如何通过perf_event_open（）确保dtlb命中率和dtlb遗漏率？

Question

我想测量缓存未命中率和dtlb未命中率。我已经完成了第一部分。但是我找不到如何设置配置来获取dtlb miss和dtlb hits。当我测量缓存未命中时，我会这样：

    pe.type = PERF_TYPE_HARDWARE;
    pe.size = sizeof(struct perf_event_attr);
    pe.config = PERF_COUNT_HW_CACHE_MISSES;

Answer 1

perf中没有'直接'PMU事件，这将有助于您测量dTLB hits。有单独的dTLB miss事件用于内存加载和存储，运行以下命令时可以看到这些事件，

sudo perf list | grep 'Hardware cache'

dTLB-load-misses                                   [Hardware cache event]
dTLB-loads                                         [Hardware cache event]
dTLB-store-misses                                  [Hardware cache event]
dTLB-stores                                        [Hardware cache event]

每个事件的含义已经提到here。它们取决于您使用的微体系结构，这对dTLB-hits的计算很重要。

例如，您要对事件dTLB-load-misses的发生进行采样，

       pe.type = PERF_TYPE_HW_CACHE;
       pe.size = sizeof(struct perf_event_attr);
       pe.config = PERF_COUNT_HW_CACHE_DTLB <<  0 | PERF_COUNT_HW_CACHE_OP_READ <<  8 | PERF_COUNT_HW_CACHE_RESULT_MISS << 16;

并且如果您要测量事件dTLB-loads的发生，

       pe.type = PERF_TYPE_HW_CACHE;
       pe.size = sizeof(struct perf_event_attr);
       pe.config = PERF_COUNT_HW_CACHE_DTLB <<  0 | PERF_COUNT_HW_CACHE_OP_READ <<  8 | PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16;

为了测量dTLB-store-misses和dTLB-stores，您需要在上述配置中将PERF_COUNT_HW_CACHE_OP_READ替换为PERF_COUNT_HW_CACHE_OP_WRITE。

[测量任何硬件缓存事件时，配置应始终采用-]的形式>

pe.config = (perf_hw_cache_id << 0) | (perf_hw_cache_op_id << 8) | (perf_hw_cache_op_result_id << 16) 
perf_hw_cache_id，perf_hw_cache_op_id和perf_hw_cache_op_result_id的含义和不同的“枚举”值在here中提及。

理想情况下，根据您的要求，您希望针对单个工作负载一起测量上述四个事件，因此，下面显示了如何一起测量dTLB-load-misses和dTLB-loads的示例-

#define _GNU_SOURCE
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/syscall.h>
#include <string.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include <linux/hw_breakpoint.h>
#include <asm/unistd.h>
#include <errno.h>
#include <stdint.h>
#include <inttypes.h>


struct read_format {
  uint64_t nr;
  struct {
    uint64_t value;
    uint64_t id;
  } values[];
};

int main(int argc, char* argv[]) {
  struct perf_event_attr pea;
  int fd1, fd2;
  uint64_t id1, id2;
  uint64_t val1, val2;
  char buf[4096];
  struct read_format* rf = (struct read_format*) buf;
  int i;

  memset(&pea, 0, sizeof(struct perf_event_attr));
  pea.type = PERF_TYPE_HW_CACHE;
  pea.size = sizeof(struct perf_event_attr);
  pea.config = PERF_COUNT_HW_CACHE_DTLB <<  0 | PERF_COUNT_HW_CACHE_OP_READ <<  8 | PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16;
  pea.disabled = 1;
  pea.exclude_kernel = 1;
  pea.exclude_hv = 1;
  pea.read_format = PERF_FORMAT_GROUP | PERF_FORMAT_ID;
  fd1 = syscall(__NR_perf_event_open, &pea, 0, -1, -1, 0);
  ioctl(fd1, PERF_EVENT_IOC_ID, &id1);

  memset(&pea, 0, sizeof(struct perf_event_attr));
  pea.type = PERF_TYPE_HW_CACHE;
  pea.size = sizeof(struct perf_event_attr);
  pea.config = PERF_COUNT_HW_CACHE_DTLB <<  0 | PERF_COUNT_HW_CACHE_OP_READ <<  8 | PERF_COUNT_HW_CACHE_RESULT_MISS << 16;;
  pea.disabled = 1;
  pea.exclude_kernel = 1;
  pea.exclude_hv = 1;
  pea.read_format = PERF_FORMAT_GROUP | PERF_FORMAT_ID;
  fd2 = syscall(__NR_perf_event_open, &pea, 0, -1, fd1 /*!!!*/, 0);
  ioctl(fd2, PERF_EVENT_IOC_ID, &id2);


  ioctl(fd1, PERF_EVENT_IOC_RESET, PERF_IOC_FLAG_GROUP);
  ioctl(fd1, PERF_EVENT_IOC_ENABLE, PERF_IOC_FLAG_GROUP);
  sleep(10);
  ioctl(fd1, PERF_EVENT_IOC_DISABLE, PERF_IOC_FLAG_GROUP);


  read(fd1, buf, sizeof(buf));
  for (i = 0; i < rf->nr; i++) {
    if (rf->values[i].id == id1) {
      val1 = rf->values[i].value;
    } else if (rf->values[i].id == id2) {
      val2 = rf->values[i].value;
    }
  }

  printf("dTLB-loads: %"PRIu64"\n", val1);
  printf("dTLB-load-misses: %"PRIu64"\n", val2);

  return 0;

提到使用perf_event_open监视多个事件时涉及的一些想法，并复制了上述程序。here。>