diff --git a/Makefile b/Makefile
index 55146a0..1f05858 100644
--- a/Makefile
+++ b/Makefile
@@ -40,7 +40,7 @@ ifneq ($(arch), x86_64)
 endif
 
 biscuit:
-ifeq ($(arch), x86_64)
+ifeq ($(arch), $(filter $(arch), x86_64 aarch64))
 	@echo Building biscuit programs
 	@mkdir -p biscuit/build
 	@cd biscuit/build && cmake $(cmake_build_args) .. && make
diff --git a/biscuit/CMakeLists.txt b/biscuit/CMakeLists.txt
index 238e1be..b94f1c0 100644
--- a/biscuit/CMakeLists.txt
+++ b/biscuit/CMakeLists.txt
@@ -14,12 +14,12 @@ if (${ARCH} STREQUAL i386)
     if(APPLE)
         set(PREFIX          i386-elf-)
     endif ()
-    set(CMAKE_C_FLAGS       "-m32")
+    set(CMAKE_C_FLAGS       "-m32 -mno-red-zone")
 elseif (${ARCH} STREQUAL x86_64)
     if(APPLE)
         set(PREFIX          x86_64-elf-)
     endif ()
-    set(CMAKE_C_FLAGS       "-m64")
+    set(CMAKE_C_FLAGS       "-m64 -mno-red-zone")
 elseif (${ARCH} STREQUAL riscv32)
     set(PREFIX              riscv64-unknown-elf-)
     set(CMAKE_C_FLAGS       "-march=rv32imac -mabi=ilp32 -mcmodel=medany")
@@ -28,7 +28,6 @@ elseif (${ARCH} STREQUAL riscv64)
     set(CMAKE_C_FLAGS       "-march=rv64imac -mabi=lp64 -mcmodel=medany")
 elseif (${ARCH} STREQUAL aarch64)
     set(PREFIX              aarch64-none-elf-)
-    set(CMAKE_C_FLAGS       "-mgeneral-regs-only")
     set(LINK_FLAGS          "-Ttext 0xffff000000000000")
 else()
     message("Unsupported arch: ${ARCH}")
@@ -36,7 +35,7 @@ endif ()
 set(CMAKE_ASM_COMPILER      ${PREFIX}gcc)
 set(CMAKE_C_COMPILER        ${PREFIX}gcc)
 set(CMAKE_RANLIB            ${PREFIX}ranlib)
-set(CMAKE_C_FLAGS           "${CMAKE_C_FLAGS} -Wall -Werror -MMD -MP -O -g -ffreestanding -nostdlib -nostdinc -fno-builtin -mno-red-zone -fno-stack-protector -fPIC -std=gnu11")
+set(CMAKE_C_FLAGS           "${CMAKE_C_FLAGS} -Wall -Werror -MMD -MP -O -g -ffreestanding -nostdlib -nostdinc -fno-builtin -fno-stack-protector -fPIC -std=gnu11")
 set(CMAKE_ASM_FLAGS         ${CMAKE_C_FLAGS})
 set(CMAKE_C_LINK_FLAGS      "${LINK_FLAGS} -nostdlib")  # override default value to get rid of '-Wl,-search_paths_first -Wl,-headerpad_max_install_names'
 set(CMAKE_SHARED_LIBRARY_LINK_C_FLAGS)                  # override default value to get rid of '-rdynamic' on Linux
@@ -48,5 +47,5 @@ add_library(ulib ${LIBS})
 foreach (PATH ${SRCS})
     get_filename_component(NAME ${PATH} NAME_WE)
     add_executable(${NAME} ${PATH})
-    target_link_libraries(${NAME} ulib)
+    target_link_libraries(${NAME} ulib gcc)
 endforeach ()
diff --git a/biscuit/c/bmsmallfile.c b/biscuit/c/bmsmallfile.c
index a0813c6..b466f5d 100644
--- a/biscuit/c/bmsmallfile.c
+++ b/biscuit/c/bmsmallfile.c
@@ -22,6 +22,7 @@ struct timeval before, end;
 
 void rgettimeofday(struct timeval *h, void *p)
 {
+#if defined(__x86_64__)
 	time_t hi, lo, v;
 	asm(
 	    "rdtsc\n"
@@ -31,6 +32,9 @@ void rgettimeofday(struct timeval *h, void *p)
 	v = hi << 32 | lo;
 	if (h)
 		h->tv_usec = v;
+#else
+	// TODO: aarch64
+#endif
 }
 
 void start()
@@ -117,7 +121,7 @@ write_test(char *name, int n, int size)
       exit(1);
     }
   }
-    
+
   if ((r = close(fd)) < 0) {
     printf("%s: close failed %d %d\n", prog_name, r, errno);
   }
@@ -164,13 +168,13 @@ read_test(int n, int size)
   printf("%s: read took %ld usec\n", prog_name, time);
 }
 
-void 
+void
 delete_test(int n)
-{	
+{
   int i;
   int r;
   int j;
- 
+
   start();
   for (i = 0, j = 0; i < n; i ++) {
 
@@ -208,11 +212,11 @@ int main(int argc, char *argv[])
   printf("%s %d %d %s\n", prog_name, n, size, topdir);
 
   creat_dir();
-  
+
   //printstats(topdir, 1);
-  
+
   creat_test(n, size);
-  
+
   //printstats(topdir, 0);
 
   read_test(n, size);
diff --git a/biscuit/c/libs/litc.c b/biscuit/c/libs/litc.c
index 24d9d5e..eeb60a8 100644
--- a/biscuit/c/libs/litc.c
+++ b/biscuit/c/libs/litc.c
@@ -86,13 +86,20 @@ static struct kinfo_t *kinfo;
 // stack is not used after munmapping it, but before calling exit(2). we use
 // this macro to make sure the clobbers are coherent for these three pieces of
 // code using syscalls.
+#if defined(__x86_64__)
 #define SYSCALL_CLOBBERS "cc", "memory", "r9", "r10", "r11", "r12", "r13", \
 			 "r14", "r15"
+#elif defined(__aarch64__)
+#define SYSCALL_CLOBBERS "cc", "memory"
+#endif
+
 long
 syscall(long a1, long a2, long a3, long a4,
     long a5, long trap)
 {
 	long ret;
+
+#if defined(__x86_64__)
 	register long r8 asm("r8") = a5;
 
 	// we may want to follow the sys5 abi and have the kernel restore
@@ -104,6 +111,20 @@ syscall(long a1, long a2, long a3, long a4,
 		: "=a"(ret)
 		: "0"(trap), "D"(a1), "S"(a2), "d"(a3), "c"(a4), "r"(r8)
 		: SYSCALL_CLOBBERS);
+#elif defined(__aarch64__)
+	register long x8 asm("x8") = trap;
+	register long x0 asm("x0") = a1;
+	register long x1 asm("x1") = a2;
+	register long x2 asm("x2") = a3;
+	register long x3 asm("x3") = a4;
+	register long x4 asm("x4") = a5;
+
+	asm volatile(
+		"svc	0"
+		: "=r"(ret)
+		: "r"(x8), "0"(x0), "r"(x1), "r"(x2), "r"(x3), "r"(x4)
+		: SYSCALL_CLOBBERS);
+#endif
 
 	return ret;
 }
@@ -837,6 +858,7 @@ tfork_thread(struct tfork_t *args, long (*fn)(void *), void *fnarg)
 	int tid;
 	long flags = FORK_THREAD;
 
+#if defined(__x86_64__)
 	// rbx and rbp are preserved across syscalls. i don't know how to
 	// specify rbp as a register contraint.
 	register ulong rbp asm("rbp") = (ulong)fn;
@@ -859,6 +881,30 @@ tfork_thread(struct tfork_t *args, long (*fn)(void *), void *fnarg)
 	    : "=a"(tid)
 	    : "D"(args), "S"(flags), "0"(SYS_FORK), "r"(rbp), "r"(rbx)
 	    : SYSCALL_CLOBBERS);
+#elif defined(__aarch64__)
+	// all registers are preserved across syscalls for aarch64.
+	register ulong x8 asm("x8") = SYS_FORK;
+	register ulong x0 asm("x0") = (ulong)args;
+	register ulong x1 asm("x1") = flags;
+
+	asm volatile(
+	    "svc	0\n"
+	    "cmp	x0, #0\n"
+	    // parent or error
+	    "b.ne	1f\n"
+	    // child
+	    "ldr	x0, %5\n"
+	    "ldr	x9, %4\n"
+	    "blr	x9\n"
+	    "bl 	tfork_done\n"
+	    "mov	x0, #0\n"
+	    "str	xzr, [x0]\n"
+	    "1:\n"
+	    : "=r"(tid)
+	    : "r"(x8), "0"(x0), "r"(x1), "m"(fn), "m"(fnarg)
+	    : SYSCALL_CLOBBERS);
+#endif
+
 	return tid;
 }
 
@@ -907,6 +953,7 @@ _pcreate(void *vpcarg)
 	status = (long)(pcargs.fn(pcargs.arg));
 	free(pcargs.tls);
 
+#if defined(__x86_64__)
 	// rbx and rbp are preserved across syscalls. i don't know how to
 	// specify rbp as a register contraint.
 	register ulong rbp asm("rbp") = SYS_THREXIT;
@@ -929,6 +976,29 @@ _pcreate(void *vpcarg)
 	    : "a"(SYS_MUNMAP), "D"(pcargs.stack), "S"(pcargs.stksz),
 	      "r"(rbp), "r"(rbx)
 	    : SYSCALL_CLOBBERS);
+#elif defined(__aarch64__)
+	register ulong x8 asm("x8") = SYS_MUNMAP;
+	register ulong x0 asm("x0") = (ulong)pcargs.stack;
+	register ulong x1 asm("x1") = (ulong)pcargs.stksz;
+
+	asm volatile(
+	    "svc	0\n"
+	    "cmp	x0, #0\n"
+	    "b.eq	1f\n"
+	    "mov	x0, #0\n"
+	    "str	xzr, [x0]\n"
+	    "1:\n"
+	    "mov	x8, %3\n"
+	    "ldr	x0, %4\n"
+	    "svc	0\n"
+	    "mov	x0, #1\n"
+	    "str	xzr, [x0]\n"
+	    :
+	    : "r"(x8), "r"(x0), "r"(x1),
+	      "X"(SYS_THREXIT), "m"(status)
+	    : SYSCALL_CLOBBERS);
+#endif
+
 	// not reached
 	return 0;
 }
@@ -1107,7 +1177,11 @@ pthread_barrier_wait(pthread_barrier_t *b)
 		uint o = b->current;
 		uint n = o + 1;
 		if ((o & m) != 0) {
+#if defined(__x86_64__)
 			asm volatile("pause":::"memory");
+#elif defined(__aarch64__)
+			asm volatile("yield":::"memory");
+#endif
 			continue;
 		}
 		c = n;
@@ -1126,7 +1200,11 @@ pthread_barrier_wait(pthread_barrier_t *b)
 	}
 
 	while ((b->current & m) == 0)
+#if defined(__x86_64__)
 		asm volatile("pause":::"memory");
+#elif defined(__aarch64__)
+		asm volatile("yield":::"memory");
+#endif
 
 	c = __sync_add_and_fetch(&b->current, -1);
 	if (c == m)
@@ -2159,7 +2237,7 @@ DIR *
 fdopendir(int fd)
 {
         #define BSIZE 4096
-  
+
 	struct stat st;
 	if (fstat(fd, &st) == -1)
 		return NULL;
@@ -2785,6 +2863,7 @@ sscanf(const char *src, const char *fmt, ...)
 ulong
 rdtsc(void)
 {
+#if defined(__x86_64__)
 	ulong low, hi;
 	asm volatile(
 	    "rdtsc\n"
@@ -2792,6 +2871,10 @@ rdtsc(void)
 	    :
 	    :);
 	return hi << 32 | low;
+#else
+	// TODO: aarch64
+	return 0;
+#endif
 }
 
 static char readlineb[256];
@@ -3534,17 +3617,25 @@ __start(int argc, char **argv, struct kinfo_t *k)
 void
 _start(void)
 {
+#if defined(__x86_64__)
 	// make sure that the stack is 16-byte aligned, as gcc assumes, after
 	// _start's function prologue. gcc emits SSE instructions that require
 	// 16-byte alignment (misalignment generates #GP).
 	asm(
 		"movq   (%%rsp), %%rdi\n"  // argc
 		"leaq   8(%%rsp), %%rsi\n" // argv
-	    "andq	$0xfffffffffffffff0, %%rsp\n"
-	    "subq	$8, %%rsp\n"
-	    "movabs 	$__start, %%rax\n"
-	    "jmpq	*%%rax\n"
-	    ::: "memory", "cc");
+		"andq	$0xfffffffffffffff0, %%rsp\n"
+		"subq	$8, %%rsp\n"
+		"movabs 	$__start, %%rax\n"
+		"jmpq	*%%rax\n"
+		::: "memory", "cc");
+#elif defined(__aarch64__)
+	asm(
+		"ldr	x0, [sp]\n"   // argc
+		"add	x1, sp, #8\n" // argv
+		"bl 	__start\n"
+		::: "memory", "cc");
+#endif
 }
 
 /* NGINX STUFF */
diff --git a/biscuit/c/pthtests.c b/biscuit/c/pthtests.c
index 252c29e..271c5db 100644
--- a/biscuit/c/pthtests.c
+++ b/biscuit/c/pthtests.c
@@ -48,7 +48,11 @@ void *groupfault(void *a)
 	int b = (int)(long)a;
 
 	while (!go)
-		asm volatile("pause");
+#if defined(__x86_64__)
+		asm volatile("pause":::"memory");
+#elif defined(__aarch64__)
+		asm volatile("yield":::"memory");
+#endif
 
 	blah = b;
 
diff --git a/biscuit/c/sfork.c b/biscuit/c/sfork.c
index 9535682..e2a8a31 100644
--- a/biscuit/c/sfork.c
+++ b/biscuit/c/sfork.c
@@ -118,6 +118,7 @@ pid_t
 _getppid(void)
 {
 	pid_t ret;
+#if defined(__x86_64__)
 	asm volatile(
 		"movq	%%rsp, %%r10\n"
 		"leaq	2(%%rip), %%r11\n"
@@ -125,6 +126,14 @@ _getppid(void)
 		: "=a"(ret)
 		: "0"(40ul)
 		: "cc", "memory", "r9", "r10", "r11", "edi", "esi", "edx", "ecx", "r8");
+#elif defined(__aarch64__)
+	register long x8 asm("x8") = 40ul;
+	asm volatile(
+		"svc	0"
+		: "=r"(ret)
+		: "r"(x8)
+		: "cc", "memory");
+#endif
 	return ret;
 }
 
@@ -135,6 +144,7 @@ void *igetpids(void *idp)
 
 	long total = 0;
 	while (!cease) {
+#if defined(__x86_64__)
 		asm volatile(
 			"movl	$40, %%eax\n"
 			"movq	%%rsp, %%r10\n"
@@ -143,6 +153,14 @@ void *igetpids(void *idp)
 			:
 			:
 			: SYSCALL_CLOBBERS, "eax", "edi", "esi", "edx", "ecx", "r8");
+#elif defined(__aarch64__)
+		asm volatile(
+			"mov	x8, #40\n"
+			"svc	0"
+			:
+			:
+			: "cc", "memory");
+#endif
 		total++;
 	}
 	return (void *)total;
@@ -692,10 +710,14 @@ void *locks(void *_arg)
 	pthread_barrier_wait(&bar);
 
 	while (!cease) {
+#if defined(__x86_64__)
 		asm("lock incq %0\n"
 			:
 			: "m"(tot)
 			: "cc", "memory");
+#else
+	// TODO: aarch64
+#endif
 	}
 	return (void *)tot;
 }
diff --git a/biscuit/c/usertests.c b/biscuit/c/usertests.c
index ae01e23..bb47c67 100644
--- a/biscuit/c/usertests.c
+++ b/biscuit/c/usertests.c
@@ -568,7 +568,7 @@ fourfiles(void)
   for(pi = 0; pi < 4; pi++){
     fname = names[pi];
     unlink(fname);
-      
+
     pid = fork();
     if(pid < 0){
       printf("fork failed\n");
@@ -581,7 +581,7 @@ fourfiles(void)
         printf("create failed\n");
         exit(0);
       }
-      
+
       memset(buf, '0'+pi, BSIZE);
       for(i = 0; i < 12; i++){
         if((n = write(fd, buf, BSIZE)) != BSIZE){
@@ -938,7 +938,7 @@ linkunlink()
 
   if(pid)
     wait(NULL);
-  else 
+  else
     exit(0);
 
   printf("linkunlink ok\n");
@@ -1007,7 +1007,7 @@ subdir(void)
   }
   write(fd, "ff", 2);
   close(fd);
-  
+
   if(rmdir("dd") >= 0){
     printf("unlink dd (non-empty dir) succeeded!\n");
     exit(0);
@@ -1231,7 +1231,7 @@ bigfile(void)
   fsync(fd);
   close(fd);
 
-  
+
   fd = open("bigfile", 0);
   if(fd < 0){
     printf("cannot open bigfile\n");
@@ -1466,7 +1466,7 @@ forktest(void)
     if(pid == 0)
       exit(0);
   }
-  
+
   if(n == ulim){
     printf("fork claimed to work %d times!\n", ulim);
     exit(0);
@@ -1489,7 +1489,7 @@ forktest(void)
     if (pthread_create(&t[n], NULL, _ptexit, NULL))
       break;
   }
-  
+
   if(n == ulim)
     errx(-1, "pthread_create claimed to work %d times!\n", ulim);
 
@@ -1517,7 +1517,7 @@ forktest(void)
 //  // can one sbrk() less than a page?
 //  a = sbrk(0);
 //  int i;
-//  for(i = 0; i < 5000; i++){ 
+//  for(i = 0; i < 5000; i++){
 //    b = sbrk(1);
 //    if(b != a){
 //      printf("sbrk test failed %d %x %x\n", i, a, b);
@@ -1546,7 +1546,7 @@ forktest(void)
 //  a = sbrk(0);
 //  amt = (BIG) - (uint)a;
 //  p = sbrk(amt);
-//  if (p != a) { 
+//  if (p != a) {
 //    printf("sbrk test failed to grow big address space; enough phys mem?\n");
 //    exit(0);
 //  }
@@ -1585,7 +1585,7 @@ forktest(void)
 //    printf("sbrk downsize failed, a %x c %x\n", a, c);
 //    exit(0);
 //  }
-//  
+//
 //  // can we read the kernel's memory?
 //  for(a = (char*)(KERNBASE); a < (char*) (KERNBASE+2000000); a += 50000){
 //    ppid = getpid();
@@ -1643,14 +1643,26 @@ void
 validateint(int *p)
 {
 	ulong ret;
+
+#define SYS_PIPE2         293
+#if defined(__x86_64__)
 	asm volatile(
 		"movq	%%rsp, %%r10\n"
 		"leaq	2(%%rip), %%r11\n"
 		"sysenter\n"
 		: "=a"(ret)
-#define SYS_PIPE2         293
 		: "0"(SYS_PIPE2), "D"(p)
 		: "cc", "memory");
+#elif defined(__aarch64__)
+	register ulong x8 asm("x8") = SYS_PIPE2;
+	register ulong x0 asm("x0") = (ulong)p;
+	asm volatile(
+		"svc 0\n"
+		: "=r"(ret)
+		: "r"(x8), "0"(x0)
+		: "cc", "memory");
+#endif
+
 	if (ret == 0)
 		errx(-1, "bad int passed?");
 }
@@ -2789,7 +2801,11 @@ static volatile int go;
 static void *_locker(void *v)
 {
 	while (go != 1)
+#if defined(__x86_64__)
 		asm volatile("pause\n":::"memory");
+#elif defined(__aarch64__)
+		asm volatile("yield\n":::"memory");
+#endif
 	pthread_mutex_t *m = (pthread_mutex_t *)v;
 	int i;
 	for (i = 0; i < ltimes; i++) {
@@ -2904,7 +2920,11 @@ static void _condtest(const int nt)
 		if (pthread_create(&t[i], NULL, _condsleep, &args[i]))
 			errx(-1, "pthread_ create");
 		while (go == 0)
+#if defined(__x86_64__)
 			asm volatile("pause\n":::"memory");
+#elif defined(__aarch64__)
+			asm volatile("yield\n":::"memory");
+#endif
 	}
 
 	for (i = 0; i < nt; i++)
@@ -2953,7 +2973,11 @@ static void _condbctest(const int nt)
 	for (i = 0; i < bctimes; i++) {
 		volatile int *p = &lcounter;
 		while (*p < enext)
+#if defined(__x86_64__)
 			asm volatile("pause\n":::"memory");
+#elif defined(__aarch64__)
+			asm volatile("yield\n":::"memory");
+#endif
 		if (pthread_mutex_lock(&m))
 			err(-1, "lock");
 		if (i == bctimes - 1)
@@ -2985,7 +3009,7 @@ int pthreadsharedfd;
 void *threadfd(void *arg)
 {
   long n = (long) arg;
-  
+
   for (int i = 0; i < 1000; i++) {
     if (n == 0) {
       pthreadsharedfd = open("sharedfdf", O_CREATE|O_RDWR);
@@ -3007,13 +3031,13 @@ void _pthreadfd(void) {
   const int nthreads = 2;
 
   printf("pthread shared fd\n");
-	
+
   int i;
   pthread_t t[nthreads];
   for (int i = 0; i < nthreads; i++) {
     if (pthread_create(&t[i], NULL, threadfd, (void *) (long) i))
       errx(-1, "pthread create");
-  }    
+  }
   for (i = 0; i < nthreads; i++) {
     if (pthread_join(t[i], NULL))
       errx(-1, "pthread join");
@@ -3620,7 +3644,7 @@ logtest()
   }
 
   free(buf);
-    
+
   printf("log test OK\n");
 }
 
@@ -3835,9 +3859,9 @@ main(int argc, char *argv[])
   posixtest();
   barriertest();
   threadwait();
-  
+
   pthreadfd();
-  
+
   fnonblock();
   preadwrite();
   stdiotest();
diff --git a/biscuit/cxx/libutil.cc b/biscuit/cxx/libutil.cc
index 48c151d..b5b1d3c 100644
--- a/biscuit/cxx/libutil.cc
+++ b/biscuit/cxx/libutil.cc
@@ -101,6 +101,7 @@ setaffinity(int c)
 ulong
 rdtsc(void)
 {
+#if defined(__x86_64__)
 	ulong low, hi;
 	asm volatile(
 	    "rdtsc\n"
@@ -108,5 +109,9 @@ rdtsc(void)
 	    :
 	    :);
 	return hi << 32 | low;
+#else
+	// TODO: aarch64
+	return 0;
+#endif
 }
 #endif
diff --git a/biscuit/cxx/spinbarrier.hh b/biscuit/cxx/spinbarrier.hh
index 3447589..1c54271 100644
--- a/biscuit/cxx/spinbarrier.hh
+++ b/biscuit/cxx/spinbarrier.hh
@@ -39,7 +39,11 @@ public:
   {
     // Wait if the barrier is in the exit phase
     while (entered_ & phase_mask)
+#if defined(__x86_64__)
       asm volatile("pause":::);
+#elif defined(__aarch64__)
+      asm volatile("yield":::);
+#endif
 
     // Enter the barrier
     auto v = ++entered_;
@@ -52,7 +56,11 @@ public:
 
     // Wait until the barrier switches to the exit phase
     while (!(entered_.load(std::memory_order_relaxed) & phase_mask))
+#if defined(__x86_64__)
       asm volatile("pause":::);
+#elif defined(__aarch64__)
+      asm volatile("yield":::);
+#endif
 
     // Exit the batter
     if ((v = --entered_) == phase_mask)