Linux Kernel Architecture

Chapter3:MemoryManagement

Registering Regions on IA-32

Besides callingadd_active_range, the functionzone_sizes_initstores the boundaries of the different

memory zones in terms of page frames.

arch/x86/kernel/setup_32.c

void __init zone_sizes_init(void)

{

unsigned long max_zone_pfns[MAX_NR_ZONES];

memset(max_zone_pfns, 0, sizeof(max_zone_pfns));

max_zone_pfns[ZONE_DMA] =

virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;

max_zone_pfns[ZONE_NORMAL] = max_low_pfn;

#ifdef CONFIG_HIGHMEM

max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;

add_active_range(0, 0, highend_pfn);

#else

add_active_range(0, 0, max_low_pfn);

#endif

free_area_init_nodes(max_zone_pfns);

}

MAX_DMA_ADDRESSis the highest suitable memory address for DMA operations. The constant is

declared asPAGE_OFFSET+0x1000000. Recall that the physical pages are mapped into the virtual

starting fromPAGE_OFFSET, and the first 16 MiB — hexadecimal0x1000000— are suitable for DMA

operations. Conversion withvirt_to_physyields the address in physical memory, and shifting

right byPAGE_SHIFTbits effectively divides this figure by the page size and produces the number

of pages that can be used for DMA. Unsurprisingly, the result is 4,096 since IA-32 uses pages

of 4 KiB.

max_low_pfnandhighend_pfnare global constants to specify the highest page number in the low

(usually≤896 MiB if 3 : 1 split of the address space is used) and high memory ranges that were

filled before.

Notice thatfree_area_init_nodeswill combine the information inearly_mem_mapandmax_zone_pfns:

The active ranges for each memory region are selected,and architecture-independent data structures are

constructed.

Registering Regions on AMD64

Registering the available memory is split between two functions on AMD64. The active memory regions

are registered as follows:

arch/x86/kernel/e820_64.c

e820_register_active_regions(int nid, unsigned long start_pfn,

unsigned long end_pfn)

{

unsigned long ei_startpfn;

unsigned long ei_endpfn;

int i;