Re: [Qemu-devel] [RFC PATCH v2 6/8] dma: Implement .cancel_async

[Fam Zheng <famz@redhat.com>]

On Tue, 08/26 10:46, Paolo Bonzini wrote:
> Il 26/08/2014 08:08, Fam Zheng ha scritto:
> > +    if (dbs->cancelled) {
> > +        ret = -ECANCELED;
> > +    }
> 
> Why is dbs->cancelled necessary?

Request may complete after bdrv_aio_cancel_async with other status, this flag
is checked to fix the status to -ECANCELED.

> 
> >      dma_bdrv_unmap(dbs);
> >      if (dbs->common.cb) {
> >          dbs->common.cb(dbs->common.opaque, ret);
> > @@ -141,6 +148,9 @@ static void dma_bdrv_cb(void *opaque, int ret)
> >  
> >      trace_dma_bdrv_cb(dbs, ret);
> >  
> > +    if (dbs->cancelled) {
> > +        ret = -ECANCELED;
> > +    }
> >      dbs->acb = NULL;
> >      dbs->sector_num += dbs->iov.size / 512;
> >  
> > @@ -185,6 +195,7 @@ static void dma_aio_cancel(BlockDriverAIOCB *acb)
> >  
> >      trace_dma_aio_cancel(dbs);
> >  
> > +    dbs->cancelled = true;
> >      if (dbs->acb) {
> >          BlockDriverAIOCB *acb = dbs->acb;
> >          dbs->acb = NULL;
> > @@ -196,9 +207,25 @@ static void dma_aio_cancel(BlockDriverAIOCB *acb)
> >      dma_complete(dbs, 0);
> >  }
> >  
> > +static void dma_aio_cancel_async(BlockDriverAIOCB *acb)
> > +{
> > +    DMAAIOCB *dbs = container_of(acb, DMAAIOCB, common);
> > +
> > +    trace_dma_aio_cancel(dbs);
> > +
> > +    dbs->cancelled = true;
> > +    if (dbs->acb) {
> > +        acb = dbs->acb;
> > +        dbs->acb = NULL;
> 
> Why do you need to set dbs->acb to NULL, since the callback is going to
> be called?

Just copied from dma_aio_cancel, but seems not particularly useful. It reminds
me that the one in dma_aio_cancel also looks suspicious.

Fam