The drivers were popping the slave request FIFO, whether it was empty or not.
This seemed safe since empty FIFOs do nothing when popped. However, if the
FIFO receives an element between the empty check and the pop, we might pop
an element without ever seeing it. Under load this led to lost interrupts.

Trying to manage an external repository that was optionally built
or not depending on the kernel version was a nightmare.

Including it lets us use a single Makefile that does the right thing.

        against the vme in-tree  kernle module, it's not ready though

        fixing merge conflict

        address

        The chip tundra doesn't support 8 bit data width transfer.
        The wb select in the VME slave is written from wb_byteenable,
        who calculates it. The write and read is always 32 bit.
        Replaced swapbe32 by the macro cpu_to_be32.

better if it looks like one

the IRQ subscription is done using the VME driver
handling of the IRQ, request and response is done using
the etherbone library

if need be

since KBUILD_EXTMOD and KBUILD_EXTRA_SYMBOLS compilations
options don't work, I have create the one manually,
it seems to work

	Now it uses three windows:
	1) CR/CSR
	2) WB Control, 24A access
	3) WB Data,    32A access

	  clones from  the repo

It allows to use from a vme master eb commands (e.g. eb-ls) over the vme interface.
It follows the philosophy of pci-wb. So far only reading eb commands are supported.
The development is in a test phase, for that reason there are hardcode values and paths.

Main files:
vme_wb.*: register the kernel module and creates two VME windows, the first for the
          setting of the CSR/CR parameters in the vme slave, and second for data
          communication.
Module.symvers: it is needed for the correct compilation of the module sice it uses functions
                from wishbone.ko and vmebus.ko.
scripts for inserting and removing the module.