Re: [PATCH] intel-iommu: Obey coherent_dma_mask for alloc_coherent on passthrough

[Alex Williamson <alex.williamson@hp.com>]

On Fri, 2009-11-06 at 11:41 +0900, FUJITA Tomonori wrote:
> On Wed, 04 Nov 2009 15:59:34 -0700
> Alex Williamson <alex.williamson@hp.com> wrote:
> > @@ -2582,7 +2582,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
> >  	BUG_ON(dir == DMA_NONE);
> >  
> >  	if (iommu_no_mapping(hwdev))
> > -		return paddr;
> > +		return paddr + size > dma_mask ? 0 : paddr;
> 
> You can use dma_capable(hwdev, paddr, size) here.

Good thought, however __intel_map_single() gets called with either the
dma_mask or the coherent_dma_mask.  dma_capable() only checks dma_mask,
so would only work for one of the callers.

> >  	domain = get_valid_domain_for_dev(pdev);
> >  	if (!domain)
> > @@ -2767,7 +2767,15 @@ static void *intel_alloc_coherent(struct device *hwdev, size_t size,
> >  
> >  	size = PAGE_ALIGN(size);
> >  	order = get_order(size);
> > -	flags &= ~(GFP_DMA | GFP_DMA32);
> > +
> > +	if (!iommu_no_mapping(hwdev))
> > +		flags &= ~(GFP_DMA | GFP_DMA32);
> > +	else if (hwdev->coherent_dma_mask != DMA_BIT_MASK(64)) {
> > +		if (hwdev->coherent_dma_mask < DMA_BIT_MASK(32))
> > +			flags |= GFP_DMA;
> > +		else
> > +			flags |= GFP_DMA32;
> > +	}
> 
> This is fine for 2.6.32 but we'll cleanly fix this by using
> swiotlb_dma_ops later, right?

I'm open to suggestions.  I don't really understand why we dropped
swiotlb for passthrough mode in 2.6.32 to start with.  It seems like we
now have a couple corner cases where we have to either hope for the best
or effectively ignore the request to use passthrough.  Thanks,

Alex