video: tegra: nvmap: treat writecombine requests as dmacoherent
NvMap always allocates DRAM or IRAM memory. It never allocates real
Device memory. From Arm v8 onwards writecombine memory is real
writecombine, which is of less performace for DRAM accesses and
unnecessary for DRAM.
Treat writecombine requests as dmacoherent, which is normal memory
non-cacheable.