Linux Kernel Architecture

Chapter3:MemoryManagement

rmv_page_order(page);

__mod_zone_page_state(zone, NR_FREE_PAGES,

-(1UL << order));

...

expand(zone, page, order, current_order, area, migratetype);

return page;

}

}

...

Notice that the new migrate type is used byexpandif the kernel has decided to change it before. Other-

wise, the remainders are put back onto their original migrate list.

Finally, one more scenario must be considered: What if the allocation cannot be satisfied despite all page

orders and all migrate types have been taken into account? In this case, the kernel can try to fulfill the

allocation from theMIGRATE_RESERVElist, which serves as a last resort:

mm/page_alloc.c

/* Use MIGRATE_RESERVE rather than fail an allocation */

return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);

}

3.5.6 Freeing Pages

__free_pagesis the base function used to implement all functions of the kernel API. Its code flow dia-

gram is shown in Figure 3-35.

Yes

No

__free_pages

_ _free_pages_ok __free_one_page

Single page? free_hot_page

Figure 3-35: Code flow diagram for__free_pages.

__free_pagesfirst establishes whether a single page or a larger contiguous block is to be freed. If a single

page is freed, it is not returned to the buddy system but is placed in the per-CPU cache — in thewarmlist

for all pages that are highly likely to reside in the CPU cache. For this purpose, the kernel provides the

free_hot_pagehelper function, which is a parameter conversion function forfree_hot_cold_pagethat

is invoked in turn.

Iffree_hot_cold_pagedetermines that the number of pages in the per-CPU cache exceeds the limit

set bypcp->count, a whole batch of pages — whose size is specified bypcp->batch—isreturned

to the buddy system. This strategy is known aslazy coalescing.Itpreventslargenumbersofwasteful

coalescing operations that would be carried out if single pages were returned to the buddy system and

then immediately split to satisfy subsequent allocation requests. Thefree_pages_bulkfunction is used

to return pages to the buddy system.