Functions not getting picked up by dtrace

[Mark Johnston]

On Mon, Feb 29, 2016 at 02:07:45PM -0800, hiren panchasara wrote:
> On 02/29/16 at 01:49P, hiren panchasara wrote:
> > On 02/29/16 at 01:49P, Mark Johnston wrote:
> > > On Mon, Feb 29, 2016 at 01:39:33PM -0800, hiren panchasara wrote:
> > > > I've seen this earlier with others too but this one is the latest
> > > > confusing me: em_xmit() in $src/dev/e1000/if_em.c
> > > > 
> > > > Other functions with similar signature are listed in 'dtrace -l'.
> > > 
> > > em_xmit() is a static function with a single call site, so it's getting
> > > inlined into its caller. DTrace FBT can't instrument inlined functions.
> > > 
> > > > 
> > > > Is is because of some optimization? How do I undo it for testing
> > > > purposes?
> > > 
> > > There's a __noinline attribute that you can use for this.
> > 
> > Thanks, Mark.
> > 
> > cem on irc also explained the "why" part.
> 
> Davide pointed out that it _might_ be possible that compiler change the
> way it decides what to inline and non-static functions can also get
> hidden that way from dtrace.

Indeed: the compiler may inline a copy of a non-static function at call
sites in the same compilation unit as the definition of the function.
This case is more insidious since a fbt:::entry probe will exist for
the function, and the only way to determine whether the function was
inlined at a given call site is by examining the disassembly.