diff --git a/include/acl_hal.h b/include/acl_hal.h
index 483980a1..95f038fb 100644
--- a/include/acl_hal.h
+++ b/include/acl_hal.h
@@ -247,6 +247,10 @@ typedef struct {
                                            const std::string &signal_name,
                                            unsigned int &finish_counter);
 
+  void (*simulation_set_kernel_cra_address_map)(
+      unsigned int physical_device_id,
+      const std::vector<uintptr_t> &kernel_csr_address_map);
+
   size_t (*read_csr)(unsigned int physical_device_id, uintptr_t offset,
                      void *ptr, size_t size);
 
diff --git a/include/acl_hal_mmd.h b/include/acl_hal_mmd.h
index d1a301b3..136815a5 100644
--- a/include/acl_hal_mmd.h
+++ b/include/acl_hal_mmd.h
@@ -149,6 +149,12 @@ typedef struct {
   // accounted and returned in a subsequent invocation of this function.
   void (*aocl_mmd_simulation_streaming_kernel_done)(
       int handle, const std::string &signal_name, unsigned int &finish_counter);
+
+  // Pass kernel-id to csr-address mapping read from the current binary
+  // to the simulation runtime, so that it can detect which start register
+  // has been written by the runtime.
+  void (*aocl_mmd_simulation_set_kernel_cra_address_map)(
+      int handle, const std::vector<uintptr_t> &kernel_csr_address_map);
 } acl_mmd_dispatch_t;
 
 typedef struct {
diff --git a/src/acl_hal_mmd.cpp b/src/acl_hal_mmd.cpp
index 8d9810a0..81793ad1 100644
--- a/src/acl_hal_mmd.cpp
+++ b/src/acl_hal_mmd.cpp
@@ -161,6 +161,10 @@ void acl_hal_mmd_simulation_streaming_kernel_done(
     unsigned int physical_device_id, const std::string &kernel_name,
     unsigned int &finish_counter);
 
+void acl_hal_mmd_simulation_set_kernel_cra_address_map(
+    unsigned int physical_device_id,
+    const std::vector<uintptr_t> &kernel_csr_address_map);
+
 size_t acl_hal_mmd_read_csr(unsigned int physical_device_id, uintptr_t offset,
                             void *ptr, size_t size);
 
@@ -360,8 +364,9 @@ static acl_hal_t acl_hal_mmd = {
     acl_hal_mmd_shared_alloc,                     // shared_alloc
     acl_hal_mmd_simulation_streaming_kernel_start, // simulation_streaming_kernel_start
     acl_hal_mmd_simulation_streaming_kernel_done, // simulation_streaming_kernel_done
-    acl_hal_mmd_read_csr,                         // read_csr
-    acl_hal_mmd_write_csr,                        // write_csr
+    acl_hal_mmd_simulation_set_kernel_cra_address_map, // simulation_set_kernel_cra_address_map
+    acl_hal_mmd_read_csr,                              // read_csr
+    acl_hal_mmd_write_csr,                             // write_csr
 };
 
 // This will contain the device physical id to tell us which device across all
@@ -2865,6 +2870,14 @@ void acl_hal_mmd_simulation_streaming_kernel_done(
           device_info[physical_device_id].handle, kernel_name, finish_counter);
 }
 
+void acl_hal_mmd_simulation_set_kernel_cra_address_map(
+    unsigned int physical_device_id,
+    const std::vector<uintptr_t> &kernel_csr_address_map) {
+  device_info[physical_device_id]
+      .mmd_dispatch->aocl_mmd_simulation_set_kernel_cra_address_map(
+          device_info[physical_device_id].handle, kernel_csr_address_map);
+}
+
 size_t acl_hal_mmd_read_csr(unsigned int physical_device_id, uintptr_t offset,
                             void *ptr, size_t size) {
   return device_info[physical_device_id].mmd_dispatch->aocl_mmd_read(
diff --git a/src/acl_kernel_if.cpp b/src/acl_kernel_if.cpp
index 6ea575d9..c75bc913 100644
--- a/src/acl_kernel_if.cpp
+++ b/src/acl_kernel_if.cpp
@@ -921,16 +921,22 @@ int acl_kernel_if_update(const acl_device_def_autodiscovery_t &devdef,
     // The Kernel CSR registers
     // The new and improved config ROM give us the address *offsets* from
     // the first kernel CSR, and the range of each kernel CSR.
+    std::vector<uintptr_t> kernel_csr_address_map;
     for (unsigned ii = 0; ii < devdef.accel.size(); ++ii) {
       kern->accel_csr[ii].address =
           OFFSET_KERNEL_CRA + devdef.hal_info[ii].csr.address;
       kern->accel_csr[ii].bytes = devdef.hal_info[ii].csr.num_bytes;
+      kernel_csr_address_map.push_back(kern->accel_csr[ii].address);
 
       ACL_KERNEL_IF_DEBUG_MSG(
           kern, "Kernel_%s CSR { 0x%08" PRIxPTR ", 0x%08" PRIxPTR " }\n",
           devdef.accel[ii].iface.name.c_str(), kern->accel_csr[ii].address,
           kern->accel_csr[ii].bytes);
     }
+    if (acl_platform.offline_mode == ACL_CONTEXT_MPSIM) {
+      acl_get_hal()->simulation_set_kernel_cra_address_map(
+          kern->physical_device_id, kernel_csr_address_map);
+    }
 
     // The Kernel performance monitor registers
     for (unsigned ii = 0; ii < devdef.accel.size(); ++ii) {