> On Wed, Aug 13, 2008 at 09:45:54AM +0900, FUJITA Tomonori wrote:
> > On Tue, 12 Aug 2008 17:24:11 +0200
> > Joerg Roedel <joerg.roedel@amd.com> wrote:
> >
> > > Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> > > ---
> > > arch/x86/kernel/pci-gart_64.c | 21 +++++++++++++++++++++
> > > 1 files changed, 21 insertions(+), 0 deletions(-)
> > >
> > > diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
> > > index cdab678..55cc388 100644
> > > --- a/arch/x86/kernel/pci-gart_64.c
> > > +++ b/arch/x86/kernel/pci-gart_64.c
> > > @@ -499,6 +499,26 @@ error:
> > > return 0;
> > > }
> > >
> > > +/* allocate and map a coherent mapping */
> > > +static void *
> > > +gart_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_addr,
> > > + gfp_t flag)
> > > +{
> > > + void *vaddr;
> > > +
> > > + vaddr = (void *)__get_free_pages(flag, get_order(size));
> > > + if (!vaddr)
> > > + return NULL;
> > > +
> > > + *dma_addr = gart_map_single(dev, __pa(vaddr), size, DMA_BIDIRECTIONAL);
> > > + if (*dma_addr != bad_dma_address)
> > > + return vaddr;
> > > +
> > > + free_pages((unsigned long)vaddr, get_order(size));
> > > +
> > > + return NULL;
> > > +}
> > > +
> > > static int no_agp;
> >
> > It would be better to return a size-aligned memory as DMA-mapping.txt
> > says (though I don't think that it doesn't matter much):
> >
> >
http://lkml.org/lkml/2008/8/8/555
>
> Agreed. I try to change the patchset so it returns size aligned dma
> addresses.
>
> > I also think that x86 IOMMUs need to handle DMA_*BIT_MASK properly,
> > don't we?
>
> Shouldn't this be done by the IOMMUs using your iommu_area_alloc()
> function? Or do I misunderstand something?
