Reference hardware: TS-253D with a QM2-2P-244A-A1

Everything here was observed on the project’s reference box in October 2026 unless it says otherwise. It is the ground truth for defaults and tests; other models are expected to differ in counts, not in kind.

Platform

Item	Value
NAS	QNAP TS-253D, Intel Celeron J4125 (4C, 10 W TDP, Tjunction 105 °C per Intel ark), one 70 mm chassis fan, 2 SATA bays
OS	OpenMediaVault 8.5 on Debian 13 (Trixie), Proxmox kernel 7.0.x-pve (headers: proxmox-headers-*)
Boot disk	USB SSD (sdc; non-removable flag, so it shows up as an sdX disk). Not a fan source since 2026-10-17: qnapd skips disks on a USB path
Internal eMMC	4 GB (mmc-…), holds the original QTS firmware; ignored by the plugin
Expansion	QM2-2P-244A-A1 in the PCIe slot: 2 × M.2 NVMe, ASMedia ASM1812 switch (1b21:1812), own fan, PIC firmware 1.13
Drivers	qnap8528-dkms 1.27, qm2mod-dkms 1.1 (plugin release packages, 2026-10-18; hand-built 1.24 before), loaded by the Drivers page

qnap8528 on this box

• Chip check passes without skip_hw_check (dmesg: IT8528 EC device found successfully), contrary to the upstream list of ENE-EC models.
• Model detection: MB=70-0Q05H0201 + backplane 70-1QY570111 → device model is TS-253D (kernel log only; not in sysfs).
• hwmon qnap8528: fan1_input, pwm1, temp1_input (chassis, ≈21–28 °C idle), temp6_input (backplane, ≈25–32 °C). No junk channels.
• LEDs: status, usb, hdd1, hdd2, panel_brightness. hdd1 = bay 1 = SATA port 1 (ata1), hdd2 = bay 2 = ata2 (verified by lighting each LED and by reading from one disk only).
• VPD strings are NUL padded to 32 bytes; VPD reads cost ~5 ms per byte (qnapd info took 1.9 s uncached, 0.14 s with the per-boot cache).
• EC firmware Q04WE018; CPLD reads 0xff (not present). BIOS from DMI (Q04WAR09); DMI product/serial fields are “Default string”.
• ec/power_recovery was 2 (restore last state): after an unplanned power cut the box powered itself back on.

Buttons

Input device qnap8528 (event4 on the reference box): Reset (rear pinhole) = BTN_1, USB Copy (front) = BTN_2, chassis open = BTN_0 (not tested). Press and release are separate events; a 7 s Reset hold measured correctly. The power button is ACPI and shuts the box down.

Buzzer

The buzzer is the PC-speaker line: the in-tree pcspkr driver (loaded by default) drives it through its input device /dev/input/by-path/platform-pcspkr-event-spkr. An EV_SND/SND_TONE event of 1000 Hz, then 0 after 0.2 s, beeps audibly (2026-10-18). qnap8528 has no buzzer interface, and none is needed.

qm2mod on this card

• swap=0 (exact model match), 2 slots, 2 temperature sensors (one per slot, ≈28–38 °C), I2C adapter number appears in LED names (qm2-<n>:{green,red}:slot{1,2}).
• Slot 1 = switch downstream port 02:00.0 = nvme0, slot 2 = 02:08.0 = nvme1 (verified by lighting each LED and by reading from nvme0 only).
• LEDs are all off until something sets them; the PIC drives nothing itself.
• Brightness 2 (qm2mod’s “PIC blink”, register value 3) blinks slowly, red or green; register value 2 blinks fast (PIC 1.13, retested 2026-10-20; an earlier “stays dark” result did not reproduce). A bicolour slot with green on and red blinking looks red/green alternating. qnapd uses red 1 with the timer trigger for Locate.
• fan_default hands the fan to the PIC, which idles it at PWM 64 / ≈1260 RPM.

Measured fan curves

PWM written vs RPM, 15 s per step, idle system:

PWM	255	192	128	96	80	64	56	48	40	32	16	8
Chassis (70 mm)	3631	2804	1905	1407	1191	944	786	663	546	397	342	stall
QM2-2P-244A-A1	5860	4740	3320	2440	1840	1260	880	stall	0	0	0	0

• Chassis fan stalls between PWM 16 and 8 and restarts only at 32.
• QM2 fan stalls at 48 and barely restarts at 48 (160 RPM).
• qnap8528 stores PWM as a percentage, so read-back drifts (128 → 127, 56 → 53).
• QM2 RPM comes in 60 RPM steps.

Noise (listener at the box)

• QM2 fan at 64: not audible over the chassis fan at 48. Audible from 96.
• Chassis around 1300 RPM (PWM ≈86) at idle was acceptable.
• What was distracting: the chassis fan jumping with CPU bursts, and the QM2 fan at full speed (≈5900 RPM) for a few seconds on every restart. Both are fixed (see fan-control.md).

Temperatures and load seen

• CPU package idles 45–52 °C; bursty background load (ZFS kworkers, an open OMV dashboard) spikes single samples to 60–68 °C.
• An open OMV dashboard with many widgets kept two php-fpm workers at ≈80 % of a core. Close or navigate away from it when measuring.
• HDDs (WD 12 TB) 33–39 °C, NVMe 40–42 °C, QM2 slot sensors 37–38 °C.

Disks and pools

• 2 × 12 TB SATA disks in a ZFS mirror; 2 × NVMe on the QM2 in a ZFS mirror.
• OMV spindown configured (APM 1, 10 min), applied at boot by smartctl-hdparm (Standby timer set to 120). Anything calling smartctl without -n standby wakes them (a tempmon setup did), and even -n standby queries on an awake disk restart its standby timer: with qnapd polling every 60 s the idle disks never slept (2026-10-17). Fixed in 8.0.1 (idle disks are not queried); they now reach standby 12–14 min after their last wake. OMV’s smartd here still uses power mode Never and wakes them every 30 min; Standby would let them stay asleep. The general list: ../disk-sleep.md.
• The USB boot SSD runs hot: smartd logged 62 °C. The chassis fans cannot cool it, so it is not a fan source.