Map pages properly maybe

include/memory.hpp

@@ -8,8 +8,12 @@
 #include "helpers.hpp"
 #include "handles.hpp"
 #include "loader/ncsd.hpp"
+#include "MemArena/memory_arena.hpp"
 #include "services/shared_font.hpp"
 
+// Comment this out to disable fastmem
+#define PANDA3DS_HARDWARE_FASTMEM
+
 namespace PhysicalAddrs {
 	enum : u32 {
 		VRAM = 0x18000000,
@@ -100,7 +104,7 @@ class Memory {
 	u64& cpuTicks; // Reference to the CPU tick counter
 	using SharedMemoryBlock = KernelMemoryTypes::SharedMemoryBlock;
 
-	// Our dynarmic core uses page tables for reads and writes with 4096 byte pages
+	// Our memory core uses page tables for reads and writes with 4096 byte pages
 	std::vector<uintptr_t> readTable, writeTable;
 
 	// This tracks our OS' memory allocations
@@ -128,6 +132,36 @@ public:
 	static constexpr u32 DSP_DATA_MEMORY_OFFSET = 256_KB;
 
 private:
+	// We also use MMU-accelerated fastmem for fast memory emulation
+	// This means that we've got a 4GB memory arena which is organized the same way as the emulated 3DS' memory map
+	// And we can access this directly instead of calling the memory read/write functions, which would be slower
+	// Regions that are not mapped or can't be accelerated this way will segfault, and the caller (eg dynarmic), will
+	// handle this segfault and call the Slower memory read/write functions
+	bool useFastmem = false;
+	u8* fastmemArenaBase = nullptr;
+	static constexpr size_t FASTMEM_FCRAM_OFFSET = 0;  // Offset of FCRAM in the fastmem arena
+	static constexpr size_t FASTMEM_DSP_RAM_OFFSET = FASTMEM_FCRAM_OFFSET + FCRAM_SIZE;  // Offset of DSP RAM
+
+#ifdef PANDA3DS_HARDWARE_FASTMEM
+	Common::MemoryArena arena;
+	std::vector<Common::MemoryArena::View> fastmemViews;
+
+	void resetFastmemViews() { fastmemViews.clear(); }
+	void addFastmemView(u32 guestVaddr, size_t arenaOffset, size_t size, bool w, bool x = false) {
+		if (useFastmem) {
+			const auto hostAddr = fastmemArenaBase + guestVaddr;
+			auto view = arena.CreateView(arenaOffset, size, w, x, hostAddr);
+
+			if (view) {
+				fastmemViews.push_back(std::move(view.value()));
+			}
+		}
+	}
+#else
+	void resetFastmemViews() {}
+	void addFastmemView(u32 guestVaddr, size_t arenaOffset, size_t size, bool r, bool w, bool x = false) {}
+#endif
+
 	std::bitset<FCRAM_PAGE_COUNT> usedFCRAMPages;
 	std::optional<u32> findPaddr(u32 size);
 	u64 timeSince3DSEpoch();
@@ -202,6 +236,9 @@ public:
 		return (addr & pageMask) == 0;
 	}
 
+	bool isFastmemEnabled() { return useFastmem; }
+	u8* getFastmemArenaBase() { return fastmemArenaBase; }
+
 	// Allocate "size" bytes of RAM starting from FCRAM index "paddr" (We pick it ourself if paddr == 0)
 	// And map them to virtual address "vaddr" (We also pick it ourself if vaddr == 0).
 	// If the "linear" flag is on, the paddr pages must be adjacent in FCRAM

src/core/CPU/cpu_dynarmic.cpp

@@ -12,7 +12,8 @@ CPU::CPU(Memory& mem, Kernel& kernel) : mem(mem), env(mem, kernel, *this) {
     config.define_unpredictable_behaviour = true;
     config.global_monitor = &exclusiveMonitor;
     config.processor_id = 0;
-    
+    config.fastmem_pointer = mem.isFastmemEnabled() ? mem.getFastmemArenaBase() : nullptr;
+
     jit = std::make_unique<Dynarmic::A32::Jit>(config);
 }
 

src/core/memory.cpp

@@ -13,6 +13,51 @@ Memory::Memory(u64& cpuTicks) : cpuTicks(cpuTicks) {
 	readTable.resize(totalPageCount, 0);
 	writeTable.resize(totalPageCount, 0);
 	memoryInfo.reserve(32); // Pre-allocate some room for memory allocation info to avoid dynamic allocs
+
+#ifdef PANDA3DS_HARDWARE_FASTMEM
+	constexpr size_t FASTMEM_MEMORY_SIZE = 0x100000000; // Total size of the virtual address space we will occupy (4GB)
+	u8* arenaFcram = nullptr;
+	u8* arenaDSPRam = nullptr;
+
+	constexpr size_t MEMORY_ARENA_SIZE = FCRAM_SIZE + DSP_RAM_SIZE;
+	if (!arena.Create(MEMORY_ARENA_SIZE, true, false) || !arena.IsValid()) {
+		printf("Failed to create fastmem memory arena\n");
+		goto fastmemSetupFail;
+	}
+
+	fastmemArenaBase = (u8*)arena.FindBaseAddressForMapping(FASTMEM_MEMORY_SIZE);
+	if (!fastmemArenaBase) {
+		printf("Couldn't find base address for fastmem\n");
+		goto fastmemSetupFail;
+	}
+
+	arenaFcram = (u8*)arena.CreateViewPtr(FASTMEM_FCRAM_OFFSET, FCRAM_SIZE, true, false);
+	if (!arenaFcram) {
+		printf("Couldn't allocate FCRAM for fastmem");
+		goto fastmemSetupFail;
+	}
+
+	arenaDSPRam = (u8*)arena.CreateViewPtr(FASTMEM_DSP_RAM_OFFSET, DSP_RAM_SIZE, true, false);
+	if (!arenaDSPRam) {
+		printf("Couldn't allocate FCRAM for fastmem");
+		goto fastmemSetupFail;
+	}
+
+	fcram = arenaFcram;
+	dspRam = arenaDSPRam;
+	useFastmem = true;
+	return;
+
+fastmemSetupFail:
+	if (arenaFcram) arena.ReleaseViewPtr(arenaFcram, FCRAM_SIZE);
+	if (arenaDSPRam) arena.ReleaseViewPtr(arenaDSPRam, DSP_RAM_SIZE);
+	useFastmem = false;
+	fastmemArenaBase = nullptr;
+	return;
+#else
+	useFastmem = false;
+	fastmemArenaBase = nullptr;
+#endif
 }
 
 void Memory::reset() {
@@ -27,6 +72,9 @@ void Memory::reset() {
 		writeTable[i] = 0;
 	}
 
+	resetFastmemViews(); // Reset the entire fastmem state
+	addFastmemView(VirtualAddrs::DSPMemStart, FASTMEM_DSP_RAM_OFFSET, DSP_RAM_SIZE, true, false); // Allocate RW mapping for DSP RAM
+
 	// Map stack pages as R/W
 	// We have 16KB for the stack, so we allocate the last 16KB of APPLICATION FCRAM for the stack
 	u32 basePaddrForStack = FCRAM_APPLICATION_SIZE - VirtualAddrs::DefaultStackSize;
@@ -284,6 +332,9 @@ std::optional<u32> Memory::allocateMemory(u32 vaddr, u32 paddr, u32 size, bool l
 
 		// Mark FCRAM page as allocated and go on
 		usedFCRAMPages[physPage] = true;
+		// Add mapping to the fastmem arena
+		addFastmemView(size_t(virtualPage) * pageSize, FASTMEM_FCRAM_OFFSET + size_t(physPage) * pageSize, pageSize, w, false);
+
 		virtualPage++;
 		physPage++;
 	}

third_party/Duckstation/memory_arena.cpp

@@ -125,7 +125,6 @@ static std::string GetFileMappingName()
 #endif
 
   std::string ret = std::string("Alber_") + std::to_string(pid);
-  printf("File mapping name: %s\n", ret.c_str());
   return ret;
 }