[PATCH v4 09/16] dma-mapping: handle MMIO flow in dma_map|unmap_page

[Leon Romanovsky]

On Thu, Aug 28, 2025 at 12:17:30PM -0300, Jason Gunthorpe wrote:
> On Tue, Aug 19, 2025 at 08:36:53PM +0300, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro at nvidia.com>
> > 
> > Extend base DMA page API to handle MMIO flow and follow
> > existing dma_map_resource() implementation to rely on dma_map_direct()
> > only to take DMA direct path.
> 
> I would reword this a little bit too
> 
> dma-mapping: implement DMA_ATTR_MMIO for dma_(un)map_page_attrs()
> 
> Make dma_map_page_attrs() and dma_map_page_attrs() respect
> DMA_ATTR_MMIO.
> 
> DMA_ATR_MMIO makes the functions behave the same as dma_(un)map_resource():
>  - No swiotlb is possible
>  - Legacy dma_ops arches use ops->map_resource()
>  - No kmsan
>  - No arch_dma_map_phys_direct()
> 
> The prior patches have made the internl funtions called here support
> DMA_ATTR_MMIO.
> 
> This is also preparation for turning dma_map_resource() into an inline
> calling dma_map_phys(DMA_ATTR_MMIO) to consolidate the flows.
> 
> > @@ -166,14 +167,25 @@ dma_addr_t dma_map_page_attrs(struct device *dev, struct page *page,
> >  		return DMA_MAPPING_ERROR;
> >  
> >  	if (dma_map_direct(dev, ops) ||
> > -	    arch_dma_map_phys_direct(dev, phys + size))
> > +	    (!is_mmio && arch_dma_map_phys_direct(dev, phys + size)))
> >  		addr = dma_direct_map_phys(dev, phys, size, dir, attrs);
> 
> PPC is the only user of arch_dma_map_phys_direct() and it looks like
> it should be called on MMIO memory. Seems like another inconsistency
> with map_resource. I'd leave it like the above though for this series.
> 
> >  	else if (use_dma_iommu(dev))
> >  		addr = iommu_dma_map_phys(dev, phys, size, dir, attrs);
> > -	else
> > +	else if (is_mmio) {
> > +		if (!ops->map_resource)
> > +			return DMA_MAPPING_ERROR;
> 
> Probably written like:
> 
> 		if (ops->map_resource)
> 			addr = ops->map_resource(dev, phys, size, dir, attrs);
> 		else
> 			addr = DMA_MAPPING_ERROR;

I'm big fan of "if (!ops->map_resource)" coding style and prefer to keep it.

> 
> As I think some of the design here is to run the trace even on the
> failure path?

Yes, this is how it worked before.

> 
> Otherwise looks OK
> 
> Reviewed-by: Jason Gunthorpe <jgg at nvidia.com>
> 
> Jason