dtrace discuss - Jun 2008 - DTrace double probe firing on MacOS X with -dead_strip

G''Day,

This issue was news to me, and may be useful for readers of this list.

In brief:

USDT providers on MacOS X that are built with the -dead_strip linker option
have their probes fired twice by mistake.  The workaround is not to use
the -dead_strip option.

Some more details:

After integration of the JavaScript DTrace provider into Firefox, the
Mozilla developers discovered that probes were firing twice by mistake -
so results such as @[probename]=count() were doubled.

The most likely reason was that probe placement had become incorrect -
during provider development the Firefox code was changing, and I had to
tweak the probe locations a few times.  It sounded like more of the same,
but I couldn''t see why this was happening in the code.

The Mozilla developers discovered it happened with the -dead_strip linker
option!:

	https://bugzilla.mozilla.org/show_bug.cgi?id=403132

which from a quick search looks like it is a known issue:

	http://www.opensource.apple.com/darwinsource/10.5/ld64-77/unit-tests/test-cases/dtrace-static-probes/Makefile

as -dead_strip is part of the Apple sanity checks.

That''s all I know.  If you know more details, please reply. :-)  I also
don''t know if this was fixed in the recent release of MacOS X (if it is
even considered a bug?).

cheers,

Brendan

-- 
Brendan
[CA, USA]

On Jun 11, 2008, at 1:22 PM, Brendan Gregg - Sun Microsystems wrote:
> G''Day,
>
> This issue was news to me, and may be useful for readers of this list.
>
> In brief:
>
> USDT providers on MacOS X that are built with the -dead_strip linker  
> option
> have their probes fired twice by mistake.  The workaround is not to  
> use
> the -dead_strip option.
>
> Some more details:
>
> After integration of the JavaScript DTrace provider into Firefox, the
> Mozilla developers discovered that probes were firing twice by  
> mistake -
> so results such as @[probename]=count() were doubled.
>
> The most likely reason was that probe placement had become incorrect -
> during provider development the Firefox code was changing, and I had  
> to
> tweak the probe locations a few times.  It sounded like more of the  
> same,
> but I couldn''t see why this was happening in the code.
>
> The Mozilla developers discovered it happened with the -dead_strip  
> linker
> option!:
>
> 	https://bugzilla.mozilla.org/show_bug.cgi?id=403132
>
> which from a quick search looks like it is a known issue:
>
> 
http://www.opensource.apple.com/darwinsource/10.5/ld64-77/unit-tests/test-cases/dtrace-static-probes/Makefile
>
> as -dead_strip is part of the Apple sanity checks.
>
> That''s all I know.  If you know more details, please reply. :-)  I
> also
> don''t know if this was fixed in the recent release of MacOS X (if
it
> is
> even considered a bug?).
Brendan,

	This is <rdar://problem/5628149>.

	OS X 10.5.3 does not have the fix, but if you download the free  
iPhone SDK, it has a
patched linker which fixes this issue.

	James M