NAME¶

get_hugepage_region, free_hugepage_region - Allocate and free regions of memory that use hugepages where possible

SYNOPSIS¶

#include <hugetlbfs.h>

void *get_hugepage_region(size_t len, ghr_t flags);
void free_hugepage_region(void *ptr);

DESCRIPTION¶

get_hugepage_region() allocates a memory region len bytes in size backed by hugepages. Hugepages may be of benefit to applications that use large amounts of address space and suffer a performance hit due to TLB misses. Wall-clock time or oprofile can be used to determine if there is a performance benefit from using hugepages or not.

Unlike get_huge_pages(), len does not have to be hugepage-aligned although memory may be wasted due to alignment. The caller may also specify that base pages be used in the event there are no hugepages available.

The flags argument changes the behaviour of the function. Flags may be or'd together.

GHR_FALLBACK

Use base pages if there are an insufficient number of huge pages.

GHR_STRICT

Use hugepages or return NULL.

GHR_COLOR

When specified, bytes that would be wasted due to alignment are used to color the buffer by offsetting it by a random cacheline within the hugepage. This avoids a performance problem whereby multiple buffers use the same cache lines at the same offsets. If it is not important that the start of the buffer be page-aligned, specify this flag.

GHR_DEFAULT

The library chooses a sensible combination of flags for allocating a region of memory. The current default is: GHR_FALLBACK | GHR_COLOR

free_hugepage_region() frees a region of memory allocated by get_hugepage_region(). The behaviour of the function if another pointer is used, valid or otherwise, is undefined.

RETURN VALUE¶

On success, a pointer is returned for to the allocated memory. On error, NULL is returned. errno will be set based on what the failure of mmap() was due to.

SEE ALSO¶

oprofile(1) , gethugepagesize(3) , get_huge_pages(3) , libhugetlbfs(7)

AUTHORS¶

libhugetlbfs was written by various people on the libhugetlbfs-devel mailing list.