> > - not all filepages are reclaimable.
> 
> Yah, I'd also like to back off by filemin worth of pages, without
> duplicating the code that turns the %age to a page count. Suggestions
> welcome for better metrics and fine tuning.

what's "%age"?

> > - you seem to misuse uvmexp values in diff #2.
> 
> How so? It seems to give the same results and behaviour as the
> previous code (ignoring the filepages contribution) when I add and
> remove swap, and cause swapping.  I don't think interrupt races
> between taking each value are likely to be important either.

usually, swpgavail == swpages.

> > - if you take this route, eventually you need to account buffer cache,
> >   pool idle pages, etc.
> 
> Nah. I'm not trying to find every last free page - if you're
> scrounging for scraps like this, you need to add swap or reduce usage.
> I just want to find a better approximation that recognises cache pages
> as "fair game".

please don't assume that other caches are negligible.

your patch is simple because it only convers very simple cases.

YAMAMOTO Takashi