dma-iommu: Check that swiotlb is active before trying to use it
authorMario Limonciello <mario.limonciello@amd.com>
Mon, 4 Apr 2022 20:47:23 +0000 (15:47 -0500)
committerJoerg Roedel <jroedel@suse.de>
Thu, 28 Apr 2022 08:26:49 +0000 (10:26 +0200)
If the IOMMU is in use and an untrusted device is connected to an external
facing port but the address requested isn't page aligned will cause the
kernel to attempt to use bounce buffers.

If for some reason the bounce buffers have not been allocated this is a
problem that should be made apparent to the user.

Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220404204723.9767-3-mario.limonciello@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
drivers/iommu/dma-iommu.c

index 09f6e1c0f9c07c85652ae63ca185df7e74151b14..1ca85d37eeab210fd5e4d778d51664f8682b0929 100644 (file)
@@ -971,6 +971,11 @@ static dma_addr_t iommu_dma_map_page(struct device *dev, struct page *page,
                void *padding_start;
                size_t padding_size, aligned_size;
 
+               if (!is_swiotlb_active(dev)) {
+                       dev_warn_once(dev, "DMA bounce buffers are inactive, unable to map unaligned transaction.\n");
+                       return DMA_MAPPING_ERROR;
+               }
+
                aligned_size = iova_align(iovad, size);
                phys = swiotlb_tbl_map_single(dev, phys, size, aligned_size,
                                              iova_mask(iovad), dir, attrs);