Hi all,
I read the latest verion (vB.a) of the manual Arm Realm Management Extension (RME) System Architecture (https://developer.arm.com/documentation/den0129/latest/). Now I have several questions on page 17, Figure A1.1: RME System Architecture.
Q1: It seems that RME-DA supports PCIe devices but does not support on-chip devices. Is it correct? If yes, will RME-DA support it in the future?
Q2: Another problem is: How to define the on-chip device or peripheral? For example, if I use a Hikey960/JunoR2 board, will the SoC Mali GPU be an on-chip device, or a peripheral, or neither?
Can someone help me?
Thanks for your reply! Based on this, I still have 4 more problems.
Martin Weidmann said:The important thing is that the device is PCIe.
1. So does it mean RME-DA does not support the AXI-connected device? These AXI-connected devices seem to be the "On-chip Device (with white background color)" on that figure.
2. I am not sure what kind of on-chip devices (actually, even the on-chip hardware) support the PCIe. It seems that most on-chip devices connect to an AXI bus. For example on Juno boards, the GPU, peripherals and CPU are connected toghether via a NIC-400 Network Interconnect, which seems to support AXI, not a PCIe bus. If my thought is correct, could you give some examples?
3. I am a bit interested in RME-CDA. It serves the fully coherent device. But I am not sure what is "fully coherent device". Also, I cannot find much maunals related to RME-CDA. Could you give some examples on what RME-CDA supports?
4. For a device (assume a NVIDIA GPU), if we want it to support RME-DA, do we need some hardware modification? I ask this question because I see some hardware called as "RME-DA PCIe device" on that figure, and others are "PCIe device". Just want to know whether they are different.
Looking forward to your reply.