Linux Kernel Development

Robert Love is an open source hacker who has used Linux since the early days. Robert is active in and passionate about both the Linux kernel and the GNOME communities. Robert currently works as Senior Kernel Engineer in the Ximian Desktop Group at Novell. Before that, he was a kernel engineer at MontaVista Software. Robert's kernel projects include the preemptive kernel, the process scheduler, the kernel events layer, VM enhancements, and multiprocessing improvements. He is the author and maintainer of schedutils and GNOME Volume Manager. Robert has given numerous talks on and has written multiple articles about the Linux kernel. He is a Contributing Editor for Linux Journal. Robert received a B.A. in Mathematics and a B.S. in Computer Science from the University of Florida. Born in South Florida, Robert currently calls Cambridge, Massachusetts home. He enjoys college football, photography, and cooking.

1. Introduction to the Linux Kernel.

    Along Came Linus: Introduction to Linux 

    Overview of Operating Systems and Kernels 

    Linux Versus Classic Unix Kernels 

    Linux Kernel Versions 

    The Linux Kernel Development Community 

    Before We Begin 

2. Getting Started with the Kernel.

    Obtaining the Kernel Source 

      Installing the Kernel Source 

      Using Patches 

    The Kernel Source Tree 

    Building the Kernel 

      Minimizing Build Noise 

      Spawning Multiple Build Jobs 

      Installing the Kernel 

    A Beast of a Different Nature 

      No libc 

      GNU C 

      No Memory Protection 

      No (Easy) Use of Floating Point 

      Small, Fixed-Size Stack 

      Synchronization and Concurrency 

      Portability Is Important 

    So Here We Are 

3. Process Management.

    Process Descriptor and the Task Structure 

      Allocating the Process Descriptor 

      Storing the Process Descriptor 

      Process State 

      Manipulating the Current Process State 

      Process Context 

      The Process Family Tree 

    Process Creation 

      Copy-on-Write 

      fork() 

      vfork() 

    The Linux Implementation of Threads 

      Kernel Threads 

    Process Termination 

      Removal of the Process Descriptor 

      The Dilemma of the Parentless Task 

    Process Wrap Up 

4. Process Scheduling.

    Policy 

      I/O-Bound Versus Processor-Bound Processes 

      Process Priority 

      Timeslice 

      Process Preemption 

      The Scheduling Policy in Action 

    The Linux Scheduling Algorithm 

      Runqueues 

      The Priority Arrays 

      Recalculating Timeslices 

      schedule() 

      Calculating Priority and Timeslice 

      Sleeping and Waking Up 

      The Load Balancer 

    Preemption and Context Switching 

      User Preemption 

      Kernel Preemption 

    Real-Time 

    Scheduler-Related System Calls 

      Scheduling Policy and Priority-Related System Calls 

      Processor Affinity System Calls 

      Yielding Processor Time 

    Scheduler Finale 

5. System Calls.

    APIs, POSIX, and the C Library 

    Syscalls 

      System Call Numbers 

      System Call Performance 

    System Call Handler 

      Denoting the Correct System Call 

      Parameter Passing 

    System Call Implementation 

Verifying the Parameters 

    System Call Context 

      Final Steps in Binding a System Call 

      Accessing the System Call from User-Space 

      Why Not to Implement a System Call 

    System Calls in Conclusion 

6. Interrupts and Interrupt Handlers.

    Interrupts 

    Interrupt Handlers 

      Top Halves Versus Bottom Halves 

    Registering an Interrupt Handler 

      Freeing an Interrupt Handler 

    Writing an Interrupt Handler 

      Shared Handlers 

      A Real-Life Interrupt Handler 

    Interrupt Context 

    Implementation of Interrupt Handling 

      /proc/interrupts 

    Interrupt Control 

      Disabling and Enabling Interrupts 

      Disabling a Specific Interrupt Line 

      Status of the Interrupt System 

    Don’t Interrupt Me; We’re Almost Done! 

7. Bottom Halves and Deferring Work.

    Bottom Halves 

      Why Bottom Halves? 

      A World of Bottom Halves 

    Softirqs 

      Implementation of Softirqs 

      Using Softirqs 

    Tasklets 

      Implementation of Tasklets 

      Using Tasklets 

      ksoftirqd 

      The Old BH Mechanism 

    Work Queues 

      Implementation of Work Queues 

      Using Work Queues 

      The Old Task Queue Mechanism 

    Which Bottom Half Should I Use? 

    Locking Between the Bottom Halves 

      Disabling Bottom Halves 

    The Bottom of Bottom-Half Processing 

8. Kernel Synchronization Introduction.

    Critical Regions and Race Conditions 

      Why Do We Need Protection? 

Locking    g 

      What Causes Concurrency, Anyway? 

      So, How Do I Know What Needs Protecting? 

    Deadlocks 

    Contention and Scalability 

    Locking and Your Code 

9. Kernel Synchronization Methods.

    Atomic Operations 

      Atomic Integer Operations 

      Atomic Bitwise Operations 

    Spin Locks 

      Other Spin Lock Methods 

      Spin Locks and Bottom Halves 

    Reader-Writer Spin Locks 

    Semaphores 

      Creating and Initializing Semaphores 

      Using Semaphores 

    Reader-Writer Semaphores 

    Spin Locks Versus Semaphores 

    Completion Variables 

    BKL: The Big Kernel Lock 

      Seq Locks 

    Preemption Disabling 

    Ordering and Barriers 

    Synchronization Summarization 

10. Timers and Time Management.

    Kernel Notion of Time 

    The Tick Rate: HZ 

      The Ideal HZ Value 

    Jiffies 

      Internal Representation of Jiffies 

      Jiffies Wraparound 

      User-Space and HZ 

    Hardware Clocks and Timers 

      Real-Time Clock 

      System Timer 

    The Timer Interrupt Handler 

    The Time of Day 

    Timers 

      Using Timers 

      Timer Race Conditions 

      The Timer Implementation 

    Delaying Execution 

      Busy Looping 

      Small Delays 

      schedule_timeout() 

    Out of Time 

11. Memory Management.

    Pages 

    Zones 

    Getting Pages 

      Getting Zeroed Pages 

      Freeing pages 

    kmalloc() 

      gfp_mask Flags 

      kfree() 

    vmalloc() 

    Slab Layer 

      Design of the Slab Layer 

    Slab Allocator Interface 

    Statically Allocating on the Stack 

      Playing Fair on the Stack 

    High Memory Mappings 

      Permanent Mappings 

      Temporary Mappings 

    Per-CPU Allocations 

    The New percpu Interface 

      Per-CPU Data at Compile-Time 

      Per-CPU Data at Runtime 

    Reasons for Using Per-CPU Data 

    Which Allocation Method Should I Use? 

12. The Virtual Filesystem.

    Common Filesystem Interface 

    Filesystem Abstraction Layer 

    Unix Filesystems 

    VFS Objects and Their Data Structures 

      Other VFS Objects 

    The Superblock Object 

      Superblock Operations 

    The Inode Object 

      Inode Operations 

    The Dentry Object 

      Dentry State 

      The Dentry Cache 

      Dentry Operations 

    The File Object 

      File Operations 

    Data Structures Associated with Filesystems 

    Data Structures Associated with a Process 

    Filesystems in Linux 

13. The Block I/O Layer.

    Anatomy of a Block Device 

    Buffers and Buffer Heads 

    The bio structure 

      The Old Versus the New 

    Request Queues 

      Requests 

    I/O Schedulers 

      The Job of an I/O Scheduler 

      The Linus Elevator 

      The Deadline I/O Scheduler 

      The Anticipatory I/O Scheduler 

      The Complete Fair Queuing I/O Scheduler 

      The Noop I/O Scheduler 

      I/O Scheduler Selection 

    Summary 

14. The Process Address Space.

    The Memory Descriptor 

      Allocating a Memory Descriptor 

      Destroying a Memory Descriptor 

      The mm_struct and Kernel Threads 

    Memory Areas 

      VMA Flags 

      VMA Operations 

      Lists and Trees of Memory Areas 

      Memory Areas in Real Life 

    Manipulating Memory Areas 

      find_vma() 

      find_vma_prev() 

      find_vma_intersection() 

    mmap() and do_mmap(): Creating an Address Interval 

      The mmap() System Call 

    munmap() and do_munmap(): Removing an Address Interval 

      The munmap() System Call 

    Page Tables 

    Conclusion 

15. The Page Cache and Page Writeback.

    Page Cache 

      The address_space Object 

    Radix Tree 

      The Old Page Hash Table 

    The Buffer Cache 

    The pdflush Daemon 

      Laptop Mode 

      bdflush and kupdated 

      Congestion Avoidance: Why We Have Multiple Threads 

    To Make a Long Story Short 

16. Modules.

      Hello, World! 

    Building Modules 

      At Home in the Source Tree 

      Living Externally 

    Installing Modules 

    Generating Module Dependencies 

    Loading Modules 

    Managing Configuration Options 

    Module Parameters 

    Exported Symbols 

    Wrapping Up Modules 

17. kobjects and sysfs.

    kobjects 

    ktypes 

    ksets 

    Subsystems 

    Structure Confusion 

    Managing and Manipulating kobjects 

    Reference Counts 

      krefs 

    sysfs 

      Adding and Removing kobjects from sysfs 

      Adding Files to sysfs 

    The Kernel Events Layer 

    kobjects and sysfs in a Nutshell 

18. Debugging.

    What You Need to Start 

    Bugs in the Kernel 

    printk() 

      The Robustness of printk() 

      Loglevels 

      The Log Buffer 

      syslogd and klogd 

      A Note About printk() and Kernel Hacking 

    Oops 

      ksymoops 

      kallsyms 

    Kernel Debugging Options 

      Atomicity Debugging 

    Asserting Bugs and Dumping Information 

    Magic SysRq Key 

    The Saga of a Kernel Debugger 

      gdb 

      kgdb 

      kdb 

    Poking and Probing the System 

      Using UID as a Conditional 

      Using Condition Variables 

      Using Statistics 

      Rate Limiting Your Debugging 

    Binary Searching to Find the Culprit Change 

    When All Else Fails: The Community 

19. Portability.

    History of Portability in Linux 

    Word Size and Data Types 

      Opaque Types 

      Special Types 

      Explicitly Sized Types 

      Signedness of Chars 

    Data Alignment 

      Avoiding Alignment Issues 

      Alignment of Nonstandard Types 

      Structure Padding 

    Byte Order 

      History of Big- and Little-Endian 

      Byte Ordering in the Kernel 

    Time 

    Page Size 

    Processor Ordering 

    SMP, Kernel Preemption, and High Memory 

      Portability Is Fun 

20. Patches, Hacking, and the Community.

    The Community 

    Linux Coding Style 

      Indention 

      Braces 

      Line Size 

      Naming 

      Functions 

      Comments 

      Typedefs 

      Using What Is Already Provided 

      No ifdefs in the Source 

      Structure Initializers 

      Fixing Code Up Ex Post Facto 

    Chain of Command 

    Submitting Bug Reports 

    Generating Patches 

    Submitting Patches 

    Conclusion 

Appendix A. Linked Lists.

    Circular Linked Lists 

      Moving Through a Linked List 

    The Linux Kernel’s Implementation 

      The Linked-List Structure 

    Manipulating Linked Lists 

    Traversing Linked Lists 

Appendix B. Kernel Random Number Generator.

    Design and Implementation 

      The Dilemma of System Startup 

    Interfaces to Input Entropy 

    Interfaces to Output Entropy 

Appendix C. Algorithmic Complexity.

    Algorithms 

    Big-O Notation 

    Big Theta Notation 

    Putting It All Together 

    Perils of Time Complexity 

Bibliography and Reading List.

    Books on Operating System Design 

    Books on Unix Kernels 

    Books on Linux Kernels 

    Books on Other Kernels 

    Books on the Unix API 

    Books on the C Programming Language 

    Other Works 

    Websites 

Index.