Re: [Qemu-devel] [PATCH v3 2/8] block: Add bdrv_aio_cancel_async

[Stefan Hajnoczi]

On Wed, Aug 27, 2014 at 10:49:10AM +0800, Fam Zheng wrote:
> +/* Async version of aio cancel. The caller is not blocked if the acb 
> implements
> + * cancel_async, otherwise fall back to bdrv_aio_cancel. In both cases, 
> acb->cb
> + * is guarenteed to be called, before or after function returns. */
> +void bdrv_aio_cancel_async(BlockDriverAIOCB *acb)
> +{
> +    if (acb->aiocb_info->cancel_async) {
> +        acb->aiocb_info->cancel_async(acb);
> +    } else {
> +        /* Mask the cb and cancel, we'll call it manually once the 
> synchronous
> +         * cancel is done. */
> +        BlockDriverCompletionFunc *cb = acb->cb;
> +        void *opaque = acb->opaque;
> +        acb->cb = bdrv_aio_cancel_cb_nop;
> +        acb->opaque = NULL;
> +        qemu_aio_ref(acb);
> +        acb->aiocb_info->cancel(acb);
> +        cb(opaque, -ECANCELED);
> +        qemu_aio_release(acb);
> +    }
> +}

It is not totally obvious why we hijack the callback.  If you respin,
please rephrase the comment along the lines of:

/* bdrv_aio_cancel() does not guarantee to invoke cb() so mask it during
 * bdrv_aio_cancel() and always invoke it ourselves.
 */

Stefan

pgpHEXFErO_Qk.pgp
Description: PGP signature