[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;

	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);
			ring->user_tail = kh;
			break;
		}

		/*
		 * block waiting for at least one event
		 */
		ret = syscall(__NR_syslet_ring_wait, ring, ring->user_tail);
		assert(!ret);
	} while (1);
}

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;

	/*
	 * While we have less events than requested, block waiting for them
	 */
	while (sd->nr_events < (unsigned int) min)
		fio_syslet_wait_for_events(td);

	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	77	struct syslet_ring *ring = sd->ring;
a4f4fdd7 JA	78
a4f4fdd7 JA	79	do {
12cbb469 JA	80	unsigned int kh = ring->kernel_head;
12cbb469 JA	81	int ret;
a4f4fdd7	82
12cbb469 JA	83	/*
	84	* first reap events that are already completed
	85	*/
	86	if (ring->user_tail != kh) {
	87	unsigned int nr = kh - ring->user_tail;
a4f4fdd7	88
c6effc3e	89	fio_syslet_add_events(td, nr);
12cbb469	90	ring->user_tail = kh;
4a88edf7	91	break;
12cbb469	92	}
a4f4fdd7	93
12cbb469 JA	94	/*
	95	* block waiting for at least one event
	96	*/
	97	ret = syscall(__NR_syslet_ring_wait, ring, ring->user_tail);
	98	assert(!ret);
4a88edf7	99	} while (1);
a4f4fdd7 JA	100	}
	101
	102	static int fio_syslet_getevents(struct thread_data *td, int min,
	103	int fio_unused max,
	104	struct timespec fio_unused *t)
	105	{
	106	struct syslet_data *sd = td->io_ops->data;
a4f4fdd7 JA	107	long ret;
a4f4fdd7 JA	108
4a88edf7 JA	109	/*
	110	* While we have less events than requested, block waiting for them
	111	*/
	112	while (sd->nr_events < (unsigned int) min)
12cbb469	113	fio_syslet_wait_for_events(td);
a4f4fdd7 JA	114
	115	ret = sd->nr_events;
	116	sd->nr_events = 0;
	117	return ret;
	118	}
	119
	120	static struct io_u fio_syslet_event(struct thread_data td, int event)
	121	{
	122	struct syslet_data *sd = td->io_ops->data;
	123
	124	return sd->events[event];
	125	}
	126
12cbb469 JA	127	static void fio_syslet_prep_sync(struct fio_file *f,
12cbb469 JA	128	struct indirect_registers *regs)
a4f4fdd7	129	{
12cbb469	130	FILL_IN(*regs, __NR_fsync, (long) f->fd);
a4f4fdd7 JA	131	}
a4f4fdd7 JA	132
12cbb469 JA	133	static void fio_syslet_prep_rw(struct io_u io_u, struct fio_file f,
12cbb469 JA	134	struct indirect_registers *regs)
a4f4fdd7	135	{
12cbb469	136	long nr;
a4f4fdd7	137
a4f4fdd7 JA	138	/*
	139	* prepare rw
	140	*/
	141	if (io_u->ddir == DDIR_READ)
1760e679	142	nr = __NR_fio_pread;
a4f4fdd7	143	else
1760e679	144	nr = __NR_fio_pwrite;
a4f4fdd7	145
12cbb469 JA	146	FILL_IN(*regs, nr, (long) f->fd, (long) io_u->xfer_buf,
12cbb469 JA	147	(long) io_u->xfer_buflen, (long) io_u->offset);
a4f4fdd7 JA	148	}
a4f4fdd7 JA	149
12cbb469	150	static void fio_syslet_prep(struct io_u io_u, struct indirect_registers regs)
a4f4fdd7 JA	151	{
	152	struct fio_file *f = io_u->file;
	153
	154	if (io_u->ddir == DDIR_SYNC)
12cbb469	155	fio_syslet_prep_sync(f, regs);
a4f4fdd7	156	else
12cbb469	157	fio_syslet_prep_rw(io_u, f, regs);
a0a930ef JA	158	}
a0a930ef JA	159
12cbb469	160	static void ret_func(void)
a4f4fdd7	161	{
12cbb469	162	syscall(__NR_exit);
9ff9de69 JA	163	}
	164
	165	static int fio_syslet_queue(struct thread_data td, struct io_u io_u)
	166	{
	167	struct syslet_data *sd = td->io_ops->data;
12cbb469 JA	168	union indirect_params params;
	169	struct indirect_registers regs;
	170	int ret;
bf0dc8fa	171
7101d9c2 JA	172	fio_ro_check(td, io_u);
7101d9c2 JA	173
12cbb469 JA	174	memset(&params, 0, sizeof(params));
12cbb469 JA	175	fill_syslet_args(&params.syslet, sd->ring, (long)io_u, ret_func, sd->stack);
a4f4fdd7	176
12cbb469	177	fio_syslet_prep(io_u, &regs);
2ca50be4	178
12cbb469 JA	179	ret = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	180	if (ret == (int) io_u->xfer_buflen) {
	181	/*
	182	* completed sync, account. this also catches fsync().
	183	*/
	184	return FIO_Q_COMPLETED;
	185	} else if (ret < 0) {
	186	/*
	187	* queued for async execution
	188	*/
	189	if (errno == ESYSLETPENDING)
	190	return FIO_Q_QUEUED;
	191	}
a4f4fdd7	192
12cbb469 JA	193	io_u->error = errno;
	194	td_verror(td, io_u->error, "xfer");
	195	return FIO_Q_COMPLETED;
a4f4fdd7 JA	196	}
a4f4fdd7 JA	197
76f58b92 JA	198	static int check_syslet_support(struct syslet_data *sd)
76f58b92 JA	199	{
12cbb469 JA	200	union indirect_params params;
	201	struct indirect_registers regs;
	202	pid_t pid, my_pid = getpid();
76f58b92	203
12cbb469 JA	204	memset(&params, 0, sizeof(params));
12cbb469 JA	205	fill_syslet_args(&params.syslet, sd->ring, 0, ret_func, sd->stack);
76f58b92	206
12cbb469 JA	207	FILL_IN(regs, __NR_getpid);
	208
	209	pid = syscall(__NR_indirect, &regs, &params, sizeof(params), 0);
	210	if (pid == my_pid)
	211	return 0;
	212
	213	return 1;
76f58b92 JA	214	}
76f58b92 JA	215
a4f4fdd7 JA	216	static void fio_syslet_cleanup(struct thread_data *td)
	217	{
	218	struct syslet_data *sd = td->io_ops->data;
	219
	220	if (sd) {
a4f4fdd7	221	free(sd->events);
12cbb469 JA	222	free(sd->ring);
12cbb469 JA	223	free(sd->stack);
a4f4fdd7 JA	224	free(sd);
	225	td->io_ops->data = NULL;
	226	}
	227	}
	228
	229	static int fio_syslet_init(struct thread_data *td)
	230	{
	231	struct syslet_data *sd;
e81f7340	232	void ring = NULL, stack = NULL;
05f6a172	233	unsigned int ring_size, ring_nr;
a4f4fdd7 JA	234
	235	sd = malloc(sizeof(*sd));
	236	memset(sd, 0, sizeof(*sd));
12cbb469	237
2dc1bbeb JA	238	sd->events = malloc(sizeof(struct io_u ) td->o.iodepth);
2dc1bbeb JA	239	memset(sd->events, 0, sizeof(struct io_u ) td->o.iodepth);
05f6a172 JA	240
	241	/*
	242	* The ring needs to be a power-of-2, so round it up if we have to
	243	*/
	244	ring_nr = td->o.iodepth;
	245	if (ring_nr & (ring_nr - 1)) {
	246	int bits = 1;
	247
	248	while (ring_nr >>= 1)
	249	bits++;
	250
	251	ring_nr = 1 << bits;
	252	}
	253
	254	ring_size = sizeof(struct syslet_ring) +
	255	ring_nr * sizeof(struct syslet_completion);
	256	if (posix_memalign(&ring, sizeof(uint64_t), ring_size))
e81f7340	257	goto err_mem;
12cbb469	258	if (posix_memalign(&stack, page_size, page_size))
e81f7340	259	goto err_mem;
12cbb469 JA	260
12cbb469 JA	261	sd->ring = ring;
db6defc7	262	sd->ring_mask = ring_nr - 1;
12cbb469 JA	263	sd->stack = stack;
12cbb469 JA	264
05f6a172 JA	265	memset(sd->ring, 0, ring_size);
05f6a172 JA	266	sd->ring->elements = ring_nr;
db64e9bc	267
e81f7340 JA	268	if (!check_syslet_support(sd)) {
	269	td->io_ops->data = sd;
	270	return 0;
76f58b92 JA	271	}
76f58b92 JA	272
e81f7340 JA	273	log_err("fio: syslets do not appear to work\n");
	274	err_mem:
	275	free(sd->events);
	276	if (ring)
	277	free(ring);
	278	if (stack)
	279	free(stack);
	280	free(sd);
	281	return 1;
a4f4fdd7 JA	282	}
	283
	284	static struct ioengine_ops ioengine = {
	285	.name = "syslet-rw",
	286	.version = FIO_IOOPS_VERSION,
	287	.init = fio_syslet_init,
a4f4fdd7 JA	288	.queue = fio_syslet_queue,
	289	.getevents = fio_syslet_getevents,
	290	.event = fio_syslet_event,
	291	.cleanup = fio_syslet_cleanup,
b5af8293 JA	292	.open_file = generic_open_file,
b5af8293 JA	293	.close_file = generic_close_file,
a4f4fdd7 JA	294	};
	295
	296	#else /* FIO_HAVE_SYSLET */
	297
	298	/*
	299	* When we have a proper configure system in place, we simply wont build
	300	* and install this io engine. For now install a crippled version that
	301	* just complains and fails to load.
	302	*/
	303	static int fio_syslet_init(struct thread_data fio_unused *td)
	304	{
	305	fprintf(stderr, "fio: syslet not available\n");
	306	return 1;
	307	}
	308
	309	static struct ioengine_ops ioengine = {
	310	.name = "syslet-rw",
	311	.version = FIO_IOOPS_VERSION,
	312	.init = fio_syslet_init,
	313	};
	314
	315	#endif /* FIO_HAVE_SYSLET */
	316
	317	static void fio_init fio_syslet_register(void)
	318	{
	319	register_ioengine(&ioengine);
	320	}
	321
	322	static void fio_exit fio_syslet_unregister(void)
	323	{
	324	unregister_ioengine(&ioengine);
	325	}