Very basic, super slow, nasty indirection.

src/xenia/cpu/global_exports.cc

@@ -9,6 +9,7 @@
 
 #include <xenia/cpu/global_exports.h>
 
+#include <xenia/cpu/processor.h>
 #include <xenia/cpu/sdb.h>
 #include <xenia/cpu/ppc/instr.h>
 #include <xenia/cpu/ppc/state.h>
@@ -30,11 +31,12 @@ void _cdecl XeTrap(
   XEASSERTALWAYS();
 }
 
-void _cdecl XeIndirectBranch(
+void* _cdecl XeIndirectBranch(
     xe_ppc_state_t* state, uint64_t target, uint64_t br_ia) {
-  XELOGCPU("INDIRECT BRANCH %.8X -> %.8X",
-           (uint32_t)br_ia, (uint32_t)target);
-  XEASSERTALWAYS();
+  // TODO(benvanik): track this statistic - this path is very slow!
+  Processor* processor = (Processor*)state->processor;
+  void* target_ptr = processor->GetFunctionPointer((uint32_t)target);
+  return target_ptr;
 }
 
 void _cdecl XeInvalidInstruction(

src/xenia/cpu/global_exports.h

@@ -25,7 +25,7 @@ namespace cpu {
 typedef struct {
   void (_cdecl *XeTrap)(
       xe_ppc_state_t* state, uint64_t cia);
-  void (_cdecl *XeIndirectBranch)(
+  void* (_cdecl *XeIndirectBranch)(
       xe_ppc_state_t* state, uint64_t target, uint64_t br_ia);
   void (_cdecl *XeInvalidInstruction)(
       xe_ppc_state_t* state, uint64_t cia, uint64_t data);

src/xenia/cpu/jit.h

@@ -34,6 +34,7 @@ public:
   virtual int InitModule(ExecModule* module) = 0;
   virtual int UninitModule(ExecModule* module) = 0;
 
+  virtual void* GetFunctionPointer(sdb::FunctionSymbol* fn_symbol) = 0;
   virtual int Execute(xe_ppc_state_t* ppc_state,
                       sdb::FunctionSymbol* fn_symbol) = 0;
 

src/xenia/cpu/processor.cc

@@ -185,27 +185,12 @@ void Processor::DeallocThread(ThreadState* thread_state) {
 }
 
 int Processor::Execute(ThreadState* thread_state, uint32_t address) {
-  // Attempt to grab the function symbol from the global lookup table.
-  FunctionSymbol* fn_symbol = sym_table_->GetFunction(address);
+  // Attempt to get the function.
+  FunctionSymbol* fn_symbol = GetFunction(address);
   if (!fn_symbol) {
-    // Search all modules for the function symbol.
-    // Each module will see if the address is within its code range and if the
-    // symbol is not found (likely) it will do analysis on it.
-    // TODO(benvanik): make this more efficient. Could use a binary search or
-    //     something more clever.
-    sdb::FunctionSymbol* fn_symbol = NULL;
-    for (std::vector<ExecModule*>::iterator it = modules_.begin();
-        it != modules_.end(); ++it) {
-      fn_symbol = (*it)->FindFunctionSymbol(address);
-      if (fn_symbol) {
-        break;
-      }
-    }
-    if (!fn_symbol) {
-      // Symbol not found in any module.
-      XELOGCPU("Execute(%.8X): failed to find function", address);
-      return NULL;
-    }
+    // Symbol not found in any module.
+    XELOGCPU("Execute(%.8X): failed to find function", address);
+    return 1;
   }
 
   xe_ppc_state_t* ppc_state = thread_state->ppc_state();
@@ -230,3 +215,39 @@ uint64_t Processor::Execute(ThreadState* thread_state, uint32_t address,
   }
   return ppc_state->r[3];
 }
+
+FunctionSymbol* Processor::GetFunction(uint32_t address) {
+  // Attempt to grab the function symbol from the global lookup table.
+  FunctionSymbol* fn_symbol = sym_table_->GetFunction(address);
+  if (fn_symbol) {
+    return fn_symbol;
+  }
+
+  // Search all modules for the function symbol.
+  // Each module will see if the address is within its code range and if the
+  // symbol is not found (likely) it will do analysis on it.
+  // TODO(benvanik): make this more efficient. Could use a binary search or
+  //     something more clever.
+  for (std::vector<ExecModule*>::iterator it = modules_.begin();
+      it != modules_.end(); ++it) {
+    fn_symbol = (*it)->FindFunctionSymbol(address);
+    if (fn_symbol) {
+      return fn_symbol;
+    }
+  }
+  
+  // Not found at all? That seems wrong...
+  XEASSERTALWAYS();
+  return NULL;
+}
+
+void* Processor::GetFunctionPointer(uint32_t address) {
+  // Attempt to get the function.
+  FunctionSymbol* fn_symbol = GetFunction(address);
+  if (!fn_symbol) {
+    return NULL;
+  }
+
+  // Grab the pointer.
+  return jit_->GetFunctionPointer(fn_symbol);
+}

src/xenia/cpu/processor.h

@@ -48,6 +48,8 @@ public:
   void DeallocThread(ThreadState* thread_state);
   int Execute(ThreadState* thread_state, uint32_t address);
   uint64_t Execute(ThreadState* thread_state, uint32_t address, uint64_t arg0);
+  sdb::FunctionSymbol* GetFunction(uint32_t address);
+  void* GetFunctionPointer(uint32_t address);
 
 private:
   xe_memory_ref       memory_;

src/xenia/cpu/sdb/symbol_database.cc

@@ -448,18 +448,6 @@ int SymbolDatabase::AnalyzeFunction(FunctionSymbol* fn) {
   // TODO(benvanik): queue for add without expensive locks?
   sym_table_->AddFunction(fn->start_address, fn);
 
-  XELOGSDB("Finished analyzing %.8X", fn->start_address);
-  return 0;
-}
-
-int SymbolDatabase::CompleteFunctionGraph(FunctionSymbol* fn) {
-  // Find variable accesses.
-  // TODO(benvanik): data analysis to find variable accesses.
-
-  // A list of function targets that were undefined.
-  // This will run another analysis pass and it'd be best to avoid this.
-  std::vector<uint32_t> new_fns;
-
   // For each basic block:
   // - find outgoing target block or function
   for (std::map<uint32_t, FunctionBlock*>::iterator it = fn->blocks.begin();
@@ -485,26 +473,19 @@ int SymbolDatabase::CompleteFunctionGraph(FunctionSymbol* fn) {
         // Function call.
         block->outgoing_type = FunctionBlock::kTargetFunction;
         block->outgoing_function = GetFunction(block->outgoing_address);
-        if (block->outgoing_function) {
-          FunctionSymbol::AddCall(fn, block->outgoing_function);
-        } else {
-          XELOGE("call target not found: %.8X -> %.8X",
-                 block->end_address, block->outgoing_address);
-          new_fns.push_back(block->outgoing_address);
-        }
+        XEASSERTNOTNULL(block->outgoing_function);
+        FunctionSymbol::AddCall(fn, block->outgoing_function);
       }
     }
   }
 
-  if (new_fns.size()) {
-    XELOGW("Repeat analysis required to find %d new functions",
-           (uint32_t)new_fns.size());
-    for (std::vector<uint32_t>::iterator it = new_fns.begin();
-        it != new_fns.end(); ++it) {
-      GetOrInsertFunction(*it, fn);
-    }
-    return 1;
-  }
+  XELOGSDB("Finished analyzing %.8X", fn->start_address);
+  return 0;
+}
+
+int SymbolDatabase::CompleteFunctionGraph(FunctionSymbol* fn) {
+  // Find variable accesses.
+  // TODO(benvanik): data analysis to find variable accesses.
 
   return 0;
 }

src/xenia/cpu/sdb/symbol_table.cc

@@ -16,17 +16,26 @@ using namespace xe::cpu::sdb;
 
 
 SymbolTable::SymbolTable() {
+  lock_ = xe_mutex_alloc(10000);
+  XEASSERTNOTNULL(lock_);
 }
 
 SymbolTable::~SymbolTable() {
+  xe_mutex_free(lock_);
+  lock_ = NULL;
 }
 
 int SymbolTable::AddFunction(uint32_t address, FunctionSymbol* symbol) {
+  xe_mutex_lock(lock_);
   map_[address] = symbol;
+  xe_mutex_unlock(lock_);
   return 0;
 }
 
 FunctionSymbol* SymbolTable::GetFunction(uint32_t address) {
+  xe_mutex_lock(lock_);
   FunctionMap::const_iterator it = map_.find(address);
-  return it != map_.end() ? it->second : NULL;
+  FunctionSymbol* result = it != map_.end() ? it->second : NULL;
+  xe_mutex_unlock(lock_);
+  return result;
 }

src/xenia/cpu/sdb/symbol_table.h

@@ -29,6 +29,8 @@ public:
   sdb::FunctionSymbol* GetFunction(uint32_t address);
 
 private:
+  // TODO(benvanik): replace with a better data structure.
+  xe_mutex_t* lock_;
   typedef std::tr1::unordered_map<uint32_t, sdb::FunctionSymbol*> FunctionMap;
   FunctionMap map_;
 };

src/xenia/cpu/x64/x64_emitter.cc

@@ -450,10 +450,21 @@ void X64Emitter::GenerateSharedBlocks() {
     SpillRegisters();
     X86CompilerFuncCall* call = c.call(global_exports_.XeIndirectBranch);
     call->setPrototype(kX86FuncConvDefault,
-        FuncBuilder3<void, void*, uint64_t, uint64_t>());
+        FuncBuilder3<void*, void*, uint64_t, uint64_t>());
     call->setArgument(0, c.getGpArg(0));
     call->setArgument(1, locals_.indirection_target);
     call->setArgument(2, locals_.indirection_cia);
+    GpVar target_ptr(c.newGpVar());
+    call->setReturn(target_ptr);
+
+    // Call target.
+    // void fn(ppc_state*, uint64_t)
+    call = c.call(target_ptr);
+    call->setComment("Indirection branch");
+    call->setPrototype(kX86FuncConvDefault,
+        FuncBuilder2<void, void*, uint64_t>());
+    call->setArgument(0, c.getGpArg(0));
+    call->setArgument(1, locals_.indirection_cia);
     c.ret();
   }
 
@@ -936,16 +947,27 @@ int X64Emitter::GenerateIndirectionBranch(uint32_t cia, GpVar& target,
     // Spill registers. We could probably share this.
     SpillRegisters();
 
-    // Issue the full indirection branch.
+    // Grab the target of the indirection.
     // TODO(benvanik): remove once fixed: https://code.google.com/p/asmjit/issues/detail?id=86
     GpVar arg2 = c.newGpVar(kX86VarTypeGpq);
     c.mov(arg2, imm(cia));
     X86CompilerFuncCall* call = c.call(global_exports_.XeIndirectBranch);
     call->setPrototype(kX86FuncConvDefault,
-        FuncBuilder3<void, void*, uint64_t, uint64_t>());
+        FuncBuilder3<void*, void*, uint64_t, uint64_t>());
     call->setArgument(0, c.getGpArg(0));
     call->setArgument(1, target);
     call->setArgument(2, arg2);
+    GpVar target_ptr(c.newGpVar());
+    call->setReturn(target_ptr);
+
+    // Call target.
+    // void fn(ppc_state*, uint64_t)
+    call = c.call(target_ptr);
+    call->setComment("Indirection branch");
+    call->setPrototype(kX86FuncConvDefault,
+        FuncBuilder2<void, void*, uint64_t>());
+    call->setArgument(0, c.getGpArg(0));
+    call->setArgument(1, arg2);
 
     // TODO(benvanik): next_block/is_last_block/etc
     //if (next_block) {

src/xenia/cpu/x64/x64_jit.cc

@@ -138,23 +138,32 @@ int X64JIT::UninitModule(ExecModule* module) {
   return 0;
 }
 
-int X64JIT::Execute(xe_ppc_state_t* ppc_state, FunctionSymbol* fn_symbol) {
-  XELOGCPU("Execute(%.8X): %s...", fn_symbol->start_address, fn_symbol->name());
-
+void* X64JIT::GetFunctionPointer(sdb::FunctionSymbol* fn_symbol) {
   // Check function.
   x64_function_t fn_ptr = (x64_function_t)fn_symbol->impl_value;
   if (!fn_ptr) {
     // Function hasn't been prepped yet - make it now inline.
     // The emitter will lock and do other fancy things, if required.
     if (emitter_->PrepareFunction(fn_symbol)) {
-      XELOGCPU("Execute(%.8X): unable to make function %s",
-          fn_symbol->start_address, fn_symbol->name());
-      return 1;
+      return NULL;
     }
     fn_ptr = (x64_function_t)fn_symbol->impl_value;
     XEASSERTNOTNULL(fn_ptr);
   }
 
+  return fn_ptr;
+}
+
+int X64JIT::Execute(xe_ppc_state_t* ppc_state, FunctionSymbol* fn_symbol) {
+  XELOGCPU("Execute(%.8X): %s...", fn_symbol->start_address, fn_symbol->name());
+
+  x64_function_t fn_ptr = (x64_function_t)GetFunctionPointer(fn_symbol);
+  if (!fn_ptr) {
+    XELOGCPU("Execute(%.8X): unable to make function %s",
+        fn_symbol->start_address, fn_symbol->name());
+    return 1;
+  }
+
   // Call into the function. This will compile it if needed.
   fn_ptr(ppc_state, ppc_state->lr);
 

src/xenia/cpu/x64/x64_jit.h

@@ -33,6 +33,7 @@ public:
   virtual int InitModule(ExecModule* module);
   virtual int UninitModule(ExecModule* module);
 
+  virtual void* GetFunctionPointer(sdb::FunctionSymbol* fn_symbol);
   virtual int Execute(xe_ppc_state_t* ppc_state,
                       sdb::FunctionSymbol* fn_symbol);