On Tue, Nov 01, 2022 at 06:24:16PM +0100, Christoph Hellwig wrote: > On Tue, Nov 01, 2022 at 05:19:46PM +0000, Catalin Marinas wrote: > > > > The main downside of bouncing is mobile phones where those vendors are > > in the habit of passing noswiotlb on the kernel command line or they > > want a very small bounce buffer (hard to pick a universally small size). > > I guess we can go ahead with this and, depending on how bad the problem > > is, we can look at optimising swiotlb to use a kmem_cache (or aligned > > kmalloc) as fall-back for bouncing. > > Theses phones setups are completely broken already. There is a reason > why it needs a debug option to disable swiotlb, and that reason is > that Linux guarantees that the 32-bit DMA always works. If they force > swiotlb off they can keep the pieces as this is not a supported > configuration. There's also the case of low-end phones with all RAM below 4GB and arm64 doesn't allocate the swiotlb. Not sure those vendors would go with a recent kernel anyway. So the need for swiotlb now changes from 32-bit DMA to any DMA (non-coherent but we can't tell upfront when booting, devices may be initialised pretty late). > > As a less than optimal solution, we can force bouncing for the whole > > list if any of the sg elements is below the alignment size. Hopefully we > > won't have many such mixed size cases. > > I suspect we won't see any such cases. The scatterlist is usually used > for large data transfers, and many devices won't like unaligned buffers > for SG operations to start with. So I think it is a perfectly fine > tradeoff. I'll give it a try this week, hopefully post patches soon(ish). -- Catalin
