svn commit: r252192 - projects/bhyve_npt_pmap/sys/amd64/amd64

[Neel Natu]

Author: neel
Date: Tue Jun 25 03:57:27 2013
New Revision: 252192
URL: http://svnweb.freebsd.org/changeset/base/252192

Log:
  vtopte()/vtopde() can never be used to lookup the PTE/PDE associated with
  a guest physical address since the page tables pointed to by the nested pmap
  are never installed in the host %cr3.
  
  The recursive mapping used by these functions will end up using the page
  tables associated with the host thread that provides the execution context
  for the vcpu - which is clearly wrong!
  
  Since the user virtual address space is numerically identical to the
  guest physical address space this restriction can be expressed in terms
  of VM_MAXUSER_ADDRESS.
  
  This is safe because the amd64/pmap code never uses vtopte()/vtopde() for
  user virtual addresses.

Modified:
  projects/bhyve_npt_pmap/sys/amd64/amd64/pmap.c

Modified: projects/bhyve_npt_pmap/sys/amd64/amd64/pmap.c
==============================================================================
--- projects/bhyve_npt_pmap/sys/amd64/amd64/pmap.c	Tue Jun 25 02:48:36 2013	(r252191)
+++ projects/bhyve_npt_pmap/sys/amd64/amd64/pmap.c	Tue Jun 25 03:57:27 2013	(r252192)
@@ -473,6 +473,8 @@ vtopte(vm_offset_t va)
 {
 	u_int64_t mask = ((1ul << (NPTEPGSHIFT + NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT)) - 1);
 
+	KASSERT(va >= VM_MAXUSER_ADDRESS, ("vtopte on a uva/gpa 0x%0lx", va));
+
 	return (PTmap + ((va >> PAGE_SHIFT) & mask));
 }
 
@@ -481,6 +483,8 @@ vtopde(vm_offset_t va)
 {
 	u_int64_t mask = ((1ul << (NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT)) - 1);
 
+	KASSERT(va >= VM_MAXUSER_ADDRESS, ("vtopde on a uva/gpa 0x%0lx", va));
+
 	return (PDmap + ((va >> PDRSHIFT) & mask));
 }