Allocating Memory

Chapter 8

Story of kmalloc

• Similar to malloc
• Allocates contiguous memory

The Flags Argument

void *kmalloc(size_t size, int flags);

• GFP_ATOMIC
  • Allocate memory outside of a process context. Never sleeps
• GFP_KERNEL
  • Normal allocation of kernel memory
• GFP_USER
  • Used to allocate memory for user-space pages. It may sleep

Memory zones

• DMA-capable memory
  • Used for DMA access
• Normal memory
  • Normal allocation
• High memory
  • Used to allow access to large amounts of memory on 32 bit platforms
• __GFP_DMA (only the dma zone is searched)
• If no flags present, DMA and and normal memory are searched
• __GFP_HIGMEM (all 3 zones are searched)

Size Argument

• Kernel memory available page sized chunks
• Kernel creates set pools of memory objects of fixed size
• Allocation is handled by returning a memory chunk from a sufficiently large pool back to the requester
• memory allocated is fixed size byte arrays and requester will likely get back more memory than asked up to twice as much

Lookaside Caches

• Special memory pool for high volume objects
• kmem_cache_t *kmem_cache_create(...)
  • creates a new cache object that can host any number of memory areas all of the same size, specified by the size argument
• void *kmem_cache_alloc(kmem_cache_t *cache, flags)
  • Allocate objects from cache object
  • cache ( cache created by kmem_cache_create )
• void kmem_cache_free(kmem_cache_t *cache, void *obj)
  • Free object memory into a cache
• kmem_cache_destroy(kmem_cache_t *cache)
  • Frees cache data structure; no longer usable

Memory pools

Form of lookaside cache in kernel where memory allocation is not allowed to fail

get_free_page and friends

Used to allocate memory in page sized chunks

• for when driver needs big chunks of memory
• void *get_zeroed_page(unsigned int flags);
  • Returns pointer to a new page and fills the page with zeros
• __get_free_page
  • Similar to get_zeroed_page, but doesn’t clear the page
• Flags argument works similarly to kmalloc
  • Usually GFP_KERNEL or GFP_ATOMIC is used
• free_page(...), free_pages(...)

vmalloc

Allocates contiguous memory in virtual address space

• Pages not consecutive. Each page retrieved by a call to alloc_page
• Address range used by vmalloc is synthetic. Sets up page tables
• void *vamalloc(unsigned long size);
• void free(void * addr);

Per-CPU Variables

• When a per-cpu variable is created, each processor gets a copy
• Used in networking subsystem. Require no locking and are fast
• DEFINE_PER_CPU(type, name)
• Dynamically allocated
  • void *alloc_percpu(type);
  • void * __alloc_percpu(size_t size, size_t align);

Obtaining Large Buffers

• Allocations of large contiguous memory buffers are prone to failure
• Best way of performing large I/O operations is through scatter/gather operations (chapter 15)

Acquiring Dedicated Buffer at Boot Time

• If you need a huge buffer of physical contiguous memory, request memory at boot time.
• Dirty technique that bypasses memory management policies
• Modules can’t allocate memory at boot time, only drivers directly linked to the kernel