On Fri, 21 Nov 2014, Matthew Wilcox wrote: > We could further optimise it by knowing that we need 0 pages if the I/O > is <= 8k in size: I think that'll break if the first page has a non-zero offset, so we'd have to allow the overestimate if we want to keep this function simple. We could micro-optimize to return 0 if it's <= 4k, though.