Merge branch 'master' into d3d12

src/xenia/cpu/lzx.cc

@@ -0,0 +1,183 @@
+/**
+ ******************************************************************************
+ * Xenia : Xbox 360 Emulator Research Project                                 *
+ ******************************************************************************
+ * Copyright 2013 Ben Vanik. All rights reserved.                             *
+ * Released under the BSD license - see LICENSE in the root for more details. *
+ ******************************************************************************
+ */
+
+#include "xenia/cpu/lzx.h"
+
+#include <algorithm>
+#include <climits>
+
+#include "xenia/base/byte_order.h"
+#include "xenia/base/logging.h"
+#include "xenia/base/math.h"
+#include "xenia/base/memory.h"
+#include "xenia/kernel/util/xex2_info.h"
+
+#include "third_party/mspack/lzx.h"
+#include "third_party/mspack/mspack.h"
+
+typedef struct mspack_memory_file_t {
+  struct mspack_system sys;
+  void* buffer;
+  off_t buffer_size;
+  off_t offset;
+} mspack_memory_file;
+mspack_memory_file* mspack_memory_open(struct mspack_system* sys, void* buffer,
+                                       const size_t buffer_size) {
+  assert_true(buffer_size < INT_MAX);
+  if (buffer_size >= INT_MAX) {
+    return NULL;
+  }
+  mspack_memory_file* memfile =
+      (mspack_memory_file*)calloc(1, sizeof(mspack_memory_file));
+  if (!memfile) {
+    return NULL;
+  }
+  memfile->buffer = buffer;
+  memfile->buffer_size = (off_t)buffer_size;
+  memfile->offset = 0;
+  return memfile;
+}
+void mspack_memory_close(mspack_memory_file* file) {
+  mspack_memory_file* memfile = (mspack_memory_file*)file;
+  free(memfile);
+}
+int mspack_memory_read(struct mspack_file* file, void* buffer, int chars) {
+  mspack_memory_file* memfile = (mspack_memory_file*)file;
+  const off_t remaining = memfile->buffer_size - memfile->offset;
+  const off_t total = std::min(static_cast<off_t>(chars), remaining);
+  std::memcpy(buffer, (uint8_t*)memfile->buffer + memfile->offset, total);
+  memfile->offset += total;
+  return (int)total;
+}
+int mspack_memory_write(struct mspack_file* file, void* buffer, int chars) {
+  mspack_memory_file* memfile = (mspack_memory_file*)file;
+  const off_t remaining = memfile->buffer_size - memfile->offset;
+  const off_t total = std::min(static_cast<off_t>(chars), remaining);
+  std::memcpy((uint8_t*)memfile->buffer + memfile->offset, buffer, total);
+  memfile->offset += total;
+  return (int)total;
+}
+void* mspack_memory_alloc(struct mspack_system* sys, size_t chars) {
+  return std::calloc(chars, 1);
+}
+void mspack_memory_free(void* ptr) { free(ptr); }
+void mspack_memory_copy(void* src, void* dest, size_t chars) {
+  std::memcpy(dest, src, chars);
+}
+struct mspack_system* mspack_memory_sys_create() {
+  struct mspack_system* sys =
+      (struct mspack_system*)std::calloc(1, sizeof(struct mspack_system));
+  if (!sys) {
+    return NULL;
+  }
+  sys->read = mspack_memory_read;
+  sys->write = mspack_memory_write;
+  sys->alloc = mspack_memory_alloc;
+  sys->free = mspack_memory_free;
+  sys->copy = mspack_memory_copy;
+  return sys;
+}
+void mspack_memory_sys_destroy(struct mspack_system* sys) { free(sys); }
+
+int lzx_decompress(const void* lzx_data, size_t lzx_len, void* dest,
+                   size_t dest_len, uint32_t window_size, void* window_data,
+                   size_t window_data_len) {
+  uint32_t window_bits = 0;
+  uint32_t temp_sz = window_size;
+  for (size_t m = 0; m < 32; m++, window_bits++) {
+    temp_sz >>= 1;
+    if (temp_sz == 0x00000000) {
+      break;
+    }
+  }
+
+  int result_code = 1;
+
+  mspack_system* sys = mspack_memory_sys_create();
+  mspack_memory_file* lzxsrc =
+      mspack_memory_open(sys, (void*)lzx_data, lzx_len);
+  mspack_memory_file* lzxdst = mspack_memory_open(sys, dest, dest_len);
+  lzxd_stream* lzxd =
+      lzxd_init(sys, (struct mspack_file*)lzxsrc, (struct mspack_file*)lzxdst,
+                window_bits, 0, 0x8000, (off_t)dest_len, 0);
+
+  if (lzxd) {
+    if (window_data) {
+      // zero the window and then copy window_data to the end of it
+      std::memset(lzxd->window, 0, window_data_len);
+      std::memcpy(lzxd->window + (window_size - window_data_len), window_data,
+                  window_data_len);
+      lzxd->ref_data_size = (uint32_t)window_data_len;
+    }
+
+    result_code = lzxd_decompress(lzxd, (off_t)dest_len);
+
+    lzxd_free(lzxd);
+    lzxd = NULL;
+  }
+  if (lzxsrc) {
+    mspack_memory_close(lzxsrc);
+    lzxsrc = NULL;
+  }
+  if (lzxdst) {
+    mspack_memory_close(lzxdst);
+    lzxdst = NULL;
+  }
+  if (sys) {
+    mspack_memory_sys_destroy(sys);
+    sys = NULL;
+  }
+
+  return result_code;
+}
+
+int lzxdelta_apply_patch(xe::xex2_delta_patch* patch, size_t patch_len,
+                         uint32_t window_size, void* dest) {
+  void* patch_end = (char*)patch + patch_len;
+  auto* cur_patch = patch;
+
+  while (patch_end > cur_patch) {
+    int patch_sz = -4;  // 0 byte patches need us to remove 4 byte from next
+                        // patch addr because of patch_data field
+    if (cur_patch->compressed_len == 0 && cur_patch->uncompressed_len == 0 &&
+        cur_patch->new_addr == 0 && cur_patch->old_addr == 0)
+      break;
+    switch (cur_patch->compressed_len) {
+      case 0:  // fill with 0
+        std::memset((char*)dest + cur_patch->new_addr, 0,
+                    cur_patch->uncompressed_len);
+        break;
+      case 1:  // copy from old -> new
+        std::memcpy((char*)dest + cur_patch->new_addr,
+                    (char*)dest + cur_patch->old_addr,
+                    cur_patch->uncompressed_len);
+        break;
+      default:  // delta patch
+        patch_sz =
+            cur_patch->compressed_len - 4;  // -4 because of patch_data field
+
+        int result = lzx_decompress(
+            cur_patch->patch_data, cur_patch->compressed_len,
+            (char*)dest + cur_patch->new_addr, cur_patch->uncompressed_len,
+            window_size, (char*)dest + cur_patch->old_addr,
+            cur_patch->uncompressed_len);
+
+        if (result) {
+          return result;
+        }
+        break;
+    }
+
+    cur_patch++;
+    cur_patch = (xe::xex2_delta_patch*)((char*)cur_patch +
+                                        patch_sz);  // TODO: make this less ugly
+  }
+
+  return 0;
+}

src/xenia/cpu/lzx.h

@@ -0,0 +1,29 @@
+/**
+ ******************************************************************************
+ * Xenia : Xbox 360 Emulator Research Project                                 *
+ ******************************************************************************
+ * Copyright 2013 Ben Vanik. All rights reserved.                             *
+ * Released under the BSD license - see LICENSE in the root for more details. *
+ ******************************************************************************
+ */
+
+#ifndef XENIA_CPU_LZX_H_
+#define XENIA_CPU_LZX_H_
+
+#include <string>
+#include <vector>
+
+#include "xenia/cpu/module.h"
+
+namespace xe {
+struct xex2_delta_patch;
+}  // namespace xe
+
+int lzx_decompress(const void* lzx_data, size_t lzx_len, void* dest,
+                   size_t dest_len, uint32_t window_size, void* window_data,
+                   size_t window_data_len);
+
+int lzxdelta_apply_patch(xe::xex2_delta_patch* patch, size_t patch_len,
+                         uint32_t window_size, void* dest);
+
+#endif  // XENIA_CPU_LZX_H_

src/xenia/cpu/xex_module.cc

@@ -17,6 +17,7 @@
 #include "xenia/base/memory.h"
 #include "xenia/cpu/cpu_flags.h"
 #include "xenia/cpu/export_resolver.h"
+#include "xenia/cpu/lzx.h"
 #include "xenia/cpu/processor.h"
 #include "xenia/kernel/kernel_state.h"
 #include "xenia/kernel/xmodule.h"
@@ -24,8 +25,6 @@
 #include "third_party/crypto/TinySHA1.hpp"
 #include "third_party/crypto/rijndael-alg-fst.c"
 #include "third_party/crypto/rijndael-alg-fst.h"
-#include "third_party/mspack/lzx.h"
-#include "third_party/mspack/mspack.h"
 #include "third_party/pe/pe_image.h"
 
 static const uint8_t xe_xex2_retail_key[16] = {
@@ -35,165 +34,6 @@ static const uint8_t xe_xex2_devkit_key[16] = {
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00};
 
-typedef struct mspack_memory_file_t {
-  struct mspack_system sys;
-  void* buffer;
-  off_t buffer_size;
-  off_t offset;
-} mspack_memory_file;
-mspack_memory_file* mspack_memory_open(struct mspack_system* sys, void* buffer,
-                                       const size_t buffer_size) {
-  assert_true(buffer_size < INT_MAX);
-  if (buffer_size >= INT_MAX) {
-    return NULL;
-  }
-  mspack_memory_file* memfile =
-      (mspack_memory_file*)calloc(1, sizeof(mspack_memory_file));
-  if (!memfile) {
-    return NULL;
-  }
-  memfile->buffer = buffer;
-  memfile->buffer_size = (off_t)buffer_size;
-  memfile->offset = 0;
-  return memfile;
-}
-void mspack_memory_close(mspack_memory_file* file) {
-  mspack_memory_file* memfile = (mspack_memory_file*)file;
-  free(memfile);
-}
-int mspack_memory_read(struct mspack_file* file, void* buffer, int chars) {
-  mspack_memory_file* memfile = (mspack_memory_file*)file;
-  const off_t remaining = memfile->buffer_size - memfile->offset;
-  const off_t total = std::min(static_cast<off_t>(chars), remaining);
-  memcpy(buffer, (uint8_t*)memfile->buffer + memfile->offset, total);
-  memfile->offset += total;
-  return (int)total;
-}
-int mspack_memory_write(struct mspack_file* file, void* buffer, int chars) {
-  mspack_memory_file* memfile = (mspack_memory_file*)file;
-  const off_t remaining = memfile->buffer_size - memfile->offset;
-  const off_t total = std::min(static_cast<off_t>(chars), remaining);
-  memcpy((uint8_t*)memfile->buffer + memfile->offset, buffer, total);
-  memfile->offset += total;
-  return (int)total;
-}
-void* mspack_memory_alloc(struct mspack_system* sys, size_t chars) {
-  return calloc(chars, 1);
-}
-void mspack_memory_free(void* ptr) { free(ptr); }
-void mspack_memory_copy(void* src, void* dest, size_t chars) {
-  memcpy(dest, src, chars);
-}
-struct mspack_system* mspack_memory_sys_create() {
-  struct mspack_system* sys =
-      (struct mspack_system*)calloc(1, sizeof(struct mspack_system));
-  if (!sys) {
-    return NULL;
-  }
-  sys->read = mspack_memory_read;
-  sys->write = mspack_memory_write;
-  sys->alloc = mspack_memory_alloc;
-  sys->free = mspack_memory_free;
-  sys->copy = mspack_memory_copy;
-  return sys;
-}
-void mspack_memory_sys_destroy(struct mspack_system* sys) { free(sys); }
-
-int lzx_decompress(const void* lzx_data, size_t lzx_len, void* dest,
-                   size_t dest_len, uint32_t window_size, void* window_data,
-                   size_t window_data_len) {
-  uint32_t window_bits = 0;
-  uint32_t temp_sz = window_size;
-  for (size_t m = 0; m < 32; m++, window_bits++) {
-    temp_sz >>= 1;
-    if (temp_sz == 0x00000000) {
-      break;
-    }
-  }
-
-  int result_code = 1;
-
-  mspack_system* sys = mspack_memory_sys_create();
-  mspack_memory_file* lzxsrc =
-      mspack_memory_open(sys, (void*)lzx_data, lzx_len);
-  mspack_memory_file* lzxdst = mspack_memory_open(sys, dest, dest_len);
-  lzxd_stream* lzxd =
-      lzxd_init(sys, (struct mspack_file*)lzxsrc, (struct mspack_file*)lzxdst,
-                window_bits, 0, 0x8000, (off_t)dest_len, 0);
-
-  if (lzxd) {
-    if (window_data) {
-      // zero the window and then copy window_data to the end of it
-      memset(lzxd->window, 0, window_data_len);
-      memcpy(lzxd->window + (window_size - window_data_len), window_data,
-             window_data_len);
-    }
-
-    result_code = lzxd_decompress(lzxd, (off_t)dest_len);
-
-    lzxd_free(lzxd);
-    lzxd = NULL;
-  }
-  if (lzxsrc) {
-    mspack_memory_close(lzxsrc);
-    lzxsrc = NULL;
-  }
-  if (lzxdst) {
-    mspack_memory_close(lzxdst);
-    lzxdst = NULL;
-  }
-  if (sys) {
-    mspack_memory_sys_destroy(sys);
-    sys = NULL;
-  }
-
-  return result_code;
-}
-
-int lzxdelta_apply_patch(xe::xex2_delta_patch* patch, size_t patch_len,
-                         uint32_t window_size, void* dest) {
-  void* patch_end = (char*)patch + patch_len;
-  auto* cur_patch = patch;
-
-  while (patch_end > cur_patch) {
-    int patch_sz = -4;  // 0 byte patches need us to remove 4 byte from next
-                        // patch addr because of patch_data field
-    if (cur_patch->compressed_len == 0 && cur_patch->uncompressed_len == 0 &&
-        cur_patch->new_addr == 0 && cur_patch->old_addr == 0)
-      break;
-    switch (cur_patch->compressed_len) {
-      case 0:  // fill with 0
-        memset((char*)dest + cur_patch->new_addr, 0,
-               cur_patch->uncompressed_len);
-        break;
-      case 1:  // copy from old -> new
-        memcpy((char*)dest + cur_patch->new_addr,
-               (char*)dest + cur_patch->old_addr, cur_patch->uncompressed_len);
-        break;
-      default:  // delta patch
-        patch_sz =
-            cur_patch->compressed_len - 4;  // -4 because of patch_data field
-
-        int result = lzx_decompress(
-            cur_patch->patch_data, cur_patch->compressed_len,
-            (char*)dest + cur_patch->new_addr, cur_patch->uncompressed_len,
-            window_size, (char*)dest + cur_patch->old_addr,
-            cur_patch->uncompressed_len);
-
-        if (result) {
-          return result;
-        }
-        break;
-    }
-
-    cur_patch++;
-    cur_patch = (xe::xex2_delta_patch*)((char*)cur_patch +
-                                        patch_sz);  // TODO: make this less ugly
-  }
-
-  return 0;
-}
-
 void aes_decrypt_buffer(const uint8_t* session_key, const uint8_t* input_buffer,
                         const size_t input_size, uint8_t* output_buffer,
                         const size_t output_size) {
@@ -1144,9 +984,8 @@ bool XexModule::LoadContinue() {
       }
     }
 
-    auto library_data = reinterpret_cast<uint8_t*>(opt_import_libraries) +
+    auto library_data = reinterpret_cast<uint8_t*>(opt_import_libraries);
-                        opt_import_libraries->string_table.size + 12;
+    uint32_t library_offset = opt_import_libraries->string_table.size + 12;
-    uint32_t library_offset = 0;
     while (library_offset < opt_import_libraries->size) {
       auto library =
           reinterpret_cast<xex2_import_library*>(library_data + library_offset);

src/xenia/kernel/util/shim_utils.h

@@ -564,6 +564,10 @@ using xe::cpu::ExportTag;
   DECLARE_EXPORT(xboxkrnl, name, category,                             \
                  xe::cpu::ExportTag::tag1 | xe::cpu::ExportTag::tag2 | \
                      xe::cpu::ExportTag::tag3)
+#define DECLARE_XBOXKRNL_EXPORT4(name, category, tag1, tag2, tag3, tag4) \
+  DECLARE_EXPORT(xboxkrnl, name, category,                               \
+                 xe::cpu::ExportTag::tag1 | xe::cpu::ExportTag::tag2 |   \
+                     xe::cpu::ExportTag::tag3 | xe::cpu::ExportTag::tag4)
 
 }  // namespace kernel
 }  // namespace xe

src/xenia/kernel/xam/xam_info.cc

@@ -28,28 +28,30 @@ namespace xam {
 constexpr uint32_t X_LANGUAGE_ENGLISH = 1;
 constexpr uint32_t X_LANGUAGE_JAPANESE = 2;
 
+// Empty stub schema binary.
+uint8_t schema_bin[] = {
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x00,
+    0x00, 0x2C, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x2C, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x2C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x18,
+};
+
 dword_result_t XamGetOnlineSchema() {
   static uint32_t schema_guest = 0;
-  static uint32_t schema_ptr_guest = 0;
 
   if (!schema_guest) {
-    // create a dummy schema, 8 bytes of 0 seems to work fine
+    schema_guest =
-    // (with another 8 bytes for schema ptr/schema size)
+        kernel_state()->memory()->SystemHeapAlloc(8 + sizeof(schema_bin));
-    schema_guest = kernel_state()->memory()->SystemHeapAlloc(16);
-    schema_ptr_guest = schema_guest + 8;
-
     auto schema = kernel_state()->memory()->TranslateVirtual(schema_guest);
-    memset(schema, 0, 16);
+    std::memcpy(schema + 8, schema_bin, sizeof(schema_bin));
-
+    xe::store_and_swap<uint32_t>(schema + 0, schema_guest + 8);
-    // store schema ptr + size
+    xe::store_and_swap<uint32_t>(schema + 4, sizeof(schema_bin));
-    xe::store_and_swap<uint32_t>(schema + 0x8, schema_guest);
-    xe::store_and_swap<uint32_t>(schema + 0xC, 0x8);
   }
 
   // return pointer to the schema ptr/schema size struct
-  return schema_ptr_guest;
+  return schema_guest;
 }
-DECLARE_XAM_EXPORT2(XamGetOnlineSchema, kNone, kImplemented, kSketchy);
+DECLARE_XAM_EXPORT1(XamGetOnlineSchema, kNone, kImplemented);
 
 void XamFormatDateString(dword_t unk, qword_t filetime, lpvoid_t buffer,
                          dword_t buffer_length) {

src/xenia/kernel/xboxkrnl/xboxkrnl_threading.cc

@@ -373,8 +373,7 @@ void KeInitializeEvent(pointer_t<X_KEVENT> event_ptr, dword_t event_type,
 }
 DECLARE_XBOXKRNL_EXPORT1(KeInitializeEvent, kThreading, kImplemented);
 
-dword_result_t KeSetEvent(pointer_t<X_KEVENT> event_ptr, dword_t increment,
+uint32_t keSetEvent(X_KEVENT* event_ptr, uint32_t increment, uint32_t wait) {
-                          dword_t wait) {
   auto ev = XObject::GetNativeObject<XEvent>(kernel_state(), event_ptr);
   if (!ev) {
     assert_always();
@@ -383,6 +382,11 @@ dword_result_t KeSetEvent(pointer_t<X_KEVENT> event_ptr, dword_t increment,
 
   return ev->Set(increment, !!wait);
 }
+
+dword_result_t KeSetEvent(pointer_t<X_KEVENT> event_ptr, dword_t increment,
+                          dword_t wait) {
+  return keSetEvent(event_ptr, increment, wait);
+}
 DECLARE_XBOXKRNL_EXPORT2(KeSetEvent, kThreading, kImplemented, kHighFrequency);
 
 dword_result_t KePulseEvent(pointer_t<X_KEVENT> event_ptr, dword_t increment,
@@ -508,9 +512,8 @@ void KeInitializeSemaphore(pointer_t<X_KSEMAPHORE> semaphore_ptr, dword_t count,
 }
 DECLARE_XBOXKRNL_EXPORT1(KeInitializeSemaphore, kThreading, kImplemented);
 
-dword_result_t KeReleaseSemaphore(pointer_t<X_KSEMAPHORE> semaphore_ptr,
+uint32_t keReleaseSemaphore(X_KSEMAPHORE* semaphore_ptr, uint32_t increment,
-                                  dword_t increment, dword_t adjustment,
+                            uint32_t adjustment, uint32_t wait) {
-                                  dword_t wait) {
   auto sem =
       XObject::GetNativeObject<XSemaphore>(kernel_state(), semaphore_ptr);
   if (!sem) {
@@ -523,6 +526,12 @@ dword_result_t KeReleaseSemaphore(pointer_t<X_KSEMAPHORE> semaphore_ptr,
 
   return sem->ReleaseSemaphore(adjustment);
 }
+
+dword_result_t KeReleaseSemaphore(pointer_t<X_KSEMAPHORE> semaphore_ptr,
+                                  dword_t increment, dword_t adjustment,
+                                  dword_t wait) {
+  return keReleaseSemaphore(semaphore_ptr, increment, adjustment, wait);
+}
 DECLARE_XBOXKRNL_EXPORT1(KeReleaseSemaphore, kThreading, kImplemented);
 
 dword_result_t NtCreateSemaphore(lpdword_t handle_ptr,
@@ -719,9 +728,9 @@ dword_result_t NtCancelTimer(dword_t timer_handle,
 }
 DECLARE_XBOXKRNL_EXPORT1(NtCancelTimer, kThreading, kImplemented);
 
-dword_result_t KeWaitForSingleObject(lpvoid_t object_ptr, dword_t wait_reason,
+uint32_t keWaitForSingleObject(void* object_ptr, uint32_t wait_reason,
-                                     dword_t processor_mode, dword_t alertable,
+                               uint32_t processor_mode, uint32_t alertable,
-                                     lpqword_t timeout_ptr) {
+                               uint64_t* timeout) {
   auto object = XObject::GetNativeObject<XObject>(kernel_state(), object_ptr);
 
   if (!object) {
@@ -730,12 +739,19 @@ dword_result_t KeWaitForSingleObject(lpvoid_t object_ptr, dword_t wait_reason,
     return X_STATUS_ABANDONED_WAIT_0;
   }
 
-  uint64_t timeout = timeout_ptr ? static_cast<uint64_t>(*timeout_ptr) : 0u;
+  X_STATUS result =
-  X_STATUS result = object->Wait(wait_reason, processor_mode, alertable,
+      object->Wait(wait_reason, processor_mode, alertable, timeout);
-                                 timeout_ptr ? &timeout : nullptr);
 
   return result;
 }
+
+dword_result_t KeWaitForSingleObject(lpvoid_t object_ptr, dword_t wait_reason,
+                                     dword_t processor_mode, dword_t alertable,
+                                     lpqword_t timeout_ptr) {
+  uint64_t timeout = timeout_ptr ? static_cast<uint64_t>(*timeout_ptr) : 0u;
+  return keWaitForSingleObject(object_ptr, wait_reason, processor_mode,
+                               alertable, &timeout);
+}
 DECLARE_XBOXKRNL_EXPORT3(KeWaitForSingleObject, kThreading, kImplemented,
                          kBlocking, kHighFrequency);
 
@@ -844,13 +860,12 @@ dword_result_t NtSignalAndWaitForSingleObjectEx(dword_t signal_handle,
 DECLARE_XBOXKRNL_EXPORT3(NtSignalAndWaitForSingleObjectEx, kThreading,
                          kImplemented, kBlocking, kHighFrequency);
 
-dword_result_t KfAcquireSpinLock(lpdword_t lock_ptr) {
+uint32_t keKfAcquireSpinLock(uint32_t* lock) {
   // XELOGD(
   //     "KfAcquireSpinLock(%.8X)",
   //     lock_ptr);
 
   // Lock.
-  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   while (!xe::atomic_cas(0, 1, lock)) {
     // Spin!
     // TODO(benvanik): error on deadlock?
@@ -863,18 +878,27 @@ dword_result_t KfAcquireSpinLock(lpdword_t lock_ptr) {
 
   return old_irql;
 }
+
+dword_result_t KfAcquireSpinLock(lpdword_t lock_ptr) {
+  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
+  return keKfAcquireSpinLock(lock);
+}
 DECLARE_XBOXKRNL_EXPORT3(KfAcquireSpinLock, kThreading, kImplemented, kBlocking,
                          kHighFrequency);
 
-void KfReleaseSpinLock(lpdword_t lock_ptr, dword_t old_irql) {
+void keKfReleaseSpinLock(uint32_t* lock, dword_t old_irql) {
   // Restore IRQL.
   XThread* thread = XThread::GetCurrentThread();
   thread->LowerIrql(old_irql);
 
   // Unlock.
-  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
   xe::atomic_dec(lock);
 }
+
+void KfReleaseSpinLock(lpdword_t lock_ptr, dword_t old_irql) {
+  auto lock = reinterpret_cast<uint32_t*>(lock_ptr.host_address());
+  keKfReleaseSpinLock(lock, old_irql);
+}
 DECLARE_XBOXKRNL_EXPORT2(KfReleaseSpinLock, kThreading, kImplemented,
                          kHighFrequency);
 
@@ -1092,7 +1116,7 @@ struct X_ERWLOCK {
   be<uint32_t> readers_entry_count;    // 0xC
   X_KEVENT writer_event;               // 0x10
   X_KSEMAPHORE reader_semaphore;       // 0x20
-  be<uint32_t> spin_lock;              // 0x34
+  uint32_t spin_lock;                  // 0x34
 };
 
 void ExInitializeReadWriteLock(pointer_t<X_ERWLOCK> lock_ptr) {
@@ -1105,6 +1129,51 @@ void ExInitializeReadWriteLock(pointer_t<X_ERWLOCK> lock_ptr) {
 }
 DECLARE_XBOXKRNL_EXPORT1(ExInitializeReadWriteLock, kThreading, kImplemented);
 
+void ExAcquireReadWriteLockExclusive(pointer_t<X_ERWLOCK> lock_ptr) {
+  auto old_irql = keKfAcquireSpinLock(&lock_ptr->spin_lock);
+  lock_ptr->lock_count++;
+  keKfReleaseSpinLock(&lock_ptr->spin_lock, old_irql);
+
+  if (!lock_ptr->lock_count) {
+    return;
+  }
+
+  lock_ptr->writers_waiting_count++;
+  keWaitForSingleObject(&lock_ptr->writer_event, 0, 0, 0, nullptr);
+}
+DECLARE_XBOXKRNL_EXPORT4(ExAcquireReadWriteLockExclusive, kThreading,
+                         kImplemented, kBlocking, kHighFrequency, kSketchy);
+
+void ExReleaseReadWriteLock(pointer_t<X_ERWLOCK> lock_ptr) {
+  auto old_irql = keKfAcquireSpinLock(&lock_ptr->spin_lock);
+  lock_ptr->lock_count--;
+
+  if (lock_ptr->lock_count < 0) {
+    keKfReleaseSpinLock(&lock_ptr->spin_lock, old_irql);
+    return;
+  }
+
+  if (!lock_ptr->readers_entry_count) {
+    auto readers_waiting_count = lock_ptr->readers_waiting_count;
+    if (readers_waiting_count) {
+      lock_ptr->readers_waiting_count = 0;
+      lock_ptr->readers_entry_count = readers_waiting_count;
+      keKfReleaseSpinLock(&lock_ptr->spin_lock, old_irql);
+      keReleaseSemaphore(&lock_ptr->reader_semaphore, 1, readers_waiting_count,
+                         0);
+      return;
+    }
+  }
+
+  auto count = lock_ptr->readers_entry_count--;
+  keKfReleaseSpinLock(&lock_ptr->spin_lock, old_irql);
+  if (!count) {
+    keSetEvent(&lock_ptr->writer_event, 1, 0);
+  }
+}
+DECLARE_XBOXKRNL_EXPORT2(ExReleaseReadWriteLock, kThreading, kImplemented,
+                         kSketchy);
+
 // NOTE: This function is very commonly inlined, and probably won't be called!
 pointer_result_t InterlockedPushEntrySList(
     pointer_t<X_SLIST_HEADER> plist_ptr, pointer_t<X_SINGLE_LIST_ENTRY> entry) {