ACRN Project Technical Community Meeting Minutes - 11/28/2018
Meeting Minutes Archive: https://drive.google.com/drive/u/1/folders/1_XTGu6re3BOQAjtPx636Mz1T5lOy2UWC
ACRN Project TCM - 28th November 2018
1. ACRN project update
2. Chen, Zide: PCIe configuration space emulation
Download foil from ACRN Presentation->ACRN_TCM->WW48’18
Q: do we need to passthrough PCIe bridge and the device?
A: don’t need to pass-thru PCIe bridge to the guest, but we can pass-thru PCIe device to the guests.
Q: For partition mode, will we use the same or different BDF?
A: it doesn’t matter virtual BDF is identical to physical BDF or not, as long as the virtual BDF is unique in the guest and it’s bus is 0.
Q: for example a PCIe end device (bdf 1:0:0) is on the PCIe bridge (bdf 0:7:0), how do we pass thru end device? do we only do -s 1/0/0? or do we pass thru the PCIe bridge 0/7/0 as well? or do we need to emulate the bridge like host bridge?
A: We can pass through PCIe device to the guest, in this case “-s 1/0/0”, but don’t need to pass through PCIe bridge (0/7/0). We don’t need to emulate the PCIe bridge and attach the PCIe device to the virtual host bridge.
However, we didn’t test PCIe extensively, and it’s possible that it may need some fixes.
Q: Any performance data regarding to the HV interception? For emulated devices and pass-through devices.
A: In hypervisor, it only intercepts configuration I/O ports and the MSI-X table MMIO, and the guests are supposed to have access them mostly in the system initialization stage. So we don’t think there will have noticeable performance penalties for pass-through devices.
Q: What’s the next step for PCIe configuration space emulation?
A: Currently the PCI configuration emulation for UOS is done in device model and we are planning to move this functionality to hypervisor; Some improvements to the MSI remapping, for example, support the per-vector masking for MSI devices.
3. All: Community open discussion.
4. Next meeting agenda proposal:
Tel: +86-21-6116 7445
MP: +86-1364 1793 689