>
>I think what we need here is to fix the return value from the specific
>handler ... i.e. return EJUSTRETURN if additional sense information should
>NOT be printed (i.e. "it should just return").  Otherwise, the generic
>handler does print the sense info.
>

This is already what FreeBSD does, sort of, but it really isn't
any different from the current semantics.

We're handling 2 things here: error recovery actions and error
printout actions. If you have a device specific handler, you
have it handle error recovery actions (and printouts) fully,
or it cannot worry terribly much if it returns in a fashion
such that the generic handler takes over. It does not, in
this framework, have to be more complicated than this.