Explainer
How operating systems work: kernels, processes, memory and drivers
At the centre of every operating system is the kernel, which talks to the underlying hardware.[1] Applications run in a restricted user mode and ask the kernel for services through system calls, while the kernel schedules which program runs on the processor and gives each one its own private view of memory.[2][3][4][5] Drivers and hardware abstraction layers let the same operating system run on very different devices.[6][7]
This page uses Linux, Windows and Android documentation as examples; other systems differ in detail.
What an operating system is for
Modern operating systems run many programs at once on a limited set of hardware.[8] The operating system shares those resources out and keeps programs from interfering with one another.[9][4] At its centre is the kernel, which Android’s documentation calls the central part of any operating system, the layer that talks to the underlying hardware.[1]
Linux is a widely documented example. Its kernel documentation describes it as a Unix clone with the features of a modern Unix, including true multitasking, virtual memory, shared libraries and IPv4 and IPv6 networking, distributed under the GNU General Public License v2.[8]
Two modes: user and kernel
The processor has a restricted mode for ordinary apps and a privileged mode for the operating system’s core.[10] On Windows, applications run in user mode and core operating system components run in kernel mode.[2] Each app gets its own private space, so if one crashes it does not take the others down with it.[9]
Privilege levels are defined by the processor’s instruction set architecture.[10] Microsoft’s documentation explains that a user-mode process gets a private virtual address space and handle table, isolating it from other applications.[9] All kernel-mode code shares a single virtual address space, so a faulty kernel-mode driver can corrupt operating system data, and if it crashes the whole operating system crashes.[11]
When an application needs a service from the kernel, it makes a system call.[3] The Linux manual describes the system call as the fundamental interface between an application and the kernel; programs usually go through thin wrapper functions in a library such as glibc rather than calling the kernel directly.[3]
Sharing the processor: scheduling
The kernel’s scheduler decides which runnable task gets the processor next.[4] Linux began moving from its Completely Fair Scheduler to a newer design, EEVDF (Earliest Eligible Virtual Deadline First), in kernel version 6.6.[4] Like its predecessor, EEVDF aims to give equal CPU time to runnable tasks of the same priority; it tracks whether each task has had its fair share and runs the eligible task with the earliest virtual deadline.[4] This lets latency-sensitive tasks with short time slices go first, which helps responsiveness.[12]
Sharing memory: virtual memory
Programs do not work directly with the computer’s physical memory.[5] Each one is given its own imaginary address space, and the operating system and processor quietly map those imaginary addresses to real ones.[5][13] This keeps programs apart and lets the system keep only the data that is actually needed in memory.[5]
With virtual memory, every memory access uses a virtual address, which the CPU translates into a physical address; physical memory is divided into pages, and the mappings live in hierarchical page tables.[13] The scheme abstracts physical memory, supports demand paging and provides protection and controlled sharing between processes.[5] Because walking page tables takes several slow memory accesses, CPUs cache recent translations in a Translation Lookaside Buffer (TLB).[14] The CPU’s memory management unit performs the translation in hardware.[15]
The kernel also uses spare memory to speed up storage. Linux places file data it reads in the page cache, so later reads avoid slow disk access, and written data goes into the page cache before it reaches the disk.[16] If memory does run out and the kernel cannot reclaim enough, Linux invokes the OOM (out-of-memory) killer, which chooses a task to kill so the rest of the system survives.[17]
Talking to hardware: drivers and abstraction layers
An operating system must work with many kinds of devices, each operated by a driver. The Linux kernel’s device model gives drivers a common, uniform way to describe a bus and the devices that can appear on it.[7] On Windows many drivers run in kernel mode, though some can run in user mode,[18] and a crash in a kernel-mode driver brings down the whole operating system.[11]
Android adds another layer. Its hardware abstraction layer (HAL) is a standard interface that hardware vendors implement, so Android does not depend on the details of lower-level drivers.[6] Where possible, the Android kernel is split into hardware-agnostic modules and vendor-specific ones.[19] Android apps themselves run on the Android runtime (ART), which translates app bytecode into instructions for the device’s processor.[20]
Where operating systems run today
The same ideas run from phones to data centres. In cloud computing, a hypervisor uses the processor’s hardware virtualization features to split one physical server into many virtual machines, on which customers can run their own operating systems; those customers remain responsible for patching their guest operating system.[21][22][23] See how cloud computing works.
Read next
How cloud computing works: data centres, virtual machines and SaaS
How cloud computing works: the NIST definition, IaaS, PaaS and SaaS, public and private clouds, availability zones, virtualization and who secures what.
Questions readers ask
What is a kernel?
The kernel is the central part of any operating system and talks to the underlying hardware of a device. Core operating system components run in kernel mode, while applications run in user mode.[1][2]
Why doesn't one crashing app take down the whole computer?
Each application runs as a process with its own private virtual address space, so if it crashes it does not affect other applications or the operating system. A crash in kernel-mode code, such as a driver, is different and can bring down the whole system.[9][11]
What is a system call?
The fundamental interface between an application and the kernel. Programs usually make system calls through thin wrapper functions in a library such as glibc on Linux.[3]
What is virtual memory?
A system in which every program uses virtual addresses that the processor translates into physical ones through page tables. It hides the details of physical memory, keeps only needed data in memory and protects processes from one another.[5][13]
What happens when a computer runs out of memory?
On Linux, if memory is exhausted and the kernel cannot reclaim enough, it invokes the out-of-memory (OOM) killer, which picks a task to kill so the rest of the system can keep running.[17]
Sources
· 23 claims from 7 publishers
Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.
- [1]
The kernel is the central part of any operating system and talks to the underlying hardware of a device. confirmedas of 2026-11-07
- Architecture overview · Android Open Source Project · Primary source · Architecture overview, Kernel (retrieved 2026-11-07)
- [2]
On Windows, the processor runs in two modes; applications run in user mode and core operating system components run in kernel mode. confirmedas of 2025-10-29
- User mode and kernel mode · Microsoft Learn · Primary source · User mode and kernel mode, introduction (retrieved 2026-11-07)
- [3]
The system call is the fundamental interface between an application and the Linux kernel; programs usually make system calls through thin wrapper functions in a library such as glibc. confirmedas of 2026-11-07
- syscalls(2): Linux system calls · Linux man-pages project (man7.org) · Primary source · syscalls(2), DESCRIPTION (retrieved 2026-11-07)
- syscalls(2): Linux system calls · Linux man-pages project (man7.org) · Primary source · syscalls(2), System calls and library wrapper functions (retrieved 2026-11-07)
- [4]
Linux began moving from its Completely Fair Scheduler to the EEVDF scheduler in kernel version 6.6; like CFS, EEVDF aims to share CPU time equally among runnable tasks of the same priority, uses a lag value to judge whether each task has received its fair share, and runs next the eligible task with the earliest virtual deadline. confirmedas of 2026-11-07
- EEVDF Scheduler · The Linux Kernel documentation · Primary source · EEVDF Scheduler (retrieved 2026-11-07)
- EEVDF Scheduler · The Linux Kernel documentation · Primary source · EEVDF Scheduler (retrieved 2026-11-07)
- EEVDF Scheduler · The Linux Kernel documentation · Primary source · EEVDF Scheduler (retrieved 2026-11-07)
- EEVDF Scheduler · The Linux Kernel documentation · Primary source · EEVDF Scheduler (retrieved 2026-11-07)
- [5]
Virtual memory hides the details of physical memory from applications, keeps only needed information in physical memory (demand paging), and protects and controls sharing of data between processes. confirmedas of 2026-11-07
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, Virtual Memory Primer (retrieved 2026-11-07)
- [6]
Android uses a hardware abstraction layer (HAL), a standard interface that hardware vendors implement, so that Android does not depend on lower-level driver details. confirmedas of 2026-11-07
- Architecture overview · Android Open Source Project · Primary source · Architecture overview, Hardware abstraction layer (HAL) (retrieved 2026-11-07)
- [7]
The Linux kernel's device model gives a common, uniform way to describe a bus and the devices that can appear on it. confirmedas of 2026-11-07
- The Linux Kernel Device Model · The Linux Kernel documentation · Primary source · The Linux Kernel Device Model, Overview (retrieved 2026-11-07)
- [8]
The Linux kernel documentation describes Linux as a Unix clone with the features of a modern Unix, including true multitasking, virtual memory, shared libraries and networking with IPv4 and IPv6, distributed under the GNU General Public License v2. confirmedas of 2026-11-07
- Linux kernel release 6.x · The Linux Kernel documentation · Primary source · Linux kernel README, What is Linux? (retrieved 2026-11-07)
- Linux kernel release 6.x · The Linux Kernel documentation · Primary source · Linux kernel README, What is Linux? (retrieved 2026-11-07)
- [9]
When an application starts in user mode, Windows creates a process with its own private virtual address space, so a crash in one application does not affect other applications or the operating system. confirmedas of 2025-10-29
- User mode and kernel mode · Microsoft Learn · Primary source · User mode and kernel mode, User mode (retrieved 2026-11-07)
- User mode and kernel mode · Microsoft Learn · Primary source · User mode and kernel mode, User mode (retrieved 2026-11-07)
- [10]
An ISA typically defines privilege levels, such as user mode and kernel mode, which enable secure operating system functions. confirmedas of 2026-11-07
- What is Instruction Set Architecture (ISA)? · Arm · Primary source · What are the Key ISA Components or Features? (retrieved 2026-11-07)
- [12]
EEVDF lets latency-sensitive tasks with shorter time slices be prioritized, which helps their responsiveness. confirmedas of 2026-11-07
- EEVDF Scheduler · The Linux Kernel documentation · Primary source · EEVDF Scheduler (retrieved 2026-11-07)
- [13]
With virtual memory, every memory access uses a virtual address that the CPU translates to a physical one; physical memory is divided into pages, and the mappings are described by hierarchical page tables. confirmedas of 2026-11-07
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, Virtual Memory Primer (retrieved 2026-11-07)
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, Virtual Memory Primer (retrieved 2026-11-07)
- [14]
Because address translation needs several slow memory accesses, CPUs keep a cache of recent translations called the Translation Lookaside Buffer (TLB). confirmedas of 2026-11-07
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, Huge Pages (retrieved 2026-11-07)
- [15]
The main parts of a modern CPU include a control unit that interprets instructions, an arithmetic logic unit that does arithmetic and logic, very fast registers, L1 to L3 caches near the core, a clock that synchronizes operations, and a memory management unit that translates virtual addresses to physical ones. confirmedas of 2026-11-07
- What is a CPU? · Arm · Primary source · What are the Main Components of a CPU? (retrieved 2026-11-07)
- What is a CPU? · Arm · Primary source · What are the Main Components of a CPU? (retrieved 2026-11-07)
- [16]
Linux puts file data it reads into the page cache to avoid slow disk access on later reads, and written data is placed in the page cache before eventually reaching the storage device. confirmedas of 2026-11-07
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, Page cache (retrieved 2026-11-07)
- [17]
If a loaded Linux machine runs out of memory and cannot reclaim enough, the kernel invokes the OOM killer, which selects and kills a task to free memory for the rest of the system. confirmedas of 2026-11-07
- Concepts overview (memory management) · The Linux Kernel documentation · Primary source · Memory management concepts, OOM killer (retrieved 2026-11-07)
- [18]
Microsoft's documentation notes that many device drivers run in kernel mode, although some can run in user mode. confirmedas of 2025-10-29
- User mode and kernel mode · Microsoft Learn · Primary source · User mode and kernel mode, introduction (retrieved 2026-11-07)
- [19]
Where possible, the Android Open Source Project kernel is split into hardware-agnostic modules and vendor-specific modules. confirmedas of 2026-11-07
- Architecture overview · Android Open Source Project · Primary source · Architecture overview, Kernel (retrieved 2026-11-07)
- [20]
Android apps run on the Android runtime (ART), which translates an app's bytecode into processor-specific instructions. confirmedas of 2026-11-07
- Architecture overview · Android Open Source Project · Primary source · Architecture overview, Android runtime (ART) (retrieved 2026-11-07)
- [21]
In AWS's Nitro System, the hypervisor receives virtual machine commands such as start and stop and partitions memory and CPU resources between virtual machines using the hardware virtualization features of the server processor. confirmedas of 2026-11-07
- The components of the Nitro System · Amazon Web Services · Primary source · The Security Design of the AWS Nitro System, The Nitro Hypervisor (retrieved 2026-11-07)
- [22]
NIST's three service models are software as a service (SaaS), where customers use the provider's applications; platform as a service (PaaS), where they deploy their own applications onto the provider's platform; and infrastructure as a service (IaaS), where they provision processing, storage and networks and can run arbitrary software, including operating systems. confirmedas of 2011-09-28
- The NIST Definition of Cloud Computing, SP 800-145 · National Institute of Standards and Technology · Primary source · 2011-09-28 · SP 800-145, Service Models, SaaS (retrieved 2026-11-07)
- The NIST Definition of Cloud Computing, SP 800-145 · National Institute of Standards and Technology · Primary source · 2011-09-28 · SP 800-145, Service Models, SaaS (retrieved 2026-11-07)
- The NIST Definition of Cloud Computing, SP 800-145 · National Institute of Standards and Technology · Primary source · 2011-09-28 · SP 800-145, Service Models, SaaS (retrieved 2026-11-07)
- The NIST Definition of Cloud Computing, SP 800-145 · National Institute of Standards and Technology · Primary source · 2011-09-28 · SP 800-145, Service Models, PaaS (retrieved 2026-11-07)
- The NIST Definition of Cloud Computing, SP 800-145 · National Institute of Standards and Technology · Primary source · 2011-09-28 · SP 800-145, Service Models, IaaS (retrieved 2026-11-07)
Revision history (1)
- Page created.
Created Nov 7, 2026. Last reviewed by an editor on Nov 8, 2026. Next scheduled review: May 7, 2027.
Cite this page
"How operating systems work: kernels, processes, memory and drivers." ContentLora, updated Nov 7, 2026. https://contentlora.com/explain/how-operating-systems-work
Spotted an error? Suggest a correction or emailcorrections@contentlora.com.