Xen project Mailing List

I'm doing some research on xen 4.1 platform.I have two question which I'm trying to make it convenient for every expert to answer.

Question 1:

Here are some codes in the file xen-4.1.2/tools/ioemu-qemu-xen/block-raw-posix.c,

....

629 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,

630 int64_t sector_num, uint8_t *buf, int nb_sectors,

631 BlockDriverCompletionFunc *cb, void *opaque)

632 {

633 RawAIOCB *acb;

634

635 /*

636 * If O_DIRECT is used and the buffer is not aligned fall back

637 * to synchronous IO.

638 */

639 BDRVRawState *s = bs->opaque;

640

641 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {

642 QEMUBH *bh;

643 acb = qemu_aio_get(bs, cb, opaque);

644 acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);

645 bh = qemu_bh_new(raw_aio_em_cb, acb);

646 qemu_bh_schedule(bh);

647 return &acb->common;

648 }

649

650 acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);

651 if (!acb)

652 return NULL;

653 if (qemu_paio_read(&acb->aiocb) < 0) {

654 raw_aio_remove(acb);

655 return NULL;

656 }

657 return &acb->common;

658 }

......

In the notes between the codes , the sentence show that If O_DIRECT is used and the buffer is not aligned fall back to synchronous IO. What does it mean?I just want to make these function returning with the exact data in its field which the parameter buf point to , that is to say, I want to make the asynchronous read into a synchronous one.I have make the O_DIRECT used in file xenstore.c and the flags will be in effect along the way to this function above.

Question 2:

Here are some codes in /xen-4.1.2/tools/block-qcow.c

...

594 static void qcow_aio_read_cb(void *opaque, int ret)

595 {

596 QCowAIOCB *acb = opaque;

597 BlockDriverState *bs = acb->common.bs;

598 BDRVQcowState *s = bs->opaque;

599 int index_in_cluster;

600

601 acb->hd_aiocb = NULL;

602 if (ret < 0) {

603 fail:

604 acb->common.cb(acb->common.opaque, ret);

605 qemu_aio_release(acb);

606 return;

607 }

.........................

642 if (!acb->cluster_offset) {

643 if (bs->backing_hd) {

644 /* read from the base image */

645 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd,

646 acb->sector_num, acb->buf, acb->n, qcow_aio_read_cb, acb);

647 if (acb->hd_aiocb == NULL)

648 goto fail;

649 } else {

650 /* Note: in this case, no need to wait */

651 memset(acb->buf, 0, 512 * acb->n);

652 goto redo;

653 }

654 } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {

655 /* add AIO support for compressed blocks ? */

656 if (decompress_cluster(s, acb->cluster_offset) < 0)

657 goto fail;

658 memcpy(acb->buf,

659 s->cluster_cache + index_in_cluster * 512, 512 * acb->n);

660 goto redo;

661 } else {

662 if ((acb->cluster_offset & 511) != 0) {

663 ret = -EIO;

664 goto fail;

665 }

666 acb->hd_aiocb = bdrv_aio_read(s->hd,

667 (acb->cluster_offset >> 9) + index_in_cluster,

668 acb->buf, acb->n, qcow_aio_read_cb, acb);

In the code note on the line 650, it says that "in this case, no need to wait".

what does it mean? Here the code just would wait for what?

If I have some time ,I will learn the whole direction about how to ask a smart question which a expert send to me in a earlier mail.

Thanks.

A Newbie

[Xen-devel] !!!!!two pressing questions in the xen code notes(I'm dying!!!HELP!)