Add a new type of VM object: OBJT_SG. An OBJT_SG object is very similar to

a device pager (OBJT_DEVICE) object in that it uses fictitious pages to
provide aliases to other memory addresses.  The primary difference is that
it uses an sglist(9) to determine the physical addresses for a given offset
into the object instead of invoking the d_mmap() method in a device driver.

Reviewed by:	alc
Approved by:	re (kensmith)
MFC after:	2 weeks

sys/amd64/amd64/pmap.c

@@ -3359,7 +3359,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object,
 	int pat_mode;
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 	if ((addr & (NBPDR - 1)) == 0 && (size & (NBPDR - 1)) == 0) {
 		if (!vm_object_populate(object, pindex, pindex + atop(size)))

sys/arm/arm/pmap.c

@@ -3101,7 +3101,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object,
 {
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 }
 

sys/conf/files

@@ -2637,6 +2637,7 @@ vm/default_pager.c		standard
 vm/device_pager.c		standard
 vm/phys_pager.c			standard
 vm/redzone.c			optional DEBUG_REDZONE
+vm/sg_pager.c			standard
 vm/swap_pager.c			standard
 vm/uma_core.c			standard
 vm/uma_dbg.c			standard

sys/fs/procfs/procfs_map.c

@@ -174,6 +174,7 @@ procfs_doprocmap(PFS_FILL_ARGS)
 				type = "swap";
 				vp = NULL;
 				break;
+			case OBJT_SG:
 			case OBJT_DEVICE:
 				type = "device";
 				vp = NULL;

sys/i386/i386/pmap.c

@@ -3448,7 +3448,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object,
 	int pat_mode;
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 	if (pseflag && 
 	    (addr & (NBPDR - 1)) == 0 && (size & (NBPDR - 1)) == 0) {

sys/i386/xen/pmap.c

@@ -3093,7 +3093,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr,
 	vm_page_t p;
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 	if (pseflag && 
 	    ((addr & (NBPDR - 1)) == 0) && ((size & (NBPDR - 1)) == 0)) {

sys/ia64/ia64/pmap.c

@@ -1751,7 +1751,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr,
 {
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 }
 

sys/kern/kern_proc.c

@@ -1486,6 +1486,9 @@ sysctl_kern_proc_ovmmap(SYSCTL_HANDLER_ARGS)
 			case OBJT_DEAD:
 				kve->kve_type = KVME_TYPE_DEAD;
 				break;
+			case OBJT_SG:
+				kve->kve_type = KVME_TYPE_SG;
+				break;
 			default:
 				kve->kve_type = KVME_TYPE_UNKNOWN;
 				break;
@@ -1658,6 +1661,9 @@ sysctl_kern_proc_vmmap(SYSCTL_HANDLER_ARGS)
 			case OBJT_DEAD:
 				kve->kve_type = KVME_TYPE_DEAD;
 				break;
+			case OBJT_SG:
+				kve->kve_type = KVME_TYPE_SG;
+				break;
 			default:
 				kve->kve_type = KVME_TYPE_UNKNOWN;
 				break;

sys/mips/mips/pmap.c

@@ -2148,7 +2148,7 @@ pmap_object_init_pt(pmap_t pmap, vm_offset_t addr,
     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
 {
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 }
 

sys/powerpc/booke/pmap.c

@@ -2582,7 +2582,7 @@ mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
 {
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("mmu_booke_object_init_pt: non-device object"));
 }
 

sys/sparc64/sparc64/pmap.c

@@ -1499,7 +1499,7 @@ pmap_object_init_pt(pmap_t pm, vm_offset_t addr, vm_object_t object,
 {
 
 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
-	KASSERT(object->type == OBJT_DEVICE,
+	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
 	    ("pmap_object_init_pt: non-device object"));
 }
 

sys/sys/user.h

@@ -340,6 +340,7 @@ struct kinfo_file {
 #define	KVME_TYPE_DEVICE	4
 #define	KVME_TYPE_PHYS		5
 #define	KVME_TYPE_DEAD		6
+#define	KVME_TYPE_SG		7
 #define	KVME_TYPE_UNKNOWN	255
 
 #define	KVME_PROT_READ		0x00000001

sys/vm/sg_pager.c

@@ -0,0 +1,263 @@
+/*-
+ * Copyright (c) 2009 Advanced Computing Technologies LLC
+ * Written by: John H. Baldwin <jhb@FreeBSD.org>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+/*
+ * This pager manages OBJT_SG objects.  These objects are backed by
+ * a scatter/gather list of physical address ranges.
+ */
+
+#include <sys/param.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/sglist.h>
+#include <vm/vm.h>
+#include <vm/vm_object.h>
+#include <vm/vm_page.h>
+#include <vm/vm_pager.h>
+#include <vm/uma.h>
+
+static void sg_pager_init(void);
+static vm_object_t sg_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
+    vm_ooffset_t, struct ucred *);
+static void sg_pager_dealloc(vm_object_t);
+static int sg_pager_getpages(vm_object_t, vm_page_t *, int, int);
+static void sg_pager_putpages(vm_object_t, vm_page_t *, int, 
+		boolean_t, int *);
+static boolean_t sg_pager_haspage(vm_object_t, vm_pindex_t, int *,
+		int *);
+
+static uma_zone_t fakepg_zone;
+
+static vm_page_t sg_pager_getfake(vm_paddr_t, vm_memattr_t);
+static void sg_pager_putfake(vm_page_t);
+
+struct pagerops sgpagerops = {
+	.pgo_init =	sg_pager_init,
+	.pgo_alloc =	sg_pager_alloc,
+	.pgo_dealloc =	sg_pager_dealloc,
+	.pgo_getpages =	sg_pager_getpages,
+	.pgo_putpages =	sg_pager_putpages,
+	.pgo_haspage =	sg_pager_haspage,
+};
+
+static void
+sg_pager_init(void)
+{
+
+	fakepg_zone = uma_zcreate("SG fakepg", sizeof(struct vm_page),
+	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
+	    UMA_ZONE_NOFREE|UMA_ZONE_VM); 
+}
+
+static vm_object_t
+sg_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
+    vm_ooffset_t foff, struct ucred *cred)
+{
+	struct sglist *sg;
+	vm_object_t object;
+	vm_pindex_t npages, pindex;
+	int i;
+
+	/*
+	 * Offset should be page aligned.
+	 */
+	if (foff & PAGE_MASK)
+		return (NULL);
+
+	/*
+	 * The scatter/gather list must only include page-aligned
+	 * ranges.
+	 */
+	npages = 0;
+	sg = handle;
+	for (i = 0; i < sg->sg_nseg; i++) {
+		if ((sg->sg_segs[i].ss_paddr % PAGE_SIZE) != 0 ||
+		    (sg->sg_segs[i].ss_len % PAGE_SIZE) != 0)
+			return (NULL);
+		npages += sg->sg_segs[i].ss_len / PAGE_SIZE;
+	}
+
+	/*
+	 * The scatter/gather list has a fixed size.  Refuse requests
+	 * to map beyond that.
+	 */
+	size = round_page(size);
+	pindex = OFF_TO_IDX(foff + size);
+	if (pindex > npages)
+		return (NULL);
+
+	/*
+	 * Allocate a new object and associate it with the
+	 * scatter/gather list.  It is ok for our purposes to have
+	 * multiple VM objects associated with the same scatter/gather
+	 * list because scatter/gather lists are static.  This is also
+	 * simpler than ensuring a unique object per scatter/gather
+	 * list.
+	 */
+	object = vm_object_allocate(OBJT_SG, npages);
+	object->handle = sglist_hold(sg);
+	TAILQ_INIT(&object->un_pager.sgp.sgp_pglist);
+	return (object);
+}
+
+static void
+sg_pager_dealloc(vm_object_t object)
+{
+	struct sglist *sg;
+	vm_page_t m;
+
+	/*
+	 * Free up our fake pages.
+	 */
+	while ((m = TAILQ_FIRST(&object->un_pager.sgp.sgp_pglist)) != 0) {
+		TAILQ_REMOVE(&object->un_pager.sgp.sgp_pglist, m, pageq);
+		sg_pager_putfake(m);
+	}
+	
+	sg = object->handle;
+	sglist_free(sg);
+}
+
+static int
+sg_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
+{
+	struct sglist *sg;
+	vm_page_t m_paddr, page;
+	vm_pindex_t offset;
+	vm_paddr_t paddr;
+	vm_memattr_t memattr;
+	size_t space;
+	int i;
+
+	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
+	sg = object->handle;
+	memattr = object->memattr;
+	VM_OBJECT_UNLOCK(object);
+	offset = m[reqpage]->pindex;
+
+	/*
+	 * Lookup the physical address of the requested page.  An initial
+	 * value of '1' instead of '0' is used so we can assert that the
+	 * page is found since '0' can be a valid page-aligned physical
+	 * address.
+	 */
+	space = 0;
+	paddr = 1;
+	for (i = 0; i < sg->sg_nseg; i++) {
+		if (space + sg->sg_segs[i].ss_len <= (offset * PAGE_SIZE)) {
+			space += sg->sg_segs[i].ss_len;
+			continue;
+		}
+		paddr = sg->sg_segs[i].ss_paddr + offset * PAGE_SIZE - space;
+		break;
+	}
+	KASSERT(paddr != 1, ("invalid SG page index"));
+
+	/* If "paddr" is a real page, perform a sanity check on "memattr". */
+	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
+	    pmap_page_get_memattr(m_paddr) != memattr) {
+		memattr = pmap_page_get_memattr(m_paddr);
+		printf(
+	    "WARNING: A device driver has set \"memattr\" inconsistently.\n");
+	}
+
+	/* Return a fake page for the requested page. */
+	KASSERT(!(m[reqpage]->flags & PG_FICTITIOUS),
+	    ("backing page for SG is fake"));
+
+	/* Construct a new fake page. */
+	printf("SG: getting fake page for paddr %lx\n", paddr);
+	page = sg_pager_getfake(paddr, memattr);
+	VM_OBJECT_LOCK(object);
+	TAILQ_INSERT_TAIL(&object->un_pager.sgp.sgp_pglist, page, pageq);
+
+	/* Free the original pages and insert this fake page into the object. */
+	vm_page_lock_queues();
+	for (i = 0; i < count; i++) {
+		printf("SG: freeing VM page %p\n", m[i]);
+		vm_page_free(m[i]);
+	}
+	vm_page_unlock_queues();
+	printf("SG: Inserting new fake page\n");
+	vm_page_insert(page, object, offset);
+	m[reqpage] = page;
+
+	return (VM_PAGER_OK);
+}
+
+static void
+sg_pager_putpages(vm_object_t object, vm_page_t *m, int count,
+    boolean_t sync, int *rtvals)
+{
+
+	panic("sg_pager_putpage called");
+}
+
+static boolean_t
+sg_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
+    int *after)
+{
+
+	if (before != NULL)
+		*before = 0;
+	if (after != NULL)
+		*after = 0;
+	return (TRUE);
+}
+
+/*
+ * Create a fictitious page with the specified physical address and memory
+ * attribute.  The memory attribute is the only the machine-dependent aspect
+ * of a fictitious page that must be initialized.
+ */
+static vm_page_t
+sg_pager_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
+{
+	vm_page_t m;
+
+	m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
+	m->phys_addr = paddr;
+	/* Fictitious pages don't use "segind". */
+	m->flags = PG_FICTITIOUS;
+	/* Fictitious pages don't use "order" or "pool". */
+	m->oflags = VPO_BUSY;
+	m->wire_count = 1;
+	pmap_page_set_memattr(m, memattr);
+	return (m);
+}
+
+static void
+sg_pager_putfake(vm_page_t m)
+{
+
+	if (!(m->flags & PG_FICTITIOUS))
+		panic("sg_pager_putfake: bad page");
+	uma_zfree(fakepg_zone, m);
+}

sys/vm/vm.h

@@ -89,7 +89,7 @@ typedef u_char vm_prot_t;	/* protection codes */
 #define	VM_PROT_DEFAULT		VM_PROT_ALL
 
 enum obj_type { OBJT_DEFAULT, OBJT_SWAP, OBJT_VNODE, OBJT_DEVICE, OBJT_PHYS,
-		OBJT_DEAD };
+		OBJT_DEAD, OBJT_SG };
 typedef u_char objtype_t;
 
 union vm_map_object;

sys/vm/vm_fault.c

@@ -472,7 +472,8 @@ RetryFault:;
 			    (fs.first_object == fs.object ||
 			     (is_first_object_locked = VM_OBJECT_TRYLOCK(fs.first_object))) &&
 			    fs.first_object->type != OBJT_DEVICE &&
-			    fs.first_object->type != OBJT_PHYS) {
+			    fs.first_object->type != OBJT_PHYS &&
+			    fs.first_object->type != OBJT_SG) {
 				vm_pindex_t firstpindex, tmppindex;
 
 				if (fs.first_pindex < 2 * VM_FAULT_READ)

sys/vm/vm_map.c

@@ -1719,7 +1719,7 @@ vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
 	if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
 		return;
 	VM_OBJECT_LOCK(object);
-	if (object->type == OBJT_DEVICE) {
+	if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
 		pmap_object_init_pt(map->pmap, addr, object, pindex, size);
 		goto unlock_return;
 	}
@@ -2247,7 +2247,8 @@ vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
 				 */
 				vm_fault_unwire(map, entry->start, entry->end,
 				    entry->object.vm_object != NULL &&
-				    entry->object.vm_object->type == OBJT_DEVICE);
+				    (entry->object.vm_object->type == OBJT_DEVICE ||
+				    entry->object.vm_object->type == OBJT_SG));
 			}
 		}
 		KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
@@ -2366,7 +2367,8 @@ vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
 			saved_start = entry->start;
 			saved_end = entry->end;
 			fictitious = entry->object.vm_object != NULL &&
-			    entry->object.vm_object->type == OBJT_DEVICE;
+			    (entry->object.vm_object->type == OBJT_DEVICE ||
+			    entry->object.vm_object->type == OBJT_SG);
 			/*
 			 * Release the map lock, relying on the in-transition
 			 * mark.
@@ -2462,7 +2464,8 @@ vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
 				 */
 				vm_fault_unwire(map, entry->start, entry->end,
 				    entry->object.vm_object != NULL &&
-				    entry->object.vm_object->type == OBJT_DEVICE);
+				    (entry->object.vm_object->type == OBJT_DEVICE ||
+				    entry->object.vm_object->type == OBJT_SG));
 			}
 		}
 	next_entry_done:
@@ -2595,7 +2598,8 @@ vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
 {
 	vm_fault_unwire(map, entry->start, entry->end,
 	    entry->object.vm_object != NULL &&
-	    entry->object.vm_object->type == OBJT_DEVICE);
+	    (entry->object.vm_object->type == OBJT_DEVICE ||
+	    entry->object.vm_object->type == OBJT_SG));
 	entry->wired_count = 0;
 }
 

sys/vm/vm_meter.c

@@ -205,7 +205,7 @@ vmtotal(SYSCTL_HANDLER_ARGS)
 		 * synchronization should not impair the accuracy of
 		 * the reported statistics. 
 		 */
-		if (object->type == OBJT_DEVICE) {
+		if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
 			/*
 			 * Devices, like /dev/mem, will badly skew our totals.
 			 */

sys/vm/vm_object.c

@@ -309,6 +309,7 @@ vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
 	case OBJT_DEFAULT:
 	case OBJT_DEVICE:
 	case OBJT_PHYS:
+	case OBJT_SG:
 	case OBJT_SWAP:
 	case OBJT_VNODE:
 		if (!TAILQ_EMPTY(&object->memq))

sys/vm/vm_object.h

@@ -123,6 +123,15 @@ struct vm_object {
 			TAILQ_HEAD(, vm_page) devp_pglist;
 		} devp;
 
+		/*
+		 * SG pager
+		 *
+		 *	sgp_pglist - list of allocated pages
+		 */
+		struct {
+			TAILQ_HEAD(, vm_page) sgp_pglist;
+		} sgp;
+
 		/*
 		 * Swap pager
 		 *

sys/vm/vm_page.c

@@ -1163,7 +1163,7 @@ vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
 	if (object != NULL) {
 		/* Ignore device objects; the pager sets "memattr" for them. */
 		if (object->memattr != VM_MEMATTR_DEFAULT &&
-		    object->type != OBJT_DEVICE)
+		    object->type != OBJT_DEVICE && object->type != OBJT_SG)
 			pmap_page_set_memattr(m, object->memattr);
 		vm_page_insert(m, object, pindex);
 	} else

sys/vm/vm_pageout.c

@@ -513,7 +513,9 @@ vm_pageout_object_deactivate_pages(pmap, first_object, desired)
 	int actcount, rcount, remove_mode;
 
 	VM_OBJECT_LOCK_ASSERT(first_object, MA_OWNED);
-	if (first_object->type == OBJT_DEVICE || first_object->type == OBJT_PHYS)
+	if (first_object->type == OBJT_DEVICE ||
+	    first_object->type == OBJT_SG ||
+	    first_object->type == OBJT_PHYS)
 		return;
 	for (object = first_object;; object = backing_object) {
 		if (pmap_resident_count(pmap) <= desired)

sys/vm/vm_pager.c

@@ -160,7 +160,8 @@ struct pagerops *pagertab[] = {
 	&vnodepagerops,		/* OBJT_VNODE */
 	&devicepagerops,	/* OBJT_DEVICE */
 	&physpagerops,		/* OBJT_PHYS */
-	&deadpagerops		/* OBJT_DEAD */
+	&deadpagerops,		/* OBJT_DEAD */
+	&sgpagerops		/* OBJT_SG */
 };
 
 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]);

sys/vm/vm_pager.h

@@ -70,6 +70,7 @@ extern struct pagerops swappagerops;
 extern struct pagerops vnodepagerops;
 extern struct pagerops devicepagerops;
 extern struct pagerops physpagerops;
+extern struct pagerops sgpagerops;
 
 /*
  * get/put return values

usr.bin/procstat/procstat_vm.c

@@ -93,6 +93,9 @@ procstat_vm(pid_t pid, struct kinfo_proc *kipp __unused)
 		case KVME_TYPE_DEAD:
 			str = "dd";
 			break;
+		case KVME_TYPE_SG:
+			str = "sg";
+			break;
 		case KVME_TYPE_UNKNOWN:
 		default:
 			str = "??";