This is a cleanup patch to Peter's new OBJT_PHYS VM object type

    and sysv shared memory support for it.  It implements a new
    PG_UNMANAGED flag that has slightly different characteristics
    from PG_FICTICIOUS.

    A new sysctl, kern.ipc.shm_use_phys has been added to enable the
    use of physically-backed sysv shared memory rather then swap-backed.
    Physically backed shm segments are not tracked with PV entries,
    allowing programs which use a large shm segment as a rendezvous
    point to operate without eating an insane amount of KVM in the
    PV entry management.  Read: Oracle.

    Peter's OBJT_PHYS object will also allow us to eventually implement
    page-table sharing and/or 4MB physical page support for such segments.
    We're half way there.

10 files changed, 118 insertions, 47 deletions

diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
index 5e3d6b5..7b27446 100644
--- a/sys/amd64/amd64/pmap.c
+++ b/sys/amd64/amd64/pmap.c
@@ -2095,7 +2095,8 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
 	 * raise IPL while manipulating pv_table since pmap_enter can be
 	 * called at interrupt time.
 	 */
-	if (pmap_initialized && (m->flags & PG_FICTITIOUS) == 0) {
+	if (pmap_initialized && 
+	    (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
 		pmap_insert_entry(pmap, va, mpte, m);
 		pa |= PG_MANAGED;
 	}
@@ -2223,7 +2224,8 @@ retry:
 	 * raise IPL while manipulating pv_table since pmap_enter can be
 	 * called at interrupt time.
 	 */
-	pmap_insert_entry(pmap, va, mpte, m);
+	if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
+		pmap_insert_entry(pmap, va, mpte, m);
 
 	/*
 	 * Increment counters
@@ -2235,7 +2237,10 @@ retry:
 	/*
 	 * Now validate mapping with RO protection
 	 */
-	*pte = pa | PG_V | PG_U | PG_MANAGED;
+	if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
+		*pte = pa | PG_V | PG_U;
+	else
+		*pte = pa | PG_V | PG_U | PG_MANAGED;
 
 	return mpte;
 }
diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c
index 5e3d6b5..7b27446 100644
--- a/sys/i386/i386/pmap.c
+++ b/sys/i386/i386/pmap.c
@@ -2095,7 +2095,8 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot,
 	 * raise IPL while manipulating pv_table since pmap_enter can be
 	 * called at interrupt time.
 	 */
-	if (pmap_initialized && (m->flags & PG_FICTITIOUS) == 0) {
+	if (pmap_initialized && 
+	    (m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0) {
 		pmap_insert_entry(pmap, va, mpte, m);
 		pa |= PG_MANAGED;
 	}
@@ -2223,7 +2224,8 @@ retry:
 	 * raise IPL while manipulating pv_table since pmap_enter can be
 	 * called at interrupt time.
 	 */
-	pmap_insert_entry(pmap, va, mpte, m);
+	if ((m->flags & (PG_FICTITIOUS|PG_UNMANAGED)) == 0)
+		pmap_insert_entry(pmap, va, mpte, m);
 
 	/*
 	 * Increment counters
@@ -2235,7 +2237,10 @@ retry:
 	/*
 	 * Now validate mapping with RO protection
 	 */
-	*pte = pa | PG_V | PG_U | PG_MANAGED;
+	if (m->flags & (PG_FICTITIOUS|PG_UNMANAGED))
+		*pte = pa | PG_V | PG_U;
+	else
+		*pte = pa | PG_V | PG_U | PG_MANAGED;
 
 	return mpte;
 }
diff --git a/sys/kern/sysv_shm.c b/sys/kern/sysv_shm.c
index 16019dc..11a5568 100644
--- a/sys/kern/sysv_shm.c
+++ b/sys/kern/sysv_shm.c
@@ -126,12 +126,15 @@ struct	shminfo shminfo = {
 	SHMALL
 };
 
+static int shm_use_phys;
+
 SYSCTL_DECL(_kern_ipc);
 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
 SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
 SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0, "");
 SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
+SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, "");
 
 static int
 shm_find_segment_by_key(key)
@@ -528,13 +531,13 @@ shmget_allocate_segment(p, uap, mode)
 	 * We make sure that we have allocated a pager before we need
 	 * to.
 	 */
-#ifdef SHM_PHYS_BACKED
-	shm_handle->shm_object =
-		vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
-#else
-	shm_handle->shm_object =
-		vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
-#endif
+	if (shm_use_phys) {
+		shm_handle->shm_object =
+		    vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
+	} else {
+		shm_handle->shm_object =
+		    vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
+	}
 	vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
 	vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
 
diff --git a/sys/vm/phys_pager.c b/sys/vm/phys_pager.c
index a48126c..586844f 100644
--- a/sys/vm/phys_pager.c
+++ b/sys/vm/phys_pager.c
@@ -104,7 +104,9 @@ phys_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
 		object = vm_object_allocate(OBJT_PHYS,
 			OFF_TO_IDX(foff + size));
 		object->handle = handle;
+#if 0
 		TAILQ_INIT(&object->un_pager.physp.physp_pglist);
+#endif
 		TAILQ_INSERT_TAIL(&phys_pager_object_list, object,
 		    pager_object_list);
 	} else {
@@ -131,20 +133,6 @@ phys_pager_dealloc(object)
 	int s;
 
 	TAILQ_REMOVE(&phys_pager_object_list, object, pager_object_list);
-	/*
-	 * Free up our fake pages.
-	 */
-	s = splvm();
-	while ((m = TAILQ_FIRST(&object->un_pager.physp.physp_pglist)) != 0) {
-		TAILQ_REMOVE(&object->un_pager.physp.physp_pglist, m, pageq);
-		/* return the page back to normal */
-		m->flags &= ~PG_FICTITIOUS;
-		m->dirty = 0;
-		vm_page_unwire(m, 0);
-		vm_page_flag_clear(m, PG_ZERO);
-		vm_page_free(m);
-	}
-	splx(s);
 }
 
 static int
@@ -165,8 +153,7 @@ phys_pager_getpages(object, m, count, reqpage)
 			vm_page_zero_fill(m[i]);
 		vm_page_flag_set(m[i], PG_ZERO);
 		/* Switch off pv_entries */
-		vm_page_wire(m[i]);
-		vm_page_flag_set(m[i], PG_FICTITIOUS);
+		vm_page_unmanage(m[i]);
 		m[i]->valid = VM_PAGE_BITS_ALL;
 		m[i]->dirty = 0;
 		/* The requested page must remain busy, the others not. */
@@ -174,8 +161,6 @@ phys_pager_getpages(object, m, count, reqpage)
 			vm_page_flag_clear(m[i], PG_BUSY);
 			m[i]->busy = 0;
 		}
-		TAILQ_INSERT_TAIL(&object->un_pager.physp.physp_pglist, m[i],
-		    pageq);
 	}
 	splx(s);
 
diff --git a/sys/vm/vm_fault.c b/sys/vm/vm_fault.c
index 14133fa..af4fe35 100644
--- a/sys/vm/vm_fault.c
+++ b/sys/vm/vm_fault.c
@@ -423,7 +423,7 @@ readrest:
 					if (mt == NULL || (mt->valid != VM_PAGE_BITS_ALL))
 						break;
 					if (mt->busy ||
-						(mt->flags & (PG_BUSY | PG_FICTITIOUS)) ||
+						(mt->flags & (PG_BUSY | PG_FICTITIOUS | PG_UNMANAGED)) ||
 						mt->hold_count ||
 						mt->wire_count) 
 						continue;
diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index 1b33f78..e5403d1 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -833,12 +833,14 @@ shadowlookup:
 
 		/*
 		 * If the page is busy or not in a normal active state,
-		 * we skip it.  Things can break if we mess with pages
-		 * in any of the below states.
+		 * we skip it.  If the page is not managed there are no
+		 * page queues to mess with.  Things can break if we mess
+		 * with pages in any of the below states.
 		 */
 		if (
 		    m->hold_count ||
 		    m->wire_count ||
+		    (m->flags & PG_UNMANAGED) ||
 		    m->valid != VM_PAGE_BITS_ALL
 		) {
 			continue;
@@ -1394,6 +1396,13 @@ vm_object_page_remove(object, start, end, clean_only)
 
 	all = ((end == 0) && (start == 0));
 
+	/*
+	 * Since physically-backed objects do not use managed pages, we can't
+	 * remove pages from the object (we must instead remove the page
+	 * references, and then destroy the object).
+	 */
+	KASSERT(object->type != OBJT_PHYS, ("attempt to remove pages from a physical object"));
+
 	vm_object_pip_add(object, 1);
 again:
 	size = end - start;
diff --git a/sys/vm/vm_object.h b/sys/vm/vm_object.h
index 62a3bbe..c9f239c 100644
--- a/sys/vm/vm_object.h
+++ b/sys/vm/vm_object.h
@@ -123,6 +123,7 @@ struct vm_object {
 			TAILQ_HEAD(, vm_page) devp_pglist;
 		} devp;
 
+#if 0
 		/*
 		 * Physmem pager
 		 *
@@ -131,6 +132,7 @@ struct vm_object {
 		struct {
 			TAILQ_HEAD(, vm_page) physp_pglist;
 		} physp;
+#endif
 
 		/*
 		 * Swap pager
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 6b2b320..9701150 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -688,7 +688,7 @@ vm_page_select_cache(object, pindex)
 		    (pindex + object->pg_color) & PQ_L2_MASK,
 		    FALSE
 		);
-		if (m && ((m->flags & PG_BUSY) || m->busy ||
+		if (m && ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy ||
 			       m->hold_count || m->wire_count)) {
 			vm_page_deactivate(m);
 			continue;
@@ -997,7 +997,7 @@ vm_page_activate(m)
 
 		vm_page_unqueue(m);
 
-		if (m->wire_count == 0) {
+		if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
 			m->queue = PQ_ACTIVE;
 			vm_page_queues[PQ_ACTIVE].lcnt++;
 			TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl, m, pageq);
@@ -1128,9 +1128,17 @@ vm_page_free_toq(vm_page_t m)
 		}
 	}
 
+	/*
+	 * Clear the UNMANAGED flag when freeing an unmanaged page.
+	 */
+
+	if (m->flags & PG_UNMANAGED) {
+	    m->flags &= ~PG_UNMANAGED;
+	} else {
 #ifdef __alpha__
-	pmap_page_is_free(m);
+	    pmap_page_is_free(m);
 #endif
+	}
 
 	m->queue = PQ_FREE + m->pc;
 	pq = &vm_page_queues[m->queue];
@@ -1155,6 +1163,39 @@ vm_page_free_toq(vm_page_t m)
 }
 
 /*
+ *	vm_page_unmanage:
+ *
+ * 	Prevent PV management from being done on the page.  The page is
+ *	removed from the paging queues as if it were wired, and as a 
+ *	consequence of no longer being managed the pageout daemon will not
+ *	touch it (since there is no way to locate the pte mappings for the
+ *	page).  madvise() calls that mess with the pmap will also no longer
+ *	operate on the page.
+ *
+ *	Beyond that the page is still reasonably 'normal'.  Freeing the page
+ *	will clear the flag.
+ *
+ *	This routine is used by OBJT_PHYS objects - objects using unswappable
+ *	physical memory as backing store rather then swap-backed memory and
+ *	will eventually be extended to support 4MB unmanaged physical 
+ *	mappings.
+ */
+
+void
+vm_page_unmanage(vm_page_t m)
+{
+	int s;
+
+	s = splvm();
+	if ((m->flags & PG_UNMANAGED) == 0) {
+		if (m->wire_count == 0)
+			vm_page_unqueue(m);
+	}
+	vm_page_flag_set(m, PG_UNMANAGED);
+	splx(s);
+}
+
+/*
  *	vm_page_wire:
  *
  *	Mark this page as wired down by yet
@@ -1170,9 +1211,15 @@ vm_page_wire(m)
 {
 	int s;
 
+	/*
+	 * Only bump the wire statistics if the page is not already wired,
+	 * and only unqueue the page if it is on some queue (if it is unmanaged
+	 * it is already off the queues).
+	 */
 	s = splvm();
 	if (m->wire_count == 0) {
-		vm_page_unqueue(m);
+		if ((m->flags & PG_UNMANAGED) == 0)
+			vm_page_unqueue(m);
 		cnt.v_wire_count++;
 	}
 	m->wire_count++;
@@ -1218,7 +1265,9 @@ vm_page_unwire(m, activate)
 		m->wire_count--;
 		if (m->wire_count == 0) {
 			cnt.v_wire_count--;
-			if (activate) {
+			if (m->flags & PG_UNMANAGED) {
+				;
+			} else if (activate) {
 				TAILQ_INSERT_TAIL(&vm_page_queues[PQ_ACTIVE].pl, m, pageq);
 				m->queue = PQ_ACTIVE;
 				vm_page_queues[PQ_ACTIVE].lcnt++;
@@ -1259,7 +1308,7 @@ _vm_page_deactivate(vm_page_t m, int athead)
 		return;
 
 	s = splvm();
-	if (m->wire_count == 0) {
+	if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
 		if ((m->queue - m->pc) == PQ_CACHE)
 			cnt.v_reactivated++;
 		vm_page_unqueue(m);
@@ -1293,7 +1342,7 @@ vm_page_cache(m)
 {
 	int s;
 
-	if ((m->flags & PG_BUSY) || m->busy || m->wire_count) {
+	if ((m->flags & (PG_BUSY|PG_UNMANAGED)) || m->busy || m->wire_count) {
 		printf("vm_page_cache: attempting to cache busy page\n");
 		return;
 	}
diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h
index e61be7f..a25c6ac 100644
--- a/sys/vm/vm_page.h
+++ b/sys/vm/vm_page.h
@@ -225,6 +225,13 @@ extern struct vpgqueues vm_page_queues[PQ_COUNT];
  * These are the flags defined for vm_page.
  *
  * Note: PG_FILLED and PG_DIRTY are added for the filesystems.
+ *
+ * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
+ * 	 not under PV management but otherwise should be treated as a
+ *	 normal page.  Pages not under PV management cannot be paged out
+ *	 via the object/vm_page_t because there is no knowledge of their
+ *	 pte mappings, nor can they be removed from their objects via 
+ *	 the object, and such pages are also not on any PQ queue.
  */
 #define	PG_BUSY		0x0001		/* page is in transit (O) */
 #define	PG_WANTED	0x0002		/* someone is waiting for page (O) */
@@ -236,6 +243,7 @@ extern struct vpgqueues vm_page_queues[PQ_COUNT];
 #define PG_CLEANCHK	0x0100		/* page will be checked for cleaning */
 #define PG_SWAPINPROG	0x0200		/* swap I/O in progress on page	     */
 #define PG_NOSYNC	0x0400		/* do not collect for syncer */
+#define PG_UNMANAGED	0x0800		/* No PV management for page */
 
 /*
  * Misc constants.
@@ -399,6 +407,7 @@ void vm_page_remove __P((vm_page_t));
 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t));
 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t));
 vm_page_t vm_add_new_page __P((vm_offset_t pa));
+void vm_page_unmanage __P((vm_page_t));
 void vm_page_unwire __P((vm_page_t, int));
 void vm_page_wire __P((vm_page_t));
 void vm_page_unqueue __P((vm_page_t));
diff --git a/sys/vm/vm_pageout.c b/sys/vm/vm_pageout.c
index 1627118..307dd0b 100644
--- a/sys/vm/vm_pageout.c
+++ b/sys/vm/vm_pageout.c
@@ -233,11 +233,12 @@ vm_pageout_clean(m)
 	 */
 
 	/*
-	 * Don't mess with the page if it's busy.
+	 * Don't mess with the page if it's busy, held, or special
 	 */
 	if ((m->hold_count != 0) ||
-	    ((m->busy != 0) || (m->flags & PG_BUSY)))
+	    ((m->busy != 0) || (m->flags & (PG_BUSY|PG_UNMANAGED)))) {
 		return 0;
+	}
 
 	mc[vm_pageout_page_count] = m;
 	pageout_count = 1;
@@ -279,7 +280,7 @@ more:
 			break;
 		}
 		if (((p->queue - p->pc) == PQ_CACHE) ||
-		    (p->flags & PG_BUSY) || p->busy) {
+		    (p->flags & (PG_BUSY|PG_UNMANAGED)) || p->busy) {
 			ib = 0;
 			break;
 		}
@@ -309,7 +310,7 @@ more:
 		if ((p = vm_page_lookup(object, pindex + is)) == NULL)
 			break;
 		if (((p->queue - p->pc) == PQ_CACHE) ||
-		    (p->flags & PG_BUSY) || p->busy) {
+		    (p->flags & (PG_BUSY|PG_UNMANAGED)) || p->busy) {
 			break;
 		}
 		vm_page_test_dirty(p);
@@ -474,7 +475,7 @@ vm_pageout_object_deactivate_pages(map, object, desired, map_remove_only)
 			if (p->wire_count != 0 ||
 			    p->hold_count != 0 ||
 			    p->busy != 0 ||
-			    (p->flags & PG_BUSY) ||
+			    (p->flags & (PG_BUSY|PG_UNMANAGED)) ||
 			    !pmap_page_exists(vm_map_pmap(map), p)) {
 				p = next;
 				continue;
@@ -1047,7 +1048,10 @@ rescan0:
 		m = vm_page_list_find(PQ_CACHE, cache_rover, FALSE);
 		if (!m)
 			break;
-		if ((m->flags & PG_BUSY) || m->busy || m->hold_count || m->wire_count) {
+		if ((m->flags & (PG_BUSY|PG_UNMANAGED)) || 
+		    m->busy || 
+		    m->hold_count || 
+		    m->wire_count) {
 #ifdef INVARIANTS
 			printf("Warning: busy page %p found in cache\n", m);
 #endif