On 26.2.2025 9.02, Michal Pecio wrote:
These patches reduce latency of error reporting in some cases related
to 'error mid TD' and Missed Service events and sometimes fix a failure
to give back such TDs altogether until they are cancelled.
Also included are fixes for potential packet loss or memory corruption
due to obscure races. Whether it causes problems IRL is not known and
the worst case would be hard to reproduce, but exactly for this reason
if the worst case actually happens, it could be hard to debug too.
The first three should be safe. The fourth should also be safe, but it
relies on HC functionality Linux never relied on before, so I placed it
towards the end in case it would need some tweaks. I tested it on all
hardware I have and it worked just fine.
The last one is perhaps the most controversial, though it should be
OK with typical "complete -> resubmit" drivers. It's the only one here
which increases latency in some severe error cases. The intent is to
avoid potentially giving back URBs not yet executed by hardware.
New in v3:
- dropped the cleanup patch
- added Don't skip on Stopped - Length Invalid
New in v3:
- fixed spurious empty list warning on xrun
- clear skip flag if one skipped event was the last
Michal Pecio (5):
usb: xhci: Don't skip on Stopped - Length Invalid
usb: xhci: Complete 'error mid TD' transfers when handling Missed
Service
usb: xhci: Fix isochronous Ring Underrun/Overrun event handling
usb: xhci: Expedite skipping missed isoch TDs on modern HCs
usb: xhci: Skip only one TD on Ring Underrun/Overrun
Updated my for-usb-next branch to this v3 version
Thanks
Mathias
