[$] SA_IMMUTABLE and the hazards of messing with signals

There are some parts of the kernel where even the most experienced and
capable developers fear to tread; one of those is surely the code that
implements signals. The nature of the signal API almost guarantees that
any implementation will be full of subtle interactions and complexities,
and the version in Linux doesn’t disappoint. So the inclusion of a
signal-handling change late in the 5.16 merge window might have been
expected to have the potential for difficulties; it didn’t disappoint
either.