Instead just loop on submit, since there are no events for us
to reap. This is usually a kernel bug, violating the principle
of forward progress guarantee.
If we can't submit anything in 30 seconds, error out.
Signed-off-by: Jens Axboe <axboe@fb.com>
struct libaio_data *ld = td->io_ops->data;
struct iocb **iocbs;
struct io_u **io_us;
struct libaio_data *ld = td->io_ops->data;
struct iocb **iocbs;
struct io_u **io_us;
+ struct timeval tv;
+ int ret, wait_start = 0;
if (!ld->queued)
return 0;
if (!ld->queued)
return 0;
/*
* If we get EAGAIN, we should break out without
* error and let the upper layer reap some
/*
* If we get EAGAIN, we should break out without
* error and let the upper layer reap some
+ * events for us. If we have no queued IO, we
+ * must loop here. If we loop for more than 30s,
+ * just error out, something must be buggy in the
+ * IO path.
+ if (ld->queued) {
+ ret = 0;
+ break;
+ }
+ if (!wait_start) {
+ fio_gettime(&tv, NULL);
+ wait_start = 0;
+ } else if (mtime_since_now(&tv) > 30000) {
+ log_err("fio: aio appears to be stalled, giving up\n");
+ break;
+ }
+ usleep(1);
+ continue;
} else
break;
} while (ld->head != ld->tail);
} else
break;
} while (ld->head != ld->tail);