Hi Jacob, On Mon, 6 Mar 2023 15:45:04 -0800, Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx> wrote: > Hi Jason, > > On Mon, 6 Mar 2023 15:02:37 -0400, Jason Gunthorpe <jgg@xxxxxxxxxx> wrote: > > > On Mon, Mar 06, 2023 at 11:04:43AM -0800, Jacob Pan wrote: > > > > > > and probably this is the right thing to do as in the end DMA type > > > > will be removed with Jason's cleanup > > > > > > so, let me recap. set_dev_pasid() should make no assumptions of > > > ordering, i.e. it is equal to iommu_domain_ops.attach_dev(). > > > > Absolutely yes. > > > > You should factor out all the "prepare the domain to be used" code and > > call it in both places. > > > I think this has been done by Baolu > https://lore.kernel.org/linux-iommu/20190325013036.18400-1-baolu.lu@xxxxxxxxxxxxxxx/T/#m8c980357a39dc75dc360ff0f71dc7ebe1e49f9a6 > iommu/vt-d: Move common code out of iommu_attch_device() > > This part of code could be used by both normal and aux > domain specific attach entries. Hence move them into a > common function to avoid duplication. > > set_dev_pasid() will call prepare_domain_attach_device() as well. Actually, there are more to be factored to common code if we take that assumption away. attach_dev() can be viewed as a special case for set_dev_pasid() except the PASID is RID_PASID. Thanks, Jacob