What are chunky pages?

What type of pages are reclaimed via page reclaim algorithm?

• I meant an answer for below scenario: Consider  we have lot of Page Cache pages in system. Also there are quite a good  number of CMA free pages, but very less unmovable pages. 1. Now,  since we have more Page cache pages means lowmemory killer will not  trigger and since we have more CMA pages, zone_watermarks will not be  ok. So, does linux Page reclaim algorithm reclaims page cache pages ? 2. If  lowmemory killer considers Page cache pages as free pages then does  that mean that in respect to other type of pages page cache are  reclaimed first ? 3. What sort of differentiation linux kernel does for page cache pages when it comes to page reclaim ?

• Answer:

  Page reclamation algorithm claims pages in following order: 1> Inactive pages in the page cache: These are the pages with no reference count. i.e. No application accesses them but are kept to satisfy near future reads and writes. Since this is only for performance optimization, these pages are cleaned up first. 2> Active pages: These are pages in page cache that have some reference count attached to them. They are cleaned up in NRU (Not Recently Used)  fashion. There are patches to implement ARC and CLOCK algorithms. Not sure which algorithm is used on mainline kernel. 3> Dirty pages: These are page cache entries with dirty data in them. They are flushed to backing media and released back to free pool. 4> Process text pages: Since text pages are already present on backing store, pages used (at least partially) for application text can be reclaimed without harming the system. 5> In case swapping is enabled and enough free swap space is present, data sections of the applications are swapped out to free up the pages.