Master Linux Memory Management: Virtual, Physical, and Kernel Spaces Explained

Virtual Address

Linux provides each 32‑bit process with an isolated 4 GB virtual address space. Physical pages are allocated on demand when a process accesses a virtual address, and the MMU translates the virtual address to a physical address.

Benefits

Prevents user code from accessing physical memory directly, protecting the kernel.

Allows a process to use an address space larger than the available RAM.

The 4 GB space is divided into user space (0x00000000‑0xBFFFFFFF) and kernel space (0xC0000000‑0xFFFFFFFF).

Physical Address Translation

When a page fault occurs, the kernel loads the required page and uses the MMU’s segment‑page translation to map the virtual address to a physical address.

Physical memory is organized into three zones:

ZONE_DMA : 0 MB – 16 MB, used by legacy ISA devices.

ZONE_NORMAL : 16 MB – 896 MB, directly mapped into the kernel address space.

ZONE_HIGHMEM : > 896 MB, not directly mapped; accessed via temporary or permanent mappings.

User Space Layout

User space (0x00000000‑0xBFFFFFFF) is split into five regions with distinct access attributes:

Code segment : executable instructions, read‑only.

Data segment : initialized global/static variables.

BSS segment : uninitialized globals, zero‑filled at load time.

Heap : memory allocated by malloc / free, grows upward.

Stack : local variables and call frames, grows downward.

On i386 the stack grows toward lower addresses while the heap expands toward higher addresses.

Inspect a binary’s memory layout with the size command:

[lemon ~]# size /usr/local/sbin/sshd
   text   data    bss    dec    hex  filename
1924532  12412 4268962363840 2411c0 /usr/local/sbin/sshd

Kernel Space Layout (32‑bit x86)

Kernel virtual addresses range from 0xC0000000 to 0xFFFFFFFF (1 GB).

Direct Mapping Region

The first 896 MB of kernel space is a direct‑mapping region where linear addresses map one‑to‑one with physical addresses using the offset PAGE_OFFSET = 0xC0000000. The conversion is linear = PAGE_OFFSET + physical. The kernel helper virt_to_phys() performs this translation.

High‑Memory Linear Address Space

The remaining 128 MB (0xF8000000‑0xFFFFFFFF) is reserved for high‑memory mappings, allowing the kernel to address physical memory beyond the direct‑mapping limit.

Dynamic Kernel Mapping (vmalloc)

The vmalloc API allocates virtually contiguous kernel memory whose underlying physical pages may be scattered. Allocations reside between vmalloc_start and vmalloc_end and are described by a vm_struct object.

Persistent Kernel Mapping

High‑memory pages can be accessed via alloc_page(_GFP_HIGHMEM) and then mapped with kmap().

Fixed Mapping Region

A small 4 KB window at the top of the kernel address space holds fixed mappings for special purposes (e.g., ACPI_BASE).

Memory Management Data Structures

User‑Space

Each user‑space region is represented by a vm_area_struct. A process may have multiple vm_area_struct instances linked in a doubly‑linked list and indexed in a red‑black tree for fast lookup.

Kernel‑Space Dynamic Allocation

Dynamic kernel mappings allocated by vmalloc are described by a vm_struct. Each vm_struct occupies a 4 KB guard area and is linked between vmalloc_start and vmalloc_end. Physical pages are attached on demand.

References

《Linux内核设计与实现(原书第3版)》

Linux内存管理 https://cloud.tencent.com/developer/article/1515762

linux 内存管理初探 https://cloud.tencent.com/developer/article/1005671

linux内存管理源码分析 - 页框分配器 https://www.cnblogs.com/tolimit/p/4551428.html

Linux内核--内核地址空间分布和进程地址空间 https://my.oschina.net/wuqingyi/blog/854382

Linux内存管理 http://gityuan.com/2015/10/30/kernel-memory/

Linux Used内存到底哪里去了？ http://blog.yufeng.info/archives/2456