dtrace discuss - Sep 2005 - Probe firing question

Hi !

While playing with DTrace to find out why xterm -C doesn''t work on
Solaris
I noticed an ioctl() call that fails with EINVAL and I was curious to see 
where exactly that happens.

So I ran
       dtrace -c xterm -C -s /tmp/tr.d  | less 
where tr.d is the following script:

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

#pragma D option flowindent
#pragma D option quiet

syscall::ioctl:entry
/ curpsinfo->pr_pid == $target /
{
  self->spec = speculation();
}

syscall::ioctl:return
/self->spec && errno == EINVAL/
{
   commit(self->spec);
   self->spec = 0;
}

syscall::ioctl:return
/self->spec && errno != EINVAL/
{
   discard(self->spec);
   self->spec = 0;
}

fbt:::entry
/self->spec/
{
   speculate(self->spec);
   printf(" %s:%s \n",  probeprov, probefunc);

}

fbt:::return
/self->spec/
{
   speculate(self->spec);
   printf(" %s:%s - %d \n",  probeprov, probefunc, arg1);
}
----------

In the output I get:
----------
 0    -> fifo_ioctl                           fbt:fifo_ioctl
  0    <- fifo_ioctl                           fbt:fifo_ioctl - 3298589757952
  0    -> fifo_fastioctl                       fbt:fifo_fastioctl
  0    <- fifo_fastioctl                       fbt:fifo_fastioctl - 22
  0    -> releasef                             fbt:releasef
  0      -> clear_active_fd                    fbt:clear_active_fd
  0      <- clear_active_fd                    fbt:clear_active_fd - 1
----------

The output shows fifo_ioctl() returning before fifo_fastioctl(). Looking at 
the code in
http://cvs.opensolaris.org/source/xref/usr/src/uts/common/fs/fifofs/fifovnops.c
I see

----------
   1017 static int
   1018 fifo_ioctl(vnode_t *vp, int cmd, intptr_t arg, int mode,
   1019 	cred_t *cr, int *rvalp)
   1020 {
                  ....
   1027 	return ((VTOF(vp)->fn_flag & FIFOFAST) ?
   1028 		fifo_fastioctl(vp, cmd, arg, mode, cr, rvalp) :
   1029 		fifo_strioctl(vp, cmd, arg, mode, cr, rvalp));
   1030 }
   1031 
----------

How can fifo_ioctl() return before fifo_fastioctl() ?

Vlad.
This message posted from opensolaris.org

This is because of tail call optimisation. 
Your not the first to be surprised by this.

Basically fifo_ioctl is done with whatever it has to do.
The return code is the one that is about to come out of fifo_fastioctl();
so it arranges to call that function without leaving it''s own trace.
fifo_fastioctl() will then return into the caller of fifo_ioctl().

HTH

-r


Vlad Grama writes:
 > Hi !
 > 
 > While playing with DTrace to find out why xterm -C doesn''t work
on Solaris
 > I noticed an ioctl() call that fails with EINVAL and I was curious to see 
 > where exactly that happens.
 > 
 > So I ran
 >        dtrace -c xterm -C -s /tmp/tr.d  | less 
 > where tr.d is the following script:
 > 
 > ----------
 > #!/usr/sbin/dtrace
 > 
 > #pragma D option flowindent
 > #pragma D option quiet
 > 
 > syscall::ioctl:entry
 > / curpsinfo->pr_pid == $target /
 > {
 >   self->spec = speculation();
 > }
 > 
 > syscall::ioctl:return
 > /self->spec && errno == EINVAL/
 > {
 >    commit(self->spec);
 >    self->spec = 0;
 > }
 > 
 > syscall::ioctl:return
 > /self->spec && errno != EINVAL/
 > {
 >    discard(self->spec);
 >    self->spec = 0;
 > }
 > 
 > fbt:::entry
 > /self->spec/
 > {
 >    speculate(self->spec);
 >    printf(" %s:%s \n",  probeprov, probefunc);
 > 
 > }
 > 
 > fbt:::return
 > /self->spec/
 > {
 >    speculate(self->spec);
 >    printf(" %s:%s - %d \n",  probeprov, probefunc, arg1);
 > }
 > ----------
 > 
 > In the output I get:
 > ----------
 >  0    -> fifo_ioctl                           fbt:fifo_ioctl
 >   0    <- fifo_ioctl                           fbt:fifo_ioctl -
3298589757952
 >   0    -> fifo_fastioctl                       fbt:fifo_fastioctl
 >   0    <- fifo_fastioctl                       fbt:fifo_fastioctl - 22
 >   0    -> releasef                             fbt:releasef
 >   0      -> clear_active_fd                    fbt:clear_active_fd
 >   0      <- clear_active_fd                    fbt:clear_active_fd - 1
 > ----------
 > 
 > The output shows fifo_ioctl() returning before fifo_fastioctl(). Looking
at
 > the code in
http://cvs.opensolaris.org/source/xref/usr/src/uts/common/fs/fifofs/fifovnops.c
 > I see
 > 
 > ----------
 >    1017 static int
 >    1018 fifo_ioctl(vnode_t *vp, int cmd, intptr_t arg, int mode,
 >    1019 	cred_t *cr, int *rvalp)
 >    1020 {
 >                   ....
 >    1027 	return ((VTOF(vp)->fn_flag & FIFOFAST) ?
 >    1028 		fifo_fastioctl(vp, cmd, arg, mode, cr, rvalp) :
 >    1029 		fifo_strioctl(vp, cmd, arg, mode, cr, rvalp));
 >    1030 }
 >    1031 
 > ----------
 > 
 > How can fifo_ioctl() return before fifo_fastioctl() ?
 > 
 > Vlad.
 > This message posted from opensolaris.org
 > _______________________________________________
 > dtrace-discuss mailing list
 > dtrace-discuss at opensolaris.org