Fix some bugs in device-mapper handling of 64-bit values, replacing
dm_div_up() and dm_round_up() inlines with macros and removing some
avoidable divisions of 64-bit numbers.

The mirror region size is the granularity used to manage and monitor the
data copying, typically 512KB, so 32 bits (of sectors) should be plenty to
hold this.

Taken together with the two earlier patches:
  "fix TB stripe data corruption" (a missing cast) and
  "stripe_width should be sector_t",
I've now had 3 independent reports that this fixes device-mapper
for devices with large stripes.

Still awaiting test results for dm-crypt and dm-raid1.
Signed-Off-By: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>