On Mon, Dec 06, 2021 at 06:36:27AM -0800, Christoph Hellwig wrote: > I really hate the amount of boilerplate code that having this in each > bus type causes. +1 I liked the first version of this series better with the code near really_probe(). Can we go back to that with some device_configure_dma() wrapper condtionally called by really_probe as we discussed? > Between that and the suggestion from Joerg I wonder if we could do the > following again: > > - add new no_kernel_dma flag to struct device_driver > - set this flag for the various vfio drivers > - skip claiming the kernel dma ownership for those (or rather release > it if the suggestion from Joerg works out) v1 did exactly this. Jason
