preempt rt对pagefault的处理

最新推荐文章于 2023-08-04 19:07:30 发布
最新推荐文章于 2023-08-04 19:07:30 发布
阅读量329 收藏
点赞数
文章标签： 操作系统 实时操作系统
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/chensong_2000/article/details/121138675
版权

在项目中我们发现了实时系统的pagefault开销比非实时要高，所以本文对pagefault做一下分析和对比。

1，pagefault调用栈（以arm64为例，内核版本为4.19.90）

el1_da (arch/arm64/kernel/entry.S) -- arm64发生data abort的异常处理入口

     do_mem_abort

          inf = esr_to_fault_info (fault_info, 异常向量表，见arch/arm64/mm/fault.c +661)

           inf->fn(也就是 do_page_fault)

                     __do_page_fault

                          find_vma //如果找不到相应的vma，说明地址不合法

                          handle_mm_fault

                                __handle_mm_fault

                                     pgd = pgd_offset(mm, address);

                                     p4d = p4d_alloc(mm, pgd, address);//获取该进程的pgd(p4d一样)

                                    vmf.pud = pud_alloc(mm, p4d, address); //获取pud，没有分配

                                    vmf.pmd = pmd_alloc(mm, vmf.pud, address);//获取pmd，没有分配

                                    handle_pte_fault（不知为什么，该函数没有出现在function_graph中）      

                                          do_anonymous_page//分配一个物理页，并把页表信息写到pte中

                                               （本例是堆栈中的pagefault，进入函数do_anonymous_page，还有可能进入  swap等函数的major pagefault）

                                                 pte_alloc(vma->vm_mm, vmf->pmd...//获取或分配pte

                                                 vmf->pte = pte_offset_map_lock(vma->vm_mm,//得到pte索引

                                                 page = alloc_zeroed_user_highpage_movable//分配一页物理内存

                                                 alloc_page_vma

                                                 entry = mk_pte(page, vma->vm_page_prot);//设置页表属性

                                                 set_pte_at(...vmf->pte, entry); //将该页表项放到pte的索引位置上

异常处理结束，然后返回用户空间，用户进程继续访问该地址，这次已经映射好，可以正常使用。

2，测试程序

这个很简单，应用程序中，申请几页内存，然后循环写入，通常在第二页就发生pagefault。

3，测试方法

trace-cmd start -e all;./mem_test

trace-cmd start -e all -p function -l "do_page_fault";./mem_test

trace-cmd start -p function_graph -g do_page_fault;./mem_test

4，通过function_graph来对比pagefault在实时和标准内核的区别

1）实时系统

mem_test-2096    2....0    83.211342: funcgraph_entry:                   |  do_page_fault() {
mem_test-2096    2....0    83.211342: funcgraph_entry:                   |    down_read_trylock() {
mem_test-2096    2....0    83.211343: funcgraph_entry:                   |      __down_read_trylock() {
mem_test-2096    2....0    83.211343: funcgraph_exit:         0.542 us   |      }
mem_test-2096    2....0    83.211344: funcgraph_exit:         1.646 us   |    }
mem_test-2096    2....0    83.211344: funcgraph_entry:                   |    find_vma() {
mem_test-2096    2....0    83.211345: funcgraph_entry:                   |      vmacache_find() {
mem_test-2096    2....0    83.211346: funcgraph_exit:         0.667 us   |      }
mem_test-2096    2....0    83.211346: funcgraph_exit:         1.792 us   |    }
mem_test-2096    2....0    83.211347: funcgraph_entry:                   |    handle_mm_fault() {
mem_test-2096    2....0    83.211347: funcgraph_entry:                   |      preempt_count_add() {
mem_test-2096    2...10    83.211348: funcgraph_exit:         0.562 us   |      }
mem_test-2096    2...10    83.211348: funcgraph_entry:                   |      preempt_count_sub() {
mem_test-2096    2....0    83.211349: funcgraph_exit:         0.521 us   |      }
mem_test-2096    2....0    83.211350: funcgraph_entry:                   |      __rcu_read_lock() {
mem_test-2096    2....0    83.211350: funcgraph_exit:         0.542 us   |      }
mem_test-2096    2....0    83.211351: funcgraph_entry:                   |      mem_cgroup_from_task() {
mem_test-2096    2....0    83.211351: funcgraph_exit:         0.542 us   |      }
mem_test-2096    2....0    83.211352: funcgraph_entry:                   |      __rcu_read_unlock() {
mem_test-2096    2....0    83.211352: funcgraph_exit:         0.521 us   |      }
mem_test-2096    2....0    83.211353: funcgraph_entry:                   |      __handle_mm_fault() {
mem_test-2096    2....0    83.211354: funcgraph_entry:                   |        do_anonymous_page() {
mem_test-2096    2....0    83.211354: funcgraph_entry:                   |          alloc_pages_vma() {
mem_test-2096    2....0    83.211355: funcgraph_entry:                   |            __get_vma_policy() {
mem_test-2096    2....0    83.211356: funcgraph_exit:         0.542 us   |            }
mem_test-2096    2....0    83.211356: funcgraph_entry:                   |            get_vma_policy.part.0() {
mem_test-2096    2....0    83.211357: funcgraph_entry:                   |              get_task_policy.part.0() {
mem_test-2096    2....0    83.211357: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....0    83.211358: funcgraph_exit:         1.646 us   |            }
mem_test-2096    2....0    83.211358: funcgraph_entry:                   |            policy_nodemask() {
mem_test-2096    2....0    83.211359: funcgraph_exit:         0.521 us   |            }
mem_test-2096    2....0    83.211359: funcgraph_entry:                   |            policy_node() {
mem_test-2096    2....0    83.211360: funcgraph_exit:         0.541 us   |            }
mem_test-2096    2....0    83.211360: funcgraph_entry:                   |            __alloc_pages_nodemask() {
mem_test-2096    2....0    83.211361: funcgraph_entry:                   |              get_page_from_freelist() {
mem_test-2096    2....0    83.211362: funcgraph_entry:                   |                rmqueue_pcplist.isra.0() {
mem_test-2096    2....0    83.211362: funcgraph_entry:                   |                  migrate_disable() {
mem_test-2096    2....0    83.211363: funcgraph_entry:                   |                    preempt_count_add() {
mem_test-2096    2...10    83.211363: funcgraph_exit:         0.541 us   |                    }
mem_test-2096    2...10    83.211364: funcgraph_entry:                   |                    pin_current_cpu() {
mem_test-2096    2...10    83.211364: funcgraph_entry:                   |                      __read_rt_trylock() {
mem_test-2096    2...10    83.211365: funcgraph_exit:         0.541 us   |                      }
mem_test-2096    2...10    83.211366: funcgraph_exit:         1.625 us   |                    }
mem_test-2096    2...10    83.211366: funcgraph_entry:                   |                    task_rq_lock() {
mem_test-2096    2...10    83.211367: funcgraph_entry:                   |                      _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211367: funcgraph_entry:                   |                        preempt_count_add() {
mem_test-2096    2d..20    83.211368: funcgraph_exit:         0.500 us   |                        }
mem_test-2096    2d..20    83.211368: funcgraph_exit:         1.584 us   |                      }
mem_test-2096    2d..20    83.211369: funcgraph_entry:                   |                      _raw_spin_lock() {
mem_test-2096    2d..20    83.211369: funcgraph_entry:                   |                        preempt_count_add() {
mem_test-2096    2d..30    83.211370: funcgraph_exit:         0.500 us   |                        }
mem_test-2096    2d..30    83.211370: funcgraph_exit:         1.541 us   |                      }
mem_test-2096    2d..30    83.211371: funcgraph_exit:         4.666 us   |                    }
mem_test-2096    2d..30    83.211371: funcgraph_entry:                   |                    preempt_count_sub() {
mem_test-2096    2d..20    83.211372: funcgraph_exit:         0.542 us   |                    }
mem_test-2096    2d..20    83.211372: funcgraph_entry:                   |                    _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211373: funcgraph_entry:                   |                      preempt_count_sub() {
mem_test-2096    2...10    83.211373: funcgraph_exit:         0.521 us   |                      }
mem_test-2096    2...10    83.211374: funcgraph_exit:         1.541 us   |                    }
mem_test-2096    2...11    83.211374: funcgraph_entry:                   |                    preempt_count_sub() {
mem_test-2096    2....1    83.211375: funcgraph_exit:         0.542 us   |                    }
mem_test-2096    2....1    83.211375: funcgraph_exit:       + 13.250 us  |                  }
mem_test-2096    2....1    83.211376: funcgraph_entry:                   |                  rt_spin_lock() {
mem_test-2096    2....1    83.211377: funcgraph_entry:                   |                    migrate_disable() {
mem_test-2096    2....2    83.211377: funcgraph_exit:         0.542 us   |                    }
mem_test-2096    2....2    83.211378: funcgraph_exit:         1.646 us   |                  }
mem_test-2096    2....2    83.211378: funcgraph_entry:                   |                  preempt_count_add() {
mem_test-2096    2...12    83.211379: funcgraph_exit:         0.563 us   |                  }
mem_test-2096    2...12    83.211379: funcgraph_entry:                   |                  preempt_count_sub() {
mem_test-2096    2....2    83.211380: funcgraph_exit:         0.520 us   |                  }
mem_test-2096    2....2    83.211381: funcgraph_entry:                   |                  __inc_numa_state() {
mem_test-2096    2....2    83.211381: funcgraph_exit:         0.583 us   |                  }
mem_test-2096    2....2    83.211382: funcgraph_entry:                   |                  __inc_numa_state() {
mem_test-2096    2....2    83.211382: funcgraph_exit:         0.562 us   |                  }
mem_test-2096    2....2    83.211383: funcgraph_entry:                   |                  rt_spin_unlock() {
mem_test-2096    2....2    83.211383: funcgraph_entry:                   |                    migrate_enable() {
mem_test-2096    2....1    83.211384: funcgraph_exit:         0.562 us   |                    }
mem_test-2096    2....1    83.211384: funcgraph_exit:         1.666 us   |                  }
mem_test-2096    2....1    83.211385: funcgraph_entry:                   |                  migrate_enable() {
mem_test-2096    2....1    83.211386: funcgraph_entry:                   |                    preempt_count_add() {
mem_test-2096    2...11    83.211386: funcgraph_exit:         0.625 us   |                    }
mem_test-2096    2...10    83.211387: funcgraph_entry:                   |                    task_rq_lock() {
mem_test-2096    2...10    83.211387: funcgraph_entry:                   |                      _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211388: funcgraph_entry:                   |                        preempt_count_add() {
mem_test-2096    2d..20    83.211388: funcgraph_exit:         0.500 us   |                        }
mem_test-2096    2d..20    83.211389: funcgraph_exit:         1.562 us   |                      }
mem_test-2096    2d..20    83.211389: funcgraph_entry:                   |                      _raw_spin_lock() {
mem_test-2096    2d..20    83.211390: funcgraph_entry:                   |                        preempt_count_add() {
mem_test-2096    2d..30    83.211390: funcgraph_exit:         0.500 us   |                        }
mem_test-2096    2d..30    83.211391: funcgraph_exit:         1.542 us   |                      }
mem_test-2096    2d..30    83.211391: funcgraph_exit:         4.646 us   |                    }
mem_test-2096    2d..30    83.211392: funcgraph_entry:                   |                    preempt_count_sub() {
mem_test-2096    2d..20    83.211393: funcgraph_exit:         0.500 us   |                    }
mem_test-2096    2d..20    83.211393: funcgraph_entry:                   |                    _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211394: funcgraph_entry:                   |                      preempt_count_sub() {
mem_test-2096    2...10    83.211394: funcgraph_exit:         0.500 us   |                      }
mem_test-2096    2...10    83.211395: funcgraph_exit:         1.562 us   |                    }
mem_test-2096    2...10    83.211395: funcgraph_entry:                   |                    unpin_current_cpu() {
mem_test-2096    2...10    83.211396: funcgraph_entry:                   |                      __read_rt_unlock() {
mem_test-2096    2...10    83.211396: funcgraph_exit:         0.521 us   |                      }
mem_test-2096    2...10    83.211397: funcgraph_exit:         1.604 us   |                    }
mem_test-2096    2...10    83.211397: funcgraph_entry:                   |                    preempt_count_sub() {
mem_test-2096    2....0    83.211398: funcgraph_exit:         0.500 us   |                    }
mem_test-2096    2....0    83.211398: funcgraph_exit:       + 13.312 us  |                  }
mem_test-2096    2....0    83.211399: funcgraph_exit:       + 37.229 us  |                }
mem_test-2096    2....0    83.211399: funcgraph_exit:       + 38.333 us  |              }
mem_test-2096    2....0    83.211400: mm_page_alloc:        page=0xffa09b71 pfn=2139273 order=0 migratetype=1 gfp_flags=GFP_HIGHUSER_MOVABLE
mem_test-2096    2....0    83.211400: funcgraph_exit:       + 39.979 us  |            }
mem_test-2096    2....0    83.211401: funcgraph_exit:       + 46.584 us  |          }
mem_test-2096    2....0    83.211402: funcgraph_entry:                   |          migrate_disable() {
mem_test-2096    2....0    83.211402: funcgraph_entry:                   |            preempt_count_add() {
mem_test-2096    2...10    83.211403: funcgraph_exit:         0.521 us   |            }
mem_test-2096    2...10    83.211403: funcgraph_entry:                   |            pin_current_cpu() {
mem_test-2096    2...10    83.211404: funcgraph_entry:                   |              __read_rt_trylock() {
mem_test-2096    2...10    83.211404: funcgraph_exit:         0.562 us   |              }
mem_test-2096    2...10    83.211405: funcgraph_exit:         1.625 us   |            }
mem_test-2096    2...10    83.211405: funcgraph_entry:                   |            task_rq_lock() {
mem_test-2096    2...10    83.211406: funcgraph_entry:                   |              _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211406: funcgraph_entry:                   |                preempt_count_add() {
mem_test-2096    2d..20    83.211407: funcgraph_exit:         0.479 us   |                }
mem_test-2096    2d..20    83.211407: funcgraph_exit:         1.563 us   |              }
mem_test-2096    2d..20    83.211408: funcgraph_entry:                   |              _raw_spin_lock() {
mem_test-2096    2d..20    83.211409: funcgraph_entry:                   |                preempt_count_add() {
mem_test-2096    2d..30    83.211409: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2d..30    83.211409: funcgraph_exit:         1.541 us   |              }
mem_test-2096    2d..30    83.211410: funcgraph_exit:         4.666 us   |            }
mem_test-2096    2d..30    83.211410: funcgraph_entry:                   |            preempt_count_sub() {
mem_test-2096    2d..20    83.211411: funcgraph_exit:         0.500 us   |            }
mem_test-2096    2d..20    83.211412: funcgraph_entry:                   |            _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211412: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2...10    83.211413: funcgraph_exit:         0.521 us   |              }
mem_test-2096    2...10    83.211413: funcgraph_exit:         1.563 us   |            }
mem_test-2096    2...11    83.211414: funcgraph_entry:                   |            preempt_count_sub() {
mem_test-2096    2....1    83.211414: funcgraph_exit:         0.521 us   |            }
mem_test-2096    2....1    83.211415: funcgraph_exit:       + 13.146 us  |          }
mem_test-2096    2....1    83.211415: funcgraph_entry:                   |          __cpu_clear_user_page() {
mem_test-2096    2....1    83.211420: funcgraph_exit:         4.270 us   |          }
mem_test-2096    2....1    83.211420: funcgraph_entry:                   |          migrate_enable() {
mem_test-2096    2....1    83.211421: funcgraph_entry:                   |            preempt_count_add() {
mem_test-2096    2...11    83.211421: funcgraph_exit:         0.521 us   |            }
mem_test-2096    2...10    83.211422: funcgraph_entry:                   |            task_rq_lock() {
mem_test-2096    2...10    83.211422: funcgraph_entry:                   |              _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211423: funcgraph_entry:                   |                preempt_count_add() {
mem_test-2096    2d..20    83.211423: funcgraph_exit:         0.500 us   |                }
mem_test-2096    2d..20    83.211424: funcgraph_exit:         1.583 us   |              }
mem_test-2096    2d..20    83.211424: funcgraph_entry:                   |              _raw_spin_lock() {
mem_test-2096    2d..20    83.211425: funcgraph_entry:                   |                preempt_count_add() {
mem_test-2096    2d..30    83.211425: funcgraph_exit:         0.500 us   |                }
mem_test-2096    2d..30    83.211426: funcgraph_exit:         1.563 us   |              }
mem_test-2096    2d..30    83.211426: funcgraph_exit:         4.687 us   |            }
mem_test-2096    2d..30    83.211427: funcgraph_entry:                   |            preempt_count_sub() {
mem_test-2096    2d..20    83.211427: funcgraph_exit:         0.500 us   |            }
mem_test-2096    2d..20    83.211428: funcgraph_entry:                   |            _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211429: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2...10    83.211429: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2...10    83.211430: funcgraph_exit:         1.563 us   |            }
mem_test-2096    2...10    83.211430: funcgraph_entry:                   |            unpin_current_cpu() {
mem_test-2096    2...10    83.211431: funcgraph_entry:                   |              __read_rt_unlock() {
mem_test-2096    2...10    83.211431: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2...10    83.211432: funcgraph_exit:         1.605 us   |            }
mem_test-2096    2...10    83.211432: funcgraph_entry:                   |            preempt_count_sub() {
mem_test-2096    2....0    83.211433: funcgraph_exit:         0.521 us   |            }
mem_test-2096    2....0    83.211433: funcgraph_exit:       + 13.187 us  |          }
mem_test-2096    2....0    83.211434: funcgraph_entry:                   |          mem_cgroup_try_charge_delay() {
mem_test-2096    2....0    83.211434: funcgraph_entry:                   |            mem_cgroup_try_charge() {
mem_test-2096    2....0    83.211435: funcgraph_entry:                   |              get_mem_cgroup_from_mm() {
mem_test-2096    2....0    83.211436: funcgraph_entry:                   |                __rcu_read_lock() {
mem_test-2096    2....0    83.211436: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2....0    83.211437: funcgraph_entry:                   |                __rcu_read_unlock() {
mem_test-2096    2....0    83.211437: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2....0    83.211438: funcgraph_exit:         2.708 us   |              }
mem_test-2096    2....0    83.211438: funcgraph_entry:                   |              try_charge() {
mem_test-2096    2....0    83.211439: funcgraph_exit:         0.521 us   |              }
mem_test-2096    2....0    83.211439: funcgraph_exit:         4.834 us   |            }
mem_test-2096    2....0    83.211440: funcgraph_entry:                   |            mem_cgroup_throttle_swaprate() {
mem_test-2096    2....0    83.211440: funcgraph_entry:                   |              __rcu_read_lock() {
mem_test-2096    2....0    83.211441: funcgraph_exit:         0.520 us   |              }
mem_test-2096    2....0    83.211441: funcgraph_entry:                   |              kthread_blkcg() {
mem_test-2096    2....0    83.211442: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....0    83.211443: funcgraph_entry:                   |              __rcu_read_unlock() {
mem_test-2096    2....0    83.211443: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....0    83.211444: funcgraph_exit:         3.813 us   |            }
mem_test-2096    2....0    83.211444: funcgraph_exit:       + 10.229 us  |          }
mem_test-2096    2....0    83.211445: funcgraph_entry:                   |          rt_spin_lock() {
mem_test-2096    2....0    83.211445: funcgraph_entry:                   |            migrate_disable() {
mem_test-2096    2....0    83.211446: funcgraph_entry:                   |              preempt_count_add() {
mem_test-2096    2...10    83.211446: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2...10    83.211447: funcgraph_entry:                   |              pin_current_cpu() {
mem_test-2096    2...10    83.211447: funcgraph_entry:                   |                __read_rt_trylock() {
mem_test-2096    2...10    83.211448: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2...10    83.211448: funcgraph_exit:         1.605 us   |              }
mem_test-2096    2...10    83.211449: funcgraph_entry:                   |              task_rq_lock() {
mem_test-2096    2...10    83.211450: funcgraph_entry:                   |                _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211450: funcgraph_entry:                   |                  preempt_count_add() {
mem_test-2096    2d..20    83.211451: funcgraph_exit:         0.500 us   |                  }
mem_test-2096    2d..20    83.211451: funcgraph_exit:         1.583 us   |                }
mem_test-2096    2d..20    83.211452: funcgraph_entry:                   |                _raw_spin_lock() {
mem_test-2096    2d..20    83.211452: funcgraph_entry:                   |                  preempt_count_add() {
mem_test-2096    2d..30    83.211453: funcgraph_exit:         0.500 us   |                  }
mem_test-2096    2d..30    83.211453: funcgraph_exit:         1.542 us   |                }
mem_test-2096    2d..30    83.211454: funcgraph_exit:         4.688 us   |              }
mem_test-2096    2d..30    83.211454: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2d..20    83.211455: funcgraph_exit:         0.500 us   |              }
mem_test-2096    2d..20    83.211455: funcgraph_entry:                   |              _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211456: funcgraph_entry:                   |                preempt_count_sub() {
mem_test-2096    2...10    83.211456: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2...10    83.211457: funcgraph_exit:         1.584 us   |              }
mem_test-2096    2...11    83.211457: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2....1    83.211458: funcgraph_exit:         0.520 us   |              }
mem_test-2096    2....1    83.211458: funcgraph_exit:       + 13.187 us  |            }
mem_test-2096    2....1    83.211459: funcgraph_exit:       + 14.271 us  |          }
mem_test-2096    2....1    83.211459: funcgraph_entry:                   |          page_add_new_anon_rmap() {
mem_test-2096    2....1    83.211460: funcgraph_entry:                   |            __mod_node_page_state() {
mem_test-2096    2....1    83.211461: funcgraph_entry:                   |              preempt_count_add() {
mem_test-2096    2...11    83.211461: funcgraph_exit:         0.563 us   |              }
mem_test-2096    2...11    83.211462: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2....1    83.211462: funcgraph_exit:         0.521 us   |              }
mem_test-2096    2....1    83.211463: funcgraph_exit:         2.792 us   |            }
mem_test-2096    2....1    83.211463: funcgraph_entry:                   |            __page_set_anon_rmap() {
mem_test-2096    2....1    83.211464: funcgraph_exit:         0.542 us   |            }
mem_test-2096    2....1    83.211464: funcgraph_exit:         4.958 us   |          }
mem_test-2096    2....1    83.211465: funcgraph_entry:                   |          mem_cgroup_commit_charge() {
mem_test-2096    2....1    83.211466: funcgraph_entry:                   |            migrate_disable() {
mem_test-2096    2....2    83.211466: funcgraph_exit:         0.563 us   |            }
mem_test-2096    2....2    83.211467: funcgraph_entry:                   |            rt_spin_lock() {
mem_test-2096    2....2    83.211467: funcgraph_entry:                   |              migrate_disable() {
mem_test-2096    2....3    83.211468: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....3    83.211468: funcgraph_exit:         1.646 us   |            }
mem_test-2096    2....3    83.211469: funcgraph_entry:                   |            mem_cgroup_charge_statistics() {
mem_test-2096    2....3    83.211470: funcgraph_exit:         0.688 us   |            }
mem_test-2096    2....3    83.211470: funcgraph_entry:                   |            memcg_check_events() {
mem_test-2096    2....3    83.211471: funcgraph_entry:                   |              mem_cgroup_event_ratelimit.isra.0() {
mem_test-2096    2....3    83.211471: funcgraph_exit:         0.605 us   |              }
mem_test-2096    2....3    83.211472: funcgraph_exit:         1.729 us   |            }
mem_test-2096    2....3    83.211472: funcgraph_entry:                   |            rt_spin_unlock() {
mem_test-2096    2....3    83.211473: funcgraph_entry:                   |              migrate_enable() {
mem_test-2096    2....2    83.211474: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....2    83.211474: funcgraph_exit:         1.646 us   |            }
mem_test-2096    2....2    83.211475: funcgraph_entry:                   |            migrate_enable() {
mem_test-2096    2....1    83.211475: funcgraph_exit:         0.563 us   |            }
mem_test-2096    2....1    83.211476: funcgraph_exit:       + 10.708 us  |          }
mem_test-2096    2....1    83.211476: funcgraph_entry:                   |          lru_cache_add_active_or_unevictable() {
mem_test-2096    2....1    83.211477: funcgraph_entry:                   |            __lru_cache_add() {
mem_test-2096    2....1    83.211477: funcgraph_entry:                   |              migrate_disable() {
mem_test-2096    2....2    83.211478: funcgraph_exit:         0.542 us   |              }
mem_test-2096    2....2    83.211479: funcgraph_entry:                   |              rt_spin_lock() {
mem_test-2096    2....2    83.211479: funcgraph_entry:                   |                migrate_disable() {
mem_test-2096    2....3    83.211480: funcgraph_exit:         0.500 us   |                }
mem_test-2096    2....3    83.211480: funcgraph_exit:         1.625 us   |              }
mem_test-2096    2....3    83.211481: funcgraph_entry:                   |              rt_spin_unlock() {
mem_test-2096    2....3    83.211481: funcgraph_entry:                   |                migrate_enable() {
mem_test-2096    2....2    83.211482: funcgraph_exit:         0.541 us   |                }
mem_test-2096    2....2    83.211482: funcgraph_exit:         1.667 us   |              }
mem_test-2096    2....2    83.211483: funcgraph_entry:                   |              migrate_enable() {
mem_test-2096    2....1    83.211483: funcgraph_exit:         0.562 us   |              }
mem_test-2096    2....1    83.211484: funcgraph_exit:         7.229 us   |            }
mem_test-2096    2....1    83.211484: funcgraph_exit:         8.291 us   |          }
mem_test-2096    2....1    83.211485: funcgraph_entry:                   |          rt_spin_unlock() {
mem_test-2096    2....1    83.211486: funcgraph_entry:                   |            migrate_enable() {
mem_test-2096    2....1    83.211486: funcgraph_entry:                   |              preempt_count_add() {
mem_test-2096    2...11    83.211487: funcgraph_exit:         0.541 us   |              }
mem_test-2096    2...10    83.211487: funcgraph_entry:                   |              task_rq_lock() {
mem_test-2096    2...10    83.211488: funcgraph_entry:                   |                _raw_spin_lock_irqsave() {
mem_test-2096    2d..10    83.211488: funcgraph_entry:                   |                  preempt_count_add() {
mem_test-2096    2d..20    83.211489: funcgraph_exit:         0.479 us   |                  }
mem_test-2096    2d..20    83.211490: funcgraph_exit:         2.042 us   |                }
mem_test-2096    2d..20    83.211490: funcgraph_entry:                   |                _raw_spin_lock() {
mem_test-2096    2d..20    83.211491: funcgraph_entry:                   |                  preempt_count_add() {
mem_test-2096    2d..30    83.211492: funcgraph_exit:         0.500 us   |                  }
mem_test-2096    2d..30    83.211492: funcgraph_exit:         1.708 us   |                }
mem_test-2096    2d..30    83.211493: funcgraph_exit:         5.333 us   |              }
mem_test-2096    2d..30    83.211493: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2d..20    83.211494: funcgraph_exit:         0.500 us   |              }
mem_test-2096    2d..20    83.211494: funcgraph_entry:                   |              _raw_spin_unlock_irqrestore() {
mem_test-2096    2...20    83.211495: funcgraph_entry:                   |                preempt_count_sub() {
mem_test-2096    2...10    83.211495: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2...10    83.211496: funcgraph_exit:         1.583 us   |              }
mem_test-2096    2...10    83.211496: funcgraph_entry:                   |              unpin_current_cpu() {
mem_test-2096    2...10    83.211497: funcgraph_entry:                   |                __read_rt_unlock() {
mem_test-2096    2...10    83.211497: funcgraph_exit:         0.521 us   |                }
mem_test-2096    2...10    83.211498: funcgraph_exit:         1.583 us   |              }
mem_test-2096    2...10    83.211498: funcgraph_entry:                   |              preempt_count_sub() {
mem_test-2096    2....0    83.211499: funcgraph_exit:         0.520 us   |              }
mem_test-2096    2....0    83.211500: funcgraph_exit:       + 13.896 us  |            }
mem_test-2096    2....0    83.211500: funcgraph_exit:       + 14.979 us  |          }
mem_test-2096    2....0    83.211501: funcgraph_exit:       ! 146.792 us |        }
mem_test-2096    2....0    83.211501: funcgraph_exit:       ! 148.188 us |      }
mem_test-2096    2....0    83.211502: funcgraph_exit:       ! 154.813 us |    }
mem_test-2096    2....0    83.211502: funcgraph_entry:                   |    up_read() {
mem_test-2096    2....0    83.211503: funcgraph_entry:                   |      __up_read() {
mem_test-2096    2....0    83.211503: funcgraph_exit:         0.562 us   |      }
mem_test-2096    2....0    83.211504: funcgraph_exit:         1.646 us   |    }
mem_test-2096    2....0    83.211504: funcgraph_exit:       ! 162.729 us |  }

2）标准内核

mem_test-2234    3....   438.228220: funcgraph_entry:                   |  do_page_fault() {
mem_test-2234    3....   438.228220: funcgraph_entry:                   |    down_read_trylock() {
mem_test-2234    3....   438.228221: funcgraph_exit:         0.355 us   |    }
mem_test-2234    3....   438.228221: funcgraph_entry:                   |    find_vma() {
mem_test-2234    3....   438.228222: funcgraph_entry:                   |      vmacache_find() {
mem_test-2234    3....   438.228222: funcgraph_exit:         0.333 us   |      }
mem_test-2234    3....   438.228222: funcgraph_exit:         1.041 us   |    }
mem_test-2234    3....   438.228223: funcgraph_entry:                   |    handle_mm_fault() {
mem_test-2234    3....   438.228223: funcgraph_entry:                   |      mem_cgroup_from_task() {
mem_test-2234    3....   438.228223: funcgraph_exit:         0.334 us   |      }
mem_test-2234    3....   438.228224: funcgraph_entry:                   |      __handle_mm_fault() {
mem_test-2234    3....   438.228224: funcgraph_entry:                   |        pmd_devmap_trans_unstable() {
mem_test-2234    3....   438.228224: funcgraph_exit:         0.334 us   |        }
mem_test-2234    3....   438.228225: funcgraph_entry:                   |        do_anonymous_page() {
mem_test-2234    3....   438.228225: funcgraph_entry:                   |          alloc_pages_vma() {
mem_test-2234    3....   438.228226: funcgraph_entry:                   |            __get_vma_policy() {
mem_test-2234    3....   438.228226: funcgraph_exit:         0.333 us   |            }
mem_test-2234    3....   438.228226: funcgraph_entry:                   |            get_vma_policy.part.10() {
mem_test-2234    3....   438.228227: funcgraph_entry:                   |              get_task_policy.part.7() {
mem_test-2234    3....   438.228227: funcgraph_exit:         0.354 us   |              }
mem_test-2234    3....   438.228227: funcgraph_exit:         1.021 us   |            }
mem_test-2234    3....   438.228228: funcgraph_entry:                   |            policy_nodemask() {
mem_test-2234    3....   438.228228: funcgraph_exit:         0.354 us   |            }
mem_test-2234    3....   438.228228: funcgraph_entry:                   |            policy_node() {
mem_test-2234    3....   438.228229: funcgraph_exit:         0.334 us   |            }
mem_test-2234    3....   438.228229: funcgraph_entry:                   |            __alloc_pages_nodemask() {
mem_test-2234    3....   438.228229: funcgraph_entry:                   |              get_page_from_freelist() {
mem_test-2234    3d...   438.228230: funcgraph_entry:                   |                __inc_numa_state() {
mem_test-2234    3d...   438.228230: funcgraph_exit:         0.313 us   |                }
mem_test-2234    3d...   438.228231: funcgraph_entry:                   |                __inc_numa_state() {
mem_test-2234    3d...   438.228231: funcgraph_exit:         0.334 us   |                }
mem_test-2234    3....   438.228231: funcgraph_exit:         1.833 us   |              }
mem_test-2234    3....   438.228232: mm_page_alloc:        page=0xffa0b329 pfn=2145349 order=0 migratetype=1 gfp_flags=GFP_HIGHUSER_MOVABLE
mem_test-2234    3....   438.228232: funcgraph_exit:         2.750 us   |            }
mem_test-2234    3....   438.228232: funcgraph_exit:         6.917 us   |          }
mem_test-2234    3....   438.228232: funcgraph_entry:                   |          __cpu_clear_user_page() {
mem_test-2234    3....   438.228237: funcgraph_exit:         4.000 us   |          }
mem_test-2234    3....   438.228237: funcgraph_entry:                   |          mem_cgroup_try_charge_delay() {
mem_test-2234    3....   438.228237: funcgraph_entry:                   |            mem_cgroup_try_charge() {
mem_test-2234    3....   438.228238: funcgraph_entry:                   |              get_mem_cgroup_from_mm() {
mem_test-2234    3....   438.228238: funcgraph_exit:         0.334 us   |              }
mem_test-2234    3....   438.228238: funcgraph_entry:                   |              try_charge() {
mem_test-2234    3....   438.228239: funcgraph_exit:         0.333 us   |              }
mem_test-2234    3....   438.228239: funcgraph_exit:         1.730 us   |            }
mem_test-2234    3....   438.228239: funcgraph_entry:                   |            mem_cgroup_throttle_swaprate() {
mem_test-2234    3....   438.228240: funcgraph_entry:                   |              kthread_blkcg() {
mem_test-2234    3....   438.228240: funcgraph_exit:         0.354 us   |              }
mem_test-2234    3....   438.228240: funcgraph_exit:         1.021 us   |            }
mem_test-2234    3....   438.228241: funcgraph_exit:         3.792 us   |          }
mem_test-2234    3....   438.228241: funcgraph_entry:                   |          page_add_new_anon_rmap() {
mem_test-2234    3....   438.228242: funcgraph_entry:                   |            __mod_node_page_state() {
mem_test-2234    3....   438.228242: funcgraph_exit:         0.333 us   |            }
mem_test-2234    3....   438.228242: funcgraph_entry:                   |            __page_set_anon_rmap() {
mem_test-2234    3....   438.228243: funcgraph_exit:         0.333 us   |            }
mem_test-2234    3....   438.228243: funcgraph_exit:         1.729 us   |          }
mem_test-2234    3....   438.228243: funcgraph_entry:                   |          mem_cgroup_commit_charge() {
mem_test-2234    3d...   438.228244: funcgraph_entry:                   |            mem_cgroup_charge_statistics() {
mem_test-2234    3d...   438.228244: funcgraph_exit:         0.333 us   |            }
mem_test-2234    3d...   438.228244: funcgraph_entry:                   |            memcg_check_events() {
mem_test-2234    3d...   438.228245: funcgraph_entry:                   |              mem_cgroup_event_ratelimit.isra.2() {
mem_test-2234    3d...   438.228245: funcgraph_exit:         0.333 us   |              }
mem_test-2234    3d...   438.228245: funcgraph_exit:         1.000 us   |            }
mem_test-2234    3....   438.228246: funcgraph_exit:         2.354 us   |          }
mem_test-2234    3....   438.228246: funcgraph_entry:                   |          lru_cache_add_active_or_unevictable() {
mem_test-2234    3....   438.228246: funcgraph_entry:                   |            __lru_cache_add() {
mem_test-2234    3....   438.228247: funcgraph_exit:         0.354 us   |            }
mem_test-2234    3....   438.228247: funcgraph_exit:         1.042 us   |          }
mem_test-2234    3....   438.228247: funcgraph_exit:       + 22.541 us  |        }
mem_test-2234    3....   438.228248: funcgraph_exit:       + 23.980 us  |      }
mem_test-2234    3....   438.228248: funcgraph_exit:       + 25.437 us  |    }
mem_test-2234    3....   438.228248: funcgraph_entry:                   |    up_read() {
mem_test-2234    3....   438.228249: funcgraph_exit:         0.333 us   |    }
mem_test-2234    3....   438.228249: funcgraph_exit:       + 29.021 us  |  }

5，分析

从上述log的对比来看，实时内核的do_page_fault比非实时内核耗时多100+us，但主要是由于function_graph的开销引起的，实测没有多那么多，大约多5us左右。

实时内核调用了get_page_from_freelist--rmqueue_pcplist--rmqueue--local_lock_irqsave--rt_spin_lock--migrate_disable，migrate_disable里包含了很多preempt的操作，同时还有其成对出现的migrate_enable，也有同样的操作，所以导致了额外的开销以及耗时。

在申请大量内存并使用的情况下，系统会发生多次这种minor pagefault，累计起来，会有一定的性能损失。同时，migrate disable/enable也会使用_raw_spin_lock，task_rq_lock，如果申请的资源没有空闲的话，还有额外的等待时间。

6，参考

本文参考了链接“（十四）Linux内存管理之page fault处理【转】 - sky-heaven - 博客园”对pagefault的分析，结合代码来看，分析的很准确。

mozart1756
关注
页错误 Page Fault /缺页异常 详解
LETFLY'S LAB
01-26 1943
页错误 Page Fault/缺页异常 详解 1. 第一部分：如果你看得懂 1.1 页错误定义 1.2 页错误的处理 2. 第二部分：如果你看不懂上面的，请看这里 2.1. 举例子（背景） 2.1.1 进程及页映射 2.1.2 映射关系和数据结构 2.2 结合图说明什么是页错误 2.2.1 页错误产生 2.2.2 页错误处理 1. 第一部分：如果你看得懂 1.1 页错误定义 当cpu执行进程的某个页面时，发现他要访问的页(虚拟地
loongarch集成preempt rt后ltpstress死机的问题分析
chensong_2000的博客
07-05 711
loongarch集成preempt rt后ltpstress死机的问题分析
LWN：在mainline中关闭task migration！
Linux News搬运工
11-19 1026
关注了就能看到更多这么棒的文章哦～Migration disable for the mainlineBy Jonathan CorbetNovember 9, 2020DeepL as...
ftrace基础知识学习+使用实例 （仅作学习记录）
wangyezi19930928的专栏
02-22 5352
ftrace 简介 Trace 对于软件的维护和性能分析至关重要，ftrace 是当前 Linux 内核中一种新的 trace 工具。本文介绍 ftrace 的使用和实现原理，并将 ftrace 和 systemTap，LTTng 等软件进行对比，希望读者能够对 ftrace 有一个全面的了解。
一步一步学linux操作系统: 13 进程调度二_主动调度
墨1024
06-30 831
进程调度可分为主动调度与抢占式调度，主动调度 、schedule 函数、进程上下文切换
RTPREEMPT rt实时补丁
Atlas2013的专栏
05-15 4683
RT PREEMPT HOWTO Intro Authors Luotao Fu (l.fu AT pengutronix DOT de), Pengutronix e.K., Kernel Development Group  Robert Schwebel (r.schwebel AT pengutronix DOT de), Pengutronix e.K., Ker
关于操作系统设计的基本原理和设计原则
tugouxp的专栏
03-09 6900
操作系统设计的精妙之处就在于，在底层硬件之上创造了新的抽象， 对于系统初始化来说， 它呈现了执行流到线程转化这一概念，该概念远远比它的实现细节更加重要，处理器以“取指-执行”为周期开始串行执行指令，而初始化代码将自身转化为一个并发处理系统， 这里的关键之处在于，初始化代码并不是创建一个独立的并发系统，然后跳转到新的系统。
有利于提高xenomai /PREEMPT-RT 实时性的一些配置建议
qq_38350702的博客
08-04 805
实时”一词在许多应用领域中使用，人们它有不同的解释，并不总是正确的。人们常说，如果控制系统能够对外部事件做出快速反应，那么它就是实时运行的。根据这种解释，如果系统速度快，则系统被认为是实时的。然而，“快”具有相对含义，并未涵盖表征这些类型系统的主要属性。我们来看一下，在自然界中，生物在栖息地中的实时行为，这些行为与它们的速度无关。例如，乌龟对来自其栖息地的外部刺激的反应，与猫对其栖息地的外部反应一样有效。
ARM平台Linux内核空指针异常处理流程
zangdongming的专栏
08-13 4703
ARM平台Linux内核空指针异常处理流程 平台：Linux 3.10.40 + ARM V7 一 从异常向量入口到__do_kernel_fault 访问空指针在ARM平台上属于data abort异常，对应异常向量中的vector_dabt。 文件：arch/arm/kernel/entry-armv.S __vectors_start:          W(b)  
arm linux内核实时补丁,宋宝华： Linux实时补丁的原理和实践
weixin_32968155的博客
05-02 1442
2012年的文章，重新在微信公众号发表。第一章：硬实时Linux(RT-Preempt Patch)在PC上的编译、使用和测试第二章：硬实时Linux(RT-Preempt Patch)的中断线程化硬实时Linux(RT-Preempt Patch)在PC上的编译、使用和测试Vanilla kernel的问题Linux kernel在spinlock、irq上下文方面无法抢占，因此高优先级任务被唤...
从 oom-killer 到 cgroup
永远相信美好的事情即将发生
10-14 773
最近线上遇到一个问题，运行的好好的Java应用莫名其妙的不见了，记录的日志也到了某个时刻"戛然而止",就好像应用对应的进程突然"人间蒸发"了似的。 经过一番查找，终于找到的系统日志中找到了一些猫腻： [8314616.900000] java invoked oom-killer: gfp_mask=0xd0, order=0, oom_score_adj=0 [8314616.901123] java cpuset=docker-4da60884c4c56e1c58f86fcd853ad7510f1f1e
php artisan migration 数据库操作的使用说明
lw545034502的博客
03-02 6581
一、命名规范 1，创建表命名 php artisan make:migration create_tablename_table --create="tablename" 2，修改字段命名 php artisan make:migration alter_tablename_table --table=tablename 3，添加字段命名 php artisan make:mig...
migrate设置
weixin_30387799的博客
02-21 171
migrate配置 进入项目manage所在的目录中执行如下命令，否则会报错：no such table: django_session 进入cmd 输入盘符比如：E: 切换目录 cd E:\progect\Firstdjango 执行migrate python manage.py migrate 转载于:https://www....
linux rt 线程调度
shenhuxi_yu的专栏
07-18 1226
aarch64 linux4.9 最近遇到问题需要trace某个rt线程调度情况 trace event filter: echo 1 > sched/sched_switch/enable echo "next_comm == xxx" >sched/sched_switch/filter (filter 格式可以cat format 同目录下获得) echo 1 ...
Linux进程管理 (7)实时调度
weixin_34130269的博客
08-11 147
关键词：RT、preempt_count、RT patch。   除了CFS调度器之外，还包括重要的实时调度器，有两种RR和FIFO调度策略。本章只是一个简单的介绍。 更详细的介绍参考《Linux进程管理 (9)实时调度类分析，以及FIFO和RR对比实验》。 同时为了提高Linux的实时性，Linux社区还维护了realtime相关的补丁。这些补丁的介绍在《Linux实时补丁及其分析》。   1....
linux内存管理笔记
r7ronalshun的专栏
11-16 1086
1.     地址空间划分 Start              End        Use -------------------------------------------------------------------------- ffff8000  ffffffff      copy_user_page/ clear_user_page use.
linux那些事之page fault（AMD64架构)(user space)（2）
weixin_42730667的博客
01-26 3012
do_user_addr_fault 用户空间地址处理是page fault主要处理流程，x86 64位系统主要是do_user_addr_fault（）函数 该处理部分是x86架构特有部分 即与架构有关处理，整个函数处理逻辑过程可以总结为如下图所示： do_user_addr_fault源码 /* Handle faults in the user portion of the address space */ static inline void do_user_addr_faul..
linux内核虚拟化之路(一) cgroup机制
eleven_xiy的博客
02-08 9085
PREEMPT_RT
最新发布
03-06
PREEMPT_RT是一个用于实时操作系统（Real-Time Operating System，RTOS）的Linux内核补丁，它旨在提供更可预测和可靠的实时性能。PREEMPT_RT补丁使得Linux内核能够在实时任务和普通任务之间进行快速切换，从而降低实时任务的响应时间和延迟。 PREEMPT_RT补丁的主要特性包括： 1. 实时抢占：PREEMPT_RT补丁允许实时任务在任何时刻抢占正在运行的普通任务，以确保实时任务能够及时响应。 2. 内核自旋锁替代：PREEMPT_RT补丁使用更高效的自旋锁实现，减少了自旋锁带来的延迟。 3. 实时调度器：PREEMPT_RT补丁引入了一个实时调度器，它可以根据任务的优先级和截止时间来进行任务调度，以满足实时任务的需求。 4. 实时中断处理：PREEMPT_RT补丁改进了中断处理机制，使得中断能够更快地响应和处理。 PREEMPT_RT补丁对于需要高实时性能的应用非常有用，例如工业自动化、机器人控制、音视频处理等领域。它可以提供更可靠和可预测的实时性能，确保实时任务能够按时完成。
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值