On 06/08/2016 08:10 AM, Kevin Wolf wrote:
> The raw-posix block driver actually supports byte-aligned requests now
> on non-O_DIRECT images, like it already (and previously incorrectly)
> claimed in bs->request_alignment.
> 
> For some block drivers this means that a RMW cycle can be avoided when
> they write sub-sector metadata e.g. for cluster allocation.

[well, there's still probably a RMW going on, but it's being done by the
kernel, rather than qemu - and choice of caching may let the kernel
optimize things... not worth cluttering the commit message with this,
though]

> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---

> +++ b/block/linux-aio.c
> @@ -272,14 +272,12 @@ static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
>  }
>  
>  int laio_submit_co(BlockDriverState *bs, LinuxAioState *s, int fd,
> -        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, int type)
> +                   uint64_t offset, QEMUIOVector *qiov, int type)
>  {
> -    off_t offset = sector_num * 512;
>      int ret;
> -
>      struct qemu_laiocb laiocb = {
>          .co         = qemu_coroutine_self(),
> -        .nbytes     = nb_sectors * 512,
> +        .nbytes     = qiov->size,

So for this interface, we require non-NULL qiov and no duplicated
length; I guess it isn't used for write_zeroes.  We may still want to do
some consistency sweep to decide what level of NULL-ness we want for
representing write_zeroes, rather than ad hoc decisions at each layer of
the call stack, but that's a task for another day.

> @@ -1344,26 +1344,27 @@ static int coroutine_fn raw_co_rw(BlockDriverState *bs, int64_t sector_num,
>              type |= QEMU_AIO_MISALIGNED;
>  #ifdef CONFIG_LINUX_AIO
>          } else if (s->use_aio) {
> -            return laio_submit_co(bs, s->aio_ctx, s->fd, sector_num, qiov,
> -                               nb_sectors, type);
> +            assert(qiov->size == bytes);

Worth hoisting the assertion outside of the #ifdef?...

> +            return laio_submit_co(bs, s->aio_ctx, s->fd, offset, qiov, type);
>  #endif
>          }
>      }
>  
> -    return paio_submit_co(bs, s->fd, sector_num * BDRV_SECTOR_SIZE, qiov,
> -                          nb_sectors * BDRV_SECTOR_SIZE, type);
> +    return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);

...then again, paio_submit_co() also does the assert - and this is more
evidence of our inconsistency on whether we duplicate a separate bytes
parameter or reuse qiov->size.

>  
> -static int coroutine_fn raw_co_readv(BlockDriverState *bs, int64_t sector_num,
> -                                     int nb_sectors, QEMUIOVector *qiov)
> +static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
> +                                      uint64_t bytes, QEMUIOVector *qiov,
> +                                      int flags)
>  {
> -    return raw_co_rw(bs, sector_num, nb_sectors, qiov, QEMU_AIO_READ);
> +    return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);

We ignore flags, but that's not a change in semantics.  (Maybe someday
we need .supported_read_flags)

>  }
>  
> -static int coroutine_fn raw_co_writev(BlockDriverState *bs, int64_t sector_num,
> -                                      int nb_sectors, QEMUIOVector *qiov)
> +static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
> +                                       uint64_t bytes, QEMUIOVector *qiov,
> +                                       int flags)
>  {
> -    return raw_co_rw(bs, sector_num, nb_sectors, qiov, QEMU_AIO_WRITE);
> +    return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);

And here, we could assert(!flags) (since we intentionally don't set
.supported_write_flags) - but I won't insist.

None of my comments require a code change, other than a possible added
assertion, so:

Reviewed-by: Eric Blake <eblake@redhat.com>

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org