On Wed, Apr 17, 2024 at 05:35:47PM +0800, Yi Liu wrote: > > do we want to force iommu driver to block translation upon > > any error in @set_dev_pasid (then rely on the core to recover > > it correctly) or tolerate duplicated attaches? > > The second one seems better? The first option looks a too heavy especially > considering the atomic requirement in certain scenarios. On set_dev_pasid() failure the driver should make no change to the translation. It is the only sane answer. Jason
