Fix vm::lock_range wrong check

Minor header refactoring.

Utilities/Thread.cpp

@@ -74,6 +74,7 @@
 
 #include "sync.h"
 #include "util/logs.hpp"
+#include "Emu/Memory/vm_locking.h"
 
 LOG_CHANNEL(sig_log, "SIG");
 LOG_CHANNEL(sys_log, "SYS");

rpcs3/Emu/Cell/lv2/lv2.cpp

@@ -1,6 +1,7 @@
 #include "stdafx.h"
 #include "Emu/System.h"
 #include "Emu/Memory/vm_ptr.h"
+#include "Emu/Memory/vm_locking.h"
 
 #include "Emu/Cell/PPUFunction.h"
 #include "Emu/Cell/ErrorCodes.h"
@@ -1114,6 +1115,26 @@ DECLARE(lv2_obj::g_waiting);
 
 thread_local DECLARE(lv2_obj::g_to_awake);
 
+void lv2_obj::sleep(cpu_thread& cpu, const u64 timeout)
+{
+	vm::temporary_unlock(cpu);
+	std::lock_guard{g_mutex}, sleep_unlocked(cpu, timeout);
+	g_to_awake.clear();
+}
+
+bool lv2_obj::awake(cpu_thread* const thread, s32 prio)
+{
+	vm::temporary_unlock();
+	std::lock_guard lock(g_mutex);
+	return awake_unlocked(thread, prio);
+}
+
+bool lv2_obj::yield(cpu_thread& thread)
+{
+	vm::temporary_unlock(thread);
+	return awake(&thread, yield_cmd);
+}
+
 void lv2_obj::sleep_unlocked(cpu_thread& thread, u64 timeout)
 {
 	const u64 start_time = get_guest_system_time();

rpcs3/Emu/Cell/lv2/sys_ppu_thread.cpp

@@ -6,6 +6,7 @@
 #include "Emu/Cell/ErrorCodes.h"
 #include "Emu/Cell/PPUThread.h"
 #include "Emu/Cell/PPUCallback.h"
+#include "Emu/Memory/vm_locking.h"
 #include "sys_event.h"
 #include "sys_process.h"
 #include "sys_mmapper.h"

rpcs3/Emu/Cell/lv2/sys_rsx.cpp

@@ -3,6 +3,7 @@
 
 #include "Emu/Cell/PPUModule.h"
 #include "Emu/Cell/ErrorCodes.h"
+#include "Emu/Memory/vm_locking.h"
 #include "Emu/RSX/RSXThread.h"
 #include "sys_event.h"
 

rpcs3/Emu/Cell/lv2/sys_sync.h

@@ -4,7 +4,6 @@
 #include "Utilities/sema.h"
 #include "Utilities/cond.h"
 
-#include "Emu/Memory/vm_locking.h"
 #include "Emu/CPU/CPUThread.h"
 #include "Emu/Cell/ErrorCodes.h"
 #include "Emu/IdManager.h"
@@ -156,26 +155,12 @@ private:
 	static bool awake_unlocked(cpu_thread*, s32 prio = enqueue_cmd);
 
 public:
-	static void sleep(cpu_thread& cpu, const u64 timeout = 0)
-	{
-		vm::temporary_unlock(cpu);
-		std::lock_guard{g_mutex}, sleep_unlocked(cpu, timeout);
-		g_to_awake.clear();
-	}
+	static void sleep(cpu_thread& cpu, const u64 timeout = 0);
 
-	static inline bool awake(cpu_thread* const thread, s32 prio = enqueue_cmd)
-	{
-		vm::temporary_unlock();
-		std::lock_guard lock(g_mutex);
-		return awake_unlocked(thread, prio);
-	}
+	static bool awake(cpu_thread* const thread, s32 prio = enqueue_cmd);
 
 	// Returns true on successful context switch, false otherwise
-	static bool yield(cpu_thread& thread)
-	{
-		vm::temporary_unlock(thread);
-		return awake(&thread, yield_cmd);
-	}
+	static bool yield(cpu_thread& thread);
 
 	static void set_priority(cpu_thread& thread, s32 prio)
 	{

rpcs3/Emu/Memory/vm.cpp

@@ -166,29 +166,29 @@ namespace vm
 		for (u64 i = 0;; i++)
 		{
 			const u64 lock_val = g_range_lock.load();
+			const u64 is_shared = g_shareable[begin >> 16].load();
 			const u64 lock_addr = static_cast<u32>(lock_val); // -> u64
 			const u32 lock_size = static_cast<u32>(lock_val >> 35);
 
 			u64 addr = begin;
 
-			// See range_lock()
-			if (g_shareable[begin >> 16] | (((lock_val >> 32) & (range_full_mask >> 32)) ^ (range_locked >> 32)))
+			if (is_shared)
 			{
 				addr = addr & 0xffff;
 			}
 
-			if (addr + size <= lock_addr || addr >= lock_addr + lock_size) [[likely]]
+			if ((lock_val & range_full_mask) != range_locked || addr + size <= lock_addr || addr >= lock_addr + lock_size) [[likely]]
 			{
 				range_lock->store(begin | (u64{size} << 32));
 
 				const u64 new_lock_val = g_range_lock.load();
 
-				if (!(new_lock_val | (new_lock_val != lock_val))) [[likely]]
+				if (!new_lock_val || new_lock_val == lock_val) [[likely]]
 				{
 					break;
 				}
 
-				range_lock->release(0);
+				range_lock->store(0);
 			}
 
 			std::shared_lock lock(g_mutex, std::try_to_lock);

rpcs3/Emu/Memory/vm_locking.h

@@ -43,6 +43,10 @@ namespace vm
 	FORCE_INLINE void range_lock(atomic_t<u64, 64>* range_lock, u32 begin, u32 size)
 	{
 		const u64 lock_val = g_range_lock.load();
+		#ifndef _MSC_VER
+		__asm__(""); // Tiny barrier
+		#endif
+		const u64 is_shared = g_shareable[begin >> 16].load();
 		const u64 lock_addr = static_cast<u32>(lock_val); // -> u64
 		const u32 lock_size = static_cast<u32>(lock_val >> 35);
 
@@ -50,24 +54,25 @@ namespace vm
 
 		// Optimization: if range_locked is not used, the addr check will always pass
 		// Otherwise, g_shareable is unchanged and its value is reliable to read
-		if (g_shareable[begin >> 16] | (((lock_val >> 32) & (range_full_mask >> 32)) ^ (range_locked >> 32)))
+		if (is_shared)
 		{
 			addr = addr & 0xffff;
 		}
 
-		if (addr + size <= lock_addr || addr >= lock_addr + lock_size) [[likely]]
+		if (addr + size <= lock_addr || addr >= lock_addr + lock_size || ((lock_val >> 32) ^ (range_locked >> 32)) & (range_full_mask >> 32)) [[likely]]
 		{
-			// Optimistic locking
+			// Optimistic locking.
+			// Note that we store the range we will be accessing, without any clamping.
 			range_lock->store(begin | (u64{size} << 32));
 
 			const u64 new_lock_val = g_range_lock.load();
 
-			if (!(new_lock_val | (new_lock_val != lock_val))) [[likely]]
+			if (!new_lock_val || new_lock_val == lock_val) [[likely]]
 			{
 				return;
 			}
 
-			range_lock->release(0);
+			range_lock->store(0);
 		}
 
 		// Fallback to slow path

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -4,6 +4,7 @@
 #include "GLGSRender.h"
 #include "GLCompute.h"
 #include "GLVertexProgram.h"
+#include "Emu/Memory/vm_locking.h"
 
 #define DUMP_VERTEX_DATA 0
 

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -6,6 +6,7 @@
 #include "VKRenderPass.h"
 #include "VKResourceManager.h"
 #include "VKCommandStream.h"
+#include "Emu/Memory/vm_locking.h"
 
 namespace vk
 {