hotspot dtrace support in soylatte

[Michael Allman]

K.  After reading up some more, including some rtfm, I believe the 
essential problem with porting the solaris hotspot dtrace probes to 
leopard is that they use a "basic" syntax (using the DTRACE_PROBE() 
macros) that dtrace on leopard does not support.  More more info, see the 
thread on the missing "-G" option starting here:

http://mail.opensolaris.org/pipermail/dtrace-discuss/2007-October/004845.html

and continuing here:

http://mail.opensolaris.org/pipermail/dtrace-discuss/2007-November/004912.html

So it seems that before the hotspot dtrace probes can be ported to 
leopard, they must be converted to use the newer syntax with "dtrace -h" 
as described here

http://blogs.sun.com/ahl/entry/user_land_tracing_gets_better

Does anyone know if this project is underway somewhere else?

Cheers,

Michael

On Fri, 11 Jan 2008, Michael Allman wrote:

> Hello All,
>
> After working on this problem for a few weeks, I was hoping this could be a 
> great big happy announcement that I've ported the Solaris Hotspot dtrace 
> probes to Soylatte on Leopard.  However, I have to admit I am stuck and at a 
> loss for how to proceed.  If this were a Java problem, I'd have it solved by 
> now.  But frankly, at this level of C/C++ I don't know WTF I'm doing, and 
> what I have done is more like a half blind merge of code from os/solaris to 
> os/bsd.  So apologies for coming here with little to offer.
>
> FWIW, this is the error I get where I'm stuck:
>
> Compiling dtrace.d
> /usr/sbin/dtrace  -C -I. -h -o dtrace.o -s dtrace.d \
>     classLoadingService.o hashtable.o java.o jni.o memoryManager.o 
> objectMonitor_bsd.o runtimeService.o sharedRuntime.o synchronizer.o thread.o 
> vmCMSOperations.o vmGCOperations.o ciEnv.o compileBroker.o nmethod.o  ||\
>  STATUS=$?;\
> 	if [ x"$STATUS" = x"1" -a \
>       x`uname -r` = x"5.10" -a \
>       x`uname -p` = x"sparc" ]; then\
>    echo 
> "*****************************************************************";\
>    echo "* If you are building server compiler, and the error message is ";\
>    echo "* \"incorrect ELF machine type...\", you have run into solaris bug 
> ";\
>    echo "* 6213962, \"dtrace -G doesn't work on sparcv8+ object files\".";\
>    echo "* Either patch/upgrade your system (>= S10u1_15), or set the ";\
>    echo "* environment variable HOTSPOT_DISABLE_DTRACE_PROBES to disable ";\
>    echo "* dtrace probes for this build.";\
>    echo 
> "*****************************************************************";\
>  fi;\
>  exit $STATUS
> dtrace: failed to compile script dtrace.d: line 963: extraneous argument 
> 'nmethod.o' ($15 is not referenced)
>
> I had to hack quite a few things just to get this far.  I can probably 
> provide a patch against Landon Fuller's branch at 
> http://hg.bikemonkey.org/javasrc_1_6_jrl_darwin/ to anyone who's seriously 
> interested, but I think that someone with the right skills and motivation 
> could probably reproduce my work --- and more to the point, do it correctly 
> --- on their own in a matter of hours.
>
> Perhaps someone with dtrace expertise can explain the problem here.  I fear 
> it's a limitation of Leopard's port of DTrace.  But anyway, what would be 
> extra super would be someone who knows WTF they're doing taking a look at 
> soylatte and dtrace on Leopard and letting me know if this port has any 
> chance of working.
>
> Thanks.
>
> Michael
>