Section 4.6. Memory Request Path

4.6. Memory Request Path

Until now, we have approached the description of the slab allocator as though it were independent of any actual memory request. With the exception of the cache initialization functions, we have not tied together how all these functions come to be called. Now, we look at the flow of control associated with memory requests. When the kernel must obtain memory in byte-sized groupings, it uses the kmalloc() function, which eventually makes the call to kmem_getpages as follows:

 kmalloc()->__cache_alloc()->kmem_cache_grow()->kmem_getpages()

4.6.1. kmalloc()

The kmalloc() function allocates memory objects in the kernel:

 ----------------------------------------------------------------------------- mm/slab.c 2098  void * __kmalloc (size_t size, int flags) 2099  { 2100   struct cache_sizes *csizep = malloc_sizes; 2101 2102   for (; csizep->cs_size; csizep++) { 2103    if (size > csizep->cs_size) 2104     continue; ... 2112    return __cache_alloc(flags & GFP_DMA ? 2113      csizep->cs_dmacachep : csizep->cs_cachep, flags); 2114   } 2115   return NULL; 2116  } -----------------------------------------------------------------------------

4.6.1.1. size

This is the number of bytes requested.

4.6.1.2. flags

Indicates the type of memory requested. These flags are passed on to the buddy system without affecting the behavior of kmalloc().Table 4.6 shows the flags, and they are covered in detail in the "Buddy System" section.

Table 4.6. vm_area_struct->vm_flags Values
Flag	Description
`VM_READ`	Pages in this region can be read.
`VM_WRITE`	Pages in this region can be written.
`VM_EXEC`	Pages in this region can be executed.
`VM_SHARED`	Pages in this region are shared with another process.
`VM_GROWSDOWN`	The linear addresses are added onto the low side.
`VM_GROWSUP`	The linear addresses are added onto the high side.
`VM_DENYWRITE`	These pages cannot be written.
`VM_EXECUTABLE`	Pages in this region consist of executable code.
`VM_LOCKED`	Pages are locked.
`VM_DONTCOPY`	These pages cannot be cloned.
`VM_DNTEXPAND`	Do not expand this virtual memory area.

Lines 21022104

Find the first cache with objects greater than the size requested.

Lines 21122113

Allocate an object from the memory zone specified by the flags parameter.

4.6.2. kmem_cache_alloc()

This is a wrapper function around __cache_alloc(). It does not perform any additional functionality because its parameters are passed as is:

 ----------------------------------------------------------------------------- mm/slab.c 2070  void * kmem_cache_alloc (kmem_cache_t *cachep, int flags) 2071  { 2072   return __cache_alloc(cachep, flags); 2073  } -----------------------------------------------------------------------------

4.6.2.1. cachep

The cachep parameter is the cache descriptor of the cache from which we want to allocate objects.

4.6.2.2. flags

The type of memory requested. This is passed directly as indicated to kmalloc().

To free byte-sized memory allocated with kmalloc(), the kernel provides the kfree() interface, which takes as a parameter the pointer to the memory returned by kmalloc(). Figure 4.9 illustrates the flow from kfree to kmem_freepages.