I'm no expert, but I believe this is a firmware problem you just have to find some acceptable workaround for. It's not a kernel bug.
The differences you have found by comparing kernels and building you own all boil down to the fact that Debian enables DMAR devices by default. That's not a regression, or a bug at all. It's just a different default which happens to expose latent bugs in drivers and/or device firmware. You should be able to boot the Debian kernels with "intel_iommu=off", achieving exactly the same result as all those other working kernels. See https://bugzilla.kernel.org/show_bug.cgi?id=215906 for further details on the specific Via firmware bug. Mathias Nyman (who is the xhci driver maintainer) suggests that a workaround might be possible in the driver, provided that the exact details of the bug can be identified. Maybe you can work with him to help find a driver workaround? Otherwise the best option is probably to disable DMAR on this system, like you've already figured out. Either i BIOS settings, or simply by booting with "intel_iommu=off". Just my .02 € as another end user with buggy firmwares all around me. Bjørn
