Add instruction address metadata to vivisect broker (#28)

Author: andrej

MetadataCategories.h

@@ -2,17 +2,26 @@
 #define LLVM_MCA_METADATACATEGORIES_H
 namespace llvm {
 namespace mcad {
+
+enum MetadataCategories {
+
 // Metadata for LSUnit
-static constexpr unsigned MD_LSUnit_MemAccess = 0;
+MD_LSUnit_MemAccess = 0,
 
 // Metadata for Branch Prediction
-static constexpr unsigned MD_FrontEnd_BranchFlow = 1;
+MD_FrontEnd_BranchFlow = 1,
+
+// Virtual address at which an instruction is located in memory
+MD_InstrAddr = 2,
 
 // Used for marking the start of custom MD Category
-static constexpr unsigned MD_LAST = MD_FrontEnd_BranchFlow;
+MD_LAST,
 
 // Metadata categories (custom)
-static constexpr unsigned MD_BinaryRegionMarkers = MD_LAST + 1;
+MD_BinaryRegionMarkers
+
+};
+
 } // end namespace mcad
 } // end namespace llvm
 #endif

plugins/vivisect-broker/Broker.cpp

@@ -128,26 +128,40 @@ class VivisectBroker : public Broker {
 
             MCIS[i] = &MCI;  // return a pointer into our MCI_Pool
 
-            if (MDE && insn.has_memory_access()) {
-                auto MemAccess = insn.memory_access();
-                auto &Registry = MDE->MDRegistry;
-                auto &IndexMap = MDE->IndexMap;
-                auto &MemAccessCat = Registry[MD_LSUnit_MemAccess];
-                IndexMap[i] = TotalNumTraces;
-                MemAccessCat[TotalNumTraces] = std::move(MDMemoryAccess{
-                    MemAccess.is_store(),
-                    MemAccess.vaddr(),
-                    MemAccess.size(),
-                });
-            }
+            // Add metadata to the fetched instruction if metadata exchanger
+            // is available
+            if (MDE) {
+              
+              // The registry stores the metadata
+              auto &Registry = MDE->MDRegistry;
+
+              // The IndexMap maps instruction index (within this region) to
+              // the identifier we chose for our metadata. We chose a
+              // monotonically increasing counter as the identifier for each
+              // metadata entry.
+              auto &IndexMap = MDE->IndexMap;
+              IndexMap[i] = TotalNumTraces;
+
+              auto &InstrAddrCat = Registry[MD_InstrAddr];
+              InstrAddrCat[TotalNumTraces] = insn.addr();
+
+              if (insn.has_memory_access()) {
+                  auto MemAccess = insn.memory_access();
+                  auto &MemAccessCat = Registry[MD_LSUnit_MemAccess];
+                  MemAccessCat[TotalNumTraces] = std::move(MDMemoryAccess{
+                      MemAccess.is_store(),
+                      MemAccess.vaddr(),
+                      MemAccess.size(),
+                  });
+              }
+
+              if (insn.has_branch_flow()) {
+                  auto BranchFlow = insn.branch_flow();
+                  auto &Registry = MDE->MDRegistry;
+                  auto &BranchFlowCat = Registry[MD_FrontEnd_BranchFlow];
+                  BranchFlowCat[TotalNumTraces] = BranchFlow.is_mispredict();
+              }
 
-            if (MDE && insn.has_branch_flow()) {
-                auto BranchFlow = insn.branch_flow();
-                auto &Registry = MDE->MDRegistry;
-                auto &IndexMap = MDE->IndexMap;
-                auto &BranchFlowCat = Registry[MD_FrontEnd_BranchFlow];
-                IndexMap[i] = TotalNumTraces;
-                BranchFlowCat[TotalNumTraces] = BranchFlow.is_mispredict();
             }
 
             ++TotalNumTraces;

plugins/vivisect-broker/README.md

@@ -9,7 +9,7 @@ C++-based server and a python client. To update them, edit the interface
 `emulator.proto` and rerun the following.
 
 ```
-python3 -m grpc_tools.protoc -I. --python_out=. --pyi_out=. --grpc_python_out=. emulator.proto
+python3 -m grpc_tools.protoc -I. --python_out=grpc_client --pyi_out=grpc_client --grpc_python_out=grpc_client emulator.proto
 protoc --grpc_out=. --cpp_out=. --plugin=protoc-gen-grpc=`which grpc_cpp_plugin` emulator.proto
 ```