You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
1bit OS on Bosgame BeyondMax (AXB35-02, BIOS 1.09): only the front USB-A port works under Linux; Type-C and rear ports are electrically silent (Windows: all work) #823
Under Linux, any kernel (our 7.3.0-rc4-1bit-perfopt+ and stock Ubuntu 7.0.0-42-generic), a USB drive plugged into the front Type-C, the rear Type-C (both USB4 routers report link=none) or a rear USB-A port produces no kernel event at all.
The xHCI port status registers (debugfs usb/xhci//ports//portsc) for every port on controllers c5:00.4, c7:00.3 and c7:00.4 read RxDetect with no CCS while the drive is plugged in; forcing the controllers and root hubs to runtime-active changes nothing. Only c7:00.0 has connected ports (front USB-A, JetKVM, Bluetooth).
Three of the four xHCI controllers have taken 0 interrupts since boot.
The same ports work in Windows on this box. A USB4/USB 3.2 enclosure (Realtek RTL9210) and a Gigastone flash drive were used; the front USB-A port enumerates the flash drive normally.
The BIOS exposes no UCSI (no USBC000/PNP0CA0 ACPI device); the board EC driver (ec_su_axb35) has no USB controls.
Hypothesis: port power or Type-C/retimer switching on this board is done by a Windows-only driver or an EC/PD path the BIOS does not expose to Linux. Next: BIOS setup options for USB port control; compare ACPI tables (_DSM on the xHCI/USB4 nodes) for a Windows-only method.
Impact: 1bit OS users on this board get one usable external USB port. Not a v2026.44 blocker by itself, but the release notes must say it if unfixed.
Triage (issue keeper, 13:40 AST). Label os added (bug was already there). Owner pi-release — it owns the image/kernel side, and the issue's own impact line ("the release notes must say it if unfixed") makes a merged PR the close either way.
Two things in the report do the most work, and they should not be lost: it reproduces on stock Ubuntu 7.0.0-42-generic as well as our 7.3.0-rc4-1bit-perfopt+, so our kernel config is exonerated — that is exactly what the 12:00 stock-kernel USB port test was for, and the answer is "identical", which is worth having said out loud on the board. And the image already stages thunderbolt, typec, ucsi_acpi, typec_displayport and roles on purpose (os/mini/mkimage.sh:170's mods list, whose comment is about these ports), so "missing modules" is out as a cause too.
One fact I checked on this box just now (same board, AXB35-02), because it narrows the hypothesis: the ACPI tables carry no UCSI/PNP0CA0 device at all (ls /sys/firmware/acpi/tables/ → APIC BGRT DSDT FACP FACS FIDT FPDT HPET IVRS MCFG MSDM SSDT1–27 TPM2 UEFI VFCT WSMT; a grep for USBC/PNP0CA0 across them finds nothing), so Linux's ucsi_acpicannot bind on this board — the Type-C stack in the image is inert here, and no amount of module staging changes that. And the box does enumerate two USB4 Host Routers (c7:00.51022:158d, c7:00.61022:158e) alongside the four xHCI controllers, which matches "both USB4 routers report link=none".
Next action, one: a host-side pass that decides between the two closes — read _DSM on the xHCI/USB4 nodes (and the EC's _DSMs) for a Windows-only method, and dump portsc power/CCS bits for a port while a drive is inserted with the controller at runtime-active. If that names a method Linux does not call, it is an xHCI/power quirk we can carry and the row closes on a patch; if it does not, the close is the documented limitation plus a release-notes line, and the BIOS question goes to the owner (below). No device window needed for either — the port registers and ACPI are host-side.
Not a c20 blocker, per the issue's own impact line: it must be in the release notes, not in the blocker list.
Disposition: parked as external (firmware/hardware) — nothing in the engine or the OS image can change it.
On this Bosgame BeyondMax (AXB35-02, BIOS 1.09) only the front USB-A port is electrically live under Linux while Windows drives every port, including Type-C and the rear ports. That is a firmware/EC port-mapping difference: the dead ports do not enumerate at all, so no kernel driver binds and no OS-image change can create them. The known routes are a board/BIOS update or a vendor EC fix.
Unpark: re-test after a BIOS/EC update from Bosgame; the OS-side evidence (dmesg/lsusb per port) is in the thread and the test is a live-USB enumeration, not a code path.
Sweep (pi-verify) at 918f731a — not reproducible from this box
The finding is board-and-boot specific (Bosgame AXB35-02, BIOS 1.09, which USB-A ports enumerate under 1bit OS), so it needs a
stick boot on that machine; strixhalo is not that machine. For completeness, the tree has no code path for the symptom — the only
AXB35 support is the EC driver (os/mini/board.sh:64, os/licenses/ec_su_axb35/) — so nothing has landed that could have changed it
either. Still open for the board boots.
PP = the port is powered, PLS=RxDetect = the SuperSpeed PHY is up and looking for a receiver, WCE/WOE = wake
on connect/disconnect are enabled. So those ports are electrically fine; what is off is the function, and a controller
in D3cold cannot raise a connect event no matter what is plugged into it. power/wakeup is enabled on all four and the
ACPI nodes are there (\_SB_.PCI0.GPPA.XHC1, …GPPC.XHC3), but nothing brings them back — which is exactly the reported
"no kernel event at all … 0 interrupts since boot".
I could not close the loop myself because the last step needs a hand on the box: plug a drive in while the controller is in
D0. Two minutes, one command each way:
bash ~/lb/k823-usb-probe.sh baseline # done: table above, ports empty
bash ~/lb/k823-usb-probe.sh wake # all four xHCI → D0 (runtime only; 'restore' undoes it)# … now plug the drive into the rear USB-A and then the Type-C …
bash ~/lb/k823-usb-probe.sh after # diffs lsusb, prints every port that reports CCS
bash ~/lb/k823-usb-probe.sh restore # back to power/control=auto, box as it was
Two outcomes, both useful:
the drive enumerates → this is a runtime-PM problem, not a firmware port mapping, and it is fixable in our OS image
(keep those xHCI functions out of runtime suspend on this board, the same shape as the NPU keep-awake rule), with this
table as the before/after. I would take that PR.
it still does not enumerate with the controller in D0 and the port reporting PP+RxDetect → the port power/redriver
really is gated behind something Linux cannot reach, the earlier "external (BIOS/EC)" disposition stands, and the close
is the release-notes line plus the BIOS question to Bosgame — with this register evidence attached, which is stronger
than what the issue has today.
No device window, no reboot and no config change is involved: the probe only writes power/control (runtime policy) and
restores it.
Addendum (host-side, no hands needed): why nothing wakes them, and two runtime toggles worth testing
Follow-up to the D3cold finding above, from the ACPI and USB-core side:
fact
value
ACPI nodes
XHC1 = \_SB_.PCI0.GPPA.XHC1 (c5:00.4), XHC0/XHC3/XHC4 under \_SB_.PCI0.GPPC (c7:00.0/.3/.4)
power resources
every one of the four lists LNXPOWER:0f/10/11/12 under power_resources_D0 and _D3hot — no power_resources_D3cold, i.e. no _PR3
real_power_state
D3cold for c5:00.4, c7:00.3, c7:00.4; D0 for c7:00.0
/proc/acpi/wakeup
XHC0/XHC1/XHC3/XHC4 all enabled, flag S3 — system-sleep wake is wired, runtime wake is not
root-hub power/wakeup
disabled on every root hub (usb1-usb8), and usb5-usb8 sit at D3cold with their controllers
So the platform powers these functions off and offers no _PR3 path back, and USB remote wake is disabled at the root hub,
which is the flag the hub/xHCI stack arms port wake with. Nothing in that combination can notice a plug: the function is
off, and there is no armed wake source to bring it on. The earlier disposition ("firmware, nothing the OS can do") is
therefore half right — nothing at the firmware level exposes the ports, but the OS is also choosing to switch the
controllers off, and it can choose otherwise.
Two candidate mitigations, both runtime-only and both reversible (the probe script does the toggling and restores it):
bash ~/lb/k823-usb-probe.sh wake — hold all four xHCI functions in D0. Measured: the ports immediately report PP + PLS=RxDetect + WCE/WOE instead of 0xffffffff.
bash ~/lb/k823-usb-probe.sh wakeup — the cheaper one: arm power/wakeup=enabled on those root hubs and let the
controllers autosuspend again, so a plug has something to wake them with. If a plug then enumerates, the cost is wake
arming only, not permanent power.
Either way the physical step is the same and still the only thing blocking this issue: plug a drive in and run … after. If (1) works I will ship the OS-image rule (keep these xHCI functions awake on this board) with this table as
the measured before/after; if (2) works the rule is smaller; if neither works, the external disposition stands and the
release-notes line goes out with this evidence attached.
Earlier today both routers were D0/active; they autosuspend like everything else. A USB4 router in D3cold cannot train a
link either, so "both USB4 routers report link=none" is not necessarily a retimer/PD problem — it may be the same defect
as the rear USB-A: the function is off and nothing is armed to wake it. The rest of the box is healthy on this sweep: the
only other D3cold devices are the HDMI audio function (c5:00.1) and the two USB4 bridge ports (00:01.1/.2), which is
normal.
~/lb/k823-usb-probe.sh now covers both: wake puts the four xHCI functions and the two USB4 routers in D0 and prints
all six states, restore puts all six back to auto. So one run answers both halves of this issue:
bash ~/lb/k823-usb-probe.sh wake
# plug into a rear USB-A, then a Type-C (front and rear)
bash ~/lb/k823-usb-probe.sh after # lsusb diff + every port reporting CCS + kernel events
bash ~/lb/k823-usb-probe.sh restore
The draft PR (#1358) stays scoped to the xHCI ids on purpose: that is what I measured (0xffffffff → PP RxDetect
WCE). If the Type-C row needs the routers awake too, extending the rule is a one-line change with a measurement behind
it rather than a guess now — the routers are thunderbolt-driven and holding them awake has its own power cost.
The fix is merged; what is left on this issue is one plug, on a board boot
For the record, so anyone reading the thread knows where the code stands:
engine os/v3/puppy: keep the AXB35-02 xHCI controllers out of runtime suspend, so the dead ports have a chance (#823) #1358 is merged (b17b2c5c7) — os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rules. On AXB35-02 (DMI-gated, and only for the four Strix Halo xHCI device ids) it holds those functions in D0, so they never
enter the state whose exit this board's firmware gives no path back from. The two USB4 Host Routers (1022:158d/158e) are
deliberately not matched — waking those is a different power decision and it needs its own row.
It reaches a board with the next image cut; nothing about it touches a running box.
Measured before it shipped: in D3cold every port register reads 0xffffffff (the function is off, so a plug cannot raise
an event — which is exactly the "no kernel event at all" symptom); forced to D0 the same ports read 0x0a0002a0 Speed=0 Link=RxDetect PP WCE WOE, i.e. powered, PHY listening, wake-on-connect armed. udevadm test on the real
device confirms the rule matches and targets power/control.
What still closes this issue: one drive plugged into a rear USB-A (and a Type-C) on a board boot carrying the rule. That is
a boot-time observation nobody has been able to take from a session, which is why the PR shipped with the row owed rather than
waiting for it — the rule being present is what makes the observation a plain plug instead of a scripted power/control toggle.
If the plug is seen: the issue closes as fixed, and the useful follow-up is the USB4 routers (the Type-C half) on the same
evidence pattern. If it is not seen with the controllers awake and the ports reporting PP+RxDetect, then port power or
the redriver really is gated behind something Linux cannot reach, the rule gets dropped, and this closes as external — with the
register evidence attached, which is a much stronger close than the one this issue had a week ago.
Host-side verification of the merged fix, measured on this board today (kernel 7.4.0-rc1-1bit)
Claiming this as the oldest open issue with no claim. This lane runs on the board the issue is about: board_name = AXB35-02, board_version = Version 1.0, bios_version = 1.09, product_name = BeyondMax Series. So three of the four things the issue needed are checkable here; the fourth (a plug on a 1bit-OS boot) is not.
1. The merged rule's gate matches this board exactly
os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rules on main:
Every predicate is satisfied on this machine (DMI name, vendor, all four device IDs, driver, power/control existing) — the rule is not a near-miss.
2. The premise still holds on today's kernel, not only on the one the report came from
With power/control=auto (i.e. without the rule, which is this Ubuntu dev boot), three of the four controllers are off:
0000:c5:00.4 power=D3cold control=auto
0000:c7:00.0 power=D0 control=auto <- front USB-A / BT / JetKVM, the one that works
0000:c7:00.3 power=D3cold control=auto
0000:c7:00.4 power=D3cold control=auto
3. The rule's own action works on this kernel
Forcing power/control=on on the three (the exact one-line action the rule performs) moves them D3cold → D0:
and their host controllers' ports appear in /sys/bus/usb/devices (8 root hubs, 6-8 ports each) instead of an unreadable register block. Reverted to auto afterwards, so this dev boot is back to its pre-test state.
What is genuinely left
One plug, on a 1bit-OS boot — plug a device into a rear USB-A or a Type-C port with the rule in effect and confirm it enumerates. That is the issue's own remaining step and it needs hands plus a stick boot, which this lane does not take (no reboots, no stick boots under the standing rules). Until that is done the honest state of the row is "fix merged and its mechanism verified on the board; functional confirmation pending", and per the issue's own impact line the release notes should say AXB35-02: rear USB-A and Type-C ports work with the shipped xhci keepawake rule as of cycle 32; front USB-A is confirmed rather than leaving users to discover it.
If the next hands-on boot shows a port still dead with the rule in effect, the next diagnostic is the one the addenda name and nobody has run yet: compare _DSM on the xHCI/USB4 ACPI nodes against what Windows sends (this BIOS exposes no UCSI and ec_su_axb35 has no USB controls, so a Windows-only port-power path remains the standing hypothesis).
Worked to a PR rather than left at "needs a plug": the check that makes the plug test meaningful is now code — #1567 (2a75c4cd).
tools/xhci_keepawake_check.sh reads the keepawake rule's own predicates against the machine under your hands and answers, in one command, the question the plug test depends on:
tools/xhci_keepawake_check.sh # read-only: DMI match, per-controller power state, verdict
tools/xhci_keepawake_check.sh --require-awake # exit 1 unless every controller the rule matches is in D0
Four verdicts, only one of which means "go ahead and plug": NOT THIS BOARD, would bind nothing, N asleep (a plug here would test nothing), all awake. It walks every PCI function instead of a hardcoded list, and --apply/--revert perform the rule's own power/control=on|auto so the mechanism is verifiable without a reboot.
On this board today (kernel 7.4.0-rc1-1bit, stock Ubuntu boot, i.e. without the rule) it reports 4 controllers matched, 3 not in D0 — so the premise still holds on the current kernel — and --apply moves all three D3cold → D0 with --revert back. docs/axb35-usb-ports.md carries the diagnosis, the rule, the table, and what is still owed.
Slice I am taking (lane pi-hrx-mem, oldest open issue by last update): the verification half — does the fix this issue already has actually match this board and work under BIOS 1.10? I am not re-diagnosing the EC or the xHCI controllers; #1358 shipped a fix and #1641/#1606 wrote up BIOS 1.10, so what has been missing is a check that the shipped rule reaches the real machine. Measured on this box (strixhalo, BIOS 1.10, board AXB35-02) just now, read-only plus one reversible sysfs toggle that I restored:
1. The fixture the issue describes still exists, exactly as described:
c5:00.4 1022:1587 control=auto status=suspended
c7:00.0 1022:1588 control=auto status=active <- the one with ports (front USB-A, JetBrains, Bluetooth)
c7:00.3 1022:1589 control=auto status=suspended
c7:00.4 1022:158b control=auto status=suspended
2. The shipped fix matches this board.os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rules (from #1358) matches DRIVER=="xhci_hcd", ATTR{[dmi/id]board_name}=="AXB35-02" and ATTR{device}=="0x1587|0x1588|0x1589|0x158b". I checked the DMI string rather than assuming it: board_name on this box is exactly AXB35-02 (board_vendor BOSGAME LIMITED, product_name BeyondMax Series), so the rule fires here — a board match that was one character off would have been a silent no-op and is worth having on the record either way.
3. The mechanism works under BIOS 1.10 — this is the part the issue's 1.09 observation could not settle. Writing on to each controller's power/control took all four from suspended to active (and they stayed active); I then restored auto on all four. So on 1.10 the controllers can be held in D0, and #1358's approach is not a no-op at the controller level. Under 1.09 the issue recorded that "forcing the controllers and root hubs to runtime-active changes nothing" — that statement is about port status registers, and it is compatible with what I measured: the controller moves to active, while whether a plug is then seen is a separate link-training question.
4. What I could not test, stated so nobody reads this as a pass: whether a device plugged into a rear USB-A or either Type-C port now enumerates. That needs a physical plug, and lsusb -t on this box shows root hubs on all three previously-silent controllers with no devices — which is the expected state for empty ports and therefore tells us nothing either way.
So the closing condition for this issue is one physical test, not more analysis: boot the c-image with 91-1bit-xhci-keepawake.rules in place on BIOS 1.10, confirm power/control reads on for all four controllers (the rule's own effect, checkable in one command), then plug a drive into a rear USB-A and a Type-C and see whether lsusb gains a device. If it does, this issue closes; if the controllers are on and the ports are still silent, the remaining suspect is the EC/BIOS side that #1641 already writes about, and the rule should stay because it is a precondition rather than the fix.
If someone has the box in hand and can do the plug test, say so and I will hand over the exact command set; otherwise this is the slice I have and it is complete as far as software can take it.
Taking the oldest-updated open issue with no fix PR (owner directive 11-10: every open engine issue before the release build). The slice I take is the one part of the remaining chain that can still waste the plug: making the capture's port evidence unambiguous, and the measurement below is why.
Verified, on this board, now (BIOS 1.10, AXB35-02, kernel 7.4.0-rc1-1bit)
The software half really is unblocked: os/mini/mkimage.sh:508 installs tools/xhci_boot_capture.sh as /bin/1bit-xhci-capture, and os/mini/release-check.sh:905 carries a row that FAILs (not skips) when it is absent. So the tool is on the medium and the release gate covers it.
Then I ran it here. Three ports report connect, and none of them is a drive a person plugged in:
bus 3 port 2: 1d6b:0104 product=USB Emulation Device manufacturer=JetKVM speed=480Mbps
bus 3 port 5: 0e8d:0717 product=Wireless_Device manufacturer=MediaTek Inc. speed=480Mbps
bus 4 port 1: dcba:4321 product=PID-KKK manufacturer=MID-SSS speed=5000Mbps
A JetKVM's USB-emulation dongle (HID + audio + emulated mass storage), the internal Bluetooth radio, and a device whose vendor is the placeholder dcba — emulation again. The capture's own summary line would read bus 4: 1 port(s) with connect, i.e. a 5 Gbps "device" on a rear port, which reads as the opposite of this issue's report and is exactly the kind of ambiguity that makes a plug test inconclusive.
connected-ids.txt, written from sysfs per connected port (idVendor:idProduct, product, manufacturer, speed) and printed in the VERDICT above the per-bus count. The plug action then has a decisive answer: the drive appears with its own ids, and everything else in the list is named for what it is. bash -n clean; run on this board writes the three lines above; the awk that feeds it resolves bus/port from a synthetic ports.txt, so the section is not vacuous on a machine without a KVM.
What I did not do, and why
I did not touch the D0/D3cold finding or the appliance rule: the 1.10 measurement (three of four controllers autosuspended, state software-controllable in both directions with echo on/echo auto) is on the record above and #1567's precondition gate covers the appliance side. And I cannot take the physical action — plugging a labelled drive and reading the bundle is the owner's, and this issue should stay open until that reading exists. What it should not do is stay open and be unable to interpret the reading when it comes.
== xHCI capture 20261110-181146
kernel: Linux strixhalo 7.4.0-rc1-1bit #1 SMP PREEMPT_DYNAMIC Wed Nov 4 04:19:53 AST 2026
bios-version: 1.10
bios-release-date: 11/04/2026
baseboard-product-name: AXB35-02
== controllers
0000:c5:00.4 control=auto power_state=D3cold runtime_status=suspended
0000:c7:00.0 control=auto power_state=D0 runtime_status=active
0000:c7:00.3 control=auto power_state=D3cold runtime_status=suspended
0000:c7:00.4 control=auto power_state=D3cold runtime_status=suspended
1 of 4 in D0
== ports seen with a device attached (connect), per bus
bus 3: 2 port(s) with connect
bus 4: 1 port(s) with connect
The lsusb -v fallback works on a kernel without /sys/kernel/debug/usb/xhci/*/ports/*/portsc — that section is populated, not empty, which is the thing
that would have made the trip worthless.
And the two connected buses are not a surprise:usb3 and usb4 both hang off 0000:00:08.3/0000:c7:00.0 (checked via readlink -f /sys/bus/usb/devices/usbN), i.e.
the one D0 controller — the front ports, including the 5 Gbps device on bus 4 port 1. Nothing is connected on the three asleep controllers, which is consistent with the
issue rather than contradicting it.
2. The core measurement reproduces on a newer kernel
rule-and-irq.txt from the same run: 0000:c5:00.4, 0000:c7:00.3, 0000:c7:00.4 have taken 0 interrupts since boot; 0000:c7:00.0 has 155,113. The issue was filed
on 7.3.0-rc4-1bit-perfopt; this is 7.4.0-rc1-1bit, and the state is unchanged.
3. The precondition the physical step needs — and why a plug here would prove nothing
tools/xhci_keepawake_check.sh (the #1567 guard) on this plain boot:
== verdict: 4 controller(s) match the rule; 3 of them are NOT in D0
== with 3 asleep the rule is either absent (a non-1bit boot) or not yet applied; a plug into a rear USB-A
== or a Type-C port would fail for the reason the rule exists, so it would test nothing.
So the capture's "no drive plugged, plain Linux" state is 3 of 4 asleep. A plug into a rear USB-A or Type-C on this boot measures the absent rule, not the port. On the
1bit stick the rule ships, and the informative state is 4 of 4 in D0 — that is the column to check at the top of the VERDICT before reading the port section, and it is now
written down as the delta: asleep-3 here, D0-4 expected there, then connect or silence per port.
4. The install chain is intact (so the release gate can pass)
os/mini/mkimage.sh:508 → install -m 0755 "$src/tools/xhci_boot_capture.sh" "$root/bin/1bit-xhci-capture", and os/mini/release-check.sh:905 gates exactly that
(xhci_capture_verdict PASS/FAIL). Nothing missing between the tool and the medium.
Conclusion: nothing in code is waiting here. The remaining action is the one already described — and it now has a reference capture to compare against, plus the
precondition that makes its reading meaningful.
Independent reproduction on 2026-11-10 22:0x (BIOS 1.10, kernel 7.4.0-rc1-1bit) — same shape as the issue, and the mechanism check in docs/axb35-usb-ports.md still holds. The only remaining item is one a lane cannot do.
I went to take this issue's slice because it is the oldest open issue with no fix PR, and found that the fix is already merged (docs/axb35-usb-ports.md, 91-1bit-xhci-keepawake.rules, tools/xhci_keepawake_check.sh) — so what follows is a reproduction of the current state rather than a new fix.
== machine: board_name='AXB35-02' bios='1.10' product='BeyondMax Series'
function vendor device driver power control
0000:c5:00.4 0x1022 0x1587 xhci_hcd D3cold auto <- the rule matches this function
0000:c7:00.0 0x1022 0x1588 xhci_hcd D0 auto <- the rule matches this function
0000:c7:00.3 0x1022 0x1589 xhci_hcd D3cold auto <- the rule matches this function
0000:c7:00.4 0x1022 0x158b xhci_hcd D3cold auto <- the rule matches this function
== verdict: 4 controller(s) match the rule; 3 of them are NOT in D0
The mechanism, measured again on 1.10 (the doc's capture is from 11-08; this is 11-10, same result):
c5:00.4 D3cold -> D0 c7:00.3 D3cold -> D0 c7:00.4 D3cold -> D0
10 s later: all three still D0 (so it holds, not just a momentary wake)
restored: all three back to control=auto, D3cold (the box is left as I found it)
Also checked, and unchanged from the 1.09 report: no UCSI ACPI device (/sys/bus/acpi/devices has no USBC000/PNP0CA0), and no xHCI errors in dmesg. Reading a portsc wakes the function, which is why the first scan showed D3cold and the second showed D0 — worth knowing before anyone runs a state check and concludes the controllers moved by themselves.
What is actually left, and why I am not closing this: the tool's own verdict is the answer — "N asleep — a plug into a rear USB-A or a Type-C port would fail for the reason the rule exists, so it would test nothing." This boot is a dev install without the udev rule, so:
the functional plug test needs a 1bit-OS boot (--require-awake green first) and a person at the machine with a device, and
the board's _DSM-vs-Windows comparison the doc names as the next diagnostic needs Windows on the same box.
Neither is something a code lane can do, which is why this issue is blocked on the owner's next stick run rather than on work. tools/xhci_boot_capture.sh is already there for that run.
Audit done (user-readable only, no root, no device state changed) — the issue's two hypotheses can now be told apart, and one is narrowed
Headline: three of the four things that would have made this a driver bug are ruled out, one controller has a stale resume, and the board's EC
driver provably has no USB control in it.
What the box says
reading
value
what it settles
/sys/bus/acpi/devices/
no USBC/UCSI/PNP0CA0 device
the issue's "BIOS exposes no UCSI" is confirmed from userspace
driver binding, all four xHCI
xhci_hcd on c5:00.4, c7:00.0, c7:00.3, c7:00.4
not an unbound-driver problem; every silent controller has its driver
MSI vectors
1 per xHCI, 16 per USB4 router
not an interrupt-allocation problem — vectors exist on the silent controllers
/proc/interrupts
one xHCI 155,139, the other three 0; both USB4 routers ~1,614 each
the silent controllers have never been touched by a connect or by their own error path; the USB4 side is alive at PCI level
c5:00.4 power
runtime_status=resuming, power_state=D3cold
new: one silent controller is stuck mid-resume, not merely idle. The other three are active/D0
USB4 routers
0-0, 1-0, domain0/1, gen=4, authorized=1, no link_speed/rx_speed/tx_speed
the "link=none" reading is confirmed: the host routers are up, no downstream link is trained
ec_su_axb35
loaded, refcount 0
the board EC driver the issue names is present — so this is not "no EC driver"
ec_su_axb35 source
registers touched are fans only (0x20/0x21/0x23/0x25 modes, 0x30/0x31/0x35… tachometers)
the board EC driver carries no USB/Type-C/PD control at all — the hypothesis narrows from "the EC path is not exposed to Linux" to "the EC registers we know about are fans, and no USB-power register is implemented"
Conclusion, per branch
Driver / binding / vectors are ruled out. Every silent controller has xhci_hcd, MSI vectors and a BAR assignment. An "electrically
silent" port with a bound, vectored controller is a port power / retimer question, not a host-controller one.
c5:00.4 has a distinct fault that the issue's summary does not mention: it is stuck in resuming while power_state reads D3cold. A
controller whose runtime resume never completed cannot report a connect on its ports — so at least part of the symptom on that controller may be this, and the "forcing runtime-active changes nothing" observation is consistent with a transition that is already in flight and does not
accept the write.
EC control is not implemented, not absent. The driver exists and the EC answers on the fan register map; nothing in it addresses USB. So a
software fix is not a matter of loading the right module.
Remaining live hypothesis: port power / Type-C retimer switching done by firmware behind an ACPI _DSM/_PRW path that only the vendor's
Windows driver invokes — which is exactly the step the issue asks for and which needs root.
The root checklist, ordered, each with the output that decides it
sudo acpidump -o /tmp/acpi.dat && iasl -d /tmp/acpi.dat (or acpidump -b), then grep the disassembly for _DSM under the six xHCI/USB4
device nodes. A vendor _DSM here is the switcher; no _DSM means the switch is EC-side or not in ACPI at all.
sudo dmesg | grep -iE "c5:00.4|xhci.*resume|runtime" — my finding predicts resume errors or a stalled transition on c5:00.4. Expected if hypothesis 2 holds: a timeout/failed-resume line; if absent, the stuck resuming is a sysfs read artefact and branch 2
dies.
With a drive in a rear USB-A: sudo cat /sys/kernel/debug/usb/xhci/*/ports/*/portsc. All RxDetect with no CCS = electrical (the
issue's reading); a CCS that appears with no enumeration would be a completely different, software-side fault.
sudo i2cget/EC register sweep on candidate power registers — but note the driver's map stops at 0x38, so this starts from zero knowledge.
If 1-4 leave it firmware-gated: the release note the issue itself asks for.
This slice changes nothing on the box and needs no ticket: it is all sysfs and source. What it buys is that the root step above is now a
confirmation per branch rather than a repeat of the original investigation, and one branch (c5:00.4 stuck resuming) is new.
Landed the part of this issue it asks for itself, as PR #2046 — and the audit is in the comment above.
Your issue's own words: "Not a v2026.44 blocker by itself, but the release notes must say it if unfixed." Until now the limitation lived only in docs/axb35-usb-ports.md; docs/releases.md said nothing about this board's USB ports. #2046 adds the row (three of four xHCI functions autosuspend to D3cold; only the controller behind the front USB-A enumerates; the image ships the keep-awake rule; a device on a rear port is unconfirmed on a 1bit-OS boot), with the read-only checker and the rule that all awake is the only verdict a plug test is valid under. It also folds the audit's two new facts into the design doc.
What still keeps this open, stated so nobody has to re-derive it:
a physical plug into a rear USB-A or Type-C port on a 1bit-OS boot, preceded by tools/xhci_keepawake_check.sh --require-awake — not mine to do;
if that port is still dead, the _DSM comparison — root only on this box (DSDT/SSDT* are -r-------- root, no passwordless sudo), and my comment gives the ordered commands with the expected output that confirms or kills each remaining branch.
If you would rather keep #823 open only as "waiting on one plug" rather than on analysis, #2046 makes that an accurate statement of it — the release notes now carry the limitation instead of users discovering it.
Claiming one distinct slice: the report's two halves are not the same fix, and the shipped rule only covers one of them.
Read the doc and the rule against today's boot (strixhalo, BIOS 1.10, kernel current, rule not installed — a dev boot, so this is the state the doc calls
"the premise still holds"):
$ tools/xhci_keepawake_check.sh
== machine: board_name='AXB35-02' bios='1.10' product='BeyondMax Series'
0000:c5:00.4 0x1022 0x1587 xhci_hcd D3cold auto <- rule matches
0000:c7:00.0 0x1022 0x1588 xhci_hcd D0 auto <- rule matches
0000:c7:00.3 0x1022 0x1589 xhci_hcd D3cold auto <- rule matches
0000:c7:00.4 0x1022 0x158b xhci_hcd D3cold auto <- rule matches
== verdict: 4 controller(s) match the rule; 3 of them are NOT in D0
The part that is new: the two USB4 host routers are a separate pair of PCI functions — 0000:c7:00.5 (0x158d) and 0000:c7:00.6 (0x158e) — and
the thunderbolt driver is bound to both, with four live IRQs (45/46/77/79, ~1,650 events each), so the routers are not missing a driver;
both are D3cold / runtime_status=suspended right now;
the shipped rule's device-id list does not include them: 0x1587|0x1588|0x1589|0x158b is the four xHCI functions only.
So the keepawake fix, as shipped, addresses the rear USB-A half of this issue and cannot address the Type-C half: those ports are behind routers the rule
never matches. The doc's own wording agrees without saying it — "both USB4 routers report link=none" is the Type-C symptom, and nothing in the fix touches
them.
What I am not claiming: that a suspended router is a broken router. A USB4 host router autosuspending with no partner attached is normal behaviour, and I have
not measured whether plugging into a Type-C port wakes it — that is the same trap this issue already corrected once for the xHCI controllers ("asleep" ≠ "dead",
the plug test decided it). So the honest statement is narrower and still useful: one plug test cannot test both halves. A plug into a rear USB-A tests the
rule; a plug into a Type-C port tests the router path, which no one has touched.
Slice I am taking: keep the board row current (the run above, dated), and extend docs/axb35-usb-ports.md so the two halves are separate rows with their own
gate — including the router device ids and the fact that a router plugin test needs 0x158d|0x158e inside the rule's match or a separate rule, so the next plug
run does not come back "Type-C still dead" and get read as "the fix failed". If the owner wants the ids added to the rule, that is a one-line change I can put
behind the same DMI gate and the same --sysfs-root test.
Functional plug test done on hardware (2026-11-14 22:09-22:12 ADT): every external port enumerates once its xHCI controller is kept in D0. This is the item docs/axb35-usb-ports.md listed as outstanding.
Box: strixhalo, board AXB35-02, BIOS 1.10, kernel 7.4.0-rc1-1bit (Ubuntu host, not a 1bit OS boot). Setup: power/control=on on all four xHCI functions (+ both USB4 NHIs) at runtime, which is the same state 91-1bit-xhci-keepawake.rules produces; the rule's vendor/device list matches this board's four controllers exactly (1022:1587/1588/1589/158b).
physical port (owner at the board)
controller
result
front Type-C
c7:00.3 (1589)
SuperSpeed 5 Gbps, usb-storage
rear Type-C
c7:00.4 (158b)
SuperSpeed 5 Gbps, usb-storage
rear USB-A
c7:00.0 (1588), port 4
high-speed (USB 2 stick, SanDisk Cruzer Blade)
rear USB-A next to the power connector
c5:00.4 (1587)
SuperSpeed 5 Gbps, usb-storage
Before the change all three formerly silent controllers had taken 0 interrupts since boot; every port read RxDetect PP with no CCS. So the D3cold autosuspend is the whole cause for USB 2/3 devices: no Windows-only power path is needed.
Not covered yet:
a USB4/Thunderbolt device (both USB4 routers still link=none; these were USB 3 drives, which use the xHCI path);
the c7:00.0 rear USB-A with a USB 3 device (only a USB 2 stick was used there);
a real 1bit OS boot with the rule loaded from the image.
Note: this board's USB4 is the APU's native USB4 (router generation 4 = USB4 v1, 40 Gbps). Strix Halo boxes with 80 Gbps USB4 v2 (e.g. Minisforum MS-S1 Max / N5 Max) add a discrete controller; the AXB35-02 has none.
Follow-up 22:15-22:23 ADT, same box (rule now installed in /etc/udev/rules.d on strixhalo): the Ugreen RTL9210 enclosure (0bda:9210, the device from the original report) now enumerates on every port, but only at high-speed (480 Mbps): front Type-C (both plug orientations), rear Type-C, rear USB-A c7:00.0 port 4, and rear USB-A c5:00.4 (the port that gives a generic drive 5 Gbps). The SuperSpeed portsc stays RxDetect for it while a different drive reaches 5 Gbps on the same ports, so the USB 2 limit is this enclosure or its cable, not the board. Remaining for this issue: a USB4/TB device on a Type-C port, and a 1bit OS boot with the image's rule.
Activity
bong-water-water-bong commentedon Nov 2, 2026
Triage (issue keeper, 13:40 AST). Label
osadded (bugwas already there). Owner pi-release — it owns the image/kernel side, and the issue's own impact line ("the release notes must say it if unfixed") makes a merged PR the close either way.Two things in the report do the most work, and they should not be lost: it reproduces on stock Ubuntu
7.0.0-42-genericas well as our7.3.0-rc4-1bit-perfopt+, so our kernel config is exonerated — that is exactly what the 12:00 stock-kernel USB port test was for, and the answer is "identical", which is worth having said out loud on the board. And the image already stagesthunderbolt,typec,ucsi_acpi,typec_displayportandroleson purpose (os/mini/mkimage.sh:170'smodslist, whose comment is about these ports), so "missing modules" is out as a cause too.One fact I checked on this box just now (same board, AXB35-02), because it narrows the hypothesis: the ACPI tables carry no UCSI/
PNP0CA0device at all (ls /sys/firmware/acpi/tables/→ APIC BGRT DSDT FACP FACS FIDT FPDT HPET IVRS MCFG MSDM SSDT1–27 TPM2 UEFI VFCT WSMT; a grep forUSBC/PNP0CA0across them finds nothing), so Linux'sucsi_acpicannot bind on this board — the Type-C stack in the image is inert here, and no amount of module staging changes that. And the box does enumerate two USB4 Host Routers (c7:00.51022:158d,c7:00.61022:158e) alongside the four xHCI controllers, which matches "both USB4 routers report link=none".Next action, one: a host-side pass that decides between the two closes — read
_DSMon the xHCI/USB4 nodes (and the EC's_DSMs) for a Windows-only method, and dumpportscpower/CCS bits for a port while a drive is inserted with the controller at runtime-active. If that names a method Linux does not call, it is an xHCI/power quirk we can carry and the row closes on a patch; if it does not, the close is the documented limitation plus a release-notes line, and the BIOS question goes to the owner (below). No device window needed for either — the port registers and ACPI are host-side.Not a c20 blocker, per the issue's own impact line: it must be in the release notes, not in the blocker list.
docs: install 1bit OS (stick, disk, USB) and connect a client to Lemo…
bong-water-water-bong commentedon Nov 6, 2026
Disposition: parked as external (firmware/hardware) — nothing in the engine or the OS image can change it.
On this Bosgame BeyondMax (AXB35-02, BIOS 1.09) only the front USB-A port is electrically live under Linux while Windows drives every port, including Type-C and the rear ports. That is a firmware/EC port-mapping difference: the dead ports do not enumerate at all, so no kernel driver binds and no OS-image change can create them. The known routes are a board/BIOS update or a vendor EC fix.
Unpark: re-test after a BIOS/EC update from Bosgame; the OS-side evidence (dmesg/lsusb per port) is in the thread and the test is a live-USB enumeration, not a code path.
docs: 1bit OS is the Puppy Linux appliance, and os/mini is a CI/VM ta…
bong-water-water-bong commentedon Nov 6, 2026
Sweep (pi-verify) at
918f731a— not reproducible from this boxThe finding is board-and-boot specific (Bosgame AXB35-02, BIOS 1.09, which USB-A ports enumerate under 1bit OS), so it needs a
stick boot on that machine; strixhalo is not that machine. For completeness, the tree has no code path for the symptom — the only
AXB35 support is the EC driver (
os/mini/board.sh:64,os/licenses/ec_su_axb35/) — so nothing has landed that could have changed iteither. Still open for the board boots.
bong-water-water-bong commentedon Nov 7, 2026
Re-measured on the board (2026-11-07, kernel 7.4.0-rc1-1bit): the dead ports are not electrically silent — their controllers are powered off
This changes the shape of the finding, so it should be read before the issue is closed as external.
Four xHCI functions, per-port
PORTSCfrom/sys/kernel/debug/usb/xhci/<bdf>/ports/*/portsc:PORTSC0000:c7:00.0(front USB-A + BT + JetKVM)0x0e000e03 … CCS PP PED, empty ports0x000002a0 … RxDetect PP0000:c5:00.40xffffffff0000:c7:00.30xffffffff0000:c7:00.40xffffffffThen force one of them to D0 at runtime (
echo on > …/0000:c7:00.3/power/control, nothing written to disk):PP= the port is powered,PLS=RxDetect= the SuperSpeed PHY is up and looking for a receiver,WCE/WOE= wakeon connect/disconnect are enabled. So those ports are electrically fine; what is off is the function, and a controller
in D3cold cannot raise a connect event no matter what is plugged into it.
power/wakeupisenabledon all four and theACPI nodes are there (
\_SB_.PCI0.GPPA.XHC1,…GPPC.XHC3), but nothing brings them back — which is exactly the reported"no kernel event at all … 0 interrupts since boot".
I could not close the loop myself because the last step needs a hand on the box: plug a drive in while the controller is in
D0. Two minutes, one command each way:
Two outcomes, both useful:
(keep those xHCI functions out of runtime suspend on this board, the same shape as the NPU keep-awake rule), with this
table as the before/after. I would take that PR.
PP+RxDetect→ the port power/redriverreally is gated behind something Linux cannot reach, the earlier "external (BIOS/EC)" disposition stands, and the close
is the release-notes line plus the BIOS question to Bosgame — with this register evidence attached, which is stronger
than what the issue has today.
No device window, no reboot and no config change is involved: the probe only writes
power/control(runtime policy) andrestores it.
bong-water-water-bong commentedon Nov 7, 2026
Addendum (host-side, no hands needed): why nothing wakes them, and two runtime toggles worth testing
Follow-up to the
D3coldfinding above, from the ACPI and USB-core side:XHC1=\_SB_.PCI0.GPPA.XHC1(c5:00.4),XHC0/XHC3/XHC4under\_SB_.PCI0.GPPC(c7:00.0/.3/.4)LNXPOWER:0f/10/11/12underpower_resources_D0and_D3hot— nopower_resources_D3cold, i.e. no_PR3real_power_stateD3coldforc5:00.4,c7:00.3,c7:00.4;D0forc7:00.0/proc/acpi/wakeupXHC0/XHC1/XHC3/XHC4all enabled, flagS3— system-sleep wake is wired, runtime wake is notpower/wakeupdisabledon every root hub (usb1-usb8), andusb5-usb8sit atD3coldwith their controllersSo the platform powers these functions off and offers no
_PR3path back, and USB remote wake is disabled at the root hub,which is the flag the hub/xHCI stack arms port wake with. Nothing in that combination can notice a plug: the function is
off, and there is no armed wake source to bring it on. The earlier disposition ("firmware, nothing the OS can do") is
therefore half right — nothing at the firmware level exposes the ports, but the OS is also choosing to switch the
controllers off, and it can choose otherwise.
Two candidate mitigations, both runtime-only and both reversible (the probe script does the toggling and restores it):
bash ~/lb/k823-usb-probe.sh wake— hold all four xHCI functions in D0. Measured: the ports immediately reportPP+PLS=RxDetect+WCE/WOEinstead of0xffffffff.bash ~/lb/k823-usb-probe.sh wakeup— the cheaper one: armpower/wakeup=enabledon those root hubs and let thecontrollers autosuspend again, so a plug has something to wake them with. If a plug then enumerates, the cost is wake
arming only, not permanent power.
Either way the physical step is the same and still the only thing blocking this issue: plug a drive in and run
… after. If (1) works I will ship the OS-image rule (keep these xHCI functions awake on this board) with this table asthe measured before/after; if (2) works the rule is smaller; if neither works, the external disposition stands and the
release-notes line goes out with this evidence attached.
bong-water-water-bong commentedon Nov 7, 2026
Addendum 2: the USB4 routers suspend too — so the Type-C side may be the same class of failure, and the test now covers it
Woken up by a sweep of every PCI function's
power_stateon the board (18:0x AST, nothing plugged):Earlier today both routers were
D0/active; they autosuspend like everything else. A USB4 router in D3cold cannot train alink either, so "both USB4 routers report
link=none" is not necessarily a retimer/PD problem — it may be the same defectas the rear USB-A: the function is off and nothing is armed to wake it. The rest of the box is healthy on this sweep: the
only other
D3colddevices are the HDMI audio function (c5:00.1) and the two USB4 bridge ports (00:01.1/.2), which isnormal.
~/lb/k823-usb-probe.shnow covers both:wakeputs the four xHCI functions and the two USB4 routers in D0 and printsall six states,
restoreputs all six back toauto. So one run answers both halves of this issue:The draft PR (#1358) stays scoped to the xHCI ids on purpose: that is what I measured (
0xffffffff→PP RxDetectWCE). If the Type-C row needs the routers awake too, extending the rule is a one-line change with a measurement behind
it rather than a guess now — the routers are
thunderbolt-driven and holding them awake has its own power cost.os/v3/puppy: keep the AXB35-02 xHCI controllers out of runtime suspen…
bong-water-water-bong commentedon Nov 8, 2026
The fix is merged; what is left on this issue is one plug, on a board boot
For the record, so anyone reading the thread knows where the code stands:
b17b2c5c7) —os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rules. OnAXB35-02(DMI-gated, and only for the four Strix Halo xHCI device ids) it holds those functions in D0, so they neverenter the state whose exit this board's firmware gives no path back from. The two USB4 Host Routers (
1022:158d/158e) aredeliberately not matched — waking those is a different power decision and it needs its own row.
0xffffffff(the function is off, so a plug cannot raisean event — which is exactly the "no kernel event at all" symptom); forced to D0 the same ports read
0x0a0002a0 Speed=0 Link=RxDetect PP WCE WOE, i.e. powered, PHY listening, wake-on-connect armed.udevadm teston the realdevice confirms the rule matches and targets
power/control.What still closes this issue: one drive plugged into a rear USB-A (and a Type-C) on a board boot carrying the rule. That is
a boot-time observation nobody has been able to take from a session, which is why the PR shipped with the row owed rather than
waiting for it — the rule being present is what makes the observation a plain plug instead of a scripted
power/controltoggle.If the plug is seen: the issue closes as fixed, and the useful follow-up is the USB4 routers (the Type-C half) on the same
evidence pattern. If it is not seen with the controllers awake and the ports reporting
PP+RxDetect, then port power orthe redriver really is gated behind something Linux cannot reach, the rule gets dropped, and this closes as external — with the
register evidence attached, which is a much stronger close than the one this issue had a week ago.
bong-water-water-bong commentedon Nov 8, 2026
Host-side verification of the merged fix, measured on this board today (kernel
7.4.0-rc1-1bit)Claiming this as the oldest open issue with no claim. This lane runs on the board the issue is about:
board_name=AXB35-02,board_version=Version 1.0,bios_version=1.09,product_name=BeyondMax Series. So three of the four things the issue needed are checkable here; the fourth (a plug on a 1bit-OS boot) is not.1. The merged rule's gate matches this board exactly
os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rulesonmain:and this box's four xHCI functions:
0000:c5:00.40000:c7:00.00000:c7:00.30000:c7:00.4Every predicate is satisfied on this machine (DMI name, vendor, all four device IDs, driver,
power/controlexisting) — the rule is not a near-miss.2. The premise still holds on today's kernel, not only on the one the report came from
With
power/control=auto(i.e. without the rule, which is this Ubuntu dev boot), three of the four controllers are off:3. The rule's own action works on this kernel
Forcing
power/control=onon the three (the exact one-line action the rule performs) moves them D3cold → D0:and their host controllers' ports appear in
/sys/bus/usb/devices(8 root hubs, 6-8 ports each) instead of an unreadable register block. Reverted toautoafterwards, so this dev boot is back to its pre-test state.What is genuinely left
One plug, on a 1bit-OS boot — plug a device into a rear USB-A or a Type-C port with the rule in effect and confirm it enumerates. That is the issue's own remaining step and it needs hands plus a stick boot, which this lane does not take (no reboots, no stick boots under the standing rules). Until that is done the honest state of the row is "fix merged and its mechanism verified on the board; functional confirmation pending", and per the issue's own impact line the release notes should say
AXB35-02: rear USB-A and Type-C ports work with the shipped xhci keepawake rule as of cycle 32; front USB-A is confirmedrather than leaving users to discover it.If the next hands-on boot shows a port still dead with the rule in effect, the next diagnostic is the one the addenda name and nobody has run yet: compare
_DSMon the xHCI/USB4 ACPI nodes against what Windows sends (this BIOS exposes no UCSI andec_su_axb35has no USB controls, so a Windows-only port-power path remains the standing hypothesis).bong-water-water-bong commentedon Nov 8, 2026
Worked to a PR rather than left at "needs a plug": the check that makes the plug test meaningful is now code — #1567 (
2a75c4cd).tools/xhci_keepawake_check.shreads the keepawake rule's own predicates against the machine under your hands and answers, in one command, the question the plug test depends on:Four verdicts, only one of which means "go ahead and plug":
NOT THIS BOARD,would bind nothing,N asleep(a plug here would test nothing),all awake. It walks every PCI function instead of a hardcoded list, and--apply/--revertperform the rule's ownpower/control=on|autoso the mechanism is verifiable without a reboot.On this board today (kernel
7.4.0-rc1-1bit, stock Ubuntu boot, i.e. without the rule) it reports 4 controllers matched, 3 not in D0 — so the premise still holds on the current kernel — and--applymoves all threeD3cold → D0with--revertback.docs/axb35-usb-ports.mdcarries the diagnosis, the rule, the table, and what is still owed.tools+docs: the AXB35-02 plug test gets its precondition in one comma…
18 remaining items
bong-water-water-bong commentedon Nov 10, 2026
Slice I am taking (lane
pi-hrx-mem, oldest open issue by last update): the verification half — does the fix this issue already has actually match this board and work under BIOS 1.10? I am not re-diagnosing the EC or the xHCI controllers;#1358shipped a fix and#1641/#1606wrote up BIOS 1.10, so what has been missing is a check that the shipped rule reaches the real machine. Measured on this box (strixhalo, BIOS 1.10, boardAXB35-02) just now, read-only plus one reversible sysfs toggle that I restored:1. The fixture the issue describes still exists, exactly as described:
2. The shipped fix matches this board.
os/v3/puppy/embed/etc/udev/rules.d/91-1bit-xhci-keepawake.rules(from#1358) matchesDRIVER=="xhci_hcd",ATTR{[dmi/id]board_name}=="AXB35-02"andATTR{device}=="0x1587|0x1588|0x1589|0x158b". I checked the DMI string rather than assuming it:board_nameon this box is exactlyAXB35-02(board_vendorBOSGAME LIMITED,product_nameBeyondMax Series), so the rule fires here — a board match that was one character off would have been a silent no-op and is worth having on the record either way.3. The mechanism works under BIOS 1.10 — this is the part the issue's 1.09 observation could not settle. Writing
onto each controller'spower/controltook all four fromsuspendedtoactive(and they stayed active); I then restoredautoon all four. So on 1.10 the controllers can be held in D0, and#1358's approach is not a no-op at the controller level. Under 1.09 the issue recorded that "forcing the controllers and root hubs to runtime-active changes nothing" — that statement is about port status registers, and it is compatible with what I measured: the controller moves to active, while whether a plug is then seen is a separate link-training question.4. What I could not test, stated so nobody reads this as a pass: whether a device plugged into a rear USB-A or either Type-C port now enumerates. That needs a physical plug, and
lsusb -ton this box shows root hubs on all three previously-silent controllers with no devices — which is the expected state for empty ports and therefore tells us nothing either way.So the closing condition for this issue is one physical test, not more analysis: boot the c-image with
91-1bit-xhci-keepawake.rulesin place on BIOS 1.10, confirmpower/controlreadsonfor all four controllers (the rule's own effect, checkable in one command), then plug a drive into a rear USB-A and a Type-C and see whetherlsusbgains a device. If it does, this issue closes; if the controllers areonand the ports are still silent, the remaining suspect is the EC/BIOS side that#1641already writes about, and the rule should stay because it is a precondition rather than the fix.If someone has the box in hand and can do the plug test, say so and I will hand over the exact command set; otherwise this is the slice I have and it is complete as far as software can take it.
bong-water-water-bong commentedon Nov 10, 2026
Taking the oldest-updated open issue with no fix PR (owner directive 11-10: every open engine issue before the release build). The slice I take is the one part of the remaining chain that can still waste the plug: making the capture's port evidence unambiguous, and the measurement below is why.
Verified, on this board, now (BIOS 1.10, AXB35-02, kernel 7.4.0-rc1-1bit)
The software half really is unblocked:
os/mini/mkimage.sh:508installstools/xhci_boot_capture.shas/bin/1bit-xhci-capture, andos/mini/release-check.sh:905carries a row that FAILs (not skips) when it is absent. So the tool is on the medium and the release gate covers it.Then I ran it here. Three ports report
connect, and none of them is a drive a person plugged in:A JetKVM's USB-emulation dongle (HID + audio + emulated mass storage), the internal Bluetooth radio, and a device whose vendor is the placeholder
dcba— emulation again. The capture's own summary line would readbus 4: 1 port(s) with connect, i.e. a 5 Gbps "device" on a rear port, which reads as the opposite of this issue's report and is exactly the kind of ambiguity that makes a plug test inconclusive.What I changed — PR #2044 (
9446af6d2)connected-ids.txt, written from sysfs per connected port (idVendor:idProduct, product, manufacturer, speed) and printed in theVERDICTabove the per-bus count. The plug action then has a decisive answer: the drive appears with its own ids, and everything else in the list is named for what it is.bash -nclean; run on this board writes the three lines above; the awk that feeds it resolves bus/port from a syntheticports.txt, so the section is not vacuous on a machine without a KVM.What I did not do, and why
I did not touch the D0/D3cold finding or the appliance rule: the 1.10 measurement (three of four controllers autosuspended, state software-controllable in both directions with
echo on/echo auto) is on the record above and #1567's precondition gate covers the appliance side. And I cannot take the physical action — plugging a labelled drive and reading the bundle is the owner's, and this issue should stay open until that reading exists. What it should not do is stay open and be unable to interpret the reading when it comes.bong-water-water-bong commentedon Nov 10, 2026
Slice done — the tool works and the block really is only the physical action, with one precondition now measured that the trip depends on.
1. The capture runs on this kernel and produces an interpretable verdict
bash tools/xhci_boot_capture.sh <outdir>→ rc 0, complete VERDICT, bundle written (xhci-capture-*.tar.gz):The
lsusb -vfallback works on a kernel without/sys/kernel/debug/usb/xhci/*/ports/*/portsc— that section is populated, not empty, which is the thingthat would have made the trip worthless.
And the two connected buses are not a surprise:
usb3andusb4both hang off0000:00:08.3/0000:c7:00.0(checked viareadlink -f /sys/bus/usb/devices/usbN), i.e.the one D0 controller — the front ports, including the 5 Gbps device on bus 4 port 1. Nothing is connected on the three asleep controllers, which is consistent with the
issue rather than contradicting it.
2. The core measurement reproduces on a newer kernel
rule-and-irq.txtfrom the same run:0000:c5:00.4,0000:c7:00.3,0000:c7:00.4have taken 0 interrupts since boot;0000:c7:00.0has 155,113. The issue was filedon
7.3.0-rc4-1bit-perfopt; this is7.4.0-rc1-1bit, and the state is unchanged.3. The precondition the physical step needs — and why a plug here would prove nothing
tools/xhci_keepawake_check.sh(the #1567 guard) on this plain boot:So the capture's "no drive plugged, plain Linux" state is 3 of 4 asleep. A plug into a rear USB-A or Type-C on this boot measures the absent rule, not the port. On the
1bit stick the rule ships, and the informative state is 4 of 4 in D0 — that is the column to check at the top of the VERDICT before reading the port section, and it is now
written down as the delta: asleep-3 here, D0-4 expected there, then
connector silence per port.4. The install chain is intact (so the release gate can pass)
os/mini/mkimage.sh:508→install -m 0755 "$src/tools/xhci_boot_capture.sh" "$root/bin/1bit-xhci-capture", andos/mini/release-check.sh:905gates exactly that(
xhci_capture_verdictPASS/FAIL). Nothing missing between the tool and the medium.Conclusion: nothing in code is waiting here. The remaining action is the one already described — and it now has a reference capture to compare against, plus the
precondition that makes its reading meaningful.
bong-water-water-bong commentedon Nov 10, 2026
Independent reproduction on 2026-11-10 22:0x (BIOS 1.10, kernel
7.4.0-rc1-1bit) — same shape as the issue, and the mechanism check indocs/axb35-usb-ports.mdstill holds. The only remaining item is one a lane cannot do.I went to take this issue's slice because it is the oldest open issue with no fix PR, and found that the fix is already merged (
docs/axb35-usb-ports.md,91-1bit-xhci-keepawake.rules,tools/xhci_keepawake_check.sh) — so what follows is a reproduction of the current state rather than a new fix.The mechanism, measured again on 1.10 (the doc's capture is from 11-08; this is 11-10, same result):
Also checked, and unchanged from the 1.09 report: no UCSI ACPI device (
/sys/bus/acpi/deviceshas noUSBC000/PNP0CA0), and no xHCI errors in dmesg. Reading aportscwakes the function, which is why the first scan showed D3cold and the second showed D0 — worth knowing before anyone runs a state check and concludes the controllers moved by themselves.What is actually left, and why I am not closing this: the tool's own verdict is the answer — "
N asleep— a plug into a rear USB-A or a Type-C port would fail for the reason the rule exists, so it would test nothing." This boot is a dev install without the udev rule, so:--require-awakegreen first) and a person at the machine with a device, and_DSM-vs-Windows comparison the doc names as the next diagnostic needs Windows on the same box.Neither is something a code lane can do, which is why this issue is blocked on the owner's next stick run rather than on work.
tools/xhci_boot_capture.shis already there for that run.bong-water-water-bong commentedon Nov 10, 2026
Audit done (user-readable only, no root, no device state changed) — the issue's two hypotheses can now be told apart, and one is narrowed
Headline: three of the four things that would have made this a driver bug are ruled out, one controller has a stale resume, and the board's EC
driver provably has no USB control in it.
What the box says
/sys/bus/acpi/devices/USBC/UCSI/PNP0CA0devicexhci_hcdon c5:00.4, c7:00.0, c7:00.3, c7:00.4/proc/interruptsc5:00.4powerruntime_status=resuming,power_state=D3coldactive/D00-0,1-0,domain0/1, gen=4,authorized=1, nolink_speed/rx_speed/tx_speedec_su_axb35ec_su_axb35source0x20/0x21/0x23/0x25modes,0x30/0x31/0x35…tachometers)Conclusion, per branch
xhci_hcd, MSI vectors and a BAR assignment. An "electricallysilent" port with a bound, vectored controller is a port power / retimer question, not a host-controller one.
c5:00.4has a distinct fault that the issue's summary does not mention: it is stuck inresumingwhilepower_statereadsD3cold. Acontroller whose runtime resume never completed cannot report a connect on its ports — so at least part of the symptom on that controller may be
this, and the "forcing runtime-active changes nothing" observation is consistent with a transition that is already in flight and does not
accept the write.
software fix is not a matter of loading the right module.
_DSM/_PRWpath that only the vendor'sWindows driver invokes — which is exactly the step the issue asks for and which needs root.
The root checklist, ordered, each with the output that decides it
sudo acpidump -o /tmp/acpi.dat && iasl -d /tmp/acpi.dat(oracpidump -b), then grep the disassembly for_DSMunder the six xHCI/USB4device nodes. A vendor
_DSMhere is the switcher; no_DSMmeans the switch is EC-side or not in ACPI at all.sudo dmesg | grep -iE "c5:00.4|xhci.*resume|runtime"— my finding predicts resume errors or a stalled transition onc5:00.4.Expected if hypothesis 2 holds: a timeout/failed-resume line; if absent, the stuck
resumingis a sysfs read artefact and branch 2dies.
sudo cat /sys/kernel/debug/usb/xhci/*/ports/*/portsc. AllRxDetectwith no CCS = electrical (theissue's reading); a CCS that appears with no enumeration would be a completely different, software-side fault.
sudo i2cget/EC register sweep on candidate power registers — but note the driver's map stops at0x38, so this starts from zero knowledge.This slice changes nothing on the box and needs no ticket: it is all sysfs and source. What it buys is that the root step above is now a
confirmation per branch rather than a repeat of the original investigation, and one branch (
c5:00.4stuck resuming) is new.bong-water-water-bong commentedon Nov 10, 2026
Landed the part of this issue it asks for itself, as PR #2046 — and the audit is in the comment above.
Your issue's own words: "Not a v2026.44 blocker by itself, but the release notes must say it if unfixed." Until now the limitation lived only in
docs/axb35-usb-ports.md;docs/releases.mdsaid nothing about this board's USB ports. #2046 adds the row (three of four xHCI functions autosuspend toD3cold; only the controller behind the front USB-A enumerates; the image ships the keep-awake rule; a device on a rear port is unconfirmed on a 1bit-OS boot), with the read-only checker and the rule thatall awakeis the only verdict a plug test is valid under. It also folds the audit's two new facts into the design doc.What still keeps this open, stated so nobody has to re-derive it:
tools/xhci_keepawake_check.sh --require-awake— not mine to do;_DSMcomparison — root only on this box (DSDT/SSDT*are-r-------- root, no passwordless sudo), and my comment gives the ordered commands with the expected output that confirms or kills each remaining branch.If you would rather keep #823 open only as "waiting on one plug" rather than on analysis, #2046 makes that an accurate statement of it — the release notes now carry the limitation instead of users discovering it.
bong-water-water-bong commentedon Nov 10, 2026
Claiming one distinct slice: the report's two halves are not the same fix, and the shipped rule only covers one of them.
Read the doc and the rule against today's boot (strixhalo, BIOS 1.10, kernel current, rule not installed — a dev boot, so this is the state the doc calls
"the premise still holds"):
The part that is new: the two USB4 host routers are a separate pair of PCI functions —
0000:c7:00.5(0x158d) and0000:c7:00.6(0x158e) — andthunderboltdriver is bound to both, with four live IRQs (45/46/77/79, ~1,650 events each), so the routers are not missing a driver;D3cold/runtime_status=suspendedright now;0x1587|0x1588|0x1589|0x158bis the four xHCI functions only.So the keepawake fix, as shipped, addresses the rear USB-A half of this issue and cannot address the Type-C half: those ports are behind routers the rule
never matches. The doc's own wording agrees without saying it — "both USB4 routers report link=none" is the Type-C symptom, and nothing in the fix touches
them.
What I am not claiming: that a suspended router is a broken router. A USB4 host router autosuspending with no partner attached is normal behaviour, and I have
not measured whether plugging into a Type-C port wakes it — that is the same trap this issue already corrected once for the xHCI controllers ("asleep" ≠ "dead",
the plug test decided it). So the honest statement is narrower and still useful: one plug test cannot test both halves. A plug into a rear USB-A tests the
rule; a plug into a Type-C port tests the router path, which no one has touched.
Slice I am taking: keep the board row current (the run above, dated), and extend
docs/axb35-usb-ports.mdso the two halves are separate rows with their owngate — including the router device ids and the fact that a router plugin test needs
0x158d|0x158einside the rule's match or a separate rule, so the next plugrun does not come back "Type-C still dead" and get read as "the fix failed". If the owner wants the ids added to the rule, that is a one-line change I can put
behind the same DMI gate and the same
--sysfs-roottest.xhci capture: record WHAT is on each connected port, so a plugged dri…
xhci_boot_capture: say which of engine#823's two faults the boot show…
xhci: arms for the verdict function that says it was written to be te…
bong-water-water-bong commentedon Nov 15, 2026
Functional plug test done on hardware (2026-11-14 22:09-22:12 ADT): every external port enumerates once its xHCI controller is kept in D0. This is the item
docs/axb35-usb-ports.mdlisted as outstanding.Box: strixhalo, board AXB35-02, BIOS 1.10, kernel 7.4.0-rc1-1bit (Ubuntu host, not a 1bit OS boot). Setup:
power/control=onon all four xHCI functions (+ both USB4 NHIs) at runtime, which is the same state91-1bit-xhci-keepawake.rulesproduces; the rule's vendor/device list matches this board's four controllers exactly (1022:1587/1588/1589/158b).Before the change all three formerly silent controllers had taken 0 interrupts since boot; every port read
RxDetect PPwith no CCS. So the D3cold autosuspend is the whole cause for USB 2/3 devices: no Windows-only power path is needed.Not covered yet:
link=none; these were USB 3 drives, which use the xHCI path);Note: this board's USB4 is the APU's native USB4 (router generation 4 = USB4 v1, 40 Gbps). Strix Halo boxes with 80 Gbps USB4 v2 (e.g. Minisforum MS-S1 Max / N5 Max) add a discrete controller; the AXB35-02 has none.
bong-water-water-bong commentedon Nov 15, 2026
Follow-up 22:15-22:23 ADT, same box (rule now installed in /etc/udev/rules.d on strixhalo): the Ugreen RTL9210 enclosure (0bda:9210, the device from the original report) now enumerates on every port, but only at high-speed (480 Mbps): front Type-C (both plug orientations), rear Type-C, rear USB-A c7:00.0 port 4, and rear USB-A c5:00.4 (the port that gives a generic drive 5 Gbps). The SuperSpeed portsc stays RxDetect for it while a different drive reaches 5 Gbps on the same ports, so the USB 2 limit is this enclosure or its cable, not the board. Remaining for this issue: a USB4/TB device on a Type-C port, and a 1bit OS boot with the image's rule.
KNOWN-ISSUES: #823 is D3cold autosuspend, mitigated by the shipped xH…
KNOWN-ISSUES: the BeyondMax board's dead Type-C and rear USB ports (e…