On Tue, May 19, 2026 at 08:37:54PM +0530, Aneesh Kumar K.V wrote: > if we get force_dma_unencrypted(dev) correct, we won't need the above. > > for dma_direct_alloc and dma_direct_alloc_pages() we have > > if (force_dma_unencrypted(dev)) > attrs |= DMA_ATTR_CC_SHARED; > > > for dma_direct_map_phys(), if we have swiotlb bouncing forced, > > swiotlb_tbl_map_single(): > > if ((attrs & DMA_ATTR_CC_SHARED) || force_dma_unencrypted(dev)) > require_decrypted = true;
IMHO I really do prefer the DMA_ATTR_CC_SHARED flows closer to the thing that did the decryption. While the above is possibly sound it is very obtuse to be guessing what kind of memory swiotlb decided to return.. Can we pass a pointer to the attrs into the swiotlb stuff and it can update it based on the kind of memory it has allocated? Jason
