From mboxrd@z Thu Jan  1 00:00:00 1970
From: catalin.marinas@arm.com (Catalin Marinas)
Date: Tue, 21 Aug 2012 13:59:48 +0100
Subject: [PATCH v2 14/31] arm64: DMA mapping API
In-Reply-To: <201208151616.01114.arnd@arndb.de>
References: <1344966752-16102-1-git-send-email-catalin.marinas@arm.com>
 <1344966752-16102-15-git-send-email-catalin.marinas@arm.com>
 <201208151616.01114.arnd@arndb.de>
Message-ID: <20120821125948.GC12708@arm.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Wed, Aug 15, 2012 at 05:16:00PM +0100, Arnd Bergmann wrote:
> On Tuesday 14 August 2012, Catalin Marinas wrote:
> > +static struct dma_map_ops arm64_swiotlb_dma_ops = {
> > +       .alloc = arm64_swiotlb_alloc_coherent,
> > +       .free = arm64_swiotlb_free_coherent,
> > +       .map_page = arm64_swiotlb_map_page,
> > +       .unmap_page = arm64_swiotlb_unmap_page,
> > +       .map_sg = arm64_swiotlb_map_sg_attrs,
> > +       .unmap_sg = arm64_swiotlb_unmap_sg_attrs,
> > +       .sync_single_for_cpu = arm64_swiotlb_sync_single_for_cpu,
> > +       .sync_single_for_device = arm64_swiotlb_sync_single_for_device,
> > +       .sync_sg_for_cpu = arm64_swiotlb_sync_sg_for_cpu,
> > +       .sync_sg_for_device = arm64_swiotlb_sync_sg_for_device,
> > +       .dma_supported = swiotlb_dma_supported,
> > +       .mapping_error = swiotlb_dma_mapping_error,
> > +};
> > +
> > +void __init swiotlb_init_with_default_size(size_t default_size, int verbose);
> > +
> > +void __init arm64_swiotlb_init(size_t max_size)
> > +{
> > +       dma_ops = &arm64_swiotlb_dma_ops;
> > +       swiotlb_init_with_default_size(min((size_t)SZ_64M, max_size), 1);
> > +}
> 
> Why is swiotlb the default? I would expect that most devices can in fact
> use the entire 64 bit address space, so you can use a simple linear
> implementation for those.

That was my worry, devices not capable of accessing the full 64-bit
address space. We can hope that those SoCs would have an IOMMU but I
can't tell for sure at this stage.

The default implementation could be simpler. I can even drop it
altogether from the initial patchset given that no SoC makes use of it
yet.

-- 
Catalin