Pony-Alpha-2-Dataset-Training/skills/skill-kernel-engineering.md

Kernel Engineering Expert Skill

Activation Criteria

Activate this skill when the user:

• Requests kernel module development or loading
• Needs device driver implementation (char, block, network)
• Asks for kernel debugging techniques (ftrace, perf, eBPF)
• Requires memory management optimization
• Needs synchronization primitives (spinlocks, mutexes, RCU)
• Asks for filesystem development or VFS integration
• Requires network stack programming
• Needs interrupt handling or bottom half implementation
• Asks for kernel security mechanisms (LSM, SELinux)
• Requires kernel performance optimization
• Needs kernel-space / user-space communication
• Is working on: embedded systems, high-performance computing, custom hardware drivers, real-time systems

Core Methodology

1. Kernel Module Development

Basic Module Structure

// hello_kernel.c - Basic Kernel Module
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/init.h>

#define MODULE_NAME "hello_kernel"
#define MODULE_LICENSE "GPL"
#define MODULE_AUTHOR "Kernel Engineer"
#define MODULE_DESCRIPTION "A simple kernel module"
#define MODULE_VERSION "1.0"

// Module parameters
static int debug_level = 0;
module_param(debug_level, int, 0644);
MODULE_PARM_DESC(debug_level, "Debug level (0-3)");

static char *device_name = "kernel0";
module_param(device_name, charp, 0644);
MODULE_PARM_DESC(device_name, "Device name");

// Module initialization
static int __init hello_init(void)
{
    pr_info("%s: Module loaded\n", MODULE_NAME);
    pr_info("%s: Device name: %s, debug level: %d\n",
            MODULE_NAME, device_name, debug_level);
    return 0;
}

// Module cleanup
static void __exit hello_exit(void)
{
    pr_info("%s: Module unloaded\n", MODULE_NAME);
}

module_init(hello_init);
module_exit(hello_exit);

MODULE_INFO(intree, "Y");
MODULE_LICENSE(MODULE_LICENSE);
MODULE_AUTHOR(MODULE_AUTHOR);
MODULE_DESCRIPTION(MODULE_DESCRIPTION);
MODULE_VERSION(MODULE_VERSION);

Makefile for Kernel Modules

# Makefile for kernel module
obj-m += hello_kernel.o

# Kernel build directory
KDIR := /lib/modules/$(shell uname -r)/build

# Current directory
PWD := $(shell pwd)

# Default target
all:
	$(MAKE) -C $(KDIR) M=$(PWD) modules

# Clean build artifacts
clean:
	$(MAKE) -C $(KDIR) M=$(PWD) clean
	rm -f Module.symvers modules.order

# Load module
load:
	insmod hello_kernel.ko
	dmesg | tail -10

# Unload module
unload:
	rmmod hello_kernel
	dmesg | tail -10

# Reload module
reload: unload load

# Show module info
info:
	modinfo hello_kernel.ko

.PHONY: all clean load unload reload info

2. Character Device Driver

Complete Character Driver Implementation

// char_dev.c - Character Device Driver
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/cdev.h>
#include <linux/uaccess.h>
#include <linux/slab.h>
#include <linux/device.h>
#include <linux/mutex.h>

#define DEVICE_NAME "chardev"
#define CLASS_NAME "chardev_class"
#define BUF_LEN 1024

MODULE_LICENSE("GPL");
MODULE_AUTHOR("Kernel Engineer");
MODULE_DESCRIPTION("Character device driver");
MODULE_VERSION("1.0");

// Global variables
static int major_number;
static struct class *chardev_class = NULL;
static struct device *chardev_device = NULL;
static struct cdev chardev_cdev;

static char device_buffer[BUF_LEN];
static int buffer_pointer = 0;
static DEFINE_MUTEX(chardev_mutex);

// Function prototypes
static int chardev_open(struct inode *, struct file *);
static int chardev_release(struct inode *, struct file *);
static ssize_t chardev_read(struct file *, char __user *, size_t, loff_t *);
static ssize_t chardev_write(struct file *, const char __user *, size_t, loff_t *);
static loff_t chardev_llseek(struct file *, loff_t, int);

// File operations structure
static const struct file_operations fops = {
    .owner = THIS_MODULE,
    .open = chardev_open,
    .release = chardev_release,
    .read = chardev_read,
    .write = chardev_write,
    .llseek = chardev_llseek,
};

// Device open
static int chardev_open(struct inode *inodep, struct file *filep)
{
    if (!mutex_trylock(&chardev_mutex)) {
        pr_alert("%s: Device busy\n", DEVICE_NAME);
        return -EBUSY;
    }

    buffer_pointer = 0;
    pr_info("%s: Device opened\n", DEVICE_NAME);
    return 0;
}

// Device release
static int chardev_release(struct inode *inodep, struct file *filep)
{
    mutex_unlock(&chardev_mutex);
    pr_info("%s: Device closed\n", DEVICE_NAME);
    return 0;
}

// Device read
static ssize_t chardev_read(struct file *filep, char __user *buffer,
                           size_t len, loff_t *offset)
{
    int bytes_read = 0;

    if (*offset >= buffer_pointer) {
        return 0; // EOF
    }

    if (*offset + len > BUF_LEN) {
        len = BUF_LEN - *offset;
    }

    if (copy_to_user(buffer, device_buffer + *offset, len) != 0) {
        return -EFAULT;
    }

    *offset += len;
    bytes_read = len;

    pr_info("%s: Sent %d bytes to user\n", DEVICE_NAME, bytes_read);
    return bytes_read;
}

// Device write
static ssize_t chardev_write(struct file *filep, const char __user *buffer,
                            size_t len, loff_t *offset)
{
    if (*offset + len > BUF_LEN) {
        len = BUF_LEN - *offset;
    }

    if (copy_from_user(device_buffer + *offset, buffer, len) != 0) {
        return -EFAULT;
    }

    *offset += len;
    buffer_pointer = *offset;

    pr_info("%s: Received %zu bytes from user\n", DEVICE_NAME, len);
    return len;
}

// Device seek
static loff_t chardev_llseek(struct file *filep, loff_t offset, int orig)
{
    loff_t new_pos = 0;

    switch (orig) {
    case 0: // SEEK_SET
        new_pos = offset;
        break;
    case 1: // SEEK_CUR
        new_pos = filep->f_pos + offset;
        break;
    case 2: // SEEK_END
        new_pos = buffer_pointer + offset;
        break;
    default:
        return -EINVAL;
    }

    if (new_pos < 0 || new_pos > BUF_LEN) {
        return -EINVAL;
    }

    filep->f_pos = new_pos;
    return new_pos;
}

// Module initialization
static int __init chardev_init(void)
{
    int ret = 0;

    pr_info("%s: Initializing\n", DEVICE_NAME);

    // Allocate dynamic major number
    major_number = register_chrdev(0, DEVICE_NAME, &fops);
    if (major_number < 0) {
        pr_err("%s: Failed to register major number\n", DEVICE_NAME);
        return major_number;
    }

    // Create device class
    chardev_class = class_create(CLASS_NAME);
    if (IS_ERR(chardev_class)) {
        unregister_chrdev(major_number, DEVICE_NAME);
        pr_err("%s: Failed to register device class\n", DEVICE_NAME);
        return PTR_ERR(chardev_class);
    }

    // Create device
    chardev_device = device_create(chardev_class, NULL,
                                   MKDEV(major_number, 0), NULL, DEVICE_NAME);
    if (IS_ERR(chardev_device)) {
        class_destroy(chardev_class);
        unregister_chrdev(major_number, DEVICE_NAME);
        pr_err("%s: Failed to create device\n", DEVICE_NAME);
        return PTR_ERR(chardev_device);
    }

    mutex_init(&chardev_mutex);
    pr_info("%s: Device created with major %d\n", DEVICE_NAME, major_number);
    return 0;
}

// Module cleanup
static void __exit chardev_exit(void)
{
    device_destroy(chardev_class, MKDEV(major_number, 0));
    class_unregister(chardev_class);
    class_destroy(chardev_class);
    unregister_chrdev(major_number, DEVICE_NAME);
    mutex_destroy(&chardev_mutex);
    pr_info("%s: Exiting\n", DEVICE_NAME);
}

module_init(chardev_init);
module_exit(chardev_exit);

3. Memory Management

Kernel Memory Allocation Patterns

// memory_management.c - Kernel Memory Management
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/mm.h>
#include <linux/gfp.h>
#include <linux/highmem.h>

// Small fixed-size allocation
static void *small_buffer;

// Large allocation example
static void *large_buffer;

// DMA-capable allocation example
static void *dma_buffer;
dma_addr_t dma_handle;

struct page *pages;
void *vaddr;

// Allocation strategies
static int allocate_memory(void)
{
    // 1. Small allocations (< 128 bytes) - use kmalloc
    small_buffer = kmalloc(64, GFP_KERNEL);
    if (!small_buffer) {
        pr_err("Failed to allocate small buffer\n");
        return -ENOMEM;
    }
    pr_info("Allocated small buffer: %p\n", small_buffer);

    // 2. Medium allocations (128 bytes - 4KB) - use kmalloc
    // GFP flags:
    // - GFP_KERNEL: Normal kernel allocation, can sleep
    // - GFP_ATOMIC: Atomic allocation, cannot sleep (interrupt context)
    // - GFP_DMA: DMA-able memory (low 16MB)
    // - GFP_HIGHUSER: High memory for userspace pages
    large_buffer = kmalloc(8192, GFP_KERNEL);
    if (!large_buffer) {
        pr_err("Failed to allocate large buffer\n");
        kfree(small_buffer);
        return -ENOMEM;
    }
    pr_info("Allocated large buffer: %p\n", large_buffer);

    // 3. Very large allocations (> 128KB) - use vmalloc
    // Note: vmalloc memory is not contiguous in physical memory
    void *very_large = vmalloc(1024 * 1024); // 1 MB
    if (!very_large) {
        pr_err("Failed to allocate very large buffer\n");
        kfree(small_buffer);
        kfree(large_buffer);
        return -ENOMEM;
    }
    pr_info("Allocated very large buffer: %p\n", very_large);

    // 4. DMA-coherent allocation
    // For device DMA, must use physically contiguous memory
    dma_buffer = dma_alloc_coherent(NULL, 4096, &dma_handle, GFP_KERNEL);
    if (!dma_buffer) {
        pr_err("Failed to allocate DMA buffer\n");
        vfree(very_large);
        kfree(small_buffer);
        kfree(large_buffer);
        return -ENOMEM;
    }
    pr_info("Allocated DMA buffer: %p (phys: %pad)\n", dma_buffer, &dma_handle);

    // 5. Page-based allocation
    pages = alloc_pages(GFP_KERNEL, 2); // 2^2 = 4 pages
    if (!pages) {
        pr_err("Failed to allocate pages\n");
        dma_free_coherent(NULL, 4096, dma_buffer, dma_handle);
        vfree(very_large);
        kfree(small_buffer);
        kfree(large_buffer);
        return -ENOMEM;
    }
    vaddr = page_address(pages);
    pr_info("Allocated pages: %p\n", vaddr);

    return 0;
}

static void free_memory(void)
{
    // Free in reverse order
    if (pages) {
        __free_pages(pages, 2);
    }

    if (dma_buffer) {
        dma_free_coherent(NULL, 4096, dma_buffer, dma_handle);
    }

    if (large_buffer) {
        kfree(large_buffer);
    }

    if (small_buffer) {
        kfree(small_buffer);
    }
}

// Memory copy example
static void memory_copy_example(void)
{
    char *src, *dst;
    size_t size = 1024;

    src = kmalloc(size, GFP_KERNEL);
    dst = kmalloc(size, GFP_KERNEL);

    if (src && dst) {
        // Kernel-to-kernel copy
        memcpy(dst, src, size);

        // User-kernel copy (in read/write operations)
        // copy_from_user(to, from, n);
        // copy_to_user(to, from, n);

        // Optimized copy for large data
        // __memcpy(dst, src, size);

        kfree(src);
        kfree(dst);
    }
}

// Memory mapping example (mmap support)
static void *mmap_buffer;
static size_t mmap_size = PAGE_SIZE;

static int mmap_allocate(void)
{
    // Allocate memory that will be mapped to userspace
    mmap_buffer = kmalloc(mmap_size, GFP_KERNEL);
    if (!mmap_buffer) {
        return -ENOMEM;
    }
    memset(mmap_buffer, 0, mmap_size);
    return 0;
}

static void mmap_free(void)
{
    kfree(mmap_buffer);
}

static int chardev_mmap(struct file *filp, struct vm_area_struct *vma)
{
    unsigned long size = vma->vm_end - vma->vm_start;

    if (size > mmap_size) {
        return -EINVAL;
    }

    // Set page attributes
    vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);

    // Map memory to userspace
    if (remap_pfn_range(vma, vma->vm_start,
                        virt_to_phys(mmap_buffer) >> PAGE_SHIFT,
                        size, vma->vm_page_prot)) {
        return -EAGAIN;
    }

    return 0;
}

4. Synchronization and Concurrency

Synchronization Primitives

// synchronization.c - Kernel Synchronization
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/spinlock.h>
#include <linux/mutex.h>
#include <linux/rwlock.h>
#include <linux/semaphore.h>
#include <linux/completion.h>
#include <linux/atomic.h>
#include <linux/wait.h>

// 1. Spinlock - for short critical sections, cannot sleep
static DEFINE_SPINLOCK(my_spinlock);

void spinlock_example(void)
{
    unsigned long flags;

    // Spinlock variant: can be used in interrupt context
    spin_lock_irqsave(&my_spinlock, flags);
    // Critical section - cannot sleep, no blocking calls
    pr_info("In spinlock critical section\n");
    spin_unlock_irqrestore(&my_spinlock, flags);

    // Regular spinlock (not in interrupt context)
    spin_lock(&my_spinlock);
    // Critical section
    spin_unlock(&my_spinlock);

    // Try lock - non-blocking
    if (spin_trylock(&my_spinlock)) {
        // Got the lock
        spin_unlock(&my_spinlock);
    }
}

// 2. Mutex - for longer critical sections, can sleep
static DEFINE_MUTEX(my_mutex);

void mutex_example(void)
{
    // Lock mutex - can sleep
    mutex_lock(&my_mutex);
    // Critical section - can sleep, blocking calls allowed
    msleep(100); // OK to sleep
    mutex_unlock(&my_mutex);

    // Try lock - non-blocking
    if (mutex_trylock(&my_mutex)) {
        // Got the lock
        mutex_unlock(&my_mutex);
    }
}

// 3. Read-Write Semaphore
static DECLARE_RWSEM(rwsem);

void rwsem_example(void)
{
    // Reader lock - multiple readers allowed
    down_read(&rwsem);
    // Read critical section
    up_read(&rwsem);

    // Writer lock - exclusive access
    down_write(&rwsem);
    // Write critical section
    up_write(&rwsem);

    // Try variants
    if (down_read_trylock(&rwsem)) {
        up_read(&rwsem);
    }

    if (down_write_trylock(&rwsem)) {
        up_write(&rwsem);
    }
}

// 4. Atomic Operations - lock-free synchronization
static atomic_t counter = ATOMIC_INIT(0);

void atomic_example(void)
{
    int old, new;

    // Atomic increment
    atomic_inc(&counter);

    // Atomic decrement
    atomic_dec(&counter);

    // Atomic add
    atomic_add(5, &counter);

    // Atomic read
    old = atomic_read(&counter);

    // Atomic exchange
    new = atomic_xchg(&counter, 100);

    // Atomic compare and exchange
    old = atomic_read(&counter);
    atomic_cmpxchg(&counter, old, 200);

    // Atomic add and return old value
    old = atomic_fetch_add(&counter, 10);
}

// 5. Completion - one-to-one synchronization
static struct completion my_completion;

static int wait_thread(void *data)
{
    pr_info("Worker: Waiting for signal\n");
    wait_for_completion(&my_completion);
    pr_info("Worker: Received signal, proceeding\n");
    return 0;
}

void completion_example(void)
{
    init_completion(&my_completion);

    // Start thread that waits
    // kthread_run(wait_thread, NULL, "waiter");

    msleep(1000);

    // Signal completion
    complete(&my_completion);
}

// 6. Wait Queue - producer-consumer pattern
static DECLARE_WAIT_QUEUE_HEAD(my_wait_queue);
static int data_ready = 0;

void wait_queue_example(void)
{
    // Consumer - wait for data
    wait_event_interruptible(my_wait_queue, data_ready != 0);
    pr_info("Consumer: Data is ready\n");
    data_ready = 0;

    // Producer - signal data ready
    data_ready = 1;
    wake_up_interruptible(&my_wait_queue);
}

// 7. RCU (Read-Copy-Update) - read-mostly data structures
#include <linux/rcupdate.h>

struct rcu_data {
    int value;
    struct rcu_head rcu;
};

static struct rcu_data *global_data;

static void rcu_reclaim(struct rcu_head *rp)
{
    struct rcu_data *data = container_of(rp, struct rcu_data, rcu);
    kfree(data);
}

void rcu_example(void)
{
    struct rcu_data *new_data, *old_data;

    // Reader side - very fast, no locks
    rcu_read_lock();
    if (global_data) {
        pr_info("RCU data: %d\n", global_data->value);
    }
    rcu_read_unlock();

    // Writer side - update with grace period
    new_data = kmalloc(sizeof(*new_data), GFP_KERNEL);
    if (new_data) {
        new_data->value = 42;

        // Swap pointers atomically
        old_data = global_data;
        rcu_assign_pointer(global_data, new_data);

        // Wait for readers to finish
        call_rcu(&old_data->rcu, rcu_reclaim);
    }
}

// 8. Seqlock - for data with many readers and few writers
static DEFINE_SEQLOCK(seqlock_data);
static unsigned long seqlock_timestamp;

void seqlock_example(void)
{
    unsigned int seq;
    unsigned long timestamp;

    // Reader
    do {
        seq = read_seqbegin(&seqlock_data);
        timestamp = seqlock_timestamp;
    } while (read_seqretry(&seqlock_data, seq));

    // Writer
    write_seqlock(&seqlock_data);
    seqlock_timestamp = jiffies;
    write_sequnlock(&seqlock_data);
}

// Per-CPU variables - avoid locking entirely
#include <linux/percpu.h>

static DEFINE_PER_CPU(unsigned long, per_cpu_counter);

void percpu_example(void)
{
    unsigned long *counter;
    unsigned long sum = 0;
    int cpu;

    // Access per-CPU variable
    preempt_disable(); // Prevent CPU migration
    counter = this_cpu_ptr(&per_cpu_counter);
    (*counter)++;
    preempt_enable();

    // Sum all per-CPU counters
    for_each_possible_cpu(cpu) {
        counter = per_cpu_ptr(&per_cpu_counter, cpu);
        sum += *counter;
    }

    pr_info("Total count: %lu\n", sum);
}

5. Interrupt Handling

Interrupt Handler Implementation

// interrupt_handler.c - Interrupt Handling
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/interrupt.h>
#include <linux/gpio.h>
#include <linux/irq.h>
#include <linux/delay.h>

#define IRQ_NUMBER 42  // Example IRQ number

// Shared data between interrupt handler and process context
static atomic_t irq_count = ATOMIC_INIT(0);
static DECLARE_WAIT_QUEUE_HEAD(irq_wait_queue);

// Top-half: Interrupt Service Routine (ISR)
static irqreturn_t my_isr(int irq, void *dev_id)
{
    atomic_inc(&irq_count);
    wake_up_interruptible(&irq_wait_queue);

    pr_info("IRQ %d handled\n", irq);

    // Return IRQ_HANDLED if we handled this interrupt
    // Return IRQ_NONE if we didn't handle it
    return IRQ_HANDLED;
}

// Alternative: Threaded interrupt handler
static irqreturn_t my_isr_thread(int irq, void *dev_id)
{
    pr_info("Threaded IRQ handler running\n");
    msleep(100); // Can sleep in threaded handler

    return IRQ_HANDLED;
}

// Request interrupt
static int request_interrupt(void)
{
    int ret;

    // Request standard IRQ
    ret = request_irq(IRQ_NUMBER,
                      my_isr,
                      IRQF_TRIGGER_RISING | IRQF_SHARED,
                      "my_interrupt",
                      (void *)my_isr);
    if (ret) {
        pr_err("Failed to request IRQ %d\n", IRQ_NUMBER);
        return ret;
    }

    // Alternative: Request threaded IRQ
    // ret = request_threaded_irq(IRQ_NUMBER,
    //                           my_isr,           // Top half
    //                           my_isr_thread,    // Bottom half (thread)
    //                           IRQF_TRIGGER_RISING | IRQF_ONESHOT,
    //                           "my_interrupt",
    //                           (void *)my_isr);

    return 0;
}

// Free interrupt
static void free_interrupt(void)
{
    free_irq(IRQ_NUMBER, (void *)my_isr);
}

// Bottom half: Softirq (not commonly used directly)
static void softirq_handler(struct softirq_action *h)
{
    pr_info("Softirq handler\n");
}

// Bottom half: Tasklet
static void my_tasklet_handler(unsigned long data);
static DECLARE_TASKLET(my_tasklet, my_tasklet_handler, 0);

static void my_tasklet_handler(unsigned long data)
{
    pr_info("Tasklet handler\n");
}

// Schedule tasklet
static void schedule_tasklet(void)
{
    tasklet_schedule(&my_tasklet);
}

// Bottom half: Workqueue
#include <linux/workqueue.h>

static struct workqueue_struct *my_workqueue;
static void my_work_handler(struct work_struct *work);

static DECLARE_WORK(my_work, my_work_handler);
static DECLARE_DELAYED_WORK(my_delayed_work, my_work_handler);

static void my_work_handler(struct work_struct *work)
{
    pr_info("Workqueue handler\n");
    // Can sleep here
    msleep(100);
}

// Initialize workqueue
static int init_workqueue(void)
{
    // Create dedicated workqueue
    my_workqueue = create_singlethread_workqueue("my_workqueue");
    if (!my_workqueue) {
        return -ENOMEM;
    }

    // Queue work on system workqueue
    schedule_work(&my_work);

    // Queue delayed work
    schedule_delayed_work(&my_delayed_work, msecs_to_jiffies(1000));

    // Queue work on dedicated workqueue
    // queue_work(my_workqueue, &my_work);

    return 0;
}

// Cleanup workqueue
static void cleanup_workqueue(void)
{
    // Wait for work to complete
    cancel_work_sync(&my_work);
    cancel_delayed_work_sync(&my_delayed_work);

    // Destroy workqueue
    if (my_workqueue) {
        destroy_workqueue(my_workqueue);
    }
}

// GPIO Interrupt example
#define GPIO_PIN 17

static irqreturn_t gpio_isr(int irq, void *data)
{
    int state = gpio_get_value(GPIO_PIN);
    pr_info("GPIO %d changed to %d\n", GPIO_PIN, state);
    return IRQ_HANDLED;
}

static int setup_gpio_interrupt(void)
{
    int ret, irq;

    // Request GPIO
    ret = gpio_request(GPIO_PIN, "my_gpio");
    if (ret) {
        pr_err("Failed to request GPIO %d\n", GPIO_PIN);
        return ret;
    }

    // Set direction
    gpio_direction_input(GPIO_PIN);

    // Get IRQ number for GPIO
    irq = gpio_to_irq(GPIO_PIN);

    // Request IRQ
    ret = request_irq(irq, gpio_isr,
                     IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING,
                     "gpio_irq", NULL);
    if (ret) {
        gpio_free(GPIO_PIN);
        return ret;
    }

    return 0;
}

static void cleanup_gpio_interrupt(void)
{
    int irq = gpio_to_irq(GPIO_PIN);
    free_irq(irq, NULL);
    gpio_free(GPIO_PIN);
}

6. Kernel Debugging

Ftrace Usage

#!/bin/bash
# Kernel Tracing with Ftrace

# Enable ftrace
echo 1 > /proc/sys/kernel/ftrace_enabled

# Available tracers
cat /sys/kernel/debug/tracing/available_tracers

# Use function tracer
echo function > /sys/kernel/debug/tracing/current_tracer

# Filter functions
echo '*sched*' > /sys/kernel/debug/tracing/set_ftrace_filter

# View trace
cat /sys/kernel/debug/tracing/trace

# Clear trace
echo > /sys/kernel/debug/tracing/trace

# Function graph tracer
echo function_graph > /sys/kernel/debug/tracing/current_tracer

# Set graph depth
echo 3 > /sys/kernel/debug/tracing/max_graph_depth

# View graph trace
cat /sys/kernel/debug/tracing/trace_graph

# Trace specific function
echo do_sys_open > /sys/kernel/debug/tracing/set_ftrace_pid

Kernel Code with Tracepoints

// tracing_example.c - Kernel Tracing
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/ftrace.h>

// Tracepoint example
#define CREATE_TRACE_POINTS
#include <trace/events/sample.h>

// Trace events
static void trace_events_example(void)
{
    int value = 42;

    // Custom trace event
    trace_sample_event(value, "example data");

    pr_info("Generated trace event\n");
}

// Using trace_printk() (for debugging only)
static void trace_printk_example(void)
{
    // Only works when tracing is enabled
    trace_printk("Debug message: value=%d\n", 42);
}

// Function tracing
static noinline void traced_function(void)
{
    pr_info("This function will be traced\n");
}

// Performance counters
static void perf_counter_example(void)
{
    // Use perf events to count occurrences
    trace_printk("Performance marker\n");
}

eBPF Programs

// eBPF program example (requires bpf() syscall)
// This would be loaded via bpftool or libbpf
// Example: Socket filter eBPF program

#include <linux/bpf.h>
#include <linux/pkt_cls.h>

// BPF program for packet counting
SEC("socket")
int bpf_prog1(struct __sk_buff *skb)
{
    void *data_end = (void *)(long)skb->data_end;
    void *data = (void *)(long)skb->data;
    __u32 nh_off;

    nh_off = sizeof(struct ethhdr);

    if (data + nh_off > data_end)
        return 0;

    // Count packets
    __u64 *counter = bpf_map_lookup_elem(&my_map, &key);
    if (counter)
        __sync_fetch_and_add(counter, 1);

    return 0;
}

// Maps definition
struct bpf_map_def SEC("maps") my_map = {
    .type = BPF_MAP_TYPE_ARRAY,
    .key_size = sizeof(__u32),
    .value_size = sizeof(__u64),
    .max_entries = 256,
};

char _license[] SEC("license") = "GPL";

7. Kernel Security

Linux Security Module (LSM) Hook

// lsm_example.c - Simple LSM Hook
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/security.h>
#include <linux/binfmts.h>

// Security hook for file execution
static int my_security_bprm_check(struct linux_binprm *bprm)
{
    const char *filename = bprm->filename;

    pr_info("LSM: Executing %s\n", filename);

    // Security check logic here
    // Return 0 to allow, negative error to deny

    return 0;
}

// Security operations list
static struct security_hook_list my_hooks[] = {
    LSM_HOOK_INIT(bprm_check, my_security_bprm_check),
};

// Initialize LSM
static int __init my_lsm_init(void)
{
    pr_info("LSM: Initializing\n");
    security_add_hooks(my_hooks, ARRAY_SIZE(my_hooks), "my_lsm");
    return 0;
}

// Cleanup LSM
static void __exit my_lsm_exit(void)
{
    pr_info("LSM: Exiting\n");
}

module_init(my_lsm_init);
module_exit(my_lsm_exit);

MODULE_LICENSE("GPL");

8. Network Programming

Kernel Socket Programming

// kernel_socket.c - Kernel Network Programming
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/net.h>
#include <net/sock.h>
#include <linux/tcp.h>
#include <linux/in.h>
#include <asm/uaccess.h>
#include <linux/socket.h>

// Create socket in kernel
static int create_kernel_socket(void)
{
    struct socket *sock;
    struct sockaddr_in addr;
    int ret;

    // Create TCP socket
    ret = sock_create_kern(&init_net, AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
    if (ret < 0) {
        pr_err("Failed to create socket: %d\n", ret);
        return ret;
    }

    // Set address
    memset(&addr, 0, sizeof(addr));
    addr.sin_family = AF_INET;
    addr.sin_port = htons(8080);
    addr.sin_addr.s_addr = htonl(INADDR_ANY);

    // Bind socket
    ret = kernel_bind(sock, (struct sockaddr *)&addr, sizeof(addr));
    if (ret < 0) {
        pr_err("Failed to bind socket: %d\n", ret);
        sock_release(sock);
        return ret;
    }

    // Listen
    ret = kernel_listen(sock, 10);
    if (ret < 0) {
        pr_err("Failed to listen: %d\n", ret);
        sock_release(sock);
        return ret;
    }

    pr_info("Socket created and bound\n");
    return 0;
}

// Send data from kernel
static int send_kernel_data(struct socket *sock, const char *data, size_t len)
{
    struct msghdr msg;
    struct kvec iov;
    int ret;

    memset(&msg, 0, sizeof(msg));
    iov.iov_base = (void *)data;
    iov.iov_len = len;

    ret = kernel_sendmsg(sock, &msg, &iov, 1, len);
    if (ret < 0) {
        pr_err("Failed to send data: %d\n", ret);
        return ret;
    }

    return ret;
}

// Receive data in kernel
static int recv_kernel_data(struct socket *sock, char *data, size_t len)
{
    struct msghdr msg;
    struct kvec iov;
    int ret;

    memset(&msg, 0, sizeof(msg));
    iov.iov_base = data;
    iov.iov_len = len;

    ret = kernel_recvmsg(sock, &msg, &iov, 1, len, 0);
    if (ret < 0) {
        pr_err("Failed to receive data: %d\n", ret);
        return ret;
    }

    return ret;
}

// Netfilter hook example
#include <linux/netfilter.h>
#include <linux/netfilter_ipv4.h>
#include <linux/ip.h>

static unsigned int nf_hook_func(void *priv, struct sk_buff *skb,
                                 const struct nf_hook_state *state)
{
    struct iphdr *iph;

    if (!skb)
        return NF_ACCEPT;

    iph = ip_hdr(skb);
    pr_info("Packet: protocol=%d, saddr=%pI4, daddr=%pI4\n",
            iph->protocol, &iph->saddr, &iph->daddr);

    return NF_ACCEPT;
}

static struct nf_hook_ops nf_hook_ops = {
    .hook = nf_hook_func,
    .pf = NFPROTO_IPV4,
    .hooknum = NF_INET_PRE_ROUTING,
    .priority = NF_IP_PRI_FIRST,
};

static int __init netfilter_init(void)
{
    return nf_register_net_hook(&init_net, &nf_hook_ops);
}

static void __exit netfilter_exit(void)
{
    nf_unregister_net_hook(&init_net, &nf_hook_ops);
}

9. Decision Trees

Synchronization Primitive Selection

Critical section characteristics?
│
├─ Very short (< microsecond) → Spinlock
├─ Can sleep, exclusive access → Mutex
├─ Many readers, few writers → RCU or Seqlock
├─ Readers and writers → RW Semaphore
├─ Simple signal → Completion
└─ Producer-consumer → Wait queue

Memory Allocation Strategy

Allocation size and context?
│
├─ < 128 bytes → kmalloc with appropriate GFP flags
├─ 128 bytes - 128 KB → kmalloc
├─ > 128 KB → vmalloc (or alloc_pages if contiguous)
├─ DMA required → dma_alloc_coherent
├─ Highmem → __get_free_pages with GFP_HIGHUSER
└─ Interrupt context → GFP_ATOMIC (no sleep)

10. Anti-Patterns to Avoid

1. Sleeping in atomic context: Never use sleeping functions while holding spinlock
2. Race conditions: Always use proper synchronization
3. Memory leaks: Track and free all allocations
4. Use after free: Be careful with RCU and freeing
5. Integer overflow: Check arithmetic operations
6. Buffer overflows: Validate all user input
7. Deadlocks: Acquire locks in consistent order
8. Priority inversion: Use proper priority inheritance
9. Ignoring return values: Always check error codes
10. Missing module_put: Match get/put operations

11. Quality Checklist

Before considering kernel code production-ready:

• All error paths properly handled
• Memory allocations checked for failure
• Synchronization primitives correctly used
• No sleeping in atomic context
• No use of deprecated APIs
• Module metadata complete
• Coding style follows kernel standards
• Sparse checking passes
• Tested with lockdep
• Memory leak testing performed
• Performance testing completed
• Security review conducted
• Documentation complete
• Backward compatibility considered
• API stability maintained
• Tested on multiple architectures
• Kernel version compatibility verified
• Static analysis performed
• Stress testing completed
• Integration testing done

This comprehensive skill definition provides complete guidance for kernel engineering across Linux environments.