Merge c4c7165c4f into 391478b307

src/Cafe/GamePatch.cpp

@@ -245,6 +245,10 @@ uint8 botw_crashFuncMask[] = { 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF
 
 uint8 ffl_floatArrayEndianSwap[] = { 0x7C,0x08,0x02,0xA6,0x94,0x21,0xFF,0xE8,0x93,0xC1,0x00,0x10,0x7C,0x7E,0x1B,0x78,0x93,0xE1,0x00,0x14,0x93,0x81,0x00,0x08,0x7C,0x9F,0x23,0x78,0x93,0xA1,0x00,0x0C,0x90,0x01,0x00,0x1C,0x3B,0xA0,0x00,0x00,0x7C,0x1D,0xF8,0x40,0x40,0x80,0x00,0x20,0x57,0xBC,0x10,0x3A,0x7C,0x3E,0xE4,0x2E };
 
+// Alternative FFL endian swap signature (Mario Kart 8 and possibly other titles)
+// Same algorithm as above but with a different compiler-scheduled prologue ordering
+uint8 ffl_floatArrayEndianSwap_v2[] = { 0x7C,0x08,0x02,0xA6,0x94,0x21,0xFF,0xE8,0x93,0xE1,0x00,0x14,0x93,0xA1,0x00,0x0C,0x7C,0x9F,0x23,0x78,0x93,0x81,0x00,0x08,0x93,0xC1,0x00,0x10,0x90,0x01,0x00,0x1C,0x3B,0xA0,0x00,0x00,0x7C,0x1D,0xF8,0x40,0x7C,0x7E,0x1B,0x78,0x40,0x80,0x00,0x20,0x57,0xBC,0x10,0x3A,0x7C,0x3E,0xE4,0x2E };
+
 uint8 xcx_enterCriticalSectionSignature[] = { 0x94,0x21,0xFF,0xE0,0xBF,0x41,0x00,0x08,0x7C,0x08,0x02,0xA6,0x90,0x01,0x00,0x24,0x7C,0x7E,0x1B,0x78,0x80,0x1E,0x00,0x08,0x2C,0x00,0x00,0x00,0x41,0x82,0x00,0xC0,0x48,0x01,0xD7,0xA1,0x7C,0x7A,0x1B,0x79,0x41,0x82,0x00,0xB4,0x81,0x3E,0x00,0x04,0x7C,0x09,0xD0,0x40,0x40,0x82,0x00,0x2C,0x7D,0x20,0xF0,0x28,0x7C,0x00,0xF0,0x6C };
 uint8 xcx_enterCriticalSectionMask[] =      { 0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0x00,0x00,0x00,0x00,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF };
 
@@ -361,11 +365,23 @@ void GamePatch_scan()
 	hleAddr = hle_locate(ffl_floatArrayEndianSwap, NULL, sizeof(ffl_floatArrayEndianSwap));
 	if (hleAddr)
 	{
-		cemuLog_logDebug(LogType::Force, "HLE: Hook FFL float array endian swap function at 0x{:08x}", hleAddr);
+		cemuLog_logDebug(LogType::Force, "HLE: Hook FFL float array endian swap function (v1) at 0x{:08x}", hleAddr);
 		sint32 functionIndex = hleIndex_h000000003;
 		uint32 opcode = (1 << 26) | (functionIndex); // opcode for HLE: 0x1000 + FunctionIndex
 		memory_write<uint32>(hleAddr, opcode);
 	}
+	else
+	{
+		// Try v2 signature (Mario Kart 8 and possibly other titles with reordered prologue)
+		hleAddr = hle_locate(ffl_floatArrayEndianSwap_v2, NULL, sizeof(ffl_floatArrayEndianSwap_v2));
+		if (hleAddr)
+		{
+			cemuLog_logDebug(LogType::Force, "HLE: Hook FFL float array endian swap function (v2) at 0x{:08x}", hleAddr);
+			sint32 functionIndex = hleIndex_h000000003;
+			uint32 opcode = (1 << 26) | (functionIndex); // opcode for HLE: 0x1000 + FunctionIndex
+			memory_write<uint32>(hleAddr, opcode);
+		}
+	}
 
 	// XCX freeze workaround
 	//hleAddr = hle_locate(xcx_enterCriticalSectionSignature, xcx_enterCriticalSectionMask, sizeof(xcx_enterCriticalSectionSignature));

src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp

@@ -336,6 +336,134 @@ void VulkanRenderer::GetDeviceFeatures()
 	cemuLog_log(LogType::Force, fmt::format("VulkanLimits: UBAlignment {0} nonCoherentAtomSize {1}", prop2.properties.limits.minUniformBufferOffsetAlignment, prop2.properties.limits.nonCoherentAtomSize));
 }
 
+#if BOOST_OS_LINUX
+#include <sys/wait.h>
+#include "resource/IconsFontAwesome5.h"
+
+int BreathOfTheWildChildProcessMain()
+{
+	InitializeGlobalVulkan();
+	struct sigaction sa{.sa_handler = [](int unused){_exit(1);}};
+	int ret = sigaction(SIGABRT, &sa, nullptr);
+
+	freopen("/dev/null", "w", stderr);
+
+	setenv("RADV_DEBUG", "llvm", 1);
+
+	VkInstanceCreateInfo create_info{};
+	create_info.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
+	VkInstance instance = VK_NULL_HANDLE;
+	if (vkCreateInstance(&create_info, nullptr, &instance) != VK_SUCCESS)
+		return 1;
+	InitializeInstanceVulkan(instance);
+
+	// this function will abort() when LLVM is absent
+	uint32_t count = 0;
+	vkEnumeratePhysicalDevices(instance, &count, nullptr);
+
+	vkDestroyInstance(instance, nullptr);
+	return 0;
+}
+
+static void LinuxBreathOfTheWildWorkaround(VkInstance& instance, const VkInstanceCreateInfo* create_info)
+{
+
+	// if the user specified either shader backend, do nothing.
+	// should parse the flag list but there are currently no other flags containing llvm or aco as a substring
+	const char* debugEnvC = getenv("RADV_DEBUG");
+	std::string_view debugEnv = debugEnvC != nullptr ? debugEnvC : "";
+	if (debugEnv.find("aco") != std::string_view::npos || debugEnv.find("llvm") != std::string_view::npos)
+		return;
+
+	uint32_t count = 0;
+	vkEnumeratePhysicalDevices(instance, &count, nullptr);
+
+	std::vector<VkPhysicalDevice> physicalDevices{count};
+	vkEnumeratePhysicalDevices(instance, &count, physicalDevices.data());
+
+	// Find the first AMD device using a RADV driver and store its version
+	int version = 0;
+	for (auto& i : physicalDevices)
+	{
+		VkPhysicalDeviceDriverProperties driverProps{};
+		driverProps.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES;
+		VkPhysicalDeviceProperties2 prop{};
+		prop.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
+		prop.pNext = &driverProps;
+		vkGetPhysicalDeviceProperties2(i, &prop);
+		if (prop.properties.vendorID != 0x1002 || driverProps.driverID != VK_DRIVER_ID_MESA_RADV)
+			continue;
+
+		version = prop.properties.driverVersion;
+		break;
+	}
+
+	if (version == 0)
+		return;
+
+
+	int major = VK_API_VERSION_MAJOR(version);
+	int minor = VK_API_VERSION_MINOR(version);
+	int patch = VK_API_VERSION_PATCH(version);
+
+	// If the driver is unaffected skip the workaround.
+	// affected drivers:
+	// 25.3.0 - 26.0.4
+	if ((major <= 25 && minor < 3) || (major == 26 && (minor > 0 || patch >= 5)) || major > 26)
+		return;
+
+	// check if running with LLVM would crash because mesa is LLVM-less.
+	int childID = fork();
+	if (childID == 0) // inside this if statement runs in child
+	{
+		setenv("CEMU_DETECT_RADV","1", 1);
+		execl("/proc/self/exe", "/proc/self/exe", nullptr);
+		_exit(2); // exec failed so err on the safe side and signal failure
+	}
+
+	int childStatus = 0;
+	waitpid(childID,  &childStatus, 0);
+
+	// if the process didn't exit cleanly or failed to determine LLVM status
+	if (!WIFEXITED(childStatus) || WEXITSTATUS(childStatus) == 2)
+	{
+		cemuLog_log(LogType::Force, "BOTW/RADV workaround not applied because detecting LLVM presence failed unexpectedly");
+		return;
+	}
+
+	if (WEXITSTATUS(childStatus) == 1)
+		cemuLog_log(LogType::Force, "BOTW/RADV workaround not applied because mesa was built without LLVM");
+
+	// only continue if the process exits with code zero, which means it didn't crash
+	if (WEXITSTATUS(childStatus) != 0)
+		return;
+
+	cemuLog_log(LogType::Force, "BOTW/RADV workaround active. Adding \"llvm\" to RADV_DEBUG environment variable");
+	if (debugEnv.empty())
+	{
+		setenv("RADV_DEBUG", "llvm", 1);
+	}
+	else
+	{
+		std::string appendedDebugEnv{debugEnv};
+		appendedDebugEnv.append(",llvm");
+		setenv("RADV_DEBUG", appendedDebugEnv.c_str(), 1);
+	}
+
+	// recreate the vulkan instance to update debug setting
+	vkDestroyInstance(instance, nullptr);
+	VkResult err = vkCreateInstance(create_info, nullptr, &instance);
+	// re-check for errors just in case.
+	if (err != VK_SUCCESS)
+		throw std::runtime_error(fmt::format("Unable to re-create a Vulkan instance after RADV/LLVM workaround: {}", err));
+	InitializeInstanceVulkan(instance);
+
+	LatteOverlay_pushNotification(std::string{(const char*)ICON_FA_EXCLAMATION_TRIANGLE} + "RADV_DEBUG=llvm set automatically to avoid crashing due to a driver bug. If possible update mesa to 26.0.5 or newer", 10'000);
+
+}
+
+#endif
+
 VulkanRenderer::VulkanRenderer()
 {
 	glslang::InitializeProcess();
@@ -395,6 +523,15 @@ VulkanRenderer::VulkanRenderer()
 	if (!InitializeInstanceVulkan(m_instance))
 		throw std::runtime_error("Unable to load instanced Vulkan functions");
 
+	// Workaround for BOTW + RADV. Runes like Magnesis and the camera cause GPU crashes.
+#if BOOST_OS_LINUX
+	uint64 currentTitleId = CafeSystem::GetForegroundTitleId();
+	if (currentTitleId == 0x00050000101c9500 || currentTitleId == 0x00050000101c9400 || currentTitleId == 0x00050000101c9300)
+	{
+		LinuxBreathOfTheWildWorkaround(m_instance, &create_info);
+	}
+#endif
+
 	uint32_t device_count = 0;
 	vkEnumeratePhysicalDevices(m_instance, &device_count, nullptr);
 	if (device_count == 0)

src/Cafe/OS/libs/nn_ac/nn_ac.cpp

@@ -5,6 +5,9 @@
 
 #if BOOST_OS_WINDOWS
 #include <iphlpapi.h>
+#elif BOOST_OS_LINUX
+#include <ifaddrs.h>
+#include <net/if.h>
 #endif
 
 // AC lib (manages internet connection)
@@ -81,6 +84,37 @@ void _GetLocalIPAndSubnetMask(uint32& localIp, uint32& subnetMask)
 	cemuLog_logDebug(LogType::Force, "_GetLocalIPAndSubnetMask(): Failed to find network IP and subnet mask");
 	_GetLocalIPAndSubnetMaskFallback(localIp, subnetMask);
 }
+#elif BOOST_OS_LINUX
+void _GetLocalIPAndSubnetMask(uint32& localIp, uint32& subnetMask)
+{
+	struct ifaddrs *ifaddr;
+	if (getifaddrs(&ifaddr) == -1)
+	{
+		cemuLog_log(LogType::Force, "Failed to acquire local IP and subnet mask");
+		_GetLocalIPAndSubnetMaskFallback(localIp, subnetMask);
+	}
+	stdx::scope_exit _ifa([&]{ freeifaddrs(ifaddr); });
+
+	for (const struct ifaddrs* ifa = ifaddr; ifa != nullptr; ifa = ifa->ifa_next)
+	{
+		if (ifa->ifa_addr == nullptr || ifa->ifa_addr->sa_family != AF_INET)
+			continue;
+
+		if (!(ifa->ifa_flags & IFF_UP) || !(ifa->ifa_flags & IFF_RUNNING))
+			continue;
+
+		if (ifa->ifa_flags & IFF_LOOPBACK)
+			continue;
+
+		const auto* addr_in = reinterpret_cast<struct sockaddr_in*>(ifa->ifa_addr);
+		localIp = ntohl(addr_in->sin_addr.s_addr);
+		const auto* mask_in = reinterpret_cast<struct sockaddr_in*>(ifa->ifa_netmask);
+		subnetMask = ntohl(mask_in->sin_addr.s_addr);
+		return;
+	}
+	cemuLog_logDebug(LogType::Force, "_GetLocalIPAndSubnetMask(): Failed to find network IP and subnet mask");
+	_GetLocalIPAndSubnetMaskFallback(localIp, subnetMask);
+}
 #else
 void _GetLocalIPAndSubnetMask(uint32& localIp, uint32& subnetMask)
 {

src/gui/wxgui/MainWindow.cpp

@@ -621,35 +621,6 @@ bool MainWindow::FileLoad(const fs::path launchPath, wxLaunchGameEvent::INITIATE
             cemuLog_log(LogType::Force, "GameMode has been started.");
         }
     }
-#endif
-
-	// Workaround for BOTW + Mesa. Runes like Magnesis and the camera cause GPU crashes.
-	// Using the LLVM shader compiler prevents crashes which points to an issue with the default ACO compiler.
-	// Either that or Cemu is violating a specification (GLSL?) causing the shaders to be broken after compilation.
-#if BOOST_OS_LINUX
-	uint64 currentTitleId = CafeSystem::GetForegroundTitleId();
-	if (currentTitleId == 0x00050000101c9500 || currentTitleId == 0x00050000101c9400 || currentTitleId == 0x00050000101c9300)
-	{
-		// if the variable is empty set it to llvm, otherwise check if it contains llvm/aco and if not append it
-		if (const char* value; (value = getenv("RADV_DEBUG")) != NULL && strlen(value) != 0)
-		{
-			std::string valueStr{value};
-			// only append ,llvm when llvm or aco are not already passed as flags.
-			// "aco" is not a mesa flag (anymore, it used to be when llvm was default)
-			// but it will provide users with a way to override this workaround by setting RADV_DEBUG=aco
-			// should parse the flag list but there are currently no other flags containing llvm or aco as a substring
-			if (valueStr.find("llvm") == std::string::npos && valueStr.find("aco") == std::string::npos)
-			{
-				valueStr.append(",llvm");
-				setenv("RADV_DEBUG", valueStr.c_str(), 1);
-			}
-		}
-		else
-		{
-			setenv("RADV_DEBUG", "llvm", 1);
-		}
-	}
-
 #endif
 
 	CreateCanvas();

src/main.cpp

@@ -255,8 +255,14 @@ int main(int argc, char* argv[])
 
 #else
 
+int BreathOfTheWildChildProcessMain();
 int main(int argc, char *argv[])
 {
+#if BOOST_OS_LINUX
+	if (getenv("CEMU_DETECT_RADV") != nullptr)
+		return BreathOfTheWildChildProcessMain();
+#endif
+
 #if BOOST_OS_LINUX || BOOST_OS_BSD
     XInitThreads();
 #endif