largely working

src/asm_snippets.S

@@ -136,7 +136,9 @@ red_asm_pop_all_regs_start:
   m_pop_all_regs
 red_asm_pop_all_regs_end:
 
-
+  .global red_asm_jump_rsi
+red_asm_jump_rsi:
+  jmp *%rsi
 
 // we don't need executable stack
   .section .note.GNU-stack,"",%progbits

src/main.cc

@@ -52,9 +52,9 @@ int main(int argc, char* argv[]) {
   while(1) {
     switch(program[pc]) {
     case PRINT_NUM:
-      //redmagic_temp_disable();
+      redmagic_temp_disable();
       cout << program[pc + 1] << endl;
-      //redmagic_temp_enable();
+      redmagic_temp_enable();
       //printf("%i\n", program[pc + 1]);
       pc += 2;
       break;

src/manager.cc

@@ -129,6 +129,9 @@ static const char *avoid_inlining_methods[] = {
   "redmagic_ensure_not_traced",
   "redmagic_temp_disable",
   "redmagic_temp_enable",
+  "redmagic_is_traced",
+  "redmagic_disable_branch",
+  "redmagic_do_not_trace_function"
 };
 
 // namespace redmagic {
@@ -319,6 +322,7 @@ void* Manager::backwards_branch(void *id, void *ret_addr) {
       head->is_compiled = true;
       head->is_traced = true;
       head->trace_id = id;
+      red_printf("entering trace %x\n", id);
       return info->starting_point;
     }
     // don't care about atomic since we are just trying to get an estimate, so if we lose some counts it is fine

src/tracer.cc

@@ -179,6 +179,7 @@ extern "C" void* red_end_trace(mem_loc_t normal_end_address) {
   } else {
     protected_malloc = false;
   }
+  red_printf("exiting trace %x\n", head.trace_id);
   return ret;
 }
 
@@ -459,6 +460,9 @@ void* Tracer::TempDisableTrace() {
   compiler.mov(asmjit::x86::r15, asmjit::imm_u(0xdeadcafe));
   compiler.bind(label);
   auto written = compiler.finalize();
+  // SimpleCompiler compiler2(buffer.get());
+  // compiler2.mov(asmjit::x86::rax, asmjit::x86::ptr(asmjit::x86::rsp, -8));
+  // compiler2.TestRegister(RAX)
   write_interrupt_block();
 
   //temp_disable_resume = (void*)(written.getRawBuffer() + written.getOffset());
@@ -467,6 +471,19 @@ void* Tracer::TempDisableTrace() {
   return (void*)last_call_ret_addr;
 }
 
+extern "C" void red_asm_jump_rsi();
+
+void Tracer::TempEnableTrace(void *resume_pc) {
+  // check that the temp enable instruction is coming in at a expected spot, otherwise fork a new trace
+  set_pc((mem_loc_t)resume_pc);
+  SimpleCompiler compiler(buffer.get());
+  // the "normal" return address will be set to ris when this returns from the temp disabled region
+  //compiler.mov(asmjit::x86::rax, asmjit::x86::ptr(asmjit::x86::rsp, -8));
+  compiler.TestRegister((mem_loc_t)&red_asm_jump_rsi, RSI, (register_t)resume_pc);
+  auto written = compiler.finalize();
+  write_interrupt_block();
+}
+
 extern "C" void red_asm_start_nested_trace();
 
 void Tracer::JumpToNestedLoop(void *nested_trace_id) {
@@ -1345,6 +1362,7 @@ void Tracer::replace_rip_instruction() {
     //case UD_Ipush: // have to do push independently since the stack is moving
   case UD_Iadd:
   case UD_Isub:
+  case UD_Iimul:
   case UD_Icmp:
   case UD_Itest:
   case UD_Ixor:

src/tracer.h

@@ -40,7 +40,7 @@ namespace redmagic {
 
     // generate a temp disable command, sets the thread local where to resume to address
     void* TempDisableTrace();
-    void TempEnableTrace(void *resume_pc) { set_pc((uint64_t)resume_pc); }
+    void TempEnableTrace(void *resume_pc);// { set_pc((uint64_t)resume_pc); }
 
     void* ReplaceIsTracedCall();
 

src/user_interface.c

@@ -8,7 +8,7 @@
 #ifndef DISABLE_REDMAGIC
 
 __asm__("jmp_rax: \n"
-        "add $8, %rsp \n"
+        "pop %rsi \n"
         "jmp *%rax \n"
         );