On 27/03/2019 08:04, Christoph Hellwig wrote:
arch_dma_prep_coherent can handle physically contiguous ranges larger than PAGE_SIZE just fine, which means we don't need a page-based iterator.
Heh, I got several minutes into writing a "but highmem..." reply before finding csky's arch_dma_prep_coherent() implementation. And of course that's why it specifically takes a page instead of any addresses. In hindsight I now have no idea why I didn't just write the flush_page() logic to work that way in the first place...
Reviewed-by: Robin Murphy <robin.mur...@arm.com>
Signed-off-by: Christoph Hellwig <h...@lst.de> --- drivers/iommu/dma-iommu.c | 14 +++++--------- 1 file changed, 5 insertions(+), 9 deletions(-) diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c index 77d704c8f565..f915cb7c46e6 100644 --- a/drivers/iommu/dma-iommu.c +++ b/drivers/iommu/dma-iommu.c @@ -577,15 +577,11 @@ struct page **iommu_dma_alloc(struct device *dev, size_t size, gfp_t gfp, goto out_free_iova;if (!(prot & IOMMU_CACHE)) {- struct sg_mapping_iter miter; - /* - * The CPU-centric flushing implied by SG_MITER_TO_SG isn't - * sufficient here, so skip it by using the "wrong" direction. - */ - sg_miter_start(&miter, sgt.sgl, sgt.orig_nents, SG_MITER_FROM_SG); - while (sg_miter_next(&miter)) - arch_dma_prep_coherent(miter.page, PAGE_SIZE); - sg_miter_stop(&miter); + struct scatterlist *sg; + int i; + + for_each_sg(sgt.sgl, sg, sgt.orig_nents, i) + arch_dma_prep_coherent(sg_page(sg), sg->length); }if (iommu_map_sg(domain, iova, sgt.sgl, sgt.orig_nents, prot)
_______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
