From 77535b7cecf33d58b98294fae304b24bb6c2940a Mon Sep 17 00:00:00 2001
From: "Herman S." <429230+has207@users.noreply.github.com>
Date: Tue, 10 Feb 2026 10:10:03 +0900
Subject: [PATCH] [Base] Replace CRITICAL_SECTION with SRWLOCK based mutexes

xe_fast_mutex now also asserts on recursive lock attempts
---
 src/xenia/base/mutex.cc | 82 ++++++++++++++++++++++-------------------
 src/xenia/base/mutex.h  | 35 +++++++++---------
 2 files changed, 62 insertions(+), 55 deletions(-)

diff --git a/src/xenia/base/mutex.cc b/src/xenia/base/mutex.cc
index 4fc37de9b..5f62ae39b 100644
--- a/src/xenia/base/mutex.cc
+++ b/src/xenia/base/mutex.cc
@@ -14,55 +14,61 @@
 
 namespace xe {
 #if XE_PLATFORM_WIN32 == 1 && XE_ENABLE_FAST_WIN32_MUTEX == 1
-// default spincount for entercriticalsection is insane on windows, 0x20007D0i64
-// (33556432 times!!) when a lock is highly contended performance degrades
-// sharply on some processors todo: perhaps we should have a set of optional
-// jobs that processors can do instead of spinning, for instance, sorting a list
-// so we have better locality later or something
-#define XE_CRIT_SPINCOUNT 128
-/*
-chrispy: todo, if a thread exits before releasing the global mutex we need to
-check this and release the mutex one way to do this is by using FlsAlloc and
-PFLS_CALLBACK_FUNCTION, which gets called with the fiber local data when a
-thread exits
-*/
-
-static CRITICAL_SECTION* global_critical_section(xe_global_mutex* mutex) {
-  return reinterpret_cast<CRITICAL_SECTION*>(mutex);
-}
-
-xe_global_mutex::xe_global_mutex() {
-  InitializeCriticalSectionEx(global_critical_section(this), XE_CRIT_SPINCOUNT,
-                              CRITICAL_SECTION_NO_DEBUG_INFO);
-}
-xe_global_mutex ::~xe_global_mutex() {
-  DeleteCriticalSection(global_critical_section(this));
-}
 
+// xe_global_mutex: recursive mutex via SRWLOCK
 void xe_global_mutex::lock() {
-  EnterCriticalSection(global_critical_section(this));
+  DWORD self = GetCurrentThreadId();
+  if (owner_thread_ == self) {
+    ++recursion_count_;
+    return;
+  }
+  AcquireSRWLockExclusive(&srwlock_);
+  owner_thread_ = self;
+  recursion_count_ = 1;
 }
+
 void xe_global_mutex::unlock() {
-  LeaveCriticalSection(global_critical_section(this));
+  if (--recursion_count_ == 0) {
+    owner_thread_ = 0;
+    ReleaseSRWLockExclusive(&srwlock_);
+  }
 }
+
 bool xe_global_mutex::try_lock() {
-  BOOL success = TryEnterCriticalSection(global_critical_section(this));
-  return success;
+  DWORD self = GetCurrentThreadId();
+  if (owner_thread_ == self) {
+    ++recursion_count_;
+    return true;
+  }
+  if (TryAcquireSRWLockExclusive(&srwlock_)) {
+    owner_thread_ = self;
+    recursion_count_ = 1;
+    return true;
+  }
+  return false;
 }
 
-CRITICAL_SECTION* fast_crit(xe_fast_mutex* mutex) {
-  return reinterpret_cast<CRITICAL_SECTION*>(mutex);
+// xe_fast_mutex: non-recursive mutex via SRWLOCK
+void xe_fast_mutex::lock() {
+  DWORD self = GetCurrentThreadId();
+  if (owner_thread_ == self) {
+    assert_always("xe_fast_mutex: recursive lock detected");
+  }
+  AcquireSRWLockExclusive(&srwlock_);
+  owner_thread_ = self;
+}
+
+void xe_fast_mutex::unlock() {
+  owner_thread_ = 0;
+  ReleaseSRWLockExclusive(&srwlock_);
 }
-xe_fast_mutex::xe_fast_mutex() {
-  InitializeCriticalSectionEx(fast_crit(this), XE_CRIT_SPINCOUNT,
-                              CRITICAL_SECTION_NO_DEBUG_INFO);
-}
-xe_fast_mutex::~xe_fast_mutex() { DeleteCriticalSection(fast_crit(this)); }
 
-void xe_fast_mutex::lock() { EnterCriticalSection(fast_crit(this)); }
-void xe_fast_mutex::unlock() { LeaveCriticalSection(fast_crit(this)); }
 bool xe_fast_mutex::try_lock() {
-  return TryEnterCriticalSection(fast_crit(this));
+  if (TryAcquireSRWLockExclusive(&srwlock_)) {
+    owner_thread_ = GetCurrentThreadId();
+    return true;
+  }
+  return false;
 }
 #endif
 global_mutex_type& global_critical_region::mutex() {
diff --git a/src/xenia/base/mutex.h b/src/xenia/base/mutex.h
index d5e691c60..665f38f97 100644
--- a/src/xenia/base/mutex.h
+++ b/src/xenia/base/mutex.h
@@ -10,27 +10,27 @@
 #ifndef XENIA_BASE_MUTEX_H_
 #define XENIA_BASE_MUTEX_H_
 #include <mutex>
-#include "memory.h"
+#include <thread>
 #include "platform.h"
+#if XE_PLATFORM_WIN32
+#include "platform_win.h"
+#elif XE_PLATFORM_LINUX
+#include <sys/types.h>
+#endif
+#include "memory.h"
 #define XE_ENABLE_FAST_WIN32_MUTEX 1
 namespace xe {
 
 #if XE_PLATFORM_WIN32 == 1 && XE_ENABLE_FAST_WIN32_MUTEX == 1
-/*
-   must conform to
-   BasicLockable:https://en.cppreference.com/w/cpp/named_req/BasicLockable as
-   well as Lockable: https://en.cppreference.com/w/cpp/named_req/Lockable
-
-   this emulates a recursive mutex, except with far less overhead
-*/
-
+// Recursive mutex using SRWLOCK.
 class alignas(4096) xe_global_mutex {
-  XE_MAYBE_UNUSED
-  char detail[64];
+  SRWLOCK srwlock_ = SRWLOCK_INIT;
+  DWORD owner_thread_ = 0;
+  uint32_t recursion_count_ = 0;
 
  public:
-  xe_global_mutex();
-  ~xe_global_mutex();
+  xe_global_mutex() = default;
+  ~xe_global_mutex() = default;
 
   void lock();
   void unlock();
@@ -38,13 +38,14 @@ class alignas(4096) xe_global_mutex {
 };
 using global_mutex_type = xe_global_mutex;
 
+// Non-recursive mutex using SRWLOCK.
 class alignas(64) xe_fast_mutex {
-  XE_MAYBE_UNUSED
-  char detail[64];
+  SRWLOCK srwlock_ = SRWLOCK_INIT;
+  DWORD owner_thread_ = 0;
 
  public:
-  xe_fast_mutex();
-  ~xe_fast_mutex();
+  xe_fast_mutex() = default;
+  ~xe_fast_mutex() = default;
 
   void lock();
   void unlock();