http://www.cisco.com/en/US/tech/tk827/tk831/tk102/tech_protocol_home.html
But I think the best explanation of CEF available is the one in Inside Cisco IOS Software Architecture, on Cisco Press. That book is being rewritten currently, by the way, to include more information and different platforms. It'll take about a year, I think, for it to hit the market.
Now, to the specific question at hand: The fast cache would build out a /32 entry under numerous circumstances:
1. If there are multiple paths to the same destination installed in the routing table.
10.1.2.0/24 is reachable via x
10.1.2.0/24 is reachable via y
2. If the subnet is directly connected, the fast cache will build out /32's for each host which it actually switches traffic to within the subnet.
10.1.1.0/24 is connected....
3. If you had a /32 contained within a given section of the tree (think block of ip addresses), every adress within that block would be built as a /32. So, for instance, if you had this:
10.1.3.0/24 is subnetted
10.1.3.1/32 is reachable via x
10.1.3.0/24 is reachable via y
Actually, this rule is a little more confusing than the fast cache building out a /32, it would build out to the longest length subnet mask within the block.
CEF still builds out /32's in the case of directly connected hosts, as in the case of 10.1.1.0/24 is directly connected. CEF also continues to build out to the longest length within any given block, so it would still build /32's for the last case (and even here, the rules are different, resulting in fewer CEF table entries than you would have fast cache entries, and there are some other differences). But CEF doesn't build out to /32's for the first case, load sharing.
Hope that helps...
:-)
Russ.W
