Re: [PATCH, RFC] xfstests: fail test if it causes a kernel taint

[Eric Sandeen <sandeen@sandeen.net>]

On 5/13/13 10:15 PM, Dave Chinner wrote:
> On Mon, May 13, 2013 at 09:00:45PM -0500, Eric Sandeen wrote:
>> On 5/13/13 8:15 PM, Dave Chinner wrote:
>>> On Mon, May 13, 2013 at 04:55:18PM -0500, Eric Sandeen wrote:
>>>> This should, in theory, fail a test if it introduces one of
>>>> a handful of "serious" kernel taints.  I mask on a few taint
>>>> values because using an out of tree module or a non-GPL module
>>>> should never fail a test, for example.
>>>>
>>>> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
>>>
>>> I like the idea - it should catch lockdep failures and other such
>>> problems.
>>>
>>>> ---
>>>>
>>>> diff --git a/check b/check
>>>> index a79747e..a9cac4e 100755
>>>> --- a/check
>>>> +++ b/check
>>>> @@ -446,6 +445,11 @@ do
>>>>  	fi
>>>>  	rm -f core $seqres.notrun
>>>>  
>>>> +	if [ "$HOSTOS" == "Linux" ]; then
>>>> +		tainted=`sysctl -n kernel.tainted`
>>>> +		let "tainted &= $TAINT_FAIL"
>>>> +	fi
>>>> +
>>>>  	start=`_wallclock`
>>>>  	$timestamp && echo -n "	["`date "+%T"`"]"
>>>>  	[ ! -x $seq ] && chmod u+x $seq # ensure we can run it
>>>> @@ -507,6 +511,19 @@ do
>>>>  			 "entire diff)"
>>>>  		    err=true
>>>>  		fi
>>>> +
>>>> +		# See if this run tainted the kernel due to oops, etc
>>>> +		if [ "$HOSTOS" == "Linux" ]
>>>> +		then
>>>> +			tainted2=`sysctl -n kernel.tainted`
>>>> +			let "tainted2 &= $TAINT_FAIL"
>>>> +			if [ "$tainted" != "$tainted2" ]
>>>> +			then
>>>> +				echo " Kernel taint changed from $tainted to $tainted2."
>>>> +				echo " See dmesg for details."
>>>> +				err=true
>>>> +			fi
>>>
>>> Should we dump the last 100 lines of dmesg into the output file
>>> here? And then rely on the golden image match failing to fail the
>>> test?
>>
>> Or into $seq.full . .  I dunno.  It's hard to know how much dmesg you'd need to get the right bits.
>>
>> Dumping all of it into $seq.full might be best; it only trips when it's a new taint, so won't
>> fire for every test.
> 
> Only problem with $seqres.full is that it is internal to the test
> and so many tests will not remove one that is there before the test
> starts. Hence you'll get growing files the never get removed from
> repeated failures. 

Oh, I see what you mean, yeah.

> Even just dumping the above error message into
> the output file would be sufficient, I think, as the 10 line diff
> output will state it pretty clearly as the first difference....

*shrug* I don't see the advantage of failing the test explicitly vs.
breaking it via output, but if you like that better I don't really mind,
I can change it to do that.

-Eric

> Cheers,
> 
> Dave.
> 

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs