On 25/09/18 21:16, Christoph Hellwig wrote:
Looking at the code I think this commit is simply broken for architectures not using arch_setup_dma_ops, but instead setting up the dma ops through arch specific magic. I'll revert the patch.
Ugh, sorry about missing that too. Ack to a revert - thinking about those PPC symptoms, it might actually be that other architectures could also get into the same pickle just by unbinding and rebinding a driver (e.g. switching to VFIO then back again).
Robin.
