2023-09-07 05:27:30

by Zheng Yejian

[permalink] [raw]
Subject: Re: "[PATCH] selftests/ftrace: Correctly enable event in instance-event.tc"

On 2023/9/7 05:37, Steven Rostedt wrote:
> On Wed, 6 Sep 2023 10:37:18 -0400
> Steven Rostedt <[email protected]> wrote:
>
>>> Log details,
>>> -------------
>>> # ok 45 ftrace - test tracing error log support
>>> <47>[ 1373.662292] systemd-journald[90]: Sent WATCHDOG=1 notification.
>>> # ok 46 Test creation and deletion of trace instances while setting an event
>>
>> It's definitely a race with the creation and deletion of instances.
>>
>> I'm going to run it on my laptop VM and see if that reproduces it. My other
>> VM is on a pretty powerful machine, and perhaps that's keeping it from
>> hitting the race.
>
> Putting in a while loop of:
>
> # while :; do ./ftracetest test.d/instances/instance-event.tc ; done
>

In test.d/instances/instance-event.tc, concurrently create/delete/access
intances just about 1 second and then killed them, it may cause missing
the race.

No longer sleep and kill, following testcase can also reproduce:
```
#!/bin/bash

cd /sys/kernel/tracing/instances

instance_slam() {
while :; do
mkdir foo 2> /dev/null
rmdir foo 2> /dev/null
done
}

instance_set() {
while :; do
echo 1 > foo/events/sched/sched_switch/enable
done 2> /dev/null
}

instance_slam &
p1=$!
echo $p1

instance_set &
p2=$!
echo $p2
```

--

Thanks,
Zheng Yejian

> eventually triggered the bug. Looks like this is really an existing bug not
> related to the eventfs, but the eventfs code actually opened up the window
> of this race.
>
> Hopefully I'll have a fix shortly.
>
> -- Steve
>