Arch Linux User Repository

• Leaving eclairvoyant's comment pinned because it contains some useful information.
• Patch policy is different, but similar.
  • Primary purpose of this package are the ACS override and i915 VGA arbiter patches.
  • Upstream Arch patches will be applied to maintain parity with the standard kernel.
  • No other patches will be added.
• Package does have options that can be changed, like building with clang or disabling Arch patches.
  • Defaults will match Arch packages, except when incompatible with this package's primary patches.
  • Options are subject to change. Read PKGBUILD for details.
• Avoid commenting and flagging at the same time for the same issue.
  • Flag for common issues with standard solutions, like new releases, key changes, etc.
  • Comment for issues requiring explanation or debugging.
    • Use a pastebin for blocks of text more than a few lines.

This package exists for the specific purpose of adding ported patches based on those originally created by Alex Williamson for:

• Bypassing the default IOMMU grouping by overriding PCIe ACS support, and

• Enabling VGA arbitration support on Intel iGPUs.

Arbitrary patches will not be added.

Refer to the wiki on PCI passthrough and this blog post on IOMMU groups for risks/caveats before using this package.

Regular AUR etiquette applies as well (knowledge of makepkg and searching the wiki/Arch forums is expected, and AUR helpers or Arch-based distros that are not Arch Linux are unsupported).

Just so that there's at least one other person showing good signs, this version (4.4-3) works beautifully with my setup. Without the hugepages and with an Nvidia card. It works so well that I put the package on my repo.

I am only running 8 GB of guest memory with the remaining 24 GB for the host, but I am using 1 GB hugepages. I also have only 4 cores / 8 threads (of 6/12) used by the guest. Not sure if it makes a difference for what we're talking about, but I have the cpu pinning set up so each vcpu is pinned to a different host thread and the emulator is pinned to the remaining 4 host threads. AMD gpu too, so maybe things would be different with nvidia.

Hmm, that's weird. It seems that I'm the only one who has this issue. My guest is a Windows 10 guest, I put all my cores/threads on this guest (6 cores with 12 threads). The Tianocore splash screen took the same amount of time as with 4.1.15, but Windows loading took a while with all my threads on the Linux host are maxed. I put 24 GB of memory to it. Probably that's the case, what are you guest memory? Let's see if I can improve the boot performance if I reduce the memory size.

I haven't tried the latest mainline. I did notice with 4.4 that cpu usage maxes out 2 cores for a second or 2 during boot, but it seems to actually boot faster for me. The tianocore screen doesn't even time to show now when it used to appear for 3-5 seconds. Running windows 8.1 in the guest.

The patch for the bug has been merged since 4.4rc7 [1]. Although it can now boot successfully, I still felt the boot time is considerably slower and took higher CPU usage than in previous 4.1.15 kernel. Have you experienced the similar behavior in the latest mainline kernel? [1] https://bugzilla.kernel.org/show_bug.cgi?id=107561

I usually keep this package in sync with the main Arch package, but I have recently built 4.4 and for the first time since 4.1 it is actually working to boot a VM with OVMF and pcie passthrough. I am running git version of OVMF (17560.ccdf8bc), qemu (2.5.0.r43284.5a57acb), and libvirt (1.3.1.rc2.3.g8f0a157 - custom package from my github), so it may or may not work with the repo versions. Before upgrading, consider that this package is very closely based on the main Arch package which hasn't been released yet. You are taking the same risk as running the testing repo. It is possible that it could contain some system-destroying bug that you would normally not be exposed to because it would be fixed by the time it is released. You have been warned.

Pssst, the PKGBUILD states you are at pkgrel=2 but the .SRCINFO isn't reflecting that change. Might you re-run mksrcinfo?

@electricprism, that information is in the PKGBUILD.

Requires importing GPG Keys to build # Linus Torvalds gpg --recv-keys 79BE3E4300411886 # Greg Kroah-Hartman (Linux kernel stable release signing key) gpg --recv-keys 38DBBDC86092693E

You need to actually read the error messages above to see what the problem is. You are probably getting the same issue that gets comes up constantly for any aur package using signed sources - you don't have the public keys in your key ring. https://wiki.archlinux.org/index.php/Makepkg#Signature_checking