[dtrace-discuss] issue dtracing function return value in pid provider

[Andrea Cucciarre']

Hello,

I need to use the pid provider to trace the inputs and return value of 
the following function


calc_guest_util(rs_guest_util_t *gutil, int ncpu, int fcpu)
  {
snip
   return ((100.0 * (double)used_cycles) / (double)(total_cycles));
  }

I can trace the input arguments with the following dscript (if there's a 
quicker and cleaver dscript to do that please correct me):


#!/usr/sbin/dtrace -qs

typedef struct rs_guest_util {
uint64_tlifespan;
uint64_twallclock_delta;
uint64_tactive_delta;
uint64_tstopped_cycles;
uint64_tyielded_cycles;
 } rs_guest_util_t;

pid$1::calc_guest_util:entry
{
self-guest=((rs_guest_util_t *)(copyin(arg0,sizeof(rs_guest_util_t;
printf(arg0 %x\n,arg0);
printf(ncpu %d\n,arg1);
printf(fcup %d\n,arg2);
printf(active delta %d\n,self-guest-active_delta);
printf(stopped cycles %d\n,self-guest-stopped_cycles);
printf(yielded cycles %d\n,self-guest-yielded_cycles);
}

Then I added the following to trace the return value of the function

pid$1::calc_guest_util:return
{
printf(return %x\n,arg1);
}

But I get the following output:

arg0 ffbffb80
ncpu 8
fcup 0
active delta 583970560
stopped cycles 0
yielded cycles 4652066048
total cycles 4671764480
return ffbffb80

As you can see the return value is an address, actually the same address 
as arg0 in entry.
I suspect I'm doing something wrong in dtrace but I can't realize 
whatCould you please advice?


Regards
Andrea


/
/
___
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org

Re: [dtrace-discuss] issue dtracing function return value in pid provider

[Adam Leventhal]

Hey Andrea,

 As you can see the return value is an address, actually the same address as
 arg0 in entry.
 I suspect I'm doing something wrong in dtrace but I can't realize
 whatCould you please advice?

The return type of calc_guest_util() was cut off, but I infer that
it's a double. DTrace doesn't have much by way of support for
floating-point values -- adding access to the floating-point registers
and support for manipulating floating-point datatypes have long been
RFEs for DTrace. I'd offer to add them to illumos, but I doubt that
would be of much help to users of Oracle Solaris.

Adam

On Sun, Jun 12, 2011 at 4:04 AM, Andrea Cucciarre'
andrea.cuccia...@oracle.com wrote:
 Hello,

 I need to use the pid provider to trace the inputs and return value of the
 following function

 calc_guest_util(rs_guest_util_t *gutil, int ncpu, int fcpu)
   {
    snip
    return ((100.0 * (double)used_cycles) / (double)(total_cycles));
   }

 I can trace the input arguments with the following dscript (if there's a
 quicker and cleaver dscript to do that please correct me):

 #!/usr/sbin/dtrace -qs

 typedef struct rs_guest_util {
     uint64_t    lifespan;
     uint64_t    wallclock_delta;
     uint64_t    active_delta;
     uint64_t    stopped_cycles;
     uint64_t    yielded_cycles;
  } rs_guest_util_t;

 pid$1::calc_guest_util:entry
 {
 self-guest=((rs_guest_util_t *)(copyin(arg0,sizeof(rs_guest_util_t;
 printf(arg0 %x\n,arg0);
 printf(ncpu %d\n,arg1);
 printf(fcup %d\n,arg2);
 printf(active delta %d\n,self-guest-active_delta);
 printf(stopped cycles %d\n,self-guest-stopped_cycles);
 printf(yielded cycles %d\n,self-guest-yielded_cycles);
 }

 Then I added the following to trace the return value of the function

 pid$1::calc_guest_util:return
 {
 printf(return %x\n,arg1);
 }

 But I get the following output:

 arg0 ffbffb80
 ncpu 8
 fcup 0
 active delta 583970560
 stopped cycles 0
 yielded cycles 4652066048
 total cycles 4671764480
 return ffbffb80

 As you can see the return value is an address, actually the same address as
 arg0 in entry.
 I suspect I'm doing something wrong in dtrace but I can't realize
 whatCould you please advice?

 Regards
 Andrea




 ___
 dtrace-discuss mailing list
 dtrace-discuss@opensolaris.org




-- 
Adam Leventhal, Delphix
http://dtrace.org/blogs/ahl

275 Middlefield Road, Suite 50
Menlo Park, CA 94025
http://www.delphix.com
___
dtrace-discuss mailing list
dtrace-discuss@opensolaris.org