kernel/linux: fix using custom location

Currently, extract and patch are skipped as thus:
- using a custom directory of pre-installed headers
- a correctly named directory already exists

Otherwise, extract and patch are done.

The current second condition is wrong, because it allows the following
sequence to happen:
- a non-custom kernel is used
- a previous build only partially extracted the non-custom sources
- that p[revious build broke during extraction (eg. incomplete tarball...)
- a subsequent build will find a properly named directory, and will
thus skip extract and patch, which is wrong

Fix that by following the conditions in this table:

Type | Extract | Patch
----------------------+---------+-------
Pre-installed headers | N | N
custom directory | N | N
custom tarball | Y | N
mainstream tarball | Y | Y

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: David Holsgrove <david.holsgrove@xilinx.com>

reporter_name="Anthony Foiani <anthony.foiani@gmail.com>"

reporter_url="http://sourceware.org/ml/crossgcc/2010-09/msg00100.html"

reporter_comment="This is a sample config file for Freescale e500v2 processors (e.g., MPC8548,

MPC8572). It uses eglibc (for e500/SPE patches) and a recent gcc (4.6.0,

for e500v2 DPFP support) and will generate appropriate dual-precision

floating point instructions by default.

Note: If building a Linux kernel with this toolchain, you will want to make

sure -mno-spe AND -mspe=no are passed to gcc to prevent SPE ABI/instructions

from getting into the kernel (which is currently unsupported). At this time,

the kernel build system properly passes those two options, but older kernels

were only passing -mno-spe by default."