Bug 612788
| Summary: | 82576 physical function device assignment doesn't work with window os when both two ports are pass-through to guest. | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 6 | Reporter: | juzhang <juzhang> |
| Component: | qemu-kvm | Assignee: | Don Dutile (Red Hat) <ddutile> |
| Status: | CLOSED WONTFIX | QA Contact: | Virtualization Bugs <virt-bugs> |
| Severity: | medium | Docs Contact: | |
| Priority: | low | ||
| Version: | 6.0 | CC: | alex.williamson, ddugger, lihuang, llim, michen, mkenneth, rpacheco, tburke, virt-maint |
| Target Milestone: | beta | ||
| Target Release: | 6.1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | Bug Fix | |
| Doc Text: |
A dual function, 82576 interface (codename: Kawela, PCI Vendor/Device ID: 8086:10c9)cannot have both physical functions (PF's) device-assigned to a Windows 2008 guest. Either physical function can be device assigned to a Windows 2008 guest (PCI function 0 or function 1), but not both.
|
Story Points: | --- |
| Clone Of: | Environment: | ||
| Last Closed: | 2011-01-26 18:47:31 UTC | Type: | --- |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
| Bug Depends On: | |||
| Bug Blocks: | 580953 | ||
|
Description
juzhang
2010-07-09 02:26:38 UTC
could you try the following:
(1) specify 03:00.0 in the qemu-kvm cmdline first, then 03:00.1;
by (PCI) definition, fcn 0 must appear before fcn1, but not sure
if qem-kvm cares or not.
The devices should appear in the windows guest as devices having
zero-based functions with different device numbers (like 00:4.0, 00:5.0,
or the Windows-equiv), but interesting to know
(2) Could you do the assignment using a properly qualified xml spec
and use virsh define <guest-name>.xml & virst start <guest-name> ?
e.g.,
(a) start kvm guest as listed above _without_ the pci dev assignments.
(b) exec: virsh dumpxml <guest-name> > <guest-name>.xml
i) edit <guest-name>.xml and add:
<hostdev mode='subsystem' type='pci' managed='yes'>
<source>
<address domain='0x0000' bus='0x03' slot='0x00' function='0x0'/>
</source>
</hostdev>
<hostdev mode='subsystem' type='pci' managed='yes'>
<source>
<address domain='0x0000' bus='0x03' slot='0x00' function='0x1'/>
</source>
</hostdev>
(c) stop <guest-name> : virsh shutdown <guest-name>
(d) virsh undefine <guest-name>
(e) virsh define <guest-name> < <guest-name>.xml
(f) virsh start <guest-name>
Note, also, that Windows reports it as a 'dual-port server adapter',
yet qemu-kvm is probably exporting it as two, single devices.
Not sure if Windows driver is having 'heartburn' that a dual-port is showing up as a single port.
What does Device Manager show/say about each port -- does it show the first one functional, and a yellow/warning label on the second?
Finally, can you try kernel-2.6.32-44 and qemu-kvm 0.12.1.2-2.93 ?
Minor tweaks were done in this area around MSI-X mapping, which this adapter performs. Seen as a multi-VF problem, but wondering if multi-PF has same problem.
I dug up a dual Kawela 82576, id 8086:10c9.
Able to duplicate this bz.
note that dual Kawela have shared eeprom access, and when assigning
function 1 of PF's to a windows guest, it appears as a unique function 0 device.
I dont know how Linux drivers work (as claimed in description; I should try duping those cases), but this looks like a Windows driver 'feature'.
I can assign either of the PF's (PF.0 or PF.1), but cannot assign both to the
guest as the same time.
Other things found while testing:
(a) if I swap the order of assignment, i.e., PF.0 first, PF.1 second, the first-spec'd dev-assignment is the one that connects successfully in win2k8.
(b) The successfully configured dev in win2k8 is always named
"Intel(R) Gigabit ET Dual Port Server Adapter #2"
The failing one is always "Intel(R) Gigabit ET Dual Port Server Adapter"
(c) the successfully configured dev is always pci dev 00:04.0;
the failing one is always pci dev 00:05.0 (in the guest/win2k8 vm).
Adding Intel Partner & Ron Pacheco to see if we should be using
an updated driver on win2k8 for this situation, since Linux drivers (5.5 & 6.0)
work.
This issue has been proposed when we are only considering blocker issues in the current Red Hat Enterprise Linux release. It has been denied for the current Red Hat Enterprise Linux release. ** If you would still like this issue considered for the current release, ask your support representative to file as a blocker on your behalf. Otherwise ask that it be considered for the next Red Hat Enterprise Linux release. ** Moving to rhel6.1, & Tier3 for now. Intel claims device assigning any dual-fcn, Kawela PF to a Windows guest is not supported. Still trying to debug to get to final resolution, but current belief is w2k8 driver isnt as clean/robust as linux driver. Technical Note: =============== A dual function, 82576 (codename: Kawela, PCI Vendor/Device ID: 8086:10c9) interface cannot have both physical functions (PF's) device-assigned to a Window 2008 guest. Either function can be device assigned to a Windows 2008 guest (PCI function 0 or function 1), but not both. Since dual-function device assignment works for Linux guests that support the 82576, it appears that there is a bug in the Windows 2008 driver that affects dual-function operation when device-assigned to a KVM guest.
Technical note added. If any revisions are required, please edit the "Technical Notes" field
accordingly. All revisions will be proofread by the Engineering Content Services team.
New Contents:
A dual function, 82576 (codename: Kawela, PCI Vendor/Device ID: 8086:10c9)
interface cannot have both physical functions (PF's) device-assigned to a
Window 2008 guest. Either function can be device assigned to a Windows 2008
guest (PCI function 0 or function 1), but not both. Since dual-function device
assignment works for Linux guests that support the 82576, it appears that there
is a bug in the Windows 2008 driver that affects dual-function operation when
device-assigned to a KVM guest.
Technical note updated. If any revisions are required, please edit the "Technical Notes" field
accordingly. All revisions will be proofread by the Engineering Content Services team.
Diffed Contents:
@@ -1,7 +1 @@
-A dual function, 82576 (codename: Kawela, PCI Vendor/Device ID: 8086:10c9)
+A dual function, 82576 interface (codename: Kawela, PCI Vendor/Device ID: 8086:10c9)cannot have both physical functions (PF's) device-assigned to a Windows 2008 guest. Either physical function can be device assigned to a Windows 2008 guest (PCI function 0 or function 1), but not both.-interface cannot have both physical functions (PF's) device-assigned to a
-Window 2008 guest. Either function can be device assigned to a Windows 2008
-guest (PCI function 0 or function 1), but not both. Since dual-function device
-assignment works for Linux guests that support the 82576, it appears that there
-is a bug in the Windows 2008 driver that affects dual-function operation when
-device-assigned to a KVM guest.
Note that Intel does not support assignment of PF devices to a guest (that's what VFs were designed for). Even though assigning 2 PFs works for a Linux guest that is still an unsupported configuration and not recommended. Closing due to lack of Windows driver support. |