dtrace discuss - Sep 2009 - dtrace predicat

Hi,

I''m a nub in dtrace, could you please help me to find the mistake?

I need to implement the logic:

- if PID and TID are defined, fire the probe for the particular thread of
particular process;
- if only PID is defined, fire the probe for the process including all threads.

The predicat  [i]syscall:::entry / (pid == PID && tid == TID )[/i] /
works ok and return syscall for PID/TID process only.

Then I set a predicat:

[i] syscall:::entry
 /(pid == PID && tid == TID) ||
  pid == PID/[/i]
 
and it works for the whole process, as if I set only / pid == PID/.
What am I missing? 
Thanks in advance.
-- 
This message posted from opensolaris.org

On Thu, Sep 17, 2009 at 9:17 AM, Eugenias <redtigrareg at gmail.com>
wrote:
> Hi,
>
> I''m a nub in dtrace, could you please help me to find the mistake?
>
> I need to implement the logic:
>
> - if PID and TID are defined, fire the probe for the particular thread of
particular process;
> - if only PID is defined, fire the probe for the process including all
threads.
>
> The predicat ?[i]syscall:::entry / (pid == PID && tid == TID )[/i]
/ works ok and return syscall for PID/TID process only.
>
> Then I set a predicat:
>
> [i] syscall:::entry
> ?/(pid == PID && tid == TID) ||
> ?pid == PID/[/i]
>
> and it works for the whole process, as if I set only / pid == PID/.
> What am I missing?
I''m a little bit confused about what you want to do.  If you want to
perform one set of actions if the tid matches and a different set of
actions if it doesn''t, you want this:

syscall:::entry
/ pid == PID && tid == TID /
{
    [ one set of actions ]
}

syscall:::entry
/ pid == PID && tid != TID /
{
    [ a different set of actions ]
}

If you want to perform one set of actions no matter the tid and an
additional set of actions if the tid matches, you want this:

syscall:::entry
/ pid == PID /
{
    [ one set of actions ]
}

syscall:::entry
/ pid == PID && tid == TID /
{
    [ additional set of actions ]
}

(It''s probably worth noting that "/ (pid == PID && tid ==
TID) || pid
== PID /" is logically equivalent to "/ pid == PID /".)

Chad

Thanks a lot! You gave me an idea I needed.
I tried to enhance the dtruss tool (http://www.brendangregg.com/DTrace/dtruss)
to be possible to monitor the particular thread of the process.

The final predicat looks in that way:

 syscall:::entry
  /(OPT_command && pid == $target) ||
  (OPT_tid && (pid == PID && tid == TID)) ||
  ((OPT_pid && pid == PID) && OPT_tid == 0) ||
  (OPT_name && execname == NAME) ||
  (self->child)/

So now the command like:
./dtruss.d -a -l -p 5506 -w 28

Returns me syscolls for the pid=5506, tid=28

THANKS A LOT!
-- 
This message posted from opensolaris.org

Thanks for spelling that out what you were working with.

As a matter of style, splitting the test logic across probe declarations sure
would be easier to read (and reread).

All that opt_* test logic in one predicate requires the maintainer to understand
the preamble. For example, when is OPT_command set? When opt_pid and opt_name
are not, which means a command should be given instead (otherwise the help
option triggers). If there is a command, it is read from $* which, judging by
the documented comment may throw off a reader from time to time.

While I trust there isn''t a substantively cleaner way to add so many
options, in this case I would opt for separating concerns and choosing clarity
of expression over economy.

Just a US$0.02 observation based on the time it took me to unravel your
solution...

Regards,

Michael
-- 
This message posted from opensolaris.org