!23405 Add remaining isolation for dump

Merge pull request !23405 from sabrinasun_59ee/macro

mindspore/ccsrc/backend/kernel_compiler/ascend_kernel_mod.h

@@ -21,7 +21,9 @@
 #include <memory>
 #include "runtime/device/ascend/ge_runtime/task_info.h"
 #include "backend/kernel_compiler/kernel.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 
 using TaskInfoPtr = std::shared_ptr<mindspore::ge::model_runner::TaskInfo>;
 namespace mindspore {

mindspore/ccsrc/backend/session/ascend_session.cc

@@ -55,8 +55,10 @@
 #include "backend/optimizer/common/helper.h"
 #include "runtime/device/kernel_runtime_manager.h"
 #include "utils/config_manager.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
 #include "debug/data_dump/e2e_dump.h"
+#endif
 #include "debug/anf_ir_utils.h"
 #include "backend/optimizer/graph_kernel/graph_kernel_optimization.h"
 #include "backend/session/ascend_auto_monad.h"
@@ -99,6 +101,7 @@ constexpr char SR_TAG[] = "sr_tag";
 constexpr char BACKWARD[] = "backward";
 constexpr auto kUnknowErrorString = "Unknown error occurred";
 namespace {
+#ifndef ENABLE_SECURITY
 void DumpGraphExeOrder(const std::vector<CNodePtr> &execution_order, const std::string &tag = "") {
   MS_LOG(INFO) << "Dump execution_order size " << execution_order.size();
   MS_LOG(INFO) << "[index][stream_label][graph_id][node string]";
@@ -129,6 +132,7 @@ void DumpGraphExeOrder(const std::vector<CNodePtr> &execution_order, const std::
   }
   buf << "================== execution order ==================\n";
 }
+#endif
 
 bool IsVMGraphTaskSink() {
   auto ms_context = MsContext::GetInstance();
@@ -1153,6 +1157,7 @@ void AscendSession::SelectKernel(const KernelGraph &kernel_graph) const {
   MS_LOG(INFO) << "Finish!";
 }
 
+#ifndef ENABLE_SECURITY
 void DumpInit(uint32_t device_id) {
   auto &json_parser = DumpJsonParser::GetInstance();
   json_parser.Parse();
@@ -1165,6 +1170,7 @@ void DumpInit(uint32_t device_id) {
     }
   }
 }
+#endif
 
 void AscendSession::InitRuntimeResource() {
   MS_LOG(INFO) << "Start!";
@@ -1180,7 +1186,9 @@ void AscendSession::InitRuntimeResource() {
     // get actual rank id if it's distribution training case.
     rank_id_ = GetRankId();
   }
+#ifndef ENABLE_SECURITY
   DumpInit(rank_id_);
+#endif
   MS_LOG(INFO) << "Finish!";
 }
 
@@ -1470,11 +1478,15 @@ void AscendSession::Execute(const std::shared_ptr<KernelGraph> &kernel_graph, bo
   auto runtime_instance = device::KernelRuntimeManager::Instance().GetKernelRuntime(kAscendDevice, device_id_);
   MS_EXCEPTION_IF_NULL(runtime_instance);
   if (is_task && is_task_sink) {
+#ifndef ENABLE_SECURITY
     DumpSetup(kernel_graph);
+#endif
   }
   bool ret_ok = runtime_instance->Run(kernel_graph.get(), is_task_sink);
   if (is_task && is_task_sink) {
+#ifndef ENABLE_SECURITY
     Dump(kernel_graph);
+#endif
   }
   if (!ret_ok) {
 #ifdef ENABLE_DUMP_IR
@@ -1485,6 +1497,7 @@ void AscendSession::Execute(const std::shared_ptr<KernelGraph> &kernel_graph, bo
   MS_LOG(DEBUG) << "Finish!";
 }
 
+#ifndef ENABLE_SECURITY
 void AscendSession::DumpSetup(const std::shared_ptr<KernelGraph> &kernel_graph) const {
   MS_LOG(DEBUG) << "Start!";
   MS_EXCEPTION_IF_NULL(kernel_graph);
@@ -1498,6 +1511,7 @@ void AscendSession::Dump(const std::shared_ptr<KernelGraph> &kernel_graph) const
   E2eDump::DumpData(kernel_graph.get(), rank_id_);
   MS_LOG(DEBUG) << "Finish!";
 }
+#endif
 
 void AscendSession::DumpAllGraphs(const std::vector<KernelGraphPtr> &all_graphs) {
 #ifdef ENABLE_DUMP_IR
@@ -1636,7 +1650,9 @@ void AscendSession::MergeGraphExecOrder() {
   }
   // set final_exec_order into final graph
   MS_EXCEPTION_IF_NULL(final_graph);
+#ifndef ENABLE_SECURITY
   DumpGraphExeOrder(final_exec_order);
+#endif
   final_graph->set_execution_order(final_exec_order);
 }
 

mindspore/ccsrc/backend/session/ascend_session.h

@@ -109,8 +109,10 @@ class AscendSession : public SessionBasic {
   void RunOpGenKernelEvent(const KernelGraph *graph) const;
   void Load(const std::shared_ptr<KernelGraph> &kernel_graph) const;
   void Execute(const std::shared_ptr<KernelGraph> &kernel_graph, bool is_task) const;
+#ifndef ENABLE_SECURITY
   void Dump(const std::shared_ptr<KernelGraph> &kernel_graph) const;
   void DumpSetup(const std::shared_ptr<KernelGraph> &kernel_graph) const;
+#endif
   void DumpAllGraphs(const std::vector<KernelGraphPtr> &all_graphs);
   void LoadTensor(const std::shared_ptr<KernelGraph> &kernel_graph) const;
   // below functions are used for run op

mindspore/ccsrc/backend/session/cpu_session.cc

@@ -33,7 +33,9 @@
 #include "backend/optimizer/pass/erase_visit_attr.h"
 #include "debug/anf_ir_dump.h"
 #include "debug/dump_proto.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 #if ((defined ENABLE_CPU) && (!defined _WIN32))
 #include "ps/util.h"
 #include "ps/ps_context.h"

mindspore/ccsrc/runtime/device/ascend/ascend_device_address.cc

@@ -31,7 +31,9 @@
 #include "abstract/utils.h"
 #include "utils/utils.h"
 #include "common/trans.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 #ifdef ENABLE_DEBUGGER
 #include "debug/tensor_load.h"
 #endif
@@ -507,7 +509,9 @@ bool AscendDeviceAddress::DumpMemToFile(const std::string &filepath, const std::
       MS_LOG(ERROR) << "Copy device mem to host failed";
       return ret;
     }
+#ifndef ENABLE_SECURITY
     ret = DumpJsonParser::DumpToFile(path, out_tensor->data_c(), host_size, host_shape, host_type);
+#endif
   } else {
     auto host_tmp = std::vector<uint8_t>(size_);
     auto ret_rt_memcpy = rtMemcpy(host_tmp.data(), size_, ptr_, size_, RT_MEMCPY_DEVICE_TO_HOST);
@@ -516,7 +520,9 @@ bool AscendDeviceAddress::DumpMemToFile(const std::string &filepath, const std::
     }
     std::string path = filepath + '.' + format_;
     MS_LOG(INFO) << "E2E Dump path is " << path;
+#ifndef ENABLE_SECURITY
     ret = DumpJsonParser::DumpToFile(path, host_tmp.data(), size_, host_shape, type_id_);
+#endif
   }
 
   return ret;

mindspore/ccsrc/runtime/device/ascend/ascend_kernel_runtime.cc

@@ -20,7 +20,6 @@
 #include <utility>
 #include <algorithm>
 #include "utils/signal_util.h"
-#include "debug/data_dump/e2e_dump.h"
 #include "runtime/device/ascend/ascend_device_address.h"
 #include "runtime/device/ascend/distribute/ascend_collective.h"
 #include "utils/ms_context.h"
@@ -39,7 +38,10 @@
 #endif
 #include "runtime/device/ascend/ascend_memory_manager.h"
 #include "runtime/device/ascend/ascend_event.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#include "debug/data_dump/e2e_dump.h"
+#endif
 #include "toolchain/adx_datadump_server.h"
 #include "utils/trace_base.h"
 #include "graphengine/inc/external/acl/error_codes/rt_error_codes.h"
@@ -226,6 +228,7 @@ bool AscendKernelRuntime::NeedDestroyHccl() {
   return true;
 }
 
+#ifndef ENABLE_SECURITY
 void AsyncDataDumpUninit() {
   if (DumpJsonParser::GetInstance().async_dump_enabled()) {
     if (AdxDataDumpServerUnInit() != 0) {
@@ -234,7 +237,6 @@ void AsyncDataDumpUninit() {
   }
 }
 
-#ifndef ENABLE_SECURITY
 void AscendKernelRuntime::ReportProfilingData() {
   auto context = MsContext::GetInstance();
   MS_EXCEPTION_IF_NULL(context);
@@ -268,7 +270,9 @@ void AscendKernelRuntime::ReleaseDeviceRes() {
   // release ge runtime
   ClearGraphModelMap();
 
+#ifndef ENABLE_SECURITY
   AsyncDataDumpUninit();
+#endif
 
   auto context_ptr = MsContext::GetInstance();
   MS_EXCEPTION_IF_NULL(context_ptr);
@@ -382,6 +386,7 @@ bool AscendKernelRuntime::LoadData(mindspore::session::KernelGraph *graph) {
 bool AscendKernelRuntime::KernelMemNotReuse(const AnfNodePtr &node) {
   MS_EXCEPTION_IF_NULL(node);
   bool need_dump = false;
+#ifndef ENABLE_SECURITY
   auto &dump_json_parser = DumpJsonParser::GetInstance();
   if (dump_json_parser.e2e_dump_enabled() && dump_json_parser.dump_mode() == 1) {
     auto op_name = node->fullname_with_scope();
@@ -389,6 +394,7 @@ bool AscendKernelRuntime::KernelMemNotReuse(const AnfNodePtr &node) {
       need_dump = true;
     }
   }
+#endif
   return need_dump;
 }
 
@@ -447,14 +453,18 @@ bool AscendKernelRuntime::GenTask(const session::KernelGraph *graph) {
     if (ConfigManager::GetInstance().dataset_mode() == DS_SINK_MODE && (ConfigManager::GetInstance().iter_num() > 1)) {
       MS_LOG(EXCEPTION) << "Dynamic shape is not supported with dataset_sink_mode.";
     }
+#ifndef ENABLE_SECURITY
     if (DumpJsonParser::GetInstance().async_dump_enabled()) {
       MS_LOG(EXCEPTION) << "Dynamic shape is not supported with Asynchronous Dump. Please use Synchronous Dump.";
     }
+#endif
     MS_LOG(INFO) << "Dynamic Shape Graph Generate Dynamic kernel";
     return GenDynamicKernel(graph);
   }
   MS_LOG(INFO) << "GenTask start. GraphId:" << graph->graph_id();
+#ifndef ENABLE_SECURITY
   DumpJsonParser::GetInstance().UpdateNeedDumpKernels(NOT_NULL(graph));
+#endif
 #ifdef MEM_REUSE_DEBUG
   if (!EnvConfigParser::GetInstance().GetSysMemreuse()) {
     // Get normal graph ir for memreuse
@@ -656,6 +666,7 @@ std::string AscendKernelRuntime::GetDumpPath() {
   return path;
 }
 
+#ifndef ENABLE_SECURITY
 void AscendKernelRuntime::DumpTaskExceptionInfo(const session::KernelGraph *graph) {
   MS_EXCEPTION_IF_NULL(graph);
   const std::string path = GetDumpPath();
@@ -680,6 +691,7 @@ void AscendKernelRuntime::DumpTaskExceptionInfo(const session::KernelGraph *grap
     E2eDump::DumpOutputImpl(node, false, path, &full_scope_name, nullptr);
   }
 }
+#endif
 
 bool AscendKernelRuntime::Run(session::KernelGraph *const graph, bool is_task_sink) {
   const uint64_t kUSecondInSecond = 1000000;
@@ -947,7 +959,9 @@ bool AscendKernelRuntime::RunTask(const session::KernelGraph *graph) {
   try {
     ModelRunner::Instance().RunModel(graph->graph_id());
   } catch (const std::exception &) {
+#ifndef ENABLE_SECURITY
     DumpTaskExceptionInfo(graph);
+#endif
 #ifdef ENABLE_TDTQUE
     // Run task error, we should call TdtHostDestroy to release tdt to avoid DeviceQueueOp hostPush hung
     // case1: cpu usage 100% cause thread/process exit, but some tdt thread remain in backend

mindspore/ccsrc/runtime/device/ascend/ascend_kernel_runtime.h

@@ -102,7 +102,9 @@ class AscendKernelRuntime : public KernelRuntime {
 #endif
   static CNodePtr GetErrorNodeName(uint32_t streamid, uint32_t taskid);
   static std::string GetDumpPath();
+#ifndef ENABLE_SECURITY
   static void DumpTaskExceptionInfo(const session::KernelGraph *graph);
+#endif
   static void TaskFailCallback(rtExceptionInfo *task_fail_info);
   static bool DeleteDumpDir(const std::string &path);
   static int DeleteDumpFile(std::string path);

mindspore/ccsrc/runtime/device/ascend/dump/data_dumper.cc

@@ -29,7 +29,9 @@
 #include "proto/op_mapping_info.pb.h"
 #include "utils/comm_manager.h"
 #include "utils/ms_context.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 #ifdef ENABLE_DEBUGGER
 #include "debug/debugger/debugger.h"
 #endif
@@ -67,6 +69,7 @@ DataDumper::~DataDumper() {
   ReleaseDevMem(&op_debug_dump_args_);
 }
 
+#ifndef ENABLE_SECURITY
 void DataDumper::GetNeedDumpKernelList(NotNull<std::map<std::string, CNodePtr> *> kernel_map) const {
   for (const auto &kernel : kernel_graph_->execution_order()) {
     if (AnfAlgo::GetKernelType(kernel) == HCCL_KERNEL &&
@@ -196,6 +199,7 @@ bool DataDumper::KernelNeedDump(const CNodePtr &kernel) const {
   // dump all kernel if mode is set 0 in data_dump.json
   return DumpJsonParser::GetInstance().NeedDump(kernel->fullname_with_scope());
 }
+#endif
 
 void DataDumper::UnloadDumpInfo() {
   if (!load_flag_) {
@@ -246,7 +250,9 @@ void DataDumper::ConstructDumpTask(NotNull<const CNodePtr &> kernel, NotNull<aic
   MS_EXCEPTION_IF_NULL(iter->second);
   auto task_id = std::get<kTupleTaskId>(*iter->second);
   auto stream_id = std::get<kTupleStreamId>(*iter->second);
+#ifndef ENABLE_SECURITY
   auto args = std::get<kTupleArgs>(*iter->second);
+#endif
   MS_LOG(INFO) << "[DataDump] Get runtime info task_id:" << task_id << " stream_id:" << stream_id;
 
   dump_task->set_task_id(task_id);
@@ -255,8 +261,10 @@ void DataDumper::ConstructDumpTask(NotNull<const CNodePtr &> kernel, NotNull<aic
   dump_task->mutable_op()->set_op_name(kernel->fullname_with_scope());
   dump_task->mutable_op()->set_op_type(AnfAlgo::GetCNodeName(kernel.get()));
 
+#ifndef ENABLE_SECURITY
   DumpKernelOutput(kernel, args, dump_task);
   DumpKernelInput(kernel, args, dump_task);
+#endif
 }
 
 void DataDumper::SetOpDebugMappingInfo(const NotNull<aicpu::dump::OpMappingInfo *> dump_info) const {
@@ -287,6 +295,7 @@ void DataDumper::SetOpDebugMappingInfo(const NotNull<aicpu::dump::OpMappingInfo
   dump_info->mutable_task()->Add(std::move(task));
 }
 
+#ifndef ENABLE_SECURITY
 void DataDumper::OpDebugRegister() {
   uint32_t op_debug_mode = DumpJsonParser::GetInstance().op_debug_mode();
   auto iter = kOverflowModeStr.find(op_debug_mode);
@@ -336,6 +345,7 @@ void DataDumper::OpDebugUnregister() {
     MS_LOG(EXCEPTION) << "[DataDump] Call rtDebugUnRegister failed, ret = " << rt_ret;
   }
 }
+#endif
 
 void DataDumper::RtLoadDumpData(const aicpu::dump::OpMappingInfo &dump_info, void **ptr) {
   std::string proto_str;
@@ -372,6 +382,7 @@ void SetDumpShape(const std::vector<size_t> &ms_shape, NotNull<aicpu::dump::Shap
   }
 }
 
+#ifndef ENABLE_SECURITY
 void DataDumper::DumpKernelOutput(const CNodePtr &kernel, void *args, NotNull<aicpu::dump::Task *> task) {
   if (!DumpJsonParser::GetInstance().OutputNeedDump()) {
     MS_LOG(INFO) << "Skip dump output";
@@ -452,6 +463,7 @@ void DataDumper::DumpKernelInput(const CNodePtr &kernel, void *args, NotNull<aic
     offset += sizeof(void *);
   }
 }
+#endif
 
 std::string DataDumper::StripUniqueId(const std::string node_name) {
   size_t last_underscore = node_name.find_last_of('_');

mindspore/ccsrc/runtime/device/ascend/dump/data_dumper.h

@@ -53,20 +53,26 @@ class DataDumper {
   void set_runtime_info(const std::map<std::string, std::shared_ptr<RuntimeInfo>> &runtime_info) {
     runtime_info_map_ = runtime_info;
   }
+#ifndef ENABLE_SECURITY
   void LoadDumpInfo();
-  void UnloadDumpInfo();
   void OpDebugRegister();
   void OpDebugUnregister();
+#endif
+  void UnloadDumpInfo();
 
  private:
   void ReleaseDevMem(void **ptr) const noexcept;
+#ifndef ENABLE_SECURITY
   bool KernelNeedDump(const CNodePtr &kernel) const;
   void SetOpMappingInfo(NotNull<aicpu::dump::OpMappingInfo *> dump_info) const;
+#endif
   void SetOpDebugMappingInfo(const NotNull<aicpu::dump::OpMappingInfo *> dump_info) const;
   void ConstructDumpTask(NotNull<const CNodePtr &> kernel, NotNull<aicpu::dump::Task *> dump_task) const;
+#ifndef ENABLE_SECURITY
   void GetNeedDumpKernelList(NotNull<std::map<std::string, CNodePtr> *> kernel_map) const;
   static void DumpKernelOutput(const CNodePtr &kernel, void *args, NotNull<aicpu::dump::Task *> task);
   static void DumpKernelInput(const CNodePtr &kernel, void *args, NotNull<aicpu::dump::Task *> task);
+#endif
   static std::string StripUniqueId(const std::string node_name);
   static void RtLoadDumpData(const aicpu::dump::OpMappingInfo &dump_info, void **ptr);
 

mindspore/ccsrc/runtime/device/cpu/cpu_device_address.cc

@@ -18,7 +18,9 @@
 #include <memory>
 #include "runtime/device/convert_tensor_utils.h"
 #include "runtime/hardware/cpu/cpu_memory_pool.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 
 namespace mindspore {
 namespace device {

mindspore/ccsrc/runtime/framework/graph_compiler.cc

@@ -33,7 +33,9 @@
 #include "debug/anf_ir_dump.h"
 #include "debug/rdr/running_data_recorder.h"
 #endif
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 
 namespace mindspore {
 namespace runtime {

mindspore/ccsrc/runtime/framework/graph_scheduler.cc

@@ -30,7 +30,9 @@
 #if !defined(_WIN32) && !defined(_WIN64)
 #include "utils/signal_util.h"
 #endif
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 #ifdef ENABLE_DUMP_IR
 #include "debug/rdr/recorder_manager.h"
 #endif
@@ -246,12 +248,15 @@ void GraphScheduler::BuildAndScheduleGlobalActor() {
   (void)actorMgr->Spawn(base_recorder_actor, true);
 
   // Create and schedule debug actor.
+#ifndef ENABLE_SECURITY
   bool debugger_actor_need = DumpJsonParser::GetInstance().e2e_dump_enabled();
+#endif
 #ifdef ENABLE_DEBUGGER
   if (Debugger::GetInstance()->DebuggerBackendEnabled()) {
     debugger_actor_need = true;
   }
 #endif
+#ifndef ENABLE_SECURITY
   if (debugger_actor_need) {
     auto debug_actor = std::make_shared<DebugActor>();
     MS_EXCEPTION_IF_NULL(debug_actor);
@@ -259,6 +264,7 @@ void GraphScheduler::BuildAndScheduleGlobalActor() {
     auto base_debug_actor = static_cast<ActorReference>(debug_actor);
     (void)actorMgr->Spawn(base_debug_actor, true);
   }
+#endif
 }
 
 ActorSet *GraphScheduler::Transform(const GraphCompilerInfo &graph_compiler_info) {

mindspore/ccsrc/runtime/hardware/cpu/cpu_device_context.cc

@@ -30,7 +30,9 @@
 #include "backend/optimizer/pass/replace_node_by_proxy.h"
 #include "backend/optimizer/pass/erase_visit_attr.h"
 #include "profiler/device/cpu/cpu_profiling.h"
+#ifndef ENABLE_SECURITY
 #include "debug/data_dump/dump_json_parser.h"
+#endif
 
 namespace mindspore {
 namespace device {

mindspore/offline_debug/__init__.py

@@ -16,11 +16,6 @@ This module provides APIs to load and process dump data, i.e. read tensors, chec
 for watchpoints and other debugging services.
 """
 
-from mindspore._c_expression import security
 from . import dbg_services
 from . import mi_validator_helpers
 from . import mi_validators
-
-if security.enable_security():
-    raise ModuleNotFoundError("Offline debugger is not supported in security mode."\
-                              "Please recompile mindspore without `-s on`.")

mindspore/offline_debug/dbg_services.py

@@ -17,12 +17,13 @@ The module DbgServices provides offline debugger APIs.
 """
 
 from mindspore._c_expression import security
-import mindspore._mindspore_offline_debug as cds
 from mindspore.offline_debug.mi_validators import check_init, check_initialize, check_add_watchpoint,\
      check_remove_watchpoint, check_check_watchpoints, check_read_tensor_info, check_initialize_done, \
          check_tensor_info_init, check_tensor_data_init, check_tensor_base_data_init, check_tensor_stat_data_init,\
               check_watchpoint_hit_init, check_parameter_init
 from mindspore.offline_debug.mi_validator_helpers import replace_minus_one
+if not security.enable_security():
+    import mindspore._mindspore_offline_debug as cds
 
 
 def get_version():