto submit;
	329	continue;
	330	} else if (ret < to_submit) {
	331	int diff = to_submit - ret;
	332
	333	s->done += ret;
	334	prepped -= diff;
	335	goto submit_more;
	336	}
	337	s->done += ret;
	338	prepped = 0;
	339	continue;
	340	} else if (ret < 0) {
	341	if (errno == EAGAIN) {
	342	if (s->finish)
	343	break;
	344	if (this_reap)
	345	goto submit;
	346	to_submit = 0;
	347	goto submit;
	348	}
	349	printf("io_submit: %s\n", strerror(errno));
	350	break;
	351	}
	352	} while (!s->finish);
	353
	354	finish = 1;
	355	return NULL;
	356	}
	357
	358	static void sig_int(int sig)
	359	{
	360	printf("Exiting on signal %d\n", sig);
	361	submitters[0].finish = 1;
	362	finish = 1;
	363	}
	364
	365	static void arm_sig_int(void)
	366	{
	367	struct sigaction act;
	368
	369	memset(&act, 0, sizeof(act));
	370	act.sa_handler = sig_int;
	371	act.sa_flags = SA_RESTART;
	372	sigaction(SIGINT, &act, NULL);
	373	}
	374
	375	static int setup_ring(struct submitter *s)
	376	{
	377	struct io_sq_ring *sring = &s->sq_ring;
	378	struct io_cq_ring *cring = &s->cq_ring;
	379	struct io_uring_params p;
	380	int ret, fd;
	381	void *ptr;
	382
	383	memset(&p, 0, sizeof(p));
	384
	385	if (polled)
	386	p.flags \|= IORING_SETUP_IOPOLL;
	387	if (sq_thread_poll) {
	388	p.flags \|= IORING_SETUP_SQPOLL;
	389	if (sq_thread_cpu != -1)
	390	p.flags \|= IORING_SETUP_SQ_AFF;
	391	}
	392
	393	fd = io_uring_setup(DEPTH, &p);
	394	if (fd < 0) {
	395	perror("io_uring_setup");
	396	return 1;
	397	}
	398	s->ring_fd = fd;
	399
	400	if (fixedbufs) {
	401	ret = io_uring_register_buffers(s);
	402	if (ret < 0) {
	403	perror("io_uring_register_buffers");
	404	return 1;
	405	}
	406	}
	407
	408	ret = io_uring_register_files(s);
	409	if (ret < 0) {
	410	perror("io_uring_register_files");
	411	return 1;
	412	}
	413
	414	ptr = mmap(0, p.sq_off.array + p.sq_entries * sizeof(__u32),
	415	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
	416	IORING_OFF_SQ_RING);
	417	printf("sq_ring ptr = 0x%p\n", ptr);
	418	sring->head = ptr + p.sq_off.head;
	419	sring->tail = ptr + p.sq_off.tail;
	420	sring->ring_mask = ptr + p.sq_off.ring_mask;
	421	sring->ring_entries = ptr + p.sq_off.ring_entries;
	422	sring->flags = ptr + p.sq_off.flags;
	423	sring->array = ptr + p.sq_off.array;
	424	sq_ring_mask = *sring->ring_mask;
	425
	426	s->sqes = mmap(0, p.sq_entries * sizeof(struct io_uring_sqe),
	427	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
	428	IORING_OFF_SQES);
	429	printf("sqes ptr = 0x%p\n", s->sqes);
	430
	431	ptr = mmap(0, p.cq_off.cqes + p.cq_entries * sizeof(struct io_uring_cqe),
	432	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
	433	IORING_OFF_CQ_RING);
	434	printf("cq_ring ptr = 0x%p\n", ptr);
	435	cring->head = ptr + p.cq_off.head;
	436	cring->tail = ptr + p.cq_off.tail;
	437	cring->ring_mask = ptr + p.cq_off.ring_mask;
	438	cring->ring_entries = ptr + p.cq_off.ring_entries;
	439	cring->cqes = ptr + p.cq_off.cqes;
	440	cq_ring_mask = *cring->ring_mask;
	441	return 0;
	442	}
	443
	444	int main(int argc, char *argv[])
	445	{
	446	struct submitter *s = &submitters[0];
	447	unsigned long done, calls, reap, cache_hit, cache_miss;
	448	int err, i, flags, fd;
	449	struct rlimit rlim;
	450	void *ret;
	451
	452	if (argc < 2) {
	453	printf("%s: filename\n", argv[0]);
	454	return 1;
	455	}
	456
	457	flags = O_RDONLY \| O_NOATIME;
	458	if (!buffered)
	459	flags \|= O_DIRECT;
	460
	461	i = 1;
	462	while (i < argc) {
	463	struct file *f = &s->files[s->nr_files];
	464
	465	fd = open(argv[i], flags);
	466	if (fd < 0) {
	467	perror("open");
	468	return 1;
	469	}
	470	f->real_fd = fd;
	471	if (get_file_size(f)) {
	472	printf("failed getting size of device/file\n");
	473	return 1;
	474	}
	475	if (f->max_blocks <= 1) {
	476	printf("Zero file/device size?\n");
	477	return 1;
	478	}
	479	f->max_blocks--;
	480
	481	printf("Added file %s\n", argv[i]);
	482	s->nr_files++;
	483	i++;
	484	}
	485
	486	rlim.rlim_cur = RLIM_INFINITY;
	487	rlim.rlim_max = RLIM_INFINITY;
	488	if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
	489	perror("setrlimit");
	490	return 1;
	491	}
	492
	493	arm_sig_int();
	494
	495	for (i = 0; i < DEPTH; i++) {
	496	void *buf;
	497
	498	if (posix_memalign(&buf, BS, BS)) {
	499	printf("failed alloc\n");
	500	return 1;
	501	}
	502	s->iovecs[i].iov_base = buf;
	503	s->iovecs[i].iov_len = BS;
	504	}
	505
	506	err = setup_ring(s);
	507	if (err) {
	508	printf("ring setup failed: %s, %d\n", strerror(errno), err);
	509	return 1;
	510	}
	511	printf("polled=%d, fixedbufs=%d, buffered=%d", polled, fixedbufs, buffered);
	512	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", DEPTH, s->sq_ring.ring_entries, s->cq_ring.ring_entries);
	513
	514	pthread_create(&s->thread, NULL, submitter_fn, s);
	515
	516	cache_hit = cache_miss = reap = calls = done = 0;
	517	do {
	518	unsigned long this_done = 0;
	519	unsigned long this_reap = 0;
	520	unsigned long this_call = 0;
	521	unsigned long this_cache_hit = 0;
	522	unsigned long this_cache_miss = 0;
	523	unsigned long rpc = 0, ipc = 0;
	524	double hit = 0.0;
	525
	526	sleep(1);
	527	this_done += s->done;
	528	this_call += s->calls;
	529	this_reap += s->reaps;
	530	this_cache_hit += s->cachehit;
	531	this_cache_miss += s->cachemiss;
	532	if (this_cache_hit && this_cache_miss) {
	533	unsigned long hits, total;
	534
	535	hits = this_cache_hit - cache_hit;
	536	total = hits + this_cache_miss - cache_miss;
	537	hit = (double) hits / (double) total;
	538	hit *= 100.0;
	539	}
	540	if (this_call - calls) {
	541	rpc = (this_done - done) / (this_call - calls);
	542	ipc = (this_reap - reap) / (this_call - calls);
	543	} else
	544	rpc = ipc = -1;
	545	printf("IOPS=%lu, IOS/call=%ld/%ld, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n",
	546	this_done - done, rpc, ipc, s->inflight,
	547	s->cq_ring.head, s->cq_ring.tail, hit);
	548	done = this_done;
	549	calls = this_call;
	550	reap = this_reap;
	551	cache_hit = s->cachehit;
	552	cache_miss = s->cachemiss;
	553	} while (!finish);
	554
	555	pthread_join(s->thread, &ret);
	556	close(s->ring_fd);
	557	return 0;
	558	}