We reserve the 3.75GB - 4GB region of PCIe address space for device to
device transfers, making the corresponding physical memory under
direct mapping unavailable for DMA.

To allow for PCIe DMA to all physical memory we map this chunk of
physical memory with BAR1.  Because of the resulting discontinuity in
the mapping function, we remove a page of memory at each end of the
range so multi-page DMA buffers can never be allocated that span the
range.
Signed-off-by: David Daney <ddaney@caviumnetworks.com>
To: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/1535/Signed-off-by: Ralf Baechle <ralf@linux-mips.org>