On Thu, Oct 17, 2024 at 02:12:00PM +0300, Dmitry Baryshkov wrote: > On Wed, Oct 16, 2024 at 10:43:19AM +0530, Krishna Chaitanya Chundru wrote: > > > > > > On 10/14/2024 4:55 AM, Dmitry Baryshkov wrote: > > > On Sat, Oct 12, 2024 at 06:13:34PM +0530, Manivannan Sadhasivam wrote: > > > > On Fri, Oct 11, 2024 at 05:24:29PM +0530, Krishna Chaitanya Chundru wrote: > > > > > > > > [...] > > > > > > > > > > > The logic here is that the fixed endpoints in the switch will get an unique SID > > > > > > > and the devices getting attached to slots will share the same SID of the bus > > > > > > > (this is the usual case with all Qcom SoCs). > > > > > > > > > > > > > > But I guess we would need 'iommu-map-mask' as well. Hope this addresses your > > > > > > > concern. > > > > > > > > > > > > Yes, thank you! > > > > > > > > > > > Hi dimitry & mani, > > > > > > > > > > This particular board variant doesn't expose any open slots to connect > > > > > a different endpoints like another switch(which might have BDF unknown > > > > > to us) so static table should be fine for this board variant. > > > > > > > > > > I tries to add iommu-map-mask property, the issue with that property is > > > > > that the driver is applying the mask to the bdf before searching for the > > > > > entry in the table. If I use a mask value which satisfies all the > > > > > entries in the table ( mask as 0x718) and if a new bdf is enumerated > > > > > lets say 0x600 due to mask 0x718 its value is again 0x600 only. > > > > > > > > > > Can we skip iommu-map-mask property and use only static table for this > > > > > board as we know this board doesn't expose any open slots. > > > > > > > > > > > > > Hmm, I was not aware that it doesn't have open slots. Fine with me then. > > > > > > It doesn't feature open slots, but it has two PCIe connections on HS2 / > > > HS3. Users might attach external PCIe devices. > > > > > > Krishna, could you please clarify, how those two connections are routed? > > > > > For this qps615 board to one of the downstream port (pcie to usb) usb > > hub is connected and to the other downstream port NVMe will be > > connected. > > The board has two PCIe links routed to the HS2 and HS3 connectors. Are > they routed to the PCIe switch? > > Yes, they are not standard slots, but still the board is expandable and > it is possible to connect external PCIe devices. As such it is not > possible to have static SID mapping. > Sorry, I think the conversation got deviated. We have concluded that the endpoints fixed (soldered) in the board will get a fixed SID (because we know what they are) and all other devices going to get connected to HS/LS connectors will get shared SID (because we don't know what they are). Any concern with that? - Mani -- மணிவண்ணன் சதாசிவம்
