diff --git a/lib.rs b/lib.rs
index da193a0..87a9b52 100644
--- a/lib.rs
+++ b/lib.rs
@@ -133,6 +133,13 @@ pub fn cpu_id() -> usize {
     CPU_ID.load(Ordering::Relaxed)
 }
 
+static CPU_COUNT : AtomicUsize = ATOMIC_USIZE_INIT;
+
+#[inline(always)]
+pub fn cpu_count() -> usize {
+    CPU_COUNT.load(Ordering::Relaxed)
+}
+
 pub extern fn userspace_init() {
     assert_eq!(syscall::chdir(b"initfs:bin"), Ok(0));
 
@@ -146,13 +153,14 @@ pub extern fn userspace_init() {
 }
 
 #[no_mangle]
-pub extern fn kmain() {
+pub extern fn kmain(cpus: usize) {
     CPU_ID.store(0, Ordering::SeqCst);
+    CPU_COUNT.store(cpus, Ordering::SeqCst);
 
     context::init();
 
     let pid = syscall::getpid();
-    println!("BSP: {:?}", pid);
+    println!("BSP: {:?} {}", pid, cpus);
 
     match context::contexts_mut().spawn(userspace_init) {
         Ok(context_lock) => {
diff --git a/syscall/process.rs b/syscall/process.rs
index 5d35af6..39dd822 100644
--- a/syscall/process.rs
+++ b/syscall/process.rs
@@ -346,7 +346,7 @@ pub fn clone(flags: usize, stack_base: usize) -> Result<usize> {
                 context.grants = grants;
             } else {
                 // Copy percpu mapping
-                {
+                for cpu_id in 0..::cpu_count() {
                     extern {
                         /// The starting byte of the thread data segment
                         static mut __tdata_start: u8;
@@ -356,7 +356,7 @@ pub fn clone(flags: usize, stack_base: usize) -> Result<usize> {
 
                     let size = unsafe { & __tbss_end as *const _ as usize - & __tdata_start as *const _ as usize };
 
-                    let start = arch::KERNEL_PERCPU_OFFSET + arch::KERNEL_PERCPU_SIZE * ::cpu_id();
+                    let start = arch::KERNEL_PERCPU_OFFSET + arch::KERNEL_PERCPU_SIZE * cpu_id;
                     let end = start + size;
 
                     let start_page = Page::containing_address(VirtualAddress::new(start));