Re: [Qemu-devel] [PATCH 04/11] aio: introduce aio_co_schedule

[Stefan Hajnoczi]

On Tue, May 17, 2016 at 04:57:28PM +0200, Paolo Bonzini wrote:
> That said, there is at least another case where it will be used.  In the
> dataplane branch, where AIO callbacks take the AioContext mutex
> themselves, we have:
> 
> static void bdrv_co_io_em_complete(void *opaque, int ret)
> {
>     CoroutineIOCompletion *co = opaque;
> 
>     co->ret = ret;
>     aio_context_acquire(co->ctx);
>     qemu_coroutine_enter(co->coroutine, NULL);
>     aio_context_release(co->ctx);
> }
> 
> ...
> 
>     acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
>                                   bdrv_co_io_em_complete, &co);
>     qemu_coroutine_yield();
> 
> bdrv_co_io_em_complete here can be called before the coroutine has
> yielded.  To prepare for the replacement of the AioContext mutex with
> fine-grained mutexes, I think bdrv_co_io_em_complete should do something
> like
> 
>     if (ctx != qemu_get_current_aio_context()) {
>         aio_co_schedule(ctx, co->coroutine);
>         return;
>     }
> 
>     aio_context_acquire(ctx);
>     qemu_coroutine_enter(co->coroutine, NULL);
>     aio_context_release(ctx);

Okay, that makes sense.

signature.asc
Description: PGP signature