cxgb3: map entire Rx page, feed map+offset to Rx ring.

DMA mapping can be expensive in the presence of iommus.
Reduce the Rx iommu activity by mapping an entire page, and provide the H/W
the mapped address + offset of the current page chunk.
Reserve bits at the end of the page to track mapping references, so the page
can be unmapped.

Signed-off-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>