Memory rework pt 2 (#801)

* Memory: Rework FCRAM management entirely

Disables a lot of functionality... but I didn't want to commit too much to this commit
Also reworks virtual memory management somewhat (but needs more work)

* Accurately handle MemoryState for virtual memory

Previously all non-free blocks were marked as Reserved

* Memory: Consolidate state and permission changes

Can now use a single function to change either state, permissions, or both
Also merge vmem blocks that have the same state and permissions

* Memory: Fix double reset for FCRAM manager

Fix minor bug with permission tracking

* Memory: Implement Protect operation in ControlMemory

* Memory: Implement Unmap in ControlMemory

Also do a sanity check to make sure the memory region is free for linear allocations

* Memory: Make TLS only 0x200 bytes for each thread

Also move TLS to Base region

* RO: Unmap CROs when unloaded

Thanks @noumidev

* Kernel: Return used app memory for Commit ResourceLimit

Not quite correct, but nothing to be done until process management is improved
Also remove the stack limit for CXIs (thanks amogus)

* Kernel: Report used app memory for GetProcessInfo 2

Not really correct, but it should be accurate for applications at least

* Formatting changes

* Initial fastmem support

* PCSX2 fastmem depression

* Move away from PCSX2 fastmem

* Add enum_flag_ops.hpp

* Finally building on Windows

* Almost got a PoC

* Fix arm64 builds

* This somehow works

* This also works...

* Properly fix fastmem

* Add free region manager

* Update boost

* Add ScopeExit

* Comment out asserts on Linux/Mac/Android

* Comment out ASSERT_MSG asserts too

* Fix derp

* Attempt to fix Android

* Disable fastmem on Android

* Fix Android again maybe pt 2

* android pls

* AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

* AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

* Update host_memory.cpp

* Properly reset memory arena on reset

* Proper ashmem code for Android

* more

* Add temporary Android buildjet script for faster prototype builds

* Fix fastmem (again)

* Clean up shared memory

* Remove Android BuildJet runner

* a

* Revert "a"

This reverts commit 5443ad6f2a794c19c9b1a1567ca1c7f58eed78cd.

* Re-add ELF support

* Re-add 3DSX support

* GetSystemInfo, GetProcessInfo: Memory sizes should be in bytes

* Update Boost

* Update metal-cpp

* Fix metal renderer compilation

* Fix fastmem mapping

* Clean up fastmem code

* Fix oopsie again

* Emulator: Reorder struct

* Kernel types: Cleanup

* Cleanup

* More cleanup

* Make invalid mprotects warn instead of panicking

* Add setting for toggling fastmem

* More cleanup

* Properly initialize BSS to zeroes

* Remove unused code

* Formatting

* Cleanup

* Memory/CRO: Workaround for Pokemon XY

* NCSD loader: Fix BSS (again)

* NCSD loader: Fix BSS (again) (again)

* More memory fixes

* Memory: Remove unused code

* FS: Warn on unimplemented functions instead of panic

* Update software_keyboard.cpp

* Libretro: Add fastmem option

* FRD: Stub SaveLocalAccountData

---------

Co-authored-by: PSI-Rockin <PSI-Rockin@users.noreply.github.com>

src/config.cpp

@@ -48,6 +48,7 @@ void EmulatorConfig::load() {
 
 			printAppVersion = toml::find_or<toml::boolean>(general, "PrintAppVersion", true);
 			circlePadProEnabled = toml::find_or<toml::boolean>(general, "EnableCirclePadPro", true);
+			fastmemEnabled = toml::find_or<toml::boolean>(general, "EnableFastmem", enableFastmemDefault);
 			systemLanguage = languageCodeFromString(toml::find_or<std::string>(general, "SystemLanguage", "en"));
 		}
 	}
@@ -180,6 +181,7 @@ void EmulatorConfig::save() {
 	data["General"]["PrintAppVersion"] = printAppVersion;
 	data["General"]["SystemLanguage"] = languageCodeToString(systemLanguage);
 	data["General"]["EnableCirclePadPro"] = circlePadProEnabled;
+	data["General"]["EnableFastmem"] = fastmemEnabled;
 
 	data["Window"]["AppVersionOnWindow"] = windowSettings.showAppVersion;
 	data["Window"]["RememberWindowPosition"] = windowSettings.rememberPosition;

src/core/CPU/cpu_dynarmic.cpp

@@ -6,6 +6,7 @@
 
 CPU::CPU(Memory& mem, Kernel& kernel, Emulator& emu) : mem(mem), emu(emu), scheduler(emu.getScheduler()), env(mem, kernel, emu.getScheduler()) {
 	cp15 = std::make_shared<CP15>();
+	mem.setCPUTicks(getTicksRef());
 
 	Dynarmic::A32::UserConfig config;
 	config.arch_version = Dynarmic::A32::ArchVersion::v6K;
@@ -15,6 +16,12 @@ CPU::CPU(Memory& mem, Kernel& kernel, Emulator& emu) : mem(mem), emu(emu), sched
 	config.global_monitor = &exclusiveMonitor;
 	config.processor_id = 0;
 
+	if (mem.isFastmemEnabled()) {
+		config.fastmem_pointer = u64(mem.getFastmemArenaBase());
+	} else {
+		config.fastmem_pointer = std::nullopt;
+	}
+
 	jit = std::make_unique<Dynarmic::A32::Jit>(config);
 }
 

src/core/applets/software_keyboard.cpp

@@ -55,7 +55,7 @@ Result::HorizonResult SoftwareKeyboardApplet::start(const MemoryBlock* sharedMem
 	}
 	mem.write16(textAddress, 0);  // Write UTF-16 null terminator
 
-	// Temporarily hardcode the pressed button to be the firs tone
+	// Temporarily hardcode the pressed button to be the first one
 	switch (config.numButtonsM1) {
 		case SoftwareKeyboardButtonConfig::SingleButton: config.returnCode = SoftwareKeyboardResult::D0Click; break;
 		case SoftwareKeyboardButtonConfig::DualButton: config.returnCode = SoftwareKeyboardResult::D1Click1; break;

src/core/kernel/fcram.cpp

@@ -0,0 +1,102 @@
+#include "fcram.hpp"
+
+#include "memory.hpp"
+
+void KFcram::Region::reset(u32 start, size_t size) {
+	this->start = start;
+	pages = size >> 12;
+	freePages = pages;
+
+	Block initialBlock(pages, 0);
+	blocks.clear();
+	blocks.push_back(initialBlock);
+}
+
+void KFcram::Region::alloc(std::list<FcramBlock>& out, s32 allocPages, bool linear) {
+	for (auto it = blocks.begin(); it != blocks.end(); it++) {
+		if (it->used) continue;
+
+		// On linear allocations, only a single contiguous block may be used
+		if (it->pages < allocPages && linear) continue;
+
+		// If the current block is bigger than the allocation, split it
+		if (it->pages > allocPages) {
+			Block newBlock(it->pages - allocPages, it->pageOffset + allocPages);
+			it->pages = allocPages;
+			blocks.insert(it, newBlock);
+		}
+
+		// Mark the block as allocated and add it to the output list
+		it->used = true;
+		allocPages -= it->pages;
+		freePages -= it->pages;
+
+		u32 paddr = start + (it->pageOffset << 12);
+		FcramBlock outBlock(paddr, it->pages);
+		out.push_back(outBlock);
+
+		if (allocPages < 1) {
+			return;
+		}
+	}
+
+	// Official kernel panics here
+	Helpers::panic("Failed to allocate FCRAM, not enough guest memory");
+}
+
+u32 KFcram::Region::getUsedCount() { return pages - freePages; }
+u32 KFcram::Region::getFreeCount() { return freePages; }
+
+KFcram::KFcram(Memory& mem) : mem(mem) {}
+
+void KFcram::reset(size_t ramSize, size_t appSize, size_t sysSize, size_t baseSize) {
+	fcram = mem.getFCRAM();
+	refs = std::unique_ptr<u32>(new u32[ramSize >> 12]);
+	std::memset(refs.get(), 0, (ramSize >> 12) * sizeof(u32));
+
+	appRegion.reset(0, appSize);
+	sysRegion.reset(appSize, sysSize);
+	baseRegion.reset(appSize + sysSize, baseSize);
+}
+
+void KFcram::alloc(FcramBlockList& out, s32 pages, FcramRegion region, bool linear) {
+	switch (region) {
+		case FcramRegion::App: appRegion.alloc(out, pages, linear); break;
+		case FcramRegion::Sys: sysRegion.alloc(out, pages, linear); break;
+		case FcramRegion::Base: baseRegion.alloc(out, pages, linear); break;
+		default: Helpers::panic("Invalid FCRAM region chosen for allocation!"); break;
+	}
+
+	incRef(out);
+}
+
+void KFcram::incRef(FcramBlockList& list) {
+	for (auto it = list.begin(); it != list.end(); it++) {
+		for (int i = 0; i < it->pages; i++) {
+			u32 index = (it->paddr >> 12) + i;
+			refs.get()[index]++;
+		}
+	}
+}
+
+void KFcram::decRef(FcramBlockList& list) {
+	for (auto it = list.begin(); it != list.end(); it++) {
+		for (int i = 0; i < it->pages; i++) {
+			u32 index = (it->paddr >> 12) + i;
+			refs.get()[index]--;
+
+			if (!refs.get()[index]) {
+				Helpers::panic("TODO: Freeing FCRAM");
+			}
+		}
+	}
+}
+
+u32 KFcram::getUsedCount(FcramRegion region) {
+	switch (region) {
+		case FcramRegion::App: return appRegion.getUsedCount();
+		case FcramRegion::Sys: return sysRegion.getUsedCount();
+		case FcramRegion::Base: return baseRegion.getUsedCount();
+		default: Helpers::panic("Invalid FCRAM region in getUsedCount!");
+	}
+}

src/core/kernel/idle_thread.cpp

@@ -17,6 +17,8 @@ idle_thread_main:
 	b idle_thread_main
 */
 
+using namespace KernelMemoryTypes;
+
 static constexpr u8 idleThreadCode[] = {
 	0x00, 0x00, 0xA0, 0xE3,  // mov r0, #0
 	0x00, 0x10, 0xA0, 0xE3,  // mov r1, #0
@@ -27,18 +29,16 @@ static constexpr u8 idleThreadCode[] = {
 // Set up an idle thread to run when no thread is able to run
 void Kernel::setupIdleThread() {
 	Thread& t = threads[idleThreadIndex];
-	constexpr u32 codeAddress = 0xBFC00000;
 
-	// Reserve some memory for the idle thread's code. We map this memory to vaddr BFC00000 which is not userland-accessible
+	// Reserve some memory for the idle thread's code. We map this memory to vaddr 3FC00000 which shouldn't be accessed by applications
 	// We only allocate 4KB (1 page) because our idle code is pretty small
-	const u32 fcramIndex = mem.allocateSysMemory(Memory::pageSize);
-	auto vaddr = mem.allocateMemory(codeAddress, fcramIndex, Memory::pageSize, true, true, false, true, false, true);
-	if (!vaddr.has_value() || vaddr.value() != codeAddress) {
+	constexpr u32 codeAddress = 0x3FC00000;
+	if (!mem.allocMemory(codeAddress, 1, FcramRegion::Base, true, true, false, MemoryState::Locked)) {
 		Helpers::panic("Failed to setup idle thread");
 	}
 
 	// Copy idle thread code to the allocated FCRAM
-	std::memcpy(&mem.getFCRAM()[fcramIndex], idleThreadCode, sizeof(idleThreadCode));
+	mem.copyToVaddr(codeAddress, idleThreadCode, sizeof(idleThreadCode));
 
 	t.entrypoint = codeAddress;
 	t.initialSP = 0;

src/core/kernel/kernel.cpp

@@ -6,7 +6,7 @@
 #include "kernel_types.hpp"
 
 Kernel::Kernel(CPU& cpu, Memory& mem, GPU& gpu, const EmulatorConfig& config, LuaManager& lua)
-	: cpu(cpu), regs(cpu.regs()), mem(mem), handleCounter(0), serviceManager(regs, mem, gpu, currentProcess, *this, config, lua) {
+	: cpu(cpu), regs(cpu.regs()), mem(mem), handleCounter(0), serviceManager(regs, mem, gpu, currentProcess, *this, config, lua), fcramManager(mem) {
 	objects.reserve(512);  // Make room for a few objects to avoid further memory allocs later
 	mutexHandles.reserve(8);
 	portHandles.reserve(32);
@@ -180,9 +180,7 @@ void Kernel::reset() {
 }
 
 // Get pointer to thread-local storage
-u32 Kernel::getTLSPointer() {
-	return VirtualAddrs::TLSBase + currentThreadIndex * VirtualAddrs::TLSSize;
-}
+u32 Kernel::getTLSPointer() { return VirtualAddrs::TLSBase + currentThreadIndex * VirtualAddrs::TLSSize; }
 
 // Result CloseHandle(Handle handle)
 void Kernel::svcCloseHandle() {
@@ -271,7 +269,7 @@ void Kernel::getProcessInfo() {
 		// According to 3DBrew: Amount of private (code, data, heap) memory used by the process + total supervisor-mode
 		// stack size + page-rounded size of the external handle table
 		case 2:
-			regs[1] = mem.getUsedUserMem();
+			regs[1] = fcramManager.getUsedCount(FcramRegion::App) * Memory::pageSize;
 			regs[2] = 0;
 			break;
 
@@ -364,7 +362,7 @@ void Kernel::getSystemInfo() {
 			switch (subtype) {
 				// Total used memory size in the APPLICATION memory region
 				case 1:
-					regs[1] = mem.getUsedUserMem();
+					regs[1] = fcramManager.getUsedCount(FcramRegion::App) * Memory::pageSize;
 					regs[2] = 0;
 					break;
 

src/core/kernel/memory_management.cpp

@@ -30,10 +30,10 @@ namespace MemoryPermissions {
 	};
 }
 
+using namespace KernelMemoryTypes;
+
 // Returns whether "value" is aligned to a page boundary (Ie a boundary of 4096 bytes)
-static constexpr bool isAligned(u32 value) {
-	return (value & 0xFFF) == 0;
-}
+static constexpr bool isAligned(u32 value) { return (value & 0xFFF) == 0; }
 
 // Result ControlMemory(u32* outaddr, u32 addr0, u32 addr1, u32 size,
 //						MemoryOperation operation, MemoryPermission permissions)
@@ -44,6 +44,7 @@ void Kernel::controlMemory() {
 	u32 addr0 = regs[1];
 	u32 addr1 = regs[2];
 	u32 size = regs[3];
+	u32 pages = size >> 12;  // Official kernel truncates nonaligned sizes
 	u32 perms = regs[4];
 
 	if (perms == MemoryPermissions::DontCare) {
@@ -61,7 +62,7 @@ void Kernel::controlMemory() {
 		Helpers::panic("ControlMemory: attempted to allocate executable memory");
 	}
 
-	if (!isAligned(addr0) || !isAligned(addr1) || !isAligned(size)) {
+	if (!isAligned(addr0) || !isAligned(addr1)) {
 		Helpers::panic("ControlMemory: Unaligned parameters\nAddr0: %08X\nAddr1: %08X\nSize: %08X", addr0, addr1, size);
 	}
 
@@ -72,22 +73,54 @@ void Kernel::controlMemory() {
 
 	switch (operation & 0xFF) {
 		case Operation::Commit: {
-			std::optional<u32> address = mem.allocateMemory(addr0, 0, size, linear, r, w, x, true);
-			if (!address.has_value()) {
-				Helpers::panic("ControlMemory: Failed to allocate memory");
+			// TODO: base this from the exheader
+			auto region = FcramRegion::App;
+
+			u32 outAddr = 0;
+			if (linear) {
+				if (!mem.allocMemoryLinear(outAddr, addr0, pages, region, r, w, false)) {
+					Helpers::panic("ControlMemory: Failed to allocate linear memory");
+				}
+			} else {
+				if (!mem.allocMemory(addr0, pages, region, r, w, false, MemoryState::Private)) {
+					Helpers::panic("ControlMemory: Failed to allocate memory");
+				}
+
+				outAddr = addr0;
 			}
 
-			regs[1] = address.value();
+			regs[1] = outAddr;
 			break;
 		}
 
-		case Operation::Map: mem.mirrorMapping(addr0, addr1, size); break;
+		case Operation::Map:
+			// Official kernel only allows Private regions to be mapped to Free regions. An Alias or Aliased region cannot be mapped again
+			if (!mem.mapVirtualMemory(
+					addr0, addr1, pages, r, w, false, MemoryState::Free, MemoryState::Private, MemoryState::Alias, MemoryState::Aliased
+				))
+				Helpers::panic("ControlMemory: Failed to map memory");
+			break;
+
+		case Operation::Unmap:
+			// The same as a Map operation, except in reverse
+			if (!mem.mapVirtualMemory(
+					addr0, addr1, pages, false, false, false, MemoryState::Alias, MemoryState::Aliased, MemoryState::Free, MemoryState::Private
+				)) {
+				Helpers::panic("ControlMemory: Failed to unmap memory");
+			}
+			break;
 
 		case Operation::Protect:
-			Helpers::warn(
-				"Ignoring mprotect! Hope nothing goes wrong but if the game accesses invalid memory or crashes then we prolly need to implement "
-				"this\n"
-			);
+			// Official kernel has an internal state bit to indicate that the region's permissions may be changed
+			// But this should account for all cases
+			if (!mem.testMemoryState(addr0, pages, MemoryState::Private) && !mem.testMemoryState(addr0, pages, MemoryState::Alias) &&
+				!mem.testMemoryState(addr0, pages, MemoryState::Aliased) && !mem.testMemoryState(addr0, pages, MemoryState::AliasCode)) {
+				Helpers::warn("Tried to mprotect invalid region!");
+				return;
+			}
+
+			mem.changePermissions(addr0, pages, r, w, false);
+			regs[1] = addr0;
 			break;
 
 		default: Helpers::warn("ControlMemory: unknown operation %X\n", operation); break;
@@ -104,10 +137,11 @@ void Kernel::queryMemory() {
 
 	logSVC("QueryMemory(mem info pointer = %08X, page info pointer = %08X, addr = %08X)\n", memInfo, pageInfo, addr);
 
-	const auto info = mem.queryMemory(addr);
-	regs[0] = Result::Success;
+	KernelMemoryTypes::MemoryInfo info;
+	const auto result = mem.queryMemory(info, addr);
+	regs[0] = result;
 	regs[1] = info.baseAddr;
-	regs[2] = info.size;
+	regs[2] = info.pages * Memory::pageSize;
 	regs[3] = info.perms;
 	regs[4] = info.state;
 	regs[5] = 0;  // page flags

src/core/kernel/resource_limits.cpp

@@ -82,7 +82,9 @@ void Kernel::getResourceLimitCurrentValues() {
 s32 Kernel::getCurrentResourceValue(const KernelObject* limit, u32 resourceName) {
 	const auto data = static_cast<ResourceLimits*>(limit->data);
 	switch (resourceName) {
-		case ResourceType::Commit: return mem.usedUserMemory;
+		// TODO: needs to use the current amount of memory allocated by the process
+		case ResourceType::Commit: return fcramManager.getUsedCount(FcramRegion::App) * Memory::pageSize;
+
 		case ResourceType::Thread: return threadIndices.size();
 		default: Helpers::panic("Attempted to get current value of unknown kernel resource: %d\n", resourceName);
 	}

src/core/loader/3dsx.cpp

@@ -6,6 +6,8 @@
 
 #include "memory.hpp"
 
+using namespace KernelMemoryTypes;
+
 namespace {
 	struct LoadInfo {
 		u32 codeSegSizeAligned;
@@ -53,12 +55,6 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 	// suum of aligned values is always aligned, have an extra RW page for libctru
 	const u32 totalSize = hbInfo.codeSegSizeAligned + hbInfo.rodataSegSizeAligned + hbInfo.dataSegSizeAligned + 4_KB;
 
-	const auto opt = findPaddr(totalSize);
-	if (!opt.has_value()) {
-		Helpers::panic("Failed to find paddr to map 3DSX file's code to");
-		return false;
-	}
-
 	// Map the ROM on the kernel side
 	const u32 textOffset = 0;
 	const u32 rodataOffset = textOffset + hbInfo.codeSegSizeAligned;
@@ -213,7 +209,8 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 		{
 			pst->heapSize = u32(48_MB);
 			pst->linearHeapSize = u32(64_MB);
-		} else */ {
+		} else */
+		{
 			pst.heapSize = u32(24_MB);
 			pst.linearHeapSize = u32(32_MB);
 		}
@@ -221,12 +218,17 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 		std::memcpy(&code[4], &pst, sizeof(pst));
 	}
 
-	const auto paddr = opt.value();
-	std::memcpy(&fcram[paddr], &code[0], totalSize);  // Copy the 3 segments + BSS to FCRAM
+	// Text is R-X
+	allocMemory(textSegAddr, hbInfo.codeSegSizeAligned / Memory::pageSize, FcramRegion::App, true, false, true, MemoryState::Code);
+	copyToVaddr(textSegAddr, &code[textOffset], hbInfo.codeSegSizeAligned);
 
-	allocateMemory(textSegAddr, paddr + textOffset, hbInfo.codeSegSizeAligned, true, true, false, true);           // Text is R-X
-	allocateMemory(rodataSegAddr, paddr + rodataOffset, hbInfo.rodataSegSizeAligned, true, true, false, false);    // Rodata is R--
-	allocateMemory(dataSegAddr, paddr + dataOffset, hbInfo.dataSegSizeAligned + 0x1000, true, true, true, false);  // Data+BSS+Extra is RW-
+	// Rodata is R--
+	allocMemory(rodataSegAddr, hbInfo.rodataSegSizeAligned / Memory::pageSize, FcramRegion::App, true, false, false, MemoryState::Code);
+	copyToVaddr(rodataSegAddr, &code[rodataOffset], hbInfo.rodataSegSizeAligned);
+
+	// Data + BSS + Extra is RW-. We allocate 1 extra page (4KB) which is not initialized to anything.
+	allocMemory(dataSegAddr, (hbInfo.dataSegSizeAligned + 4_KB) / Memory::pageSize, FcramRegion::App, true, true, false, MemoryState::Private);
+	copyToVaddr(dataSegAddr, &code[dataOffset], hbInfo.dataSegSizeAligned);
 
 	return true;
 }

src/core/loader/elf.cpp

@@ -4,6 +4,7 @@
 #include "memory.hpp"
 
 using namespace ELFIO;
+using namespace KernelMemoryTypes;
 
 std::optional<u32> Memory::loadELF(std::ifstream& file) {
 	loadedCXI = std::nullopt;  // ELF files don't have a CXI, so set this to null
@@ -24,6 +25,7 @@ std::optional<u32> Memory::loadELF(std::ifstream& file) {
 	auto segNum = reader.segments.size();
 	printf("Number of segments: %d\n", segNum);
 	printf(" #  Perms       Vaddr           File Size       Mem Size\n");
+
 	for (int i = 0; i < segNum; ++i) {
 		const auto seg = reader.segments[i];
 		const auto flags = seg->get_flags();
@@ -55,12 +57,8 @@ std::optional<u32> Memory::loadELF(std::ifstream& file) {
 			Helpers::warn("Rounding ELF segment size to %08X\n", memorySize);
 		}
 
-		// This should also assert that findPaddr doesn't fail
-		u32 fcramAddr = findPaddr(memorySize).value();
-		std::memcpy(&fcram[fcramAddr], data, fileSize);
-
-		// Allocate the segment on the OS side
-		allocateMemory(vaddr, fcramAddr, memorySize, true, r, w, x);
+		allocMemory(vaddr, memorySize / Memory::pageSize, FcramRegion::App, r, w, x, MemoryState::Code);
+		copyToVaddr(vaddr, data, fileSize);
 	}
 
 	// ELF can't specify a region, make it default to USA

src/core/loader/ncsd.cpp

@@ -3,8 +3,11 @@
 #include <cstring>
 #include <optional>
 
+#include "kernel/fcram.hpp"
 #include "memory.hpp"
 
+using namespace KernelMemoryTypes;
+
 bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 	printf("Text address = %08X, size = %08X\n", cxi.text.address, cxi.text.size);
 	printf("Rodata address = %08X, size = %08X\n", cxi.rodata.address, cxi.rodata.size);
@@ -24,12 +27,6 @@ bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 	// Round up the size of the CXI stack size to a page (4KB) boundary, as the OS can only allocate memory this way
 	u32 stackSize = (cxi.stackSize + pageSize - 1) & -pageSize;
 
-	if (stackSize > 512_KB) {
-		// TODO: Figure out the actual max stack size
-		Helpers::warn("CXI stack size is %08X which seems way too big. Clamping to 512KB", stackSize);
-		stackSize = 512_KB;
-	}
-
 	// Allocate stack
 	if (!allocateMainThreadStack(stackSize)) {
 		// Should be unreachable
@@ -42,40 +39,41 @@ bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 	u32 bssSize = (cxi.bssSize + 0xfff) & ~0xfff;  // Round BSS size up to a page boundary
 	// Total memory to allocate for loading
 	u32 totalSize = (cxi.text.pageCount + cxi.rodata.pageCount + cxi.data.pageCount) * pageSize + bssSize;
-	code.resize(code.size() + bssSize, 0);  // Pad the .code file with zeroes for the BSS segment
 
-	if (code.size() < totalSize) {
+	if (code.size() + bssSize < totalSize) {
 		Helpers::panic("Total code size as reported by the exheader is larger than the .code file");
 		return false;
 	}
 
-	const auto opt = findPaddr(totalSize);
-	if (!opt.has_value()) {
-		Helpers::panic("Failed to find paddr to map CXI file's code to");
-		return false;
-	}
-
-	const auto paddr = opt.value();
-	std::memcpy(&fcram[paddr], &code[0], totalSize);  // Copy the 3 segments + BSS to FCRAM
-
 	// Map the ROM on the kernel side
-	u32 textOffset = 0;
 	u32 textAddr = cxi.text.address;
 	u32 textSize = cxi.text.pageCount * pageSize;
 
-	u32 rodataOffset = textOffset + textSize;
 	u32 rodataAddr = cxi.rodata.address;
 	u32 rodataSize = cxi.rodata.pageCount * pageSize;
 
-	u32 dataOffset = rodataOffset + rodataSize;
 	u32 dataAddr = cxi.data.address;
 	u32 dataSize = cxi.data.pageCount * pageSize + bssSize;  // We're merging the data and BSS segments, as BSS is just pre-initted .data
 
-	allocateMemory(textAddr, paddr + textOffset, textSize, true, true, false, true);         // Text is R-X
-	allocateMemory(rodataAddr, paddr + rodataOffset, rodataSize, true, true, false, false);  // Rodata is R--
-	allocateMemory(dataAddr, paddr + dataOffset, dataSize, true, true, true, false);         // Data+BSS is RW-
+	// TODO: base this off the exheader
+	auto region = FcramRegion::App;
+	u32 bssAddr = dataAddr + (cxi.data.pageCount << 12);
 
-	ncsd.entrypoint = textAddr;
+	allocMemory(textAddr, cxi.text.pageCount, region, true, false, true, MemoryState::Code);
+	allocMemory(rodataAddr, cxi.rodata.pageCount, region, true, false, false, MemoryState::Code);
+	allocMemory(dataAddr, cxi.data.pageCount, region, true, true, false, MemoryState::Private);
+	allocMemory(bssAddr, bssSize >> 12, region, true, true, false, MemoryState::Private);
+
+	// Copy .code file to FCRAM
+	copyToVaddr(textAddr, code.data(), textSize);
+	copyToVaddr(rodataAddr, code.data() + textSize, rodataSize);
+	copyToVaddr(dataAddr, code.data() + textSize + rodataSize, cxi.data.pageCount << 12);
+
+	// Set BSS to zeroes
+	std::vector<u8> bss(bssSize, 0);
+	copyToVaddr(bssAddr, bss.data(), bssSize);
+
+	ncsd.entrypoint = cxi.text.address;
 
 	// Back the IOFile for accessing the ROM, as well as the ROM's CXI partition, in the memory class.
 	CXIFile = ncsd.file;
@@ -85,7 +83,9 @@ bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 
 std::optional<NCSD> Memory::loadNCSD(Crypto::AESEngine& aesEngine, const std::filesystem::path& path) {
 	NCSD ncsd;
-	if (!ncsd.file.open(path, "rb")) return std::nullopt;
+	if (!ncsd.file.open(path, "rb")) {
+		return std::nullopt;
+	}
 
 	u8 magic[4];  // Must be "NCSD"
 	ncsd.file.seek(0x100);

src/core/memory.cpp

@@ -6,6 +6,7 @@
 #include <ctime>
 
 #include "config_mem.hpp"
+#include "kernel/fcram.hpp"
 #include "resource_limits.hpp"
 #include "services/fonts.hpp"
 #include "services/ptm.hpp"
@@ -14,38 +15,43 @@ CMRC_DECLARE(ConsoleFonts);
 
 using namespace KernelMemoryTypes;
 
-Memory::Memory(u64& cpuTicks, const EmulatorConfig& config) : cpuTicks(cpuTicks), config(config) {
-	fcram = new uint8_t[FCRAM_SIZE]();
+Memory::Memory(KFcram& fcramManager, const EmulatorConfig& config) : fcramManager(fcramManager), config(config) {
+	const bool fastmemEnabled = config.fastmemEnabled;
+	arena = new Common::HostMemory(FASTMEM_BACKING_SIZE, FASTMEM_VIRTUAL_SIZE, fastmemEnabled);
 
 	readTable.resize(totalPageCount, 0);
 	writeTable.resize(totalPageCount, 0);
-	memoryInfo.reserve(32);  // Pre-allocate some room for memory allocation info to avoid dynamic allocs
+	paddrTable.resize(totalPageCount, 0);
+
+	fcram = arena->BackingBasePointer() + FASTMEM_FCRAM_OFFSET;
+	// arenaDSPRam = arena->BackingBasePointer() + FASTMEM_DSP_RAM_OFFSET;
+	useFastmem = fastmemEnabled && arena->VirtualBasePointer() != nullptr;
 }
 
 void Memory::reset() {
-	// Unallocate all memory
+	// Mark the entire process address space as free
+	constexpr static int MAX_USER_PAGES = 0x40000000 >> 12;
 	memoryInfo.clear();
-	usedFCRAMPages.reset();
-	usedUserMemory = u32(0_MB);
-	usedSystemMemory = u32(0_MB);
+	memoryInfo.push_back(MemoryInfo(0, MAX_USER_PAGES, 0, KernelMemoryTypes::Free));
+
+	// TODO: remove this, only needed to make the subsequent allocations work for now
+	fcramManager.reset(FCRAM_SIZE, FCRAM_APPLICATION_SIZE, FCRAM_SYSTEM_SIZE, FCRAM_BASE_SIZE);
+
+	if (useFastmem) {
+		// Unmap any mappings when resetting
+		arena->Unmap(0, 4_GB, false);
+	}
 
 	for (u32 i = 0; i < totalPageCount; i++) {
 		readTable[i] = 0;
 		writeTable[i] = 0;
+		paddrTable[i] = 0;
 	}
 
-	// Map (32 * 4) KB of FCRAM before the stack for the TLS of each thread
-	std::optional<u32> tlsBaseOpt = findPaddr(32 * 4_KB);
-	if (!tlsBaseOpt.has_value()) {  // Should be unreachable but still good to have
-		Helpers::panic("Failed to allocate memory for thread-local storage");
-	}
-
-	u32 basePaddrForTLS = tlsBaseOpt.value();
-	for (u32 i = 0; i < appResourceLimits.maxThreads; i++) {
-		u32 vaddr = VirtualAddrs::TLSBase + i * VirtualAddrs::TLSSize;
-		allocateMemory(vaddr, basePaddrForTLS, VirtualAddrs::TLSSize, true);
-		basePaddrForTLS += VirtualAddrs::TLSSize;
-	}
+	// Allocate 512 bytes of TLS for each thread. Since the smallest allocatable unit is 4 KB, that means allocating one page for every 8 threads
+	// Note that TLS is always allocated in the Base region
+	s32 tlsPages = (appResourceLimits.maxThreads + 7) >> 3;
+	allocMemory(VirtualAddrs::TLSBase, tlsPages, FcramRegion::Base, true, true, false, MemoryState::Locked);
 
 	// Initialize shared memory blocks and reserve memory for them
 	for (auto& e : sharedMemBlocks) {
@@ -56,19 +62,23 @@ void Memory::reset() {
 		}
 
 		e.mapped = false;
-		e.paddr = allocateSysMemory(e.size);
+		FcramBlockList memBlock;
+		fcramManager.alloc(memBlock, e.size >> 12, FcramRegion::Sys, false);
+		e.paddr = memBlock.begin()->paddr;
 	}
 
 	// Map DSP RAM as R/W at [0x1FF00000, 0x1FF7FFFF]
-	constexpr u32 dspRamPages = DSP_RAM_SIZE / pageSize;               // Number of DSP RAM pages
-	constexpr u32 initialPage = VirtualAddrs::DSPMemStart / pageSize;  // First page of DSP RAM in the virtual address space
+	constexpr u32 dspRamPages = DSP_RAM_SIZE / pageSize;  // Number of DSP RAM pages
 
-	for (u32 i = 0; i < dspRamPages; i++) {
-		auto pointer = uintptr_t(&dspRam[i * pageSize]);
+	u32 vaddr = VirtualAddrs::DSPMemStart;
+	u32 paddr = PhysicalAddrs::DSP_RAM;
 
-		readTable[i + initialPage] = pointer;
-		writeTable[i + initialPage] = pointer;
-	}
+	Operation op{.newState = MemoryState::Static, .r = true, .w = true, .changeState = true, .changePerms = true};
+	changeMemoryState(vaddr, dspRamPages, op);
+	mapPhysicalMemory(vaddr, paddr, dspRamPages, true, true, false);
+
+	// Allocate RW mapping for DSP RAM
+	// addFastmemView(VirtualAddrs::DSPMemStart, FASTMEM_DSP_RAM_OFFSET, DSP_RAM_SIZE, true, false);
 
 	// Later adjusted based on ROM header when possible
 	region = Regions::USA;
@@ -76,14 +86,9 @@ void Memory::reset() {
 
 bool Memory::allocateMainThreadStack(u32 size) {
 	// Map stack pages as R/W
-	std::optional<u32> basePaddr = findPaddr(size);
-	if (!basePaddr.has_value()) {  // Should also be unreachable but still good to have
-		return false;
-	}
-
+	// TODO: get the region from the exheader
 	const u32 stackBottom = VirtualAddrs::StackTop - size;
-	std::optional<u32> result = allocateMemory(stackBottom, basePaddr.value(), size, true);  // Should never be nullopt
-	return result.has_value();
+	return allocMemory(stackBottom, size >> 12, FcramRegion::App, true, true, false, MemoryState::Locked);
 }
 
 u8 Memory::read8(u32 vaddr) {
@@ -120,7 +125,7 @@ u8 Memory::read8(u32 vaddr) {
 			case ConfigMem::FirmRevision: return firm.revision;
 			case ConfigMem::FirmVersionMinor: return firm.minor;
 			case ConfigMem::FirmVersionMajor: return firm.major;
-			case ConfigMem::WifiLevel: return 0; // No wifi :(
+			case ConfigMem::WifiLevel: return 0;  // No wifi :(
 
 			case ConfigMem::WifiMac:
 			case ConfigMem::WifiMac + 1:
@@ -163,8 +168,8 @@ u32 Memory::read32(u32 vaddr) {
 			case ConfigMem::Datetime0 + 4:
 				return u32(timeSince3DSEpoch() >> 32);  // top 32 bits
 			// Ticks since time was last updated. For now we return the current tick count
-			case ConfigMem::Datetime0 + 8: return u32(cpuTicks);
-			case ConfigMem::Datetime0 + 12: return u32(cpuTicks >> 32);
+			case ConfigMem::Datetime0 + 8: return u32(*cpuTicks);
+			case ConfigMem::Datetime0 + 12: return u32(*cpuTicks >> 32);
 			case ConfigMem::Datetime0 + 16: return 0xFFB0FF0;  // Unknown, set by PTM
 			case ConfigMem::Datetime0 + 20:
 			case ConfigMem::Datetime0 + 24:
@@ -172,11 +177,10 @@ u32 Memory::read32(u32 vaddr) {
 
 			case ConfigMem::AppMemAlloc: return appResourceLimits.maxCommit;
 			case ConfigMem::SyscoreVer: return 2;
-			case 0x1FF81000: return 0;                   // TODO: Figure out what this config mem address does
+			case 0x1FF81000:
+				return 0;  // TODO: Figure out what this config mem address does
 			// Wifi MAC: First 4 bytes of MAC Address
-			case ConfigMem::WifiMac:
-				return (u32(MACAddress[3]) << 24) | (u32(MACAddress[2]) << 16) | (u32(MACAddress[1]) << 8) |
-					   MACAddress[0];
+			case ConfigMem::WifiMac: return (u32(MACAddress[3]) << 24) | (u32(MACAddress[2]) << 16) | (u32(MACAddress[1]) << 8) | MACAddress[0];
 
 			// 3D slider. Float in range 0.0 = off, 1.0 = max.
 			case ConfigMem::SliderState3D: return Helpers::bit_cast<u32, float>(0.0f);
@@ -186,7 +190,7 @@ u32 Memory::read32(u32 vaddr) {
 			default:
 				if (vaddr >= VirtualAddrs::VramStart && vaddr < VirtualAddrs::VramStart + VirtualAddrs::VramSize) {
 					static int shutUpCounter = 0;
-					if (shutUpCounter < 5) { // Stop spamming about VRAM reads after the first 5
+					if (shutUpCounter < 5) {  // Stop spamming about VRAM reads after the first 5
 						shutUpCounter++;
 						Helpers::warn("VRAM read!\n");
 					}
@@ -296,149 +300,254 @@ std::string Memory::readString(u32 address, u32 maxSize) {
 // thanks to the New 3DS having more FCRAM
 u32 Memory::getLinearHeapVaddr() { return (kernelVersion < 0x22C) ? VirtualAddrs::LinearHeapStartOld : VirtualAddrs::LinearHeapStartNew; }
 
-std::optional<u32> Memory::allocateMemory(u32 vaddr, u32 paddr, u32 size, bool linear, bool r, bool w, bool x, bool adjustAddrs, bool isMap) {
-	// Kernel-allocated memory & size must always be aligned to a page boundary
-	// Additionally assert we don't OoM and that we don't try to allocate physical FCRAM past what's available to userland
-	// If we're mapping there's no fear of OoM, because we're not really allocating memory, just binding vaddrs to specific paddrs
-	assert(isAligned(vaddr) && isAligned(paddr) && isAligned(size));
-	assert(size <= FCRAM_APPLICATION_SIZE || isMap);
-	assert(usedUserMemory + size <= FCRAM_APPLICATION_SIZE || isMap);
-	assert(paddr + size <= FCRAM_APPLICATION_SIZE || isMap);
+void Memory::changeMemoryState(u32 vaddr, s32 pages, const Operation& op) {
+	assert(!(vaddr & 0xFFF));
 
-	// Amount of available user FCRAM pages and FCRAM pages to allocate respectively
-	const u32 availablePageCount = (FCRAM_APPLICATION_SIZE - usedUserMemory) / pageSize;
-	const u32 neededPageCount = size / pageSize;
+	if (!op.changePerms && !op.changeState) Helpers::panic("Invalid op passed to changeMemoryState!");
 
-	assert(availablePageCount >= neededPageCount || isMap);
+	bool blockFound = false;
 
-	// If the paddr is 0, that means we need to select our own
-	// TODO: Fix. This method always tries to allocate blocks linearly.
-	// However, if the allocation is non-linear, the panic will trigger when it shouldn't.
-	// Non-linear allocation needs special handling
-	if (paddr == 0 && adjustAddrs) {
-		std::optional<u32> newPaddr = findPaddr(size);
-		if (!newPaddr.has_value()) {
-			Helpers::panic("Failed to find paddr");
+	for (auto it = memoryInfo.begin(); it != memoryInfo.end(); it++) {
+		// Find the block that the memory region is located in
+		u32 blockStart = it->baseAddr;
+		u32 blockEnd = it->end();
+
+		u32 reqStart = vaddr;
+		u32 reqEnd = vaddr + (pages << 12);
+
+		if (!(reqStart >= blockStart && reqEnd <= blockEnd)) continue;
+
+		// Now that the block has been found, fill it with the necessary info
+		auto oldState = it->state;
+		u32 oldPerms = it->perms;
+		it->baseAddr = reqStart;
+		it->pages = pages;
+		if (op.changePerms) it->perms = (op.r ? PERMISSION_R : 0) | (op.w ? PERMISSION_W : 0) | (op.x ? PERMISSION_X : 0);
+		if (op.changeState) it->state = op.newState;
+
+		// If the requested memory region is smaller than the block found, the block must be split
+		if (blockStart < reqStart) {
+			MemoryInfo startBlock(blockStart, (reqStart - blockStart) >> 12, oldPerms, oldState);
+			memoryInfo.insert(it, startBlock);
 		}
 
-		paddr = newPaddr.value();
-		assert(paddr + size <= FCRAM_APPLICATION_SIZE || isMap);
-	}
-
-	// If the vaddr is 0 that means we need to select our own
-	// Depending on whether our mapping should be linear or not we allocate from one of the 2 typical heap spaces
-	// We don't plan on implementing freeing any time soon, so we can pick added userUserMemory to the vaddr base to
-	// Get the full vaddr.
-	// TODO: Fix this
-	if (vaddr == 0 && adjustAddrs) {
-		// Linear memory needs to be allocated in a way where you can easily get the paddr by subtracting the linear heap base
-		// In order to be able to easily send data to hardware like the GPU
-		if (linear) {
-			vaddr = getLinearHeapVaddr() + paddr;
-		} else {
-			vaddr = usedUserMemory + VirtualAddrs::NormalHeapStart;
-		}
-	}
-
-	if (!isMap) {
-		usedUserMemory += size;
-	}
-
-	// Do linear mapping
-	u32 virtualPage = vaddr >> pageShift;
-	u32 physPage = paddr >> pageShift;  // TODO: Special handle when non-linear mapping is necessary
-	for (u32 i = 0; i < neededPageCount; i++) {
-		if (r) {
-			readTable[virtualPage] = uintptr_t(&fcram[physPage * pageSize]);
-		}
-		if (w) {
-			writeTable[virtualPage] = uintptr_t(&fcram[physPage * pageSize]);
+		if (reqEnd < blockEnd) {
+			auto itAfter = std::next(it);
+			MemoryInfo endBlock(reqEnd, (blockEnd - reqEnd) >> 12, oldPerms, oldState);
+			memoryInfo.insert(itAfter, endBlock);
 		}
 
-		// Mark FCRAM page as allocated and go on
-		usedFCRAMPages[physPage] = true;
-		virtualPage++;
-		physPage++;
+		blockFound = true;
+		break;
 	}
 
-	// Back up the info for this allocation in our memoryInfo vector
-	u32 perms = (r ? PERMISSION_R : 0) | (w ? PERMISSION_W : 0) | (x ? PERMISSION_X : 0);
-	memoryInfo.push_back(std::move(MemoryInfo(vaddr, size, perms, KernelMemoryTypes::Reserved)));
+	if (!blockFound) Helpers::panic("Unable to find block in changeMemoryState!");
 
-	return vaddr;
+	// Merge all blocks with the same state and permissions
+	for (auto it = memoryInfo.begin(); it != memoryInfo.end();) {
+		auto next = std::next(it);
+		if (next == memoryInfo.end()) break;
+
+		if (it->state != next->state || it->perms != next->perms) {
+			it++;
+			continue;
+		}
+
+		next->baseAddr = it->baseAddr;
+		next->pages += it->pages;
+		it = memoryInfo.erase(it);
+	}
 }
 
-// Find a paddr which we can use for allocating "size" bytes
-std::optional<u32> Memory::findPaddr(u32 size) {
-	assert(isAligned(size));
-	const u32 neededPages = size / pageSize;
+void Memory::queryPhysicalBlocks(FcramBlockList& outList, u32 vaddr, s32 pages) {
+	s32 srcPages = pages;
+	for (auto& alloc : memoryInfo) {
+		u32 blockStart = alloc.baseAddr;
+		u32 blockEnd = alloc.end();
 
-	// The FCRAM page we're testing to see if it's appropriate to use
-	u32 candidatePage = 0;
-	// The number of linear available pages we could find starting from this candidate page.
-	// If this ends up >= than neededPages then the paddr is good (ie we can use the candidate page as a base address)
-	u32 counter = 0;
+		if (!(vaddr >= blockStart && vaddr < blockEnd)) continue;
 
-	for (u32 i = 0; i < FCRAM_APPLICATION_PAGE_COUNT; i++) {
-		if (usedFCRAMPages[i]) {  // Page is occupied already, go to new candidate
-			candidatePage = i + 1;
-			counter = 0;
-		} else {  // The paddr we're testing has 1 more free page
-			counter++;
-			// Check if there's enough free memory to use this page
-			// We use == instead of >= because some software does 0-byte allocations
-			if (counter >= neededPages) {
-				return candidatePage * pageSize;
+		s32 blockPaddr = paddrTable[vaddr >> 12];
+		s32 blockPages = alloc.pages - ((vaddr - blockStart) >> 12);
+		blockPages = std::min(srcPages, blockPages);
+		FcramBlock physicalBlock(blockPaddr, blockPages);
+		outList.push_back(physicalBlock);
+
+		vaddr += blockPages << 12;
+		srcPages -= blockPages;
+		if (srcPages == 0) break;
+	}
+
+	if (srcPages != 0) Helpers::panic("Unable to find virtual pages to map!");
+}
+
+void Memory::mapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages, bool r, bool w, bool x) {
+	assert(!(vaddr & 0xFFF));
+	assert(!(paddr & 0xFFF));
+
+	// TODO: make this a separate function
+	u8* hostPtr = nullptr;
+	if (paddr < FCRAM_SIZE) {
+		hostPtr = fcram + paddr;  // FIXME: FCRAM doesn't actually start from physical address 0, but from 0x20000000
+
+		if (useFastmem) {
+			addFastmemView(vaddr, FASTMEM_FCRAM_OFFSET + paddr, usize(pages) * pageSize, w);
+		}
+	} else if (paddr >= VirtualAddrs::DSPMemStart && paddr < VirtualAddrs::DSPMemStart + DSP_RAM_SIZE) {
+		hostPtr = dspRam + (paddr - VirtualAddrs::DSPMemStart);
+	}
+
+	for (int i = 0; i < pages; i++) {
+		u32 index = (vaddr >> 12) + i;
+		paddrTable[index] = paddr + (i << 12);
+		if (r)
+			readTable[index] = (uintptr_t)(hostPtr + (i << 12));
+		else
+			readTable[index] = 0;
+
+		if (w)
+			writeTable[index] = (uintptr_t)(hostPtr + (i << 12));
+		else
+			writeTable[index] = 0;
+	}
+}
+
+void Memory::unmapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages) {
+	for (int i = 0; i < pages; i++) {
+		u32 index = (vaddr >> 12) + i;
+		paddrTable[index] = 0;
+		readTable[index] = 0;
+		writeTable[index] = 0;
+	}
+
+	if (useFastmem) {
+		arena->Unmap(vaddr, pages * pageSize, false);
+	}
+}
+
+bool Memory::allocMemory(u32 vaddr, s32 pages, FcramRegion region, bool r, bool w, bool x, MemoryState state) {
+	auto res = testMemoryState(vaddr, pages, MemoryState::Free);
+	if (res.isFailure()) return false;
+
+	FcramBlockList memList;
+	fcramManager.alloc(memList, pages, region, false);
+
+	for (auto it = memList.begin(); it != memList.end(); it++) {
+		Operation op{.newState = state, .r = r, .w = w, .x = x, .changeState = true, .changePerms = true};
+		changeMemoryState(vaddr, it->pages, op);
+		mapPhysicalMemory(vaddr, it->paddr, it->pages, r, w, x);
+		vaddr += it->pages << 12;
+	}
+
+	return true;
+}
+
+bool Memory::allocMemoryLinear(u32& outVaddr, u32 inVaddr, s32 pages, FcramRegion region, bool r, bool w, bool x) {
+	if (inVaddr) Helpers::panic("inVaddr specified for linear allocation!");
+
+	FcramBlockList memList;
+	fcramManager.alloc(memList, pages, region, true);
+
+	u32 paddr = memList.begin()->paddr;
+	u32 vaddr = getLinearHeapVaddr() + paddr;
+	auto res = testMemoryState(vaddr, pages, MemoryState::Free);
+	if (res.isFailure()) Helpers::panic("Unable to map linear allocation (vaddr:%08X pages:%08X)", vaddr, pages);
+
+	Operation op{.newState = MemoryState::Continuous, .r = r, .w = w, .x = x, .changeState = true, .changePerms = true};
+	changeMemoryState(vaddr, pages, op);
+	mapPhysicalMemory(vaddr, paddr, pages, r, w, x);
+
+	outVaddr = vaddr;
+	return true;
+}
+
+bool Memory::mapVirtualMemory(
+	u32 dstVaddr, u32 srcVaddr, s32 pages, bool r, bool w, bool x, MemoryState oldDstState, MemoryState oldSrcState, MemoryState newDstState,
+	MemoryState newSrcState, bool unmapPages
+) {
+	// Check that the regions have the specified state
+	// TODO: check src perms
+	auto res = testMemoryState(srcVaddr, pages, oldSrcState);
+	if (res.isFailure()) return false;
+
+	res = testMemoryState(dstVaddr, pages, oldDstState);
+	if (res.isFailure()) return false;
+
+	// Change the virtual memory state for both regions
+	Operation srcOp{.newState = newSrcState, .changeState = true};
+	changeMemoryState(srcVaddr, pages, srcOp);
+
+	Operation dstOp{.newState = newDstState, .r = r, .w = w, .x = x, .changeState = true, .changePerms = true};
+	changeMemoryState(dstVaddr, pages, dstOp);
+
+	// Get a list of physical blocks in the source region
+	FcramBlockList physicalList;
+	queryPhysicalBlocks(physicalList, srcVaddr, pages);
+
+	// Map or unmap each physical block
+	for (auto& block : physicalList) {
+		if (newDstState == MemoryState::Free) {
+			// TODO: Games with CROs will unmap the CRO yet still continue accessing the address it was mapped to?
+			if (unmapPages) {
+				unmapPhysicalMemory(dstVaddr, block.paddr, block.pages);
 			}
+		} else {
+			mapPhysicalMemory(dstVaddr, block.paddr, block.pages, r, w, x);
 		}
+
+		dstVaddr += block.pages << 12;
 	}
 
-	// Couldn't find any page :(
-	return std::nullopt;
+	return true;
 }
 
-u32 Memory::allocateSysMemory(u32 size) {
-	// Should never be triggered, only here as a sanity check
-	if (!isAligned(size)) {
-		Helpers::panic("Memory::allocateSysMemory: Size is not page aligned (val = %08X)", size);
+void Memory::changePermissions(u32 vaddr, s32 pages, bool r, bool w, bool x) {
+	Operation op{.r = r, .w = w, .x = x, .changePerms = true};
+	changeMemoryState(vaddr, pages, op);
+
+	// Now that permissions have been changed, update the corresponding host tables
+	FcramBlockList physicalList;
+	queryPhysicalBlocks(physicalList, vaddr, pages);
+
+	for (auto& block : physicalList) {
+		mapPhysicalMemory(vaddr, block.paddr, block.pages, r, w, x);
+		vaddr += block.pages;
 	}
-
-	// We use a pretty dumb allocator for OS memory since this is not really accessible to the app and is only used internally
-	// It works by just allocating memory linearly, starting from index 0 of OS memory and going up
-	// This should also be unreachable in practice and exists as a sanity check
-	if (size > remainingSysFCRAM()) {
-		Helpers::panic("Memory::allocateSysMemory: Overflowed OS FCRAM");
-	}
-
-	const u32 pageCount = size / pageSize;                      // Number of pages that will be used up
-	const u32 startIndex = sysFCRAMIndex() + usedSystemMemory;  // Starting FCRAM index
-	const u32 startingPage = startIndex / pageSize;
-
-	for (u32 i = 0; i < pageCount; i++) {
-		if (usedFCRAMPages[startingPage + i])  // Also a theoretically unreachable panic for safety
-			Helpers::panic("Memory::reserveMemory: Trying to reserve already reserved memory");
-		usedFCRAMPages[startingPage + i] = true;
-	}
-
-	usedSystemMemory += size;
-	return startIndex;
 }
 
-// The way I understand how the kernel's QueryMemory is supposed to work is that you give it a vaddr
-// And the kernel looks up the memory allocations it's performed, finds which one it belongs in and returns its info?
-// TODO: Verify this
-MemoryInfo Memory::queryMemory(u32 vaddr) {
+Result::HorizonResult Memory::queryMemory(MemoryInfo& out, u32 vaddr) {
 	// Check each allocation
 	for (auto& alloc : memoryInfo) {
 		// Check if the memory address belongs in this allocation and return the info if so
 		if (vaddr >= alloc.baseAddr && vaddr < alloc.end()) {
-			return alloc;
+			out = alloc;
+			return Result::Success;
 		}
 	}
 
-	// Otherwise, if this vaddr was never allocated
-	// TODO: I think this is meant to return how much memory starting here is free as the size?
-	return MemoryInfo(vaddr, pageSize, 0, KernelMemoryTypes::Free);
+	// Official kernel just returns an error here
+	Helpers::warn("Failed to find block in QueryMemory!");
+	return Result::FailurePlaceholder;
+}
+
+Result::HorizonResult Memory::testMemoryState(u32 vaddr, s32 pages, MemoryState desiredState) {
+	for (auto& alloc : memoryInfo) {
+		// Don't bother checking if we're to the left of the requested region
+		if (vaddr >= alloc.end()) continue;
+		if (alloc.state != desiredState) return Result::FailurePlaceholder;  // TODO: error for state mismatch
+
+		// If the end of this block comes after the end of the requested range with no errors, it's a success
+		if (alloc.end() >= vaddr + (pages << 12)) return Result::Success;
+	}
+
+	// TODO: error for when address is outside of userland
+	return Result::FailurePlaceholder;
+}
+
+void Memory::copyToVaddr(u32 dstVaddr, const u8* srcHost, s32 size) {
+	// TODO: check for noncontiguous allocations
+	u8* dstHost = (u8*)readTable[dstVaddr >> 12] + (dstVaddr & 0xFFF);
+	memcpy(dstHost, srcHost, size);
 }
 
 u8* Memory::mapSharedMemory(Handle handle, u32 vaddr, u32 myPerms, u32 otherPerms) {
@@ -459,13 +568,11 @@ u8* Memory::mapSharedMemory(Handle handle, u32 vaddr, u32 myPerms, u32 otherPerm
 			bool w = myPerms & 0b010;
 			bool x = myPerms & 0b100;
 
-			const auto result = allocateMemory(vaddr, paddr, size, true, r, w, x, false, true);
-			e.mapped = true;
-			if (!result.has_value()) {
-				Helpers::panic("Memory::mapSharedMemory: Failed to map shared memory block");
-				return nullptr;
-			}
+			Operation op{.newState = MemoryState::Shared, .r = r, .w = x, .x = x, .changeState = true, .changePerms = true};
+			changeMemoryState(vaddr, size >> 12, op);
+			mapPhysicalMemory(vaddr, paddr, size >> 12, r, w, x);
 
+			e.mapped = true;
 			return &fcram[paddr];
 		}
 	}
@@ -475,24 +582,6 @@ u8* Memory::mapSharedMemory(Handle handle, u32 vaddr, u32 myPerms, u32 otherPerm
 	return nullptr;
 }
 
-void Memory::mirrorMapping(u32 destAddress, u32 sourceAddress, u32 size) {
-	// Should theoretically be unreachable, only here for safety purposes
-	assert(isAligned(destAddress) && isAligned(sourceAddress) && isAligned(size));
-
-	const u32 pageCount = size / pageSize;  // How many pages we need to mirror
-	for (u32 i = 0; i < pageCount; i++) {
-		// Redo the shift here to "properly" handle wrapping around the address space instead of reading OoB
-		const u32 sourcePage = sourceAddress / pageSize;
-		const u32 destPage = destAddress / pageSize;
-
-		readTable[destPage] = readTable[sourcePage];
-		writeTable[destPage] = writeTable[sourcePage];
-
-		sourceAddress += pageSize;
-		destAddress += pageSize;
-	}
-}
-
 // Get the number of ms since Jan 1 1900
 u64 Memory::timeSince3DSEpoch() {
 	using namespace std::chrono;

src/core/renderer_mtl/pica_to_mtl.cpp

@@ -16,27 +16,27 @@ namespace PICA {
 		 decodeTexelAI8ToRG8,
 		 true,
 		 {
-			 .red = MTL::TextureSwizzleRed,
-			 .green = MTL::TextureSwizzleRed,
-			 .blue = MTL::TextureSwizzleRed,
-			 .alpha = MTL::TextureSwizzleGreen,
+			 MTL::TextureSwizzleRed,
+			 MTL::TextureSwizzleRed,
+			 MTL::TextureSwizzleRed,
+			 MTL::TextureSwizzleGreen,
 		 }},                                                 // IA8
 		{MTL::PixelFormatRG8Unorm, 2, decodeTexelGR8ToRG8},  // RG8
 		{MTL::PixelFormatR8Unorm,
 		 1,
 		 decodeTexelI8ToR8,
 		 true,
-		 {.red = MTL::TextureSwizzleRed, .green = MTL::TextureSwizzleRed, .blue = MTL::TextureSwizzleRed, .alpha = MTL::TextureSwizzleOne}},  // I8
-		{MTL::PixelFormatA8Unorm, 1, decodeTexelA8ToA8},                                                                                      // A8
-		{MTL::PixelFormatABGR4Unorm, 2, decodeTexelAI4ToABGR4},                                                                               // IA4
+		 {MTL::TextureSwizzleRed, MTL::TextureSwizzleRed, MTL::TextureSwizzleRed, MTL::TextureSwizzleOne}},  // I8
+		{MTL::PixelFormatA8Unorm, 1, decodeTexelA8ToA8},                                                     // A8
+		{MTL::PixelFormatABGR4Unorm, 2, decodeTexelAI4ToABGR4},                                              // IA4
 		{MTL::PixelFormatR8Unorm,
 		 1,
 		 decodeTexelI4ToR8,
 		 true,
-		 {.red = MTL::TextureSwizzleRed, .green = MTL::TextureSwizzleRed, .blue = MTL::TextureSwizzleRed, .alpha = MTL::TextureSwizzleOne}},  // I4
-		{MTL::PixelFormatA8Unorm, 1, decodeTexelA4ToA8},                                                                                      // A4
-		{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1ToRGBA8},                                                                              // ETC1
-		{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1A4ToRGBA8},  // ETC1A4
+		 {MTL::TextureSwizzleRed, MTL::TextureSwizzleRed, MTL::TextureSwizzleRed, MTL::TextureSwizzleOne}},  // I4
+		{MTL::PixelFormatA8Unorm, 1, decodeTexelA4ToA8},                                                     // A4
+		{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1ToRGBA8},                                             // ETC1
+		{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1A4ToRGBA8},                                           // ETC1A4
 	};
 
 	void checkForMTLPixelFormatSupport(MTL::Device* device) {
@@ -57,10 +57,10 @@ namespace PICA {
 				decodeTexelAI4ToRG8,
 				true,
 				{
-					.red = MTL::TextureSwizzleRed,
-					.green = MTL::TextureSwizzleRed,
-					.blue = MTL::TextureSwizzleRed,
-					.alpha = MTL::TextureSwizzleGreen,
+					MTL::TextureSwizzleRed,
+					MTL::TextureSwizzleRed,
+					MTL::TextureSwizzleRed,
+					MTL::TextureSwizzleGreen,
 				}
 			};
 		}

src/core/services/frd.cpp

@@ -27,6 +27,7 @@ namespace FRDCommands {
 		GetFriendAttributeFlags = 0x00170042,
 		UpdateGameModeDescription = 0x001D0002,
 
+		SaveLocalAccountData = 0x04050000,
 		UpdateMii = 0x040C0800,
 	};
 }
@@ -61,6 +62,7 @@ void FRDService::handleSyncRequest(u32 messagePointer, FRDService::Type type) {
 			if (type == Type::A) {
 				switch (command) {
 					case FRDCommands::UpdateMii: updateMii(messagePointer); break;
+					case FRDCommands::SaveLocalAccountData: saveLocalAccountData(messagePointer); break;
 					default: Helpers::panic("FRD:A service requested. Command: %08X\n", command); break;
 				}
 			} else {
@@ -265,6 +267,13 @@ void FRDService::logout(u32 messagePointer) {
 	mem.write32(messagePointer + 4, Result::Success);
 }
 
+void FRDService::saveLocalAccountData(u32 messagePointer) {
+	log("FRD::SaveLocalAccountData (stubbed)\n");
+
+	mem.write32(messagePointer, IPC::responseHeader(0x405, 1, 0));
+	mem.write32(messagePointer + 4, Result::Success);
+}
+
 void FRDService::updateMii(u32 messagePointer) {
 	log("FRD::UpdateMii (stubbed)\n");
 

src/core/services/fs.cpp

@@ -194,7 +194,11 @@ void FSService::handleSyncRequest(u32 messagePointer) {
 		case FSCommands::SetThisSaveDataSecureValue: setThisSaveDataSecureValue(messagePointer); break;
 		case FSCommands::AbnegateAccessRight: abnegateAccessRight(messagePointer); break;
 		case FSCommands::TheGameboyVCFunction: theGameboyVCFunction(messagePointer); break;
-		default: Helpers::panic("FS service requested. Command: %08X\n", command);
+
+		default:
+			Helpers::warn("Unimplemented FS service requested. Command: %08X\n", command);
+			mem.write32(messagePointer + 4, Result::Success);
+			break;
 	}
 }
 

src/core/services/ldr_ro.cpp

@@ -23,7 +23,8 @@ namespace CROHeader {
 		NameOffset = 0x084,
 		NextCRO = 0x088,
 		PrevCRO = 0x08C,
-		FixedSize = 0x98,
+		FileSize = 0x090,
+		FixedSize = 0x098,
 		OnUnresolved = 0x0AC,
 		CodeOffset = 0x0B0,
 		DataOffset = 0x0B8,
@@ -146,8 +147,10 @@ static const std::string CRO_MAGIC("CRO0");
 static const std::string CRO_MAGIC_FIXED("FIXD");
 static const std::string CRR_MAGIC("CRR0");
 
+using namespace KernelMemoryTypes;
+
 class CRO {
-	Memory &mem;
+	Memory& mem;
 
 	u32 croPointer;  // Origin address of CRO in RAM
 	u32 oldDataSegmentOffset;
@@ -155,7 +158,7 @@ class CRO {
 	bool isCRO;  // False if CRS
 
   public:
-	CRO(Memory &mem, u32 croPointer, bool isCRO) : mem(mem), croPointer(croPointer), oldDataSegmentOffset(0), isCRO(isCRO) {}
+	CRO(Memory& mem, u32 croPointer, bool isCRO) : mem(mem), croPointer(croPointer), oldDataSegmentOffset(0), isCRO(isCRO) {}
 	~CRO() = default;
 
 	std::string getModuleName() {
@@ -164,25 +167,15 @@ class CRO {
 		return mem.readString(moduleName.offset, moduleName.size);
 	}
 
-	u32 getNextCRO() {
-		return mem.read32(croPointer + CROHeader::NextCRO);
-	}
-	
-	u32 getPrevCRO() {
-		return mem.read32(croPointer + CROHeader::PrevCRO);
-	}
+	u32 getNextCRO() { return mem.read32(croPointer + CROHeader::NextCRO); }
 
-	u32 getFixedSize() {
-		return mem.read32(croPointer + CROHeader::FixedSize);
-	}
+	u32 getPrevCRO() { return mem.read32(croPointer + CROHeader::PrevCRO); }
 
-	void setNextCRO(u32 nextCRO) {
-		mem.write32(croPointer + CROHeader::NextCRO, nextCRO);
-	}
+	u32 getFixedSize() { return mem.read32(croPointer + CROHeader::FixedSize); }
 
-	void setPrevCRO(u32 prevCRO) {
-		mem.write32(croPointer + CROHeader::PrevCRO, prevCRO);
-	}
+	void setNextCRO(u32 nextCRO) { mem.write32(croPointer + CROHeader::NextCRO, nextCRO); }
+	void setPrevCRO(u32 prevCRO) { mem.write32(croPointer + CROHeader::PrevCRO, prevCRO); }
+	u32 getSize() { return mem.read32(croPointer + CROHeader::FileSize); }
 
 	void write32(u32 addr, u32 value) {
 		// Note: some games export symbols to the static module, which doesn't contain any segments.
@@ -228,21 +221,17 @@ class CRO {
 		return entryOffset + offset;
 	}
 
-	u32 getOnUnresolvedAddr() {
-		return getSegmentAddr(mem.read32(croPointer + CROHeader::OnUnresolved));
-	}
+	u32 getOnUnresolvedAddr() { return getSegmentAddr(mem.read32(croPointer + CROHeader::OnUnresolved)); }
 
 	u32 getNamedExportSymbolAddr(const std::string& symbolName) {
 		// Note: The CRO contains a trie for fast symbol lookup. For simplicity,
 		// we won't use it and instead look up the symbol in the named export symbol table
 
 		const u32 exportStringSize = mem.read32(croPointer + CROHeader::ExportStringSize);
-
 		const CROHeaderEntry namedExportTable = getHeaderEntry(CROHeader::NamedExportTableOffset);
 
 		for (u32 namedExport = 0; namedExport < namedExportTable.size; namedExport++) {
 			const u32 nameOffset = mem.read32(namedExportTable.offset + 8 * namedExport + NamedExportTable::NameOffset);
-
 			const std::string exportSymbolName = mem.readString(nameOffset, exportStringSize);
 
 			if (symbolName.compare(exportSymbolName) == 0) {
@@ -437,7 +426,7 @@ class CRO {
 		return true;
 	}
 
-	bool rebaseSegmentTable(u32 dataVaddr, u32 bssVaddr, u32 *oldDataVaddr) {
+	bool rebaseSegmentTable(u32 dataVaddr, u32 bssVaddr, u32* oldDataVaddr) {
 		const CROHeaderEntry segmentTable = getHeaderEntry(CROHeader::SegmentTableOffset);
 
 		for (u32 segment = 0; segment < segmentTable.size; segment++) {
@@ -446,13 +435,16 @@ class CRO {
 			const u32 segmentID = mem.read32(segmentTable.offset + 12 * segment + SegmentTable::ID);
 			switch (segmentID) {
 				case SegmentTable::SegmentID::DATA:
-					*oldDataVaddr = segmentOffset + croPointer; oldDataSegmentOffset = segmentOffset; segmentOffset = dataVaddr; break;
+					*oldDataVaddr = segmentOffset + croPointer;
+					oldDataSegmentOffset = segmentOffset;
+					segmentOffset = dataVaddr;
+					break;
 				case SegmentTable::SegmentID::BSS: segmentOffset = bssVaddr; break;
 				case SegmentTable::SegmentID::TEXT:
 				case SegmentTable::SegmentID::RODATA:
-					if (segmentOffset != 0) segmentOffset += croPointer; break;
-				default:
-					Helpers::panic("Unknown segment ID = %u", segmentID);
+					if (segmentOffset != 0) segmentOffset += croPointer;
+					break;
+				default: Helpers::panic("Unknown segment ID = %u", segmentID);
 			}
 
 			mem.write32(segmentTable.offset + 12 * segment + SegmentTable::Offset, segmentOffset);
@@ -473,9 +465,9 @@ class CRO {
 				case SegmentTable::SegmentID::BSS: segmentOffset = 0; break;
 				case SegmentTable::SegmentID::TEXT:
 				case SegmentTable::SegmentID::RODATA:
-					if (segmentOffset != 0) segmentOffset -= croPointer; break;
-				default:
-					Helpers::panic("Unknown segment ID = %u", segmentID);
+					if (segmentOffset != 0) segmentOffset -= croPointer;
+					break;
+				default: Helpers::panic("Unknown segment ID = %u", segmentID);
 			}
 
 			mem.write32(segmentTable.offset + 12 * segment + SegmentTable::Offset, segmentOffset);
@@ -639,7 +631,9 @@ class CRO {
 			u32 relocationOffset = mem.read32(anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset);
 
 			if (relocationOffset != 0) {
-				mem.write32(anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset, relocationOffset + croPointer);
+				mem.write32(
+					anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset, relocationOffset + croPointer
+				);
 			}
 		}
 
@@ -653,7 +647,9 @@ class CRO {
 			u32 relocationOffset = mem.read32(anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset);
 
 			if (relocationOffset != 0) {
-				mem.write32(anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset, relocationOffset - croPointer);
+				mem.write32(
+					anonymousImportTable.offset + 8 * anonymousImport + AnonymousImportTable::RelocationOffset, relocationOffset - croPointer
+				);
 			}
 		}
 
@@ -673,7 +669,6 @@ class CRO {
 			const u32 addend = mem.read32(relocationPatchTable.offset + 12 * relocationPatch + RelocationPatch::Addend);
 
 			const u32 segmentAddr = getSegmentAddr(segmentOffset);
-
 			const u32 entryID = mem.read32(segmentTable.offset + 12 * (segmentOffset & 0xF) + SegmentTable::ID);
 
 			u32 relocationTarget = segmentAddr;
@@ -1198,9 +1193,7 @@ class CRO {
 	}
 };
 
-void LDRService::reset() {
-	loadedCRS = 0;
-}
+void LDRService::reset() { loadedCRS = 0; }
 
 void LDRService::handleSyncRequest(u32 messagePointer) {
 	const u32 command = mem.read32(messagePointer);
@@ -1245,7 +1238,14 @@ void LDRService::initialize(u32 messagePointer) {
 	}
 
 	// Map CRO to output address
-	mem.mirrorMapping(mapVaddr, crsPointer, size);
+	// TODO: how to handle permissions?
+	bool succeeded = mem.mapVirtualMemory(
+		mapVaddr, crsPointer, size >> 12, true, true, true, MemoryState::Free, MemoryState::Private, MemoryState::Locked, MemoryState::AliasCode
+	);
+
+	if (!succeeded) {
+		Helpers::panic("Failed to map CRS");
+	}
 
 	CRO crs(mem, mapVaddr, false);
 
@@ -1312,7 +1312,9 @@ void LDRService::loadCRO(u32 messagePointer, bool isNew) {
 	const u32 fixLevel = mem.read32(messagePointer + 40);
 	const Handle process = mem.read32(messagePointer + 52);
 
-	log("LDR_RO::LoadCRO (isNew = %d, buffer = %08X, vaddr = %08X, size = %08X, .data vaddr = %08X, .data size = %08X, .bss vaddr = %08X, .bss size = %08X, auto link = %d, fix level = %X, process = %X)\n", isNew, croPointer, mapVaddr, size, dataVaddr, dataSize, bssVaddr, bssSize, autoLink, fixLevel, process);
+	log("LDR_RO::LoadCRO (isNew = %d, buffer = %08X, vaddr = %08X, size = %08X, .data vaddr = %08X, .data size = %08X, .bss vaddr = %08X, .bss size "
+		"= %08X, auto link = %d, fix level = %X, process = %X)\n",
+		isNew, croPointer, mapVaddr, size, dataVaddr, dataSize, bssVaddr, bssSize, autoLink, fixLevel, process);
 
 	// Sanity checks
 	if (size < CRO_HEADER_SIZE) {
@@ -1332,7 +1334,14 @@ void LDRService::loadCRO(u32 messagePointer, bool isNew) {
 	}
 
 	// Map CRO to output address
-	mem.mirrorMapping(mapVaddr, croPointer, size);
+	// TODO: how to handle permissions?
+	bool succeeded = mem.mapVirtualMemory(
+		mapVaddr, croPointer, size >> 12, true, true, true, MemoryState::Free, MemoryState::Private, MemoryState::Locked, MemoryState::AliasCode
+	);
+
+	if (!succeeded) {
+		Helpers::panic("Failed to map CRO");
+	}
 
 	CRO cro(mem, mapVaddr, true);
 
@@ -1392,7 +1401,18 @@ void LDRService::unloadCRO(u32 messagePointer) {
 		Helpers::panic("Failed to unrebase CRO");
 	}
 
+	u32 size = cro.getSize();
+	bool succeeded = mem.mapVirtualMemory(
+		mapVaddr, croPointer, size >> 12, false, false, false, MemoryState::Locked, MemoryState::AliasCode, MemoryState::Free, MemoryState::Private,
+		false
+	);
+
+	if (!succeeded) {
+		Helpers::panic("Failed to unmap CRO");
+	}
+
 	kernel.clearInstructionCacheRange(mapVaddr, cro.getFixedSize());
+
 	mem.write32(messagePointer, IPC::responseHeader(0x5, 1, 0));
 	mem.write32(messagePointer + 4, Result::Success);
 }

src/dynamic_library.cpp

@@ -0,0 +1,96 @@
+// SPDX-FileCopyrightText: 2019 Dolphin Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "dynamic_library.hpp"
+
+#include <fmt/format.h>
+
+#include <string>
+#include <utility>
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <dlfcn.h>
+#endif
+
+namespace Common {
+	DynamicLibrary::DynamicLibrary() = default;
+	DynamicLibrary::DynamicLibrary(const char* filename) { void(open(filename)); }
+	DynamicLibrary::DynamicLibrary(void* handle_) : handle{handle_} {}
+	DynamicLibrary::DynamicLibrary(DynamicLibrary&& rhs) noexcept : handle{std::exchange(rhs.handle, nullptr)} {}
+
+	DynamicLibrary& DynamicLibrary::operator=(DynamicLibrary&& rhs) noexcept {
+		close();
+		handle = std::exchange(rhs.handle, nullptr);
+		return *this;
+	}
+
+	DynamicLibrary::~DynamicLibrary() { close(); }
+
+	std::string DynamicLibrary::getUnprefixedFilename(const char* filename) {
+#if defined(_WIN32)
+		return std::string(filename) + ".dll";
+#elif defined(__APPLE__)
+		return std::string(filename) + ".dylib";
+#else
+		return std::string(filename) + ".so";
+#endif
+	}
+
+	std::string DynamicLibrary::getVersionedFilename(const char* libname, int major, int minor) {
+#if defined(_WIN32)
+		if (major >= 0 && minor >= 0)
+			return fmt::format("{}-{}-{}.dll", libname, major, minor);
+		else if (major >= 0)
+			return fmt::format("{}-{}.dll", libname, major);
+		else
+			return fmt::format("{}.dll", libname);
+#elif defined(__APPLE__)
+		const char* prefix = std::strncmp(libname, "lib", 3) ? "lib" : "";
+		if (major >= 0 && minor >= 0)
+			return fmt::format("{}{}.{}.{}.dylib", prefix, libname, major, minor);
+		else if (major >= 0)
+			return fmt::format("{}{}.{}.dylib", prefix, libname, major);
+		else
+			return fmt::format("{}{}.dylib", prefix, libname);
+#else
+		const char* prefix = std::strncmp(libname, "lib", 3) ? "lib" : "";
+		if (major >= 0 && minor >= 0)
+			return fmt::format("{}{}.so.{}.{}", prefix, libname, major, minor);
+		else if (major >= 0)
+			return fmt::format("{}{}.so.{}", prefix, libname, major);
+		else
+			return fmt::format("{}{}.so", prefix, libname);
+#endif
+	}
+
+	bool DynamicLibrary::open(const char* filename) {
+#ifdef _WIN32
+		handle = reinterpret_cast<void*>(LoadLibraryA(filename));
+#else
+		handle = dlopen(filename, RTLD_NOW);
+#endif
+		return handle != nullptr;
+	}
+
+	void DynamicLibrary::close() {
+		if (!isOpen()) return;
+
+#ifdef _WIN32
+		FreeLibrary(reinterpret_cast<HMODULE>(handle));
+#else
+		dlclose(handle);
+#endif
+		handle = nullptr;
+	}
+
+	void* DynamicLibrary::getSymbolAddress(const char* name) const {
+#ifdef _WIN32
+		return reinterpret_cast<void*>(GetProcAddress(reinterpret_cast<HMODULE>(handle), name));
+#else
+		return reinterpret_cast<void*>(dlsym(handle, name));
+#endif
+	}
+
+}  // namespace Common

src/emulator.cpp

@@ -20,8 +20,8 @@ __declspec(dllexport) DWORD AmdPowerXpressRequestHighPerformance = 1;
 
 Emulator::Emulator()
 	: config(getConfigPath()), kernel(cpu, memory, gpu, config, lua), cpu(memory, kernel, *this), gpu(memory, config),
-	  memory(cpu.getTicksRef(), config), cheats(memory, kernel.getServiceManager().getHID()), audioDevice(config.audioDeviceConfig), lua(*this),
-	  running(false)
+	  memory(kernel.getFcramManager(), config), cheats(memory, kernel.getServiceManager().getHID()), audioDevice(config.audioDeviceConfig),
+	  lua(*this), running(false)
 #ifdef PANDA3DS_ENABLE_HTTP_SERVER
 	  ,
 	  httpServer(this)
@@ -159,20 +159,21 @@ void Emulator::pollScheduler() {
 		scheduler.updateNextTimestamp();
 
 		switch (eventType) {
-			case Scheduler::EventType::VBlank: [[likely]] {
-				// Signal that we've reached the end of a frame
-				frameDone = true;
-				lua.signalEvent(LuaEvent::Frame);
+			case Scheduler::EventType::VBlank:
+				[[likely]] {
+					// Signal that we've reached the end of a frame
+					frameDone = true;
+					lua.signalEvent(LuaEvent::Frame);
 
-				// Send VBlank interrupts
-				ServiceManager& srv = kernel.getServiceManager();
-				srv.sendGPUInterrupt(GPUInterrupt::VBlank0);
-				srv.sendGPUInterrupt(GPUInterrupt::VBlank1);
+					// Send VBlank interrupts
+					ServiceManager& srv = kernel.getServiceManager();
+					srv.sendGPUInterrupt(GPUInterrupt::VBlank0);
+					srv.sendGPUInterrupt(GPUInterrupt::VBlank1);
 
-				// Queue next VBlank event
-				scheduler.addEvent(Scheduler::EventType::VBlank, time + CPU::ticksPerSec / 60);
-				break;
-			}
+					// Queue next VBlank event
+					scheduler.addEvent(Scheduler::EventType::VBlank, time + CPU::ticksPerSec / 60);
+					break;
+				}
 
 			case Scheduler::EventType::UpdateTimers: kernel.pollTimers(); break;
 			case Scheduler::EventType::RunDSP: {
@@ -352,8 +353,7 @@ bool Emulator::loadELF(std::ifstream& file) {
 std::span<u8> Emulator::getSMDH() {
 	switch (romType) {
 		case ROMType::NCSD:
-		case ROMType::CXI:
-			return memory.getCXI()->smdh;
+		case ROMType::CXI: return memory.getCXI()->smdh;
 		default: {
 			return std::span<u8>();
 		}
@@ -385,7 +385,7 @@ static void dumpRomFSNode(const RomFS::RomFSNode& node, const char* romFSBase, c
 
 	for (auto& directory : node.directories) {
 		const auto newPath = path / directory->name;
-		
+
 		// Create the directory for the new folder
 		std::error_code ec;
 		std::filesystem::create_directories(newPath, ec);
@@ -464,7 +464,7 @@ void Emulator::reloadSettings() {
 		loadRenderdoc();
 	}
 
-    gpu.getRenderer()->setHashTextures(config.hashTextures);
+	gpu.getRenderer()->setHashTextures(config.hashTextures);
 
 #ifdef PANDA3DS_ENABLE_DISCORD_RPC
 	// Reload RPC setting if we're compiling with RPC support

src/libretro_core.cpp

@@ -153,6 +153,7 @@ static int fetchVariableRange(std::string key, int min, int max) { return std::c
 
 static void configInit() {
 	static const retro_variable values[] = {
+		{"panda3ds_use_fastmem", EmulatorConfig::enableFastmemDefault ? "Enable fastmem; enabled|disabled" : "Enable fastmem; disabled|enabled"},
 		{"panda3ds_use_shader_jit", EmulatorConfig::shaderJitDefault ? "Enable shader JIT; enabled|disabled" : "Enable shader JIT; disabled|enabled"},
 		{"panda3ds_accelerate_shaders",
 		 EmulatorConfig::accelerateShadersDefault ? "Run 3DS shaders on the GPU; enabled|disabled" : "Run 3DS shaders on the GPU; disabled|enabled"},
@@ -188,9 +189,10 @@ static void configUpdate() {
 	config.rendererType = RendererType::OpenGL;
 	config.vsyncEnabled = fetchVariableBool("panda3ds_use_vsync", true);
 	config.shaderJitEnabled = fetchVariableBool("panda3ds_use_shader_jit", EmulatorConfig::shaderJitDefault);
+	config.fastmemEnabled = fetchVariableBool("panda3ds_use_fastmem", EmulatorConfig::enableFastmemDefault);
+	config.systemLanguage = EmulatorConfig::languageCodeFromString(fetchVariable("panda3ds_system_language", "en"));
 	config.chargerPlugged = fetchVariableBool("panda3ds_use_charger", true);
 	config.batteryPercentage = fetchVariableRange("panda3ds_battery_level", 5, 100);
-	config.systemLanguage = EmulatorConfig::languageCodeFromString(fetchVariable("panda3ds_system_language", "en"));
 
 	config.dspType = Audio::DSPCore::typeFromString(fetchVariable("panda3ds_dsp_emulation", "null"));
 	config.audioEnabled = fetchVariableBool("panda3ds_use_audio", false);

src/panda_qt/config_window.cpp

@@ -172,6 +172,10 @@ ConfigWindow::ConfigWindow(ConfigCallback configCallback, MainWindowCallback win
 	connectCheckbox(circlePadProEnabled, config.circlePadProEnabled);
 	genLayout->addRow(circlePadProEnabled);
 
+	QCheckBox* fastmemEnabled = new QCheckBox(tr("Enable Fastmem"));
+	connectCheckbox(fastmemEnabled, config.fastmemEnabled);
+	genLayout->addRow(fastmemEnabled);
+
 	QCheckBox* discordRpcEnabled = new QCheckBox(tr("Enable Discord RPC"));
 	connectCheckbox(discordRpcEnabled, config.discordRpcEnabled);
 	genLayout->addRow(discordRpcEnabled);