vm_object: drop reverse walk from collect_flush

[dougm]

Summary

In vm_object_page_collect_flush, following a forward walk from page p seeking more pages to which vm_object_page_remove_write might be applied, a backward walk from page p is attempted. It never finds a page, and so this change removes that walk.

It never finds a page because collect_flush() is only called from one place, in vm_object_page_clean(). If there was a page p-1 to be found, then either it is less than tstart, or it would have been found in the previous iteration. If it less than tstart, then page_clean() shouldn't be messing with it. If it was found in the previous iteration, then vm_object_page_collect_flush() would have been called in that iteration, and page p would have been the first page in the forward walk from p-1, and it would have been processed then, and so p would not be passed to vm_object_page_collect_flush now.

Test Plan

I ran some buildworlds and counted the number of 'prev' pages that were processed, finding that there were none.

[markj]

I think this is ok. I might also try to signal that vm_object_page_collect_flush() is a helper function for vm_object_page_clean(), e.g., by moving it above the latter and/or renaming it to vm_object_page_clean_flush(), the notion being that the reverse walk makes some sense in a more general-purpose helper function, but not if it's intended to be called only from vm_object_page_clean().