OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
network high common DISPUTE

Stop an RTL8111/8168 gigabit NIC from flapping or dropping to 100 Mbps

omarchyarchendeavouroscachyosmanjarodesktoplaptop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Nearly everything here checks out against the cited sources and against this Omarchy 4 box. https://wiki.archlinux.org/title/Network_configuration/Ethernet carries the exact two log lines the symptom quotes ("MicroStar Motherboard with Realtek 8111/8168/8411" section), prescribes the ip link set dev <iface> down/up bounce verbatim, prescribes the AUR r8168 + blacklist r8169 route for flapping revisions, and documents iommu=soft for Gigabyte boards, so the record's "documented remedy" claim is real, not fabricated. bbs.archlinux.org/viewtopic.php?id=285421 is exactly the AER-storm mini-PC case and its accepted answer is literally r8168.aspm=0 r8168.eee_enable=0 pcie_aspm=off with r8168-dkms and r8169 blacklisted. I confirmed aspm and eee_enable are real module_param()s in r8168_n.c (lines 502 and 520), and that r8168-dkms exists in the AUR at 8.056.02-1, last updated 2026-02. The Omarchy-specific boot bits are correct for Quattro, not Omarchy 3: /etc/limine-entry-tool.d exists on this machine holding omarchy-defaults.conf and resume.conf, the KERNEL_CMDLINE[default]+=" ..." append syntax matches /etc/limine-entry-tool.conf's documented drop-in operator, sudo limine-mkinitcpio is precisely what /usr/share/omarchy/migrations/1784917531.sh and 1786482992.sh run after writing such a drop-in, and omarchy-refresh-limine really does overwrite /boot/limine.conf from $OMARCHY_PATH/default/limine/limine.conf, so the warning against hand-editing it is right. NetworkManager-dispatcher.service is enabled here, so the dispatcher hook will fire. ONE REAL GAP: ethtool is not installed on Omarchy 4. pacman -Q ethtool returns "package not found" on this machine and ethtool appears nowhere in /usr/share/omarchy/install/*.packages. A user copy-pasting the diagnostic block or Step 1 gets "command not found", and the dispatcher script silently no-ops on /usr/bin/ethtool. Corrected fix adds the install and a check that the dispatcher actually took effect. Cause is accurate and stands.

The cause above was not rewritten and may still contain the error described. The fix below is the corrected version.
RISK

r8168 is an out-of-tree DKMS module. If it fails to build against a new kernel you boot with no ethernet at all, and because you blacklisted r8169, there is no fallback. Always install the matching linux-headers before a kernel upgrade, check dkms status afterwards, and keep a USB ethernet adapter or working Wi-Fi as a way back in. Try steps 1 to 3 on the in-tree driver first. Most machines never need step 4. pcie_aspm=off disables PCIe power management system-wide and will measurably shorten laptop battery life. Prefer the per-driver r8168 aspm=0 where it is sufficient.

Symptom

Onboard Realtek ethernet is unreliable: the link goes down and up every minute or so under load, the connection negotiates 100 Mbps instead of 1 Gbps on a cable that works fine elsewhere, or there is a link light but no traffic at all. lspci shows Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller. The journal shows Generic FE-GE Realtek PHY r8169-2200:00: Downshift occurred from negotiated speed 1Gbps to actual speed 100Mbps, check cabling! and r8169 0000:22:00.0 enp34s0: Link is Up - 100Mbps/Full (downshifted), or on worse boards a repeating enp1s0: pci link is down and PCIe AER Uncorrected (Non-Fatal) error received storms.

Cause

The RTL8111/8168 family covers dozens of silicon revisions behind one PCI ID. The in-tree r8169 driver handles most of them well but some revisions misbehave, usually around PCIe ASPM and Energy Efficient Ethernet: the PHY drops into a low-power state, the link partner does not follow, and the link renegotiates or downshifts. On some mini-PCs and Gigabyte/MSI boards the PCIe root port itself throws AER errors when ASPM is active. This is distinct from the RTL8125 2.5 GbE offload bug, which is a different chip and a different fix.

Fix

**Step 0: install ethtool.** Omarchy 4 does not ship it (it is not in omarchy-base.packages), so every command below fails with command not found without this:


sudo pacman -S --needed ethtool

Identify the exact revision and the bound driver:


lspci -nnk | grep -A3 -i ethernet
sudo ethtool <iface> | grep -E 'Speed|Duplex|Link detected'
sudo ethtool --show-eee <iface>
journalctl -kb | grep -iE 'r8169|r8168|downshift|aer'

**Step 1: turn off EEE on the in-tree driver.** This alone fixes the flapping on many boards and costs you nothing but a fraction of a watt:


sudo ethtool --set-eee <iface> eee off

Test for a few minutes, then make it persistent with a NetworkManager dispatcher script:


sudo tee /etc/NetworkManager/dispatcher.d/50-realtek-eee >/dev/null <<'EOF'
#!/bin/bash
IFACE="$1"
ACTION="$2"
case "$IFACE" in en*|eth*) ;; *) exit 0 ;; esac
if [ "$ACTION" = "up" ]; then
  /usr/bin/ethtool --set-eee "$IFACE" eee off || true
fi
EOF
sudo chmod 755 /etc/NetworkManager/dispatcher.d/50-realtek-eee
sudo systemctl enable --now NetworkManager-dispatcher.service

Confirm it actually ran after the next link event, rather than assuming it did:


sudo nmcli device disconnect <iface> && sudo nmcli device connect <iface>
sudo ethtool --show-eee <iface>      # must still report EEE disabled
journalctl -u NetworkManager-dispatcher -n 20

**Step 2: if the log shows pci link is down or AER errors, disable ASPM.** On Omarchy 4, kernel parameters go into a limine-entry-tool drop-in (hand edits to /boot/limine.conf are reset by omarchy-refresh-limine, which copies the packaged default over it):


sudo mkdir -p /etc/limine-entry-tool.d
echo 'KERNEL_CMDLINE[default]+=" pcie_aspm=off"' \
  | sudo tee /etc/limine-entry-tool.d/realtek-aspm.conf
sudo limine-mkinitcpio
sudo reboot

After the reboot, check the parameter actually made it into the booted command line, because Omarchy builds a UKI, so a drop-in that was written without a rebuild will not be in effect:


grep -o 'pcie_aspm=off' /proc/cmdline

On plain Arch with GRUB, add pcie_aspm=off to GRUB_CMDLINE_LINUX_DEFAULT and run sudo grub-mkconfig -o /boot/grub/grub.cfg. On some Gigabyte boards (the wiki's example is the GA-990FXA-UD3) iommu=soft is the documented remedy instead.

**Step 3: as a one-off recovery** when it has already downshifted, bounce the link:


sudo ip link set dev <iface> down
sudo ip link set dev <iface> up

**Step 4: only if the in-tree driver still cannot hold a link**, switch to Realtek's out-of-tree driver:


yay -S linux-headers r8168-dkms

echo 'blacklist r8169' | sudo tee /etc/modprobe.d/blacklist-r8169.conf
sudo tee /etc/modprobe.d/r8168.conf >/dev/null <<'EOF'
options r8168 aspm=0 eee_enable=0
EOF

sudo dkms status                 # must show r8168 installed for your kernel
sudo mkinitcpio -P
sudo reboot

After the reboot, lspci -k should show Kernel driver in use: r8168. The aspm=0 eee_enable=0 combination (with pcie_aspm=off on the cmdline) is what fixed the AER-storm case on affected mini-PCs.

Verify

sudo ethtool <iface> | grep -E 'Speed|Link detected' reports 1000Mb/s and yes. sudo ethtool --show-eee <iface> shows EEE disabled. Then hold it under load: ping -i 0.2 -c 600 <router-ip> alongside a large transfer should complete with no loss, and journalctl -kf | grep -iE 'r816|link is' should stay silent for the duration.

Sources

r8169-rtl8111-link-flapping-r8168-dkms