NVMe drive drops out with I/O timeouts and 'failed to set APST feature (-19)' (broken APST)
Fetched the raw Arch wiki Solid_state_drive/NVMe. It supports the cause (Kingston A2000 on S5Z42105, Samsung, WD/SanDisk, SK Hynix), the log lines in the symptom's first block, Btrfs read-only and ext4 I/O errors, nvme_core.default_ps_max_latency_us=0, the pcie_aspm=off pcie_port_pm=off fallback, the BIOS power-saving step and the Kingston firmware. It does not contain 'controller is down; will reset: CSTS=0xffffffff, PCI_STATUS=0x10' anywhere, yet the title is built on it and the record cites no other source. That message and its PCI_STATUS value are unsupported precision, so the title and symptom were corrected to the messages the source gives. The Omarchy branch held on this machine: /usr/lib/limine/limine-common-functions loads /etc/limine-entry-tool.d/*.conf before /etc/default/limine and every file appends with +=, so a drop-in named nvme-apst-off.conf is applied, and limine-mkinitcpio rebuilds the UKI that carries the cmdline. CONFIG_NVME_CORE=m on linux-omarchy 7.2.5-3 and /sys/module/nvme_core/parameters/default_ps_max_latency_us exists (reads 100000 here), so the verify step is valid. Not exercised: the parameter was not applied.
Disabling APST, and especially PCIe ASPM, raises idle power draw and shortens laptop battery life. Each controller drop risks losing unwritten data, so back up before experimenting.
Symptom
Random freezes, Btrfs suddenly read-only, or ext4 I/O errors, with the kernel log showing something like:
nvme nvme0: I/O 566 QID 7 timeout, aborting
nvme nvme0: I/O 840 QID 6 timeout, reset controller
nvme nvme0: Device not ready; aborting reset, CSTS=0x1
nvme nvme0: failed to set APST feature (-19)
The drive is unusable until the system is reset.
Cause
Some NVMe drives mishandle Autonomous Power State Transitions (APST) and do not come back from a deep power state. Reports cover the Kingston A2000 on firmware S5Z42105, some Samsung, Western Digital/SanDisk and SK Hynix drives. The controller stops answering, the kernel's reset fails, and the filesystem on it is lost for the rest of the boot. When it hits during writes to /boot, it is also a common cause of the FAT errors that remount the ESP read-only.
Fix
Disable APST with the kernel parameter nvme_core.default_ps_max_latency_us=0.
Omarchy 4:
echo 'KERNEL_CMDLINE[default]+=" nvme_core.default_ps_max_latency_us=0"' | sudo tee /etc/limine-entry-tool.d/nvme-apst-off.conf
sudo limine-mkinitcpio
Plain Arch with GRUB: add it to GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub, then:
sudo grub-mkconfig -o /boot/grub/grub.cfg
systemd-boot: append it to the options line in /boot/loader/entries/*.conf.
If failures continue, the wiki's next step is to add pcie_aspm=off pcie_port_pm=off the same way, and to look for NVMe or PCIe power-saving options in the firmware setup. Also check the vendor for a drive firmware update, such as Kingston's for the A2000.
Verify
cat /sys/module/nvme_core/parameters/default_ps_max_latency_us prints 0, and journalctl -k | grep -i 'nvme.*\(timeout\|reset\)' stays empty across days of normal use.