Re: [Patch 3/3] prepopulate/cache cleared pages

[Andi Kleen]

On Friday 24 February 2006 08:01, Nick Piggin wrote:

> 
> Yeah, as I said above, the newly allocated page is fine, it is the
> page table pages I'm worried about.

page tables are easy because we zero them on free (as a side effect
of all the pte_clears)

I did a experimental hack some time ago to set a new struct page
flag when a page is known to be zeroed on freeing and use that for 
a GFP_ZERO allocation (basically skip the clear_page when that
flag was set)

The idea was to generalize the old page table reuse caches which
Ingo removed at some point.

It only works of course if the allocations and freeing
of page tables roughly matches up. In theory on could have
split the lists of the buddy allocator too into zero/non
zero pages to increase the hit rate, but I didn't attempt this.

I unfortunately don't remember the outcome, dropped it for some reason. 

-Andi

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/