[fio.git] / engines / syslet-rw.c

/*
 * syslet engine
 *
 * IO engine that does regular pread(2)/pwrite(2) to transfer data, but
 * with syslets to make the execution async.
 *
 */
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <assert.h>
#include <malloc.h>
#include <asm/unistd.h>

#include "../fio.h"
#include "../indirect.h"
#include "../syslet.h"

#ifdef FIO_HAVE_SYSLET

#ifdef __NR_pread64
#define __NR_fio_pread	__NR_pread64
#define __NR_fio_pwrite	__NR_pwrite64
#else
#define __NR_fio_pread	__NR_pread
#define __NR_fio_pwrite	__NR_pwrite
#endif

struct syslet_data {
	struct io_u **events;
	unsigned int nr_events;
	
	struct syslet_ring *ring;
	unsigned int ring_mask;
	void *stack;
};

static void fio_syslet_add_event(struct thread_data *td, struct io_u *io_u)
{
	struct syslet_data *sd = td->io_ops->data;

	assert(sd->nr_events < td->o.iodepth);
	sd->events[sd->nr_events++] = io_u;
}

static void fio_syslet_add_events(struct thread_data *td, unsigned int nr)
{
	struct syslet_data *sd = td->io_ops->data;
	unsigned int i, uidx;

	uidx = sd->ring->user_tail;
	read_barrier();

	for (i = 0; i < nr; i++) {
		unsigned int idx = (i + uidx) & sd->ring_mask;
		struct syslet_completion *comp = &sd->ring->comp[idx];
		struct io_u *io_u = (struct io_u *) (long) comp->caller_data;
		long ret;

		ret = comp->status;
		if (ret <= 0) {
			io_u->resid = io_u->xfer_buflen;
			io_u->error = -ret;
		} else {
			io_u->resid = io_u->xfer_buflen - ret;
			io_u->error = 0;
		}

		fio_syslet_add_event(td, io_u);
	}
}

static void fio_syslet_wait_for_events(struct thread_data *td)
{
	struct syslet_data *sd = td->io_ops->data;
	struct syslet_ring *ring = sd->ring;
	unsigned int events;

	events = 0;
	do {
		unsigned int kh = ring->kernel_head;
		int ret;

		/*
		 * first reap events that are already completed
		 */
		if (ring->user_tail != kh) {
			unsigned int nr = kh - ring->user_tail;

			fio_syslet_add_events(td, nr);
			events += nr;
			ring->user_tail = kh;
			continue;
		}

		/*
		 * block waiting for at least one event
		 */
		ret = syscall(__NR_syslet_ring_wait, ring, ring->user_tail);
		assert(!ret);
	} while (!events);
}

static int fio_syslet_getevents(struct thread_data *td, int min,
				int fio_unused max,
				struct timespec fio_unused *t)
{
	struct syslet_data *sd = td->io_ops->data;
	long ret;

	do {
		/*
		 * do we have enough immediate completions?
		 */
		if (sd->nr_events >= (unsigned int) min)
			break;

		fio_syslet_wait_for_events(td);
	} while (1);

	ret = sd->nr_events;
	sd->nr_events = 0;
	return ret;
}

static struct io_u *fio_syslet_event(struct thread_data *td, int event)
{
	struct syslet_data *sd = td->io_ops->data;

	return sd->events[event];
}

static void fio_syslet_prep_sync(struct fio_file *f,
				 struct indirect_registers *regs)
{
	FILL_IN(*regs, __NR_fsync, (long) f->fd);
}

static void fio_syslet_prep_rw(struct io_u *io_u, struct fio_file *f,
			       struct indirect_registers *regs)
{
	long nr;

	/*
	 * prepare rw
	 */
	if (io_u->ddir == DDIR_READ)
		nr = __NR_fio_pread;
	else
		nr = __NR_fio_pwrite;

	FILL_IN(*regs, nr, (long) f->fd, (long) io_u->xfer_buf,
		(long) io_u->xfer_buflen, (long) io_u->offset);
}

static void fio_syslet_prep(struct io_u *io_u, struct indirect_registers *regs)
{
	struct fio_file *f = io_u->file;

	if (io_u->ddir == DDIR_SYNC)
		fio_syslet_prep_sync(f, regs);
	else
		fio_syslet_prep_rw(io_u, f, regs);
}

static void ret_func(void)
{
	syscall(__NR_exit);
}

static int fio_syslet_queue(struct thread_data *td, struct io_u *io_u)
{
	struct syslet_data *sd = td->io_ops->data;
	union indirect_params params;
	struct indirect_registers regs;
	int ret;

	fio_ro_check(td, io_u);

	memset(&params, 0, sizeof(params));
	fill_syslet_args(&params.syslet, sd->ring, (long)io_u, ret_func, sd->stack);

	fio_syslet_prep(io_u, &regs);

	ret = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	if (ret == (int) io_u->xfer_buflen) {
		/*
		 * completed sync, account. this also catches fsync().
		 */
		return FIO_Q_COMPLETED;
	} else if (ret < 0) {
		/*
		 * queued for async execution
		 */
		if (errno == ESYSLETPENDING)
			return FIO_Q_QUEUED;
	}

	io_u->error = errno;
	td_verror(td, io_u->error, "xfer");
	return FIO_Q_COMPLETED;
}

static int check_syslet_support(struct syslet_data *sd)
{
	union indirect_params params;
	struct indirect_registers regs;
	pid_t pid, my_pid = getpid();

	memset(&params, 0, sizeof(params));
	fill_syslet_args(&params.syslet, sd->ring, 0, ret_func, sd->stack);

	FILL_IN(regs, __NR_getpid);

	pid = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	if (pid == my_pid)
		return 0;

	return 1;
}

static void fio_syslet_cleanup(struct thread_data *td)
{
	struct syslet_data *sd = td->io_ops->data;

	if (sd) {
		free(sd->events);
		free(sd->ring);
		free(sd->stack);
		free(sd);
		td->io_ops->data = NULL;
	}
}

static int fio_syslet_init(struct thread_data *td)
{
	struct syslet_data *sd;
	void *ring = NULL, *stack = NULL;
	unsigned int ring_size, ring_nr;

	sd = malloc(sizeof(*sd));
	memset(sd, 0, sizeof(*sd));

	sd->events = malloc(sizeof(struct io_u *) * td->o.iodepth);
	memset(sd->events, 0, sizeof(struct io_u *) * td->o.iodepth);

	/*
	 * The ring needs to be a power-of-2, so round it up if we have to
	 */
	ring_nr = td->o.iodepth;
	if (ring_nr & (ring_nr - 1)) {
		int bits = 1;

		while (ring_nr >>= 1)
			bits++;

		ring_nr = 1 << bits;
	}

	ring_size = sizeof(struct syslet_ring) +
			ring_nr * sizeof(struct syslet_completion);
	if (posix_memalign(&ring, sizeof(uint64_t), ring_size))
		goto err_mem;
	if (posix_memalign(&stack, page_size, page_size))
		goto err_mem;

	sd->ring = ring;
	sd->ring_mask = ring_nr - 1;
	sd->stack = stack;

	memset(sd->ring, 0, ring_size);
	sd->ring->elements = ring_nr;

	if (!check_syslet_support(sd)) {
		td->io_ops->data = sd;
		return 0;
	}

	log_err("fio: syslets do not appear to work\n");
err_mem:
	free(sd->events);
	if (ring)
		free(ring);
	if (stack)
		free(stack);
	free(sd);
	return 1;
}

static struct ioengine_ops ioengine = {
	.name		= "syslet-rw",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_syslet_init,
	.queue		= fio_syslet_queue,
	.getevents	= fio_syslet_getevents,
	.event		= fio_syslet_event,
	.cleanup	= fio_syslet_cleanup,
	.open_file	= generic_open_file,
	.close_file	= generic_close_file,
};

#else /* FIO_HAVE_SYSLET */

/*
 * When we have a proper configure system in place, we simply wont build
 * and install this io engine. For now install a crippled version that
 * just complains and fails to load.
 */
static int fio_syslet_init(struct thread_data fio_unused *td)
{
	fprintf(stderr, "fio: syslet not available\n");
	return 1;
}

static struct ioengine_ops ioengine = {
	.name		= "syslet-rw",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_syslet_init,
};

#endif /* FIO_HAVE_SYSLET */

static void fio_init fio_syslet_register(void)
{
	register_ioengine(&ioengine);
}

static void fio_exit fio_syslet_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
a4f4fdd7	1	/*
da751ca9 JA	2	* syslet engine
	3	*
	4	* IO engine that does regular pread(2)/pwrite(2) to transfer data, but
	5	* with syslets to make the execution async.
a4f4fdd7 JA	6	*
	7	*/
	8	#include <stdio.h>
	9	#include <stdlib.h>
	10	#include <unistd.h>
	11	#include <errno.h>
	12	#include <assert.h>
12cbb469	13	#include <malloc.h>
b8846354	14	#include <asm/unistd.h>
a4f4fdd7 JA	15
a4f4fdd7 JA	16	#include "../fio.h"
12cbb469 JA	17	#include "../indirect.h"
12cbb469 JA	18	#include "../syslet.h"
a4f4fdd7 JA	19
	20	#ifdef FIO_HAVE_SYSLET
	21
1760e679 JA	22	#ifdef __NR_pread64
	23	#define __NR_fio_pread __NR_pread64
	24	#define __NR_fio_pwrite __NR_pwrite64
	25	#else
	26	#define __NR_fio_pread __NR_pread
	27	#define __NR_fio_pwrite __NR_pwrite
	28	#endif
	29
a4f4fdd7 JA	30	struct syslet_data {
	31	struct io_u **events;
	32	unsigned int nr_events;
	33
12cbb469	34	struct syslet_ring *ring;
db6defc7	35	unsigned int ring_mask;
12cbb469	36	void *stack;
a4f4fdd7 JA	37	};
a4f4fdd7 JA	38
c6effc3e	39	static void fio_syslet_add_event(struct thread_data td, struct io_u io_u)
9ff9de69 JA	40	{
9ff9de69 JA	41	struct syslet_data *sd = td->io_ops->data;
9ff9de69	42
12cbb469 JA	43	assert(sd->nr_events < td->o.iodepth);
	44	sd->events[sd->nr_events++] = io_u;
	45	}
9ff9de69	46
c6effc3e	47	static void fio_syslet_add_events(struct thread_data *td, unsigned int nr)
12cbb469 JA	48	{
12cbb469 JA	49	struct syslet_data *sd = td->io_ops->data;
db6defc7 JA	50	unsigned int i, uidx;
	51
	52	uidx = sd->ring->user_tail;
	53	read_barrier();
9ff9de69	54
12cbb469	55	for (i = 0; i < nr; i++) {
db6defc7	56	unsigned int idx = (i + uidx) & sd->ring_mask;
12cbb469 JA	57	struct syslet_completion *comp = &sd->ring->comp[idx];
12cbb469 JA	58	struct io_u io_u = (struct io_u ) (long) comp->caller_data;
c6effc3e JA	59	long ret;
	60
	61	ret = comp->status;
	62	if (ret <= 0) {
	63	io_u->resid = io_u->xfer_buflen;
	64	io_u->error = -ret;
	65	} else {
	66	io_u->resid = io_u->xfer_buflen - ret;
	67	io_u->error = 0;
	68	}
5b38ee84	69
c6effc3e	70	fio_syslet_add_event(td, io_u);
12cbb469	71	}
9ff9de69 JA	72	}
9ff9de69 JA	73
12cbb469	74	static void fio_syslet_wait_for_events(struct thread_data *td)
a4f4fdd7 JA	75	{
a4f4fdd7 JA	76	struct syslet_data *sd = td->io_ops->data;
12cbb469 JA	77	struct syslet_ring *ring = sd->ring;
12cbb469 JA	78	unsigned int events;
a4f4fdd7	79
12cbb469	80	events = 0;
a4f4fdd7	81	do {
12cbb469 JA	82	unsigned int kh = ring->kernel_head;
12cbb469 JA	83	int ret;
a4f4fdd7	84
12cbb469 JA	85	/*
	86	* first reap events that are already completed
	87	*/
	88	if (ring->user_tail != kh) {
	89	unsigned int nr = kh - ring->user_tail;
a4f4fdd7	90
c6effc3e	91	fio_syslet_add_events(td, nr);
12cbb469 JA	92	events += nr;
	93	ring->user_tail = kh;
	94	continue;
	95	}
a4f4fdd7	96
12cbb469 JA	97	/*
	98	* block waiting for at least one event
	99	*/
	100	ret = syscall(__NR_syslet_ring_wait, ring, ring->user_tail);
	101	assert(!ret);
	102	} while (!events);
a4f4fdd7 JA	103	}
	104
	105	static int fio_syslet_getevents(struct thread_data *td, int min,
	106	int fio_unused max,
	107	struct timespec fio_unused *t)
	108	{
	109	struct syslet_data *sd = td->io_ops->data;
a4f4fdd7 JA	110	long ret;
	111
	112	do {
a4f4fdd7 JA	113	/*
	114	* do we have enough immediate completions?
	115	*/
	116	if (sd->nr_events >= (unsigned int) min)
	117	break;
	118
12cbb469	119	fio_syslet_wait_for_events(td);
a4f4fdd7 JA	120	} while (1);
	121
	122	ret = sd->nr_events;
	123	sd->nr_events = 0;
	124	return ret;
	125	}
	126
	127	static struct io_u fio_syslet_event(struct thread_data td, int event)
	128	{
	129	struct syslet_data *sd = td->io_ops->data;
	130
	131	return sd->events[event];
	132	}
	133
12cbb469 JA	134	static void fio_syslet_prep_sync(struct fio_file *f,
12cbb469 JA	135	struct indirect_registers *regs)
a4f4fdd7	136	{
12cbb469	137	FILL_IN(*regs, __NR_fsync, (long) f->fd);
a4f4fdd7 JA	138	}
a4f4fdd7 JA	139
12cbb469 JA	140	static void fio_syslet_prep_rw(struct io_u io_u, struct fio_file f,
12cbb469 JA	141	struct indirect_registers *regs)
a4f4fdd7	142	{
12cbb469	143	long nr;
a4f4fdd7	144
a4f4fdd7 JA	145	/*
	146	* prepare rw
	147	*/
	148	if (io_u->ddir == DDIR_READ)
1760e679	149	nr = __NR_fio_pread;
a4f4fdd7	150	else
1760e679	151	nr = __NR_fio_pwrite;
a4f4fdd7	152
12cbb469 JA	153	FILL_IN(*regs, nr, (long) f->fd, (long) io_u->xfer_buf,
12cbb469 JA	154	(long) io_u->xfer_buflen, (long) io_u->offset);
a4f4fdd7 JA	155	}
a4f4fdd7 JA	156
12cbb469	157	static void fio_syslet_prep(struct io_u io_u, struct indirect_registers regs)
a4f4fdd7 JA	158	{
	159	struct fio_file *f = io_u->file;
	160
	161	if (io_u->ddir == DDIR_SYNC)
12cbb469	162	fio_syslet_prep_sync(f, regs);
a4f4fdd7	163	else
12cbb469	164	fio_syslet_prep_rw(io_u, f, regs);
a0a930ef JA	165	}
a0a930ef JA	166
12cbb469	167	static void ret_func(void)
a4f4fdd7	168	{
12cbb469	169	syscall(__NR_exit);
9ff9de69 JA	170	}
	171
	172	static int fio_syslet_queue(struct thread_data td, struct io_u io_u)
	173	{
	174	struct syslet_data *sd = td->io_ops->data;
12cbb469 JA	175	union indirect_params params;
	176	struct indirect_registers regs;
	177	int ret;
bf0dc8fa	178
7101d9c2 JA	179	fio_ro_check(td, io_u);
7101d9c2 JA	180
12cbb469 JA	181	memset(&params, 0, sizeof(params));
12cbb469 JA	182	fill_syslet_args(&params.syslet, sd->ring, (long)io_u, ret_func, sd->stack);
a4f4fdd7	183
12cbb469	184	fio_syslet_prep(io_u, &regs);
2ca50be4	185
12cbb469 JA	186	ret = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	187	if (ret == (int) io_u->xfer_buflen) {
	188	/*
	189	* completed sync, account. this also catches fsync().
	190	*/
	191	return FIO_Q_COMPLETED;
	192	} else if (ret < 0) {
	193	/*
	194	* queued for async execution
	195	*/
	196	if (errno == ESYSLETPENDING)
	197	return FIO_Q_QUEUED;
	198	}
a4f4fdd7	199
12cbb469 JA	200	io_u->error = errno;
	201	td_verror(td, io_u->error, "xfer");
	202	return FIO_Q_COMPLETED;
a4f4fdd7 JA	203	}
a4f4fdd7 JA	204
76f58b92 JA	205	static int check_syslet_support(struct syslet_data *sd)
76f58b92 JA	206	{
12cbb469 JA	207	union indirect_params params;
	208	struct indirect_registers regs;
	209	pid_t pid, my_pid = getpid();
76f58b92	210
12cbb469 JA	211	memset(&params, 0, sizeof(params));
12cbb469 JA	212	fill_syslet_args(&params.syslet, sd->ring, 0, ret_func, sd->stack);
76f58b92	213
12cbb469 JA	214	FILL_IN(regs, __NR_getpid);
	215
	216	pid = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	217	if (pid == my_pid)
	218	return 0;
	219
	220	return 1;
76f58b92 JA	221	}
76f58b92 JA	222
a4f4fdd7 JA	223	static void fio_syslet_cleanup(struct thread_data *td)
	224	{
	225	struct syslet_data *sd = td->io_ops->data;
	226
	227	if (sd) {
a4f4fdd7	228	free(sd->events);
12cbb469 JA	229	free(sd->ring);
12cbb469 JA	230	free(sd->stack);
a4f4fdd7 JA	231	free(sd);
	232	td->io_ops->data = NULL;
	233	}
	234	}
	235
	236	static int fio_syslet_init(struct thread_data *td)
	237	{
	238	struct syslet_data *sd;
e81f7340	239	void ring = NULL, stack = NULL;
05f6a172	240	unsigned int ring_size, ring_nr;
a4f4fdd7 JA	241
	242	sd = malloc(sizeof(*sd));
	243	memset(sd, 0, sizeof(*sd));
12cbb469	244
2dc1bbeb JA	245	sd->events = malloc(sizeof(struct io_u ) td->o.iodepth);
2dc1bbeb JA	246	memset(sd->events, 0, sizeof(struct io_u ) td->o.iodepth);
05f6a172 JA	247
	248	/*
	249	* The ring needs to be a power-of-2, so round it up if we have to
	250	*/
	251	ring_nr = td->o.iodepth;
	252	if (ring_nr & (ring_nr - 1)) {
	253	int bits = 1;
	254
	255	while (ring_nr >>= 1)
	256	bits++;
	257
	258	ring_nr = 1 << bits;
	259	}
	260
	261	ring_size = sizeof(struct syslet_ring) +
	262	ring_nr * sizeof(struct syslet_completion);
	263	if (posix_memalign(&ring, sizeof(uint64_t), ring_size))
e81f7340	264	goto err_mem;
12cbb469	265	if (posix_memalign(&stack, page_size, page_size))
e81f7340	266	goto err_mem;
12cbb469 JA	267
12cbb469 JA	268	sd->ring = ring;
db6defc7	269	sd->ring_mask = ring_nr - 1;
12cbb469 JA	270	sd->stack = stack;
12cbb469 JA	271
05f6a172 JA	272	memset(sd->ring, 0, ring_size);
05f6a172 JA	273	sd->ring->elements = ring_nr;
db64e9bc	274
e81f7340 JA	275	if (!check_syslet_support(sd)) {
	276	td->io_ops->data = sd;
	277	return 0;
76f58b92 JA	278	}
76f58b92 JA	279
e81f7340 JA	280	log_err("fio: syslets do not appear to work\n");
	281	err_mem:
	282	free(sd->events);
	283	if (ring)
	284	free(ring);
	285	if (stack)
	286	free(stack);
	287	free(sd);
	288	return 1;
a4f4fdd7 JA	289	}
	290
	291	static struct ioengine_ops ioengine = {
	292	.name = "syslet-rw",
	293	.version = FIO_IOOPS_VERSION,
	294	.init = fio_syslet_init,
a4f4fdd7 JA	295	.queue = fio_syslet_queue,
	296	.getevents = fio_syslet_getevents,
	297	.event = fio_syslet_event,
	298	.cleanup = fio_syslet_cleanup,
b5af8293 JA	299	.open_file = generic_open_file,
b5af8293 JA	300	.close_file = generic_close_file,
a4f4fdd7 JA	301	};
	302
	303	#else /* FIO_HAVE_SYSLET */
	304
	305	/*
	306	* When we have a proper configure system in place, we simply wont build
	307	* and install this io engine. For now install a crippled version that
	308	* just complains and fails to load.
	309	*/
	310	static int fio_syslet_init(struct thread_data fio_unused *td)
	311	{
	312	fprintf(stderr, "fio: syslet not available\n");
	313	return 1;
	314	}
	315
	316	static struct ioengine_ops ioengine = {
	317	.name = "syslet-rw",
	318	.version = FIO_IOOPS_VERSION,
	319	.init = fio_syslet_init,
	320	};
	321
	322	#endif /* FIO_HAVE_SYSLET */
	323
	324	static void fio_init fio_syslet_register(void)
	325	{
	326	register_ioengine(&ioengine);
	327	}
	328
	329	static void fio_exit fio_syslet_unregister(void)
	330	{
	331	unregister_ioengine(&ioengine);
	332	}