Re: [Qemu-devel] [PATCH v2 4/4] dma-helpers: Move reschedule_dma BH to blk's AioContext

[Fam Zheng <famz@redhat.com>]

On Fri, 03/13 09:13, Paolo Bonzini wrote:
> 
> 
> On 13/03/2015 02:38, Fam Zheng wrote:
> > That if the dbs' owner is an iothread, dma should be resumed on the right
> > thread. In this case it is the AioContext of the block device.
> > 
> > Signed-off-by: Fam Zheng <famz@redhat.com>
> > ---
> >  dma-helpers.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/dma-helpers.c b/dma-helpers.c
> > index 6918572..84f61a7 100644
> > --- a/dma-helpers.c
> > +++ b/dma-helpers.c
> > @@ -95,8 +95,10 @@ static void reschedule_dma(void *opaque)
> >  static void continue_after_map_failure(void *opaque)
> >  {
> >      DMAAIOCB *dbs = (DMAAIOCB *)opaque;
> > +    AioContext *ctx;
> >  
> > -    dbs->bh = qemu_bh_new(reschedule_dma, dbs);
> > +    ctx = blk_get_aio_context(dbs->blk);
> > +    dbs->bh = aio_bh_new(ctx, reschedule_dma, dbs);
> >      qemu_bh_schedule(dbs->bh);
> >  }
> >  
> > 
> 
> This looks good.  However, I wonder if dma_aio_cancel should also call
> cpu_unregister_map_client.  In this case, it's much better to just use a
> lock for the list (though you can still use atomics for the in-use flag).

The other possibility is grab a reference for the cpu_register_map_client call,
and release it in reschedule_dma. This way the atomics can keep, but we'll need
a "finished" flag in DMAAIOCB to avoid double completion.

Fam