On 01/04/2016 11:46, Fam Zheng wrote:
> +
> +static void bdrv_co_drain_bh_cb(void *opaque)
> +{
> + BdrvCoDrainData *data = opaque;
> + Coroutine *co = data->co;
> +
> + bdrv_drain(data->bs);
> + data->done = true;
> + qemu_coroutine_enter(co, NULL);
> +}
> +
> +static void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
> +{
> + QEMUBH *bh;
> + BdrvCoDrainData data;
> +
> + assert(qemu_in_coroutine());
> + data = (BdrvCoDrainData) {
> + .co = qemu_coroutine_self(),
> + .bs = bs,
> + .done = false,
> + };
> + bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data),
> + qemu_bh_schedule(bh);
> +
> + do {
> + qemu_coroutine_yield();
> + } while (!data.done);
The loop and "done" is not necessary. Also,
> + qemu_bh_delete(bh);
this can be moved to bdrv_co_drain_bh_cb before bdrv_drain, so that the
bottom half doesn't slow down the event loop until bdrv_drain completes.
Paolo
> +}
