From reading that link, especially the Torvalds quote, it seems like all Linux kernel modules have always counted as derivative works, although Torvalds once told the AFS devs that he wasn't sure in their particular case, and he personally didn't think that theirs was a derivative work.
That seems to strongly contradict your "it plainly is not" claim.
As for GPLv2/CDDL incompatibility being "nonsense", I suggest you re-read the above. They are incompatible. The questions are:
- How much do the kernel devs care? They're the only ones in a position to complain, since as you say it's the GPL on their code being violated. The answer seems to be "not much", since CDDL is "in the spirit" of the GPL.
- How much would the legal system care if a complaint was made? As that SFLC link says, probably "not much", since it seems hard to argue that much/any physical or economic damage has been done, or any malicious intent (again, it preserves the "spirit")
- How much do Canonical care about the damage of a possible complaint? Clearly not enough to stop them going ahead.
- How much do users care about Canonical's risk-taking? Most users probably don't care.
I can imagine some 'enterprise level' (potential) customers who may run at the sight of "potential lawsuit", despite the "potential" seeming to be low, and the fallout from such a lawsuit being pretty insignificant (as SFLC says, since there's no malicious intent, it would probably just require a switch to Debian's model going forward, which would cause a little disruption but no real damage)