Merge pull request #147 from chrisps/canary_experimental

setup initial value of MSR on ppc context

src/xenia/base/main_init_win.cc

@@ -14,20 +14,32 @@
 // Includes Windows headers, so it goes after platform_win.h.
 #include "third_party/xbyak/xbyak/xbyak_util.h"
 
-class StartupAvxCheck {
+class StartupCpuFeatureCheck {
  public:
-  StartupAvxCheck() {
+  StartupCpuFeatureCheck() {
     Xbyak::util::Cpu cpu;
-    if (cpu.has(Xbyak::util::Cpu::tAVX)) {
-      return;
+    const char* error_message = nullptr;
+    if (!cpu.has(Xbyak::util::Cpu::tAVX)) {
+      error_message =
+          "Your CPU does not support AVX, which is required by Xenia. See "
+          "the "
+          "FAQ for system requirements at https://xenia.jp";
+    }
+    unsigned int data[4];
+    Xbyak::util::Cpu::getCpuid(0x80000001, data);
+    if (!(data[2] & (1U << 8))) {
+      error_message =
+          "Your cpu does not support PrefetchW, which Xenia Canary "
+          "requires.";
+    }
+    if (error_message == nullptr) {
+      return;
+    } else {
+      // TODO(gibbed): detect app type and printf instead, if needed?
+      MessageBoxA(nullptr, error_message, "Xenia Error",
+                  MB_OK | MB_ICONERROR | MB_SETFOREGROUND);
+      ExitProcess(static_cast<uint32_t>(-1));
     }
-    // TODO(gibbed): detect app type and printf instead, if needed?
-    MessageBoxA(
-        nullptr,
-        "Your CPU does not support AVX, which is required by Xenia. See the "
-        "FAQ for system requirements at https://xenia.jp",
-        "Xenia Error", MB_OK | MB_ICONERROR | MB_SETFOREGROUND);
-    ExitProcess(static_cast<uint32_t>(-1));
   }
 };
 
@@ -38,4 +50,4 @@ class StartupAvxCheck {
 // https://docs.microsoft.com/en-us/cpp/preprocessor/init-seg
 #pragma warning(suppress : 4073)
 #pragma init_seg(lib)
-static StartupAvxCheck gStartupAvxCheck;
+static StartupCpuFeatureCheck gStartupAvxCheck;

src/xenia/base/platform_amd64.cc

@@ -71,6 +71,13 @@ void InitFeatureFlags() {
     unsigned int data[4];
     Xbyak::util::Cpu::getCpuid(0x80000001, data);
     unsigned amd_flags = data[2];
+    // chrispy: do prefetchw manually, prefetchw is one of the features xbyak
+    // ignores if you have an amd cpu.
+
+    if (amd_flags & (1U << 8)) {
+      // no mask for prefetchw
+      feature_flags_ |= kX64EmitPrefetchW;
+    }
     if (amd_flags & (1U << 5)) {
       if ((cvars::x64_extension_mask & kX64EmitLZCNT) == kX64EmitLZCNT) {
         feature_flags_ |= kX64EmitLZCNT;

src/xenia/cpu/cpu_flags.cc

@@ -30,11 +30,6 @@ DEFINE_bool(trace_function_references, false,
 DEFINE_bool(trace_function_data, false,
             "Generate tracing for function result data.", "CPU");
 
-DEFINE_bool(
-    disable_global_lock, false,
-    "Disables global lock usage in guest code. Does not affect host code.",
-    "CPU");
-
 DEFINE_bool(validate_hir, false,
             "Perform validation checks on the HIR during compilation.", "CPU");
 

src/xenia/cpu/cpu_flags.h

@@ -22,8 +22,6 @@ DECLARE_bool(trace_function_coverage);
 DECLARE_bool(trace_function_references);
 DECLARE_bool(trace_function_data);
 
-DECLARE_bool(disable_global_lock);
-
 DECLARE_bool(validate_hir);
 
 DECLARE_uint64(pvr);

src/xenia/cpu/ppc/ppc_frontend.cc

@@ -11,6 +11,7 @@
 
 #include "xenia/base/atomic.h"
 #include "xenia/base/logging.h"
+#include "xenia/base/mutex.h"
 #include "xenia/cpu/ppc/ppc_context.h"
 #include "xenia/cpu/ppc/ppc_emit.h"
 #include "xenia/cpu/ppc/ppc_opcode_info.h"
@@ -56,15 +57,15 @@ Memory* PPCFrontend::memory() const { return processor_->memory(); }
 // Checks the state of the global lock and sets scratch to the current MSR
 // value.
 void CheckGlobalLock(PPCContext* ppc_context, void* arg0, void* arg1) {
-  auto global_mutex = reinterpret_cast<std::recursive_mutex*>(arg0);
+  auto global_mutex = reinterpret_cast<xe_global_mutex*>(arg0);
   auto global_lock_count = reinterpret_cast<int32_t*>(arg1);
-  std::lock_guard<std::recursive_mutex> lock(*global_mutex);
+  std::lock_guard<xe_global_mutex> lock(*global_mutex);
   ppc_context->scratch = *global_lock_count ? 0 : 0x8000;
 }
 
 // Enters the global lock. Safe to recursion.
 void EnterGlobalLock(PPCContext* ppc_context, void* arg0, void* arg1) {
-  auto global_mutex = reinterpret_cast<std::recursive_mutex*>(arg0);
+  auto global_mutex = reinterpret_cast<xe_global_mutex*>(arg0);
   auto global_lock_count = reinterpret_cast<int32_t*>(arg1);
   global_mutex->lock();
   xe::atomic_inc(global_lock_count);
@@ -72,7 +73,7 @@ void EnterGlobalLock(PPCContext* ppc_context, void* arg0, void* arg1) {
 
 // Leaves the global lock. Safe to recursion.
 void LeaveGlobalLock(PPCContext* ppc_context, void* arg0, void* arg1) {
-  auto global_mutex = reinterpret_cast<std::recursive_mutex*>(arg0);
+  auto global_mutex = reinterpret_cast<xe_global_mutex*>(arg0);
   auto global_lock_count = reinterpret_cast<int32_t*>(arg1);
   auto new_lock_count = xe::atomic_dec(global_lock_count);
   assert_true(new_lock_count >= 0);

src/xenia/cpu/thread_state.cc

@@ -95,6 +95,12 @@ ThreadState::ThreadState(Processor* processor, uint32_t thread_id,
   context_->r[1] = stack_base;
   context_->r[13] = pcr_address;
   // fixme: VSCR must be set here!
+  context_->msr = 0x9030;  // dumped from a real 360, 0x8000
+
+  //this register can be used for arbitrary data according to the PPC docs
+  //but the suggested use is to mark which vector registers are in use, for faster save/restore
+  //it seems unlikely anything uses this, especially since we have way more than 32 vrs, but setting it to all ones seems closer to correct than 0
+  context_->vrsave = ~0u;
 }
 
 ThreadState::~ThreadState() {