> There is a fair case to be made that kernels should be compiled with
> -fno-builtin, even if the sources are "fixed" to avoid namespace
> collisions.
> 
> for one, inlining them skews can profiling results... 

True, but most users aren't running profiled kernels.  Those who do
can add -no-builtin (or not, as either can be justified based on 
what you're trying to measure).

> additionally, the compiler's prototypes do not necessarily match the
> kernel's prototypes, and that may not be an error. 

It is an error. 

I understand that Technical Corrigendum 2, an official amendment to
the C standard, will state that all identifiers reserved by a hosted
implementation are also reserved for a freestanding implementation.
Thus is a freestanding implementation is going to have a str* or mem*
functions, they have to be identical to those found in the C library.

	--jtc