[Kernel] Code reentrance for guest fibers.

[Kernel] Code reentrance using exceptions for guest fibers.

src/xenia/cpu/processor.cc

@@ -358,7 +358,6 @@ bool Processor::ExecuteRaw(ThreadState* thread_state, uint32_t address) {
     return false;
   }
 
-  auto context = thread_state->context();
   return function->Call(thread_state, 0xBCBCBCBC);
 }
 

src/xenia/kernel/xboxkrnl/xboxkrnl_threading.cc

@@ -205,22 +205,30 @@ dword_result_t NtSuspendThread(dword_t handle, lpdword_t suspend_count_ptr) {
 }
 DECLARE_XBOXKRNL_EXPORT1(NtSuspendThread, kThreading, kImplemented);
 
-void KeSetCurrentStackPointers(lpvoid_t stack_ptr,
+void KeSetCurrentStackPointers(lpvoid_t stack_ptr, pointer_t<X_KTHREAD> thread,
-                               pointer_t<X_KTHREAD> cur_thread,
                                lpvoid_t stack_alloc_base, lpvoid_t stack_base,
                                lpvoid_t stack_limit) {
-  auto thread = XThread::GetCurrentThread();
+  auto current_thread = XThread::GetCurrentThread();
-  auto context = thread->thread_state()->context();
+  auto context = current_thread->thread_state()->context();
-  context->r[1] = stack_ptr.guest_address();
+  auto pcr = kernel_memory()->TranslateVirtual<X_KPCR*>(
+      static_cast<uint32_t>(context->r[13]));
 
-  auto pcr =
+  thread->stack_alloc_base = stack_alloc_base.value();
-      kernel_memory()->TranslateVirtual<X_KPCR*>((uint32_t)context->r[13]);
+  thread->stack_base = stack_base.value();
+  thread->stack_limit = stack_limit.value();
   pcr->stack_base_ptr = stack_base.guest_address();
   pcr->stack_end_ptr = stack_limit.guest_address();
+  context->r[1] = stack_ptr.guest_address();
 
-  // TODO: Do we need to set the stack info on cur_thread?
+  // If a fiber is set, and the thread matches, reenter to avoid issues with
+  // host stack overflowing.
+  if (thread->fiber_ptr &&
+      current_thread->guest_object() == thread.guest_address()) {
+    current_thread->Reenter(static_cast<uint32_t>(context->lr));
   }
-DECLARE_XBOXKRNL_EXPORT1(KeSetCurrentStackPointers, kThreading, kImplemented);
+}
+DECLARE_XBOXKRNL_EXPORT2(KeSetCurrentStackPointers, kThreading, kImplemented,
+                         kHighFrequency);
 
 dword_result_t KeSetAffinityThread(lpvoid_t thread_ptr, dword_t affinity,
                                    lpdword_t previous_affinity_ptr) {

src/xenia/kernel/xthread.cc

@@ -498,6 +498,16 @@ X_STATUS XThread::Terminate(int exit_code) {
   return X_STATUS_SUCCESS;
 }
 
+class reenter_exception {
+ public:
+  reenter_exception(uint32_t address) : address_(address){};
+  virtual ~reenter_exception(){};
+  uint32_t address() const { return address_; }
+
+ private:
+  uint32_t address_;
+};
+
 void XThread::Execute() {
   XELOGKERNEL("XThread::Execute thid {} (handle={:08X}, '{}', native={:08X})",
               thread_id_, handle(), thread_name_, thread_->system_id());
@@ -510,31 +520,61 @@ void XThread::Execute() {
   // have time to initialize shared structures AFTER CreateThread (RR).
   xe::threading::Sleep(std::chrono::milliseconds(10));
 
-  int exit_code = 0;
-
   // Dispatch any APCs that were queued before the thread was created first.
   DeliverAPCs();
 
+  uint32_t address;
+  std::vector<uint64_t> args;
+  bool want_exit_code;
+  int exit_code = 0;
+
   // If a XapiThreadStartup value is present, we use that as a trampoline.
   // Otherwise, we are a raw thread.
   if (creation_params_.xapi_thread_startup) {
-    uint64_t args[] = {creation_params_.start_address,
+    address = creation_params_.xapi_thread_startup;
-                       creation_params_.start_context};
+    args.push_back(creation_params_.start_address);
-    kernel_state()->processor()->Execute(thread_state_,
+    args.push_back(creation_params_.start_context);
-                                         creation_params_.xapi_thread_startup,
+    want_exit_code = false;
-                                         args, xe::countof(args));
   } else {
     // Run user code.
-    uint64_t args[] = {creation_params_.start_context};
+    address = creation_params_.start_address;
-    exit_code = static_cast<int>(kernel_state()->processor()->Execute(
+    args.push_back(creation_params_.start_context);
-        thread_state_, creation_params_.start_address, args,
+    want_exit_code = true;
-        xe::countof(args)));
-    // If we got here it means the execute completed without an exit being
-    // called.
-    // Treat the return code as an implicit exit code.
   }
 
-  Exit(exit_code);
+  uint32_t next_address;
+  try {
+    exit_code = static_cast<int>(kernel_state()->processor()->Execute(
+        thread_state_, address, args.data(), args.size()));
+    next_address = 0;
+  } catch (const reenter_exception& ree) {
+    next_address = ree.address();
+  }
+
+  // See XThread::Reenter comments.
+  while (next_address != 0) {
+    try {
+      kernel_state()->processor()->ExecuteRaw(thread_state_, next_address);
+      next_address = 0;
+      if (want_exit_code) {
+        exit_code = static_cast<int>(thread_state_->context()->r[3]);
+      }
+    } catch (const reenter_exception& ree) {
+      next_address = ree.address();
+    }
+  }
+
+  // If we got here it means the execute completed without an exit being called.
+  // Treat the return code as an implicit exit code (if desired).
+  Exit(!want_exit_code ? 0 : exit_code);
+}
+
+void XThread::Reenter(uint32_t address) {
+  // TODO(gibbed): Maybe use setjmp/longjmp on Windows?
+  // https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/longjmp#remarks
+  // On Windows with /EH, setjmp/longjmp do stack unwinding.
+  // Is there a better solution than exceptions for stack unwinding?
+  throw reenter_exception(address);
 }
 
 void XThread::EnterCriticalRegion() {

src/xenia/kernel/xthread.h

@@ -70,35 +70,72 @@ struct XAPC {
 // Processor Control Region
 struct X_KPCR {
   xe::be<uint32_t> tls_ptr;         // 0x0
-  char unk_04[0x2C];                // 0x4
+  uint8_t unk_04[0x2C];             // 0x4
   xe::be<uint32_t> pcr_ptr;         // 0x30
-  char unk_34[0x3C];                // 0x34
+  uint8_t unk_34[0x3C];             // 0x34
   xe::be<uint32_t> stack_base_ptr;  // 0x70 Stack base address (high addr)
   xe::be<uint32_t> stack_end_ptr;   // 0x74 Stack end (low addr)
-  char unk_78[0x88];                // 0x78
+  uint8_t unk_78[0x88];             // 0x78
   xe::be<uint32_t> current_thread;  // 0x100
-  char unk_104[0x8];                // 0x104
+  uint8_t unk_104[0x8];             // 0x104
-  xe::be<uint8_t> current_cpu;      // 0x10C
+  uint8_t current_cpu;              // 0x10C
-  char unk_10D[0x43];               // 0x10D
+  uint8_t unk_10D[0x43];            // 0x10D
   xe::be<uint32_t> dpc_active;      // 0x150
 };
 
 struct X_KTHREAD {
   X_DISPATCH_HEADER header;           // 0x0
-  char unk_10[0xAC];             // 0x10
+  xe::be<uint32_t> unk_10;            // 0x10
+  xe::be<uint32_t> unk_14;            // 0x14
+  uint8_t unk_18[0x28];               // 0x10
+  xe::be<uint32_t> unk_40;            // 0x40
+  xe::be<uint32_t> unk_44;            // 0x44
+  xe::be<uint32_t> unk_48;            // 0x48
+  xe::be<uint32_t> unk_4C;            // 0x4C
+  uint8_t unk_50[0x4];                // 0x50
+  xe::be<uint16_t> unk_54;            // 0x54
+  xe::be<uint16_t> unk_56;            // 0x56
+  uint8_t unk_58[0x4];                // 0x58
+  xe::be<uint32_t> stack_base;        // 0x5C
+  xe::be<uint32_t> stack_limit;       // 0x60
+  uint8_t unk_64[0x4];                // 0x64
+  xe::be<uint32_t> tls_address;       // 0x68
+  uint8_t unk_6C;                     // 0x6C
+  uint8_t unk_6D[0x7];                // 0x6D
+  xe::be<uint32_t> unk_74;            // 0x74
+  xe::be<uint32_t> unk_78;            // 0x78
+  xe::be<uint32_t> unk_7C;            // 0x7C
+  xe::be<uint32_t> unk_80;            // 0x80
+  xe::be<uint32_t> unk_84;            // 0x84
+  uint8_t unk_88[0x3];                // 0x88
+  uint8_t unk_8B;                     // 0x8B
+  uint8_t unk_8C[0x10];               // 0x8C
+  xe::be<uint32_t> unk_9C;            // 0x9C
+  uint8_t unk_A0[0x1C];               // 0xA0
   uint8_t suspend_count;              // 0xBC
   uint8_t unk_BD;                     // 0xBD
   uint8_t unk_BE;                     // 0xBE
   uint8_t current_cpu;                // 0xBF
-  char unk_C0[0x70];             // 0xC0
+  uint8_t unk_C0[0x10];               // 0xC0
+  xe::be<uint32_t> stack_alloc_base;  // 0xD0
+  uint8_t unk_D4[0x5C];               // 0xD4
   xe::be<uint64_t> create_time;       // 0x130
   xe::be<uint64_t> exit_time;         // 0x138
   xe::be<uint32_t> exit_status;       // 0x140
-  char unk_144[0x8];             // 0x144
+  xe::be<uint32_t> unk_144;           // 0x144
+  xe::be<uint32_t> unk_148;           // 0x148
   xe::be<uint32_t> thread_id;         // 0x14C
-  char unk_150[0x10];            // 0x150
+  xe::be<uint32_t> start_address;     // 0x150
+  xe::be<uint32_t> unk_154;           // 0x154
+  xe::be<uint32_t> unk_158;           // 0x158
+  uint8_t unk_15C[0x4];               // 0x15C
   xe::be<uint32_t> last_error;        // 0x160
-  char unk_164[0x94C];           // 0x164
+  xe::be<uint32_t> fiber_ptr;         // 0x164
+  uint8_t unk_168[0x4];               // 0x168
+  xe::be<uint32_t> creation_flags;    // 0x16C
+  uint8_t unk_170[0xC];               // 0x170
+  xe::be<uint32_t> unk_17C;           // 0x17C
+  uint8_t unk_180[0x930];             // 0x180
 
   // This struct is actually quite long... so uh, not filling this out!
 };
@@ -151,6 +188,8 @@ class XThread : public XObject, public cpu::Thread {
 
   virtual void Execute();
 
+  virtual void Reenter(uint32_t address);
+
   static void EnterCriticalRegion();
   static void LeaveCriticalRegion();
   uint32_t RaiseIrql(uint32_t new_irql);