I'm interested in implementing a page coloring mechanism in 2.6.1 (mostly fun). I have a few questions though.
In "UNIX(R) Systems for Modern Architectures", there is a function to calculate the number of colors, which is
num_colors = cache_size/(associativity*page_size).
However, I have trouble understanding why they did this because it implies that the number of sets/rows per color is
num_sets_per_color = page_size/linewidth
since the standard sets/rows calculation for cache is
num_rows = cachesize/(linewidth*associativity)
Doesn't this mean that many addresses within a page will possibly result in a lot of cache conflicts?
---------------------------------------------------------
Second, I think that if number_of_colors = num_rows, because thre would be less chance of cache conflicts within a page (assuming that one was trying to map a single process to different colors)?
-----------------------------------------------------------------------------------------------------------
Second, a clarification mostly, implementing this would mean I'd have to try and handle external fragmentation somehow, because this page allocator would take the place of the buddy allocator algorithm, right?
Thanks, Tim
-- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
