Build your own operating system-part04


In this article, we are going to explain how to do the Segmentation.

What is Segmentation in OS?

Segmentation in x86 means accessing the memory through segments. Segments are portions of the address space, possibly overlapping, specified by a base address and a limit. To address a byte in segmented memory you use a 48-bit logical address: 16 bits that specifies the segment and 32-bits that specifies what offset within that segment you want. The offset is added to the base address of the segment, and the resulting linear address is checked against the segment’s limit — see the figure below. If everything works out fine (including access-rights checks ignored for now) the result is a linear address. When paging is disabled, then the linear address space is mapped 1:1 onto the physical address space, and the physical memory can be accessed.

Accessing Memory

Most of the time when accessing memory there is no need to explicitly specify the segment to use. The processor has six 16-bit segment registers: cs, ss, ds, es, gs and fs. The register cs is the code segment register and specifies the segment to use when fetching instructions. The register ss is used whenever accessing the stack (through the stack pointer esp), and ds is used for other data accesses. The OS is free to use the registers es, gs and fs however it want.

Below is an example showing implicit use of the segment registers:

mov eax, [esp+4]
mov ebx, [eax]
add ebx, 8
mov [eax], ebx

The above example can be compared with the following one that makes explicit use of the segment registers:

mov eax, [ss:esp+4]
mov ebx, [ds:eax]
add ebx, 8
mov [ds:eax], ebx

You don’t need to use ss for storing the stack segment selector, or ds for the data segment selector. You could store the stack segment selector in ds and vice versa. However, in order to use the implicit style shown above, you must store the segment selectors in their indented registers.

The Global Descriptor Table (GDT)

A GDT/LDT is an array of 8-byte segment descriptors. The first descriptor in the GDT is always a null descriptor and can never be used to access memory. At least two segment descriptors (plus the null descriptor) are needed for the GDT, because the descriptor contains more information than just the base and limit fields. The two most relevant fields for us are the Type field and the Descriptor Privilege Level (DPL) field.

The DPL specifies the privilege levels required to use the segment. x86 allows for four privilege levels (PL), 0 to 3, where PL0 is the most privileged. In most operating systems (eg. Linux and Windows), only PL0 and PL3 are used. However, some operating system, such as MINIX, make use of all levels. The kernel should be able to do anything, therefore it uses segments with DPL set to 0 (also called kernel mode). The current privilege level (CPL) is determined by the segment selector in cs.

Note that the segments overlap — they both encompass the entire linear address space. In our minimal setup we’ll only use segmentation to get privilege levels.

Loading the GDT

Loading the GDT into the processor is done with the lgdt assembly code instruction, which takes the address of a struct that specifies the start and size of the GDT. It is easiest to encode this information using a “packed struct” as shown in the following example:

struct gdt {
unsigned int address;
unsigned short size;
} __attribute__((packed));

If the content of the eax register is the address to such a struct, then the GDT can be loaded with the assembly code shown below:

lgdt [eax]

It might be easier if you make this instruction available from C, the same way as was done with the assembly code instructions in and out.

After the GDT has been loaded the segment registers needs to be loaded with their corresponding segment selectors. The content of a segment selector is described in the figure and table below:

Bit:     | 15                                3 | 2  | 1 0 |
Content: | offset (index) | ti | rpl |

The offset of the segment selector is added to the start of the GDT to get the address of the segment descriptor: 0x08 for the first descriptor and 0x10 for the second, since each descriptor is 8 bytes. The Requested Privilege Level (RPL) should be 0 since the kernel of the OS should execute in privilege level 0.

Loading the segment selector registers is easy for the data registers — just copy the correct offsets to the registers:

mov ds, 0x10
mov ss, 0x10
mov es, 0x10

To load cs we have to do a “far jump”:

; code here uses the previous cs
jmp 0x08:flush_cs ; specify cs when jumping to flush_cs
; now we've changed cs to 0x08

A far jump is a jump where we explicitly specify the full 48-bit logical address: the segment selector to use and the absolute address to jump to. It will first set cs to 0x08 and then jump to flush_cs using its absolute address.


Following is the Github repository for this: from:




Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store