dtrace discuss - Oct 2009 - unexpected stacks with __moddi3 from ustack()

I''m getting quite strange stacks from DTrace''s ustack(). Could
please anybody explain what happens and suggest a fix/workaround?

My D-script is a trivial sampling profiler, see attached sampler.d
Program being profiled is attached as primes.c

This is a piece of DTrace script output (on OpenSolaris x86):
  1  62906                   :profile-100hz 7563566252 1 1 4 0
              primes`__moddi3+0xc4

  1  62906                   :profile-100hz 7573563203 1 1 4 0
              primes`__moddi3+0x97

  1  62906                   :profile-100hz 7592475248 1 1 4 0
              primes`__moddi3+0x66

  1  62906                   :profile-100hz 7602469821 1 1 4 0
              primes`is_prime+0x33
              primes`main+0x5b
              primes`_start+0x80

Note those "primes`__moddi3" stacks, which consist of only one stack
frame. Where are "primes`is_prime" and "primes`main"?

BTW, changing "long long" to "long" fixes the problem. But
still it''s not clear why ustack() reports __moddi3 alone.
-- 
This message posted from opensolaris.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sampler.d
Type: application/octet-stream
Size: 156 bytes
Desc: not available
URL:
<http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20091026/810ca207/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: primes.c
Type: application/octet-stream
Size: 363 bytes
Desc: not available
URL:
<http://mail.opensolaris.org/pipermail/dtrace-discuss/attachments/20091026/810ca207/attachment-0001.obj>

On Mon, Oct 26, 2009 at 08:58:27AM -0700, Alexey Vladykin
wrote:
> I''m getting quite strange stacks from DTrace''s ustack().
Could please anybody explain what happens and suggest a fix/workaround?
> 
> My D-script is a trivial sampling profiler, see attached sampler.d
> Program being profiled is attached as primes.c
> 
> This is a piece of DTrace script output (on OpenSolaris x86):
>   1  62906                   :profile-100hz 7563566252 1 1 4 0
>               primes`__moddi3+0xc4
> 
>   1  62906                   :profile-100hz 7573563203 1 1 4 0
>               primes`__moddi3+0x97
> 
>   1  62906                   :profile-100hz 7592475248 1 1 4 0
>               primes`__moddi3+0x66
> 
>   1  62906                   :profile-100hz 7602469821 1 1 4 0
>               primes`is_prime+0x33
>               primes`main+0x5b
>               primes`_start+0x80
> 
> Note those "primes`__moddi3" stacks, which consist of only one
stack
> frame. Where are "primes`is_prime" and "primes`main"?
>
> BTW, changing "long long" to "long" fixes the problem.
But still it''s
> not clear why ustack() reports __moddi3 alone.
__moddi3 is a runtime support function to allow division of 64-bit integers.
It''s a leaf function (meaning it doesn''t establish its own
stack frame),
which makes it very difficult to get a stack trace from.

32-bit division doesn''t require such a function.

If you''re on a 64-bit system, compiling with -m64 will also make things
better.

Cheers,
- jonathan