Re: [PATCH v6 10/19] xen: introduce xen_alloc/free_coherent_pages

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

On Wed, 2 Oct 2013, Catalin Marinas wrote:
> On Wed, Oct 02, 2013 at 06:03:46PM +0100, Stefano Stabellini wrote:
> > On Tue, 1 Oct 2013, Catalin Marinas wrote:
> > > On 30 Sep 2013, at 16:31, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> > > > On Fri, Sep 27, 2013 at 05:09:58PM +0100, Stefano Stabellini wrote:
> > > >> xen_swiotlb_alloc_coherent needs to allocate a coherent buffer for cpu
> > > >> and devices. On native x86 and ARMv8 is sufficient to call
> > > >> __get_free_pages in order to get a coherent buffer, while on ARM we need
> > > >> to call the native dma_ops->alloc implementation.
> > > >> 
> > > >> When arm64 stops using the swiotlb by default and starts having multiple
> > > >> dma_ops implementations, we'll use __get_dma_ops there too.
> > > > 
> > > > I presume this is a future TODO, not some further patch (in which
> > > > case you should say in here the title of it). If it is a TODO could
> > > > you stick that in the sentence here somewhere to make it crytal clear that
> > > > it is not implemented.
> > > 
> > > Thanks for pointing this out.  The arm64 patch shouldn't be different
> > > from arm here.  I thought a subsequent patch already added
> > > __get_dma_ops() for arm64.
> > 
> > Yes, the arm64 patch already added __get_dma_ops.
> > 
> > However I am not using __get_dma_ops to implement
> > xen_alloc_coherent_pages for arm64 because I can just use
> > __get_free_pages for the moment:
> 
> So why do the work twice when we'll get coherency maintenance in the DMA
> ops?

Yeah, good point.