CONFIG_XXX vs CONFIG_HAVE_XXX ?

Hi,

What is the rationale behind having, for some feature XXX, both configuration
macros CONFIG_XXX and CONFIG_HAVE_XXX ?

For example, I'd love to use ftrace's new life-saving feature "funcgraph-retval"
(which instantly shows the root cause of an error returned from a fully graphed
function). Unfortunately, on all recent Debians the config says:

  CONFIG_HAVE_FUNCTION_GRAPH_RETVAL=y
  # CONFIG_FUNCTION_GRAPH_RETVAL is not set

As a result, the feature is not available. Okay, I'll just enable and recompile.
However, I'm wondering why in the first place such a situation is possible: what
is the intention there ? Either we want the feature, or we don't (e.g. because
it may incur some CPU or memory overhead), but why sit halfway in between ?

Thanks in advance,

-Alex