Merge 53e934c7f5 into 948ae5ef4a

CMakeLists.txt

@@ -337,6 +337,7 @@ set(KERNEL_SOURCE_FILES src/core/kernel/kernel.cpp src/core/kernel/resource_limi
                         src/core/kernel/address_arbiter.cpp src/core/kernel/error.cpp
                         src/core/kernel/file_operations.cpp src/core/kernel/directory_operations.cpp
                         src/core/kernel/idle_thread.cpp src/core/kernel/timers.cpp
+                        src/core/kernel/fcram.cpp
 )
 set(SERVICE_SOURCE_FILES src/core/services/service_manager.cpp src/core/services/apt.cpp src/core/services/hid.cpp
                          src/core/services/fs.cpp src/core/services/gsp_gpu.cpp src/core/services/gsp_lcd.cpp
@@ -416,7 +417,7 @@ set(HEADER_FILES include/emulator.hpp include/helpers.hpp include/termcolor.hpp
                  include/fs/archive_twl_sound.hpp include/fs/archive_card_spi.hpp include/services/ns.hpp include/audio/audio_device.hpp
                  include/audio/audio_device_interface.hpp include/audio/libretro_audio_device.hpp include/services/ir/ir_types.hpp
                  include/services/ir/ir_device.hpp include/services/ir/circlepad_pro.hpp include/services/service_intercept.hpp
-                 include/screen_layout.hpp include/services/service_map.hpp include/audio/dsp_binary.hpp
+                 include/screen_layout.hpp include/services/service_map.hpp include/audio/dsp_binary.hpp include/kernel/fcram.hpp
 )
 
 if(IOS)

include/kernel/fcram.hpp

@@ -0,0 +1,47 @@
+#pragma once
+#include <list>
+#include <memory>
+#include "kernel_types.hpp"
+
+class Memory;
+
+using FcramBlockList = std::list<FcramBlock>;
+
+class KFcram {
+	struct Region {
+		struct Block {
+			s32 pages;
+			s32 pageOffs;
+			bool used;
+
+			Block(s32 pages, u32 pageOffs) : pages(pages), pageOffs(pageOffs), used(false) {}
+		};
+
+		std::list<Block> blocks;
+		u32 start;
+		s32 pages;
+		s32 freePages;
+	public:
+		Region() : start(0), pages(0) {}
+		void reset(u32 start, size_t size);
+		void alloc(std::list<FcramBlock>& out, s32 pages, bool linear);
+
+		u32 getUsedCount();
+		u32 getFreeCount();
+	};
+
+	Memory& mem;
+
+	Region appRegion, sysRegion, baseRegion;
+	uint8_t* fcram;
+	std::unique_ptr<u32> refs;
+public:
+	KFcram(Memory& memory);
+	void reset(size_t ramSize, size_t appSize, size_t sysSize, size_t baseSize);
+	void alloc(FcramBlockList& out, s32 pages, FcramRegion region, bool linear);
+
+	void incRef(FcramBlockList& list);
+	void decRef(FcramBlockList& list);
+
+	u32 getUsedCount(FcramRegion region);
+};

include/kernel/kernel.hpp

@@ -7,6 +7,7 @@
 #include <vector>
 
 #include "config.hpp"
+#include "fcram.hpp"
 #include "helpers.hpp"
 #include "kernel_types.hpp"
 #include "logger.hpp"
@@ -25,6 +26,8 @@ class Kernel {
 	CPU& cpu;
 	Memory& mem;
 
+	KFcram fcramManager;
+
 	// The handle number for the next kernel object to be created
 	u32 handleCounter;
 	// A list of our OS threads, the max number of which depends on the resource limit (hardcoded 32 per process on retail it seems).
@@ -247,6 +250,7 @@ class Kernel {
 	}
 
 	ServiceManager& getServiceManager() { return serviceManager; }
+	KFcram& getFcramManager() { return fcramManager; }
 	Scheduler& getScheduler();
 
 	void sendGPUInterrupt(GPUInterrupt type) { serviceManager.sendGPUInterrupt(type); }

include/kernel/kernel_types.hpp

@@ -44,6 +44,12 @@ enum class ProcessorID : s32 {
     New3DSExtra2 = 3
 };
 
+enum class FcramRegion {
+    App = 0x100,
+    Sys = 0x200,
+    Base = 0x300
+};
+
 struct AddressArbiter {};
 
 struct ResourceLimits {
@@ -251,3 +257,10 @@ struct KernelObject {
 		}
 	}
 };
+
+struct FcramBlock {
+    u32 paddr;
+    s32 pages;
+
+    FcramBlock(u32 paddr, s32 pages) : paddr(paddr), pages(pages) {}
+};

include/memory.hpp

@@ -3,6 +3,7 @@
 #include <bitset>
 #include <filesystem>
 #include <fstream>
+#include <list>
 #include <optional>
 #include <vector>
 
@@ -10,8 +11,10 @@
 #include "crypto/aes_engine.hpp"
 #include "handles.hpp"
 #include "helpers.hpp"
-#include "loader/ncsd.hpp"
+#include "kernel/kernel_types.hpp"
 #include "loader/3dsx.hpp"
+#include "loader/ncsd.hpp"
+#include "result/result.hpp"
 #include "services/region_codes.hpp"
 
 namespace PhysicalAddrs {
@@ -44,9 +47,9 @@ namespace VirtualAddrs {
 		LinearHeapStartNew = 0x30000000,
 		LinearHeapEndNew = 0x40000000,
 
-		// Start of TLS for first thread. Next thread's storage will be at TLSBase + 0x1000, and so on
+		// Start of TLS for first thread. Next thread's storage will be at TLSBase + 0x200, and so on
-		TLSBase = 0xFF400000,
+		TLSBase = 0x1FF82000,
-		TLSSize = 0x1000,
+		TLSSize = 0x200,
 
 		VramStart = 0x1F000000,
 		VramSize = 0x00600000,
@@ -77,17 +80,15 @@ namespace KernelMemoryTypes {
 		PERMISSION_X = 1 << 2
 	};
 
-	// I assume this is referring to a single piece of allocated memory? If it's for pages, it makes no sense.
-	// If it's for multiple allocations, it also makes no sense
 	struct MemoryInfo {
-		u32 baseAddr; // Base process virtual address. Used as a paddr in lockedMemoryInfo instead
+		u32 baseAddr;
-		u32 size;      // Of what?
+		u32 pages;
-		u32 perms;     // Is this referring to a single page or?
+		u32 perms;
 		u32 state;
 
-		u32 end() { return baseAddr + size; }
+		u32 end() { return baseAddr + (pages << 12); }
-		MemoryInfo(u32 baseAddr, u32 size, u32 perms, u32 state) : baseAddr(baseAddr), size(size)
+		MemoryInfo() : baseAddr(0), pages(0), perms(0), state(0) {}
-			, perms(perms), state(state) {}
+		MemoryInfo(u32 baseAddr, u32 pages, u32 perms, u32 state) : baseAddr(baseAddr), pages(pages), perms(perms), state(state) {}
 	};
 
 	// Shared memory block for HID, GSP:GPU etc
@@ -99,9 +100,20 @@ namespace KernelMemoryTypes {
 
 		SharedMemoryBlock(u32 paddr, u32 size, u32 handle) : paddr(paddr), size(size), handle(handle), mapped(false) {}
 	};
-}
+}  // namespace KernelMemoryTypes
+
+struct FcramBlock;
+class KFcram;
+enum class FcramRegion;
 
 class Memory {
+	// Used internally by changeMemoryState
+	struct Operation {
+		KernelMemoryTypes::MemoryState newState = KernelMemoryTypes::MemoryState::Free;
+		bool r = false, w = false, x = false;
+		bool changeState = false;
+		bool changePerms = false;
+	};
 	using Handle = HorizonHandle;
 
 	u8* fcram;
@@ -111,28 +123,39 @@ class Memory {
 	u64& cpuTicks;  // Reference to the CPU tick counter
 	using SharedMemoryBlock = KernelMemoryTypes::SharedMemoryBlock;
 
+	// TODO: remove this reference when Peach's excellent page table code is moved to a better home
+	KFcram& fcramManager;
+
 	// Our dynarmic core uses page tables for reads and writes with 4096 byte pages
 	std::vector<uintptr_t> readTable, writeTable;
 
+	// vaddr->paddr translation table
+	std::vector<u32> paddrTable;
+
 	// This tracks our OS' memory allocations
-	std::vector<KernelMemoryTypes::MemoryInfo> memoryInfo;
+	std::list<KernelMemoryTypes::MemoryInfo> memoryInfo;
 
 	std::array<SharedMemoryBlock, 5> sharedMemBlocks = {
-		SharedMemoryBlock(0, 0, KernelHandles::FontSharedMemHandle), // Shared memory for the system font (size is 0 because we read the size from the cmrc filesystem
+		SharedMemoryBlock(
+			0, 0, KernelHandles::FontSharedMemHandle
+		),  // Shared memory for the system font (size is 0 because we read the size from the cmrc filesystem
 		SharedMemoryBlock(0, 0x1000, KernelHandles::GSPSharedMemHandle),          // GSP shared memory
 		SharedMemoryBlock(0, 0x1000, KernelHandles::HIDSharedMemHandle),          // HID shared memory
 		SharedMemoryBlock(0, 0x3000, KernelHandles::CSNDSharedMemHandle),         // CSND shared memory
 		SharedMemoryBlock(0, 0xE7000, KernelHandles::APTCaptureSharedMemHandle),  // APT Capture Buffer memory
 	};
 
-public:
+  public:
 	static constexpr u32 pageShift = 12;
 	static constexpr u32 pageSize = 1 << pageShift;
 	static constexpr u32 pageMask = pageSize - 1;
 	static constexpr u32 totalPageCount = 1 << (32 - pageShift);
 
 	static constexpr u32 FCRAM_SIZE = u32(128_MB);
-	static constexpr u32 FCRAM_APPLICATION_SIZE = u32(80_MB);
+	static constexpr u32 FCRAM_APPLICATION_SIZE = u32(64_MB + 16_MB);
+	static constexpr u32 FCRAM_SYSTEM_SIZE = u32(44_MB - 16_MB);
+	static constexpr u32 FCRAM_BASE_SIZE = u32(20_MB);
+
 	static constexpr u32 FCRAM_PAGE_COUNT = FCRAM_SIZE / pageSize;
 	static constexpr u32 FCRAM_APPLICATION_PAGE_COUNT = FCRAM_APPLICATION_SIZE / pageSize;
 
@@ -140,9 +163,9 @@ public:
 	static constexpr u32 DSP_CODE_MEMORY_OFFSET = u32(0_KB);
 	static constexpr u32 DSP_DATA_MEMORY_OFFSET = u32(256_KB);
 
-private:
+  private:
-	std::bitset<FCRAM_PAGE_COUNT> usedFCRAMPages;
+	// std::bitset<FCRAM_PAGE_COUNT> usedFCRAMPages;
-	std::optional<u32> findPaddr(u32 size);
+	// std::optional<u32> findPaddr(u32 size);
 	u64 timeSince3DSEpoch();
 
 	// https://www.3dbrew.org/wiki/Configuration_Memory#ENVINFO
@@ -167,12 +190,15 @@ private:
 
 	static constexpr std::array<u8, 6> MACAddress = {0x40, 0xF4, 0x07, 0xFF, 0xFF, 0xEE};
 
+	void changeMemoryState(u32 vaddr, s32 pages, const Operation& op);
+	void queryPhysicalBlocks(std::list<FcramBlock>& outList, u32 vaddr, s32 pages);
+	void mapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages, bool r, bool w, bool x);
+	void unmapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages);
+
   public:
 	u16 kernelVersion = 0;
-	u32 usedUserMemory = u32(0_MB); // How much of the APPLICATION FCRAM range is used (allocated to the appcore)
-	u32 usedSystemMemory = u32(0_MB); // Similar for the SYSTEM range (reserved for the syscore)
 
-	Memory(u64& cpuTicks, const EmulatorConfig& config);
+	Memory(KFcram& fcramManager, u64& cpuTicks, const EmulatorConfig& config);
 	void reset();
 	void* getReadPointer(u32 address);
 	void* getWritePointer(u32 address);
@@ -198,22 +224,6 @@ private:
 	u32 getLinearHeapVaddr();
 	u8* getFCRAM() { return fcram; }
 
-	// Total amount of OS-only FCRAM available (Can vary depending on how much FCRAM the app requests via the cart exheader)
-	u32 totalSysFCRAM() {
-		return FCRAM_SIZE - FCRAM_APPLICATION_SIZE;
-	}
-
-	// Amount of OS-only FCRAM currently available
-	u32 remainingSysFCRAM() {
-		return totalSysFCRAM() - usedSystemMemory;
-	}
-
-	// Physical FCRAM index to the start of OS FCRAM
-	// We allocate the first part of physical FCRAM for the application, and the rest to the OS. So the index for the OS = application ram size
-	u32 sysFCRAMIndex() {
-		return FCRAM_APPLICATION_SIZE;
-	}
-
 	enum class BatteryLevel {
 		Empty = 0,
 		AlmostEmpty,
@@ -248,27 +258,19 @@ private:
 	}
 
 	// Returns whether "addr" is aligned to a page (4096 byte) boundary
-	static constexpr bool isAligned(u32 addr) {
+	static constexpr bool isAligned(u32 addr) { return (addr & pageMask) == 0; }
-		return (addr & pageMask) == 0;
-	}
 
-	// Allocate "size" bytes of RAM starting from FCRAM index "paddr" (We pick it ourself if paddr == 0)
+	bool allocMemory(u32 vaddr, s32 pages, FcramRegion region, bool r, bool w, bool x, KernelMemoryTypes::MemoryState state);
-	// And map them to virtual address "vaddr" (We also pick it ourself if vaddr == 0).
+	bool allocMemoryLinear(u32& outVaddr, u32 inVaddr, s32 pages, FcramRegion region, bool r, bool w, bool x);
-	// If the "linear" flag is on, the paddr pages must be adjacent in FCRAM
+	bool mapVirtualMemory(
-	// This function is for interacting with the *user* portion of FCRAM mainly. For OS RAM, we use other internal functions below
+		u32 dstVaddr, u32 srcVaddr, s32 pages, bool r, bool w, bool x, KernelMemoryTypes::MemoryState oldDstState,
-	// r, w, x: Permissions for the allocated memory
+		KernelMemoryTypes::MemoryState oldSrcState, KernelMemoryTypes::MemoryState newDstState, KernelMemoryTypes::MemoryState newSrcState
-	// adjustAddrs: If it's true paddr == 0 or vaddr == 0 tell the allocator to pick its own addresses. Used for eg svc ControlMemory
+	);
-	// isMap: Shows whether this is a reserve operation, that allocates memory and maps it to the addr space, or if it's a map operation,
+	void changePermissions(u32 vaddr, s32 pages, bool r, bool w, bool x);
-	// which just maps memory from paddr to vaddr without hassle. The latter is useful for shared memory mapping, the "map" ControlMemory, op, etc
+	Result::HorizonResult queryMemory(KernelMemoryTypes::MemoryInfo& out, u32 vaddr);
-	// Returns the vaddr the FCRAM was mapped to or nullopt if allocation failed
+	Result::HorizonResult testMemoryState(u32 vaddr, s32 pages, KernelMemoryTypes::MemoryState desiredState);
-	std::optional<u32> allocateMemory(u32 vaddr, u32 paddr, u32 size, bool linear, bool r = true, bool w = true, bool x = true,
-		bool adjustsAddrs = false, bool isMap = false);
-	KernelMemoryTypes::MemoryInfo queryMemory(u32 vaddr);
 
-	// For internal use
+	void copyToVaddr(u32 dstVaddr, const u8* srcHost, s32 size);
-	// Allocates a "size"-sized chunk of system FCRAM and returns the index of physical FCRAM used for the allocation
-	// Used for allocating things like shared memory and the like
-	u32 allocateSysMemory(u32 size);
 
 	// Map a shared memory block to virtual address vaddr with permissions "myPerms"
 	// The kernel has a second permission parameter in MapMemoryBlock but not sure what's used for
@@ -276,10 +278,6 @@ private:
 	// Returns a pointer to the FCRAM block used for the memory if allocation succeeded
 	u8* mapSharedMemory(Handle handle, u32 vaddr, u32 myPerms, u32 otherPerms);
 
-	// Mirrors the page mapping for "size" bytes starting from sourceAddress, to "size" bytes in destAddress
-	// All of the above must be page-aligned.
-	void mirrorMapping(u32 destAddress, u32 sourceAddress, u32 size);
-
 	// Backup of the game's CXI partition info, if any
 	std::optional<NCCH> loadedCXI = std::nullopt;
 	std::optional<HB3DSX> loaded3DSX = std::nullopt;
@@ -291,7 +289,6 @@ private:
 	u8* getDSPMem() { return dspRam; }
 	u8* getDSPDataMem() { return &dspRam[DSP_DATA_MEMORY_OFFSET]; }
 	u8* getDSPCodeMem() { return &dspRam[DSP_CODE_MEMORY_OFFSET]; }
-	u32 getUsedUserMem() { return usedUserMemory; }
 
 	void setVRAM(u8* pointer) { vram = pointer; }
 	void setDSPMem(u8* pointer) { dspRam = pointer; }

src/core/kernel/fcram.cpp

@@ -0,0 +1,103 @@
+#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->pageOffs + 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->pageOffs << 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);
+	constexpr u32 codeAddress = 0x3FC00000;
-	auto vaddr = mem.allocateMemory(codeAddress, fcramIndex, Memory::pageSize, true, true, false, true, false, true);
+	if (!mem.allocMemory(codeAddress, 1, FcramRegion::Base, true, true, false, MemoryState::Locked)) {
-	if (!vaddr.has_value() || vaddr.value() != codeAddress) {
 		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() {
+u32 Kernel::getTLSPointer() { return VirtualAddrs::TLSBase + currentThreadIndex * VirtualAddrs::TLSSize; }
-	return VirtualAddrs::TLSBase + currentThreadIndex * VirtualAddrs::TLSSize;
-}
 
 // Result CloseHandle(Handle handle)
 void Kernel::svcCloseHandle() {
@@ -271,7 +269,8 @@ 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();
+			// FIXME
+			regs[1] = fcramManager.getUsedCount(FcramRegion::App);
 			regs[2] = 0;
 			break;
 
@@ -364,7 +363,7 @@ 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);
 					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) {
+static constexpr bool isAligned(u32 value) { return (value & 0xFFF) == 0; }
-	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 +73,51 @@ void Kernel::controlMemory() {
 
 	switch (operation & 0xFF) {
 		case Operation::Commit: {
-			std::optional<u32> address = mem.allocateMemory(addr0, 0, size, linear, r, w, x, true);
+			// TODO: base this from the exheader
-			if (!address.has_value()) {
+			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");
 				}
 
-			regs[1] = address.value();
+				outAddr = addr0;
+			}
+
+			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(
+			// Official kernel has an internal state bit to indicate that the region's permissions may be changed
-				"Ignoring mprotect! Hope nothing goes wrong but if the game accesses invalid memory or crashes then we prolly need to implement "
+			// But this should account for all cases
-				"this\n"
+			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::panic("Tried to mprotect invalid region!");
+
+			mem.changePermissions(addr0, pages, r, w, false);
+			regs[1] = addr0;
 			break;
 
 		default: Helpers::warn("ControlMemory: unknown operation %X\n", operation); break;
@@ -104,10 +134,11 @@ 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);
+	KernelMemoryTypes::MemoryInfo info;
-	regs[0] = Result::Success;
+	const auto result = mem.queryMemory(info, addr);
+	regs[0] = result;
 	regs[1] = info.baseAddr;
-	regs[2] = info.size;
+	regs[2] = info.pages << 12;
 	regs[3] = info.perms;
 	regs[4] = info.state;
 	regs[5] = 0;  // page flags

src/core/kernel/resource_limits.cpp

@@ -82,7 +82,7 @@ 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;
+		case ResourceType::Commit: return fcramManager.getUsedCount(FcramRegion::App) << 12; // TODO: needs to use the current amount of memory allocated by the process
 		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

@@ -26,7 +26,7 @@ namespace {
 }  // namespace
 
 bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
-	const LoadInfo hbInfo = {
+/*	const LoadInfo hbInfo = {
 		.codeSegSizeAligned = (header.codeSegSize + 0xFFF) & ~0xFFF,
 		.rodataSegSizeAligned = (header.rodataSegSize + 0xFFF) & ~0xFFF,
 		.dataSegSizeAligned = (header.dataSegSize + 0xFFF) & ~0xFFF,
@@ -186,10 +186,10 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 				return false;
 			}
 		}
-	}
+	}*/
 
 	// Detect and fill _prm structure
-	HB3DSX::PrmStruct pst;
+	/*HB3DSX::PrmStruct pst;
 	std::memcpy(&pst, &code[4], sizeof(pst));
 	if (pst.magic[0] == '_' && pst.magic[1] == 'p' && pst.magic[2] == 'r' && pst.magic[3] == 'm') {
 		// if there was any argv to put, it would go there
@@ -205,7 +205,7 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 
 		// RUNFLAG_APTREINIT: Reinitialize APT.
 		// From libctru. Because there's no previously running software here
-		pst.runFlags |= 1 << 1;
+		pst.runFlags |= 1 << 1;*/
 
 		/* s64 dummy;
 		bool isN3DS = svcGetSystemInfo(&dummy, 0x10001, 0) == 0;
@@ -213,7 +213,7 @@ 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);
 		}
@@ -228,7 +228,8 @@ bool Memory::map3DSX(HB3DSX& hb3dsx, const HB3DSX::Header& header) {
 	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-
 
-	return true;
+	return true;*/
+return false;
 }
 
 std::optional<u32> Memory::load3DSX(const std::filesystem::path& path) {

src/core/loader/elf.cpp

@@ -6,7 +6,7 @@
 using namespace ELFIO;
 
 std::optional<u32> Memory::loadELF(std::ifstream& file) {
-	loadedCXI = std::nullopt;  // ELF files don't have a CXI, so set this to null
+	/*    loadedCXI = std::nullopt; // ELF files don't have a CXI, so set this to null
 
 		elfio reader;
 		if (!file.good() || !reader.load(file)) {
@@ -59,11 +59,8 @@ std::optional<u32> Memory::loadELF(std::ifstream& file) {
 			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);
-	}
-
 		// ELF can't specify a region, make it default to USA
 		region = Regions::USA;
-	return static_cast<u32>(reader.get_entry());
+		return static_cast<u32>(reader.get_entry());*/
+	return std::nullopt;
 }

src/core/loader/ncsd.cpp

@@ -4,6 +4,9 @@
 #include <optional>
 
 #include "memory.hpp"
+#include "kernel/fcram.hpp"
+
+using namespace KernelMemoryTypes;
 
 bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 	printf("Text address = %08X, size = %08X\n", cxi.text.address, cxi.text.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
@@ -49,15 +46,6 @@ bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 		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;
@@ -71,11 +59,20 @@ bool Memory::mapCXI(NCSD& ncsd, NCCH& cxi) {
 	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
+	// TODO: base this off the exheader
-	allocateMemory(rodataAddr, paddr + rodataOffset, rodataSize, true, true, false, false);  // Rodata is R--
+	auto region = FcramRegion::App;
-	allocateMemory(dataAddr, paddr + dataOffset, dataSize, true, true, true, false);         // Data+BSS is RW-
 
-	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(dataAddr + (cxi.data.pageCount << 12), 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);
+
+	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;

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 +15,33 @@ CMRC_DECLARE(ConsoleFonts);
 
 using namespace KernelMemoryTypes;
 
-Memory::Memory(u64& cpuTicks, const EmulatorConfig& config) : cpuTicks(cpuTicks), config(config) {
+Memory::Memory(KFcram& fcramManager, u64& cpuTicks, const EmulatorConfig& config) : fcramManager(fcramManager), cpuTicks(cpuTicks), config(config) {
 	fcram = new uint8_t[FCRAM_SIZE]();
 
 	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);
 }
 
 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();
+	memoryInfo.push_back(MemoryInfo(0, MAX_USER_PAGES, 0, KernelMemoryTypes::Free));
-	usedUserMemory = u32(0_MB);
+
-	usedSystemMemory = u32(0_MB);
+	// 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);
 
 	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
+	// 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
-	std::optional<u32> tlsBaseOpt = findPaddr(32 * 4_KB);
+	// Note that TLS is always allocated in the Base region
-	if (!tlsBaseOpt.has_value()) {  // Should be unreachable but still good to have
+	s32 tlsPages = (appResourceLimits.maxThreads + 7) >> 3;
-		Helpers::panic("Failed to allocate memory for thread-local storage");
+	allocMemory(VirtualAddrs::TLSBase, tlsPages, FcramRegion::Base, true, true, false, MemoryState::Locked);
-	}
-
-	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;
-	}
 
 	// Initialize shared memory blocks and reserve memory for them
 	for (auto& e : sharedMemBlocks) {
@@ -56,19 +52,20 @@ 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
 
-	for (u32 i = 0; i < dspRamPages; i++) {
+	u32 vaddr = VirtualAddrs::DSPMemStart;
-		auto pointer = uintptr_t(&dspRam[i * pageSize]);
+	u32 paddr = PhysicalAddrs::DSP_RAM;
 
-		readTable[i + initialPage] = pointer;
+	Operation op{ .newState = MemoryState::Static, .r = true, .w = true, .changeState = true, .changePerms = true };
-		writeTable[i + initialPage] = pointer;
+	changeMemoryState(vaddr, dspRamPages, op);
-	}
+	mapPhysicalMemory(vaddr, paddr, dspRamPages, true, true, false);
 
 	// Later adjusted based on ROM header when possible
 	region = Regions::USA;
@@ -76,14 +73,9 @@ void Memory::reset() {
 
 bool Memory::allocateMainThreadStack(u32 size) {
 	// Map stack pages as R/W
-	std::optional<u32> basePaddr = findPaddr(size);
+	// TODO: get the region from the exheader
-	if (!basePaddr.has_value()) {  // Should also be unreachable but still good to have
-		return false;
-	}
-
 	const u32 stackBottom = VirtualAddrs::StackTop - size;
-	std::optional<u32> result = allocateMemory(stackBottom, basePaddr.value(), size, true);  // Should never be nullopt
+	return allocMemory(stackBottom, size >> 12, FcramRegion::App, true, true, false, MemoryState::Locked);
-	return result.has_value();
 }
 
 u8 Memory::read8(u32 vaddr) {
@@ -296,149 +288,235 @@ 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) {
+void Memory::changeMemoryState(u32 vaddr, s32 pages, const Operation& op) {
-	// Kernel-allocated memory & size must always be aligned to a page boundary
+	assert(!(vaddr & 0xFFF));
-	// 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);
 
-	// Amount of available user FCRAM pages and FCRAM pages to allocate respectively
+	if (!op.changePerms && !op.changeState) Helpers::panic("Invalid op passed to changeMemoryState!");
-	const u32 availablePageCount = (FCRAM_APPLICATION_SIZE - usedUserMemory) / pageSize;
-	const u32 neededPageCount = size / pageSize;
 
-	assert(availablePageCount >= neededPageCount || isMap);
+	bool blockFound = false;
 
-	// If the paddr is 0, that means we need to select our own
+	for (auto it = memoryInfo.begin(); it != memoryInfo.end(); it++) {
-	// TODO: Fix. This method always tries to allocate blocks linearly.
+		// Find the block that the memory region is located in
-	// However, if the allocation is non-linear, the panic will trigger when it shouldn't.
+		u32 blockStart = it->baseAddr;
-	// Non-linear allocation needs special handling
+		u32 blockEnd = it->end();
-	if (paddr == 0 && adjustAddrs) {
+
-		std::optional<u32> newPaddr = findPaddr(size);
+		u32 reqStart = vaddr;
-		if (!newPaddr.has_value()) {
+		u32 reqEnd = vaddr + (pages << 12);
-			Helpers::panic("Failed to find paddr");
+
+		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();
+		if (reqEnd < blockEnd) {
-		assert(paddr + size <= FCRAM_APPLICATION_SIZE || isMap);
+			auto itAfter = std::next(it);
+			MemoryInfo endBlock(reqEnd, (blockEnd - reqEnd) >> 12, oldPerms, oldState);
+			memoryInfo.insert(itAfter, endBlock);
 		}
 
-	// If the vaddr is 0 that means we need to select our own
+		blockFound = true;
-	// Depending on whether our mapping should be linear or not we allocate from one of the 2 typical heap spaces
+		break;
-	// 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) {
+	if (!blockFound) Helpers::panic("Unable to find block in changeMemoryState!");
-		usedUserMemory += size;
+
+	// 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;
 		}
 
-	// Do linear mapping
+		next->baseAddr = it->baseAddr;
-	u32 virtualPage = vaddr >> pageShift;
+		next->pages += it->pages;
-	u32 physPage = paddr >> pageShift;  // TODO: Special handle when non-linear mapping is necessary
+		it = memoryInfo.erase(it);
-	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]);
-		}
-
-		// Mark FCRAM page as allocated and go on
-		usedFCRAMPages[physPage] = true;
-		virtualPage++;
-		physPage++;
-	}
-
-	// 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)));
-
-	return vaddr;
 }
 
-// Find a paddr which we can use for allocating "size" bytes
+void Memory::queryPhysicalBlocks(FcramBlockList& outList, u32 vaddr, s32 pages) {
-std::optional<u32> Memory::findPaddr(u32 size) {
+	s32 srcPages = pages;
-	assert(isAligned(size));
+	for (auto& alloc : memoryInfo) {
-	const u32 neededPages = size / pageSize;
+		u32 blockStart = alloc.baseAddr;
+		u32 blockEnd = alloc.end();
 
-	// The FCRAM page we're testing to see if it's appropriate to use
+		if (!(vaddr >= blockStart && vaddr < blockEnd)) continue;
-	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;
 
-	for (u32 i = 0; i < FCRAM_APPLICATION_PAGE_COUNT; i++) {
+		s32 blockPaddr = paddrTable[vaddr >> 12];
-		if (usedFCRAMPages[i]) {  // Page is occupied already, go to new candidate
+		s32 blockPages = alloc.pages - ((vaddr - blockStart) >> 12);
-			candidatePage = i + 1;
+		blockPages = std::min(srcPages, blockPages);
-			counter = 0;
+		FcramBlock physicalBlock(blockPaddr, blockPages);
-		} else {  // The paddr we're testing has 1 more free page
+		outList.push_back(physicalBlock);
-			counter++;
+
-			// Check if there's enough free memory to use this page
+		vaddr += blockPages << 12;
-			// We use == instead of >= because some software does 0-byte allocations
+		srcPages -= blockPages;
-			if (counter >= neededPages) {
+		if (srcPages == 0) break;
-				return candidatePage * pageSize;
-			}
-		}
 	}
 
-	// Couldn't find any page :(
+	if (srcPages != 0) Helpers::panic("Unable to find virtual pages to map!");
-	return std::nullopt;
 }
 
-u32 Memory::allocateSysMemory(u32 size) {
+void Memory::mapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages, bool r, bool w, bool x) {
-	// Should never be triggered, only here as a sanity check
+	assert(!(vaddr & 0xFFF));
-	if (!isAligned(size)) {
+	assert(!(paddr & 0xFFF));
-		Helpers::panic("Memory::allocateSysMemory: Size is not page aligned (val = %08X)", size);
+
+	// TODO: make this a separate function
+	u8* hostPtr = nullptr;
+	if (paddr < FCRAM_SIZE) {
+		hostPtr = fcram + paddr; // FIXME
+	}
+	else if (paddr >= VirtualAddrs::DSPMemStart && paddr < VirtualAddrs::DSPMemStart + DSP_RAM_SIZE) {
+		hostPtr = dspRam + (paddr - VirtualAddrs::DSPMemStart);
 	}
 
-	// We use a pretty dumb allocator for OS memory since this is not really accessible to the app and is only used internally
+	for (int i = 0; i < pages; i++) {
-	// It works by just allocating memory linearly, starting from index 0 of OS memory and going up
+		u32 index = (vaddr >> 12) + i;
-	// This should also be unreachable in practice and exists as a sanity check
+		paddrTable[index] = paddr + (i << 12);
-	if (size > remainingSysFCRAM()) {
+		if (r) readTable[index] = (uintptr_t)(hostPtr + (i << 12));
-		Helpers::panic("Memory::allocateSysMemory: Overflowed OS FCRAM");
+		else readTable[index] = 0;
+
+		if (w) writeTable[index] = (uintptr_t)(hostPtr + (i << 12));
+		else writeTable[index] = 0;
 	}
-
-	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
+void Memory::unmapPhysicalMemory(u32 vaddr, u32 paddr, s32 pages) {
-// And the kernel looks up the memory allocations it's performed, finds which one it belongs in and returns its info?
+	for (int i = 0; i < pages; i++) {
-// TODO: Verify this
+		u32 index = (vaddr >> 12) + i;
-MemoryInfo Memory::queryMemory(u32 vaddr) {
+		paddrTable[index] = 0;
+		readTable[index] = 0;
+		writeTable[index] = 0;
+	}
+}
+
+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) {
+	// 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) unmapPhysicalMemory(dstVaddr, block.paddr, block.pages);
+		else mapPhysicalMemory(dstVaddr, block.paddr, block.pages, r, w, x);
+		dstVaddr += block.pages << 12;
+	}
+
+	return true;
+}
+
+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;
+	}
+}
+
+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
+	// Official kernel just returns an error here
-	// TODO: I think this is meant to return how much memory starting here is free as the size?
+	Helpers::panic("Failed to find block in QueryMemory!");
-	return MemoryInfo(vaddr, pageSize, 0, KernelMemoryTypes::Free);
+	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 +537,11 @@ 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);
+			Operation op{ .newState = MemoryState::Shared, .r = r, .w = x, .x = x, .changeState = true, .changePerms = true };
-			e.mapped = true;
+			changeMemoryState(vaddr, size >> 12, op);
-			if (!result.has_value()) {
+			mapPhysicalMemory(vaddr, paddr, size >> 12, r, w, x);
-				Helpers::panic("Memory::mapSharedMemory: Failed to map shared memory block");
-				return nullptr;
-			}
 
+			e.mapped = true;
 			return &fcram[paddr];
 		}
 	}
@@ -475,24 +551,6 @@ 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/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,
@@ -146,6 +147,8 @@ 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;
 
@@ -164,25 +167,17 @@ class CRO {
 		return mem.readString(moduleName.offset, moduleName.size);
 	}
 
-	u32 getNextCRO() {
+	u32 getNextCRO() { return mem.read32(croPointer + CROHeader::NextCRO); }
-		return mem.read32(croPointer + CROHeader::NextCRO);
-	}
 
-	u32 getPrevCRO() {
+	u32 getPrevCRO() { return mem.read32(croPointer + CROHeader::PrevCRO); }
-		return mem.read32(croPointer + CROHeader::PrevCRO);
-	}
 
-	u32 getFixedSize() {
+	u32 getFixedSize() { return mem.read32(croPointer + CROHeader::FixedSize); }
-		return mem.read32(croPointer + CROHeader::FixedSize);
-	}
 
-	void setNextCRO(u32 nextCRO) {
+	void setNextCRO(u32 nextCRO) { mem.write32(croPointer + CROHeader::NextCRO, nextCRO); }
-		mem.write32(croPointer + CROHeader::NextCRO, nextCRO);
-	}
 
-	void setPrevCRO(u32 prevCRO) {
+	void setPrevCRO(u32 prevCRO) { mem.write32(croPointer + CROHeader::PrevCRO, 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.
@@ -190,11 +185,11 @@ class CRO {
 		// can't be accessed via mem.write32()
 		auto writePointer = mem.getWritePointer(addr);
 		if (writePointer) {
-			*(u32*)writePointer = value;
+			*(u32 *)writePointer = value;
 		} else {
 			auto readPointer = mem.getReadPointer(addr);
 			if (readPointer) {
-				*(u32*)readPointer = value;
+				*(u32 *)readPointer = value;
 			} else {
 				Helpers::panic("LDR_RO write to invalid address = %X\n", addr);
 			}
@@ -228,11 +223,9 @@ class CRO {
 		return entryOffset + offset;
 	}
 
-	u32 getOnUnresolvedAddr() {
+	u32 getOnUnresolvedAddr() { return getSegmentAddr(mem.read32(croPointer + CROHeader::OnUnresolved)); }
-		return getSegmentAddr(mem.read32(croPointer + CROHeader::OnUnresolved));
-	}
 
-	u32 getNamedExportSymbolAddr(const std::string& symbolName) {
+	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
 
@@ -446,13 +439,16 @@ 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;
+					if (segmentOffset != 0) segmentOffset += croPointer;
-				default:
+					break;
-					Helpers::panic("Unknown segment ID = %u", segmentID);
+				default: Helpers::panic("Unknown segment ID = %u", segmentID);
 			}
 
 			mem.write32(segmentTable.offset + 12 * segment + SegmentTable::Offset, segmentOffset);
@@ -473,9 +469,9 @@ class CRO {
 				case SegmentTable::SegmentID::BSS: segmentOffset = 0; break;
 				case SegmentTable::SegmentID::TEXT:
 				case SegmentTable::SegmentID::RODATA:
-					if (segmentOffset != 0) segmentOffset -= croPointer; break;
+					if (segmentOffset != 0) segmentOffset -= croPointer;
-				default:
+					break;
-					Helpers::panic("Unknown segment ID = %u", segmentID);
+				default: Helpers::panic("Unknown segment ID = %u", segmentID);
 			}
 
 			mem.write32(segmentTable.offset + 12 * segment + SegmentTable::Offset, segmentOffset);
@@ -639,7 +635,9 @@ 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 +651,9 @@ 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
+				);
 			}
 		}
 
@@ -661,7 +661,7 @@ class CRO {
 	}
 
 	bool relocateInternalSymbols(u32 oldDataVaddr) {
-		const u8* header = (u8*)mem.getReadPointer(croPointer);
+		const u8 *header = (u8 *)mem.getReadPointer(croPointer);
 
 		const CROHeaderEntry relocationPatchTable = getHeaderEntry(CROHeader::RelocationPatchTableOffset);
 		const CROHeaderEntry segmentTable = getHeaderEntry(CROHeader::SegmentTableOffset);
@@ -1198,9 +1198,7 @@ class CRO {
 	}
 };
 
-void LDRService::reset() {
+void LDRService::reset() { loadedCRS = 0; }
-	loadedCRS = 0;
-}
 
 void LDRService::handleSyncRequest(u32 messagePointer) {
 	const u32 command = mem.read32(messagePointer);
@@ -1245,7 +1243,13 @@ 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 +1316,9 @@ 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 +1338,13 @@ 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 +1404,17 @@ 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
+	);
+
+	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/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),
+	  memory(kernel.getFcramManager(), cpu.getTicksRef(), config), cheats(memory, kernel.getServiceManager().getHID()),
-	  running(false)
+	  audioDevice(config.audioDeviceConfig), lua(*this), running(false)
 #ifdef PANDA3DS_ENABLE_HTTP_SERVER
 	  ,
 	  httpServer(this)
@@ -159,7 +159,8 @@ void Emulator::pollScheduler() {
 		scheduler.updateNextTimestamp();
 
 		switch (eventType) {
-			case Scheduler::EventType::VBlank: [[likely]] {
+			case Scheduler::EventType::VBlank:
+				[[likely]] {
 					// Signal that we've reached the end of a frame
 					frameDone = true;
 					lua.signalEvent(LuaEvent::Frame);
@@ -352,8 +353,7 @@ bool Emulator::loadELF(std::ifstream& file) {
 std::span<u8> Emulator::getSMDH() {
 	switch (romType) {
 		case ROMType::NCSD:
-		case ROMType::CXI:
+		case ROMType::CXI: return memory.getCXI()->smdh;
-			return memory.getCXI()->smdh;
 		default: {
 			return std::span<u8>();
 		}