Index: sys/amd64/amd64/pmap.c
===================================================================
--- sys/amd64/amd64/pmap.c
+++ sys/amd64/amd64/pmap.c
@@ -3785,6 +3785,19 @@
  * one or two pages may be held during the wait, only to be released
  * afterwards.  This conservative approach is easily argued to avoid
  * race conditions.
+ *
+ * Page table entry at address va page index is defined as follows:
+ * - for page table (last level), ptepindex = pmap_pde_pindex(va) =
+ *   = va >> PDRSHIFT, in other words, it is just the index of the PDE.
+ * - for page directory page, ptepindex = NUPDE (number of userland PD
+ *   entries) + (pmap_pde_index(va) >> NPDEPGSHIFT)
+ *   i.e. index of PDPE is put after the last index of PDE,
+ * - for page directory pointer page, ptepindex = NUPDE + NUPDPE +
+ *   (pmap_pde_index(va) >> (NPDEPGSHIFT + NPML4EPGSHIFT),
+ *   i.e. index of pml4e is put after the last index of PDPE.
+ * In other words, is it sequential number of the corresponding paging entry
+ * in the order where all entries of the same rank are put together, then
+ * ranks are put from deepest to root.
  */
 static vm_page_t
 _pmap_allocpte(pmap_t pmap, vm_pindex_t ptepindex, struct rwlock **lockp)