PERFORCE change 94115 for review

[Peter Wemm]

http://perforce.freebsd.org/chv.cgi?CH=94115

Change 94115 by peter at peter_daintree on 2006/03/27 16:54:45

	oops.  Since we have to skip PG_W pages, we can't depend on
	pc_map[] getting cleaned out as we go.  Removal of the sva/eva
	bounding turned out to be irrelevant.

Affected files ...

.. //depot/projects/hammer/sys/amd64/amd64/pmap.c#145 edit

Differences ...

==== //depot/projects/hammer/sys/amd64/amd64/pmap.c#145 (text+ko) ====

@@ -2709,16 +2709,17 @@
 	PMAP_LOCK(pmap);
 	TAILQ_FOREACH_SAFE(pc, &pmap->pm_pvchunk, pc_list, npc) {
 		/*
-		 * XXX optimize more after removal of sva/eva!
-		 * We can use bsfq to find the next inuse pv instead of
-		 * the for loop.  Since we free as we go, we won't have
-		 * to skip unfreeable ones that sva/eva might have allowed.
+		 * XXX removal of the sva/eva bounding wasn't necessary
+		 * after all because we still have to work around the
+		 * skipping of a PG_W page.  oops.
 		 */
 		for (field = 0; field < _NPCM; field++) {
-			while ((inuse = (~(pc->pc_map[field])) & pc_freemask[field]) != 0) {
+			inuse = (~(pc->pc_map[field])) & pc_freemask[field];
+			while (inuse != 0) {
 				bit = bsfq(inuse);
 				idx = field * 64 + bit;
 				pv = &pc->pc_pventry[idx];
+				inuse &= ~(1UL << bit);
 
 				pte = vtopte(pv->pv_va);
 				tpte = *pte;