Add more counters to the page queue scans.

[markj]

Summary

This makes it easier to see whether the page daemon is skipping large
numbers of pages for some reason. Add per-scan counters to the
stack-allocated scan_state structure, and export them to the
corresponding page queue after a scan.

Add a sysctl tree for each page queue in the system and keep stats under
it.

[kib]

sys/vm/vm_pageout.c

217

The counters have a large chance of wrap, I believe. This is fine, but I would suggest make them unsigned to avoid relying on -fwrapv.

[markj]

sys/vm/vm_pageout.c

217

Note that these counters are reset on each scan, so one would need to have more than 2^31 pages in a queue to hit the possibility of wraparound. I think u_long would probably make the most sense; there is a number of existing page counters that need to be widened, and I was going to use u_long there too.