High Memory Mappings

By definition, pages in high memory might not be permanently mapped into the kernel's address space. Thus, pages obtained via alloc_pages() with the __GFP_HIGHMEM flag might not have a logical address.

On the x86 architecture, all physical memory beyond the 896MB mark is high memory and is not permanently or automatically mapped into the kernel's address space, despite x86 processors being capable of physically addressing up to 4GB (64GB with PAE^[6]) of physical RAM. After they are allocated, these pages must be mapped into the kernel's logical address space. On x86, pages in high memory are mapped somewhere between the 3 and 4GB mark.

^[6] PAE stands for Physical Address Extension. It is a feature of x86 processors that enables them to physically address 36 bits (64GB) worth of memory, despite having only a 32-bit virtual address space.

Permanent Mappings

To map a given page structure into the kernel's address space, use

void *kmap(struct page *page)

This function works on either high or low memory. If the page structure belongs to a page in low memory, the page's virtual address is simply returned. If the page resides in high memory, a permanent mapping is created and the address is returned. The function may sleep, so kmap() works only in process context.

Because the number of permanent mappings are limited (if not, we would not be in this mess and could just permanently map all memory), high memory should be unmapped when no longer needed. This is done via

void kunmap(struct page *page)

which unmaps the given page.

Temporary Mappings

For times when a mapping must be created but the current context is unable to sleep, the kernel provides temporary mappings (which are also called atomic mappings). These are a set of reserved mappings that can hold a temporary mapping. The kernel can atomically map a high memory page into one of these reserved mappings. Consequently, a temporary mapping can be used in places that cannot sleep, such as interrupt handlers, because obtaining the mapping never blocks.

Setting up a temporary mapping is done via

void *kmap_atomic(struct page *page, enum km_type type)

The type parameter is one of the following enumerations, which describe the purpose of the temporary mapping. They are defined in <asm/kmap_types.h>:

enum km_type {
        KM_BOUNCE_READ,
        KM_SKB_SUNRPC_DATA,
        KM_SKB_DATA_SOFTIRQ,
        KM_USER0,
        KM_USER1,
        KM_BIO_SRC_IRQ,
        KM_BIO_DST_IRQ,
        KM_PTE0,
        KM_PTE1,
        KM_PTE2,
        KM_IRQ0,
        KM_IRQ1,
        KM_SOFTIRQ0,
        KM_SOFTIRQ1,
        KM_TYPE_NR
};

This function does not block and thus can be used in interrupt context and other places that cannot reschedule. It also disables kernel preemption, which is needed because the mappings are unique to each processor (and a reschedule might change which task is running on which processor).

The mapping is undone via

void kunmap_atomic(void *kvaddr, enum km_type type)

This function also does not block. In fact, in many architectures it does not do anything at all except enable kernel preemption, because a temporary mapping is valid only until the next temporary mapping. Thus, the kernel can just "forget about" the kmap_atomic() mapping, and kunmap_atomic() does not need to do anything special. The next atomic mapping then simply overwrites the previous one.