[SWDEV-548460] Add RDC Policy Reset Message

* [SWDEV-548460] Add RDC Policy Reset Message

* [rdc] Bump version to 1.3.0

Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>

* chore: [rdc] Format CMakeLists.txt

Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>
[rocm-systems] ROCm/rocm-systems#2180 (commit 5bf6e36)

Author: adam360x

CMakeLists.txt

@@ -23,9 +23,13 @@
 #
 cmake_minimum_required(VERSION 3.15)
 
-set( COMP_TYPE "runtime" )
-set( BUILD_ENABLE_LINTIAN_OVERRIDES ON CACHE BOOL "Enable/Disable Lintian Overrides" )
-set( BUILD_DEBIAN_PKGING_FLAG ON CACHE BOOL "Internal Status Flag to indicate Debian Packaging Build" )
+set(COMP_TYPE "runtime")
+set(BUILD_ENABLE_LINTIAN_OVERRIDES ON CACHE BOOL "Enable/Disable Lintian Overrides")
+set(BUILD_DEBIAN_PKGING_FLAG
+    ON
+    CACHE BOOL
+    "Internal Status Flag to indicate Debian Packaging Build"
+)
 set(RDC "rdc" CACHE INTERNAL "")
 set(RDC_PACKAGE ${RDC} CACHE STRING "")
 
@@ -89,7 +93,7 @@ include(utils)
 set(PKG_VERSION_GIT_TAG_PREFIX "rdc_pkg_ver")
 # Provide git to utilities
 find_program(GIT NAMES git)
-get_version_from_tag("1.2.0" ${PKG_VERSION_GIT_TAG_PREFIX} GIT)
+get_version_from_tag("1.3.0" ${PKG_VERSION_GIT_TAG_PREFIX} GIT)
 # VERSION_* variables should be set by get_version_from_tag
 message("Package version: ${VERSION_STRING}")
 
@@ -167,8 +171,8 @@ set(CMAKE_LIBRARY_PATH ${CMAKE_LIBRARY_PATH} /usr/lib64 /usr/lib/x86_64-linux-gn
 
 # configure packaging
 # cpack version is populated with CMAKE_PROJECT_VERSION implicitly
-set(PKG_MAINTAINER_NM  "RDC Support")
-set(PKG_MAINTAINER_EMAIL  "rdc.support@amd.com")
+set(PKG_MAINTAINER_NM "RDC Support")
+set(PKG_MAINTAINER_EMAIL "rdc.support@amd.com")
 set(CPACK_PACKAGE_NAME ${RDC_PACKAGE} CACHE INTERNAL "")
 set(CPACK_PACKAGE_VENDOR "Advanced Micro Devices, Inc." CACHE STRING "")
 set(CPACK_PACKAGE_CONTACT "${PKG_MAINTAINER_NM} <${PKG_MAINTAINER_EMAIL}>")
@@ -565,7 +569,7 @@ set(CPACK_RPM_RUNTIME_POST_UNINSTALL_SCRIPT_FILE "${CMAKE_CURRENT_SOURCE_DIR}/RP
 set(CPACK_DEBIAN_RUNTIME_PACKAGE_NAME "${CPACK_PACKAGE_NAME}")
 set(CPACK_RPM_RUNTIME_PACKAGE_NAME "${CPACK_PACKAGE_NAME}")
 
-configure_pkg( ${RDC} ${COMP_TYPE} ${CPACK_PACKAGE_VERSION} ${PKG_MAINTAINER_NM} ${PKG_MAINTAINER_EMAIL} )
+configure_pkg( ${RDC} ${COMP_TYPE} ${CPACK_PACKAGE_VERSION} ${PKG_MAINTAINER_NM} ${PKG_MAINTAINER_EMAIL})
 
 include(CPack)
 

include/rdc/rdc.h

@@ -1537,6 +1537,8 @@ typedef struct {
   rdc_policy_condition_t condition;  //!< the condition that is meet
   rdc_gpu_group_t group_id;          //!< The group id trigger this callback
   int64_t value;                     //!< The current value that meet the condition
+  uint32_t gpu_index;                //!< GPU index that hit the condition
+  bool  reset_triggered;             //!< if reset was attempted
 } rdc_policy_callback_response_t;
 
 /**

protos/rdc.proto

@@ -659,6 +659,8 @@ message RegisterPolicyResponse {
   PolicyCondition condition =3;
   uint32 group_id =4;
   uint64 value=5;
+  uint32 gpu_index=6;
+  bool   reset_triggered = 7;
 }
 
 message UnRegisterPolicyResult {

rdc_libs/CMakeLists.txt

@@ -58,7 +58,7 @@ find_program(GIT NAMES git)
 
 # Debian package specific variables
 # Set a default value for the package version
-get_version_from_tag("1.2.0.0" ${SO_VERSION_GIT_TAG_PREFIX} GIT)
+get_version_from_tag("1.3.0.0" ${SO_VERSION_GIT_TAG_PREFIX} GIT)
 
 # VERSION_* variables should be set by get_version_from_tag
 set(SO_VERSION_STRING "${VERSION_MAJOR}.${VERSION_MINOR}")

rdc_libs/rdc/src/RdcPolicyImpl.cc

@@ -218,14 +218,17 @@ void RdcPolicyImpl::rdc_policy_check_condition() {
         status = metric_fetcher_->fetch_smi_field(gpu_index, map[policy.condition.type], &value);
         if (status == RDC_ST_OK) {
           if (value.value.l_int > policy.condition.value) {
+
+            bool reset = RDC_POLICY_ACTION_GPU_RESET == policy.action;
+
             // callback if needed
             if (callback) {
               rdc_policy_callback_response_t response = {1, policy.condition, group_id,
-                                                         value.value.l_int};
+                                                         value.value.l_int, gpu_index, reset};
               callback(&response);
             }
 
-            if (RDC_POLICY_ACTION_GPU_RESET == policy.action) {
+            if (reset) {
               rdc_policy_gpu_reset(gpu_index);
             }
           }

rdc_libs/rdc_client/src/RdcStandaloneHandler.cc

@@ -918,6 +918,8 @@ rdc_status_t RdcStandaloneHandler::rdc_policy_register(rdc_gpu_group_t group_id,
         response.condition.value = cond.value();
         response.group_id = reply.group_id();
         response.value = reply.value();
+        response.gpu_index       = reply.gpu_index();
+        response.reset_triggered = reply.reset_triggered();
 
         callback(&response);
       }

rdci/src/RdciPolicySubSystem.cc

@@ -195,8 +195,16 @@ int rdc_policy_callback(rdc_policy_callback_response_t* userData) {
     threshold /= 1000000;
   }
 
-  std::cout << "A " << condition_type_to_str(userData->condition.type) << " exceeds the threshold "
-            << threshold << " with the value " << value << std::endl;
+  std::cout << "A " << condition_type_to_str(userData->condition.type)
+            << " exceeds the threshold " << threshold
+            << " with the value " << value;
+
+  if (userData->reset_triggered) {
+    std::cout << " and triggered reset on GPU " << userData->gpu_index;
+  }
+
+  std::cout << std::endl;
+
   last_time = now;  // update the last time
   return 0;
 }

rdci/src/rdci.cc

@@ -41,7 +41,7 @@ THE SOFTWARE.
 #include "rdc_lib/rdc_common.h"
 
 #define RDC_CLIENT_VERSION_MAJOR 1
-#define RDC_CLIENT_VERSION_MINOR 2
+#define RDC_CLIENT_VERSION_MINOR 3
 #define RDC_CLIENT_VERSION_RELEASE 0
 
 #define RDC_CLIENT_VERSION_CREATE_STRING(MAJOR, MINOR, RELEASE) (#MAJOR "." #MINOR "." #RELEASE)

server/include/rdc/rdc_server_main.h

@@ -31,7 +31,7 @@ THE SOFTWARE.
 #include "rdc/rdc_api_service.h"
 
 #define RDC_SERVER_VERSION_MAJOR 1
-#define RDC_SERVER_VERSION_MINOR 2
+#define RDC_SERVER_VERSION_MINOR 3
 #define RDC_SERVER_VERSION_RELEASE 0
 
 #define RDC_SERVER_VERSION_CREATE_STRING(MAJOR, MINOR, RELEASE) (#MAJOR "." #MINOR "." #RELEASE)

server/src/rdc_api_service.cc

@@ -931,7 +931,8 @@ ::grpc::Status RdcAPIServiceImpl::RegisterPolicy(
             ::rdc::PolicyCondition* cond = reply.mutable_condition();
             cond->set_type(static_cast<::rdc::PolicyCondition_Type>(ctx->response.condition.type));
             cond->set_value(ctx->response.condition.value);
-
+            reply.set_gpu_index(ctx->response.gpu_index);
+            reply.set_reset_triggered(ctx->response.reset_triggered != 0);
             writer->Write(reply);
           }