The Linux Kernel code is a complicated piece of software. The more facilities gcc provides them with, they happier the coders would be.
Why would they care about portability? gcc compiles code under virtually every hardware configuration PLUS it provides them with good features. Why would they care if Linux could or could not be compiled with another compiler?
Today, portable code is such a common concept for us that we believe it should be existent everywhere. But that is not the case. For example, if a compiler provides real-time extensions to C, NASA would use it without care for portability. The important point being the features are too good to sacrifice for a portability that is never used (I mean, who would compile the kernel with MS Visual Studio for example?)
