The following algorithm for computing the number of nodes in a linked list appears in this thesis on common algorithms in the PRAM model for parallel computation.
The author then argues that this is a $log(n)$ time algorithm (where $n$ is the number of nodes in the list) because the span of nodes covered by the "far" pointer doubles on every iteration of the innermost loop. What I don't understand is why the author doesn't seem to factor in the time required for the first two lines of code after the begin statement, where a processor is assigned to each node in the list. Assuming the input to the algorithm is just a pointer to the header node to the linked list, wouldn't the assignment of processors require traversing each node in the linked list and thus take time linear in $n$? After all, each node in the linked list may be stored in some highly non-contiguous locations in memory identifiable only by following the chain of pointers from predecessor nodes in the list one at a time.