> Just a part, until its zero-page-pool is full. The patch uses
> a size of 50 pages. Bad interactions with the 1nd./2nd. level
> cache could still happen though (on the Alpha and on the Sparc
> platform too). We would at least need some benchmarks to check
> that it is worth the effort.
the zeroing has to happen anyways. Rather lets do them at once, this is
more friendly to the cache subsystem as it doesnt has to swap and out in
other code as frequently.
this is just 'speculative pre-zeroing'. Not a single 'useless' zeroing
happens currently. (if you run the system for more than 5 minutes :)
-- mingo