Mirrors_BOOKS/LDD-LinuxDeviceDrivers
1
0
Fork 0
mirror of https://github.com/gatieme/LDD-LinuxDeviceDrivers.git synced 2026-02-09 06:22:43 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

update perf bug...

Browse Source
This commit is contained in:
gatieme
2017-10-08 22:24:45 +08:00
parent 1cd7343048
commit fdba1990f0
1 changed files with 0 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
study/patch/perf/perf.md
View File

@@ -1,51 +0,0 @@
1 perf record sched:sched_wakeup_new ./bug_fork
perf/ftrace : Fix repetitious traces when specify a target task
When use perf to trace the sched_wakeup and sched_wakeup_new tracepoint, there is a bug that output the same event repetitiously, It can be reproduced by :
```cpp
perf record -e sched:sched_wakeup ./bug_fork
OR
./bug_fork # pid 1052
perf record -e sched:sched_wakeup -p 1052
```
```cpp
swapper 0 [000] 3854.085452: sched:sched_wakeup: comm=perf pid=1101 prio=120 target_c
pu=000
swapper 0 [000] 3854.085473: sched:sched_wakeup: comm=perf pid=1101 prio=120 target_c
pu=000
bug_fork 1101 [000] 3854.089933: sched:sched_wakeup_new: comm=bug_fork pid=1102 prio=120 targ
et_cpu=001
bug_fork 1101 [000] 3854.089950: sched:sched_wakeup_new: comm=bug_fork pid=1102 prio=120 targ
et_cpu=001
bug_fork 1101 [000] 3854.089959: sched:sched_wakeup_new: comm=bug_fork pid=1102 prio=120 targ
et_cpu=001
bug_fork 1102 [001] 3854.092185: sched:sched_wakeup: comm=rcu_sched pid=8 prio=120 target
_cpu=001
swapper 0 [000] 3854.665213: sched:sched_wakeup: comm=bug_fork pid=1101 prio=120 targ
et_cpu=000
swapper 0 [000] 3854.665253: sched:sched_wakeup: comm=bug_fork pid=1101 prio=120 targ
et_cpu=000
```
Some events not only monitor the current task, but also specify a target task. For example, the sched_wakeup/sched_wakeup_new tracepoint will be caught when the current task wakeup the target task which we traced on.
the commit bed5b2 has designed a method to trace these events which specify a target task. Due to these events are registered for each CPU(percpu), so perf_swevent_event will match the event multiple times, the number of repetitions is just the number of CPUs.
On the other hand perf_tp_event will only match an event event at a time, so we will return after an event matched.
2 perf sched record 支持 -p 1 但是不支持 task
3 perf/trace record 显示current有时候不正常, 只显示pid, 不显示comm
stat_runtime current进程不正确
4 pr_debug 不支持 arm64 和 arm32