On Fri, Jul 05, 2024 at 09:48:46PM +0300, Leon Romanovsky wrote: > In that case the flow is dma_map_sg()->iommu_dma_map_sg()->dev_use_sg_swiotlb(). Even for that you'll still need to check the first and last entry for being kmalloc misaligned if we assume that all middle entries are aligned (which for NVMe they have to, but we're probably better off figuring out a way to enforce that).
