#include <sys/stat.h>
#include <dirent.h>
-#include "list.h"
+#include "flist.h"
#include "fio.h"
#include "blktrace_api.h"
-#define TRACE_FIFO_SIZE 65536
+#define TRACE_FIFO_SIZE 8192
/*
* fifo refill frontend, to avoid reading data in trace sized bites
if (ret > 0)
ret = fifo_put(fifo, buf, ret);
+ dprint(FD_BLKTRACE, "refill: filled %d bytes\n", ret);
return ret;
}
if (t->pdu_len == 0)
return 0;
+ dprint(FD_BLKTRACE, "discard pdu len %u\n", t->pdu_len);
return trace_fifo_get(td, fifo, fd, NULL, t->pdu_len);
}
int fd, ret;
fd = open(filename, O_RDONLY);
- if (fd < 0) {
- perror("open blktrace");
+ if (fd < 0)
return 0;
- }
ret = read(fd, &t, sizeof(t));
close(fd);
continue;
if (maj == major(st.st_rdev) && min == minor(st.st_rdev)) {
+ dprint(FD_BLKTRACE, "device lookup: %d/%d\n", maj, min);
strcpy(path, full_path);
found = 1;
break;
return found;
}
-#define MINORBITS 20
-#define MINORMASK ((1U << MINORBITS) - 1)
-#define MAJOR(dev) ((unsigned int) ((dev) >> MINORBITS))
-#define MINOR(dev) ((unsigned int) ((dev) & MINORMASK))
+#define FMINORBITS 20
+#define FMINORMASK ((1U << FMINORBITS) - 1)
+#define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS))
+#define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK))
+
+static void trace_add_open_event(struct thread_data *td, int fileno)
+{
+ struct io_piece *ipo;
+
+ ipo = calloc(1, sizeof(*ipo));
+
+ ipo->ddir = DDIR_INVAL;
+ ipo->fileno = fileno;
+ ipo->file_action = FIO_LOG_OPEN_FILE;
+ flist_add_tail(&ipo->list, &td->io_log_list);
+}
static void trace_add_file(struct thread_data *td, __u32 device)
{
static unsigned int last_maj, last_min;
- unsigned int maj = MAJOR(device);
- unsigned int min = MINOR(device);
+ unsigned int maj = FMAJOR(device);
+ unsigned int min = FMINOR(device);
struct fio_file *f;
char dev[256];
unsigned int i;
return;
strcpy(dev, "/dev");
- if (lookup_device(dev, maj, min))
- add_file(td, dev);
+ if (lookup_device(dev, maj, min)) {
+ int fileno;
+
+ dprint(FD_BLKTRACE, "add devices %s\n", dev);
+ fileno = add_file(td, dev);
+ trace_add_open_event(td, fileno);
+ }
}
/*
struct io_piece *ipo = malloc(sizeof(*ipo));
memset(ipo, 0, sizeof(*ipo));
- INIT_LIST_HEAD(&ipo->list);
+ INIT_FLIST_HEAD(&ipo->list);
/*
* the 512 is wrong here, it should be the hardware sector size...
*/
else
ipo->ddir = DDIR_READ;
- list_add_tail(&ipo->list, &td->io_log_list);
+ dprint(FD_BLKTRACE, "store ddir=%d, off=%llu, len=%lu, delay=%lu\n",
+ ipo->ddir, ipo->offset,
+ ipo->len, ipo->delay);
+ queue_io_piece(td, ipo);
}
-/*
- * We only care for queue traces, most of the others are side effects
- * due to internal workings of the block layer.
- */
-static void handle_trace(struct thread_data *td, struct blk_io_trace *t,
- unsigned long long ttime, unsigned long *ios,
- unsigned int *bs)
+static void handle_trace_notify(struct blk_io_trace *t)
{
- int rw;
+ 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;
+ default:
+ dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action);
+ break;
+ }
+}
- if ((t->action & 0xffff) != __BLK_TA_QUEUE)
- return;
- if (t->action & BLK_TC_ACT(BLK_TC_PC))
- return;
- if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
- return;
+static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios,
+ unsigned int *bs)
+{
+ int rw;
trace_add_file(td, t->device);
store_ipo(td, t->sector, t->bytes, rw, ttime);
}
+/*
+ * We only care for queue traces, most of the others are side effects
+ * due to internal workings of the block layer.
+ */
+static void handle_trace(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios,
+ unsigned int *bs)
+{
+ if ((t->action & 0xffff) != __BLK_TA_QUEUE)
+ return;
+ if (t->action & BLK_TC_ACT(BLK_TC_PC))
+ return;
+
+ if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
+ handle_trace_notify(t);
+ else
+ handle_trace_fs(td, t, ttime, ios, bs);
+}
+
/*
* 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.
{
unsigned long long ttime, delay;
struct blk_io_trace t;
- unsigned long ios[2];
+ unsigned long ios[2], skipped_writes;
unsigned int cpu;
unsigned int rw_bs[2];
struct fifo *fifo;
ttime = 0;
ios[0] = ios[1] = 0;
rw_bs[0] = rw_bs[1] = 0;
+ skipped_writes = 0;
do {
int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t));
}
if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
- log_err("fio: bad magic in blktrace data: %x\n", t.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);
+ log_err("fio: bad blktrace version %d\n",
+ t.magic & 0xff);
goto err;
}
ret = discard_pdu(td, fifo, fd, &t);
log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
goto err;
}
- if (!ttime) {
+ if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) {
+ if (!ttime) {
+ ttime = t.time;
+ cpu = t.cpu;
+ }
+
+ delay = 0;
+ if (cpu == t.cpu)
+ delay = t.time - ttime;
+ if ((t.action & BLK_TC_ACT(BLK_TC_WRITE)) && read_only)
+ skipped_writes++;
+ else
+ handle_trace(td, &t, delay, ios, rw_bs);
+
ttime = t.time;
cpu = t.cpu;
+ } else {
+ delay = 0;
+ handle_trace(td, &t, delay, ios, rw_bs);
}
- delay = 0;
- if (cpu == t.cpu)
- delay = t.time - ttime;
- handle_trace(td, &t, delay, ios, rw_bs);
- ttime = t.time;
- cpu = t.cpu;
} while (1);
fifo_free(fifo);
close(fd);
+ if (skipped_writes)
+ log_err("fio: %s skips replay of %lu writes due to read-only\n",
+ td->o.name, skipped_writes);
+
if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) {
log_err("fio: found no ios in blktrace data\n");
return 1;