scx_lavd E2BIG error with --per-cpu-dsq

[meflove]

System & Kernel Information:

• OS: NixOS (nixpkgs master branch)
• Kernel: 6.17.3-cachyos (x86-64-v3)

scx version:

scx_lavd --version
scx_lavd 1.0.17 x86_64-unknown-linux-gnu

Problem: When attempting to start scx_lavd with the --per-cpu-dsq argument (specifically, using the command below), I encounter an error related to loading BPF programs:

Command Causing Error:

sudo scx_lavd --performance --no-core-compaction --per-cpu-dsq

-- END PROG LOAD LOG --

23:37:37 [WARN] libbpf: prog 'lavd_dispatch': failed to load: -E2BIG

23:37:37 [WARN] libbpf: failed to load object 'bpf_bpf'

23:37:37 [WARN] libbpf: failed to load BPF skeleton 'bpf_bpf': -E2BIG

Error: Failed to load BPF program

Caused by:
    Argument list too long (os error 7)

scx_lavd.log

[multics69]

It would be related to an older version of clang. Would you share the output of clang --version?

[meflove]

I use NixOS, if that matters
clang version 21.1.2

[multics69]

@meflove -- I think I fixed the problem. Could you try with #2942 ?

[meflove]

@multics69 still the same problem

[multics69]

Hmm... that is interesting. I think I should set up a NixOS environment to reproduce your problem.

[meflove]

my config for kernel

    environment.systemPackages = with pkgs; [
      config.boot.kernelPackages.v4l2loopback
      v4l-utils
    ];

    hardware.xone.enable = true;

    services.scx = {
      enable = true;
      package = pkgs.callPackage ../../../packages/scx_rustcheds {};

      scheduler = "scx_lavd";
      extraArgs = [
        "--performance"
        "--no-core-compaction"
        # "--per-cpu-dsq"
      ];
    };

    boot = {
      kernelPackages = cfg.kernelPackage; # pkgs.linuxPackages_cachyos.cachyOverride {mArch = "GENERIC_V3";} from chaotic-cx/nyx

      initrd = {
        compressor = "cat";
      };

      consoleLogLevel = 3;

      kernelPatches = [
        {
          name = "bbr";
          patch = null;
          structuredExtraConfig = with pkgs.lib.kernel; {
            TCP_CONG_CUBIC = lib.mkForce module;
            TCP_CONG_BBR = yes; # enable BBR
            DEFAULT_BBR = yes; # use it by default
            NET_SCH_FQ_CODEL = module;
            NET_SCH_FQ = yes;
          };
        }
      ];

      kernelModules = [
        "ntsync"
        "v4l2loopback"
      ];

      extraModulePackages = with config.boot.kernelPackages; [v4l2loopback];

      kernelParams = [
        "systemd.show_status=auto"
        "quiet"
        "splash"
        "mitigations=off"
        "vt.global_cursor_default=0"
        "lpj=2496000"
        "page_alloc.shuffle=1"
        "pci=pcie_bus_perf"
        "intel_idle.max_cstate=1"
        "bgrt_disable"
        "nowatchdog"
        "ibt=off"
        "kernel.sysrq=1"
        "nvidia_drm.modeset=1"
        "nvidia_drm.fbdev=1"
      ];
    };

    powerManagement.cpuFreqGovernor = cfg.cpuGovernor;
  };

[mcfd]

@meflove-- 我想我解决了这个问题。你能试试 #2942 吗？

The patches works well on Archlinux

CPU:
  Info: 14-core (6-mt/8-st) model: 12th Gen Intel Core i7-12700H bits: 64
    type: MST AMCP cache: L2: 11.5 MiB
  Speed (MHz): avg: 1826 min/max: 400/4600:4700:3500 cores: 1: 1826 2: 1826
    3: 1826 4: 1826 5: 1826 6: 1826 7: 1826 8: 1826 9: 1826 10: 1826 11: 1826
    12: 1826 13: 1826 14: 1826 15: 1826 16: 1826 17: 1826 18: 1826 19: 1826
    20: 1826

default_sched = "scx_lavd" # Edit this line to the scheduler you want scx_loader to start at boot
default_mode = "Auto" # Possible values: "Auto", "Gaming", "LowLatency", "PowerSave".

[scheds.scx_lavd]
auto_mode = ["--autopower","--preempt-shift","5","--per-cpu-dsq"]
#,"--per-cpu-dsq"]
gaming_mode = ["--performance"]
lowlatency_mode = ["--performance"]
powersave_mode = ["--powersave"]

[mcfd]

I tried the main repository today without applying any patches and found that the issue has resurfaced.
1.0.17.r201.gbd6de22c

out.txt

[dim-geo]

Similar error occurs in LTS kernel as well.
latest code

scx_lavd --performance --per-cpu-dsq

Linux gentoo 6.12.56-cachyos-x86_64 #1 SMP PREEMPT_DYNAMIC Sat Nov  8 21:39:55 EET 2025 x86_64 Intel(R) Core(TM) i5-4460 CPU @ 3.20GHz GenuineIntel GNU/Linux

20:35:46 [INFO] libbpf: struct_ops lavd_ops: member cgroup_set_bandwidth not found in kernel, skipping it as it's set to zero

20:35:46 [INFO] libbpf: struct_ops lavd_ops: member priv not found in kernel, skipping it as it's set to zero

20:35:52 [WARN] libbpf: prog 'lavd_dispatch': BPF program load failed: -E2BIG

20:35:52 [WARN] libbpf: prog 'lavd_dispatch': -- BEGIN PROG LOAD LOG --
Global function lavd_dispatch() doesn't return scalar. Only those are supported.
0: R1=ctx() R10=fp0
; void BPF_STRUCT_OPS(lavd_dispatch, s32 cpu, struct task_struct *prev) @ main.bpf.c:712

....

[multics69]

Thanks @mcfd and @dim-geo for reporting the problem. Could you test it again with the current main? #3033 addressed the verifier errors. I wonder if my fix also covers the problem you faced too.

[dim-geo]

on 6.12 kernel a new problem occurs:

22:20:39 [WARN] libbpf: struct_ops init_kern lavd_ops: Cannot find member cgroup_set_bandwidth in kernel BTF

22:20:39 [WARN] libbpf: failed to load BPF skeleton 'bpf_bpf': -EOPNOTSUPP

Error: Failed to load BPF program

Caused by:
    Operation not supported (os error 95)

without using enable_cpu_bw: false,

[multics69]

Thanks @dim-geo for testing! It was supposed to be handled by the kernel smoothly. Let me check.