Instead of allocating one big chunk of DMA-coherent
memory for the firmware and keeping it around, only
vmalloc() the firmware and copy it into a single
page of DMA-coherent memory for the upload.

The advantage is that we don't need DMA memory for
the firmware image that is stored while the driver
is operating, we only need it while uploading.

This will make it easier for the driver to work if
the system has fragmented memory.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>