While the term can refer to global variables, it is primarily used in the context of nested and anonymous functions where some variables can be in neither the local nor the global scope.
Therefore, x is non-local to inner: In the Haskell example that follows the variable c is non-local in the anonymous function \x -> x + c: Non-local variables are the primary reason it is difficult to support nested, anonymous, higher-order and thereby first-class functions in a programming language.
If the nested function or functions are (mutually) recursive, it becomes hard for the compiler to know exactly where on the call stack the non-local variable was allocated, as the frame pointer only points to the local variable of the nested function itself and there can be an arbitrary number of activation records on the stack in between.
This is generally solved using access links or display registers.
If the nested function is returned as a result from its outer function (or stored in a variable) the non-local variables will no longer be available on the stack.