Catalin's work to reduce the size of minimum allocation sizes from kmalloc on ARM64 relies on a heuristic where small buffers are always bounced to ensure DMA safety. https://lore.kernel.org/all/20230531154836.1366225-1-catalin.marinas@xxxxxxx/ Now this shouldn't break any IIO drivers but it might (at least in theory) have a performance impact. I'm curious on whether anyone cares enough about this? Long term, if this approach ends up adopted on all architectures that currently have ARCH_KMALLOC_MINALIGN set above 8 bytes we can get rid of the explicit handling of DMA buffers in IIO as they will mostly be bounced anyway - with the exception of a few that are bigger than the relevant cacheline size. Jonathan
