The existing functionality for allocation of physical functions and virtual functions has been extended. This functionality must be investigated and documented.
Assigning to Martin for review.
Sent question to SME
*** Bug 1378689 has been marked as a duplicate of this bug. ***
Please Review Martin's Doc
see comment #11
Hello Martin, Itzik is no longer working on SR-IOV, he moved to ODL team. We have two engineers that covered SR-IOV features for the latest release, they will be glad to review the documentation draft and to help answering questions, Eran Kuris and Eyal Dannon. Thanks, Toni
(In reply to Toni Freger from comment #16) > Hello Martin, > > Itzik is no longer working on SR-IOV, he moved to ODL team. > We have two engineers that covered SR-IOV features for the latest release, > they will be glad to review the documentation draft and to help answering > questions, > Eran Kuris and Eyal Dannon. > > Thanks, > Toni Although RH writing unique section for NFV. I think the manuals must depicts about DPDK applications in the Guest (SRIOV/DPDK NICS). See explanation from DPDK.org http://dpdk.org/doc/guides/nics/intel_vf.html I would like to add NFV view regards SR-IOV Guest applications could run as following over SR-IOV Host/Compute 1) Inside Guest VM which its interfaces(manages by guest kernel) interacting through IOMMU to SRIOV PF/VF 2) Inside Guest VM there is DPDK application. For option 2, the VM must run in Host user land, and DPDK application doing PathThrough from the Guest process into DPDK packets in the Host in the host 2 extra parameters vfio-pci, no kernels in the way (Guest or Host) In the grub the additional grub param iommu=pt () While in the Guest you bind dpdk_nic_bind.py --bind=vfio 0000:06:10.0 In option (2) guest kernel bypassed and there is PathThrough of the process towards SRIOV PV/VF
19.6 as far as I know you dont need to install SRIOV agent it exist and install by default from rhos 9. About dynamic allocation from PF to VF there is an issue about it : https://bugzilla.redhat.com/show_bug.cgi?id=1370047 I think we should document it until it will be fix. I know there is a patch that didn't marge yet. Those my comments for this guide.
Regarding the feedback in comment 17 and 18: I've summarised the requests in an email, and sent to Eran and Yariv for confirmation.