diff --git a/src/startup/mldr/commpage.c b/src/startup/mldr/commpage.c
index b7e17c4a5..bfcf9a0b1 100644
--- a/src/startup/mldr/commpage.c
+++ b/src/startup/mldr/commpage.c
@@ -21,6 +21,8 @@ static const char* SIGNATURE64 = "commpage 64-bit";
 
 static uint64_t get_cpu_caps(void);
 
+int PAGE_SIZE;
+
 #define CGET(p) (commpage + ((p)-_COMM_PAGE_START_ADDRESS))
 
 void commpage_setup(bool _64bit)
@@ -36,6 +38,8 @@ void commpage_setup(bool _64bit)
 	uint8_t *user_page_shift, *kernel_page_shift;
 	struct sysinfo si;
 
+	PAGE_SIZE = sysconf(_SC_PAGESIZE);
+
 	commpage = (uint8_t*) mmap((void*)(_64bit ? _COMM_PAGE64_BASE_ADDRESS : _COMM_PAGE32_BASE_ADDRESS),
 			_64bit ? _COMM_PAGE64_AREA_LENGTH : _COMM_PAGE32_AREA_LENGTH, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
 	if (commpage == MAP_FAILED)
@@ -69,7 +73,7 @@ void commpage_setup(bool _64bit)
 	// as a substitute for log2().
 	user_page_shift = (uint8_t*)CGET(_64bit ? _COMM_PAGE_USER_PAGE_SHIFT_64 : _COMM_PAGE_USER_PAGE_SHIFT_32);
 	kernel_page_shift = (uint8_t*)CGET(_COMM_PAGE_KERNEL_PAGE_SHIFT);
-	*kernel_page_shift = *user_page_shift = (uint8_t)__builtin_ctzl(sysconf(_SC_PAGESIZE));
+	*kernel_page_shift = *user_page_shift = (uint8_t)__builtin_ctzl(PAGE_SIZE);
 
 	my_caps = get_cpu_caps();
 	if (*ncpus == 1)
@@ -166,3 +170,9 @@ unsigned long commpage_address(bool _64bit)
 	return _64bit ? _COMM_PAGE64_BASE_ADDRESS : _COMM_PAGE32_BASE_ADDRESS;
 }
 
+#ifdef __aarch64__
+vm_address_t _get_commpage_priv_address(void) {
+	// Not sure if this is correct...
+	return _COMM_PAGE64_BASE_ADDRESS;
+}
+#endif
\ No newline at end of file
diff --git a/src/startup/mldr/elfcalls/threads.c b/src/startup/mldr/elfcalls/threads.c
index e789986ea..e98cafb25 100644
--- a/src/startup/mldr/elfcalls/threads.c
+++ b/src/startup/mldr/elfcalls/threads.c
@@ -268,6 +268,13 @@ static void* darling_thread_entry(void* p)
 	register uintptr_t arg6 asm("r9")  = args.arg3;
 #elif __i386__
 	uintptr_t arg3 = args.real_entry_point;
+#elif __aarch64__
+	register void*     arg1 asm("x0") = args.pth;
+	register int       arg2 asm("w1") = args.port;
+	register uintptr_t arg3 asm("x2") = args.real_entry_point;
+	register uintptr_t arg4 asm("x3") = args.arg1;
+	register uintptr_t arg5 asm("x4") = args.arg2;
+	register uintptr_t arg6 asm("x5") = args.arg3;
 #endif
 
 	if (arg3 == 0) {
@@ -314,6 +321,22 @@ static void* darling_thread_entry(void* p)
 		// Function arguments to push to the stack.
 		[args] "r"(&args), [arg3]"r"(arg3),
 
+		[entry_point] "r"(args.entry_point),
+		[stack_ptr] "r"(stack_ptr)
+	);
+#elif defined(__aarch64__)
+	asm volatile(
+		// Zero out the frame base register.
+		"mov fp, #0\n"
+		// Switch to the new stack.
+		"mov %[stack_ptr], sp\n"
+		// Push a fake return address.
+		"mov lr, #0\n"
+		// Jump to the entry point.
+		"br %[entry_point]" ::
+		// Function arguments
+		"r"(arg1),"r"(arg2),"r"(arg3),"r"(arg4),"r"(arg5),"r"(arg6),
+		
 		[entry_point] "r"(args.entry_point),
 		[stack_ptr] "r"(stack_ptr)
 	);
diff --git a/src/startup/mldr/mldr.c b/src/startup/mldr/mldr.c
index 41491f356..54454ee4f 100644
--- a/src/startup/mldr/mldr.c
+++ b/src/startup/mldr/mldr.c
@@ -383,7 +383,8 @@ static void load_fat(int fd, cpu_type_t forced_arch, bool expect_dylinker, char*
 				if (arch.cputype == CPU_TYPE_X86)
 					best_arch = arch;
 #elif defined (__aarch64__)
-	#error TODO: arm
+				if (arch.cputype == CPU_TYPE_ARM64)
+					best_arch = arch;
 #else
 	#error Unsupported CPU architecture
 #endif
@@ -420,7 +421,7 @@ static void load_fat(int fd, cpu_type_t forced_arch, bool expect_dylinker, char*
 	}
 };
 
-#ifdef __x86_64__
+#if defined(__x86_64__ ) || defined(__aarch64__)
 #define GEN_64BIT
 #include "loader.c"
 #include "stack.c"
@@ -816,9 +817,9 @@ static void setup_space(struct load_results* lr, bool is_64_bit) {
 	// Using the default stack top would cause the stack to be placed just above the commpage
 	// and would collide with it eventually.
 	// Instead, we manually allocate a new stack below the commpage.
-#if __x86_64__
+#if defined(__x86_64__) || defined(__aarch64__)
 	lr->stack_top = commpage_address(true);
-#elif __i386__
+#elif defined(__i386__)
 	lr->stack_top = commpage_address(false);
 #else
 	#error Unsupported architecture
@@ -936,6 +937,15 @@ static void start_thread(struct load_results* lr) {
 		"r"(lr->stack_top)
 		:
 	);
+#elif defined(__aarch64__)
+	__asm__ volatile(
+		"mov sp, %1\n"
+		"br %0"
+		::
+		"r"(lr->entry_point),
+		"r"(lr->stack_top)
+		:
+	);
 #else
 #       error Unsupported platform!
 #endif