Arch Linux User Repository

1) 4.15.0-1 adds zstd kernel support, so the hook for that is no longer needed, but I'm not sure if ever made it out of my personal repo.

2) The -no-pie problem in linking is present in 4.15, and there's an included patch for that. It is now the only patch needed to compile cleanly on Arch.

3) As promised, stubdom has moved from building by default to not building by default, as no one has spoken up about using it. If I continue maintaining the package, the plan is to remove it in the next major release.

4) I haven't done anything about the SDL2 deps.

5) Unless a co-maintainer is added, this is likely going to be the last release for a few months as I'm going in for surgery to have a brain tumor removed next month.

I came across this patch to get Xen building with GCC 11.

I also needed to make a few changes to support a system with split /efi and /boot (using XBOOTLDR). I ended up adding the make arguments BOOT_DIR=/boot EFI_DIR=/boot EFI_MOUNTPOINT=/efi. Technically that BOOT_DIR=/boot is redundant because it's the default, but I actually modified the package to set them as variables (e.g., _boot_dir=${boot_dir:-/boot}) and pass "${_common_make_flags[@]} to each make invocation.

EFI_DIR is where the xen.efi symlink farm gets installed along with the actual EFI image, so setting EFI_DIR=/boot looks weird but is the right thing to do on a SystemD bootctl-powered system using XBOOTLDR: the end result is that I don't need a hook to update the EFI system partition.

I also needed to refactor the symlink cleanup; specifically remove all the symlinks at the end after we've done both the multiboot and the EFI image relocations. Otherwise, if BOOT_DIR and EFI_DIR are the same thing you're gonna have a bad time.

I put my changes on github. Built it in a chroot with multilib-build, gonna go test it now.

Update: Works for me.

Thanks for your efforts. I wish you all the best with your surgery and a fast and painless recovery!

1) 4.15.0-1 adds zstd kernel support, so the hook for that is no longer needed, but I'm not sure if ever made it out of my personal repo.

2) The -no-pie problem in linking is present in 4.15, and there's an included patch for that. It is now the only patch needed to compile cleanly on Arch.

3) As promised, stubdom has moved from building by default to not building by default, as no one has spoken up about using it. If I continue maintaining the package, the plan is to remove it in the next major release.

4) I haven't done anything about the SDL2 deps.

5) Unless a co-maintainer is added, this is likely going to be the last release for a few months as I'm going in for surgery to have a brain tumor removed next month.

@juceda must set at top of PKGBUILD

Build Options

_build_stubdom=${build_stubdom:-false} _build_qemu=${build_qemu:-false}

@juceda: how to make the below modification on src/xen-4.14.1/tools/qemu-xen-build/config-host.mak file so that it is taken into account

@Guard - FYI that issue (unable to disambiguate: -no-pie) is caused by a recent binutils package: binutils-2.36-3-x86_64.pkg.tar.zst

If you downgrade it you'll be able to compile again (at least to bypass that point).

the flag is on: src/xen-4.14.1/tools/qemu-xen-build/config-host.mak changing LDFLAGS_NOPIE=-no-pie to LDFLAGS_NOPIE= will also fix the issue.

pkgbuild was successful after changing _build_stubdom and _build_qemu from "build_qemu/stubdom:-true" to "build_qemu/stubdom:-false"

"ld: Error: unable to disambiguate: -no-pie (did you mean --no-pie ?)"

happens every build, regardless of the source. perhaps I am missing a dependency? it fails to build afterwards.

SDL2 has the least dependencies when display output is needed for (pv)hvm domUs. I run xen on two Laptops and gpu passthrough is not an option. I patch the PKGBUILD and the built-in qemu myself. I can continue to do that. So for my sake the SDL2 support can be removed.

I think spice support is automatically enabled when the spice and spice-protocol packages are installed before makepkg runs. This probably also applies to GTK, VTE and other configuration switches. This could be an alternative to a toggle in PKGBUILD, but should be documented somewhere.

Thanks for your efforts :)

Hello!

In my personal repo I have an updated version of 4.14.1 with security patches and the changes suggested in the last couple weeks. With the new hook, it will boot recent kernels, but both the most recent linux and linux-lts kernels will not start any virtual machines. I'm trying to narrow down where the problem happens still, but what I know is that 5.4.88 works, and 5.4.94 doesn't. If you want to help me track down the problem, my repo is at https://github.com/refutationalist/saur

And now some future stuff where I need some feedback:

Stubdom: I'm now transitioning to using the package without stubdom support. In the comments, I think I was the only person who needed it. If that continues to be the case, I will begin to phase out stubdom starting with the next Xen release.

SDL2: The number of dependencies goes up significantly if one uses the built-in QEMU support, which is the default as the package in extra does not have xen support in it. Since I run Xen almost entirely on headless machines, I'd like to remove SDL2 support. Unless someone says otherwise, I'll have QEMU configured to not use sdl2, otherwise I'll have to turn it into a toggle.

libvirt, qemu, and getting the package into the repos: I've talked to a couple of TUs who are interested in getting the package into the repos, but not much has actually happened on that score. I could try to become a TU myself, but I do have a lot on my plate and I don't know that I have the brain space to handle that. Would it make more sense for me to just abandon the package and hope a TU picks it up? Some direction from someone who knows the political ins and outs of Arch Linux would be very helpful.