[fio.git] / t / aio-ring.c

/*
 * gcc -D_GNU_SOURCE -Wall -O2 -o aio-ring aio-ring.c  -lpthread -laio
 */
#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 <linux/fs.h>
#include <fcntl.h>
#include <unistd.h>
#include <libaio.h>
#include <string.h>
#include <pthread.h>
#include <sched.h>

#define IOCTX_FLAG_SCQRING	(1 << 0)	/* Use SQ/CQ rings */
#define IOCTX_FLAG_IOPOLL	(1 << 1)
#define IOCTX_FLAG_FIXEDBUFS	(1 << 2)
#define IOCTX_FLAG_SQTHREAD	(1 << 3)	/* Use SQ thread */
#define IOCTX_FLAG_SQWQ		(1 << 4)	/* Use SQ wq */
#define IOCTX_FLAG_SQPOLL	(1 << 5)

#define IOEV_RES2_CACHEHIT	(1 << 0)

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

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

typedef uint64_t u64;
typedef uint32_t u32;
typedef uint16_t u16;

#define IORING_OFF_SQ_RING	0ULL
#define IORING_OFF_CQ_RING	0x8000000ULL
#define IORING_OFF_IOCB		0x10000000ULL

struct aio_uring_offsets {
	u32 head;
	u32 tail;
	u32 ring_mask;
	u32 ring_entries;
	u32 flags;
	u32 elems;
};

struct aio_uring_params {
	u32 sq_entries;
	u32 cq_entries;
	u32 flags;
	u16 sq_thread_cpu;
	u16 resv[9];
	struct aio_uring_offsets sq_off;
	struct aio_uring_offsets cq_off;
};

struct aio_sq_ring {
	u32 *head;
	u32 *tail;
	u32 *ring_mask;
	u32 *ring_entries;
	u32 *array;
};

struct aio_cq_ring {
	u32 *head;
	u32 *tail;
	u32 *ring_mask;
	u32 *ring_entries;
	struct io_event *events;
};

#define IORING_ENTER_GETEVENTS	(1 << 0)

#define DEPTH			32

#define BATCH_SUBMIT		8
#define BATCH_COMPLETE		8

#define BS			4096

static unsigned sq_ring_mask, cq_ring_mask;

struct submitter {
	pthread_t thread;
	unsigned long max_blocks;
	int fd;
	struct drand48_data rand;
	struct aio_sq_ring sq_ring;
	struct iocb *iocbs;
	struct iovec iovecs[DEPTH];
	struct aio_cq_ring cq_ring;
	int inflight;
	unsigned long reaps;
	unsigned long done;
	unsigned long calls;
	unsigned long cachehit, cachemiss;
	volatile int finish;
	char filename[128];
};

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

static int polled = 0;		/* use IO polling */
static int fixedbufs = 0;	/* use fixed user buffers */
static int buffered = 1;	/* 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 io_uring_setup(unsigned entries, struct iovec *iovecs,
			  struct aio_uring_params *p)
{
	return syscall(335, entries, iovecs, p);
}

static int io_uring_enter(struct submitter *s, unsigned int to_submit,
			  unsigned int min_complete, unsigned int flags)
{
	return syscall(336, s->fd, to_submit, min_complete, flags);
}

static int gettid(void)
{
	return syscall(__NR_gettid);
}

static void init_io(struct submitter *s, int fd, unsigned index)
{
	struct iocb *iocb = &s->iocbs[index];
	unsigned long offset;
	long r;

	lrand48_r(&s->rand, &r);
	offset = (r % (s->max_blocks - 1)) * BS;

	iocb->aio_fildes = fd;
	iocb->aio_lio_opcode = IO_CMD_PREAD;
	iocb->u.c.buf = s->iovecs[index].iov_base;
	iocb->u.c.nbytes = BS;
	iocb->u.c.offset = offset;
}

static int prep_more_ios(struct submitter *s, int fd, int max_ios)
{
	struct aio_sq_ring *ring = &s->sq_ring;
	u32 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, fd, index);
		ring->array[index] = index;
		prepped++;
		tail = next_tail;
	} while (prepped < max_ios);

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

static int get_file_size(int fd, unsigned long *blocks)
{
	struct stat st;

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

		if (ioctl(fd, BLKGETSIZE64, &bytes) != 0)
			return -1;

		*blocks = bytes / BS;
		return 0;
	} else if (S_ISREG(st.st_mode)) {
		*blocks = st.st_size / BS;
		return 0;
	}

	return -1;
}

static int reap_events(struct submitter *s)
{
	struct aio_cq_ring *ring = &s->cq_ring;
	struct io_event *ev;
	u32 head, reaped = 0;

	head = *ring->head;
	do {
		barrier();
		if (head == *ring->tail)
			break;
		ev = &ring->events[head & cq_ring_mask];
		if (ev->res != BS) {
			struct iocb *iocb = ev->obj;

			printf("io: unexpected ret=%ld\n", ev->res);
			printf("offset=%lu, size=%lu\n", (unsigned long) iocb->u.c.offset, (unsigned long) iocb->u.c.nbytes);
			return -1;
		}
		if (ev->res2 & IOEV_RES2_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;
	int fd, ret, prepped, flags;

	printf("submitter=%d\n", gettid());

	flags = O_RDONLY;
	if (!buffered)
		flags |= O_DIRECT;
	fd = open(s->filename, flags);
	if (fd < 0) {
		perror("open");
		goto done;
	}

	if (get_file_size(fd, &s->max_blocks)) {
		printf("failed getting size of device/file\n");
		goto err;
	}
	if (!s->max_blocks) {
		printf("Zero file/device size?\n");
		goto err;
	}

	s->max_blocks--;

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

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

		if (!prepped && s->inflight < DEPTH)
			prepped = prep_more_ios(s, fd, min(DEPTH - s->inflight, BATCH_SUBMIT));
		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);

		ret = io_uring_enter(s, to_submit, to_wait, IORING_ENTER_GETEVENTS);
		s->calls++;

		this_reap = reap_events(s);
		if (this_reap == -1)
			break;
		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 ((ret == -1 && errno == EAGAIN) || ret == -EAGAIN) {
				if (s->finish)
					break;
				if (this_reap)
					goto submit;
				to_submit = 0;
				goto submit;
			}
			if (ret == -1)
				printf("io_submit: %s\n", strerror(errno));
			else
				printf("io_submit: %s\n", strerror(-ret));
			break;
		}
	} while (!s->finish);
err:
	close(fd);
done:
	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 aio_sq_ring *sring = &s->sq_ring;
	struct aio_cq_ring *cring = &s->cq_ring;
	struct aio_uring_params p;
	void *ptr;
	int fd;

	memset(&p, 0, sizeof(p));

	p.flags = IOCTX_FLAG_SCQRING;
	if (polled)
		p.flags |= IOCTX_FLAG_IOPOLL;
	if (fixedbufs)
		p.flags |= IOCTX_FLAG_FIXEDBUFS;
	if (buffered)
		p.flags |= IOCTX_FLAG_SQWQ;
	else if (sq_thread) {
		p.flags |= IOCTX_FLAG_SQTHREAD;
		p.sq_thread_cpu = sq_thread_cpu;
	}

	if (fixedbufs)
		fd = io_uring_setup(DEPTH, s->iovecs, &p);
	else
		fd = io_uring_setup(DEPTH, NULL, &p);
	if (fd < 0) {
		perror("io_uring_setup");
		return 1;
	}

	s->fd = fd;

	ptr = mmap(0, p.sq_off.elems + 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->array = ptr + p.sq_off.elems;
	sq_ring_mask = *sring->ring_mask;

	s->iocbs = mmap(0, p.sq_entries * sizeof(struct iocb), PROT_READ | PROT_WRITE,
			MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_IOCB);
	printf("iocbs ptr   = 0x%p\n", s->iocbs);

	ptr = mmap(0, p.cq_off.elems + p.cq_entries * sizeof(struct io_event),
			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->events = ptr + p.cq_off.elems;
	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;
	struct rlimit rlim;
	void *ret;

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

	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);
	strcpy(s->filename, argv[1]);

	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);
		}
		printf("IOPS=%lu, IOS/call=%lu/%lu, 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);
	return 0;
}
Commit	Line	Data
c9fb4c5b JA	1	/*
	2	* gcc -D_GNU_SOURCE -Wall -O2 -o aio-ring aio-ring.c -lpthread -laio
	3	*/
	4	#include <stdio.h>
	5	#include <errno.h>
	6	#include <assert.h>
	7	#include <stdlib.h>
	8	#include <stddef.h>
	9	#include <signal.h>
	10	#include <inttypes.h>
	11
	12	#include <sys/types.h>
	13	#include <sys/stat.h>
	14	#include <sys/ioctl.h>
	15	#include <sys/syscall.h>
	16	#include <sys/resource.h>
c3e2fc25	17	#include <sys/mman.h>
c9fb4c5b JA	18	#include <linux/fs.h>
	19	#include <fcntl.h>
	20	#include <unistd.h>
	21	#include <libaio.h>
	22	#include <string.h>
	23	#include <pthread.h>
	24	#include <sched.h>
	25
c3e2fc25 JA	26	#define IOCTX_FLAG_SCQRING (1 << 0) /* Use SQ/CQ rings */
c3e2fc25 JA	27	#define IOCTX_FLAG_IOPOLL (1 << 1)
c9fb4c5b	28	#define IOCTX_FLAG_FIXEDBUFS (1 << 2)
478a96a8 JA	29	#define IOCTX_FLAG_SQTHREAD (1 << 3) /* Use SQ thread */
478a96a8 JA	30	#define IOCTX_FLAG_SQWQ (1 << 4) /* Use SQ wq */
c3e2fc25	31	#define IOCTX_FLAG_SQPOLL (1 << 5)
c9fb4c5b	32
305c06ce JA	33	#define IOEV_RES2_CACHEHIT (1 << 0)
305c06ce JA	34
c9fb4c5b JA	35	#define barrier() __asm__ __volatile__("": : :"memory")
	36
	37	#define min(a, b) ((a < b) ? (a) : (b))
	38
fa3e6c25	39	typedef uint64_t u64;
c9fb4c5b	40	typedef uint32_t u32;
478a96a8	41	typedef uint16_t u16;
c9fb4c5b	42
c3e2fc25 JA	43	#define IORING_OFF_SQ_RING 0ULL
	44	#define IORING_OFF_CQ_RING 0x8000000ULL
	45	#define IORING_OFF_IOCB 0x10000000ULL
	46
	47	struct aio_uring_offsets {
	48	u32 head;
	49	u32 tail;
	50	u32 ring_mask;
	51	u32 ring_entries;
	52	u32 flags;
	53	u32 elems;
	54	};
	55
	56	struct aio_uring_params {
	57	u32 sq_entries;
	58	u32 cq_entries;
	59	u32 flags;
	60	u16 sq_thread_cpu;
	61	u16 resv[9];
	62	struct aio_uring_offsets sq_off;
	63	struct aio_uring_offsets cq_off;
	64	};
	65
fa3e6c25	66	struct aio_sq_ring {
c3e2fc25 JA	67	u32 *head;
	68	u32 *tail;
	69	u32 *ring_mask;
	70	u32 *ring_entries;
	71	u32 *array;
c9fb4c5b JA	72	};
c9fb4c5b JA	73
fa3e6c25	74	struct aio_cq_ring {
c3e2fc25 JA	75	u32 *head;
	76	u32 *tail;
	77	u32 *ring_mask;
	78	u32 *ring_entries;
	79	struct io_event *events;
c9fb4c5b JA	80	};
c9fb4c5b JA	81
c3e2fc25	82	#define IORING_ENTER_GETEVENTS (1 << 0)
c9fb4c5b	83
c9fb4c5b	84	#define DEPTH 32
c9fb4c5b JA	85
	86	#define BATCH_SUBMIT 8
	87	#define BATCH_COMPLETE 8
	88
	89	#define BS 4096
	90
c3e2fc25	91	static unsigned sq_ring_mask, cq_ring_mask;
e39c34dc	92
c9fb4c5b JA	93	struct submitter {
	94	pthread_t thread;
	95	unsigned long max_blocks;
c3e2fc25	96	int fd;
c9fb4c5b	97	struct drand48_data rand;
c3e2fc25	98	struct aio_sq_ring sq_ring;
fa3e6c25	99	struct iocb *iocbs;
c3e2fc25 JA	100	struct iovec iovecs[DEPTH];
c3e2fc25 JA	101	struct aio_cq_ring cq_ring;
c9fb4c5b JA	102	int inflight;
	103	unsigned long reaps;
	104	unsigned long done;
	105	unsigned long calls;
305c06ce	106	unsigned long cachehit, cachemiss;
c9fb4c5b JA	107	volatile int finish;
	108	char filename[128];
	109	};
	110
	111	static struct submitter submitters[1];
	112	static volatile int finish;
	113
c3e2fc25 JA	114	static int polled = 0; /* use IO polling */
	115	static int fixedbufs = 0; /* use fixed user buffers */
	116	static int buffered = 1; /* use buffered IO, not O_DIRECT */
0527b23f JA	117	static int sq_thread = 0; /* use kernel submission thread */
0527b23f JA	118	static int sq_thread_cpu = 0; /* pin above thread to this CPU */
c9fb4c5b	119
c3e2fc25 JA	120	static int io_uring_setup(unsigned entries, struct iovec *iovecs,
c3e2fc25 JA	121	struct aio_uring_params *p)
c9fb4c5b	122	{
c3e2fc25	123	return syscall(335, entries, iovecs, p);
c9fb4c5b JA	124	}
c9fb4c5b JA	125
c3e2fc25 JA	126	static int io_uring_enter(struct submitter *s, unsigned int to_submit,
c3e2fc25 JA	127	unsigned int min_complete, unsigned int flags)
c9fb4c5b	128	{
c3e2fc25	129	return syscall(336, s->fd, to_submit, min_complete, flags);
c9fb4c5b JA	130	}
	131
	132	static int gettid(void)
	133	{
	134	return syscall(__NR_gettid);
	135	}
	136
c3e2fc25	137	static void init_io(struct submitter *s, int fd, unsigned index)
c9fb4c5b	138	{
c3e2fc25	139	struct iocb *iocb = &s->iocbs[index];
c9fb4c5b JA	140	unsigned long offset;
	141	long r;
	142
	143	lrand48_r(&s->rand, &r);
	144	offset = (r % (s->max_blocks - 1)) * BS;
	145
	146	iocb->aio_fildes = fd;
	147	iocb->aio_lio_opcode = IO_CMD_PREAD;
c3e2fc25 JA	148	iocb->u.c.buf = s->iovecs[index].iov_base;
c3e2fc25 JA	149	iocb->u.c.nbytes = BS;
c9fb4c5b	150	iocb->u.c.offset = offset;
c9fb4c5b JA	151	}
	152
	153	static int prep_more_ios(struct submitter *s, int fd, int max_ios)
	154	{
c3e2fc25	155	struct aio_sq_ring *ring = &s->sq_ring;
e39c34dc	156	u32 index, tail, next_tail, prepped = 0;
c9fb4c5b	157
c3e2fc25	158	next_tail = tail = *ring->tail;
c9fb4c5b JA	159	do {
c9fb4c5b JA	160	next_tail++;
c9fb4c5b	161	barrier();
c3e2fc25	162	if (next_tail == *ring->head)
c9fb4c5b JA	163	break;
c9fb4c5b JA	164
e39c34dc	165	index = tail & sq_ring_mask;
c3e2fc25 JA	166	init_io(s, fd, index);
c3e2fc25 JA	167	ring->array[index] = index;
c9fb4c5b JA	168	prepped++;
	169	tail = next_tail;
	170	} while (prepped < max_ios);
	171
c3e2fc25	172	if (*ring->tail != tail) {
c9fb4c5b JA	173	/* order tail store with writes to iocbs above */
c9fb4c5b JA	174	barrier();
c3e2fc25	175	*ring->tail = tail;
c9fb4c5b JA	176	barrier();
	177	}
	178	return prepped;
	179	}
	180
	181	static int get_file_size(int fd, unsigned long *blocks)
	182	{
	183	struct stat st;
	184
	185	if (fstat(fd, &st) < 0)
	186	return -1;
	187	if (S_ISBLK(st.st_mode)) {
	188	unsigned long long bytes;
	189
	190	if (ioctl(fd, BLKGETSIZE64, &bytes) != 0)
	191	return -1;
	192
	193	*blocks = bytes / BS;
	194	return 0;
	195	} else if (S_ISREG(st.st_mode)) {
	196	*blocks = st.st_size / BS;
	197	return 0;
	198	}
	199
	200	return -1;
	201	}
	202
	203	static int reap_events(struct submitter *s)
	204	{
c3e2fc25	205	struct aio_cq_ring *ring = &s->cq_ring;
c9fb4c5b JA	206	struct io_event *ev;
	207	u32 head, reaped = 0;
	208
c3e2fc25	209	head = *ring->head;
c9fb4c5b JA	210	do {
c9fb4c5b JA	211	barrier();
c3e2fc25	212	if (head == *ring->tail)
c9fb4c5b	213	break;
e39c34dc	214	ev = &ring->events[head & cq_ring_mask];
c9fb4c5b	215	if (ev->res != BS) {
fa3e6c25	216	struct iocb *iocb = ev->obj;
c9fb4c5b JA	217
	218	printf("io: unexpected ret=%ld\n", ev->res);
	219	printf("offset=%lu, size=%lu\n", (unsigned long) iocb->u.c.offset, (unsigned long) iocb->u.c.nbytes);
	220	return -1;
	221	}
305c06ce JA	222	if (ev->res2 & IOEV_RES2_CACHEHIT)
	223	s->cachehit++;
	224	else
	225	s->cachemiss++;
c9fb4c5b JA	226	reaped++;
c9fb4c5b JA	227	head++;
c9fb4c5b JA	228	} while (1);
	229
	230	s->inflight -= reaped;
c3e2fc25	231	*ring->head = head;
c9fb4c5b JA	232	barrier();
	233	return reaped;
	234	}
	235
	236	static void submitter_fn(void data)
	237	{
	238	struct submitter *s = data;
	239	int fd, ret, prepped, flags;
	240
	241	printf("submitter=%d\n", gettid());
	242
	243	flags = O_RDONLY;
	244	if (!buffered)
	245	flags \|= O_DIRECT;
	246	fd = open(s->filename, flags);
	247	if (fd < 0) {
	248	perror("open");
	249	goto done;
	250	}
	251
	252	if (get_file_size(fd, &s->max_blocks)) {
	253	printf("failed getting size of device/file\n");
	254	goto err;
	255	}
	256	if (!s->max_blocks) {
	257	printf("Zero file/device size?\n");
	258	goto err;
	259	}
	260
	261	s->max_blocks--;
	262
	263	srand48_r(pthread_self(), &s->rand);
	264
	265	prepped = 0;
	266	do {
605d3b6a	267	int to_wait, to_submit, this_reap;
c9fb4c5b JA	268
	269	if (!prepped && s->inflight < DEPTH)
	270	prepped = prep_more_ios(s, fd, min(DEPTH - s->inflight, BATCH_SUBMIT));
	271	s->inflight += prepped;
	272	submit_more:
	273	to_submit = prepped;
	274	submit:
	275	if (s->inflight + BATCH_SUBMIT < DEPTH)
	276	to_wait = 0;
	277	else
	278	to_wait = min(s->inflight + to_submit, BATCH_COMPLETE);
	279
c3e2fc25	280	ret = io_uring_enter(s, to_submit, to_wait, IORING_ENTER_GETEVENTS);
c9fb4c5b JA	281	s->calls++;
	282
	283	this_reap = reap_events(s);
	284	if (this_reap == -1)
	285	break;
	286	s->reaps += this_reap;
	287
	288	if (ret >= 0) {
	289	if (!ret) {
	290	to_submit = 0;
	291	if (s->inflight)
	292	goto submit;
	293	continue;
	294	} else if (ret < to_submit) {
	295	int diff = to_submit - ret;
	296
	297	s->done += ret;
	298	prepped -= diff;
	299	goto submit_more;
	300	}
	301	s->done += ret;
	302	prepped = 0;
	303	continue;
	304	} else if (ret < 0) {
	305	if ((ret == -1 && errno == EAGAIN) \|\| ret == -EAGAIN) {
	306	if (s->finish)
	307	break;
	308	if (this_reap)
	309	goto submit;
c9fb4c5b JA	310	to_submit = 0;
	311	goto submit;
	312	}
	313	if (ret == -1)
	314	printf("io_submit: %s\n", strerror(errno));
	315	else
	316	printf("io_submit: %s\n", strerror(-ret));
	317	break;
	318	}
	319	} while (!s->finish);
	320	err:
	321	close(fd);
	322	done:
	323	finish = 1;
	324	return NULL;
	325	}
	326
	327	static void sig_int(int sig)
	328	{
	329	printf("Exiting on signal %d\n", sig);
	330	submitters[0].finish = 1;
	331	finish = 1;
	332	}
	333
	334	static void arm_sig_int(void)
	335	{
	336	struct sigaction act;
	337
	338	memset(&act, 0, sizeof(act));
	339	act.sa_handler = sig_int;
	340	act.sa_flags = SA_RESTART;
	341	sigaction(SIGINT, &act, NULL);
	342	}
	343
c3e2fc25 JA	344	static int setup_ring(struct submitter *s)
	345	{
	346	struct aio_sq_ring *sring = &s->sq_ring;
	347	struct aio_cq_ring *cring = &s->cq_ring;
	348	struct aio_uring_params p;
	349	void *ptr;
	350	int fd;
	351
	352	memset(&p, 0, sizeof(p));
	353
	354	p.flags = IOCTX_FLAG_SCQRING;
	355	if (polled)
	356	p.flags \|= IOCTX_FLAG_IOPOLL;
	357	if (fixedbufs)
	358	p.flags \|= IOCTX_FLAG_FIXEDBUFS;
	359	if (buffered)
	360	p.flags \|= IOCTX_FLAG_SQWQ;
	361	else if (sq_thread) {
	362	p.flags \|= IOCTX_FLAG_SQTHREAD;
	363	p.sq_thread_cpu = sq_thread_cpu;
	364	}
	365
	366	if (fixedbufs)
	367	fd = io_uring_setup(DEPTH, s->iovecs, &p);
	368	else
	369	fd = io_uring_setup(DEPTH, NULL, &p);
	370	if (fd < 0) {
	371	perror("io_uring_setup");
	372	return 1;
	373	}
	374
	375	s->fd = fd;
	376
	377	ptr = mmap(0, p.sq_off.elems + p.sq_entries * sizeof(u32),
	378	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE,
	379	fd, IORING_OFF_SQ_RING);
	380	printf("sq_ring ptr = 0x%p\n", ptr);
	381	sring->head = ptr + p.sq_off.head;
	382	sring->tail = ptr + p.sq_off.tail;
	383	sring->ring_mask = ptr + p.sq_off.ring_mask;
	384	sring->ring_entries = ptr + p.sq_off.ring_entries;
	385	sring->array = ptr + p.sq_off.elems;
	386	sq_ring_mask = *sring->ring_mask;
	387
	388	s->iocbs = mmap(0, p.sq_entries * sizeof(struct iocb), PROT_READ \| PROT_WRITE,
	389	MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_IOCB);
	390	printf("iocbs ptr = 0x%p\n", s->iocbs);
	391
	392	ptr = mmap(0, p.cq_off.elems + p.cq_entries * sizeof(struct io_event),
	393	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE,
	394	fd, IORING_OFF_CQ_RING);
	395	printf("cq_ring ptr = 0x%p\n", ptr);
	396	cring->head = ptr + p.cq_off.head;
	397	cring->tail = ptr + p.cq_off.tail;
	398	cring->ring_mask = ptr + p.cq_off.ring_mask;
	399	cring->ring_entries = ptr + p.cq_off.ring_entries;
	400	cring->events = ptr + p.cq_off.elems;
	401	cq_ring_mask = *cring->ring_mask;
	402	return 0;
	403	}
	404
c9fb4c5b JA	405	int main(int argc, char *argv[])
	406	{
	407	struct submitter *s = &submitters[0];
305c06ce	408	unsigned long done, calls, reap, cache_hit, cache_miss;
c3e2fc25	409	int err, i;
c9fb4c5b	410	struct rlimit rlim;
c3e2fc25	411	void *ret;
c9fb4c5b JA	412
	413	if (argc < 2) {
	414	printf("%s: filename\n", argv[0]);
	415	return 1;
	416	}
	417
	418	rlim.rlim_cur = RLIM_INFINITY;
	419	rlim.rlim_max = RLIM_INFINITY;
	420	if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
	421	perror("setrlimit");
	422	return 1;
	423	}
	424
	425	arm_sig_int();
	426
c3e2fc25 JA	427	for (i = 0; i < DEPTH; i++) {
c3e2fc25 JA	428	void *buf;
c9fb4c5b	429
c3e2fc25	430	if (posix_memalign(&buf, BS, BS)) {
c9fb4c5b JA	431	printf("failed alloc\n");
	432	return 1;
	433	}
c3e2fc25 JA	434	s->iovecs[i].iov_base = buf;
c3e2fc25 JA	435	s->iovecs[i].iov_len = BS;
c9fb4c5b JA	436	}
c9fb4c5b JA	437
c3e2fc25	438	err = setup_ring(s);
c9fb4c5b	439	if (err) {
c3e2fc25	440	printf("ring setup failed: %s, %d\n", strerror(errno), err);
c9fb4c5b JA	441	return 1;
c9fb4c5b JA	442	}
c3e2fc25 JA	443	printf("polled=%d, fixedbufs=%d, buffered=%d", polled, fixedbufs, buffered);
c3e2fc25 JA	444	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", DEPTH, s->sq_ring.ring_entries, s->cq_ring.ring_entries);
c9fb4c5b JA	445	strcpy(s->filename, argv[1]);
	446
	447	pthread_create(&s->thread, NULL, submitter_fn, s);
	448
305c06ce	449	cache_hit = cache_miss = reap = calls = done = 0;
c9fb4c5b JA	450	do {
	451	unsigned long this_done = 0;
	452	unsigned long this_reap = 0;
	453	unsigned long this_call = 0;
305c06ce JA	454	unsigned long this_cache_hit = 0;
305c06ce JA	455	unsigned long this_cache_miss = 0;
c9fb4c5b	456	unsigned long rpc = 0, ipc = 0;
305c06ce	457	double hit = 0.0;
c9fb4c5b JA	458
	459	sleep(1);
	460	this_done += s->done;
	461	this_call += s->calls;
	462	this_reap += s->reaps;
305c06ce JA	463	this_cache_hit += s->cachehit;
	464	this_cache_miss += s->cachemiss;
	465	if (this_cache_hit && this_cache_miss) {
	466	unsigned long hits, total;
	467
	468	hits = this_cache_hit - cache_hit;
	469	total = hits + this_cache_miss - cache_miss;
	470	hit = (double) hits / (double) total;
	471	hit *= 100.0;
	472	}
c9fb4c5b JA	473	if (this_call - calls) {
	474	rpc = (this_done - done) / (this_call - calls);
	475	ipc = (this_reap - reap) / (this_call - calls);
	476	}
10c4d131	477	printf("IOPS=%lu, IOS/call=%lu/%lu, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n",
c9fb4c5b	478	this_done - done, rpc, ipc, s->inflight,
c3e2fc25	479	s->cq_ring.head, s->cq_ring.tail, hit);
c9fb4c5b JA	480	done = this_done;
	481	calls = this_call;
	482	reap = this_reap;
305c06ce JA	483	cache_hit = s->cachehit;
305c06ce JA	484	cache_miss = s->cachemiss;
c9fb4c5b JA	485	} while (!finish);
	486
	487	pthread_join(s->thread, &ret);
	488	return 0;
	489	}