dtrace discuss - Oct 2007 - how to get "libmtmalloc.so.1:malloc" returned address in DTrace script

Should I use MDB to see why the return value was not carried in arg1? But how? 
Would some experts give me help or advice with this?  Thanks!

BR
Jian


--
This message posted from opensolaris.org

On Mon, Oct 15, 2007 at 07:39:16AM -0700, LiJIan wrote:
> Should I use MDB to see why the return value was not carried in arg1? But
how?
> Would some experts give me help or advice with this?  Thanks!
Unfortunately, libmtmalloc''s malloc function can make a tail-call to
either
oversize() or malloc_internal() -- both internal functions. This means that
the apparent return value in the return probe isn''t actually the value
returned from malloc(). To accurately get the return value, you''ll need
to
get the return value from those functions.

This is a tricky issue. One could imagine a class of function return probes
which would indicate if there was a normal return or a tail-call, but we
decided that that was too complicated. Perhaps it would be a good idea to
create additional probes:

			args0			args1
    return-normal	address of return site	return value
    return-call		address of tail-call	address of tail-called function

Adam

-- 
Adam Leventhal, FishWorks                        http://blogs.sun.com/ahl

Thank you very much for your help, Adam.
Now I write another script to check memory leak issue as follows, it works well,
but during runing I see some error output on screen such as "dtrace: error
on enabled probe ID 3 (ID 49838:
pid6918:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdc78000)
in action #3".
Do you know why the error occur?  Does the error affect the output of the D
script, or I still can take use of output of the D script for memory leak
analysis ?
Thanks again.

-------------------
#!/usr/sbin/dtrace -s

pid$1:libmtmalloc.so.1:malloc:entry
{
 self->trace = 1;
 self->size = arg0;
}

pid$1:libmtmalloc.so.1:free:entry
/arg0 != 0/
{
 printf("\nfree Ptr=0x%p ", arg0);
}

pid$1::malloc_internal:return,
pid$1::oversize:entry
/self->size/
{
 printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
 ustack();
 system("date");
 self->size = 0;
 self->trace = 0;
}
---------------------


BRs
LiJian


--
This message posted from opensolaris.org

The error invalid address seems to me coming when the
pid$1::oversize:entry fires. It only has one argument but
you seem to be looking for arg1 - the second argument.

Also not sure how the script runs as you would need the
-w option for enabling destructive probes to be able to
run the system("date") action.

In fact a better way to print the date is by using the
walltimestamp. Using printf("Date: %Y\n", walltimestamp);

HTHs

Angelo

LiJIan wrote:
> Thank you very much for your help, Adam.
> Now I write another script to check memory leak issue as follows, it works
well, but during runing I see some error output on screen such as "dtrace:
error on enabled probe ID 3 (ID 49838:
pid6918:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdc78000)
in action #3".
> Do you know why the error occur?  Does the error affect the output of the D
script, or I still can take use of output of the D script for memory leak
analysis ?
> Thanks again.
> 
> -------------------
> #!/usr/sbin/dtrace -s
> 
> pid$1:libmtmalloc.so.1:malloc:entry
> {
>  self->trace = 1;
>  self->size = arg0;
> }
> 
> pid$1:libmtmalloc.so.1:free:entry
> /arg0 != 0/
> {
>  printf("\nfree Ptr=0x%p ", arg0);
> }
> 
> pid$1::malloc_internal:return,
> pid$1::oversize:entry
> /self->size/
> {
>  printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
>  ustack();
>  system("date");
>  self->size = 0;
>  self->trace = 0;
> }
> ---------------------
> 
> 
> BRs
> LiJian
> 
> 
> --
> This message posted from opensolaris.org
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org

On Wed, Oct 17, 2007 at 10:49:10PM -0700, LiJIan wrote:
> pid$1::malloc_internal:return,
> pid$1::oversize:entry
I assume you want the oversize:return probe rather than the entry probe.

Adam

-- 
Adam Leventhal, FishWorks                        http://blogs.sun.com/ahl

Hi, Adam
  Yes, it should be "oversize::return" instead of
"oversize::entry". I will change and then retry it.  Thank you.

BRs
LiJian


--
This message posted from opensolaris.org

Hi Adam,
  I changed the D script to as follows. 
 
-----------------------------------------

#!/usr/sbin/dtrace -s

pid$1:libmtmalloc.so.1:malloc:entry
{
 self->trace = 1;
 self->size = arg0;
}

pid$1:libmtmalloc.so.1:free:entry
/arg0 != 0/
{
 printf("\nfree Ptr=0x%p ", arg0);
}

pid$1:libmtmalloc.so.1:malloc_internal:return
/self->size/
{
 printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
 ustack();
 self->size = 0;
 self->trace = 0;
}

pid$1:libmtmalloc.so.1:oversize:return
/self->size/
{
  printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
  ustack();
  self->size = 0;
  self->trace = 0;
}

-----------------------------------------


But during running it, the error still occured,

-----------------------

# dtrace -qs ./mem2.d 13206
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fd572000)
in action #3
dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fd572000)
in action #3

---------------------

The address "0x1fdf72000" apparently is wrong, does
"malloc_internal:return" sometimes return a wrong address ? What is
the "action #3" ?

Thanks.
 

BRs 
LiJian


--
This message posted from opensolaris.org

Can you try just doing this:

pid$1:libmtmalloc.so.1:malloc_internal:return
{
	ustack();
}

Do you see errors? What does dtrace -V show?

- ahl

On Sun, Oct 28, 2007 at 08:40:44PM -0700, LiJIan wrote:
> Hi Adam,
>   I changed the D script to as follows. 
>  
> -----------------------------------------
> 
> #!/usr/sbin/dtrace -s
> 
> pid$1:libmtmalloc.so.1:malloc:entry
> {
>  self->trace = 1;
>  self->size = arg0;
> }
> 
> pid$1:libmtmalloc.so.1:free:entry
> /arg0 != 0/
> {
>  printf("\nfree Ptr=0x%p ", arg0);
> }
> 
> pid$1:libmtmalloc.so.1:malloc_internal:return
> /self->size/
> {
>  printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
>  ustack();
>  self->size = 0;
>  self->trace = 0;
> }
> 
> pid$1:libmtmalloc.so.1:oversize:return
> /self->size/
> {
>   printf("\nmalloc Ptr=0x%p Size=%d", arg1, self->size);
>   ustack();
>   self->size = 0;
>   self->trace = 0;
> }
> 
> -----------------------------------------
> 
> 
> But during running it, the error still occured,
> 
> -----------------------
> 
> # dtrace -qs ./mem2.d 13206
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fdf72000)
in action #3
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fd572000)
in action #3
> dtrace: error on enabled probe ID 3 (ID 49838:
pid13206:libmtmalloc.so.1:malloc_internal:return): invalid address (0x1fd572000)
in action #3
> 
> ---------------------
> 
> The address "0x1fdf72000" apparently is wrong, does
"malloc_internal:return" sometimes return a wrong address ? What is
the "action #3" ?
> 
> Thanks.
>  
> 
> BRs 
> LiJian
> 
> 
> --
> This message posted from opensolaris.org
> _______________________________________________
> dtrace-discuss mailing list
> dtrace-discuss at opensolaris.org
-- 
Adam Leventhal, FishWorks                        http://blogs.sun.com/ahl