dtrace discuss - Aug 2005 - Detecting the presence of an active probe?

Is there a way in the kernel to tell if a specific static probe is currently
enabled?

It would seem useful to have a method to predict if a static kernel probe
might fire or not in order to avoid pre-processing data for the probe.
Sure, there''s a window where the probe is enabled after the check and
it fires even though the pre-processing hasn''t been done, but that
would
happen if we were gathering data from multiple probes anyway.

Thanks,

    Rich

> Is there a way in the kernel to tell if a specific static probe is
currently
> enabled?
Someone may well suggest something cleaner/smarter but you can
check if a probe is enabled using mdb to see if there any ECBs
for this probe. Taking a single nfs SDT probe as an example:

# dtrace -l -n sdt:nfs:nfs4renew:nfs4-renew-start
   ID   PROVIDER            MODULE                          FUNCTION NAME
 1260        sdt               nfs                         nfs4renew
nfs4-renew-start

Firstly, without an enabling:

# mdb -k
Loading modules: [ unix krtld genunix specfs dtrace ufs ip sctp usba
uhci s1394 random fctl nca lofs audiosup md nfs sppp ptm ipc logindmux
]
> 0t1260::id2probe | ::print dtrace_probe_t dtpr_ecb
dtpr_ecb = 0


Then with this probe enabled:
> 0t1260::id2probe | ::print dtrace_probe_t dtpr_ecb
dtpr_ecb = 0xffffffff85bd2de8


Cheers.

Jon.
> It would seem useful to have a method to predict if a static kernel probe
> might fire or not in order to avoid pre-processing data for the probe.
> Sure, there''s a window where the probe is enabled after the check
and
> it fires even though the pre-processing hasn''t been done, but that
would
> happen if we were gathering data from multiple probes anyway.
> 
> Thanks,
> 
>     Rich
> 
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org

On Wed, Aug 17, 2005 at 10:49:57AM -0500, Rich Brown
wrote:
> Is there a way in the kernel to tell if a specific static probe is
currently
> enabled?
> 
> It would seem useful to have a method to predict if a static kernel probe
> might fire or not in order to avoid pre-processing data for the probe.
> Sure, there''s a window where the probe is enabled after the check
and
> it fires even though the pre-processing hasn''t been done, but that
would
> happen if we were gathering data from multiple probes anyway.
Generally, a better approach is to not do *any* processing of arguments;
just pass in the base pointers that can''t be derived in any other way,
and
let the D script do the processing if it needs to.

This is easier to do if your module has CTF data, of course...

Cheers,
- jonathan

-- 
Jonathan Adams, Solaris Kernel Development

Rich,

I''m posting this more for the amusement factor rather than
it being of any use to you...
> Is there a way in the kernel to tell if a specific static probe is
currently
> enabled?
You can actually use DTrace to figure this one out. Just pass
the probe id into the following script:


# cat enabled.d
BEGIN
{
	printf("Probe %d %s enabled\n", $1,
	    `dtrace_probes[$1 - 1]->dtpr_ecb == 0 ? "NOT" :
"IS");
	exit(0);
}


# dtrace -qs ./enabled.d 1259
Probe 1259 NOT enabled

<enable probe id 1259>

# dtrace -qs ./enabled.d 1259
Probe 1259 IS enabled


Cheers.

Jon.

(Sorry for the delay, thought I sent this on Wednesday and I was
out yesterday.)

Hi Jonathan,

Thanks for the advice.  In general, I agree that the D script should do
as much of the work as possible.  A few of us were kicking around some
ideas where using D to format the data was a bit clumsy (e.g., deciphering
an IPv6 address.  Not impossible, just clumsy).  Hence the original 
question.

Thanks,

    Rich


Jonathan Adams wrote:
>On Wed, Aug 17, 2005 at 10:49:57AM -0500, Rich Brown wrote:
>  
>
>>Is there a way in the kernel to tell if a specific static probe is
currently
>>enabled?
>>
>>It would seem useful to have a method to predict if a static kernel
probe
>>might fire or not in order to avoid pre-processing data for the probe.
>>Sure, there''s a window where the probe is enabled after the
check and
>>it fires even though the pre-processing hasn''t been done, but
that would
>>happen if we were gathering data from multiple probes anyway.
>>    
>>
>
>Generally, a better approach is to not do *any* processing of arguments;
>just pass in the base pointers that can''t be derived in any other
way, and
>let the D script do the processing if it needs to.
>
>This is easier to do if your module has CTF data, of course...
>
>Cheers,
>- jonathan
>
>  
>

Using Dtrace to observe itself?  I like it!  :-)

    Rich


Jonathan Haslam wrote:
>Rich,
>
>I''m posting this more for the amusement factor rather than
>it being of any use to you...
>
>  
>
>>Is there a way in the kernel to tell if a specific static probe is
currently
>>enabled?
>>    
>>
>
>You can actually use DTrace to figure this one out. Just pass
>the probe id into the following script:
>
>
># cat enabled.d
>BEGIN
>{
>	printf("Probe %d %s enabled\n", $1,
>	    `dtrace_probes[$1 - 1]->dtpr_ecb == 0 ? "NOT" :
"IS");
>	exit(0);
>}
>
>
># dtrace -qs ./enabled.d 1259
>Probe 1259 NOT enabled
>
><enable probe id 1259>
>
># dtrace -qs ./enabled.d 1259
>Probe 1259 IS enabled
>
>
>Cheers.
>
>Jon.
>
>  
>