kernel/vm: Prevent access to kernel args physical ranges after page init.

Otherwise, vm_allocate_early_physical_page could try to use them to
allocate more physical pages, which isn't legal as page state
management is now the responsibility of the vm_page system.

Discovered while working on re-activating the guarded heap.
In that case, at least, the illegally allocated pages were caught
by the memory manager trying to convert them into areas,
but they weren't marked WIRED and so tripped asserts.

Change-Id: I39af70bf8a652511bb65fe2154bba16d4ba5c924
Reviewed-on: https://review.haiku-os.org/c/haiku/+/7699
Reviewed-by: waddlesplash <[email protected]>
Reviewed-by: Axel Dörfler <[email protected]>

src/system/kernel/vm/vm.cpp

@@ -4234,6 +4234,11 @@ is_page_in_physical_memory_range(kernel_args* args, phys_addr_t address)
 page_num_t
 vm_allocate_early_physical_page(kernel_args* args)
 {
+	if (args->num_physical_allocated_ranges == 0) {
+		panic("early physical page allocations no longer possible!");
+		return 0;
+	}
+
 	for (uint32 i = 0; i < args->num_physical_allocated_ranges; i++) {
 		phys_addr_t nextPage;
 
@@ -4306,9 +4311,12 @@ vm_allocate_early(kernel_args* args, size_t virtualSize, size_t physicalSize,
 
 		//dprintf("vm_allocate_early: paddr 0x%lx\n", physicalAddress);
 
-		arch_vm_translation_map_early_map(args, virtualBase + i * B_PAGE_SIZE,
+		status_t status = arch_vm_translation_map_early_map(args,
+			virtualBase + i * B_PAGE_SIZE,
 			physicalAddress * B_PAGE_SIZE, attributes,
 			&vm_allocate_early_physical_page);
+		if (status != B_OK)
+			panic("error mapping early page!");
 	}
 
 	return virtualBase;

src/system/kernel/vm/vm_page.cpp

@@ -3374,6 +3374,10 @@ vm_page_init(kernel_args *args)
 			args->physical_allocated_range[i].size / B_PAGE_SIZE, true);
 	}
 
+	// prevent future accesses to the kernel args ranges
+	args->num_physical_memory_ranges = 0;
+	args->num_physical_allocated_ranges = 0;
+
 	// The target of actually free pages. This must be at least the system
 	// reserve, but should be a few more pages, so we don't have to extract
 	// a cached page with each allocation.