Description of problem: Virtualization guide lacks information on PCI pass through. Version-Release number of selected component (if applicable): RHEL 5.4 How reproducible: N/A Steps to Reproduce: 1. 2. 3. Actual results: N/A Expected results: N/A Additional info: Complaints from partners about lack of information on how to use PCI pass through.
Trying scope this chapter. Should this cover: - Xen PCI passthrough for RHEL 5.0-5.3 - Xen PCI passthrough in RHEL 5.4 - KVM PCI passthrough - IOMMU and VT-d stuff - Using libvirt for PCI passthrough - Using the virt-manager add device option for PCI passthrough - SR/IOV Anything I've missed or anything completely wrong? The next question is can I get some information on each of those sections except libvirt and virt-manager (I've figured that side of it out already). Chris
*** Bug 545096 has been marked as a duplicate of this bug. ***
I think you have covered most issues. Just note that for KVM PCI passtrough is in tech preview only for RHEL5.4. Very few devices do work at the moment, so we may want to emphasize this. I'm not sure what is the status with Xen. Simon.
Can someone verify the content: http://documentation-stage.bne.redhat.com/docs/en-US/Red_Hat_Enterprise_Linux/5/html/Virtualization/chap-Virtualization-PCI_passthrough.html This bug should be resolved, however, I am unsure of the technical accuracy of what is written. Chris
Chapter 13: PCI devices are limited by the virtualized system architecture. Out of the 32 available PCI devices for a guest 4 are not removable. This means there are up to 28 PCI slots available for additional devices per guest. Every para-virtualized network or block device uses one slot. Each guest can use up to 28 additional devices made up of any combination of para-virtualized network, para-virtualized disk devices, or other PCI devices using VT-d. I'm not sure if that's exactly accurate. For example, I have a guest here where there's actually only 2 slots taken (yes, 4 devices, but 3 are multifunction, so 0:0.0, 0:1.0, 0:1.1 and 0:1.3). The max is 32 slots (each slot supports up to 8 functions for multifunction devices), and a couple are used up by the chipset. So it is true that we support a limited total number of devices, but the exact number is dependent upon the config. Procedure 13.2: Nothing is required to enable AMD's IOMMU, it's on by default. And it does not support the iommu=pt option (pt == Pass Through, and is a generic mode for either vendor's IOMMU where the page table translations are not used for host devices, only devices assigned to guests. It's useful to avoid the overhead of IOMMU page table translations in the host). Unfortunately for our users...13.1 and .2 are KVM centric. Xen has different syntax. The Xen grub entry needs "iommu=on", the kernel module entry for Xen doesn't need any iommu cmdline. (And for Chapter 14, SR-IOV, Xen admins need to add pci_pt_e820_access=on to the kernel module line in grub). Otherwise you've done a nice job of capturing a bunch of information that's not always intuitive and made sense of it. I noted that you used different device in the install case vs. the virsh and virt-manager cases. Unsure if that was intentional.
Thanks for the quick response Chris. I'll make those updates by Close of Business today. Chris