Change vm_malloc() to map pages in the guest physical address space in 4KB

chunks. This breaks the assumption that the entire memory segment is
contiguously allocated in the host physical address space.

This also paves the way to satisfy the 4KB page allocations by requesting
free pages from the VM subsystem as opposed to hard-partitioning host memory
at boot time.

5 files changed, 53 insertions, 20 deletions

diff --git a/sys/amd64/include/vmm_dev.h b/sys/amd64/include/vmm_dev.h
index fc64fd8..42ad236 100644
--- a/sys/amd64/include/vmm_dev.h
+++ b/sys/amd64/include/vmm_dev.h
@@ -35,7 +35,6 @@ void	vmmdev_cleanup(void);
 #endif
 
 struct vm_memory_segment {
-	vm_paddr_t	hpa;	/* out */
 	vm_paddr_t	gpa;	/* in */
 	size_t		len;	/* in */
 };
diff --git a/sys/amd64/vmm/io/ppt.c b/sys/amd64/vmm/io/ppt.c
index ace2877..e81fdbc 100644
--- a/sys/amd64/vmm/io/ppt.c
+++ b/sys/amd64/vmm/io/ppt.c
@@ -356,7 +356,6 @@ ppt_map_mmio(struct vm *vm, int bus, int slot, int func,
 				if (error == 0) {
 					seg->gpa = gpa;
 					seg->len = len;
-					seg->hpa = hpa;
 				}
 				return (error);
 			}
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
index 3dabbd6..7bd3f7f 100644
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -275,6 +275,28 @@ vm_create(const char *name)
 	return (vm);
 }
 
+static void
+vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
+{
+	size_t len;
+	vm_paddr_t hpa;
+
+	len = 0;
+	while (len < seg->len) {
+		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
+		if (hpa == (vm_paddr_t)-1) {
+			panic("vm_free_mem_segs: cannot free hpa "
+			      "associated with gpa 0x%016lx", seg->gpa + len);
+		}
+
+		vmm_mem_free(hpa, PAGE_SIZE);
+
+		len += PAGE_SIZE;
+	}
+
+	bzero(seg, sizeof(struct vm_memory_segment));
+}
+
 void
 vm_destroy(struct vm *vm)
 {
@@ -283,7 +305,9 @@ vm_destroy(struct vm *vm)
 	ppt_unassign_all(vm);
 
 	for (i = 0; i < vm->num_mem_segs; i++)
-		vmm_mem_free(vm->mem_segs[i].hpa, vm->mem_segs[i].len);
+		vm_free_mem_seg(vm, &vm->mem_segs[i]);
+
+	vm->num_mem_segs = 0;
 
 	for (i = 0; i < VM_MAXCPU; i++)
 		vcpu_cleanup(&vm->vcpu[i]);
@@ -345,6 +369,7 @@ int
 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
 {
 	int error, available, allocated;
+	struct vm_memory_segment *seg;
 	vm_paddr_t g, hpa;
 
 	const boolean_t spok = TRUE;	/* superpage mappings are ok */
@@ -380,22 +405,32 @@ vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
 		return (E2BIG);
 
-	hpa = vmm_mem_alloc(len);
-	if (hpa == 0)
-		return (ENOMEM);
+	seg = &vm->mem_segs[vm->num_mem_segs];
 
-	error = VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_WRITE_BACK,
-			   VM_PROT_ALL, spok);
-	if (error) {
-		vmm_mem_free(hpa, len);
-		return (error);
+	seg->gpa = gpa;
+	seg->len = 0;
+	while (seg->len < len) {
+		hpa = vmm_mem_alloc(PAGE_SIZE);
+		if (hpa == 0) {
+			error = ENOMEM;
+			break;
+		}
+
+		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
+				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
+		if (error)
+			break;
+
+		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
+
+		seg->len += PAGE_SIZE;
 	}
 
-	iommu_create_mapping(vm->iommu, gpa, hpa, len);
+	if (seg->len != len) {
+		vm_free_mem_seg(vm, seg);
+		return (error);
+	}
 
-	vm->mem_segs[vm->num_mem_segs].gpa = gpa;
-	vm->mem_segs[vm->num_mem_segs].hpa = hpa;
-	vm->mem_segs[vm->num_mem_segs].len = len;
 	vm->num_mem_segs++;
 
 	return (0);
diff --git a/sys/amd64/vmm/vmm_dev.c b/sys/amd64/vmm/vmm_dev.c
index b504e6b..91edbe8 100644
--- a/sys/amd64/vmm/vmm_dev.c
+++ b/sys/amd64/vmm/vmm_dev.c
@@ -299,7 +299,7 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 		break;
 	case VM_GET_MEMORY_SEG:
 		seg = (struct vm_memory_segment *)data;
-		seg->hpa = seg->len = 0;
+		seg->len = 0;
 		(void)vm_gpabase2memseg(sc->vm, seg->gpa, seg);
 		error = 0;
 		break;
diff --git a/sys/amd64/vmm/vmm_mem.c b/sys/amd64/vmm/vmm_mem.c
index 54f98ac..eb05b9d 100644
--- a/sys/amd64/vmm/vmm_mem.c
+++ b/sys/amd64/vmm/vmm_mem.c
@@ -318,9 +318,9 @@ vmm_mem_alloc(size_t size)
 	int i;
 	vm_paddr_t addr;
 
-	if ((size & PDRMASK) != 0) {
+	if ((size & PAGE_MASK) != 0) {
 		panic("vmm_mem_alloc: size 0x%0lx must be "
-		      "aligned on a 0x%0x boundary\n", size, NBPDR);
+		      "aligned on a 0x%0x boundary\n", size, PAGE_SIZE);
 	}
 
 	addr = 0;
@@ -373,9 +373,9 @@ vmm_mem_free(vm_paddr_t base, size_t length)
 {
 	int i;
 
-	if ((base & PDRMASK) != 0 || (length & PDRMASK) != 0) {
+	if ((base & PAGE_MASK) != 0 || (length & PAGE_MASK) != 0) {
 		panic("vmm_mem_free: base 0x%0lx and length 0x%0lx must be "
-		      "aligned on a 0x%0x boundary\n", base, length, NBPDR);
+		      "aligned on a 0x%0x boundary\n", base, length, PAGE_SIZE);
 	}
 
 	mtx_lock(&vmm_mem_mtx);