ACK: [SRU][R][PATCH 0/6] USB-C alt mode dropped with false firmware bug warning on Dell systems

[Rickey Castillo]

On 8/2/26 22:55, AceLan Kao via kernel-team wrote:
> From: "Chia-Lin Kao (AceLan)" <acelan.kao at canonical.com>
>
> BugLink: https://bugs.launchpad.net/bugs/2162695
>
> [Impact]
> Dell systems (CID: 202604-38600) log a firmware bug warning on every boot and
> lose one USB-C partner alternate mode:
>
>    ucsi_acpi USBC000:00: con2: Firmware bug: duplicate partner altmode SVID 0x413c
>      (VDO 0x00000001 vs 0x00000002) at offset 1, ignoring. Please update your
>      system firmware.
>
> The two altmodes are not the same. They share the Dell vendor SVID 0x413c but
> have different VDOs (0x00000001 and 0x00000002). The second one is valid and
> gets thrown away.
>
> The cause is in our own SAUCE patch set for LP: #2127960. That set added
> ucsi_altmode_is_duplicate(). For port (SOP') and plug altmodes it compares
> SVID + VDO, but for partner (SOP) altmodes it compares SVID only, so any
> second partner altmode under the same SVID is dropped.
>
> The same machines also hit this on warm reboot, about 6 out of 60 times:
>
>    ucsi_acpi USBC000:00: con2: failed to register alt modes
>
> [Fix]
> Revert the three SAUCE patches and apply the three upstream commits that
> replaced them. The upstream versions went through review and changed the
> partner matching rule.
>
> Reverts (SAUCE patches from LP: #2127960):
>    7fdb513192401 Revert "usb: typec: ucsi: Detect and skip duplicate altmodes
>                  from buggy firmware"
>    7d150034c77fc Revert "usb: typec: ucsi: Add duplicate detection to nvidia
>                  registration path"
>    fbb8fb2a11769 Revert "usb: typec: ucsi: yoga_c630: Remove redundant duplicate
>                  altmode handling"
>
> Cherry-picks, all upstream in v7.2-rc5:
>    67c92c6419ea6 usb: typec: ucsi: Detect and skip duplicate altmodes from buggy
>                  firmware
>    f1aa17f72f9b9 usb: typec: ucsi: Add duplicate detection to nvidia registration
>                  path
>    8706409bd45fb usb: typec: ucsi: yoga_c630: Remove redundant duplicate altmode
>                  handling
>
> What changed against the SAUCE version:
>
>    - Partner (SOP) altmodes are no longer matched by SVID alone. The typec class
>      binds each partner altmode to the first port altmode with the same SVID, so
>      the real limit is how many altmodes the port side advertises for that SVID.
>      The new code counts the port altmodes for the SVID and only rejects partner
>      altmodes past that count. Dell's second 0x413c altmode has a port
>      counterpart, so it now registers. The Thunderbolt firmware duplicate has
>      none, so it is still dropped and the sysfs error from LP: #2127960 stays
>      fixed.
>    - Port (CON) and plug (SOP_P) altmodes keep the SVID + VDO rule.
>    - The nvidia registration path (used by drivers with an update_altmodes()
>      callback) also runs the check now.
>    - yoga_c630 drops its own duplicate handling, since the core does it. This
>      also puts that driver back on the standard registration path.
>
> The three cherry-picks are byte for byte the same code as v7.2-rc5.
>
> Patches:
> https://lore.kernel.org/lkml/20260713084323.287516-1-acelan.kao@canonical.com/
>
> [Test Plan]
> On a Dell system that shows the warning (CID: 202604-38600):
>
> 1. Check the Dell altmodes are both there:
>
>     $ dmesg | grep -i "duplicate partner altmode"
>     $ grep -H . /sys/class/typec/port2-partner/port2-partner.*/svid
>
>     Without patch: dmesg warns about duplicate partner altmode SVID 0x413c and
>     only one 0x413c entry shows up in sysfs.
>     With patch: no warning for SVID 0x413c, and both 0x413c altmodes are listed.
>
> 2. Warm reboot 60 times and check fwts after each boot:
>
>     $ sudo reboot
>     $ sudo fwts klog oops
>
>     Without patch: "con2: failed to register alt modes" on about 6 boots out
>     of 60.
>     With patch: no "failed to register alt modes".
>
> 3. Check the original call trace has not come back. Plug in a Thunderbolt
>     device:
>
>     $ dmesg | grep -E "cannot create duplicate filename|failed to create symlinks"
>
>     Expected: no output, on both the old and the new patches.
>
> [Where problems could occur]
> Could break USB-C alternate mode registration in the ucsi driver, on any
> machine with a Type-C port.
>
> The partner rule now depends on the port altmode list being read before the
> partner list. If the port list is empty or short when the partner altmodes are
> registered, valid partner altmodes get dropped. That would show up as a missing
> mode in /sys/class/typec/portX-partner/, and DisplayPort or Thunderbolt over
> USB-C would not come up on that port.
>
> If the rule is too loose in the other direction, the old sysfs error comes
> back:
>
>    sysfs: cannot create duplicate filename '/devices/.../port0/port0.0/partner'
>    typec-thunderbolt port0-partner.1: probe with driver typec-thunderbolt failed
>    with error -17
>
> The nvidia path change also reorders the code that fills in the altmode
> descriptor. A mistake there would register altmodes with the wrong SVID or VDO
> on nvidia UCSI hardware.
>
> yoga_c630 moves back to the standard registration path. If the core check does
> not catch what its local code caught, that laptop gets duplicate altmodes
> again.
>
> Chia-Lin Kao (AceLan) (6):
>    Revert "usb: typec: ucsi: Detect and skip duplicate altmodes from
>      buggy firmware"
>    Revert "usb: typec: ucsi: Add duplicate detection to nvidia
>      registration path"
>    Revert "usb: typec: ucsi: yoga_c630: Remove redundant duplicate
>      altmode handling"
>    usb: typec: ucsi: Detect and skip duplicate altmodes from buggy
>      firmware
>    usb: typec: ucsi: Add duplicate detection to nvidia registration path
>    usb: typec: ucsi: yoga_c630: Remove redundant duplicate altmode
>      handling
>
>   drivers/usb/typec/ucsi/ucsi.c | 106 ++++++++++++++++++++++++++--------
>   1 file changed, 81 insertions(+), 25 deletions(-)
>
Signed-off-by: Rickey Castillo<rickey.castillo.valenzuela at canonical.com>