Right now the code uses cpufreq/cpuinfo_max_freq. The correct way to identify different core types is the cpu_capacity file (/sys/devices/system/cpu/cpu3/cpu_capacity). This should be 1024 for p-cores, though it's probably safer to keep the current logic of taking the max number as the p-cores.

When I export this env variable instead of using Wayland it refuses to launch

Outside the VM:

alyssa@blossom ~> locale
LANG=fr_CA.UTF-8
LC_CTYPE="fr_CA.UTF-8"
LC_NUMERIC=en_CA.UTF-8
LC_TIME=en_CA.UTF-8
LC_COLLATE="fr_CA.UTF-8"
LC_MONETARY=en_CA.UTF-8
LC_MESSAGES="fr_CA.UTF-8"
LC_PAPER=en_CA.UTF-8
LC_NAME="fr_CA.UTF-8"
LC_ADDRESS="fr_CA.UTF-8"
LC_TELEPHONE="fr_CA.UTF-8"
LC_MEASUREMENT=en_CA.UTF-8
LC_IDENTIFICATION="fr_CA.UTF-8"
LC_ALL=

Inside the VM:

alyssa@blossom ~ [1]> krun bash
...
alyssa@blossom:~$ locale
LANG=
LC_CTYPE="POSIX"
LC_NUMERIC="POSIX"
LC_TIME="POSIX"
LC_COLLATE="POSIX"
LC_MONETARY="POSIX"
LC_MESSAGES="POSIX"
LC_PAPER="POSIX"
LC_NAME="POSIX"
LC_ADDRESS="POSIX"
LC_TELEPHONE="POSIX"
LC_MEASUREMENT="POSIX"
LC_IDENTIFICATION="POSIX"
LC_ALL=

Incidentally this masked over FEX-Emu/FEX#4010 on Asahi, otherwise my French system would've hit that bug before the user did.

Zombie processes inside the VM never get reaped, I guess because init.krun doesn't do this. Although we obviously don't need a full-blown init system, it should at least do this.

Sound doesn't seem to be working even when libkrun is built with virtio-snd support, e.g.: https://github.com/teohhanhui/libkrun/tree/virtio-snd (patch adapted from @slp but updated for latest libkrun)

I think we're missing this:
https://github.com/teohhanhui/asahi-krun-container/blob/799243343a4546be9e464f9a64d7326d3655e294/entrypoint-user.sh#L7-L9 (again, adapted from @slp's container image)

Should we do this in krun-guest?

Related page: https://www.isc.org/blogs/isc-dhcp-eol/

Code: https://github.com/slp/krun/blob/3c6f9419aa170617be07d109151b53c54f45ff4e/crates/krun-guest/src/net.rs#L21

krun shouldn't rely on an EOL package, trying to package this for NixOS reveals we need to package dhclient ourselves which is not idea. Instead an alternative supported and still maintained package should be used instead.

If I open Steam and idle long enough for Plasma to autosuspend the machine (or at least turn off my monitor), when I unsuspend & turn the screen back on & unlock Plasma, the VM has crashed with

12sommelier: ../sommelier.cc:956: void sl_registry_remover(void*, wl_registry*, uint32_t): Assertion `0' failed.
(EE) failed to write to Xwayland fd: Broken pipe

Hopefully that's fixable..?
this is on a Mac Studio, not sure if reproducible on the laptops

If there was I could get the steam app working

When launching a GUI application, the window fails to appear on the screen, though it does show up in the application bar. However, the application executes as expected, and the desired output is visible in the command line output:

The issue persists regardless of what application and whether the application is launched directly or through FEXBash.

Interestingly, this problem is not present when compiling and running the legacy C version of krun.

In the VM, my /etc/resolv.conf looks like this:

; generated by /usr/sbin/dhclient-script
nameserver 192.168.0.1

Which is bad, because that's my router's IP address... (And not what's sent as DNS servers via DHCP.)

Others have reported no Internet access in the VM, probably related to this.

If I understand correctly, I think that's why krunvm set nameserver 1.1.1.1 as the default:
https://github.com/containers/krunvm/blob/5494d84a66bee3b802a0392cf8d662158ac7287d/src/main.rs#L51

But that's also not a good solution as it'd break local domains and search domains among other things...

libkrun.h is not found, krun has this problem but not krunvm.

error: failed to run custom build command for `krun-sys v1.8.1 (/build/source/crates/krun-sys)`

Caused by:
  process didn't exit successfully: `/build/source/target/release/build/krun-sys-c928935194fd9ad3/build-script-build` (exit status: 101)
  --- stdout
  cargo::rerun-if-changed=wrapper.h
  cargo::rustc-link-lib=krun

  --- stderr
  wrapper.h:1:10: fatal error: 'libkrun.h' file not found
  thread 'main' panicked at crates/krun-sys/build.rs:11:10:
  Unable to generate bindings: ClangDiagnostic("wrapper.h:1:10: fatal error: 'libkrun.h' file not found\n")
  note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Derivation:

pkgs.rustPlatform.buildRustPackage rec {
  pname = "krun";
  version = "git+${src.rev}";

  src = pkgs.fetchFromGitHub {
    owner = "slp";
    repo = "krun";
    rev = "5a8542a5066bd1a4d8bad99236f05ea3ac687ef7";
    hash = "sha256-3w1JHqzpkF11+3yHN3yPu2dKdi4qXdO8HY+7ICugAZA=";
  };

  cargoHash = "sha256-dtj3TiszBBsMtlJaaIyTmCNAGZy/zuSGSK8OOTb5Xmg=";

  buildInputs = [
    libkrun
  ];

  cargoLock.lockFile = "${src}/Cargo.lock";
  env.LIBCLANG_PATH = "${pkgs.llvmPackages_latest.clang-unwrapped.lib}/lib/libclang.so";
}

Can we set that somehow? It seems to be used by Steam.

The right click menu in regedit and drop down menus in winecfg are broken this happens in hangover and boxwine

Followed this guide and managed to install ableton live

After successful install I'm trying to run ableton, splash screen shows, init starts and then it fails due to error:

info:    Image count:  4
info:    Exclusive FS: 2
00e0:err:alsa:get_device_name_from_guid No devices found in registry?
00e0:err:alsa:get_device_name_from_guid No devices found in registry?
00e0:err:alsa:get_device_name_from_guid No devices found in registry?
ALSA lib pulse.c:242:(pulse_connect) PulseAudio: Unable to connect: Connection refused
ALSA lib pulse.c:242:(pulse_connect) PulseAudio: Unable to connect: Connection refused
ALSA lib seq_hw.c:528:(snd_seq_hw_open) open /dev/snd/seq failed: No such file or directory
00e0:err:winediag:MIDIMAP_drvOpen No software synthesizer midi port found, Midi sound output probably won't work.

I see that if I run aplaymidi -l inside vm I get

 Port    Client name                      Port name
 14:0    Midi Through                     Midi Through Port-0

But if I run it inside FEXBash I get same error when ableton fails

ALSA lib seq_hw.c:528:(snd_seq_hw_open) open /dev/snd/seq failed: No such file or directory
Cannot open sequencer - No such file or directory

On NixOS, we use /run for many runtime things such as symlinking the current system state to /run/current-system. This means the system "installed" packages will not be accessible unless /run/current-system is available. We also have the OpenGL drivers symlinked to /run/opengl-driver. Without that symlink being available, GPU related access via Mesa is not possible. Since /run isn't shared to krun, this means system packages and OpenGL are both not available.

Possible solution is to check if NIXOS_SYSTEM is an env var set and to mount the value of it to /run/current-system. A similar solution could be done for OpenGL.

Starting an interactive shell with krun can be confusing as it's not clear when you're still in the VM or you've exited to the host. To improve the UX, we should change the prompt (PS1) so users can easily differentiate between contexts.

Not 100% sure if/how this is supposed to work. Noticed in Steam. Not a big deal (low priority versus #56) but definitely a paper cut worth writing down before I forget.

How would one go about setting this up?

Obviously you need the devel packages for all the components, but where do you go from there?

I am currently trying to run an appimage inside krun but it lacks permission to access the fuse device/interface.

I'd be glad for any pointers.

When using box64 and the new wow64 version steam gets very close to letting you login but then crashes and retries it used to just show a black screen but using the env variable krun -e WINEDLLOVERRIDES=libglesv2.dll=d lets you see the steam logo

After #23 gets merged, krun will be capable of acting as a binfmt interpreter for x86_64 binaries on aarch64 machines. We should create the binfmt support files, just qemu-user-static does, to package managers can install them on the host.

(Continuing from #23 (comment))

With some real-world testing, it's entirely jarring behaviour from a user perspective.

Steps for reproduction:

1. Launch krun FEXBash
2. Launch krun FEXInterpreter steam
3. exit from FEXBash

Expected result:
Steam should continue running.

Actual / observed behaviour:
Steam is killed abruptly.

I'm getting this error if I try to run krun a second time, i.e. trying to launch a command through an existing krun-server.

Do you have any clues what might be wrong? I haven't been able to get any useful output / logs.