Recover a MediaTek MT7921/MT7922 Wi-Fi card that is dead after suspend
Every specific held, against the cited sources and against this workstation (omarchy 4.0.2-1, kernel 7.1.9-arch1-2, systemd 261.2-1). Confirmed on this machine: modinfo -p mt7921e prints exactly one parameter, disable_aspm:disable PCI ASPM support (bool), so the modprobe.d option and the /sys/module/mt7921e/parameters/disable_aspm check are right on kernel 7.1, and mainline mt76/mt7921/pci.c still carries module_param_named(disable_aspm, mt7921_disable_aspm, bool, 0644) at line 36 with the same MODULE_PARM_DESC. mem_sleep_default= is also still live: __setup("mem_sleep_default=", mem_sleep_default_setup) is line 214 of mainline kernel/power/suspend.c, and /sys/power/mem_sleep here reads s2idle [deep]. Both forum threads support the record rather than merely sitting near it. Thread 295916 carries the exact failed to resume async: error -110, dpm_run_callback(): pci_pm_resume and Message 00020007 (seq 8) timeout lines, the modprobe -r mt7921e plus systemctl restart NetworkManager bluetooth one-off recovery, one user for whom mt7921e.disable_aspm=y worked and the original poster for whom it did not, and its final post is the /usr/lib/systemd/system-sleep pre/post script the record reproduces, which is why the record is right to present step 2 as a workaround rather than a fix. Thread 284180 carries the Unable to change power state from D3cold to D0, device inaccessible line, cat /sys/power/mem_sleep returning [s2idle] deep, the loud-fan symptom the danger field warns about, and a closing post naming the solution as a BIOS update followed by mem_sleep_default=deep, so both halves of the cause are sourced. The Arch wiki section ==== mt7921 / mt7922 ==== prescribes options mt7921e disable_aspm=1 for the high latency problem and gives the AMD RZ608 and RZ616 branding, matching the cause's own caveat that the wiki documents it for latency rather than for the resume hang (the wiki writes the file as /etc/modprobe.d/wifi.conf, and the record's mt7921e.conf is equally valid because modprobe reads every .conf in that directory). The Omarchy 4 boot path is the part most likely to have been wrong and it is correct: /etc/limine-entry-tool.d/ exists here holding omarchy-defaults.conf, omarchy-uki.conf and resume.conf, the last of which uses the same KERNEL_CMDLINE[default]+=" ..." form the record writes, /etc/limine-entry-tool.conf documents += as append and a bare = as replace, and /usr/bin/limine-mkinitcpio runs /usr/share/libalpm/scripts/limine-mkinitcpio-install rebuild, which re-reads the command line with limine-entry-tool --get-cmdline and re-registers the UKI with --add-uki, so the single sudo limine-mkinitcpio in step 3 is sufficient and no separate limine-update is needed. The record's boot-limine-conf lint hit is a false positive: it names /boot/limine.conf only to tell the reader not to edit it, which is the same advice as limine-kernel-parameters-not-applying-omarchy. Two things I checked and judged not to be defects: deep-sleep.conf sorts before omarchy-defaults.conf, which is harmless because the record uses += and /proc/cmdline here shows all drop-ins merged rather than one replacing another, and writing the sleep hook into /usr/lib/systemd/system-sleep/ matches both systemd-sleep(8) on this machine, which documents only that directory, and Omarchy's own practice, since keyboard-backlight, nvidia and unmount-fuse already live there. Not exercised: no MT7921 or MT7922 card is in this machine (the wireless NIC is an Intel Wireless-AC 9560 on iwlwifi), so the resume failure, the sleep hook and the ASPM knob could not be run, and per instruction I wrote no kernel parameter, loaded or unloaded no module and did not touch the bootloader or initramfs. The only source change is adding the omacom/omarchy quattro copy of omarchy-hibernation-setup: the record's basecamp master URL still resolves and its file still writes /etc/limine-entry-tool.d/resume.conf and runs limine-mkinitcpio, so it does support the claim and is not removed, but quattro is the tree that ships Omarchy 4.
mem_sleep_default=deep on a machine whose firmware does not properly implement S3 can cause the laptop to suspend and never wake, or to wake with the fans at full speed. Test it interactively before you rely on it, and be prepared to hold the power button. Adding a bad kernel-parameter drop-in and running limine-mkinitcpio rewrites your boot entries. Keep the Limine menu reachable (do not enable Direct Boot) so you can edit the entry if the machine will not come back.
Symptom
Wi-Fi works fine from a cold boot, but after the first suspend/resume the card is gone: no networks in the list, nmcli device shows the interface as unavailable or it disappears entirely, and only a reboot brings it back. The journal shows mt7921e 0000:24:00.0: PM: failed to resume async: error -110, mt7921e 0000:24:00.0: PM: dpm_run_callback(): pci_pm_resume+0x0/0xf0 returns -110 and Message 00020007 (seq 8) timeout, or on some machines mt7921e 0000:2e:00.0: Unable to change power state from D3cold to D0, device inaccessible. Often accompanied by driver own failed / chip reset failed. These cards are also branded AMD RZ608 (MT7921) and RZ616 (MT7922).
Cause
The mt7921e driver leaves PCIe Active State Power Management (and L1 substates) enabled across suspend. On many consumer boards and laptops the card cannot be brought back out of D3cold, so the resume callback times out with -110 and the device is left inaccessible on the bus. The Arch wiki documents disabling ASPM as the only fix for the related high-latency problem on the same chipsets, and the upstream driver exposes exactly one knob for it. On some machines the real trigger is that the firmware only offers s2idle rather than real S3, in which case a BIOS update plus mem_sleep_default=deep fixes it outright.
Fix
First check what the card is and how the machine suspends:
lspci -knn | grep -A3 -i network
cat /sys/power/mem_sleep # [s2idle] means no real S3
journalctl -kb | grep -i mt7921
Step 1. Disable ASPM for the driver:
sudo tee /etc/modprobe.d/mt7921e.conf >/dev/null <<'EOF'
options mt7921e disable_aspm=1
EOF
sudo modprobe -r mt7921e && sudo modprobe mt7921e
The option is real: mt76/mt7921/pci.c declares module_param_named(disable_aspm, mt7921_disable_aspm, bool, 0644). Confirm it took:
cat /sys/module/mt7921e/parameters/disable_aspm # expect Y or 1
Step 2. If that alone does not survive a suspend cycle, unload and reload the module around sleep. Create a systemd sleep hook (this is the workaround the Arch forum thread settled on):
sudo tee /usr/lib/systemd/system-sleep/mt7921e >/dev/null <<'EOF'
#!/usr/bin/env bash
case "$1" in
pre)
modprobe -r mt7921e
;;
post)
modprobe mt7921e
;;
esac
EOF
sudo chmod +x /usr/lib/systemd/system-sleep/mt7921e
Everything executable in /usr/lib/systemd/system-sleep/ is run by systemd-sleep with pre/post as $1. Test with systemctl suspend, then journalctl -b -u systemd-suspend.service.
Step 3. If /sys/power/mem_sleep reports only [s2idle], update the BIOS first, then force real S3. On Omarchy 4, kernel parameters go in a limine-entry-tool drop-in, not by hand-editing /boot/limine.conf (which omarchy-refresh-limine resets):
sudo mkdir -p /etc/limine-entry-tool.d
echo 'KERNEL_CMDLINE[default]+=" mem_sleep_default=deep"' \
| sudo tee /etc/limine-entry-tool.d/deep-sleep.conf
sudo limine-mkinitcpio
On plain Arch with GRUB, add mem_sleep_default=deep to GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub and run sudo grub-mkconfig -o /boot/grub/grub.cfg.
As a one-off recovery without rebooting:
sudo modprobe -r mt7921e && sudo modprobe mt7921e
sudo systemctl restart NetworkManager bluetooth
Verify
Run systemctl suspend, resume, then nmcli device status. The wlan device should be connected or disconnected, never unavailable. journalctl -kb | grep -i mt7921 should show no error -110, no driver own failed and no D3cold message after the resume timestamp. cat /sys/module/mt7921e/parameters/disable_aspm should print Y.
Sources
- https://wiki.archlinux.org/title/Network_configuration/Wireless
- https://bbs.archlinux.org/viewtopic.php?id=295916
- https://bbs.archlinux.org/viewtopic.php?id=284180
- https://raw.githubusercontent.com/torvalds/linux/master/drivers/net/wireless/mediatek/mt76/mt7921/pci.c
- https://github.com/basecamp/omarchy/blob/master/bin/omarchy-hibernation-setup
- https://raw.githubusercontent.com/torvalds/linux/master/kernel/power/suspend.c
- https://github.com/omacom/omarchy/blob/quattro/bin/omarchy-hibernation-setup