[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"

typedef uint64_t u64;
typedef uint32_t u32;
typedef int32_t s32;
typedef uint16_t u16;
typedef uint8_t u8;

#include "../os/io_uring.h"

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

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

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

struct io_cq_ring {
	u32 *head;
	u32 *tail;
	u32 *ring_mask;
	u32 *ring_entries;
	struct io_uring_event *events;
};

#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 ring_fd;
	struct drand48_data rand;
	struct io_sq_ring sq_ring;
	struct io_uring_iocb *iocbs;
	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;
	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 io_uring_params *p)
{
	return syscall(__NR_sys_io_uring_setup, entries, iovecs, 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 void init_io(struct submitter *s, int fd, unsigned index)
{
	struct io_uring_iocb *iocb = &s->iocbs[index];
	unsigned long offset;
	long r;

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

	iocb->opcode = IORING_OP_READ;
	iocb->flags = 0;
	iocb->ioprio = 0;
	iocb->fd = fd;
	iocb->off = offset;
	iocb->addr = s->iovecs[index].iov_base;
	iocb->len = BS;
}

static int prep_more_ios(struct submitter *s, int fd, int max_ios)
{
	struct io_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 io_cq_ring *ring = &s->cq_ring;
	struct io_uring_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 io_uring_iocb *iocb = &s->iocbs[ev->index];

			printf("io: unexpected ret=%d\n", ev->res);
			printf("offset=%lu, size=%lu\n",
					(unsigned long) iocb->off,
					(unsigned long) iocb->len);
			return -1;
		}
		if (ev->flags & IOEV_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;
	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 <= 1) {
		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, to_prep;

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

		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 (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);
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 io_sq_ring *sring = &s->sq_ring;
	struct io_cq_ring *cring = &s->cq_ring;
	struct io_uring_params p;
	void *ptr;
	int fd;

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

	if (polled)
		p.flags |= IORING_SETUP_IOPOLL;
	if (fixedbufs)
		p.flags |= IORING_SETUP_FIXEDBUFS;
	if (buffered)
		p.flags |= IORING_SETUP_SQWQ;
	else if (sq_thread) {
		p.flags |= IORING_SETUP_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->ring_fd = fd;
	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->array = ptr + p.sq_off.array;
	sq_ring_mask = *sring->ring_mask;

	s->iocbs = mmap(0, p.sq_entries * sizeof(struct io_uring_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.events + p.cq_entries * sizeof(struct io_uring_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.events;
	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);
	close(s->ring_fd);
	return 0;
}
Commit	Line	Data
c9fb4c5b JA	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>
c3e2fc25	14	#include <sys/mman.h>
e31b8288	15	#include <sys/uio.h>
c9fb4c5b JA	16	#include <linux/fs.h>
	17	#include <fcntl.h>
	18	#include <unistd.h>
c9fb4c5b JA	19	#include <string.h>
	20	#include <pthread.h>
	21	#include <sched.h>
	22
74efb029 JA	23	#include "../arch/arch.h"
74efb029 JA	24
fa3e6c25	25	typedef uint64_t u64;
c9fb4c5b	26	typedef uint32_t u32;
e31b8288	27	typedef int32_t s32;
478a96a8	28	typedef uint16_t u16;
e31b8288	29	typedef uint8_t u8;
c9fb4c5b	30
e31b8288	31	#include "../os/io_uring.h"
ac122fea	32
e31b8288	33	#define barrier() __asm__ __volatile__("": : :"memory")
c3e2fc25	34
e31b8288	35	#define min(a, b) ((a < b) ? (a) : (b))
c3e2fc25	36
e31b8288	37	struct io_sq_ring {
c3e2fc25 JA	38	u32 *head;
	39	u32 *tail;
	40	u32 *ring_mask;
	41	u32 *ring_entries;
	42	u32 *array;
c9fb4c5b JA	43	};
c9fb4c5b JA	44
e31b8288	45	struct io_cq_ring {
c3e2fc25 JA	46	u32 *head;
	47	u32 *tail;
	48	u32 *ring_mask;
	49	u32 *ring_entries;
e31b8288	50	struct io_uring_event *events;
c9fb4c5b JA	51	};
c9fb4c5b JA	52
c9fb4c5b	53	#define DEPTH 32
c9fb4c5b JA	54
	55	#define BATCH_SUBMIT 8
	56	#define BATCH_COMPLETE 8
	57
	58	#define BS 4096
	59
c3e2fc25	60	static unsigned sq_ring_mask, cq_ring_mask;
e39c34dc	61
c9fb4c5b JA	62	struct submitter {
	63	pthread_t thread;
	64	unsigned long max_blocks;
f310970e	65	int ring_fd;
c9fb4c5b	66	struct drand48_data rand;
e31b8288 JA	67	struct io_sq_ring sq_ring;
e31b8288 JA	68	struct io_uring_iocb *iocbs;
c3e2fc25	69	struct iovec iovecs[DEPTH];
e31b8288	70	struct io_cq_ring cq_ring;
c9fb4c5b JA	71	int inflight;
	72	unsigned long reaps;
	73	unsigned long done;
	74	unsigned long calls;
305c06ce	75	unsigned long cachehit, cachemiss;
c9fb4c5b JA	76	volatile int finish;
	77	char filename[128];
	78	};
	79
	80	static struct submitter submitters[1];
	81	static volatile int finish;
	82
c3e2fc25 JA	83	static int polled = 0; /* use IO polling */
	84	static int fixedbufs = 0; /* use fixed user buffers */
	85	static int buffered = 1; /* use buffered IO, not O_DIRECT */
0527b23f JA	86	static int sq_thread = 0; /* use kernel submission thread */
0527b23f JA	87	static int sq_thread_cpu = 0; /* pin above thread to this CPU */
c9fb4c5b	88
c3e2fc25	89	static int io_uring_setup(unsigned entries, struct iovec *iovecs,
e31b8288	90	struct io_uring_params *p)
c9fb4c5b	91	{
74efb029	92	return syscall(__NR_sys_io_uring_setup, entries, iovecs, p);
c9fb4c5b JA	93	}
c9fb4c5b JA	94
c3e2fc25 JA	95	static int io_uring_enter(struct submitter *s, unsigned int to_submit,
c3e2fc25 JA	96	unsigned int min_complete, unsigned int flags)
c9fb4c5b	97	{
f310970e JA	98	return syscall(__NR_sys_io_uring_enter, s->ring_fd, to_submit,
f310970e JA	99	min_complete, flags);
c9fb4c5b JA	100	}
	101
	102	static int gettid(void)
	103	{
	104	return syscall(__NR_gettid);
	105	}
	106
c3e2fc25	107	static void init_io(struct submitter *s, int fd, unsigned index)
c9fb4c5b	108	{
e31b8288	109	struct io_uring_iocb *iocb = &s->iocbs[index];
c9fb4c5b JA	110	unsigned long offset;
	111	long r;
	112
	113	lrand48_r(&s->rand, &r);
	114	offset = (r % (s->max_blocks - 1)) * BS;
	115
e31b8288 JA	116	iocb->opcode = IORING_OP_READ;
	117	iocb->flags = 0;
	118	iocb->ioprio = 0;
	119	iocb->fd = fd;
	120	iocb->off = offset;
	121	iocb->addr = s->iovecs[index].iov_base;
	122	iocb->len = BS;
c9fb4c5b JA	123	}
	124
	125	static int prep_more_ios(struct submitter *s, int fd, int max_ios)
	126	{
e31b8288	127	struct io_sq_ring *ring = &s->sq_ring;
e39c34dc	128	u32 index, tail, next_tail, prepped = 0;
c9fb4c5b	129
c3e2fc25	130	next_tail = tail = *ring->tail;
c9fb4c5b JA	131	do {
c9fb4c5b JA	132	next_tail++;
c9fb4c5b	133	barrier();
c3e2fc25	134	if (next_tail == *ring->head)
c9fb4c5b JA	135	break;
c9fb4c5b JA	136
e39c34dc	137	index = tail & sq_ring_mask;
c3e2fc25 JA	138	init_io(s, fd, index);
c3e2fc25 JA	139	ring->array[index] = index;
c9fb4c5b JA	140	prepped++;
	141	tail = next_tail;
	142	} while (prepped < max_ios);
	143
c3e2fc25	144	if (*ring->tail != tail) {
c9fb4c5b JA	145	/* order tail store with writes to iocbs above */
c9fb4c5b JA	146	barrier();
c3e2fc25	147	*ring->tail = tail;
c9fb4c5b JA	148	barrier();
	149	}
	150	return prepped;
	151	}
	152
	153	static int get_file_size(int fd, unsigned long *blocks)
	154	{
	155	struct stat st;
	156
	157	if (fstat(fd, &st) < 0)
	158	return -1;
	159	if (S_ISBLK(st.st_mode)) {
	160	unsigned long long bytes;
	161
	162	if (ioctl(fd, BLKGETSIZE64, &bytes) != 0)
	163	return -1;
	164
	165	*blocks = bytes / BS;
	166	return 0;
	167	} else if (S_ISREG(st.st_mode)) {
	168	*blocks = st.st_size / BS;
	169	return 0;
	170	}
	171
	172	return -1;
	173	}
	174
	175	static int reap_events(struct submitter *s)
	176	{
e31b8288 JA	177	struct io_cq_ring *ring = &s->cq_ring;
e31b8288 JA	178	struct io_uring_event *ev;
c9fb4c5b JA	179	u32 head, reaped = 0;
c9fb4c5b JA	180
c3e2fc25	181	head = *ring->head;
c9fb4c5b JA	182	do {
c9fb4c5b JA	183	barrier();
c3e2fc25	184	if (head == *ring->tail)
c9fb4c5b	185	break;
e39c34dc	186	ev = &ring->events[head & cq_ring_mask];
c9fb4c5b	187	if (ev->res != BS) {
e31b8288	188	struct io_uring_iocb *iocb = &s->iocbs[ev->index];
c9fb4c5b	189
e31b8288	190	printf("io: unexpected ret=%d\n", ev->res);
f310970e	191	printf("offset=%lu, size=%lu\n",
e31b8288 JA	192	(unsigned long) iocb->off,
e31b8288 JA	193	(unsigned long) iocb->len);
c9fb4c5b JA	194	return -1;
c9fb4c5b JA	195	}
e31b8288	196	if (ev->flags & IOEV_FLAG_CACHEHIT)
305c06ce JA	197	s->cachehit++;
	198	else
	199	s->cachemiss++;
c9fb4c5b JA	200	reaped++;
c9fb4c5b JA	201	head++;
c9fb4c5b JA	202	} while (1);
	203
	204	s->inflight -= reaped;
c3e2fc25	205	*ring->head = head;
c9fb4c5b JA	206	barrier();
	207	return reaped;
	208	}
	209
	210	static void submitter_fn(void data)
	211	{
	212	struct submitter *s = data;
	213	int fd, ret, prepped, flags;
	214
	215	printf("submitter=%d\n", gettid());
	216
	217	flags = O_RDONLY;
	218	if (!buffered)
	219	flags \|= O_DIRECT;
	220	fd = open(s->filename, flags);
	221	if (fd < 0) {
	222	perror("open");
	223	goto done;
	224	}
	225
	226	if (get_file_size(fd, &s->max_blocks)) {
	227	printf("failed getting size of device/file\n");
	228	goto err;
	229	}
f310970e	230	if (s->max_blocks <= 1) {
c9fb4c5b JA	231	printf("Zero file/device size?\n");
	232	goto err;
	233	}
c9fb4c5b JA	234	s->max_blocks--;
	235
	236	srand48_r(pthread_self(), &s->rand);
	237
	238	prepped = 0;
	239	do {
f310970e	240	int to_wait, to_submit, this_reap, to_prep;
c9fb4c5b	241
f310970e JA	242	if (!prepped && s->inflight < DEPTH) {
	243	to_prep = min(DEPTH - s->inflight, BATCH_SUBMIT);
	244	prepped = prep_more_ios(s, fd, to_prep);
	245	}
c9fb4c5b JA	246	s->inflight += prepped;
	247	submit_more:
	248	to_submit = prepped;
	249	submit:
	250	if (s->inflight + BATCH_SUBMIT < DEPTH)
	251	to_wait = 0;
	252	else
	253	to_wait = min(s->inflight + to_submit, BATCH_COMPLETE);
	254
f310970e JA	255	ret = io_uring_enter(s, to_submit, to_wait,
f310970e JA	256	IORING_ENTER_GETEVENTS);
c9fb4c5b JA	257	s->calls++;
	258
	259	this_reap = reap_events(s);
	260	if (this_reap == -1)
	261	break;
	262	s->reaps += this_reap;
	263
	264	if (ret >= 0) {
	265	if (!ret) {
	266	to_submit = 0;
	267	if (s->inflight)
	268	goto submit;
	269	continue;
	270	} else if (ret < to_submit) {
	271	int diff = to_submit - ret;
	272
	273	s->done += ret;
	274	prepped -= diff;
	275	goto submit_more;
	276	}
	277	s->done += ret;
	278	prepped = 0;
	279	continue;
	280	} else if (ret < 0) {
ac122fea	281	if (errno == EAGAIN) {
c9fb4c5b JA	282	if (s->finish)
	283	break;
	284	if (this_reap)
	285	goto submit;
c9fb4c5b JA	286	to_submit = 0;
	287	goto submit;
	288	}
ac122fea	289	printf("io_submit: %s\n", strerror(errno));
c9fb4c5b JA	290	break;
	291	}
	292	} while (!s->finish);
	293	err:
	294	close(fd);
	295	done:
	296	finish = 1;
	297	return NULL;
	298	}
	299
	300	static void sig_int(int sig)
	301	{
	302	printf("Exiting on signal %d\n", sig);
	303	submitters[0].finish = 1;
	304	finish = 1;
	305	}
	306
	307	static void arm_sig_int(void)
	308	{
	309	struct sigaction act;
	310
	311	memset(&act, 0, sizeof(act));
	312	act.sa_handler = sig_int;
	313	act.sa_flags = SA_RESTART;
	314	sigaction(SIGINT, &act, NULL);
	315	}
	316
c3e2fc25 JA	317	static int setup_ring(struct submitter *s)
c3e2fc25 JA	318	{
e31b8288 JA	319	struct io_sq_ring *sring = &s->sq_ring;
	320	struct io_cq_ring *cring = &s->cq_ring;
	321	struct io_uring_params p;
c3e2fc25 JA	322	void *ptr;
	323	int fd;
	324
	325	memset(&p, 0, sizeof(p));
	326
c3e2fc25	327	if (polled)
e31b8288	328	p.flags \|= IORING_SETUP_IOPOLL;
c3e2fc25	329	if (fixedbufs)
e31b8288	330	p.flags \|= IORING_SETUP_FIXEDBUFS;
c3e2fc25	331	if (buffered)
e31b8288	332	p.flags \|= IORING_SETUP_SQWQ;
c3e2fc25	333	else if (sq_thread) {
e31b8288	334	p.flags \|= IORING_SETUP_SQTHREAD;
c3e2fc25 JA	335	p.sq_thread_cpu = sq_thread_cpu;
	336	}
	337
	338	if (fixedbufs)
	339	fd = io_uring_setup(DEPTH, s->iovecs, &p);
	340	else
	341	fd = io_uring_setup(DEPTH, NULL, &p);
	342	if (fd < 0) {
	343	perror("io_uring_setup");
	344	return 1;
	345	}
	346
f310970e	347	s->ring_fd = fd;
ac122fea	348	ptr = mmap(0, p.sq_off.array + p.sq_entries * sizeof(u32),
f310970e JA	349	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
f310970e JA	350	IORING_OFF_SQ_RING);
c3e2fc25 JA	351	printf("sq_ring ptr = 0x%p\n", ptr);
	352	sring->head = ptr + p.sq_off.head;
	353	sring->tail = ptr + p.sq_off.tail;
	354	sring->ring_mask = ptr + p.sq_off.ring_mask;
	355	sring->ring_entries = ptr + p.sq_off.ring_entries;
ac122fea	356	sring->array = ptr + p.sq_off.array;
c3e2fc25 JA	357	sq_ring_mask = *sring->ring_mask;
c3e2fc25 JA	358
e31b8288	359	s->iocbs = mmap(0, p.sq_entries * sizeof(struct io_uring_iocb),
f310970e JA	360	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
f310970e JA	361	IORING_OFF_IOCB);
c3e2fc25 JA	362	printf("iocbs ptr = 0x%p\n", s->iocbs);
c3e2fc25 JA	363
e31b8288	364	ptr = mmap(0, p.cq_off.events + p.cq_entries * sizeof(struct io_uring_event),
f310970e JA	365	PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd,
f310970e JA	366	IORING_OFF_CQ_RING);
c3e2fc25 JA	367	printf("cq_ring ptr = 0x%p\n", ptr);
	368	cring->head = ptr + p.cq_off.head;
	369	cring->tail = ptr + p.cq_off.tail;
	370	cring->ring_mask = ptr + p.cq_off.ring_mask;
	371	cring->ring_entries = ptr + p.cq_off.ring_entries;
ac122fea	372	cring->events = ptr + p.cq_off.events;
c3e2fc25 JA	373	cq_ring_mask = *cring->ring_mask;
	374	return 0;
	375	}
	376
c9fb4c5b JA	377	int main(int argc, char *argv[])
	378	{
	379	struct submitter *s = &submitters[0];
305c06ce	380	unsigned long done, calls, reap, cache_hit, cache_miss;
c3e2fc25	381	int err, i;
c9fb4c5b	382	struct rlimit rlim;
c3e2fc25	383	void *ret;
c9fb4c5b JA	384
	385	if (argc < 2) {
	386	printf("%s: filename\n", argv[0]);
	387	return 1;
	388	}
	389
	390	rlim.rlim_cur = RLIM_INFINITY;
	391	rlim.rlim_max = RLIM_INFINITY;
	392	if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
	393	perror("setrlimit");
	394	return 1;
	395	}
	396
	397	arm_sig_int();
	398
c3e2fc25 JA	399	for (i = 0; i < DEPTH; i++) {
c3e2fc25 JA	400	void *buf;
c9fb4c5b	401
c3e2fc25	402	if (posix_memalign(&buf, BS, BS)) {
c9fb4c5b JA	403	printf("failed alloc\n");
	404	return 1;
	405	}
c3e2fc25 JA	406	s->iovecs[i].iov_base = buf;
c3e2fc25 JA	407	s->iovecs[i].iov_len = BS;
c9fb4c5b JA	408	}
c9fb4c5b JA	409
c3e2fc25	410	err = setup_ring(s);
c9fb4c5b	411	if (err) {
c3e2fc25	412	printf("ring setup failed: %s, %d\n", strerror(errno), err);
c9fb4c5b JA	413	return 1;
c9fb4c5b JA	414	}
c3e2fc25 JA	415	printf("polled=%d, fixedbufs=%d, buffered=%d", polled, fixedbufs, buffered);
c3e2fc25 JA	416	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", DEPTH, s->sq_ring.ring_entries, s->cq_ring.ring_entries);
c9fb4c5b JA	417	strcpy(s->filename, argv[1]);
	418
	419	pthread_create(&s->thread, NULL, submitter_fn, s);
	420
305c06ce	421	cache_hit = cache_miss = reap = calls = done = 0;
c9fb4c5b JA	422	do {
	423	unsigned long this_done = 0;
	424	unsigned long this_reap = 0;
	425	unsigned long this_call = 0;
305c06ce JA	426	unsigned long this_cache_hit = 0;
305c06ce JA	427	unsigned long this_cache_miss = 0;
c9fb4c5b	428	unsigned long rpc = 0, ipc = 0;
305c06ce	429	double hit = 0.0;
c9fb4c5b JA	430
	431	sleep(1);
	432	this_done += s->done;
	433	this_call += s->calls;
	434	this_reap += s->reaps;
305c06ce JA	435	this_cache_hit += s->cachehit;
	436	this_cache_miss += s->cachemiss;
	437	if (this_cache_hit && this_cache_miss) {
	438	unsigned long hits, total;
	439
	440	hits = this_cache_hit - cache_hit;
	441	total = hits + this_cache_miss - cache_miss;
	442	hit = (double) hits / (double) total;
	443	hit *= 100.0;
	444	}
c9fb4c5b JA	445	if (this_call - calls) {
	446	rpc = (this_done - done) / (this_call - calls);
	447	ipc = (this_reap - reap) / (this_call - calls);
	448	}
10c4d131	449	printf("IOPS=%lu, IOS/call=%lu/%lu, inflight=%u (head=%u tail=%u), Cachehit=%0.2f%%\n",
c9fb4c5b	450	this_done - done, rpc, ipc, s->inflight,
c3e2fc25	451	s->cq_ring.head, s->cq_ring.tail, hit);
c9fb4c5b JA	452	done = this_done;
	453	calls = this_call;
	454	reap = this_reap;
305c06ce JA	455	cache_hit = s->cachehit;
305c06ce JA	456	cache_miss = s->cachemiss;
c9fb4c5b JA	457	} while (!finish);
	458
	459	pthread_join(s->thread, &ret);
e31b8288	460	close(s->ring_fd);
c9fb4c5b JA	461	return 0;
c9fb4c5b JA	462	}