On Mon, Feb 24, 2025 at 02:24:27PM -0800, Dan Williams wrote: > Xu Yilun wrote: > [..] > > > > > > Oh, when we do this, the root port gets the same devid_start/end as the > > > device which is not correct, what should be there, the rootport bdfn? Need > > > > "Indicates the lowest/highest value RID in the range > > associated with this Stream ID at the IDE *Partner* Port" > > > > My understanding is that device should fill the RP bdfn, and the RP > > should fill the device bdfn for RID association registers. Same for Addr > > association registers. > > So you expect that the endpoint programs RPs and Peer-to-Peer RIDs in > its association register? That makes sense, although I feel like once Yes. > Peer-to-Peer operation is considered the RID association loses > effectiveness because it is difficult to capture a constrained range in > that case. > > We can start with that for RID as our best current understanding and > circle back later if it causes problems. As for address association I am That's OK. > not sure Linux needs to worry about it in the first implementation. The For Intel, yes on RP side. Intel TDX Module requires OS to input ONE address association range on IDE stream create to setup RP. Thanks, Yilun > mechanism is too coarse to keep up with IOMMU entries for the device. If > it already needs to be overspecified might as well disable it.
