vm: implement lazy-attach cow-duplication of vm-objects attached to a controller

vm/address-space.c

@@ -6,6 +6,7 @@
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/task.h>
+#include <kernel/thread.h>
 #include <kernel/util.h>
 #include <kernel/vm-controller.h>
 #include <kernel/vm-object.h>
@@ -1502,6 +1503,17 @@ static kern_status_t request_missing_page(
 	vm_object_lock(object);
 	address_space_unlock(region);
 
+	kern_status_t status = KERN_OK;
+
+	if (object->vo_flags & VMO_LAZY_ATTACH) {
+		status = vm_object_attach_cow(object, irq_flags);
+	}
+
+	if (status != KERN_OK) {
+		vm_object_unlock_irqrestore(object, *irq_flags);
+		return status;
+	}
+
 	struct vm_page *pg = vm_object_get_page(
 		object,
 		object_offset,
@@ -1515,7 +1527,7 @@ static kern_status_t request_missing_page(
 
 	/* now: `region` is unlocked, and `object` is locked */
 
-	kern_status_t status = pmap_add(
+	status = pmap_add(
 		region->s_pmap,
 		addr,
 		vm_page_get_pfn(pg),
@@ -1550,9 +1562,6 @@ static kern_status_t handle_cow_access(
 	}
 
 	tracek("cow access %zx", addr);
-	if (area->vma_object->vo_ctrl) {
-		panic("COW on controlled vm-object");
-	}
 
 	off_t object_offset = addr - area->vma_base + area->vma_object_offset;
 	vm_object_lock(area->vma_object);
@@ -1663,7 +1672,7 @@ kern_status_t address_space_demand_map(
 	address_space_lock_irqsave(region, &irq_flags);
 
 	const enum pmap_fault_flags cow_flags
-		= PMAP_FAULT_WRITE | PMAP_FAULT_PRESENT | PMAP_FAULT_USER;
+		= PMAP_FAULT_WRITE | PMAP_FAULT_PRESENT;
 
 	if ((flags & cow_flags) == cow_flags) {
 		return handle_cow_access(region, addr, flags, &irq_flags);