PERFORCE change 53022 for review

[Peter Wemm]

http://perforce.freebsd.org/chv.cgi?CH=53022

Change 53022 by peter at peter_hammer on 2004/05/18 17:48:54

	clear the kernel/direct map/recursive mapping pml4 entries and
	put the pml4 page in the zero queue.  Move the PHYS_TO_VM_PAGE()
	outside the queues lock, just like on i386.

Affected files ...

.. //depot/projects/hammer/sys/amd64/amd64/pmap.c#67 edit

Differences ...

==== //depot/projects/hammer/sys/amd64/amd64/pmap.c#67 (text+ko) ====

@@ -1287,6 +1287,8 @@
 pmap_release(pmap_t pmap)
 {
 	vm_page_t m;
+	int i;
+	static int j;
 
 	KASSERT(pmap->pm_stats.resident_count == 0,
 	    ("pmap_release: pmap resident count %ld != 0",
@@ -1296,12 +1298,16 @@
 	LIST_REMOVE(pmap, pm_list);
 	mtx_unlock_spin(&allpmaps_lock);
 
+	m = PHYS_TO_VM_PAGE(pmap->pm_pml4[PML4PML4I]);
+
+	pmap->pm_pml4[KPML4I] = 0;	/* KVA */
+	pmap->pm_pml4[DMPML4I] = 0;	/* Direct Map */
+	pmap->pm_pml4[PML4PML4I] = 0;	/* Recursive Mapping */
+
 	vm_page_lock_queues();
-	m = PHYS_TO_VM_PAGE(pmap->pm_pml4[PML4PML4I]);
 	m->wire_count--;
 	atomic_subtract_int(&cnt.v_wire_count, 1);
-	/* XXX check if we can vm_page_free_zero here! */
-	vm_page_free(m);
+	vm_page_free_zero(m);
 	vm_page_unlock_queues();
 }