Re[2]: thunderbolt: AMD Phoenix USB4 host advertises Gen3 Support=0 in SB LINK_CONF with TB5 device, link stuck at Gen2

["Andrei Schapov" <andrei.schapov@gmail.com>]

Subject: Re: thunderbolt: AMD Phoenix USB4 host advertises Gen3 
Support=0 in SB LINK_CONF with TB5 device, link stuck at Gen2

Hi,

Another data point: I see the same behaviour on a different vendor’s
Hawk Point machine, with register values that match James's report
almost exactly.

Host:    Beelink SER8, AMD Ryzen 7 8745HS, BIOS V035.P8C0M0C15.16
          USB4 NHI 1022:1668, software connection manager
          No UCSI device exposed (/sys/class/typec absent)
Device:  Minisforum DEG2 (Intel JHL9480, 8086:5786), NVM 62.42,
          vendor 0x41f device 0xd002
Cable:   DEG2 bundled TB5 cable, ~0.8 m, passive
Kernel:  7.3.0 (Ubuntu 26.10 live), thunderbolt.dyndbg=+p,
          no host_reset / pcie_aspm parameters
Windows 11 on the same machine also reports "20Gbps (Gen 2, dual lane)”.

Result: link always trains TBT3 mode, 2 x Gen2:
   0-0/usb4_port2/link = tbt
   0-2: rx/tx 10.0 Gb/s, 2 lanes

Values (read from debugfs regs/sb_regs, not tbdump):

                          SER8 (mine)    UM890 Pro (James)
   Host retimer 0-0:2.1   0x7fea:0x1032  0x7fea:0x1032
   Host LANE_ADP_CS_1     0x4828003c     0x4828003c
   Host PORT_CS_18        0x00002f00     0x00002f00
   Host SB LINK_CONF 0x0c 03 d3 03       03 d3 03
   DEG2 LANE_ADP_CS_1     0x4828003e     0x4828003e
   DEG2 PORT_CS_18        0x00f02f01     0x00902f01
   DEG2 SB LINK_CONF 0x0c 03 f3 1f       03 f3 1b

The same cable and DEG2 connected to a MacBook Air M1 come up as
"Mode: USB4, Speed: 40 Gb/s", so the cable and the device can do USB4
Gen3; only the AMD host side stays at TBT3/Gen2.

The common factor between the two affected machines (different OEMs)
seems to be the host-side retimer 7fea:1032 and whatever PD controller
firmware drives it.

Possibly related: Framework issue #56 (Framework 16, Ryzen 7040)
described passive certified cables falling back to 20 Gbps because
the platform never sent SOP' Discover Mode to the cable; a 2 m active
cable worked, and a later firmware update (reportedly PD) fixed it:
https://github.com/FrameworkComputer/SoftwareFirmwareIssueTracker/issues/56

I can provide the full debugfs dump and dmesg with thunderbolt
dynamic debug, or run tbdump if that is more useful.

Thanks,
Andrei Schapov


-