Hello, A patch that was submitted to RH5.5 kernel and was accepted to the default flavor is missing in the xen flavor, as a result the mlx4 driver missfunctions in the Xen flavor of RH5.5 Theis patch was added at: https://bugzilla.redhat.com/show_bug.cgi?id=534158#c10 Can this fix be added to the errata of RH5.5? Thanks, Yevgeny
What makes you think the lack of this patch that is causing the problems in Xen? The reason I ask is that there's only one driver for both variants "flavors". Do you have pci_pt_e820_access=on on dom0's kernel command line? Here's an example config in grub.conf title Red Hat Enterprise Linux Server (2.6.18-181.el5xen) root (hd2,0) kernel /boot/xen.gz-2.6.18-181.el5 iommu=1 console=com1 com1=9600,8n1 module /boot/vmlinuz-2.6.18-181.el5xen ro root=LABEL=/ console=ttyS0,9600n8 pci_pt_e820_access=on module /boot/initrd-2.6.18-181.el5xen.img Note the "pci_pt_e820_access=on" on the first module line (dom0's kernel command line). That's needed for SR-IOV to work with Xen. Note also the iommu=1 on Xen's command line. That's needed if you want to pass a pci device (or VF) through to an HVM guest. Drew
From Yevgeny: Hello Drew, I could not update the bugzilla for some reason, but here is my comment: The errors happened without enabling SRIOV or IOMMU, we tested the driver in single function mode, without pass-through. I can now see from you kernel that for xen, the kernel version that is used is 2.6.18-181, the version where my fix was accepted is 2.6.18-191. This can explain why the fix is not included in Xen flavor. Thanks Yevgeny Petrilin
Yevgeny, You can use -191 and all the same release numbers that bare-metal has for xen as well. For every kernel release there is also a xen version. If the patch you need is in -191, then it sounds like just using RHEL 5.5 (-194) should solve your problems. Please try that, and let me know if it works. Thanks, Drew
We tried it with the GA RH5.5 so kernel version should be 2.6.18-194 Yevgeny
Will this be included in a 5.5.z or will we have to wait for RHEL 5.6?
I'm still confused to why there appears to be a problem here. From comment 2 it looked like the original testing was done with -181, which would explain the absence of the patch, since the patch was integrated into -191. However, if testing was done with -194, then that indicates a new issue, as the patch is already there. Please note that both xen and bare-metal have the same module compiled so it wasn't missed from the xen flavor.
Mellanox, please provide an update.
Per Yevgeny Petrilin at Mellanox: > It has passed on RH5.5 GA, I thought I have updated the bugzilla, will > check again. Closing bug as CURRENTRELEASE (5.5).