Fading Coder

One Final Commit for the Last Sprint

Home > Tech > Content

Deep Dive into EventPipe: A Linux Kernel-to-User Space Event Notification System

Tech May 1 10

Event Handler Registration

void register_event_handler(handler_t *handler) {
    handler->category = next_category_id++;
    TAILQ_INSERT_TAIL(&global_handler_list, handler, list_node);
}

Event Processing Logic

static int dispatch_event(event_t *evt, size_t data_len) {
    handler_t *target_handler = find_event_handler(evt->category);
    return target_handler->process_func(evt, target_handler->private_data);
}

EventPipe Initialization

int initialize_event_pipe(event_pipe_t *pipe_inst, size_t per_cpu_queue_size) {
    uint32_t cpu_idx;
    int error;

    if (debug_mode) {
        pipe_inst->bpf_map_fd = 0xffff;
        return 0;
    }

    pipe_inst->cpu_count = sysconf(_SC_NPROCESSORS_ONLN);
    pipe_inst->bpf_map_fd = bpf_map_create(BPF_MAP_TYPE_PERF_EVENT_ARRAY,
                                           sizeof(uint32_t), sizeof(int), pipe_inst->cpu_count);
    pipe_inst->cpu_queues = calloc(pipe_inst->cpu_count, sizeof(*pipe_inst->cpu_queues));
    pipe_inst->poll_fds = calloc(pipe_inst->cpu_count, sizeof(*pipe_inst->poll_fds));

    for (cpu_idx = 0; cpu_idx < pipe_inst->cpu_count; cpu_idx++) {
        error = initialize_per_cpu_event_queue(pipe_inst, cpu_idx, per_cpu_queue_size);
        if (error != 0) {
            return error;
        }
    }
    return 0;
}

Per-CPU Event Queue Initialization

int initialize_per_cpu_event_queue(event_pipe_t *pipe_inst, uint32_t cpu_idx, size_t queue_size) {
    struct perf_event_attr perf_attr = {0};
    struct per_cpu_event_queue *queue = &pipe_inst->cpu_queues[cpu_idx];
    int error;

    perf_attr.type          = PERF_TYPE_SOFTWARE;
    perf_attr.config        = PERF_COUNT_SW_BPF_OUTPUT;
    perf_attr.sample_type   = PERF_SAMPLE_RAW;
    perf_attr.wakeup_events = 1;

    queue->perf_fd = perf_event_open(&perf_attr, -1, cpu_idx, -1, 0);
    error = bpf_map_update(pipe_inst->bpf_map_fd, &cpu_idx, &queue->perf_fd, BPF_ANY);
    if (error != 0) {
        return error;
    }

    const long page_size = sysconf(_SC_PAGESIZE);
    queue_size += page_size;
    queue->shared_memory = mmap(NULL, queue_size, PROT_READ | PROT_WRITE, MAP_SHARED, queue->perf_fd, 0);
    if (queue->shared_memory == MAP_FAILED) {
        return errno;
    }

    pipe_inst->poll_fds[cpu_idx].fd     = queue->perf_fd;
    pipe_inst->poll_fds[cpu_idx].events = POLLIN;
    return 0;
}

Kernel-to-User Space Communication Mechanism

Linux perf events offer a lightweight, kernel-supported channel for BPF programs to push raw data to user space. BPF_MAP_TYPE_PERF_EVENT_ARRAY serves as a per-CPU index for perf event file descriptors, ensuring efficient CPU-local event delivery. Events triggered by BPF's bpf_perf_event_output helper are written to per-CPU shared memory regions, which user space polls via POLLIN events and reads without context switching overhead.

Related Articles

Understanding Strong and Weak References in Java

Strong References Strong reference are the most prevalent type of object referencing in Java. When an object has a strong reference pointing to it, the garbage collector will not reclaim its memory. F...

Implement Image Upload Functionality for Django Integrated TinyMCE Editor

Django’s Admin panel is highly user-friendly, and pairing it with TinyMCE, an effective rich text editor, simplifies content management significantly. Combining the two is particular useful for bloggi...

Comprehensive Guide to Hive SQL Syntax and Operations

This article provides a detailed walkthrough of Hive SQL, categorizing its features and syntax for practical use. Hive SQL is segmented into the following categories: DDL Statements: Operations on...

Leave a Comment

Anonymous

◎Feel free to join the discussion and share your thoughts.