diff --git a/rocclr/device/rocm/rocsettings.cpp b/rocclr/device/rocm/rocsettings.cpp
index e1de3a142..9982b2eb8 100644
--- a/rocclr/device/rocm/rocsettings.cpp
+++ b/rocclr/device/rocm/rocsettings.cpp
@@ -256,6 +256,8 @@ void Settings::setKernelArgImpl(const amd::Isa& isa, bool isXgmi, bool hasValidH
 
   auto kernelArgImpl = KernelArgImpl::HostKernelArgs;
 
+  hasValidHDPFlush &= DEBUG_CLR_KERNARG_HDP_FLUSH_WA;
+
   if (isXgmi) {
     // The XGMI-connected path does not require the manual memory ordering
     // workarounds that the PCIe connected path requires
@@ -284,6 +286,8 @@ void Settings::setKernelArgImpl(const amd::Isa& isa, bool isXgmi, bool hasValidH
   if (!flagIsDefault(HIP_FORCE_DEV_KERNARG)) {
     kernel_arg_impl_ = kernelArgImpl & (HIP_FORCE_DEV_KERNARG ? 0xF : 0x0);
   }
+
+  ClPrint(amd::LOG_INFO, amd::LOG_INIT, "Using dev kernel arg wa = %d", kernel_arg_impl_);
 }
 }  // namespace amd::roc
 
diff --git a/rocclr/utils/flags.hpp b/rocclr/utils/flags.hpp
index 30d286d73..72c643bf4 100644
--- a/rocclr/utils/flags.hpp
+++ b/rocclr/utils/flags.hpp
@@ -251,6 +251,8 @@ release(bool, DEBUG_HIP_GRAPH_DOT_PRINT, false,                               \
          "Enable/Disable graph debug dot print dump")                         \
 release(bool, HIP_ALWAYS_USE_NEW_COMGR_UNBUNDLING_ACTION, false,              \
         "Force to always use new comgr unbundling action")                    \
+release(bool, DEBUG_CLR_KERNARG_HDP_FLUSH_WA, false,                          \
+        "Toggle kernel arg copy workaround")                                  \
 
 namespace amd {