Re: ucsi_acpi PPM init -ETIMEDOUT on AMD Strix Halo (Ryzen AI Max+ 395) — USB4 Type-C dead; UCSI SSDT l ooks stale (INTL 20220331)

On 7/27/26 06:37, Heikki Krogerus wrote:
> Hi,
> 
> On Thu, Jul 23, 2026 at 09:57:40AM +0200, [email protected] wrote:
>> Hi,
>>
>> On an AMD "Strix Halo" (Ryzen AI Max+ 395) mini-PC, the USB4 host side initialises correctly, but the UCSI (USB Type-C System Interface) layer never comes up, so no USB4/Type-C peripheral is ever detected under Linux (USB4 NVMe enclosures, eGPUs, TB4 docks). Standard USB 2.0/3.x on the same physical ports works, and the same USB4 dock works flawlessly on macOS — so this is isolated to the UCSI/PPM software path. I'd appreciate guidance on whether this is a known Strix Halo AGESA/PMFW issue and whether a kernel-side quirk is warranted.
>>
>> ## System under test
>> - Product: GEEKOM A9 Mega
>> - Processor: AMD Ryzen AI Max+ 395 (Radeon 8060S)
>> - BIOS: 0.13 (build 03/03/2026) EC: 0.16
>> - AGESA: StrixHaloPI-FP11 1.0.0.2 PSP BL: 0.40.0.64 SMU: 10.100.06.00 MP2: 0B.00.00.2F
>> - OS: Ubuntu 24.04 LTS Kernel: 6.18.6-061806-generic (Ubuntu mainline)
>> - Peripheral: GRAUGEAR G-M2DK-U4-40G (USB4 40 Gbps NVMe dock, controller ASMedia ASM2464PD)
>>
>> ## Symptom
>> On every plug-in of the dock to either USB4 Type-C port:
>> - No PD negotiation response; the dock is unpowered (LED dark, fan never spins).
>> - Linux registers no plug events: nothing in dmesg, no entries in /sys/class/typec/, `boltctl monitor` and `udevadm monitor` capture nothing.
>> - Standard USB 2.0 / USB 3.2 Gen2 devices on the same physical port work normally.
>> - The same dock + same cable powers up instantly on an Apple Silicon Mac and reaches full PCIe Gen4 x4 — so the dock and cable are healthy.
>>
>> ## The USB4 host controllers are fine
>> $ ls /sys/bus/thunderbolt/devices/
>> 0-0 1-0 domain0 domain1
>>
>> $ boltctl domains
>> ● domain0 security: iommu+user
>> ● domain1 security: iommu+user
>>
>> $ sudo dmesg | grep USB4
>> ACPI: USB4 _OSC: OS supports USB3+ DisplayPort+ PCIe+ XDomain+
>> ACPI: USB4 _OSC: OS controls USB3+ DisplayPort+ PCIe+ XDomain+
>>
>> Both USB4 domains register with security iommu+user; the Thunderbolt module loads. The peripheral side is never triggered because Type-C plug detection never reaches Linux → points at UCSI/PPM.
>>
>> ## UCSI fails, distinctly, in BOTH BIOS UCSI modes
>> Mode "At MMIO" (UCSI Support -> Enabled -> At MMIO): ucsi_acpi binds to USBC000:00 but the PPM reset never completes — times out after ~5.4 s:
>>
>> [ 0.003739] ACPI: SSDT 0x70320000 0014E4 (v02 AMD CPMUCSI 00000001 INTL 20220331)
>> [ 1.897156] WARNING: ucsi_reset_ppm+0x1a7/0x1c0 [typec_ucsi]
>> [ 7.327857] WARNING: ucsi_reset_ppm+0x1a7/0x1c0 [typec_ucsi]
>> [ 7.350046] ucsi_acpi USBC000:00: error -ETIMEDOUT: PPM init failed
>>
>> Reproduced across multiple cold boots and after a 60 s hard power drain (mains off, power button held) — fully deterministic, not a transient PSP soft-state.
>>
>> Mode "At EC RAM": ucsi_acpi is loaded but never binds to USBC000:00 — no init attempt, no error, no /sys/class/typec/ entries:
>>
>> $ lsmod | grep -iE "ucsi|typec"
>> ucsi_acpi 12288 0
>> typec_ucsi 69632 1 ucsi_acpi
>> typec 118784 1 typec_ucsi
>>
>> $ ls /sys/class/typec/
>> (empty)
>>
>> $ cat /sys/bus/acpi/devices/USBC000:00/modalias
>> acpi:USBC000:PNP0CA0:
>>
>> $ readlink /sys/bus/acpi/devices/USBC000:00/driver
>> (no driver bound)
>>
>> The ACPI device USBC000:00 (HID PNP0CA0) is present, but ucsi_acpi does not claim it — as if the EC-RAM operations region the SBIOS exposes doesn't match what mainline expects, or the _DSM/_CRS descriptor doesn't advertise the EC-RAM transport in a way the driver recognises.
>>
>> ## Possibly-relevant ACPI clue
>> [ 0.003718] ACPI: SSDT ... (v02 AMD CPMEC 00000001 INTL 20220331)
>> [ 0.003739] ACPI: SSDT ... (v02 AMD CPMUCSI 00000001 INTL 20220331)
>>
>> The CPMUCSI SSDT is dated 20220331 (31 March 2022) — earlier than the Strix Halo AGESA build. It looks like the UCSI ACPI methods were carried forward from an older platform reference and not refreshed for Strix Halo's PPM.
>>
>> ## Behaviour summary
>> - Disabled: no ACPI device exposed; Type-C stack absent; USB4 peripheral not detected.
>> - Enabled/At MMIO: ucsi_acpi binds, PPM reset times out (-ETIMEDOUT); Type-C absent; not detected.
>> - Enabled/At EC RAM: ucsi_acpi loaded but does not bind USBC000:00; Type-C absent; not detected.
>>
>> ## Questions
>> 1. Is a PPM-init -ETIMEDOUT on Strix Halo (AGESA StrixHaloPI-FP11 1.0.0.2) a known issue? AMD has published later StrixHaloPI-FP11 revisions (1.0.0.4, 1.1.0.0, 1.2.0.x) — do any carry UCSI/PPM fixes that would resolve the MMIO PPM timeout once an OEM integrates them?
>> 2. For the "At EC RAM" mode: what does mainline ucsi_acpi require of the EC-RAM operations region / _DSM to bind, and does this SBIOS's descriptor diverge from it? Is a kernel quirk feasible?
>> 3. Do other Strix Halo systems (Framework Desktop, HP Z2 Mini G1a, GMKtec EVO-X2, Beelink GTR9) bring up USBC000/CPMUCSI on the same kernel — i.e. is this OEM-SBIOS-specific or AGESA-wide?
>>
>> ## I can provide
>> Full `sudo dmesg` from cold boot under each BIOS setting, `acpidump`, `lspci -vvv` for both USB4 domains, and `modinfo` for ucsi_acpi/typec_ucsi. I can also test a beta/engineering BIOS on this exact unit and report back. (I have an active research collaboration with RPTU Kaiserslautern-Landau and am familiar with filing useful upstream bug reports.)
>>
>> Full field report attached (with the macOS side-by-side and all captures).
>>
>> Thanks,
>> Stefan Walcz — Walcz Strategic Consulting, Mannheim, Germany
>>
>> [Attachments beim Absenden anhängen: A9Mega_Linux_USB4_UCSI_Field_Report.pdf, Graugear G-M2DK-U4-40G datasheet]
> 
> Have you contacted AMD or whatever the manufacturer is for you product?
> 
> This looks like a firmware issue. When the UCSI driver is loaded, it
> always sends the PPM_RESET command as the very first step.
> That is an optional first step in the UCSI specification (see PPM
> Initialisation), so for example Windows may never do that. But that
> does not mean that the firmware (EC, or USB Power Delivery controller)
> can fail the PPM_RESET. PPM_RESET is not an optional command, and the
> system really needs to support it.
> 
> So please contact the manufacturer.
> 
> Br,
> 

FWIW the EC and PD aren't part of AGESA.  So contacting the manufacturer 
to fix a bug in the component that implements UCSI is the correct approach.