如何在ld_preload中使用perf？

#define _GNU_SOURCE
#include <stdio.h>
#include <dlfcn.h>
#include <perf_event.h>

int open(const char *pathname, int flags) {
    // 初始化 perf 事件
    struct perf_event_attr pe;
    memset(&pe, 0, sizeof(pe));
    pe.type = PERF_TYPE_HARDWARE;
    pe.config = PERF_COUNT_HW_CPU_CYCLES;
    pe.size = sizeof(pe);
    pe.disabled = 1;
    pe.exclude_kernel = 1;
    pe.exclude_hv = 1;

    // 打开 perf 事件
    int fd = perf_event_open(&pe, 0, -1, -1, 0);
    if (fd == -1) {
        perror("perf_event_open");
        return -1;
    }

    // 启动 perf 事件
    perf_event_enable(fd);

    // 调用原始的 open 函数
    typeof(open) *original_open = dlsym(RTLD_NEXT, "open");
    int ret = original_open(pathname, flags);

    // 停止并读取 perf 事件
    perf_event_disable(fd);
    uint64_t count;
    read(fd, &count, sizeof(count));
    printf("open() called, cycles: %lu\n", count);

    // 关闭 perf 事件
    close(fd);

    return ret;
}

gcc -shared -fPIC -o libinterceptor.so libinterceptor.c -ldl -lperf

export LD_PRELOAD=./libinterceptor.so
./your_program