[PATCHv2 3/4] mm: vmalloc: add VM_DMA flag to indicate areas used by dma-mapping framework

[konrad.wilk@oracle.com (Konrad Rzeszutek Wilk)]

On Mon, May 28, 2012 at 10:19:39AM +0200, Marek Szyprowski wrote:
> Hello,
> 
> On Sunday, May 27, 2012 2:35 PM KOSAKI Motohiro wrote:
> 
> > On Thu, May 24, 2012 at 8:28 AM, Paul Mundt <lethal@linux-sh.org> wrote:
> > > On Thu, May 24, 2012 at 02:26:12PM +0200, Marek Szyprowski wrote:
> > >> On Tuesday, May 22, 2012 9:08 AM Minchan Kim wrote:
> > >> > Hmm, VM_DMA would become generic flag?
> > >> > AFAIU, maybe VM_DMA would be used only on ARM arch.
> > >>
> > >> Right now yes, it will be used only on ARM architecture, but maybe other architecture will
> > >> start using it once it is available.
> > >>
> > > There's very little about the code in question that is ARM-specific to
> > > begin with. I plan to adopt similar changes on SH once the work has
> > > settled one way or the other, so we'll probably use the VMA flag there,
> > > too.
> > 
> > I don't think VM_DMA is good idea because x86_64 has two dma zones. x86 unaware
> > patches make no sense.
> 
> I see no problems to add VM_DMA64 later if x86_64 starts using vmalloc areas for creating 
> kernel mappings for the dma buffers (I assume that there are 2 dma zones: one 32bit and one
> 64bit). Right now x86 and x86_64 don't use vmalloc areas for dma buffers, so I hardly see
> how this patch can be considered as 'x86 unaware'.

Well they do - kind off. It is usually done by calling vmalloc_32 and then using
the DMA API on top of those pages (or sometimes the non-portable virt_to_phys macro).

Introducing this and replacing the vmalloc_32 with this seems like a nice step in making
those device drivers APIs more portable?