[Kernel] Added KeSaveFloatingPointState and KeRestoreFloatingPointState from nukernel

src/xenia/kernel/xboxkrnl/xboxkrnl_misc.cc

@@ -28,94 +28,28 @@ void KeEnableFpuExceptions_entry(
   kthread->fpu_exceptions_on = static_cast<uint32_t>(ctx->r[3]) != 0;
 }
 DECLARE_XBOXKRNL_EXPORT1(KeEnableFpuExceptions, kNone, kStub);
-#if 0
-struct __declspec(align(8)) fpucontext_ptr_t {
-  char unknown_data[158];
-  __int16 field_9E;
-  char field_A0[2272];
-  unsigned __int64 saved_FPSCR;
-  double saved_fpu_regs[32];
-};
-#pragma pack(push, 1)
-struct __declspec(align(1)) r13_struct_t {
-  char field_0[6];
-  __int16 field_6;
-  char field_8[2];
-  char field_A;
-  char field_B[5];
-  int field_10;
-  char field_14[315];
-  char field_14F;
-  unsigned int field_150;
-  char field_154[427];
-  char field_2FF;
-  char field_300;
-};
-#pragma pack(pop)
 
+void KeSaveFloatingPointState_entry(const ppc_context_t& ctx) {
+  // Probably we should use: thread_fpu_related to store/restore state
+  X_KTHREAD* kthread = ctx->TranslateVirtual(
+      ctx->TranslateVirtualGPR<X_KPCR*>(ctx->r[13])->prcb_data.current_thread);
 
-static uint64_t Do_mfmsr(ppc_context_t& ctx) {
+  for (size_t i = 0; i < xe::countof(ctx->f); ++i) {
-  auto frontend = ctx->thread_state->processor()->frontend();
+    kthread->fpu_context[i] = ctx->f[i];
-  cpu::ppc::CheckGlobalLock(
-      ctx, reinterpret_cast<void*>(&xe::global_critical_region::mutex()),
-      reinterpret_cast<void*>(&frontend->builtins()->global_lock_count));
-  return ctx->scratch;
-}
-
-void KeSaveFloatingPointState_entry(ppc_context_t& ctx) {
-  xe::Memory* memory = ctx->thread_state->memory();
-  unsigned int r13 = static_cast<unsigned int>(ctx->r[13]);
-
-
-
-  
-  r13_struct_t* st = memory->TranslateVirtual<r13_struct_t*>(r13);
-  /*
-                 lwz       r10, 0x150(r13)
-                lbz       r11, 0xA(r13)
-                tweqi     r10, 0
-                twnei     r11, 0
-  */
-
-  unsigned int r10 = st->field_150;
-  unsigned char r11 = st->field_A;
-
-  if (r10 == 0 || r11 != 0) {
-	  //trap!
   }
-
-  //should do mfmsr here
-  
-  unsigned int r3 = xe::load_and_swap<unsigned int>(&st->field_10);
-  
-  //too much work to do the mfmsr/mtmsr stuff right now
-  int to_store = -2049;
-  xe::store_and_swap(&st->field_10, (unsigned int)to_store);
-  xe::store_and_swap(&st->field_6, (short)to_store);
- 
-
-
-  if (r3 != ~0u) {
-    fpucontext_ptr_t* fpucontext =
-        memory->TranslateVirtual<fpucontext_ptr_t*>(r3);
-    xe::store_and_swap<uint64_t>(&fpucontext->saved_FPSCR, ctx->fpscr.value);
-	
-    for (unsigned int i = 0; i < 32; ++i) {
-      xe::store_and_swap(&fpucontext->saved_fpu_regs[i], ctx->f[i]);
-	}
-    xe::store_and_swap<unsigned short>(&fpucontext->field_9E, 0xD7FF);
-  }
-  ctx->processor->backend()->SetGuestRoundingMode(ctx.value(), 0);
-  ctx->fpscr.value = 0;
-  st->field_A = 1;
-
-  xe::store_and_swap(&st->field_10, r13 + 0x300);
-  ctx->r[3] = r3;
-
 }
+DECLARE_XBOXKRNL_EXPORT1(KeSaveFloatingPointState, kNone, kSketchy);
+
+void KeRestoreFloatingPointState_entry(const ppc_context_t& ctx) {
+  const X_KTHREAD* kthread = ctx->TranslateVirtual(
+      ctx->TranslateVirtualGPR<X_KPCR*>(ctx->r[13])->prcb_data.current_thread);
+
+  for (size_t i = 0; i < xe::countof(ctx->f); ++i) {
+    ctx->f[i] = kthread->fpu_context[i];
+  }
+}
+DECLARE_XBOXKRNL_EXPORT1(KeRestoreFloatingPointState, kNone, kSketchy);
 
-DECLARE_XBOXKRNL_EXPORT1(KeSaveFloatingPointState, kNone, kImplemented);
-#endif
 static qword_result_t KeQueryInterruptTime_entry(const ppc_context_t& ctx) {
   auto kstate = ctx->kernel_state;
   uint32_t ts_bundle = kstate->GetKeTimestampBundle();

src/xenia/kernel/xthread.cc

@@ -199,7 +199,7 @@ void XThread::InitializeGuestObject() {
   guest_thread->stack_limit = (this->stack_limit_);
   guest_thread->stack_kernel = (this->stack_base_ - 240);
   guest_thread->tls_address = (this->tls_dynamic_address_);
-  guest_thread->thread_state = 0;
+  guest_thread->thread_state = KTHREAD_STATE_INITIALIZED;
   uint32_t process_info_block_address =
       creation_params_.guest_process ? creation_params_.guest_process
                                      : this->kernel_state_->GetTitleProcess();
@@ -244,7 +244,8 @@ void XThread::InitializeGuestObject() {
   guest_thread->last_error = 0;
   guest_thread->unk_154.blink_ptr = v9 + 340;
   guest_thread->creation_flags = this->creation_params_.creation_flags;
-  guest_thread->unk_17C = 1;
+  // According to nukernel.
+  // guest_thread->host_xthread_stash = reinterpret_cast<void*>(this);
 
   /*
    * not doing this right at all! we're not using our threads context, because

src/xenia/kernel/xthread.h

@@ -338,11 +338,33 @@ struct X_KTHREAD {
   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!
+  // we handle context differently from a native kernel, so we can stash extra
+  // data here! the first 8 bytes of vscr are unused anyway
+  union {
+    vec128_t vscr;  // 0x170
+    struct {
+      void* host_xthread_stash;
+      uintptr_t vscr_remainder;
+    };
+  };
+
+  union {
+    // 2048 bytes
+    vec128_t vmx_context[128];  // 0x180
+    struct {
+      // 1536 bytes
+      X_KWAIT_BLOCK scratch_waitblock_memory[65];
+      // space for some more data!
+      uint32_t kernel_aux_stack_base_;
+      uint32_t kernel_aux_stack_current_;
+      uint32_t kernel_aux_stack_limit_;
+    };
+  };
+  xe::be<double> fpscr;            // 0x980
+  xe::be<double> fpu_context[32];  // 0x988
+
+  XAPC unk_A88;  // 0xA88
 };
 static_assert_size(X_KTHREAD, 0xAB0);