[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 kb;

	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) {
			i->p->o.kb += (t->bytes >> 10);
			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, usec;
	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);
	usec = (o->last_ttime - o->first_ttime) / 1000ULL;
	printf("usec:\t%lu (delay=%llu)\n", usec, (unsigned long long) o->start_delay);
	printf("rate:\t%.2fKB/sec\n", ((float) o->kb * 1000.0) / ((float) usec / 1000.0));

	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;
cea475cd	49	uint64_t kb;
40bafa33	50
40bafa33 JA	51	uint64_t start_delay;
	52	};
	53
	54	struct btrace_pid {
	55	struct flist_head hash_list;
	56	struct flist_head pid_list;
	57	pid_t pid;
65ff9cd5 JA	58
	59	struct trace_file *files;
	60	int nr_files;
	61	unsigned int last_major, last_minor;
	62
40bafa33 JA	63	struct btrace_out o;
	64	};
	65
	66	struct inflight {
	67	struct flist_head list;
	68	struct btrace_pid *p;
	69	uint64_t end_sector;
	70	};
	71
	72	#define PID_HASH_BITS 10
	73	#define PID_HASH_SIZE (1U << PID_HASH_BITS)
	74
	75	static struct flist_head pid_hash[PID_HASH_SIZE];
	76	static FLIST_HEAD(pid_list);
	77
17bf0853 JA	78	#define INFLIGHT_HASH_BITS 8
	79	#define INFLIGHT_HASH_SIZE (1U << INFLIGHT_HASH_BITS)
	80	static struct flist_head inflight_hash[INFLIGHT_HASH_SIZE];
40bafa33 JA	81
	82	static uint64_t first_ttime = -1ULL;
	83
	84	static struct inflight *inflight_find(uint64_t sector)
	85	{
17bf0853	86	struct flist_head *inflight_list;
40bafa33 JA	87	struct flist_head *e;
40bafa33 JA	88
17bf0853 JA	89	inflight_list = &inflight_hash[hash_long(sector, INFLIGHT_HASH_BITS)];
	90
	91	flist_for_each(e, inflight_list) {
40bafa33 JA	92	struct inflight *i = flist_entry(e, struct inflight, list);
	93
	94	if (i->end_sector == sector)
	95	return i;
	96	}
	97
	98	return NULL;
	99	}
	100
	101	static void inflight_remove(struct inflight *i)
	102	{
	103	struct btrace_out *o = &i->p->o;
	104
	105	o->inflight--;
	106	assert(o->inflight >= 0);
	107	flist_del(&i->list);
	108	free(i);
	109	}
	110
17bf0853	111	static void __inflight_add(struct inflight *i)
40bafa33	112	{
17bf0853 JA	113	struct flist_head *list;
	114
	115	list = &inflight_hash[hash_long(i->end_sector, INFLIGHT_HASH_BITS)];
	116	flist_add_tail(&i->list, list);
40bafa33 JA	117	}
	118
	119	static void inflight_add(struct btrace_pid *p, uint64_t sector, uint32_t len)
	120	{
	121	struct btrace_out *o = &p->o;
	122	struct inflight *i;
	123
	124	i = calloc(1, sizeof(*i));
	125	i->p = p;
	126	o->inflight++;
	127	o->depth = max((int) o->depth, o->inflight);
	128	i->end_sector = sector + (len >> 9);
17bf0853 JA	129	__inflight_add(i);
	130	}
	131
	132	static void inflight_merge(struct inflight *i, int rw, unsigned int size)
	133	{
	134	i->p->o.merges[rw]++;
	135	if (size) {
	136	i->end_sector += (size >> 9);
	137	flist_del(&i->list);
	138	__inflight_add(i);
	139	}
40bafa33 JA	140	}
	141
	142	/*
	143	* fifo refill frontend, to avoid reading data in trace sized bites
	144	*/
	145	static int refill_fifo(struct fifo *fifo, int fd)
	146	{
	147	char buf[TRACE_FIFO_SIZE];
	148	unsigned int total;
	149	int ret;
	150
	151	total = sizeof(buf);
	152	if (total > fifo_room(fifo))
	153	total = fifo_room(fifo);
	154
	155	ret = read(fd, buf, total);
	156	if (ret < 0) {
	157	perror("read refill");
	158	return -1;
	159	}
	160
	161	if (ret > 0)
	162	ret = fifo_put(fifo, buf, ret);
	163
	164	return ret;
	165	}
	166
	167	/*
	168	* Retrieve 'len' bytes from the fifo, refilling if necessary.
	169	*/
	170	static int trace_fifo_get(struct fifo fifo, int fd, void buf,
	171	unsigned int len)
	172	{
	173	if (fifo_len(fifo) < len) {
	174	int ret = refill_fifo(fifo, fd);
	175
	176	if (ret < 0)
	177	return ret;
	178	}
	179
	180	return fifo_get(fifo, buf, len);
	181	}
	182
	183	/*
	184	* Just discard the pdu by seeking past it.
	185	*/
	186	static int discard_pdu(struct fifo fifo, int fd, struct blk_io_trace t)
	187	{
	188	if (t->pdu_len == 0)
	189	return 0;
	190
	191	return trace_fifo_get(fifo, fd, NULL, t->pdu_len);
	192	}
	193
	194	static void handle_trace_notify(struct blk_io_trace *t)
	195	{
	196	switch (t->action) {
	197	case BLK_TN_PROCESS:
	198	//printf("got process notify: %x, %d\n", t->action, t->pid);
	199	break;
	200	case BLK_TN_TIMESTAMP:
	201	//printf("got timestamp notify: %x, %d\n", t->action, t->pid);
	202	break;
	203	case BLK_TN_MESSAGE:
204	break;
205	default:
17bf0853	206	log_err("unknown trace act %x\n", t->action);
40bafa33 JA	207	break;
	208	}
	209	}
	210
	211	static void __add_bs(struct btrace_out *o, unsigned int len, int rw)
	212	{
	213	o->bs[rw] = realloc(o->bs[rw], (o->nr_bs[rw] + 1) * sizeof(struct bs));
	214	o->bs[rw][o->nr_bs[rw]].bs = len;
	215	o->bs[rw][o->nr_bs[rw]].nr = 1;
	216	o->nr_bs[rw]++;
	217	}
	218
	219	static void add_bs(struct btrace_out *o, unsigned int len, int rw)
	220	{
	221	struct bs *bs = o->bs[rw];
	222	int i;
	223
	224	if (!o->nr_bs[rw]) {
	225	__add_bs(o, len, rw);
	226	return;
	227	}
	228
	229	for (i = 0; i < o->nr_bs[rw]; i++) {
	230	if (bs[i].bs == len) {
	231	bs[i].nr++;
	232	return;
	233	}
	234	}
	235
	236	__add_bs(o, len, rw);
	237	}
	238
	239	#define FMINORBITS 20
	240	#define FMINORMASK ((1U << FMINORBITS) - 1)
	241	#define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS))
	242	#define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK))
	243
65ff9cd5	244	static void btrace_add_file(struct btrace_pid *p, uint32_t devno)
40bafa33 JA	245	{
	246	unsigned int maj = FMAJOR(devno);
	247	unsigned int min = FMINOR(devno);
	248	struct trace_file *f;
	249	unsigned int i;
	250	char dev[256];
	251
35661615 JA	252	if (filename)
35661615 JA	253	return;
65ff9cd5	254	if (p->last_major == maj && p->last_minor == min)
40bafa33 JA	255	return;
40bafa33 JA	256
65ff9cd5 JA	257	p->last_major = maj;
65ff9cd5 JA	258	p->last_minor = min;
40bafa33 JA	259
	260	/*
	261	* check for this file in our list
	262	*/
65ff9cd5 JA	263	for (i = 0; i < p->nr_files; i++) {
65ff9cd5 JA	264	f = &p->files[i];
40bafa33 JA	265
	266	if (f->major == maj && f->minor == min)
	267	return;
	268	}
	269
	270	strcpy(dev, "/dev");
	271	if (!blktrace_lookup_device(NULL, dev, maj, min)) {
	272	log_err("fio: failed to find device %u/%u\n", maj, min);
	273	return;
	274	}
	275
65ff9cd5 JA	276	p->files = realloc(p->files, (p->nr_files + 1) * sizeof(*f));
65ff9cd5 JA	277	f = &p->files[p->nr_files];
40bafa33 JA	278	f->name = strdup(dev);
	279	f->major = maj;
	280	f->minor = min;
65ff9cd5	281	p->nr_files++;
40bafa33 JA	282	}
40bafa33 JA	283
65ff9cd5	284	static void handle_trace_discard(struct blk_io_trace t, struct btrace_pid p)
40bafa33	285	{
65ff9cd5 JA	286	struct btrace_out *o = &p->o;
	287
	288	btrace_add_file(p, t->device);
40bafa33 JA	289
	290	if (o->first_ttime == -1ULL)
	291	o->first_ttime = t->time;
	292
	293	o->ios[DDIR_TRIM]++;
	294	add_bs(o, t->bytes, DDIR_TRIM);
	295	}
	296
65ff9cd5	297	static void handle_trace_fs(struct blk_io_trace t, struct btrace_pid p)
40bafa33	298	{
65ff9cd5	299	struct btrace_out *o = &p->o;
40bafa33 JA	300	int rw;
40bafa33 JA	301
65ff9cd5	302	btrace_add_file(p, t->device);
40bafa33 JA	303
	304	first_ttime = min(first_ttime, (uint64_t) t->time);
	305
	306	if (o->first_ttime == -1ULL)
	307	o->first_ttime = t->time;
	308
	309	rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	310
	311	add_bs(o, t->bytes, rw);
	312	o->ios[rw]++;
	313
	314	if (t->sector == o->last_end[rw] \|\| o->last_end[rw] == -1ULL)
	315	o->seq[rw]++;
	316
	317	o->last_end[rw] = t->sector + (t->bytes >> 9);
	318	}
	319
65ff9cd5	320	static void handle_queue_trace(struct blk_io_trace t, struct btrace_pid p)
40bafa33 JA	321	{
	322	if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
	323	handle_trace_notify(t);
	324	else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
65ff9cd5	325	handle_trace_discard(t, p);
40bafa33	326	else
65ff9cd5	327	handle_trace_fs(t, p);
40bafa33 JA	328	}
	329
	330	static void handle_trace(struct blk_io_trace t, struct btrace_pid p)
	331	{
	332	unsigned int act = t->action & 0xffff;
	333
	334	if (act == __BLK_TA_QUEUE) {
	335	inflight_add(p, t->sector, t->bytes);
65ff9cd5	336	handle_queue_trace(t, p);
40bafa33 JA	337	} else if (act == __BLK_TA_REQUEUE) {
	338	p->o.inflight--;
	339	} else if (act == __BLK_TA_BACKMERGE) {
	340	struct inflight *i;
	341
	342	i = inflight_find(t->sector + (t->bytes >> 9));
	343	if (i)
	344	inflight_remove(i);
	345
	346	i = inflight_find(t->sector);
	347	if (i) {
	348	int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	349
	350	inflight_merge(i, rw, t->bytes);
	351	}
	352	} else if (act == __BLK_TA_FRONTMERGE) {
	353	struct inflight *i;
	354
	355	i = inflight_find(t->sector + (t->bytes >> 9));
	356	if (i)
	357	inflight_remove(i);
	358
	359	i = inflight_find(t->sector);
	360	if (i) {
	361	int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
	362
	363	inflight_merge(i, rw, 0);
	364	}
	365	} else if (act == __BLK_TA_COMPLETE) {
	366	struct inflight *i;
	367
	368	i = inflight_find(t->sector + (t->bytes >> 9));
cea475cd JA	369	if (i) {
cea475cd JA	370	i->p->o.kb += (t->bytes >> 10);
40bafa33	371	inflight_remove(i);
cea475cd	372	}
40bafa33 JA	373	}
	374	}
	375
	376	static void byteswap_trace(struct blk_io_trace *t)
	377	{
	378	t->magic = fio_swap32(t->magic);
	379	t->sequence = fio_swap32(t->sequence);
	380	t->time = fio_swap64(t->time);
	381	t->sector = fio_swap64(t->sector);
	382	t->bytes = fio_swap32(t->bytes);
	383	t->action = fio_swap32(t->action);
	384	t->pid = fio_swap32(t->pid);
	385	t->device = fio_swap32(t->device);
	386	t->cpu = fio_swap32(t->cpu);
	387	t->error = fio_swap16(t->error);
	388	t->pdu_len = fio_swap16(t->pdu_len);
	389	}
	390
	391	static struct btrace_pid pid_hash_find(pid_t pid, struct flist_head list)
	392	{
	393	struct flist_head *e;
	394	struct btrace_pid *p;
	395
	396	flist_for_each(e, list) {
	397	p = flist_entry(e, struct btrace_pid, hash_list);
	398	if (p->pid == pid)
	399	return p;
	400	}
	401
	402	return NULL;
	403	}
	404
	405	static struct btrace_pid *pid_hash_get(pid_t pid)
	406	{
	407	struct flist_head *hash_list;
	408	struct btrace_pid *p;
	409
	410	hash_list = &pid_hash[hash_long(pid, PID_HASH_BITS)];
	411
	412	p = pid_hash_find(pid, hash_list);
	413	if (!p) {
	414	int i;
	415
	416	p = calloc(1, sizeof(*p));
	417	p->o.first_ttime = -1ULL;
	418	p->o.last_ttime = -1ULL;
	419
	420	for (i = 0; i < DDIR_RWDIR_CNT; i++)
	421	p->o.last_end[i] = -1ULL;
	422
	423	p->pid = pid;
	424	flist_add_tail(&p->hash_list, hash_list);
	425	flist_add_tail(&p->pid_list, &pid_list);
	426	}
	427
	428	return p;
	429	}
	430
	431	/*
	432	* Load a blktrace file by reading all the blk_io_trace entries, and storing
	433	* them as io_pieces like the fio text version would do.
	434	*/
	435	static int load_blktrace(const char *filename, int need_swap)
	436	{
437	struct btrace_pid *p;
438	unsigned long traces;
439	struct blk_io_trace t;
440	struct fifo *fifo;
441	int fd;
442
443	fd = open(filename, O_RDONLY);
444	if (fd < 0) {
445	perror("open trace file\n");
446	return 1;
447	}
448
449	fifo = fifo_alloc(TRACE_FIFO_SIZE);
450
451	traces = 0;
452	do {
453	int ret = trace_fifo_get(fifo, fd, &t, sizeof(t));
454
455	if (ret < 0)
456	goto err;
457	else if (!ret)
458	break;
459	else if (ret < (int) sizeof(t)) {
17bf0853	460	log_err("fio: short fifo get\n");
40bafa33 JA	461	break;
	462	}
	463
	464	if (need_swap)
	465	byteswap_trace(&t);
	466
	467	if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
17bf0853	468	log_err("fio: bad magic in blktrace data: %x\n", t.magic);
40bafa33 JA	469	goto err;
	470	}
	471	if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
17bf0853	472	log_err("fio: bad blktrace version %d\n", t.magic & 0xff);
40bafa33 JA	473	goto err;
	474	}
	475	ret = discard_pdu(fifo, fd, &t);
	476	if (ret < 0) {
17bf0853	477	log_err("blktrace lseek\n");
40bafa33 JA	478	goto err;
40bafa33 JA	479	} else if (t.pdu_len != ret) {
17bf0853	480	log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
40bafa33 JA	481	goto err;
	482	}
	483
	484	p = pid_hash_get(t.pid);
	485	handle_trace(&t, p);
	486	p->o.last_ttime = t.time;
	487	traces++;
	488	} while (1);
	489
	490	fifo_free(fifo);
	491	close(fd);
	492
	493	if (output_ascii)
	494	printf("Traces loaded: %lu\n", traces);
	495
	496	return 0;
	497	err:
	498	close(fd);
	499	fifo_free(fifo);
	500	return 1;
	501	}
	502
	503	static int bs_cmp(const void ba, const void bb)
	504	{
	505	const struct bs *bsa = ba;
	506	const struct bs *bsb = bb;
	507
	508	return bsb->nr - bsa->nr;
	509	}
	510
	511	static void __output_p_ascii(struct btrace_pid p, unsigned long ios)
	512	{
	513	const char *msg[] = { "reads", "writes", "trims" };
	514	struct btrace_out *o = &p->o;
cea475cd	515	unsigned long total, usec;
40bafa33 JA	516	int i, j;
	517
	518	printf("[pid:\t%u]\n", p->pid);
	519
	520	total = ddir_rw_sum(o->ios);
	521	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	522	float perc;
	523
	524	if (!o->ios[i])
	525	continue;
	526
	527	ios[i] += o->ios[i] + o->merges[i];
	528	printf("%s\n", msg[i]);
	529	perc = ((float) o->ios[i] * 100.0) / (float) total;
	530	printf("\tios: %lu (perc=%3.2f%%)\n", o->ios[i], perc);
	531	perc = ((float) o->merges[i] * 100.0) / (float) total;
	532	printf("\tmerges: %lu (perc=%3.2f%%)\n", o->merges[i], perc);
	533	perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
	534	printf("\tseq: %lu (perc=%3.2f%%)\n", o->seq[i], perc);
	535
	536	for (j = 0; j < o->nr_bs[i]; j++) {
	537	struct bs *bs = &o->bs[i][j];
	538
	539	perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
	540	printf("\tbs=%u, perc=%3.2f%%\n", bs->bs, perc);
	541	}
	542	}
	543
	544	printf("depth:\t%u\n", o->depth);
cea475cd JA	545	usec = (o->last_ttime - o->first_ttime) / 1000ULL;
	546	printf("usec:\t%lu (delay=%llu)\n", usec, (unsigned long long) o->start_delay);
	547	printf("rate:\t%.2fKB/sec\n", ((float) o->kb * 1000.0) / ((float) usec / 1000.0));
40bafa33 JA	548
40bafa33 JA	549	printf("files:\t");
65ff9cd5 JA	550	for (i = 0; i < p->nr_files; i++)
65ff9cd5 JA	551	printf("%s,", p->files[i].name);
40bafa33 JA	552	printf("\n");
	553
	554	printf("\n");
	555	}
	556
	557	static int __output_p_fio(struct btrace_pid p, unsigned long ios)
	558	{
	559	struct btrace_out *o = &p->o;
	560	unsigned long total;
1a8cad44	561	unsigned long long time;
40bafa33 JA	562	float perc;
	563	int i, j;
	564
	565	if ((o->ios[0] + o->ios[1]) && o->ios[2]) {
	566	log_err("fio: trace has both read/write and trim\n");
	567	return 1;
	568	}
	569
	570	printf("[pid%u]\n", p->pid);
	571	printf("direct=1\n");
	572	if (o->depth == 1)
	573	printf("ioengine=sync\n");
	574	else
	575	printf("ioengine=libaio\niodepth=%u\n", o->depth);
	576
	577	if (o->ios[0] && !o->ios[1])
	578	printf("rw=randread\n");
	579	else if (!o->ios[0] && o->ios[1])
	580	printf("rw=randwrite\n");
	581	else if (o->ios[2])
	582	printf("rw=randtrim\n");
	583	else {
	584	printf("rw=randrw\n");
	585	total = ddir_rw_sum(o->ios);
	586	perc = ((float) o->ios[0] * 100.0) / (float) total;
	587	printf("rwmixread=%u\n", (int) (perc + 0.99));
	588	}
	589
35661615	590	printf("percentage_random=");
40bafa33 JA	591	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	592	if (o->seq[i] && o->ios[i]) {
	593	perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
	594	if (perc >= 99.0)
	595	perc = 100.0;
	596	} else
	597	perc = 100.0;
	598
	599	if (i)
	600	printf(",");
35661615	601	perc = 100.0 - perc;
40bafa33 JA	602	printf("%u", (int) perc);
	603	}
	604	printf("\n");
	605
	606	printf("filename=");
65ff9cd5	607	for (i = 0; i < p->nr_files; i++) {
40bafa33 JA	608	if (i)
40bafa33 JA	609	printf(":");
65ff9cd5	610	printf("%s", p->files[i].name);
40bafa33 JA	611	}
	612	printf("\n");
	613
	614	printf("startdelay=%llus\n", o->start_delay / 1000000ULL);
	615
1a8cad44 JA	616	time = o->last_ttime - o->first_ttime;
	617	time = (time + 1000000000ULL - 1) / 1000000000ULL;
	618	printf("runtime=%llus\n", time);
	619
40bafa33 JA	620	printf("bssplit=");
	621	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	622
	623	if (i && o->nr_bs[i - 1] && o->nr_bs[i])
	624	printf(",");
	625
	626	for (j = 0; j < o->nr_bs[i]; j++) {
	627	struct bs *bs = &o->bs[i][j];
	628
	629	perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
	630	if (perc < 1.00)
	631	continue;
	632	if (j)
	633	printf(":");
	634	if (j + 1 == o->nr_bs[i])
	635	printf("%u/", bs->bs);
	636	else
	637	printf("%u/%u", bs->bs, (int) perc);
	638	}
	639	}
	640	printf("\n\n");
	641
	642	return 0;
	643	}
	644
	645	static int __output_p(struct btrace_pid p, unsigned long ios)
	646	{
	647	struct btrace_out *o = &p->o;
	648	int i, ret = 0;
	649
	650	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
	651	if (o->nr_bs[i] <= 1)
	652	continue;
	653	qsort(o->bs[i], o->nr_bs[i], sizeof(struct bs), bs_cmp);
	654	}
	655
35661615	656	if (filename) {
65ff9cd5 JA	657	p->files = malloc(sizeof(struct trace_file));
	658	p->nr_files++;
	659	p->files[0].name = filename;
35661615 JA	660	}
35661615 JA	661
40bafa33 JA	662	if (output_ascii)
	663	__output_p_ascii(p, ios);
	664	else
	665	ret = __output_p_fio(p, ios);
	666
	667	return ret;
	668	}
	669
	670	static int prune_entry(struct btrace_out *o)
	671	{
	672	uint64_t time;
	673
	674	if (ddir_rw_sum(o->ios) < ios_threshold)
	675	return 1;
	676
	677	time = (o->last_ttime - o->first_ttime) / 1000ULL;
	678	if (time < rt_threshold)
	679	return 1;
	680
	681	return 0;
	682	}
	683
	684	static int entry_cmp(void priv, struct flist_head a, struct flist_head *b)
	685	{
	686	struct btrace_pid *pa = flist_entry(a, struct btrace_pid, pid_list);
	687	struct btrace_pid *pb = flist_entry(b, struct btrace_pid, pid_list);
	688
	689	return ddir_rw_sum(pb->o.ios) - ddir_rw_sum(pa->o.ios);
	690	}
	691
17bf0853 JA	692	static void free_p(struct btrace_pid *p)
	693	{
	694	struct btrace_out *o = &p->o;
	695	int i;
	696
65ff9cd5 JA	697	for (i = 0; i < p->nr_files; i++) {
	698	if (p->files[i].name && p->files[i].name != filename)
	699	free(p->files[i].name);
17bf0853 JA	700	}
	701
	702	for (i = 0; i < DDIR_RWDIR_CNT; i++)
	703	free(o->bs[i]);
	704
65ff9cd5	705	free(p->files);
17bf0853 JA	706	flist_del(&p->pid_list);
	707	flist_del(&p->hash_list);
	708	free(p);
	709	}
	710
40bafa33 JA	711	static int output_p(void)
	712	{
	713	unsigned long ios[DDIR_RWDIR_CNT];
	714	struct flist_head e, tmp;
	715	int ret = 0;
	716
	717	flist_for_each_safe(e, tmp, &pid_list) {
	718	struct btrace_pid *p;
	719
	720	p = flist_entry(e, struct btrace_pid, pid_list);
	721	if (prune_entry(&p->o)) {
17bf0853	722	free_p(p);
40bafa33 JA	723	continue;
	724	}
	725	p->o.start_delay = (p->o.first_ttime / 1000ULL) - first_ttime;
	726	}
	727
	728	memset(ios, 0, sizeof(ios));
	729
	730	flist_sort(NULL, &pid_list, entry_cmp);
	731
	732	flist_for_each(e, &pid_list) {
	733	struct btrace_pid *p;
	734
	735	p = flist_entry(e, struct btrace_pid, pid_list);
	736	ret \|= __output_p(p, ios);
	737	}
	738
	739	if (output_ascii)
	740	printf("Total: reads=%lu, writes=%lu\n", ios[0], ios[1]);
	741
	742	return ret;
	743	}
	744
	745	static int usage(char *argv[])
	746	{
17bf0853 JA	747	log_err("%s: <blktrace bin file>\n", argv[0]);
	748	log_err("\t-t\tUsec threshold to ignore task\n");
	749	log_err("\t-n\tNumber IOS threshold to ignore task\n");
	750	log_err("\t-f\tFio job file output\n");
	751	log_err("\t-d\tUse this file/device for replay\n");
40bafa33 JA	752	return 1;
	753	}
	754
17bf0853	755	static int trace_needs_swap(const char trace_file, int swap)
40bafa33	756	{
40bafa33	757	struct blk_io_trace t;
17bf0853 JA	758	int fd, ret;
	759
	760	*swap = -1;
	761
	762	fd = open(trace_file, O_RDONLY);
	763	if (fd < 0) {
	764	perror("open");
	765	return 1;
	766	}
	767
	768	ret = read(fd, &t, sizeof(t));
	769	if (ret < 0) {
	770	perror("read");
	771	return 1;
	772	} else if (ret != sizeof(t)) {
	773	log_err("fio: short read on trace file\n");
	774	return 1;
	775	}
	776
	777	close(fd);
	778
	779	if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
	780	*swap = 0;
	781	else {
	782	/*
	783	* Maybe it needs to be endian swapped...
	784	*/
	785	t.magic = fio_swap32(t.magic);
	786	if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
	787	*swap = 1;
	788	}
	789
	790	if (*swap == -1) {
	791	log_err("fio: blktrace appears corrupt\n");
	792	return 1;
	793	}
	794
	795	return 0;
	796	}
	797
	798	int main(int argc, char *argv[])
	799	{
	800	int need_swap, i, c;
40bafa33 JA	801
	802	if (argc < 2)
	803	return usage(argv);
	804
35661615	805	while ((c = getopt(argc, argv, "t:n:fd:")) != -1) {
40bafa33 JA	806	switch (c) {
	807	case 't':
	808	rt_threshold = atoi(optarg);
	809	break;
	810	case 'n':
	811	ios_threshold = atoi(optarg);
	812	break;
	813	case 'f':
	814	output_ascii = 0;
	815	break;
35661615 JA	816	case 'd':
	817	filename = strdup(optarg);
	818	break;
40bafa33 JA	819	case '?':
	820	default:
	821	return usage(argv);
	822	}
	823	}
	824
	825	if (argc == optind)
	826	return usage(argv);
	827
17bf0853	828	if (trace_needs_swap(argv[optind], &need_swap))
40bafa33	829	return 1;
40bafa33 JA	830
	831	for (i = 0; i < PID_HASH_SIZE; i++)
	832	INIT_FLIST_HEAD(&pid_hash[i]);
17bf0853 JA	833	for (i = 0; i < INFLIGHT_HASH_SIZE; i++)
17bf0853 JA	834	INIT_FLIST_HEAD(&inflight_hash[i]);
40bafa33 JA	835
	836	load_blktrace(argv[optind], need_swap);
	837	first_ttime /= 1000ULL;
	838
	839	return output_p();
	840	}