Power over Ethernet switched from XGS: Ports on the XGS 136. Measured, 2026-10-20¶

The XGS 136's two PoE ports, labels 11 and 12, were read by hand on the coprocessor until now. The Ports page now reads them each time it opens and switches a port's power off, on, or off and on again; a port switched off stays off across a reboot. Measured on the XGS 136 with nothing plugged into either port, so what changed is what the PoE controller reports, not a device's power.

Where the controller is¶

Not on the host's SMBus. The host's SMBus (intsmb0, the AMD chipset's) was probed read-only with smbmsg -p on 2026-10-20: it answers at the memory's SPD addresses, 0x30 to 0x37 and 0x50 and 0x51 (seven-bit), and nowhere else. The PoE controller (U40, a Microchip PD69104 by its marking) is on the coprocessor's I2C, as the cages' modules are, and the vendor's tool on the coprocessor, xgs-poe, reads and drives it. The plugin runs that tool there through scripts/xgs/npu-run.sh, over the management link when it is up, else the coprocessor's console - the way the Transceivers page reads the cages. The project distributes no part of the tool; it is the one the coprocessor already carries.

What the page does¶

XGS: Ports on the XGS 136, its Power over Ethernet section: port 11 on with no device, port 12 switched off, each with Switch off or on and Power cycle

Reads (scripts/xgs/poe.py show, api/xgs/ports/poe, under the Ports page's own privilege): the module - its status and its power budget, 62 W on this one - and each port: whether it is delivering power, what the controller detects on it, the class, the drawn power, voltage and current while a device draws, and the limit.
Switches (api/xgs/poe/set, under XGS: Settings, since a device on the port loses its power): off and on are kept in the configuration (general.poe_off, the labels switched off) and applied at once; a power cycle is off, three seconds, on, and kept nowhere. Every change is in the XGS log.
At every boot a start hook (rc.syshook.d/start/66-xgs-poe) switches off again a port the owner switched off, asking every 20 s for up to three minutes while the coprocessor comes up: the controller starts with every port on, by the tool's own help. With no port switched off it does nothing.
A powered device is never cycled by an apply. A port is switched on only when it reads switched off - delivering nothing, its detection "Unknown" - so a port that is powering something is not touched.

Measured¶

images/deploy/lab-136-poe.sh, photographed; poe.py md5 34a6ec00, the driver 85064db2. The page's request was made by Api/PoeController::setAction() itself, from the command line on the appliance, because the page asks for the owner's password.

step	what the controller reports afterwards
both ports as they start	module 0 Good, 62 W; 11 and 12 not delivering, detection Open Circuit - looking, nothing there
port 12 switched off from the page	config.xml `<poe_off>12</poe_off>`; port 12's detection Unknown - it stopped looking; port 11 unchanged
the controller switched back on behind the host's back, as at its start	port 12 Open Circuit again
the start hook, eight seconds later	port 12 Unknown again; the XGS log: port 12 switched off, as XGS: Ports has it
a power cycle of port 11	taken, four seconds; port 11 Open Circuit after it; the XGS log names it
a power cycle of port 12 while it is switched off, and port 5	both refused: port 12 is switched off; switch it on instead, port 5 carries no PoE

PoE switched from XGS: Ports on the XGS 136: port 12 off from the page, switched on behind its back and off again by the start hook, a power cycle of port 11, the refusals, and what the vendor's tool refuses on this module

What the controller on this module refuses. The vendor's tool also offers a port's priority, its power limit and the module's guard band; each answers Operation not supported here, so the page does not offer them. Its own RESET is refused on a port with no powered device, which is why the page's power cycle is off and on instead.

What it does not mean¶

No powered device was on either port. That the controller stops looking for one when a port is switched off is measured; that a camera or an access point loses its power and gets it back is not yet seen here. The drawn power, voltage and current columns have shown nothing so far for the same reason.
The page itself was not looked at on the appliance - it asks for the owner's password; its request was made by its own controller, as above. The page was looked at the same day, with the owner logged in: the section above is the XGS 136's, port 12 still switched off from the run.
A reboot of the appliance with a port switched off is the next measurement: the start hook was run by hand here.

Discussion

The discussion of this page is held in the project's GitHub Discussions. Showing it here loads giscus from giscus.app and GitHub, and writing in it needs a GitHub account - what that means.