[PATCH 0/2] thunderbolt: suppress reset-induced ring interrupt warning

[Andrei Rusu de Castro <arc@empyreal.works>]

The AMD DMA-teardown quirk resets the host interface before USB4NET
stops its service rings. The reset clears the ring interrupt bits while
the rings remain logically active. Their later interrupt-disable write
is therefore a no-op and emits a warning, even though hardware reset is
the cause.

Reordering teardown is unsafe because stopping the ring first clears its
descriptor base and unmaps frame buffers before pending path traffic has
drained.

Patch 1 tracks the host-interface reset generation and excuses only a
redundant disable by a ring that crossed an eligible reset. Genuine
software-state drift retains the warning. Patch 2 covers duplicate
enable, duplicate disable, post-reset suppression, and real register
changes.

The exact series passed all 49 Thunderbolt UML KUnit cases and three
production configuration builds with warnings fatal. The resulting
Thunderbolt module booted on two AMD USB4 systems. The affected peer-host
XDomain teardown could not be reproduced because the available attached
device is a hub rather than a peer host.

Andrei Rusu de Castro (2):
  thunderbolt: Do not warn when a reset clears ring interrupts
  thunderbolt: Test ring interrupt warning after host reset

 drivers/thunderbolt/nhi.c   | 56 ++++++++++++++++++++++++-
 drivers/thunderbolt/nhi.h   |  4 ++
 drivers/thunderbolt/test.c  | 82 +++++++++++++++++++++++++++++++++++++
 include/linux/thunderbolt.h |  9 ++++
 4 files changed, 149 insertions(+), 2 deletions(-)