diff --git a/apps/maprobe/Makefile b/apps/maprobe/Makefile
index c36a3029bc23756a1bee6417d9c3083c8b3fe60b..2bcc359067f8209ebe02181e70d96f3442148a12 100644
--- a/apps/maprobe/Makefile
+++ b/apps/maprobe/Makefile
@@ -1,3 +1,3 @@
 NAME = maprobe
-SRCS = maprobe.c
+SRCS = common.c bitutils.c latency-test.c main.c
 include $(AM_HOME)/Makefile.app
diff --git a/apps/maprobe/README.md b/apps/maprobe/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..e6f712220831019e2bb7a24bafd69022642c266b
--- /dev/null
+++ b/apps/maprobe/README.md
@@ -0,0 +1,3 @@
+# Micro Architecture Probe (MAProbe)
+
+* Memory access latency test
\ No newline at end of file
diff --git a/apps/maprobe/bitutils.c b/apps/maprobe/bitutils.c
new file mode 100644
index 0000000000000000000000000000000000000000..cfad50f8ce951e25f393a6dd8befeaca3d189614
--- /dev/null
+++ b/apps/maprobe/bitutils.c
@@ -0,0 +1,36 @@
+#include "bitutils.h" 
+
+inline uint64_t _perf_get_bit_mask(int low, int high) {
+    assert(low < high);
+    assert(low >= 0);
+    assert(high < 63);
+    return ((1 << high) - 1) >> low << low;
+}
+
+inline uint64_t _perf_get_bits(uint64_t raw_data, int low, int high) {
+    assert(low < high);
+    assert(low >= 0);
+    assert(high < 63);
+    uint64_t mask = (1 << high) - 1;
+    return (raw_data & mask) >> low;
+}
+
+inline uint64_t _perf_get_bit(uint64_t raw_data, int position) {
+    assert(position >= 0);
+    assert(position <= 63);
+    return (raw_data >> position) & 1;
+}
+
+inline uint64_t _perf_set_bits(uint64_t raw_data, int low, int high, uint64_t new_value) {
+    assert(low < high);
+    assert(low >= 0);
+    assert(high < 63);
+    int mask = _perf_get_bit_mask(low, high);
+    return (raw_data & mask) | ((new_value << low) & mask);
+}
+
+inline uint64_t _perf_set_bit(uint64_t raw_data, int position, int new_value) {
+    assert(position >= 0);
+    assert(position <= 63);
+    return raw_data & ((new_value & 1) << position);
+}
diff --git a/apps/maprobe/common.c b/apps/maprobe/common.c
new file mode 100644
index 0000000000000000000000000000000000000000..9cd55a622ce606e24b641858f26c4e61c0eac56a
--- /dev/null
+++ b/apps/maprobe/common.c
@@ -0,0 +1,48 @@
+#include "maprobe.h"
+
+struct perf perf;
+
+void _perf_start_timer()
+{
+#ifndef PERF_SIM
+    perf.cycle = csr_read(CSR_MCYCLE);
+    perf.instrcnt = csr_read(CSR_MINSTRET);
+#endif
+}
+
+void _perf_end_timer()
+{
+#ifndef PERF_SIM
+    perf.cycle = csr_read(CSR_MCYCLE) - perf.cycle;
+    perf.instrcnt = csr_read(CSR_MINSTRET) - perf.instrcnt;
+#endif
+}
+
+void _perf_print_timer()
+{
+    printf("cycle %d inst %d ipc %lf\n", perf.cycle, perf.instrcnt, (float)perf.instrcnt/perf.cycle);
+}
+
+void _perf_calibrate()
+{
+#ifndef PERF_SIM
+    // csr read delay
+    uint64_t cycle_1 = csr_read(CSR_MCYCLE);
+    uint64_t cycle_2 = csr_read(CSR_MCYCLE);
+    perf.csr_read_cycle = cycle_2-cycle_1;
+    printf("perf_calibrate: csr_read_cycle %d\n", perf.csr_read_cycle);
+
+    // csr read inst cost
+    uint64_t inst_1 = csr_read(CSR_MINSTRET);
+    uint64_t inst_2 = csr_read(CSR_MINSTRET);
+    perf.csr_read_ninst = inst_2-inst_1;
+    printf("perf_calibrate: csr_read_ninst %d\n", perf.csr_read_ninst);
+#else
+    printf("running in simulation environment, hpm read disabled\n");
+#endif
+}
+
+void _perf_blackhole(uint64_t value)
+{
+    *(uint64_t*) _PERF_BLACKHOLE = value;
+}
diff --git a/apps/maprobe/include/bitutils.h b/apps/maprobe/include/bitutils.h
new file mode 100644
index 0000000000000000000000000000000000000000..f3b7f54be8711d5558f8905af664d85c3012972f
--- /dev/null
+++ b/apps/maprobe/include/bitutils.h
@@ -0,0 +1,14 @@
+// bit op utils for perf
+
+#ifndef PROBE_BITUTILS_H
+#define PROBE_BITUTILS_H
+
+#include <klib.h>
+
+extern uint64_t _perf_get_bit_mask(int low, int high);
+extern uint64_t _perf_get_bits(uint64_t raw_data, int low, int high);
+extern uint64_t _perf_get_bit(uint64_t raw_data, int position);
+extern uint64_t _perf_set_bits(uint64_t raw_data, int low, int high, uint64_t new_value);
+extern uint64_t _perf_set_bit(uint64_t raw_data, int position, int new_value);
+
+#endif
\ No newline at end of file
diff --git a/apps/maprobe/include/maprobe.h b/apps/maprobe/include/maprobe.h
index ace1682c2e3f023e9c2dd7cef841e3ed2b43064b..27a170dae43cb68f3c6c15944e4f0bf06b4f0015 100644
--- a/apps/maprobe/include/maprobe.h
+++ b/apps/maprobe/include/maprobe.h
@@ -5,6 +5,10 @@
 
 #include <klib.h>
 #include <csr.h>
+#include "bitutils.h"
+
+// config
+// #define PERF_SIM // probe run in simulatior, diaable perf counters
 
 // perf const
 #define BYTE (1)
@@ -13,14 +17,16 @@
 #define GB (1024*MB)
 
 // platform dependent const
-// #define _PERF_TEST_ADDR_BASE 0x80400000
-#define _PERF_TEST_ADDR_BASE 0x2000400000
+#ifndef _PERF_TEST_ADDR_BASE
+#define _PERF_TEST_ADDR_BASE 0x80400000
+// #define _PERF_TEST_ADDR_BASE 0x2000400000
+#endif
 #define _PERF_CACHELINE_SIZE_BYTE (64 * BYTE)
 #define _PERF_L1_NOALIAS_SIZE_BYTE (32 * KB)
-#define _PERF_L1_SIZE_BYTE (128 * KB)
+#define _PERF_L1_SIZE_BYTE (64 * KB)
 #define _PERF_L2_SIZE_BYTE (512 * KB)
 #define _PERF_L3_SIZE_BYTE (2 * MB)
-#define _PERF_L1_NUM_WAYS 8
+#define _PERF_L1_NUM_WAYS 4
 #define _PERF_SET_SIZE_BYTE (_PERF_L1_SIZE_BYTE / _PERF_L1_NUM_WAYS)
 
 // probe const
@@ -29,128 +35,29 @@
 struct perf
 {
     // const to be calibrated at run time
-    uint64_t csr_read_cycle; //# of cycles to read mcycle
+    uint64_t csr_read_cycle; // # of cycles to read mcycle
     uint64_t csr_read_ninst; // # of inst needed to read minstret
 
     // timer
     uint64_t cycle;
     uint64_t instrcnt;
-} perf;
-
-void _perf_start_timer()
-{
-    perf.cycle = csr_read(CSR_MCYCLE);
-    perf.instrcnt = csr_read(CSR_MINSTRET);
-}
-
-void _perf_end_timer()
-{
-    perf.cycle = csr_read(CSR_MCYCLE) - perf.cycle;
-    perf.instrcnt = csr_read(CSR_MINSTRET) - perf.instrcnt;
-}
-
-void _perf_print_timer()
-{
-    printf("cycle %d inst %d ipc %lf\n", perf.cycle, perf.instrcnt, (float)perf.instrcnt/perf.cycle);
-}
-
-void _perf_calibrate()
-{
-    // csr read delay
-    uint64_t cycle_1 = csr_read(CSR_MCYCLE);
-    uint64_t cycle_2 = csr_read(CSR_MCYCLE);
-    perf.csr_read_cycle = cycle_2-cycle_1;
-    printf("perf_calibrate: csr_read_cycle %d\n", perf.csr_read_cycle);
-
-    // csr read inst cost
-    uint64_t inst_1 = csr_read(CSR_MINSTRET);
-    uint64_t inst_2 = csr_read(CSR_MINSTRET);
-    perf.csr_read_ninst = inst_2-inst_1;
-    printf("perf_calibrate: csr_read_ninst %d\n", perf.csr_read_ninst);
-}
-
-void _perf_blackhole(uint64_t value)
-{
-    *(uint64_t*) _PERF_BLACKHOLE = value;
-}
-
-uint64_t setup_latency_test_linklist(uint64_t base_addr, uint64_t end_addr, uint64_t step)
-{
-    uint64_t num_valid_node = 0;
-    assert(step % 8 == 0);
-    assert(step >= 8);
-    for (uint64_t cur_addr = base_addr; cur_addr < end_addr;) {
-        uint64_t next_addr = cur_addr + step;
-        *((uint64_t*)cur_addr) = next_addr;
-        cur_addr = next_addr;
-        num_valid_node++;
-    }
-    return num_valid_node;
-}
-
-uint64_t read_latency_test_linklist(uint64_t base_addr, uint64_t num_valid_node)
-{
-    uint64_t cur_addr = base_addr;
-    for (int i = 0; i < num_valid_node; i++) {
-        cur_addr = (*(uint64_t*)cur_addr);
-    }
-    return cur_addr;
-}
-
-void warmup(uint64_t base_addr, uint64_t end_addr)
-{
-    setup_latency_test_linklist(base_addr, end_addr, _PERF_CACHELINE_SIZE_BYTE);
-}
-
-void test_latency(uint64_t size, int iter)
-{
-    volatile uint64_t result = 0; // make sure compiler will not opt read_latency_test_linklist
-    printf("range 0x%xB (%d iters) latency test\n", size, iter);
-    _perf_start_timer();
-    uint64_t nnode = setup_latency_test_linklist(_PERF_TEST_ADDR_BASE, _PERF_TEST_ADDR_BASE + size, _PERF_CACHELINE_SIZE_BYTE);
-    _perf_end_timer();
-    uint64_t total_node = nnode * iter;
-    // _perf_print_timer();
-
-    _perf_start_timer();
-    for (int i = 0; i < iter; i++) {
-        result += read_latency_test_linklist(_PERF_TEST_ADDR_BASE, nnode);
-    }
-    _perf_end_timer();
-    // _perf_print_timer();
-    printf("range 0x%xB (%d intrs) read latency %f (%d samples)\n", size, iter, (float)perf.cycle / total_node, total_node);
-
-    _perf_blackhole(result);
-}
-
-void test_mem_throughput(uint64_t iter)
-{
-    uint64_t remain = iter;
-    uint64_t result = 0;
-    uint64_t access_addr = _PERF_TEST_ADDR_BASE;
-    _perf_start_timer();
-    while (remain--) {
-        result += *(uint64_t*) access_addr;
-        access_addr += _PERF_CACHELINE_SIZE_BYTE;
-    }
-    _perf_end_timer();
-    *(uint64_t*) _PERF_BLACKHOLE = result;
-    printf("mem band width %f B/cycle (%d samples)\n", (float)iter * _PERF_CACHELINE_SIZE_BYTE / perf.cycle, iter);
-}
-
-void test_mem_throughput_same_set(uint64_t iter)
-{
-    uint64_t remain = iter;
-    uint64_t result = 0;
-    uint64_t access_addr = _PERF_TEST_ADDR_BASE;
-    _perf_start_timer();
-    while (remain--) {
-        result += *(uint64_t*) access_addr;
-        access_addr += _PERF_SET_SIZE_BYTE;
-    }
-    _perf_end_timer();
-    *(uint64_t*) _PERF_BLACKHOLE = result;
-    printf("mem band width %f B/cycle (%d samples)\n", (float)iter * _PERF_CACHELINE_SIZE_BYTE / perf.cycle, iter);
-}
+};
+
+extern struct perf perf;
+
+// common perf tools
+extern void _perf_start_timer();
+extern void _perf_end_timer();
+extern void _perf_print_timer();
+extern void _perf_calibrate();
+extern void _perf_blackhole(uint64_t value);
+
+// latency test
+extern uint64_t setup_latency_test_linklist(uint64_t base_addr, uint64_t end_addr, uint64_t step);
+extern uint64_t read_latency_test_linklist(uint64_t base_addr, uint64_t num_valid_node);
+extern void latency_test_warmup(uint64_t base_addr, uint64_t end_addr);
+extern void test_latency(uint64_t size, int iter);
+extern void test_mem_throughput(uint64_t iter);
+extern void test_mem_throughput_same_set(uint64_t iter);
 
 #endif
\ No newline at end of file
diff --git a/apps/maprobe/latency-test.c b/apps/maprobe/latency-test.c
new file mode 100644
index 0000000000000000000000000000000000000000..570f11e879ebf280fd5ea7470b16e344c071df70
--- /dev/null
+++ b/apps/maprobe/latency-test.c
@@ -0,0 +1,80 @@
+#include "maprobe.h"
+
+uint64_t setup_latency_test_linklist(uint64_t base_addr, uint64_t end_addr, uint64_t step)
+{
+    uint64_t num_valid_node = 0;
+    assert(step % 8 == 0);
+    assert(step >= 8);
+    for (uint64_t cur_addr = base_addr; cur_addr < end_addr;) {
+        uint64_t next_addr = cur_addr + step;
+        *((uint64_t*)cur_addr) = next_addr;
+        cur_addr = next_addr;
+        num_valid_node++;
+    }
+    return num_valid_node;
+}
+
+uint64_t read_latency_test_linklist(uint64_t base_addr, uint64_t num_valid_node)
+{
+    uint64_t cur_addr = base_addr;
+    for (int i = 0; i < num_valid_node; i++) {
+        cur_addr = (*(uint64_t*)cur_addr);
+    }
+    return cur_addr;
+}
+
+void latency_test_warmup(uint64_t base_addr, uint64_t end_addr)
+{
+    setup_latency_test_linklist(base_addr, end_addr, _PERF_CACHELINE_SIZE_BYTE);
+}
+
+void test_latency(uint64_t size, int iter)
+{
+    volatile uint64_t result = 0; // make sure compiler will not opt read_latency_test_linklist
+    printf("range 0x%xB (%d iters) latency test\n", size, iter);
+    _perf_start_timer();
+    uint64_t nnode = setup_latency_test_linklist(_PERF_TEST_ADDR_BASE, _PERF_TEST_ADDR_BASE + size, _PERF_CACHELINE_SIZE_BYTE);
+    _perf_end_timer();
+    uint64_t total_node = nnode * iter;
+    // _perf_print_timer();
+
+    _perf_start_timer();
+    for (int i = 0; i < iter; i++) {
+        result += read_latency_test_linklist(_PERF_TEST_ADDR_BASE, nnode);
+    }
+    _perf_end_timer();
+    // _perf_print_timer();
+    printf("range 0x%xB (%d iters) read latency %f (%d samples)\n", size, iter, (float)perf.cycle / total_node, total_node);
+
+    _perf_blackhole(result);
+}
+
+void test_mem_throughput(uint64_t iter)
+{
+    uint64_t remain = iter;
+    uint64_t result = 0;
+    uint64_t access_addr = _PERF_TEST_ADDR_BASE;
+    _perf_start_timer();
+    while (remain--) {
+        result += *(uint64_t*) access_addr;
+        access_addr += _PERF_CACHELINE_SIZE_BYTE;
+    }
+    _perf_end_timer();
+    *(uint64_t*) _PERF_BLACKHOLE = result;
+    printf("mem band width %f B/cycle (%d samples)\n", (float)iter * _PERF_CACHELINE_SIZE_BYTE / perf.cycle, iter);
+}
+
+void test_mem_throughput_same_set(uint64_t iter)
+{
+    uint64_t remain = iter;
+    uint64_t result = 0;
+    uint64_t access_addr = _PERF_TEST_ADDR_BASE;
+    _perf_start_timer();
+    while (remain--) {
+        result += *(uint64_t*) access_addr;
+        access_addr += _PERF_SET_SIZE_BYTE;
+    }
+    _perf_end_timer();
+    *(uint64_t*) _PERF_BLACKHOLE = result;
+    printf("mem band width %f B/cycle (%d samples)\n", (float)iter * _PERF_CACHELINE_SIZE_BYTE / perf.cycle, iter);
+}
diff --git a/apps/maprobe/maprobe.c b/apps/maprobe/main.c
similarity index 100%
rename from apps/maprobe/maprobe.c
rename to apps/maprobe/main.c