This looks good to my eye, but best get Kirk to confirm as well.

I like Kyle's idea better than mine...

I like the idea behind this, just have one question...

generally I like the concept. If this isn't just text motion, though, I'd split it into (1) move things to new man page with as few other changes as is needed to make them work and (2) improvements, etc as a second commit.

I'm torn on this...
I'd rather have a regexp -> modules to load table that we cruise through so we can also do zfs
But at the same time, I don't want errors for people that have zfs or the p9fs modules in the kernel (I thought my playing with this required an additional module).
So this fills a need, but it's the end of the road for things like this...

@kib something like https://reviews.freebsd.org/D54751 maybe?

Yea. inline often isn't needed these days...

Tweaks from john: sizing variables and noting we can only do up to 2^16-1
queues, not the full 2^16. That's sane for the far forseeable future.

In D54714#1250118, @jhb wrote:

(I would probably put the uint16_t before the two existing bool fields, so that the fields are somewhat sorted by size, but that doesn't really matter)

Note: we could compress retries to 16 bits and have this be 16 bits too since the number of queues is limited to 64k if this busts the cache-line sizes of this struct.

This is a nice cleanup and eliminates a void * -> void ** type smearing we had with ElfAddr and should be a nop if I read right.

I'm agnostic on the NDEBUG issue, though have 'never disable assert' leanings. But just leanings, since performance is also important and I don't know the lay of the land here to have an informed opinion.