[fio.git] / t / btrace2fio.c

#include <stdio.h>
#include <stdio.h>
#include <unistd.h>
#include <inttypes.h>
#include <assert.h>

#include "../io_ddir.h"
#include "../flist.h"
#include "../hash.h"
#include "../fifo.h"
#include "../blktrace_api.h"
#include "../os/os.h"
#include "../log.h"
#include "../lib/linux-dev-lookup.h"

#define TRACE_FIFO_SIZE	8192

static unsigned int rt_threshold = 1000000;
static unsigned int ios_threshold = 10;
static int output_ascii = 1;
static char *filename;

struct bs {
	unsigned int bs;
	unsigned int nr;
	int merges;
};

struct trace_file {
	char *name;
	int major, minor;
};

struct btrace_out {
	unsigned long ios[DDIR_RWDIR_CNT];
	unsigned long rw_bs[DDIR_RWDIR_CNT];
	unsigned long merges[DDIR_RWDIR_CNT];

	uint64_t last_end[DDIR_RWDIR_CNT];
	uint64_t seq[DDIR_RWDIR_CNT];

	struct bs *bs[DDIR_RWDIR_CNT];
	unsigned int nr_bs[DDIR_RWDIR_CNT];

	int inflight;
	unsigned int depth;
	uint64_t first_ttime;
	uint64_t last_ttime;

	uint64_t start_delay;
};

struct btrace_pid {
	struct flist_head hash_list;
	struct flist_head pid_list;
	pid_t pid;

	struct trace_file *files;
	int nr_files;
	unsigned int last_major, last_minor;

	struct btrace_out o;
};

struct inflight {
	struct flist_head list;
	struct btrace_pid *p;
	uint64_t end_sector;
};

#define PID_HASH_BITS	10
#define PID_HASH_SIZE	(1U << PID_HASH_BITS)

static struct flist_head pid_hash[PID_HASH_SIZE];
static FLIST_HEAD(pid_list);

#define INFLIGHT_HASH_BITS	8
#define INFLIGHT_HASH_SIZE	(1U << INFLIGHT_HASH_BITS)
static struct flist_head inflight_hash[INFLIGHT_HASH_SIZE];

static uint64_t first_ttime = -1ULL;

static struct inflight *inflight_find(uint64_t sector)
{
	struct flist_head *inflight_list;
	struct flist_head *e;

	inflight_list = &inflight_hash[hash_long(sector, INFLIGHT_HASH_BITS)];

	flist_for_each(e, inflight_list) {
		struct inflight *i = flist_entry(e, struct inflight, list);

		if (i->end_sector == sector)
			return i;
	}

	return NULL;
}

static void inflight_remove(struct inflight *i)
{
	struct btrace_out *o = &i->p->o;

	o->inflight--;
	assert(o->inflight >= 0);
	flist_del(&i->list);
	free(i);
}

static void __inflight_add(struct inflight *i)
{
	struct flist_head *list;

	list = &inflight_hash[hash_long(i->end_sector, INFLIGHT_HASH_BITS)];
	flist_add_tail(&i->list, list);
}

static void inflight_add(struct btrace_pid *p, uint64_t sector, uint32_t len)
{
	struct btrace_out *o = &p->o;
	struct inflight *i;

	i = calloc(1, sizeof(*i));
	i->p = p;
	o->inflight++;
	o->depth = max((int) o->depth, o->inflight);
	i->end_sector = sector + (len >> 9);
	__inflight_add(i);
}

static void inflight_merge(struct inflight *i, int rw, unsigned int size)
{
	i->p->o.merges[rw]++;
	if (size) {
		i->end_sector += (size >> 9);
		flist_del(&i->list);
		__inflight_add(i);
	}
}

/*
 * fifo refill frontend, to avoid reading data in trace sized bites
 */
static int refill_fifo(struct fifo *fifo, int fd)
{
	char buf[TRACE_FIFO_SIZE];
	unsigned int total;
	int ret;

	total = sizeof(buf);
	if (total > fifo_room(fifo))
		total = fifo_room(fifo);

	ret = read(fd, buf, total);
	if (ret < 0) {
		perror("read refill");
		return -1;
	}

	if (ret > 0)
		ret = fifo_put(fifo, buf, ret);

	return ret;
}

/*
 * Retrieve 'len' bytes from the fifo, refilling if necessary.
 */
static int trace_fifo_get(struct fifo *fifo, int fd, void *buf,
			  unsigned int len)
{
	if (fifo_len(fifo) < len) {
		int ret = refill_fifo(fifo, fd);

		if (ret < 0)
			return ret;
	}

	return fifo_get(fifo, buf, len);
}

/*
 * Just discard the pdu by seeking past it.
 */
static int discard_pdu(struct fifo *fifo, int fd, struct blk_io_trace *t)
{
	if (t->pdu_len == 0)
		return 0;

	return trace_fifo_get(fifo, fd, NULL, t->pdu_len);
}

static void handle_trace_notify(struct blk_io_trace *t)
{
	switch (t->action) {
	case BLK_TN_PROCESS:
		//printf("got process notify: %x, %d\n", t->action, t->pid);
		break;
	case BLK_TN_TIMESTAMP:
		//printf("got timestamp notify: %x, %d\n", t->action, t->pid);
		break;
	case BLK_TN_MESSAGE:
		break;
	default:
		log_err("unknown trace act %x\n", t->action);
		break;
	}
}

static void __add_bs(struct btrace_out *o, unsigned int len, int rw)
{
	o->bs[rw] = realloc(o->bs[rw], (o->nr_bs[rw] + 1) * sizeof(struct bs));
	o->bs[rw][o->nr_bs[rw]].bs = len;
	o->bs[rw][o->nr_bs[rw]].nr = 1;
	o->nr_bs[rw]++;
}

static void add_bs(struct btrace_out *o, unsigned int len, int rw)
{
	struct bs *bs = o->bs[rw];
	int i;

	if (!o->nr_bs[rw]) {
		__add_bs(o, len, rw);
		return;
	}

	for (i = 0; i < o->nr_bs[rw]; i++) {
		if (bs[i].bs == len) {
			bs[i].nr++;
			return;
		}
	}

	__add_bs(o, len, rw);
}

#define FMINORBITS	20
#define FMINORMASK	((1U << FMINORBITS) - 1)
#define FMAJOR(dev)	((unsigned int) ((dev) >> FMINORBITS))
#define FMINOR(dev)	((unsigned int) ((dev) & FMINORMASK))

static void btrace_add_file(struct btrace_pid *p, uint32_t devno)
{
	unsigned int maj = FMAJOR(devno);
	unsigned int min = FMINOR(devno);
	struct trace_file *f;
	unsigned int i;
	char dev[256];

	if (filename)
		return;
	if (p->last_major == maj && p->last_minor == min)
		return;

	p->last_major = maj;
	p->last_minor = min;

	/*
	 * check for this file in our list
	 */
	for (i = 0; i < p->nr_files; i++) {
		f = &p->files[i];

		if (f->major == maj && f->minor == min)
			return;
	}

	strcpy(dev, "/dev");
	if (!blktrace_lookup_device(NULL, dev, maj, min)) {
		log_err("fio: failed to find device %u/%u\n", maj, min);
		return;
	}

	p->files = realloc(p->files, (p->nr_files + 1) * sizeof(*f));
	f = &p->files[p->nr_files];
	f->name = strdup(dev);
	f->major = maj;
	f->minor = min;
	p->nr_files++;
}

static void handle_trace_discard(struct blk_io_trace *t, struct btrace_pid *p)
{
	struct btrace_out *o = &p->o;

	btrace_add_file(p, t->device);

	if (o->first_ttime == -1ULL)
		o->first_ttime = t->time;

	o->ios[DDIR_TRIM]++;
	add_bs(o, t->bytes, DDIR_TRIM);
}

static void handle_trace_fs(struct blk_io_trace *t, struct btrace_pid *p)
{
	struct btrace_out *o = &p->o;
	int rw;

	btrace_add_file(p, t->device);

	first_ttime = min(first_ttime, (uint64_t) t->time);

	if (o->first_ttime == -1ULL)
		o->first_ttime = t->time;

	rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;

	add_bs(o, t->bytes, rw);
	o->ios[rw]++;

	if (t->sector == o->last_end[rw] || o->last_end[rw] == -1ULL)
		o->seq[rw]++;

	o->last_end[rw] = t->sector + (t->bytes >> 9);
}

static void handle_queue_trace(struct blk_io_trace *t, struct btrace_pid *p)
{
	if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
		handle_trace_notify(t);
	else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
		handle_trace_discard(t, p);
	else
		handle_trace_fs(t, p);
}

static void handle_trace(struct blk_io_trace *t, struct btrace_pid *p)
{
	unsigned int act = t->action & 0xffff;

	if (act == __BLK_TA_QUEUE) {
		inflight_add(p, t->sector, t->bytes);
		handle_queue_trace(t, p);
	} else if (act == __BLK_TA_REQUEUE) {
		p->o.inflight--;
	} else if (act == __BLK_TA_BACKMERGE) {
		struct inflight *i;

		i = inflight_find(t->sector + (t->bytes >> 9));
		if (i)
			inflight_remove(i);

		i = inflight_find(t->sector);
		if (i) {
			int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;

			inflight_merge(i, rw, t->bytes);
		}
	} else if (act == __BLK_TA_FRONTMERGE) {
		struct inflight *i;

		i = inflight_find(t->sector + (t->bytes >> 9));
		if (i)
			inflight_remove(i);

		i = inflight_find(t->sector);
		if (i) {
			int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;

			inflight_merge(i, rw, 0);
		}
	} else if (act == __BLK_TA_COMPLETE) {
		struct inflight *i;

		i = inflight_find(t->sector + (t->bytes >> 9));
		if (i)
			inflight_remove(i);
	}
}

static void byteswap_trace(struct blk_io_trace *t)
{
	t->magic = fio_swap32(t->magic);
	t->sequence = fio_swap32(t->sequence);
	t->time = fio_swap64(t->time);
	t->sector = fio_swap64(t->sector);
	t->bytes = fio_swap32(t->bytes);
	t->action = fio_swap32(t->action);
	t->pid = fio_swap32(t->pid);
	t->device = fio_swap32(t->device);
	t->cpu = fio_swap32(t->cpu);
	t->error = fio_swap16(t->error);
	t->pdu_len = fio_swap16(t->pdu_len);
}

static struct btrace_pid *pid_hash_find(pid_t pid, struct flist_head *list)
{
	struct flist_head *e;
	struct btrace_pid *p;

	flist_for_each(e, list) {
		p = flist_entry(e, struct btrace_pid, hash_list);
		if (p->pid == pid)
			return p;
	}

	return NULL;
}

static struct btrace_pid *pid_hash_get(pid_t pid)
{
	struct flist_head *hash_list;
	struct btrace_pid *p;

	hash_list = &pid_hash[hash_long(pid, PID_HASH_BITS)];

	p = pid_hash_find(pid, hash_list);
	if (!p) {
		int i;

		p = calloc(1, sizeof(*p));
		p->o.first_ttime = -1ULL;
		p->o.last_ttime = -1ULL;

		for (i = 0; i < DDIR_RWDIR_CNT; i++)
			p->o.last_end[i] = -1ULL;

		p->pid = pid;
		flist_add_tail(&p->hash_list, hash_list);
		flist_add_tail(&p->pid_list, &pid_list);
	}

	return p;
}

/*
 * Load a blktrace file by reading all the blk_io_trace entries, and storing
 * them as io_pieces like the fio text version would do.
 */
static int load_blktrace(const char *filename, int need_swap)
{
	struct btrace_pid *p;
	unsigned long traces;
	struct blk_io_trace t;
	struct fifo *fifo;
	int fd;

	fd = open(filename, O_RDONLY);
	if (fd < 0) {
		perror("open trace file\n");
		return 1;
	}

	fifo = fifo_alloc(TRACE_FIFO_SIZE);

	traces = 0;
	do {
		int ret = trace_fifo_get(fifo, fd, &t, sizeof(t));

		if (ret < 0)
			goto err;
		else if (!ret)
			break;
		else if (ret < (int) sizeof(t)) {
			log_err("fio: short fifo get\n");
			break;
		}

		if (need_swap)
			byteswap_trace(&t);

		if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
			log_err("fio: bad magic in blktrace data: %x\n", t.magic);
			goto err;
		}
		if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
			log_err("fio: bad blktrace version %d\n", t.magic & 0xff);
			goto err;
		}
		ret = discard_pdu(fifo, fd, &t);
		if (ret < 0) {
			log_err("blktrace lseek\n");
			goto err;
		} else if (t.pdu_len != ret) {
			log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
			goto err;
		}

		p = pid_hash_get(t.pid);
		handle_trace(&t, p);
		p->o.last_ttime = t.time;
		traces++;
	} while (1);

	fifo_free(fifo);
	close(fd);

	if (output_ascii)
		printf("Traces loaded: %lu\n", traces);

	return 0;
err:
	close(fd);
	fifo_free(fifo);
	return 1;
}

static int bs_cmp(const void *ba, const void *bb)
{
	const struct bs *bsa = ba;
	const struct bs *bsb = bb;

	return bsb->nr - bsa->nr;
}

static void __output_p_ascii(struct btrace_pid *p, unsigned long *ios)
{
	const char *msg[] = { "reads", "writes", "trims" };
	struct btrace_out *o = &p->o;
	unsigned long total;
	int i, j;

	printf("[pid:\t%u]\n", p->pid);

	total = ddir_rw_sum(o->ios);
	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
		float perc;

		if (!o->ios[i])
			continue;

		ios[i] += o->ios[i] + o->merges[i];
		printf("%s\n", msg[i]);
		perc = ((float) o->ios[i] * 100.0) / (float) total;
		printf("\tios:    %lu (perc=%3.2f%%)\n", o->ios[i], perc);
		perc = ((float) o->merges[i] * 100.0) / (float) total;
		printf("\tmerges: %lu (perc=%3.2f%%)\n", o->merges[i], perc);
		perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
		printf("\tseq:    %lu (perc=%3.2f%%)\n", o->seq[i], perc);

		for (j = 0; j < o->nr_bs[i]; j++) {
			struct bs *bs = &o->bs[i][j];

			perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
			printf("\tbs=%u, perc=%3.2f%%\n", bs->bs, perc);
		}
	}

	printf("depth:\t%u\n", o->depth);
	printf("usec:\t%llu (delay=%llu)\n", (o->last_ttime - o->first_ttime) / 1000ULL, (unsigned long long) o->start_delay);

	printf("files:\t");
	for (i = 0; i < p->nr_files; i++)
		printf("%s,", p->files[i].name);
	printf("\n");

	printf("\n");
}

static int __output_p_fio(struct btrace_pid *p, unsigned long *ios)
{
	struct btrace_out *o = &p->o;
	unsigned long total;
	unsigned long long time;
	float perc;
	int i, j;

	if ((o->ios[0] + o->ios[1]) && o->ios[2]) {
		log_err("fio: trace has both read/write and trim\n");
		return 1;
	}

	printf("[pid%u]\n", p->pid);
	printf("direct=1\n");
	if (o->depth == 1)
		printf("ioengine=sync\n");
	else
		printf("ioengine=libaio\niodepth=%u\n", o->depth);

	if (o->ios[0] && !o->ios[1])
		printf("rw=randread\n");
	else if (!o->ios[0] && o->ios[1])
		printf("rw=randwrite\n");
	else if (o->ios[2])
		printf("rw=randtrim\n");
	else {
		printf("rw=randrw\n");
		total = ddir_rw_sum(o->ios);
		perc = ((float) o->ios[0] * 100.0) / (float) total;
		printf("rwmixread=%u\n", (int) (perc + 0.99));
	}

	printf("percentage_random=");
	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
		if (o->seq[i] && o->ios[i]) {
			perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
			if (perc >= 99.0)
				perc = 100.0;
		} else
			perc = 100.0;

		if (i)
			printf(",");
		perc = 100.0 - perc;
		printf("%u", (int) perc);
	}
	printf("\n");

	printf("filename=");
	for (i = 0; i < p->nr_files; i++) {
		if (i)
			printf(":");
		printf("%s", p->files[i].name);
	}
	printf("\n");

	printf("startdelay=%llus\n", o->start_delay / 1000000ULL);

	time = o->last_ttime - o->first_ttime;
	time = (time + 1000000000ULL - 1) / 1000000000ULL;
	printf("runtime=%llus\n", time);

	printf("bssplit=");
	for (i = 0; i < DDIR_RWDIR_CNT; i++) {

		if (i && o->nr_bs[i - 1] && o->nr_bs[i])
			printf(",");

		for (j = 0; j < o->nr_bs[i]; j++) {
			struct bs *bs = &o->bs[i][j];

			perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
			if (perc < 1.00)
				continue;
			if (j)
				printf(":");
			if (j + 1 == o->nr_bs[i])
				printf("%u/", bs->bs);
			else
				printf("%u/%u", bs->bs, (int) perc);
		}
	}
	printf("\n\n");

	return 0;
}

static int __output_p(struct btrace_pid *p, unsigned long *ios)
{
	struct btrace_out *o = &p->o;
	int i, ret = 0;

	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
		if (o->nr_bs[i] <= 1)
			continue;
		qsort(o->bs[i], o->nr_bs[i], sizeof(struct bs), bs_cmp);
	}

	if (filename) {
		p->files = malloc(sizeof(struct trace_file));
		p->nr_files++;
		p->files[0].name = filename;
	}

	if (output_ascii)
		__output_p_ascii(p, ios);
	else
		ret = __output_p_fio(p, ios);

	return ret;
}

static int prune_entry(struct btrace_out *o)
{
	uint64_t time;

	if (ddir_rw_sum(o->ios) < ios_threshold)
		return 1;

	time = (o->last_ttime - o->first_ttime) / 1000ULL;
	if (time < rt_threshold)
		return 1;

	return 0;
}

static int entry_cmp(void *priv, struct flist_head *a, struct flist_head *b)
{
	struct btrace_pid *pa = flist_entry(a, struct btrace_pid, pid_list);
	struct btrace_pid *pb = flist_entry(b, struct btrace_pid, pid_list);

	return ddir_rw_sum(pb->o.ios) - ddir_rw_sum(pa->o.ios);
}

static void free_p(struct btrace_pid *p)
{
	struct btrace_out *o = &p->o;
	int i;

	for (i = 0; i < p->nr_files; i++) {
		if (p->files[i].name && p->files[i].name != filename)
			free(p->files[i].name);
	}

	for (i = 0; i < DDIR_RWDIR_CNT; i++)
		free(o->bs[i]);

	free(p->files);
	flist_del(&p->pid_list);
	flist_del(&p->hash_list);
	free(p);
}

static int output_p(void)
{
	unsigned long ios[DDIR_RWDIR_CNT];
	struct flist_head *e, *tmp;
	int ret = 0;

	flist_for_each_safe(e, tmp, &pid_list) {
		struct btrace_pid *p;

		p = flist_entry(e, struct btrace_pid, pid_list);
		if (prune_entry(&p->o)) {
			free_p(p);
			continue;
		}
		p->o.start_delay = (p->o.first_ttime / 1000ULL) - first_ttime;
	}

	memset(ios, 0, sizeof(ios));

	flist_sort(NULL, &pid_list, entry_cmp);

	flist_for_each(e, &pid_list) {
		struct btrace_pid *p;

		p = flist_entry(e, struct btrace_pid, pid_list);
		ret |= __output_p(p, ios);
	}

	if (output_ascii)
		printf("Total: reads=%lu, writes=%lu\n", ios[0], ios[1]);

	return ret;
}

static int usage(char *argv[])
{
	log_err("%s: <blktrace bin file>\n", argv[0]);
	log_err("\t-t\tUsec threshold to ignore task\n");
	log_err("\t-n\tNumber IOS threshold to ignore task\n");
	log_err("\t-f\tFio job file output\n");
	log_err("\t-d\tUse this file/device for replay\n");
	return 1;
}

static int trace_needs_swap(const char *trace_file, int *swap)
{
	struct blk_io_trace t;
	int fd, ret;

	*swap = -1;
	
	fd = open(trace_file, O_RDONLY);
	if (fd < 0) {
		perror("open");
		return 1;
	}

	ret = read(fd, &t, sizeof(t));
	if (ret < 0) {
		perror("read");
		return 1;
	} else if (ret != sizeof(t)) {
		log_err("fio: short read on trace file\n");
		return 1;
	}

	close(fd);

	if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
		*swap = 0;
	else {
		/*
		 * Maybe it needs to be endian swapped...
		 */
		t.magic = fio_swap32(t.magic);
		if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
			*swap = 1;
	}

	if (*swap == -1) {
		log_err("fio: blktrace appears corrupt\n");
		return 1;
	}

	return 0;
}

int main(int argc, char *argv[])
{
	int need_swap, i, c;

	if (argc < 2)
		return usage(argv);

	while ((c = getopt(argc, argv, "t:n:fd:")) != -1) {
		switch (c) {
		case 't':
			rt_threshold = atoi(optarg);
			break;
		case 'n':
			ios_threshold = atoi(optarg);
			break;
		case 'f':
			output_ascii = 0;
			break;
		case 'd':
			filename = strdup(optarg);
			break;
		case '?':
		default:
			return usage(argv);
		}
	}

	if (argc == optind)
		return usage(argv);

	if (trace_needs_swap(argv[optind], &need_swap))
		return 1;

	for (i = 0; i < PID_HASH_SIZE; i++)
		INIT_FLIST_HEAD(&pid_hash[i]);
	for (i = 0; i < INFLIGHT_HASH_SIZE; i++)
		INIT_FLIST_HEAD(&inflight_hash[i]);

	load_blktrace(argv[optind], need_swap);
	first_ttime /= 1000ULL;

	return output_p();
}
Commit	Line	Data
40bafa33 JA	1	#include <stdio.h>
	2	#include <stdio.h>
	3	#include <unistd.h>
	4	#include <inttypes.h>
	5	#include <assert.h>
	6
	7	#include "../io_ddir.h"
	8	#include "../flist.h"
	9	#include "../hash.h"
	10	#include "../fifo.h"
	11	#include "../blktrace_api.h"
	12	#include "../os/os.h"
	13	#include "../log.h"
	14	#include "../lib/linux-dev-lookup.h"
	15
	16	#define TRACE_FIFO_SIZE 8192
	17
	18	static unsigned int rt_threshold = 1000000;
	19	static unsigned int ios_threshold = 10;
	20	static int output_ascii = 1;
35661615	21	static char *filename;
40bafa33 JA	22
	23	struct bs {
	24	unsigned int bs;
	25	unsigned int nr;
	26	int merges;
	27	};
	28
	29	struct trace_file {
	30	char *name;
	31	int major, minor;
	32	};
	33
	34	struct btrace_out {
	35	unsigned long ios[DDIR_RWDIR_CNT];
	36	unsigned long rw_bs[DDIR_RWDIR_CNT];
	37	unsigned long merges[DDIR_RWDIR_CNT];
	38
	39	uint64_t last_end[DDIR_RWDIR_CNT];
	40	uint64_t seq[DDIR_RWDIR_CNT];
	41
	42	struct bs *bs[DDIR_RWDIR_CNT];
	43	unsigned int nr_bs[DDIR_RWDIR_CNT];
	44
	45	int inflight;
	46	unsigned int depth;
	47	uint64_t first_ttime;
	48	uint64_t last_ttime;
	49
40bafa33 JA	50	uint64_t start_delay;
	51	};
	52
	53	struct btrace_pid {
	54	struct flist_head hash_list;
	55	struct flist_head pid_list;
	56	pid_t pid;
65ff9cd5 JA	57
	58	struct trace_file *files;
	59	int nr_files;
	60	unsigned int last_major, last_minor;
	61
40bafa33 JA	62	struct btrace_out o;
	63	};
	64
	65	struct inflight {
	66	struct flist_head list;
	67	struct btrace_pid *p;
	68	uint64_t end_sector;
	69	};
	70
	71	#define PID_HASH_BITS 10
	72	#define PID_HASH_SIZE (1U << PID_HASH_BITS)
	73
	74	static struct flist_head pid_hash[PID_HASH_SIZE];
	75	static FLIST_HEAD(pid_list);
	76
17bf0853 JA	77	#define INFLIGHT_HASH_BITS 8
	78	#define INFLIGHT_HASH_SIZE (1U << INFLIGHT_HASH_BITS)
	79	static struct flist_head inflight_hash[INFLIGHT_HASH_SIZE];
40bafa33 JA	80
	81	static uint64_t first_ttime = -1ULL;
	82
	83	static struct inflight *inflight_find(uint64_t sector)
	84	{
17bf0853	85	struct flist_head *inflight_list;
40bafa33 JA	86	struct flist_head *e;
40bafa33 JA	87
17bf0853 JA	88	inflight_list = &inflight_hash[hash_long(sector, INFLIGHT_HASH_BITS)];
	89
	90	flist_for_each(e, inflight_list) {
40bafa33 JA	91	struct inflight *i = flist_entry(e, struct inflight, list);
	92
	93	if (i->end_sector == sector)
	94	return i;
	95	}
	96
	97	return NULL;
	98	}
	99
	100	static void inflight_remove(struct inflight *i)
	101	{
	102	struct btrace_out *o = &i->p->o;
	103
	104	o->inflight--;
	105	assert(o->inflight >= 0);
	106	flist_del(&i->list);
	107	free(i);
	108	}
	109
17bf0853	110	static void __inflight_add(struct inflight *i)
40bafa33	111	{
17bf0853 JA	112	struct flist_head *list;
	113
	114	list = &inflight_hash[hash_long(i->end_sector, INFLIGHT_HASH_BITS)];
	115	flist_add_tail(&i->list, list);
40bafa33 JA	116	}
	117
	118	static void inflight_add(struct btrace_pid *p, uint64_t sector, uint32_t len)
	119	{
	120	struct btrace_out *o = &p->o;
	121	struct inflight *i;
	122
	123	i = calloc(1, sizeof(*i));
	124	i->p = p;
	125	o->inflight++;
	126	o->depth = max((int) o->depth, o->inflight);
	127	i->end_sector = sector + (len >> 9);
17bf0853 JA	128	__inflight_add(i);
	129	}
	130
	131	static void inflight_merge(struct inflight *i, int rw, unsigned int size)
	132	{
	133	i->p->o.merges[rw]++;
	134	if (size) {
	135	i->end_sector += (size >> 9);
	136	flist_del(&i->list);
	137	__inflight_add(i);
	138	}
40bafa33 JA	139	}
	140
	141	/*
	142	* fifo refill frontend, to avoid reading data in trace sized bites
	143	*/
	144	static int refill_fifo(struct fifo *fifo, int fd)
	145	{
	146	char buf[TRACE_FIFO_SIZE];
	147	unsigned int total;
	148	int ret;
	149
	150	total = sizeof(buf);
	151	if (total > fifo_room(fifo))
	152	total = fifo_room(fifo);
	153
	154	ret = read(fd, buf, total);
	155	if (ret < 0) {
	156	perror("read refill");
	157	return -1;
	158	}
	159
	160	if (ret > 0)
	161	ret = fifo_put(fifo, buf, ret);
	162
	163	return ret;
	164	}
	165
	166	/*
	167	* Retrieve 'len' bytes from the fifo, refilling if necessary.
	168	*/
	169	static int trace_fifo_get(struct fifo fifo, int fd, void buf,
	170	unsigned int len)
	171	{
	172	if (fifo_len(fifo) < len) {
	173	int ret = refill_fifo(fifo, fd);
	174
	175	if (ret < 0)
	176	return ret;
	177	}
	178
	179	return fifo_get(fifo, buf, len);
	180	}
	181
	182	/*
	183	* Just discard the pdu by seeking past it.
	184	*/
	185	static int discard_pdu(struct fifo fifo, int fd, struct blk_io_trace t)
	186	{
	187	if (t->pdu_len == 0)
	188	return 0;
	189
	190	return trace_fifo_get(fifo, fd, NULL, t->pdu_len);
	191	}
	192
	193	static void handle_trace_notify(struct blk_io_trace *t)
	194	{
	195	switch (t->action) {
	196	case BLK_TN_PROCESS:
	197	//printf("got process notify: %x, %d\n", t->action, t->pid);
	198	break;
	199	case BLK_TN_TIMESTAMP:
	200	//printf("got timestamp notify: %x, %d\n", t->action, t->pid);
	201	break;
	202	case BLK_TN_MESSAGE:
203	break;
204	default:
17bf0853	205	log_err("unknown trace act %x\n", t->action);
40bafa33 JA	206	break;
	207	}
	208	}
	209
	210	static void __add_bs(struct btrace_out *o, unsigned int len, int rw)
	211	{
	212	o->bs[rw] = realloc(o->bs[rw], (o->nr_bs[rw] + 1) * sizeof(struct bs));
	213	o->bs[rw][o->nr_bs[rw]].bs = len;
	214	o->bs[rw][o->nr_bs[rw]].nr = 1;
	215	o->nr_bs[rw]++;
	216	}
	217
	218	static void add_bs(struct btrace_out *o, unsigned int len, int rw)
	219	{
	220	struct bs *bs = o->bs[rw];
	221	int i;
	222
	223	if (!o->nr_bs[rw]) {
	224	__add_bs(o, len, rw);
	225	return;
	226	}
	227
	228	for (i = 0; i < o->nr_bs[rw]; i++) {
	229	if (bs[i].bs == len) {
	230	bs[i].nr++;
	231	return;
	232	}
	233	}
	234
	235	__add_bs(o, len, rw);
	236	}
	237
	238	#define FMINORBITS 20
	239	#define FMINORMASK ((1U << FMINORBITS) - 1)
	240	#define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS))
	241	#define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK))
	242
65ff9cd5	243	static void btrace_add_file(struct btrace_pid *p, uint32_t devno)
40bafa33 JA	244	{
	245	unsigned int maj = FMAJOR(devno);
	246	unsigned int min = FMINOR(devno);
	247	struct trace_file *f;
	248	unsigned int i;
	249	char dev[256];
	250
35661615 JA	251	if (filename)
35661615 JA	252	return;
65ff9cd5	253	if (p->last_major == maj && p->last_minor == min)
40bafa33 JA	254	return;
40bafa33 JA	255
65ff9cd5 JA	256	p->last_major = maj;
65ff9cd5 JA	257	p->last_minor = min;
40bafa33 JA	258
	259	/*
	260	* check for this file in our list
	261	*/
65ff9cd5 JA	262	for (i = 0; i < p->nr_files; i++) {
65ff9cd5 JA	263	f = &p->files[i];
40bafa33 JA	264
	265	if (f->major == maj && f->minor == min)
	266	return;
	267	}
	268
	269	strcpy(dev, "/dev");
	270	if (!blktrace_lookup_device(NULL, dev, maj, min)) {
	271	log_err("fio: failed to find device %u/%u\n", maj, min);
	272	return;
	273	}
	274
65ff9cd5 JA	275	p->files = realloc(p->files, (p->nr_files + 1) * sizeof(*f));
65ff9cd5 JA	276	f = &p->files[p->nr_files];
40bafa33 JA	277	f->name = strdup(dev);
	278	f->major = maj;
	279	f->minor = min;
65ff9cd5	280	p->nr_files++;
40bafa33 JA	281	}
40bafa33 JA	282
65ff9cd5	283	static void handle_trace_discard(struct blk_io_trace t, struct btrace_pid p)
40bafa33	284	{
65ff9cd5 JA	285	struct btrace_out *o = &p->o;
	286
	287	btrace_add_file(p, t->device);
40bafa33 JA	288
	289	if (o->first_ttime == -1ULL)
	290	o->first_ttime = t->time;
	291
	292	o->ios[DDIR_TRIM]++;
	293	add_bs(o, t->bytes, DDIR_TRIM);
	294	}
	295
65ff9cd5	296	static void handle_trace_fs(struct blk_io_trace t, struct btrace_pid p)
40bafa33	297	{
65ff9cd5	298	struct btrace_out *o = &p->o;
40bafa33 JA	299	int rw;
40bafa33 JA	300
65ff9cd5	301	btrace_add_file(p, t->device);
40bafa33 JA	302
	303	first_ttime = min(first_ttime, (uint64_t) t->time);
	304
	305	if (o->first_ttime == -1ULL)
	306	o->first_ttime = t->time;
	307
	308	rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	309
	310	add_bs(o, t->bytes, rw);
	311	o->ios[rw]++;
	312
	313	if (t->sector == o->last_end[rw] \|\| o->last_end[rw] == -1ULL)
	314	o->seq[rw]++;
	315
	316	o->last_end[rw] = t->sector + (t->bytes >> 9);
	317	}
	318
65ff9cd5	319	static void handle_queue_trace(struct blk_io_trace t, struct btrace_pid p)
40bafa33 JA	320	{
	321	if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
	322	handle_trace_notify(t);
	323	else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
65ff9cd5	324	handle_trace_discard(t, p);
40bafa33	325	else
65ff9cd5	326	handle_trace_fs(t, p);
40bafa33 JA	327	}
	328
	329	static void handle_trace(struct blk_io_trace t, struct btrace_pid p)
	330	{
	331	unsigned int act = t->action & 0xffff;
	332
	333	if (act == __BLK_TA_QUEUE) {
	334	inflight_add(p, t->sector, t->bytes);
65ff9cd5	335	handle_queue_trace(t, p);
40bafa33 JA	336	} else if (act == __BLK_TA_REQUEUE) {
	337	p->o.inflight--;
	338	} else if (act == __BLK_TA_BACKMERGE) {
	339	struct inflight *i;
	340
	341	i = inflight_find(t->sector + (t->bytes >> 9));
	342	if (i)
	343	inflight_remove(i);
	344
	345	i = inflight_find(t->sector);
	346	if (i) {
	347	int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	348
	349	inflight_merge(i, rw, t->bytes);
	350	}
	351	} else if (act == __BLK_TA_FRONTMERGE) {
	352	struct inflight *i;
	353
	354	i = inflight_find(t->sector + (t->bytes >> 9));
	355	if (i)
	356	inflight_remove(i);
	357
	358	i = inflight_find(t->sector);
	359	if (i) {
	360	int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	361
	362	inflight_merge(i, rw, 0);
	363	}
	364	} else if (act == __BLK_TA_COMPLETE) {
	365	struct inflight *i;
	366
	367	i = inflight_find(t->sector + (t->bytes >> 9));
	368	if (i)
	369	inflight_remove(i);
	370	}
	371	}
	372
	373	static void byteswap_trace(struct blk_io_trace *t)
	374	{
	375	t->magic = fio_swap32(t->magic);
	376	t->sequence = fio_swap32(t->sequence);
	377	t->time = fio_swap64(t->time);
	378	t->sector = fio_swap64(t->sector);
	379	t->bytes = fio_swap32(t->bytes);
	380	t->action = fio_swap32(t->action);
	381	t->pid = fio_swap32(t->pid);
	382	t->device = fio_swap32(t->device);
	383	t->cpu = fio_swap32(t->cpu);
	384	t->error = fio_swap16(t->error);
	385	t->pdu_len = fio_swap16(t->pdu_len);
	386	}
	387
	388	static struct btrace_pid pid_hash_find(pid_t pid, struct flist_head list)
	389	{
	390	struct flist_head *e;
	391	struct btrace_pid *p;
	392
	393	flist_for_each(e, list) {
	394	p = flist_entry(e, struct btrace_pid, hash_list);
	395	if (p->pid == pid)
	396	return p;
	397	}
	398
	399	return NULL;
400	}
401
402	static struct btrace_pid *pid_hash_get(pid_t pid)
403	{
404	struct flist_head *hash_list;
405	struct btrace_pid *p;
406
407	hash_list = &pid_hash[hash_long(pid, PID_HASH_BITS)];
408
409	p = pid_hash_find(pid, hash_list);
410	if (!p) {
411	int i;
412
413	p = calloc(1, sizeof(*p));
414	p->o.first_ttime = -1ULL;
415	p->o.last_ttime = -1ULL;
416
417	for (i = 0; i < DDIR_RWDIR_CNT; i++)
418	p->o.last_end[i] = -1ULL;
419
420	p->pid = pid;
421	flist_add_tail(&p->hash_list, hash_list);
422	flist_add_tail(&p->pid_list, &pid_list);
423	}
424
425	return p;
426	}
427
428	/*
429	* Load a blktrace file by reading all the blk_io_trace entries, and storing
430	* them as io_pieces like the fio text version would do.
431	*/
432	static int load_blktrace(const char *filename, int need_swap)
433	{
434	struct btrace_pid *p;
435	unsigned long traces;
436	struct blk_io_trace t;
437	struct fifo *fifo;
438	int fd;
439
440	fd = open(filename, O_RDONLY);
441	if (fd < 0) {
442	perror("open trace file\n");
443	return 1;
444	}
445
446	fifo = fifo_alloc(TRACE_FIFO_SIZE);
447
448	traces = 0;
449	do {
450	int ret = trace_fifo_get(fifo, fd, &t, sizeof(t));
451
452	if (ret < 0)
453	goto err;
454	else if (!ret)
455	break;
456	else if (ret < (int) sizeof(t)) {
17bf0853	457	log_err("fio: short fifo get\n");
40bafa33 JA	458	break;
	459	}
	460
	461	if (need_swap)
	462	byteswap_trace(&t);
	463
	464	if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
17bf0853	465	log_err("fio: bad magic in blktrace data: %x\n", t.magic);
40bafa33 JA	466	goto err;
	467	}
	468	if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
17bf0853	469	log_err("fio: bad blktrace version %d\n", t.magic & 0xff);
40bafa33 JA	470	goto err;
	471	}
	472	ret = discard_pdu(fifo, fd, &t);
	473	if (ret < 0) {
17bf0853	474	log_err("blktrace lseek\n");
40bafa33 JA	475	goto err;
40bafa33 JA	476	} else if (t.pdu_len != ret) {
17bf0853	477	log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
40bafa33 JA	478	goto err;
	479	}
	480
	481	p = pid_hash_get(t.pid);
	482	handle_trace(&t, p);
	483	p->o.last_ttime = t.time;
	484	traces++;
	485	} while (1);
	486
	487	fifo_free(fifo);
	488	close(fd);
	489
	490	if (output_ascii)
	491	printf("Traces loaded: %lu\n", traces);
	492
	493	return 0;
	494	err:
	495	close(fd);
	496	fifo_free(fifo);
	497	return 1;
	498	}
	499
	500	static int bs_cmp(const void ba, const void bb)
	501	{
	502	const struct bs *bsa = ba;
	503	const struct bs *bsb = bb;
	504
	505	return bsb->nr - bsa->nr;
	506	}
	507
	508	static void __output_p_ascii(struct btrace_pid p, unsigned long ios)
	509	{
	510	const char *msg[] = { "reads", "writes", "trims" };
	511	struct btrace_out *o = &p->o;
	512	unsigned long total;
	513	int i, j;
	514
	515	printf("[pid:\t%u]\n", p->pid);
	516
	517	total = ddir_rw_sum(o->ios);
	518	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	519	float perc;
	520
	521	if (!o->ios[i])
	522	continue;
	523
	524	ios[i] += o->ios[i] + o->merges[i];
	525	printf("%s\n", msg[i]);
	526	perc = ((float) o->ios[i] * 100.0) / (float) total;
	527	printf("\tios: %lu (perc=%3.2f%%)\n", o->ios[i], perc);
	528	perc = ((float) o->merges[i] * 100.0) / (float) total;
	529	printf("\tmerges: %lu (perc=%3.2f%%)\n", o->merges[i], perc);
	530	perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
	531	printf("\tseq: %lu (perc=%3.2f%%)\n", o->seq[i], perc);
	532
	533	for (j = 0; j < o->nr_bs[i]; j++) {
	534	struct bs *bs = &o->bs[i][j];
	535
	536	perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
	537	printf("\tbs=%u, perc=%3.2f%%\n", bs->bs, perc);
	538	}
	539	}
	540
	541	printf("depth:\t%u\n", o->depth);
542	printf("usec:\t%llu (delay=%llu)\n", (o->last_ttime - o->first_ttime) / 1000ULL, (unsigned long long) o->start_delay);
543
544	printf("files:\t");
65ff9cd5 JA	545	for (i = 0; i < p->nr_files; i++)
65ff9cd5 JA	546	printf("%s,", p->files[i].name);
40bafa33 JA	547	printf("\n");
	548
	549	printf("\n");
	550	}
	551
	552	static int __output_p_fio(struct btrace_pid p, unsigned long ios)
	553	{
	554	struct btrace_out *o = &p->o;
	555	unsigned long total;
1a8cad44	556	unsigned long long time;
40bafa33 JA	557	float perc;
	558	int i, j;
	559
	560	if ((o->ios[0] + o->ios[1]) && o->ios[2]) {
	561	log_err("fio: trace has both read/write and trim\n");
	562	return 1;
	563	}
	564
	565	printf("[pid%u]\n", p->pid);
	566	printf("direct=1\n");
	567	if (o->depth == 1)
	568	printf("ioengine=sync\n");
	569	else
	570	printf("ioengine=libaio\niodepth=%u\n", o->depth);
	571
	572	if (o->ios[0] && !o->ios[1])
	573	printf("rw=randread\n");
	574	else if (!o->ios[0] && o->ios[1])
	575	printf("rw=randwrite\n");
	576	else if (o->ios[2])
	577	printf("rw=randtrim\n");
	578	else {
	579	printf("rw=randrw\n");
	580	total = ddir_rw_sum(o->ios);
	581	perc = ((float) o->ios[0] * 100.0) / (float) total;
	582	printf("rwmixread=%u\n", (int) (perc + 0.99));
	583	}
	584
35661615	585	printf("percentage_random=");
40bafa33 JA	586	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	587	if (o->seq[i] && o->ios[i]) {
	588	perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
	589	if (perc >= 99.0)
	590	perc = 100.0;
	591	} else
	592	perc = 100.0;
	593
	594	if (i)
	595	printf(",");
35661615	596	perc = 100.0 - perc;
40bafa33 JA	597	printf("%u", (int) perc);
	598	}
	599	printf("\n");
	600
	601	printf("filename=");
65ff9cd5	602	for (i = 0; i < p->nr_files; i++) {
40bafa33 JA	603	if (i)
40bafa33 JA	604	printf(":");
65ff9cd5	605	printf("%s", p->files[i].name);
40bafa33 JA	606	}
	607	printf("\n");
	608
	609	printf("startdelay=%llus\n", o->start_delay / 1000000ULL);
	610
1a8cad44 JA	611	time = o->last_ttime - o->first_ttime;
	612	time = (time + 1000000000ULL - 1) / 1000000000ULL;
	613	printf("runtime=%llus\n", time);
	614
40bafa33 JA	615	printf("bssplit=");
	616	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	617
	618	if (i && o->nr_bs[i - 1] && o->nr_bs[i])
	619	printf(",");
	620
	621	for (j = 0; j < o->nr_bs[i]; j++) {
	622	struct bs *bs = &o->bs[i][j];
	623
	624	perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
	625	if (perc < 1.00)
	626	continue;
	627	if (j)
	628	printf(":");
	629	if (j + 1 == o->nr_bs[i])
	630	printf("%u/", bs->bs);
	631	else
	632	printf("%u/%u", bs->bs, (int) perc);
	633	}
	634	}
	635	printf("\n\n");
	636
	637	return 0;
	638	}
	639
	640	static int __output_p(struct btrace_pid p, unsigned long ios)
	641	{
	642	struct btrace_out *o = &p->o;
	643	int i, ret = 0;
	644
	645	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	646	if (o->nr_bs[i] <= 1)
	647	continue;
	648	qsort(o->bs[i], o->nr_bs[i], sizeof(struct bs), bs_cmp);
	649	}
	650
35661615	651	if (filename) {
65ff9cd5 JA	652	p->files = malloc(sizeof(struct trace_file));
	653	p->nr_files++;
	654	p->files[0].name = filename;
35661615 JA	655	}
35661615 JA	656
40bafa33 JA	657	if (output_ascii)
	658	__output_p_ascii(p, ios);
	659	else
	660	ret = __output_p_fio(p, ios);
	661
	662	return ret;
	663	}
	664
	665	static int prune_entry(struct btrace_out *o)
	666	{
	667	uint64_t time;
	668
	669	if (ddir_rw_sum(o->ios) < ios_threshold)
	670	return 1;
	671
	672	time = (o->last_ttime - o->first_ttime) / 1000ULL;
	673	if (time < rt_threshold)
	674	return 1;
	675
	676	return 0;
	677	}
	678
	679	static int entry_cmp(void priv, struct flist_head a, struct flist_head *b)
	680	{
	681	struct btrace_pid *pa = flist_entry(a, struct btrace_pid, pid_list);
	682	struct btrace_pid *pb = flist_entry(b, struct btrace_pid, pid_list);
	683
	684	return ddir_rw_sum(pb->o.ios) - ddir_rw_sum(pa->o.ios);
	685	}
	686
17bf0853 JA	687	static void free_p(struct btrace_pid *p)
	688	{
	689	struct btrace_out *o = &p->o;
	690	int i;
	691
65ff9cd5 JA	692	for (i = 0; i < p->nr_files; i++) {
	693	if (p->files[i].name && p->files[i].name != filename)
	694	free(p->files[i].name);
17bf0853 JA	695	}
	696
	697	for (i = 0; i < DDIR_RWDIR_CNT; i++)
	698	free(o->bs[i]);
	699
65ff9cd5	700	free(p->files);
17bf0853 JA	701	flist_del(&p->pid_list);
	702	flist_del(&p->hash_list);
	703	free(p);
	704	}
	705
40bafa33 JA	706	static int output_p(void)
	707	{
	708	unsigned long ios[DDIR_RWDIR_CNT];
	709	struct flist_head e, tmp;
	710	int ret = 0;
	711
	712	flist_for_each_safe(e, tmp, &pid_list) {
	713	struct btrace_pid *p;
	714
	715	p = flist_entry(e, struct btrace_pid, pid_list);
	716	if (prune_entry(&p->o)) {
17bf0853	717	free_p(p);
40bafa33 JA	718	continue;
	719	}
	720	p->o.start_delay = (p->o.first_ttime / 1000ULL) - first_ttime;
	721	}
	722
	723	memset(ios, 0, sizeof(ios));
	724
	725	flist_sort(NULL, &pid_list, entry_cmp);
	726
	727	flist_for_each(e, &pid_list) {
	728	struct btrace_pid *p;
	729
	730	p = flist_entry(e, struct btrace_pid, pid_list);
	731	ret \|= __output_p(p, ios);
	732	}
	733
	734	if (output_ascii)
	735	printf("Total: reads=%lu, writes=%lu\n", ios[0], ios[1]);
	736
	737	return ret;
	738	}
	739
	740	static int usage(char *argv[])
	741	{
17bf0853 JA	742	log_err("%s: <blktrace bin file>\n", argv[0]);
	743	log_err("\t-t\tUsec threshold to ignore task\n");
	744	log_err("\t-n\tNumber IOS threshold to ignore task\n");
	745	log_err("\t-f\tFio job file output\n");
	746	log_err("\t-d\tUse this file/device for replay\n");
40bafa33 JA	747	return 1;
	748	}
	749
17bf0853	750	static int trace_needs_swap(const char trace_file, int swap)
40bafa33	751	{
40bafa33	752	struct blk_io_trace t;
17bf0853 JA	753	int fd, ret;
	754
	755	*swap = -1;
	756
	757	fd = open(trace_file, O_RDONLY);
	758	if (fd < 0) {
	759	perror("open");
	760	return 1;
	761	}
	762
	763	ret = read(fd, &t, sizeof(t));
	764	if (ret < 0) {
	765	perror("read");
	766	return 1;
	767	} else if (ret != sizeof(t)) {
	768	log_err("fio: short read on trace file\n");
	769	return 1;
	770	}
	771
	772	close(fd);
	773
	774	if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
	775	*swap = 0;
	776	else {
	777	/*
	778	* Maybe it needs to be endian swapped...
	779	*/
	780	t.magic = fio_swap32(t.magic);
	781	if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
	782	*swap = 1;
	783	}
	784
	785	if (*swap == -1) {
	786	log_err("fio: blktrace appears corrupt\n");
	787	return 1;
	788	}
	789
	790	return 0;
	791	}
	792
	793	int main(int argc, char *argv[])
	794	{
	795	int need_swap, i, c;
40bafa33 JA	796
	797	if (argc < 2)
	798	return usage(argv);
	799
35661615	800	while ((c = getopt(argc, argv, "t:n:fd:")) != -1) {
40bafa33 JA	801	switch (c) {
	802	case 't':
	803	rt_threshold = atoi(optarg);
	804	break;
	805	case 'n':
	806	ios_threshold = atoi(optarg);
	807	break;
	808	case 'f':
	809	output_ascii = 0;
	810	break;
35661615 JA	811	case 'd':
	812	filename = strdup(optarg);
	813	break;
40bafa33 JA	814	case '?':
	815	default:
	816	return usage(argv);
	817	}
	818	}
	819
	820	if (argc == optind)
	821	return usage(argv);
	822
17bf0853	823	if (trace_needs_swap(argv[optind], &need_swap))
40bafa33	824	return 1;
40bafa33 JA	825
	826	for (i = 0; i < PID_HASH_SIZE; i++)
	827	INIT_FLIST_HEAD(&pid_hash[i]);
17bf0853 JA	828	for (i = 0; i < INFLIGHT_HASH_SIZE; i++)
17bf0853 JA	829	INIT_FLIST_HEAD(&inflight_hash[i]);
40bafa33 JA	830
	831	load_blktrace(argv[optind], need_swap);
	832	first_ttime /= 1000ULL;
	833
	834	return output_p();
	835	}