diff --git a/src/Enclave.edl b/src/Enclave.edl
index eed15d97..48a620a0 100644
--- a/src/Enclave.edl
+++ b/src/Enclave.edl
@@ -131,6 +131,10 @@ enclave {
 
         int occlum_ocall_mprotect([user_check] void* addr, size_t len, int prot);
 
+        int occlum_ocall_get_numa_topology(
+            [out, count=ncpus] uint32_t *numa_buf,
+            size_t ncpus
+        ) propagate_errno;
         void occlum_ocall_sched_yield(void);
         int occlum_ocall_sched_setaffinity(
             int host_tid,
diff --git a/src/libos/src/sched/cpu_set.rs b/src/libos/src/sched/cpu_set.rs
index 6314928a..bdedf415 100644
--- a/src/libos/src/sched/cpu_set.rs
+++ b/src/libos/src/sched/cpu_set.rs
@@ -57,6 +57,11 @@ impl CpuSet {
         self.bits.count_ones()
     }
 
+    /// Returns the first index of CPUs in set.
+    pub fn first_cpu_idx(&self) -> Option<usize> {
+        self.iter().position(|&b| b == true)
+    }
+
     // Returns if the CpuSet is a subset of available cpu set
     pub fn is_subset_of(&self, other: &CpuSet) -> bool {
         (self.bits.clone() & other.bits.clone()) == self.bits
diff --git a/src/libos/src/sched/do_getcpu.rs b/src/libos/src/sched/do_getcpu.rs
new file mode 100644
index 00000000..0ff597fb
--- /dev/null
+++ b/src/libos/src/sched/do_getcpu.rs
@@ -0,0 +1,43 @@
+use super::cpu_set::{CpuSet, NCORES};
+use crate::prelude::*;
+use crate::process::ThreadRef;
+
+pub fn do_getcpu() -> Result<(u32, u32)> {
+    let cpu = pick_cpu_within_affinity_mask();
+    let node = NUMA_TOPOLOGY[cpu as usize];
+    debug!("do_getcpu cpu = {}, node = {}", cpu, node);
+    Ok((cpu, node))
+}
+
+fn pick_cpu_within_affinity_mask() -> u32 {
+    // Always return the idx of the first bit in the affnity mask for now.
+    // TODO: randomly choose a bit in the affinity mask.
+    let thread = current!();
+    let sched = thread.sched().lock().unwrap();
+    let idx = sched.affinity().first_cpu_idx().unwrap();
+    idx as u32
+}
+
+fn validate_numa_topology(numa_topology: &Vec<u32>) -> Result<()> {
+    for node_id in numa_topology.iter() {
+        if *node_id >= numa_topology.len() as u32 {
+            return_errno!(EINVAL, "NUMA node id exceeds the core numbers");
+        }
+    }
+    Ok(())
+}
+
+lazy_static! {
+    /// The information of Non-Uniform Memory Access(NUMA) topology
+    pub static ref NUMA_TOPOLOGY: Vec<u32> = {
+        extern "C" {
+            fn occlum_ocall_get_numa_topology(ret: *mut i32, numa_buf: *mut u32, ncpus: usize) -> sgx_status_t;
+        }
+        let mut numa_topology = vec![0; *NCORES];
+        let mut retval: i32 = 0;
+        let status = unsafe { occlum_ocall_get_numa_topology(&mut retval, numa_topology.as_mut_ptr(), numa_topology.len()) };
+        assert!(status == sgx_status_t::SGX_SUCCESS);
+        validate_numa_topology(&numa_topology).expect("ocall returned invalid NUMA topology");
+        numa_topology
+    };
+}
diff --git a/src/libos/src/sched/mod.rs b/src/libos/src/sched/mod.rs
index 466a7cec..6e6fb49c 100644
--- a/src/libos/src/sched/mod.rs
+++ b/src/libos/src/sched/mod.rs
@@ -1,5 +1,6 @@
 /// CPU scheduling for threads.
 mod cpu_set;
+mod do_getcpu;
 mod do_sched_affinity;
 mod do_sched_yield;
 mod sched_agent;
diff --git a/src/libos/src/sched/syscalls.rs b/src/libos/src/sched/syscalls.rs
index bb96340e..83f584e1 100644
--- a/src/libos/src/sched/syscalls.rs
+++ b/src/libos/src/sched/syscalls.rs
@@ -56,3 +56,30 @@ pub fn do_sched_setaffinity(pid: pid_t, buf_size: size_t, buf_ptr: *const u8) ->
     super::do_sched_affinity::do_sched_setaffinity(pid, affinity)?;
     Ok(0)
 }
+
+pub fn do_getcpu(cpu_ptr: *mut u32, node_ptr: *mut u32) -> Result<isize> {
+    // Do pointers check
+    match (cpu_ptr.is_null(), node_ptr.is_null()) {
+        (true, true) => return Ok(0),
+        (false, true) => check_mut_ptr(cpu_ptr)?,
+        (true, false) => check_mut_ptr(node_ptr)?,
+        (false, false) => {
+            check_mut_ptr(cpu_ptr)?;
+            check_mut_ptr(node_ptr)?;
+        }
+    }
+    // Call the memory-safe do_getcpu
+    let (cpu, node) = super::do_getcpu::do_getcpu()?;
+    // Copy to user
+    if !cpu_ptr.is_null() {
+        unsafe {
+            cpu_ptr.write(cpu);
+        }
+    }
+    if !node_ptr.is_null() {
+        unsafe {
+            node_ptr.write(node);
+        }
+    }
+    Ok(0)
+}
diff --git a/src/libos/src/syscall/mod.rs b/src/libos/src/syscall/mod.rs
index 77b4a9d5..3a384265 100644
--- a/src/libos/src/syscall/mod.rs
+++ b/src/libos/src/syscall/mod.rs
@@ -40,7 +40,7 @@ use crate::process::{
     do_getpgid, do_getpid, do_getppid, do_gettid, do_getuid, do_prctl, do_set_tid_address,
     do_spawn, do_wait4, pid_t, FdOp, ThreadStatus,
 };
-use crate::sched::{do_sched_getaffinity, do_sched_setaffinity, do_sched_yield};
+use crate::sched::{do_getcpu, do_sched_getaffinity, do_sched_setaffinity, do_sched_yield};
 use crate::signal::{
     do_kill, do_rt_sigaction, do_rt_sigpending, do_rt_sigprocmask, do_rt_sigreturn, do_sigaltstack,
     do_tgkill, do_tkill, sigaction_t, sigset_t, stack_t,
@@ -388,7 +388,7 @@ macro_rules! process_syscall_table_with_callback {
             (Syncfs = 306) => handle_unsupported(),
             (Sendmmsg = 307) => handle_unsupported(),
             (Setns = 308) => handle_unsupported(),
-            (Getcpu = 309) => handle_unsupported(),
+            (Getcpu = 309) => do_getcpu(cpu_ptr: *mut u32, node_ptr: *mut u32),
             (ProcessVmReadv = 310) => handle_unsupported(),
             (ProcessVmWritev = 311) => handle_unsupported(),
             (Kcmp = 312) => handle_unsupported(),
diff --git a/src/pal/src/ocalls/sched.c b/src/pal/src/ocalls/sched.c
index c7213835..a9a0c4b5 100644
--- a/src/pal/src/ocalls/sched.c
+++ b/src/pal/src/ocalls/sched.c
@@ -1,5 +1,6 @@
 #define _GNU_SOURCE
 #include <sched.h>
+#include <dirent.h>
 #include <unistd.h>
 #include "ocalls.h"
 
@@ -26,3 +27,56 @@ void occlum_ocall_sched_yield(void) {
 int occlum_ocall_ncores(void) {
     return sysconf(_SC_NPROCESSORS_CONF);
 }
+
+static int is_number(const char *str) {
+    size_t len = strlen(str);
+    for (size_t i = 0; i < len; i++) {
+        if (str[i] >= '0' && str[i] <= '9') {
+            continue;
+        }
+        return 0;
+    }
+    return len > 0;
+}
+
+static int is_node_entry(struct dirent *d) {
+    return
+        d &&
+        strncmp(d->d_name, "node", 4) == 0 &&
+        is_number(d->d_name + 4);
+}
+
+// The information about NUMA topology is stored in sysfs.
+// By reading the node entries(node<id>) in /sys/devices/system/cpu/cpu<id>,
+// we can find which cpu core locates at which NUMA node.
+int occlum_ocall_get_numa_topology(uint32_t *numa_buf, size_t ncpus) {
+    uint32_t *ptr = numa_buf;
+    for (size_t i = 0; i < ncpus; i++) {
+        struct dirent *d;
+        char cpu_dir_path[128] = { 0 };
+        int ret = snprintf(cpu_dir_path, sizeof(cpu_dir_path), "/sys/devices/system/cpu/cpu%ld",
+                           i);
+        if (ret < 0 || ret >= sizeof(cpu_dir_path)) {
+            return -1;
+        }
+        DIR *dir = opendir(cpu_dir_path);
+        if (dir == NULL) {
+            return -1;
+        }
+        while ((d = readdir(dir))) {
+            if (is_node_entry(d)) {
+                errno = 0;
+                int node_id = strtol((d->d_name) + 4, (char **)NULL, 10);
+                if (errno) {
+                    closedir(dir);
+                    return -1;
+                }
+                *ptr = node_id;
+                break;
+            }
+        }
+        closedir(dir);
+        ptr++;
+    }
+    return 0;
+}
diff --git a/test/sched/main.c b/test/sched/main.c
index 960597a1..6b7f1df4 100644
--- a/test/sched/main.c
+++ b/test/sched/main.c
@@ -252,6 +252,57 @@ static int test_sched_yield() {
     return 0;
 }
 
+// ============================================================================
+// Test cases for getcpu
+// ============================================================================
+
+static int test_getcpu() {
+    int cpu, node;
+    if (syscall(__NR_getcpu, &cpu, &node, NULL) < 0) {
+        THROW_ERROR("getcpu with cpu&node fail");
+    }
+    if (syscall(__NR_getcpu, &cpu, NULL, NULL) < 0) {
+        THROW_ERROR("getcpu with cpu fail");
+    }
+    if (syscall(__NR_getcpu, NULL, &node, NULL) < 0) {
+        THROW_ERROR("getcpu with node fail");
+    }
+    if (syscall(__NR_getcpu, NULL, NULL, NULL) < 0) {
+        THROW_ERROR("getcpu with null fail");
+    }
+    return 0;
+}
+
+static int test_getcpu_after_setaffinity() {
+    int nproc = sysconf(_SC_NPROCESSORS_ONLN);
+    cpu_set_t mask_old;
+    CPU_ZERO(&mask_old);
+    for (int i = 0; i < nproc; ++i) {
+        CPU_SET(g_online_cpu_idxs[i], &mask_old);
+    }
+
+    cpu_set_t mask;
+    CPU_ZERO(&mask);
+    CPU_SET(g_online_cpu_idxs[0], &mask);
+    if (sched_setaffinity(0, sizeof(cpu_set_t), &mask) < 0) {
+        THROW_ERROR("failed to call sched_setaffinity \n");
+    }
+
+    int cpu;
+    int ret = syscall(__NR_getcpu, &cpu, NULL, NULL);
+    if (ret < 0) {
+        THROW_ERROR("getcpu fail");
+    }
+    if (cpu != g_online_cpu_idxs[0]) {
+        THROW_ERROR("check processor id fail");
+    }
+
+    if (sched_setaffinity(0, sizeof(cpu_set_t), &mask_old) < 0) {
+        THROW_ERROR("recover cpuset error");
+    }
+    return 0;
+}
+
 // ============================================================================
 // Test suite main
 // ============================================================================
@@ -268,6 +319,8 @@ static test_case_t test_cases[] = {
     TEST_CASE(test_sched_setaffinity_with_null_buffer),
     TEST_CASE(test_sched_yield),
     TEST_CASE(test_sched_xetaffinity_children_inheritance),
+    TEST_CASE(test_getcpu),
+    TEST_CASE(test_getcpu_after_setaffinity),
 };
 
 int main() {