Power: disable a bunch of things

BROKE intel arc A380 completely because it was forced into L1.1/L1.2
pcie substates. Forcewaking the device would fail and it would never come up.

So I will be more conservative on power saving tuning.

configuration.nix

@@ -120,7 +120,12 @@
     };
   };
 
-  hardware.intelgpu.driver = "xe";
+  # Intel Arc A380 (DG2, 56a5) uses the i915 driver on kernel 6.12.
+  # The xe driver's iHD media driver integration has buffer mapping
+  # failures on this GPU/kernel combination. i915 works correctly for
+  # VAAPI transcode as long as ASPM deep states are disabled for the
+  # GPU (see modules/power.nix).
+  hardware.intelgpu.driver = "i915";
 
   # Per-service 2MB hugepage budget calculated in service-configs.nix.
   boot.kernel.sysctl."vm.nr_hugepages" = service_configs.hugepages_2m.total_pages;

modules/power.nix

@@ -1,12 +1,9 @@
 {
-  lib,
-  pkgs,
   ...
 }:
 {
   powerManagement = {
     enable = true;
-    powertop.enable = true;
     cpuFreqGovernor = "powersave";
   };
 
@@ -29,14 +26,6 @@
     # work items -- irrelevant for a server whose latency-sensitive paths are
     # all in userspace (caddy, jellyfin).
     "workqueue.power_efficient=1"
-
-    # Force PCIe ASPM on even if the BIOS doesn't advertise support. ASRock
-    # B550M Pro4 BIOS defaults are conservative; the Zen 3 root complex and
-    # all downstream devices (NVMe, AHCI, Intel NIC) support L1 substates.
-    # powertop auto-tune sets the policy to powersupersave at runtime, but
-    # without `force` the kernel may refuse to enable ASPM at all if the BIOS
-    # opted out, making the policy write a no-op.
-    "pcie_aspm=force"
   ];
 
   boot.kernel.sysctl = {
@@ -45,49 +34,8 @@
     "kernel.nmi_watchdog" = 0;
   };
 
-  # Server has no audio consumers. Power-gate the HDA codec at module load
+  # Server has no audio consumers. Power-gate the HDA codec at module load.
-  # rather than waiting for powertop auto-tune to do it after boot.
   boot.extraModprobeConfig = ''
     options snd_hda_intel power_save=1 power_save_controller=Y
   '';
-
-  # Apply sysfs power knobs that powertop --auto-tune cannot reach (hardened
-  # kernel blocks debugfs mount, so powertop silently skips ASPM policy and
-  # may only lower EPP to balance_power instead of power).
-  #
-  # AMD pstate EPP "power": deepest P-states, fastest core parking. Safe because:
-  # - xmrig runs at Nice=19 / CPUSchedulingPolicy=idle and tolerates latency
-  # - web services (caddy, jellyfin) are I/O-bound; the ~50 us extra C-state
-  #   exit latency is invisible behind network RTT
-  # - Minecraft server benefits from single-thread boost, which pstate still
-  #   provides on demand even in "power" mode (just with slightly slower ramp)
-  #
-  # ASPM powersupersave: deepest PCIe link power states (L1.1/L1.2). The
-  # pcie_aspm=force boot param enables ASPM, but the runtime policy defaults
-  # to "default" which only uses L0s. powersupersave adds L1 substates for
-  # all downstream devices (NVMe, AHCI, NIC).
-  systemd.services.power-tune = {
-    description = "Apply power-saving sysfs knobs (EPP, ASPM policy)";
-    after = [ "multi-user.target" ];
-    wantedBy = [ "multi-user.target" ];
-    serviceConfig = {
-      Type = "oneshot";
-      RemainAfterExit = true;
-      ExecStart = lib.getExe (
-        pkgs.writeShellApplication {
-          name = "power-tune";
-          text = ''
-            # AMD pstate energy performance preference
-            for epp in /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference; do
-              [ -f "$epp" ] && echo power > "$epp"
-            done
-
-            # PCIe ASPM policy
-            aspm=/sys/module/pcie_aspm/parameters/policy
-            [ -f "$aspm" ] && echo powersupersave > "$aspm"
-          '';
-        }
-      );
-    };
-  };
 }