PCI passthru: intel_iommu=on and VT-d enabled in BIOS results in HardLock on boot

[FuriousGeorge]

I've read that there are issues with the Intel X58 chipset and iommu on some kernels, but I'm not sure if this is the cause.

Basicaly the problem is that I cannot enable IOMMU in grub.conf if I have VT-d enabled in the BIOS, or the CPU hardlocks very early on in the boot.

I had the same HardLock trying to boot an asus rampage ii gene with Intel_iommu=on, but I was fairly certain that board did not properly support IOMMU, as there were no reports of anyone getting it working, and known issues with Asus and VT-d support.

The mb I purchased is a ga-ex58-ud5 that has VT-d support in a beta BIOS from Gigabyte that was never officially eleased but widely used in the community.

I've read on other forums that GPU passthru has worked in Linux with that MB, so I'm leaning toward the kernel being the problem.

I'm not sure how to proceed so any help is much appreciated.

 

[FuriousGeorge]

This was caused by a patch applied to kernel 4.2ish that breaks IOMMU for certain (most?) X58 chipset motherboards (at least). This is significant because the only LGA1366 boards with x16 PCIe slots are X58.

The fix has been merged into kernel 4.7: https://git.kernel.org/cgit/linux/k.../?id=8c2561557415a298965f50d4de2ca99c7d1f1b2c

It is fairly simple to manually revert the patch from intel-iommu.c and build your kernel manually in the meantime, though you may run into some problems with ZFS. This got the GA-EX58 board to boot.

I have a Rampage II Gene which may be affected by the same. I will verify and update this post later.

The Rampage III Formula I got to replace it is unaffected (also X58), however it will panic if the onboard audio is enabled in the bios.

I realize this is a niche thing, but I figured I'd post it for posterity. It may be useful to someone outside the Proxmox community.