dtracing static symbols

[Daniel O'Connor]

On 12 Mar 2014, at 15:15, Prashanth Kumar <pra_udupi at yahoo.co.in> wrote:
> If you run 
> # env DTRACE_DEBUG=1 dtrace -Ppid\$target -l  -c ./static
> you will notice that lot of probe creation will fail, also no probes are created for instruction offsets.
>    you will have to update the libproc library and fasttrap code to trace all the 
> functions. 

I don't really care about the function offsets, just static functions. 

Or are you suggesting updating libproc and the fasttrap code will allow that (as well as instruction offsets)?

THanks.

> --------------------------------------------
> On Wed, 12/3/14, Daniel O'Connor <doconnor at gsoft.com.au> wrote:
> 
> Subject: Re: dtracing static symbols
> To: "Robert Mustacchi" <rm at joyent.com>
> Cc: freebsd-dtrace at freebsd.org
> Date: Wednesday, 12 March, 2014, 2:54 AM
> 
> 
> On 12 Mar 2014, at 2:30, Robert Mustacchi <rm at joyent.com>
> wrote:
>> On 03/10/2014 10:34 PM, Daniel O'Connor wrote:
>>> 
>>> On 11 Mar 2014, at 15:34, Prashanth Kumar <pra_udupi at yahoo.co.in>
> wrote:
>>>> If the binary being traced has static symbols
> in its symbol table, DTrace should
>>>> be able to trace the function. Can you describe
> the example where you found this
>>>> difference in FreeBSD and OSX?
>>> 
>>> Unfortunately the static symbols don't show up in
> the symbol table (as shown by nm).
>>> 
>>> Is there a compile or link flag which will change
> that?
>> 
>> Because it's a static function the compiler may inline
> it, which may be
>> why you don't actually see an entry in nm nor that it
> can be found by
>> DTrace. You'll want to look at the disassembled output
> of your program
>> to see if it was inlined. Different compilers can and
> will do different
>> things. There generally are flags you can pass to the
> compiler to tell
>> it not to inline it, but that's compiler specific.
> 
> I just realised that my test contradicted the statement I
> made earlier..
> However I checked my test program (static.c) and it the
> functions definitely appear in the symbol table.
> [mdtest 21:13] ~ >nm static|egrep '(foo|bar)'
> 0000000000400600 T bar
> 0000000000400620 t foo
> 
> I also added the noinline attribute for good measure.
> 
> It seems that _nothing_ shows up for executables, only
> shared libraries, this is OK for me since my code resides in
> a library but it is a bit surprised nonetheless..
> 
>>> (I'm not sure what the various numbers mean)
>> 
>> The pid provider can instrument any instruction in a
> function, those are
>> the instruction offsets.
> 
> Ahh, thanks.
> 
> --
> Daniel O'Connor software and network engineer
> for Genesis Software - http://www.gsoft.com.au
> "The nice thing about standards is that there
> are so many of them to choose from."
>   -- Andrew Tanenbaum
> GPG Fingerprint - 5596 B766 97C0 0E94 4347 295E E593 DC20
> 7B3F CE8C
> 
> 
> 
> 
> 
> 
> 
> 

--
Daniel O'Connor software and network engineer
for Genesis Software - http://www.gsoft.com.au
"The nice thing about standards is that there
are so many of them to choose from."
  -- Andrew Tanenbaum
GPG Fingerprint - 5596 B766 97C0 0E94 4347 295E E593 DC20 7B3F CE8C






-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.freebsd.org/pipermail/freebsd-dtrace/attachments/20140312/6be260d6/attachment.sig>