diff --git a/Makefile b/Makefile
index cab6c71e15f74823f85715b8738cc093976ede4c..b08e60ac176b5259140b81aac8a5318316a5f7a6 100644
--- a/Makefile
+++ b/Makefile
@@ -56,7 +56,7 @@ $(SIM_TOP_V): $(SCALA_FILE) $(TEST_FILE)
 	date -R
 	mill XiangShan.test.runMain $(SIMTOP) -X verilog -td $(@D) --full-stacktrace --output-file $(@F) $(SIM_ARGS)
 	sed -i '/module XSSimTop/,/endmodule/d' $(SIM_TOP_V)
-	sed -i -e 's/$$fatal/$$finish/g' $(SIM_TOP_V)
+	sed -i -e 's/$$fatal/xs_assert(`__LINE__)/g' $(SIM_TOP_V)
 	date -R
 
 EMU_TOP      = XSSimSoC
@@ -80,7 +80,7 @@ endif
 # Verilator multi-thread support
 EMU_THREADS  ?= 1
 ifneq ($(EMU_THREADS),1)
-VEXTRA_FLAGS += --threads $(EMU_THREADS) --threads-dpi none
+VEXTRA_FLAGS += --threads $(EMU_THREADS) --threads-dpi all
 endif
 
 # Verilator savable
diff --git a/src/test/csrc/assert.cpp b/src/test/csrc/assert.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..aa09f90dcef1535ddaa66dc4e650541de25036d5
--- /dev/null
+++ b/src/test/csrc/assert.cpp
@@ -0,0 +1,20 @@
+#include "common.h"
+#include "emu.h"
+
+int assert_count = 0;
+static pthread_mutex_t assert_mutex;
+
+void assert_init() {
+  pthread_mutex_init(&assert_mutex, 0);
+}
+
+void assert_finish() {
+  pthread_mutex_destroy(&assert_mutex);
+}
+
+extern "C" void xs_assert(long long line) {
+  pthread_mutex_lock(&assert_mutex);
+  printf("Assertion failed at line %lld\n.", line);
+  assert_count++;
+  pthread_mutex_unlock(&assert_mutex);
+}
diff --git a/src/test/csrc/common.h b/src/test/csrc/common.h
index c7158c524ee9e185850bc3209dd456f3ec4f1007..2cf7fcfe304d280c5151cd6cf76f4f8146893b36 100644
--- a/src/test/csrc/common.h
+++ b/src/test/csrc/common.h
@@ -6,6 +6,7 @@
 #include <cstdlib>
 #include <stdint.h>
 #include <assert.h>
+#include <pthread.h>
 
 #define ANSI_COLOR_RED     "\x1b[31m"
 #define ANSI_COLOR_GREEN   "\x1b[32m"
@@ -21,4 +22,8 @@
 #include "cosimulation.h"
 #endif
 
+extern int assert_count;
+void assert_init();
+void assert_finish();
+
 #endif // __COMMON_H
diff --git a/src/test/csrc/difftest.h b/src/test/csrc/difftest.h
index 3fecfcfa6b7c5d889499080e7144e85abb87e812..780f6359fea6db6c0c15c37a223d8358c3d93796 100644
--- a/src/test/csrc/difftest.h
+++ b/src/test/csrc/difftest.h
@@ -96,4 +96,7 @@ int difftest_step(DiffState *s);
 int difftest_store_step(uint64_t *saddr, uint64_t *sdata, uint8_t *smask);
 void difftest_display(uint8_t mode);
 
+uint64_t get_nemu_this_pc();
+void set_nemu_this_pc(uint64_t pc);
+
 #endif
diff --git a/src/test/csrc/emu.cpp b/src/test/csrc/emu.cpp
index 59bb8b64169cc0c0db7123ed26504bf6c41aad2f..82747ececa0dc0f7ed24cc16a45d54c5f61e2dff 100644
--- a/src/test/csrc/emu.cpp
+++ b/src/test/csrc/emu.cpp
@@ -8,12 +8,6 @@
 #include "zlib.h"
 #include "compress.h"
 
-void* get_ram_start();
-long get_ram_size();
-uint64_t get_nemu_this_pc();
-void set_nemu_this_pc(uint64_t pc);
-
-
 static inline void print_help(const char *file) {
   printf("Usage: %s [OPTION...]\n", file);
   printf("\n");
@@ -90,6 +84,7 @@ Emulator::Emulator(int argc, const char *argv[]):
   srand(args.seed);
   srand48(args.seed);
   Verilated::randReset(2);
+  assert_init();
 
   // init core
   reset_ncycles(10);
@@ -132,6 +127,7 @@ Emulator::Emulator(int argc, const char *argv[]):
 
 Emulator::~Emulator() {
   ram_finish();
+  assert_finish();
 
 #ifdef VM_SAVABLE
   if (args.enable_snapshot && trapCode != STATE_GOODTRAP && trapCode != STATE_LIMIT_EXCEEDED) {
@@ -382,7 +378,7 @@ uint64_t Emulator::execute(uint64_t max_cycle, uint64_t max_instr) {
 #endif
   }
 
-  if (Verilated::gotFinish()) {
+  if (assert_count > 0) {
     difftest_display(dut_ptr->io_difftest_priviledgeMode);
     eprintf("The simulation stopped. There might be some assertion failed.\n");
     trapCode = STATE_ABORT;
diff --git a/src/test/csrc/ram.cpp b/src/test/csrc/ram.cpp
index fcdddeca6c99e40e5a4428839b930ba3266fd977..b01394fa5e2e4d2688718bf3ea5049100babd5f8 100644
--- a/src/test/csrc/ram.cpp
+++ b/src/test/csrc/ram.cpp
@@ -12,6 +12,8 @@ CoDRAMsim3 *dram = NULL;
 
 static uint64_t *ram;
 static long img_size = 0;
+static pthread_mutex_t ram_mutex;
+
 void* get_img_start() { return &ram[0]; }
 long get_img_size() { return img_size; }
 void* get_ram_start() { return &ram[0]; }
@@ -155,6 +157,8 @@ void init_ram(const char *img) {
   dram = new CoDRAMsim3(DRAMSIM3_CONFIG, DRAMSIM3_OUTDIR);
 #endif
 
+  pthread_mutex_init(&ram_mutex, 0);
+
 }
 
 void ram_finish() {
@@ -162,13 +166,18 @@ void ram_finish() {
 #ifdef WITH_DRAMSIM3
   dramsim3_finish();
 #endif
+  pthread_mutex_destroy(&ram_mutex);
 }
 
+
 extern "C" uint64_t ram_read_helper(uint8_t en, uint64_t rIdx) {
   if (en && rIdx >= EMU_RAM_SIZE / sizeof(uint64_t)) {
     rIdx %= EMU_RAM_SIZE / sizeof(uint64_t);
   }
-  return (en) ? ram[rIdx] : 0;
+  pthread_mutex_lock(&ram_mutex);
+  uint64_t rdata = (en) ? ram[rIdx] : 0;
+  pthread_mutex_unlock(&ram_mutex);
+  return rdata;
 }
 
 extern "C" void ram_write_helper(uint64_t wIdx, uint64_t wdata, uint64_t wmask, uint8_t wen) {
@@ -177,7 +186,9 @@ extern "C" void ram_write_helper(uint64_t wIdx, uint64_t wdata, uint64_t wmask,
       printf("ERROR: ram wIdx = 0x%lx out of bound!\n", wIdx);
       assert(wIdx < EMU_RAM_SIZE / sizeof(uint64_t));
     }
+    pthread_mutex_lock(&ram_mutex);
     ram[wIdx] = (ram[wIdx] & ~wmask) | (wdata & wmask);
+    pthread_mutex_unlock(&ram_mutex);
   }
 }
 
diff --git a/src/test/csrc/ram.h b/src/test/csrc/ram.h
index 342036085c1b7dc361f3291b7783ffe4dd37b2ec..f6e224f220a2168b902721c701b60da4aeb623f7 100644
--- a/src/test/csrc/ram.h
+++ b/src/test/csrc/ram.h
@@ -8,6 +8,8 @@
 
 void init_ram(const char *img);
 void ram_finish();
+void* get_ram_start();
+long get_ram_size();
 
 #ifdef WITH_DRAMSIM3
 // 4*64 bits
diff --git a/src/test/vsrc/assert.v b/src/test/vsrc/assert.v
new file mode 100644
index 0000000000000000000000000000000000000000..7d4f6f224fd53d3176c4620f165c85761bed577a
--- /dev/null
+++ b/src/test/vsrc/assert.v
@@ -0,0 +1,4 @@
+import "DPI-C" function void xs_assert
+(
+  input  longint    line
+);