systems/jeeves/networking.nix

@@ -22,6 +22,11 @@
       ];
     };
     useNetworkd = true;
+    # Without this, unconfigured NICs get a catch-all DHCP network and
+    # join the LAN uninvited (how the dead atlantic card ended up
+    # ARP-poisoning 192.168.99.14). Every interface must be configured
+    # explicitly below.
+    useDHCP = false;
   };
 
   systemd.network = {
@@ -46,6 +51,20 @@
       };
     };
     networks = {
+      # Unused second port of the onboard i350; keep it down so the dead
+      # PCIe link (see dmesg "PCIe link lost") stops logging Tx errors.
+      "01-unused-igb-port" = {
+        matchConfig.Name = "enp97s0f0";
+        linkConfig.ActivationPolicy = "always-down";
+      };
+      # Aquantia 10G card (0000:21:00.0); loses PCIe link, and its
+      # unicast RX path is dead while TX still works, so when it's up it
+      # answers ARP for the host IPs and black-holes the replies. Keep
+      # it down until the card is fixed or replaced.
+      "01-unused-atlantic-port" = {
+        matchConfig.Name = "enp33s0";
+        linkConfig.ActivationPolicy = "always-down";
+      };
       "10-Primary" = {
         matchConfig.Name = "enp97s0f1";
         address = [ "192.168.99.14/24" ];

systems/jeeves/services/systemd.nix

@@ -16,6 +16,20 @@ in
           ExecStart = "${pkgs.bash}/bin/bash ${../scripts/plex_permission.sh}";
         };
       };
+      # The unused onboard i350 port hangs ~90s after every boot ("PCIe
+      # link lost", MMIO reads return all-Fs) even with pcie_aspm=off.
+      # Remove it from the PCI bus so nothing (hwmon temperature polls,
+      # networkd) touches the dead hardware. Reversible via
+      # `echo 1 > /sys/bus/pci/rescan`.
+      remove_dead_igb_port = {
+        wantedBy = [ "multi-user.target" ];
+        description = "removes hung i350 port 0000:61:00.0 from the PCI bus";
+        unitConfig.ConditionPathExists = "/sys/bus/pci/devices/0000:61:00.0/remove";
+        serviceConfig = {
+          Type = "oneshot";
+          ExecStart = "${pkgs.bash}/bin/bash -c 'echo 1 > /sys/bus/pci/devices/0000:61:00.0/remove'";
+        };
+      };
       startup_validation = {
         requires = [ "network-online.target" ];
         after = [ "network-online.target" ];