[LTP] [PATCH] commands/sysctl: Add new regression test for invalid sched_time_avg

[Xiao Yang]

On 2018/03/06 20:42, Cyril Hrubis wrote:
> Hi!
>> +TST_TESTFUNC=sysctl_test
>> +TST_NEEDS_CMDS="sysctl"
>> +
>> +. tst_test.sh
>> +
>> +get_value()
>> +{
>> +	local value=$(sysctl -n $1)
>> +	[ -z "$value" ]&&  tst_brk TBROK \
>> +		"Failed to get the value of sched_time_avg(_ms)"
>> +
>> +	echo $value
>> +}
>> +
>> +sysctl_test()
>> +{
>> +	# With commit d00535d, sched_time_avg was renamed as sched_time_avg_ms
>> +	local name=$(sysctl -a 2>&1 | egrep -o \
>> +		'kernel.(sched_time_avg|sched_time_avg_ms)')
>> +	[ -z "$name" ]&&  tst_brk TCONF \
>> +		"sched_time_avg(_ms) was not supported"
> Can't we just do?
>
> if [ -e /proc/sys/kernel/sched_time_avg_ms ]; then
> 	name=...
> fi
>
> if [ -e /proc/sys/kernel/sched_time_avg ]; then
> 	name=...
> fi
>
> [ -z "$name" ] ...
Hi Cyril,

Agreed.  Using /proc/sys/ instead of sysctl command seems better.
>> +	local orig_value=$(get_value $name)
>> +
>> +	sysctl -w $name=0>/dev/null 2>&1
>> +
>> +	# Increase the chance of exposing the problem on RHEL6
>> +	sleep 3
> What is this sleep good for? Do we wait here for the kernel to crash?
On my RHEL6, restoring the original value immediately after setting an 
invalid value always didn't
trigger a hang in time, so we just wait a moment for kernel to hang.

Thanks,
Xiao Yang
>> +	local curr_value=$(get_value $name)
>> +
>> +	if [ ${curr_value} -eq ${orig_value} ]; then
>> +		tst_res TPASS "Setting sched_time_avg(_ms) failed"
>> +	else
>> +		tst_res TFAIL "Setting sched_time_avg(_ms) succeeded"
>> +		sysctl -w $name=${orig_value}>/dev/null 2>&1
>> +	fi
>> +}
>> +
>> +tst_run
>> -- 
>> 1.8.3.1
>>
>>
>>
>>
>> -- 
>> Mailing list info: https://lists.linux.it/listinfo/ltp