Nonlinear repeated malloc/free performance

[Mark Kirkwood]

I just upped kern.maxdsize to 1G, but noticed that a test program that 
mallocs and frees in a loop (increasing sized chunks 1M -> 1024M) takes 
about 6 times longer for 1024 iterations than it does only 512 of 'em. 
Is this non-linearity expected?

I see from a profile that ifree is taking most of the time.

I guess this use-case is probably unusual (it's only a test program that 
I used to check I really can use the whole 1G...).

I'm running 6.1-STABLE from about 09 May. (prog and gprof attached).

Cheers

Mark