These are rather too large for this late in the release cycle but
they're clear, well understood and have been tested to fix a regression
which was introduced for v3.19. The details are all in Lars' changelog
and they've been cooking in -next for a while, to a large extent out
of conservatism about the size.