BuildSystem: Fix GCC compilation with new host GCC

Newer GCCs fail to compile GCC 15.2.0 because of how u8"" literals are
handled. This can be bypassed by compiling with -fno-char8_t

.gitignore

@@ -3,3 +3,4 @@
 build/
 base/
 script/fakeroot-context
+tools/update-image-perms

BAN/include/BAN/Array.h

@@ -18,78 +18,78 @@ namespace BAN
 		using const_iterator = ConstIteratorSimple<T, Array>;
 
 	public:
-		Array() = default;
-		Array(const T&);
+		constexpr Array() = default;
+		constexpr Array(const T&);
 
 		iterator begin() { return iterator(m_data); }
 		iterator end() { return iterator(m_data + size()); }
 		const_iterator begin() const { return const_iterator(m_data); }
 		const_iterator end() const { return const_iterator(m_data + size()); }
 
-		const T& operator[](size_type) const;
-		T& operator[](size_type);
+		constexpr const T& operator[](size_type) const;
+		constexpr T& operator[](size_type);
 
-		const T& back() const;
-		T& back();
-		const T& front() const;
-		T& front();
+		constexpr const T& back() const;
+		constexpr T& back();
+		constexpr const T& front() const;
+		constexpr T& front();
 
 		Span<T> span() { return Span(m_data, size()); }
-		const Span<T> span() const { return Span(m_data, size()); }
+		Span<const T> span() const { return Span(m_data, size()); }
 
 		constexpr size_type size() const;
 
-		const T* data() const { return m_data; }
-		T* data() { return m_data; }
+		constexpr const T* data() const { return m_data; }
+		constexpr T* data() { return m_data; }
 
 	private:
 		T m_data[S] {};
 	};
 
 	template<typename T, size_t S>
-	Array<T, S>::Array(const T& value)
+	constexpr Array<T, S>::Array(const T& value)
 	{
 		for (size_type i = 0; i < S; i++)
 			m_data[i] = value;
 	}
 
 	template<typename T, size_t S>
-	const T& Array<T, S>::operator[](size_type index) const
+	constexpr const T& Array<T, S>::operator[](size_type index) const
 	{
 		ASSERT(index < S);
 		return m_data[index];
 	}
 
 	template<typename T, size_t S>
-	T& Array<T, S>::operator[](size_type index)
+	constexpr T& Array<T, S>::operator[](size_type index)
 	{
 		ASSERT(index < S);
 		return m_data[index];
 	}
 
 	template<typename T, size_t S>
-	const T& Array<T, S>::back() const
+	constexpr const T& Array<T, S>::back() const
 	{
 		ASSERT(S != 0);
 		return m_data[S - 1];
 	}
 
 	template<typename T, size_t S>
-	T& Array<T, S>::back()
+	constexpr T& Array<T, S>::back()
 	{
 		ASSERT(S != 0);
 		return m_data[S - 1];
 	}
 
 	template<typename T, size_t S>
-	const T& Array<T, S>::front() const
+	constexpr const T& Array<T, S>::front() const
 	{
 		ASSERT(S != 0);
 		return m_data[0];
 	}
 
 	template<typename T, size_t S>
-	T& Array<T, S>::front()
+	constexpr T& Array<T, S>::front()
 	{
 		ASSERT(S != 0);
 		return m_data[0];

BAN/include/BAN/Atomic.h

@@ -1,5 +1,7 @@
 #pragma once
 
+#include <BAN/Traits.h>
+
 namespace BAN
 {
 
@@ -13,8 +15,36 @@ namespace BAN
 		memory_order_seq_cst = __ATOMIC_SEQ_CST,
 	};
 
-	template<typename T, MemoryOrder MEM_ORDER = MemoryOrder::memory_order_seq_cst>
-	requires requires { __atomic_always_lock_free(sizeof(T), 0); }
+	template<typename T> concept atomic_c = is_integral_v<T> || is_pointer_v<T>;
+	template<typename T> concept atomic_lockfree_c = (is_integral_v<T> || is_pointer_v<T>) && __atomic_always_lock_free(sizeof(T), 0);
+
+	template<atomic_lockfree_c T, atomic_c U>
+	inline void atomic_store(T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { __atomic_store_n(&obj, value, mem_order); }
+	template<atomic_lockfree_c T>
+	inline T atomic_load(T& obj, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_load_n(&obj, mem_order); }
+
+	template<atomic_lockfree_c T, atomic_c U>
+	inline T atomic_exchange(T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_exchange_n(&obj, value, mem_order); }
+	template<atomic_lockfree_c T, atomic_lockfree_c U, atomic_c V>
+	inline bool atomic_compare_exchange(T& obj, U& expected, V value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_compare_exchange_n(&obj, &expected, value, false, mem_order, mem_order); }
+
+#define DECL_ATOMIC_INLINE template<atomic_lockfree_c T, atomic_c U> inline
+	DECL_ATOMIC_INLINE T atomic_add_fetch (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_add_fetch (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_sub_fetch (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_sub_fetch (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_and_fetch (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_and_fetch (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_xor_fetch (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_xor_fetch (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_or_fetch  (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_or_fetch  (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_nand_fetch(T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_nand_fetch(&obj, value, mem_order); }
+
+	DECL_ATOMIC_INLINE T atomic_fetch_add (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_add (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_fetch_sub (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_sub (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_fetch_and (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_and (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_fetch_xor (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_xor (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_fetch_or  (T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_or  (&obj, value, mem_order); }
+	DECL_ATOMIC_INLINE T atomic_fetch_nand(T& obj, U value, MemoryOrder mem_order = MemoryOrder::memory_order_seq_cst) { return __atomic_fetch_nand(&obj, value, mem_order); }
+#undef DECL_ATOMIC_INLINE
+
+	template<atomic_lockfree_c T, MemoryOrder MEM_ORDER = MemoryOrder::memory_order_seq_cst>
 	class Atomic
 	{
 		Atomic(const Atomic&) = delete;
@@ -26,41 +56,41 @@ namespace BAN
 		constexpr Atomic() : m_value(0) {}
 		constexpr Atomic(T val) : m_value(val) {}
 
-		inline T load(MemoryOrder mem_order = MEM_ORDER) const volatile			{ return __atomic_load_n(&m_value, mem_order); }
-		inline void store(T val, MemoryOrder mem_order = MEM_ORDER) volatile	{ __atomic_store_n(&m_value, val, mem_order); }
+		inline T load(MemoryOrder mem_order = MEM_ORDER) const volatile			{ return atomic_load(m_value, mem_order); }
+		inline void store(T val, MemoryOrder mem_order = MEM_ORDER) volatile	{ atomic_store(m_value, val, mem_order); }
 
 		inline T operator=(T val) volatile { store(val); return val; }
 
 		inline operator T() const volatile { return load(); }
 
-		inline T operator+=(T val) volatile { return __atomic_add_fetch(&m_value, val, MEM_ORDER); }
-		inline T operator-=(T val) volatile { return __atomic_sub_fetch(&m_value, val, MEM_ORDER); }
-		inline T operator&=(T val) volatile { return __atomic_and_fetch(&m_value, val, MEM_ORDER); }
-		inline T operator^=(T val) volatile { return __atomic_xor_fetch(&m_value, val, MEM_ORDER); }
-		inline T operator|=(T val) volatile { return __atomic_or_fetch(&m_value, val, MEM_ORDER); }
+		inline T operator+=(T val) volatile { return atomic_add_fetch(m_value, val, MEM_ORDER); }
+		inline T operator-=(T val) volatile { return atomic_sub_fetch(m_value, val, MEM_ORDER); }
+		inline T operator&=(T val) volatile { return atomic_and_fetch(m_value, val, MEM_ORDER); }
+		inline T operator^=(T val) volatile { return atomic_xor_fetch(m_value, val, MEM_ORDER); }
+		inline T operator|=(T val) volatile { return atomic_or_fetch(m_value, val, MEM_ORDER); }
 
-		inline T operator--() volatile { return __atomic_sub_fetch(&m_value, 1, MEM_ORDER); }
-		inline T operator++() volatile { return __atomic_add_fetch(&m_value, 1, MEM_ORDER); }
+		inline T operator--() volatile { return atomic_sub_fetch(m_value, 1, MEM_ORDER); }
+		inline T operator++() volatile { return atomic_add_fetch(m_value, 1, MEM_ORDER); }
 
-		inline T operator--(int) volatile { return __atomic_fetch_sub(&m_value, 1, MEM_ORDER); }
-		inline T operator++(int) volatile { return __atomic_fetch_add(&m_value, 1, MEM_ORDER); }
+		inline T operator--(int) volatile { return atomic_fetch_sub(m_value, 1, MEM_ORDER); }
+		inline T operator++(int) volatile { return atomic_fetch_add(m_value, 1, MEM_ORDER); }
 
-		inline bool compare_exchange(T& expected, T desired, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_compare_exchange_n(&m_value, &expected, desired, false, mem_order, mem_order); }
-		inline T exchange(T desired, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_exchange_n(&m_value, desired, mem_order); };
+		inline bool compare_exchange(T& expected, T desired, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_compare_exchange(m_value, expected, desired, mem_order); }
+		inline T exchange(T desired, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_exchange(m_value, desired, mem_order); };
 
-		inline T add_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_add_fetch (&m_value, val, mem_order); }
-		inline T sub_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_sub_fetch (&m_value, val, mem_order); }
-		inline T and_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_and_fetch (&m_value, val, mem_order); }
-		inline T xor_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_xor_fetch (&m_value, val, mem_order); }
-		inline T or_fetch  (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_or_fetch  (&m_value, val, mem_order); }
-		inline T nand_fetch(T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_nand_fetch(&m_value, val, mem_order); }
+		inline T add_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_add_fetch (m_value, val, mem_order); }
+		inline T sub_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_sub_fetch (m_value, val, mem_order); }
+		inline T and_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_and_fetch (m_value, val, mem_order); }
+		inline T xor_fetch (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_xor_fetch (m_value, val, mem_order); }
+		inline T or_fetch  (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_or_fetch  (m_value, val, mem_order); }
+		inline T nand_fetch(T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_nand_fetch(m_value, val, mem_order); }
 
-		inline T fetch_add (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_fetch_add (&m_value, val, mem_order); }
-		inline T fetch_sub (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_fetch_sub (&m_value, val, mem_order); }
-		inline T fetch_and (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_fetch_and (&m_value, val, mem_order); }
-		inline T fetch_xor (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_fetch_xor (&m_value, val, mem_order); }
-		inline T fetch_or  (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_fetch__or  (&m_value, val, mem_order); }
-		inline T fetch_nand(T val, MemoryOrder mem_order = MEM_ORDER) volatile { return __atomic_nfetch_and(&m_value, val, mem_order); }
+		inline T fetch_add (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_add (m_value, val, mem_order); }
+		inline T fetch_sub (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_sub (m_value, val, mem_order); }
+		inline T fetch_and (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_and (m_value, val, mem_order); }
+		inline T fetch_xor (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_xor (m_value, val, mem_order); }
+		inline T fetch_or  (T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_or  (m_value, val, mem_order); }
+		inline T fetch_nand(T val, MemoryOrder mem_order = MEM_ORDER) volatile { return atomic_fetch_nand(m_value, val, mem_order); }
 
 	private:
 		T m_value;

BAN/include/BAN/CircularQueue.h

@@ -34,6 +34,11 @@ namespace BAN
 		const T& back() const;
 		T& back();
 
+		const T& operator[](size_t index) const;
+		T& operator[](size_t index);
+
+		void clear();
+
 		size_type size() const { return m_size; }
 		bool empty() const { return size() == 0; }
 		bool full() const { return size() == capacity(); }
@@ -53,8 +58,7 @@ namespace BAN
 	template<typename T, size_t S>
 	CircularQueue<T, S>::~CircularQueue()
 	{
-		for (size_type i = 0; i < m_size; i++)
-			element_at((m_first + i) % capacity())->~T();
+		clear();
 	}
 
 	template<typename T, size_t S>
@@ -115,6 +119,28 @@ namespace BAN
 		return *element_at((m_first + m_size - 1) % capacity());
 	}
 
+	template<typename T, size_t S>
+	const T& CircularQueue<T, S>::operator[](size_t index) const
+	{
+		ASSERT(index < m_size);
+		return *element_at((m_first + index) % capacity());
+	}
+
+	template<typename T, size_t S>
+	T& CircularQueue<T, S>::operator[](size_t index)
+	{
+		ASSERT(index < m_size);
+		return *element_at((m_first + index) % capacity());
+	}
+
+	template<typename T, size_t S>
+	void CircularQueue<T, S>::clear()
+	{
+		for (size_type i = 0; i < m_size; i++)
+			element_at((m_first + i) % capacity())->~T();
+		m_size = 0;
+	}
+
 	template<typename T, size_t S>
 	const T* CircularQueue<T, S>::element_at(size_type index) const
 	{

BAN/include/BAN/Debug.h

@@ -9,29 +9,35 @@
 #include <BAN/Formatter.h>
 #include <stdio.h>
 
-#define __debug_putchar [](int c) { putc(c, stddbg); }
+#define __debug_putchar [](int c) { putc_unlocked(c, stddbg); }
 
 #define dprintln(...)											\
 	do {														\
+		flockfile(stddbg);										\
 		BAN::Formatter::print(__debug_putchar, __VA_ARGS__);	\
-		BAN::Formatter::print(__debug_putchar,"\r\n");			\
+		BAN::Formatter::print(__debug_putchar,"\n");			\
 		fflush(stddbg);											\
+		funlockfile(stddbg);									\
 	} while (false)
 
 #define dwarnln(...)											\
 	do {														\
+		flockfile(stddbg);										\
 		BAN::Formatter::print(__debug_putchar, "\e[33m");		\
 		BAN::Formatter::print(__debug_putchar, __VA_ARGS__);	\
-		BAN::Formatter::print(__debug_putchar, "\e[m\r\n");		\
+		BAN::Formatter::print(__debug_putchar, "\e[m\n");		\
 		fflush(stddbg);											\
+		funlockfile(stddbg);									\
 	} while(false)
 
 #define derrorln(...)											\
 	do {														\
+		flockfile(stddbg);										\
 		BAN::Formatter::print(__debug_putchar, "\e[31m");		\
 		BAN::Formatter::print(__debug_putchar, __VA_ARGS__);	\
-		BAN::Formatter::print(__debug_putchar, "\e[m\r\n");		\
+		BAN::Formatter::print(__debug_putchar, "\e[m\n");		\
 		fflush(stddbg);											\
+		funlockfile(stddbg);									\
 	} while(false)
 
 #define dprintln_if(cond, ...)		\

BAN/include/BAN/HashMap.h

@@ -1,49 +1,156 @@
 #pragma once
 
-#include <BAN/Hash.h>
-#include <BAN/LinkedList.h>
-#include <BAN/Vector.h>
+#include <BAN/HashSet.h>
 
 namespace BAN
 {
 
-	template<typename Key, typename T, typename HASH = BAN::hash<Key>>
+	template<typename HashSetIt, typename HashMap, typename Entry>
+	class HashMapIterator
+	{
+	public:
+		HashMapIterator() = default;
+
+		Entry& operator*()
+		{
+			return const_cast<Entry&>(m_iterator.operator*());
+		}
+		const Entry& operator*() const
+		{
+			return m_iterator.operator*();
+		}
+
+		Entry* operator->()
+		{
+			return const_cast<Entry*>(m_iterator.operator->());
+		}
+		const Entry* operator->() const
+		{
+			return m_iterator.operator->();
+		}
+
+		HashMapIterator& operator++()
+		{
+			++m_iterator;
+			return *this;
+		}
+		HashMapIterator operator++(int)
+		{
+			auto temp = *this;
+			++(*this);
+			return temp;
+		}
+
+		bool operator==(HashMapIterator other) const
+		{
+			return m_iterator == other.m_iterator;
+		}
+		bool operator!=(HashMapIterator other) const
+		{
+			return m_iterator != other.m_iterator;
+		}
+
+	private:
+		explicit HashMapIterator(HashSetIt it)
+			: m_iterator(it)
+		{ }
+
+	private:
+		HashSetIt m_iterator;
+		friend HashMap;
+	};
+
+	namespace detail
+	{
+		template<typename T, typename Key, typename HASH, typename COMP>
+		concept HashMapFindable = requires(const Key& a, const T& b) { COMP()(a, b); HASH()(b); };
+	}
+
+	template<typename Key, typename T, typename HASH = BAN::hash<Key>, typename COMP = BAN::equal<Key>>
 	class HashMap
 	{
 	public:
 		struct Entry
 		{
-			template<typename... Args>
-			Entry(const Key& key, Args&&... args) requires is_constructible_v<T, Args...>
-				: key(key)
-				, value(forward<Args>(args)...)
-			{}
-
-			template<typename... Args>
-			Entry(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
-				: key(BAN::move(key))
-				, value(forward<Args>(args)...)
-			{}
-
-			Key key;
+			const Key key;
 			T value;
+
+			Entry() = delete;
+			Entry& operator=(const Entry&) = delete;
+			Entry& operator=(Entry&&) = delete;
+
+			Entry(const Entry& other)
+				: key(other.key)
+				, value(other.value)
+			{ }
+
+			Entry(Entry&& other)
+				: key(BAN::move(const_cast<Key&>(other.key)))
+				, value(BAN::move(other.value))
+			{ }
+
+			template<typename... Args>
+			Entry(Key&& key, Args&&... args)
+				: key(BAN::move(key))
+				, value(BAN::forward<Args>(args)...)
+			{ }
+		};
+
+		struct EntryHash
+		{
+			template<detail::HashMapFindable<Key, HASH, COMP> U>
+			constexpr bool operator()(const U& a)
+			{
+				return HASH()(a);
+			}
+			constexpr bool operator()(const Entry& a)
+			{
+				return HASH()(a.key);
+			}
+		};
+
+		struct EntryComp
+		{
+			template<detail::HashMapFindable<Key, HASH, COMP> U>
+			constexpr bool operator()(const Entry& a, const U& b)
+			{
+				return COMP()(a.key, b);
+			}
+			constexpr bool operator()(const Entry& a, const Entry& b)
+			{
+				return COMP()(a.key, b.key);
+			}
 		};
 
 	public:
 		using size_type      = size_t;
 		using key_type       = Key;
 		using value_type     = T;
-		using iterator = IteratorDouble<Entry, Vector, LinkedList, HashMap>;
-		using const_iterator = ConstIteratorDouble<Entry, Vector, LinkedList, HashMap>;
+		using iterator       = HashMapIterator<typename HashSet<Entry, EntryHash, EntryComp>::iterator,       HashMap, Entry>;
+		using const_iterator = HashMapIterator<typename HashSet<Entry, EntryHash, EntryComp>::const_iterator, HashMap, const Entry>;
 
 	public:
 		HashMap() = default;
-		HashMap(const HashMap<Key, T, HASH>&);
-		HashMap(HashMap<Key, T, HASH>&&);
-		~HashMap();
+		~HashMap() { clear(); }
 
-		HashMap<Key, T, HASH>& operator=(const HashMap<Key, T, HASH>&);
-		HashMap<Key, T, HASH>& operator=(HashMap<Key, T, HASH>&&);
+		HashMap(const HashMap& other) { *this = other; }
+		HashMap& operator=(const HashMap& other)
+		{
+			m_hash_set = other.m_hash_set;
+			return *this;
+		}
+
+		HashMap(HashMap&& other) { *this = BAN::move(other); }
+		HashMap& operator=(HashMap&& other)
+		{
+			m_hash_set = BAN::move(other.m_hash_set);
+			return *this;
+		}
+
+		iterator begin()             { return iterator(m_hash_set.begin()); }
+		iterator end()               { return iterator(m_hash_set.end()); }
+		const_iterator begin() const { return const_iterator(m_hash_set.begin()); }
+		const_iterator end() const   { return const_iterator(m_hash_set.end()); }
 
 		ErrorOr<iterator> insert(const Key& key, const T& value)           { return emplace(key, value); }
 		ErrorOr<iterator> insert(const Key& key, T&& value)                { return emplace(key, move(value)); }
@@ -57,263 +164,100 @@ namespace BAN
 
 		template<typename... Args>
 		ErrorOr<iterator> emplace(const Key& key, Args&&... args) requires is_constructible_v<T, Args...>
-		{ return emplace(Key(key), forward<Args>(args)...); }
+		{ return emplace(Key(key), BAN::forward<Args>(args)...); }
 		template<typename... Args>
-		ErrorOr<iterator> emplace(Key&&, Args&&...) requires is_constructible_v<T, Args...>;
+		ErrorOr<iterator> emplace(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
+		{
+			ASSERT(!contains(key));
+			auto it = TRY(m_hash_set.insert(Entry { BAN::move(key), T(BAN::forward<Args>(args)...) }));
+			return iterator(it);
+		}
 
 		template<typename... Args>
 		ErrorOr<iterator> emplace_or_assign(const Key& key, Args&&... args) requires is_constructible_v<T, Args...>
-		{ return emplace_or_assign(Key(key), forward<Args>(args)...); }
+		{ return emplace_or_assign(Key(key), BAN::forward<Args>(args)...); }
 		template<typename... Args>
-		ErrorOr<iterator> emplace_or_assign(Key&&, Args&&...) requires is_constructible_v<T, Args...>;
-
-		iterator begin() { return iterator(m_buckets.end(), m_buckets.begin()); }
-		iterator end()   { return iterator(m_buckets.end(), m_buckets.end()); }
-		const_iterator begin() const { return const_iterator(m_buckets.end(), m_buckets.begin()); }
-		const_iterator end() const   { return const_iterator(m_buckets.end(), m_buckets.end()); }
-
-		ErrorOr<void> reserve(size_type);
-
-		void remove(const Key&);
-		void remove(iterator it);
-		void clear();
-
-		T& operator[](const Key&);
-		const T& operator[](const Key&) const;
-
-		iterator find(const Key& key);
-		const_iterator find(const Key& key) const;
-		bool contains(const Key&) const;
-
-		bool empty() const;
-		size_type size() const;
-
-	private:
-		ErrorOr<void> rebucket(size_type);
-		LinkedList<Entry>& get_bucket(const Key&);
-		const LinkedList<Entry>& get_bucket(const Key&) const;
-		Vector<LinkedList<Entry>>::iterator get_bucket_iterator(const Key&);
-		Vector<LinkedList<Entry>>::const_iterator get_bucket_iterator(const Key&) const;
-
-	private:
-		Vector<LinkedList<Entry>> m_buckets;
-		size_type m_size = 0;
-
-		friend iterator;
-	};
-
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::HashMap(const HashMap<Key, T, HASH>& other)
+		ErrorOr<iterator> emplace_or_assign(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
 		{
-		*this = other;
+			if (auto it = m_hash_set.find(key); it != m_hash_set.end())
+			{
+				const_cast<T&>(it->value) = T(BAN::forward<Args>(args)...);
+				return iterator(it);
 			}
 
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::HashMap(HashMap<Key, T, HASH>&& other)
-	{
-		*this = move(other);
+			auto it = TRY(m_hash_set.insert(Entry { BAN::move(key), T(BAN::forward<Args>(args)...) }));
+			return iterator(it);
 		}
 
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::~HashMap()
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		void remove(const U& key)
 		{
-		clear();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>& HashMap<Key, T, HASH>::operator=(const HashMap<Key, T, HASH>& other)
-	{
-		clear();
-		m_buckets = other.m_buckets;
-		m_size = other.m_size;
-		return *this;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>& HashMap<Key, T, HASH>::operator=(HashMap<Key, T, HASH>&& other)
-	{
-		clear();
-		m_buckets = move(other.m_buckets);
-		m_size = other.m_size;
-		other.m_size = 0;
-		return *this;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	template<typename... Args>
-	ErrorOr<typename HashMap<Key, T, HASH>::iterator> HashMap<Key, T, HASH>::emplace(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
-	{
-		ASSERT(!contains(key));
-		TRY(rebucket(m_size + 1));
-
-		auto bucket_it = get_bucket_iterator(key);
-		TRY(bucket_it->emplace_back(move(key), forward<Args>(args)...));
-		m_size++;
-
-		return iterator(m_buckets.end(), bucket_it, prev(bucket_it->end(), 1));
-	}
-
-	template<typename Key, typename T, typename HASH>
-	template<typename... Args>
-	ErrorOr<typename HashMap<Key, T, HASH>::iterator> HashMap<Key, T, HASH>::emplace_or_assign(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
-	{
-		if (empty())
-			return emplace(move(key), forward<Args>(args)...);
-
-		auto bucket_it = get_bucket_iterator(key);
-		for (auto entry_it = bucket_it->begin(); entry_it != bucket_it->end(); entry_it++)
-		{
-			if (entry_it->key != key)
-				continue;
-			entry_it->value = T(forward<Args>(args)...);
-			return iterator(m_buckets.end(), bucket_it, entry_it);
-		}
-
-		return emplace(move(key), forward<Args>(args)...);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	ErrorOr<void> HashMap<Key, T, HASH>::reserve(size_type size)
-	{
-		TRY(rebucket(size));
-		return {};
-	}
-
-	template<typename Key, typename T, typename HASH>
-	void HashMap<Key, T, HASH>::remove(const Key& key)
-	{
-		auto it = find(key);
-		if (it != end())
+			if (auto it = find(key); it != end())
 				remove(it);
 		}
 
-	template<typename Key, typename T, typename HASH>
-	void HashMap<Key, T, HASH>::remove(iterator it)
+		iterator remove(iterator it)
 		{
-		it.outer_current()->remove(it.inner_current());
-		m_size--;
+			return iterator(m_hash_set.remove(it.m_iterator));
 		}
 
-	template<typename Key, typename T, typename HASH>
-	void HashMap<Key, T, HASH>::clear()
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		iterator find(const U& key)
 		{
-		m_buckets.clear();
-		m_size = 0;
+			return iterator(m_hash_set.find(key));
 		}
 
-	template<typename Key, typename T, typename HASH>
-	T& HashMap<Key, T, HASH>::operator[](const Key& key)
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		const_iterator find(const U& key) const
 		{
-		ASSERT(!empty());
-		auto& bucket = get_bucket(key);
-		for (Entry& entry : bucket)
-			if (entry.key == key)
-				return entry.value;
-		ASSERT_NOT_REACHED();
+			return const_iterator(m_hash_set.find(key));
 		}
 
-	template<typename Key, typename T, typename HASH>
-	const T& HashMap<Key, T, HASH>::operator[](const Key& key) const
+		void clear()
 		{
-		ASSERT(!empty());
-		const auto& bucket = get_bucket(key);
-		for (const Entry& entry : bucket)
-			if (entry.key == key)
-				return entry.value;
-		ASSERT_NOT_REACHED();
+			m_hash_set.clear();
 		}
 
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::iterator HashMap<Key, T, HASH>::find(const Key& key)
+		ErrorOr<void> reserve(size_type size)
 		{
-		if (empty())
-			return end();
-		auto bucket_it = get_bucket_iterator(key);
-		for (auto it = bucket_it->begin(); it != bucket_it->end(); it++)
-			if (it->key == key)
-				return iterator(m_buckets.end(), bucket_it, it);
-		return end();
+			return m_hash_set.reserve(size);
 		}
 
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::const_iterator HashMap<Key, T, HASH>::find(const Key& key) const
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		T& operator[](const U& key)
 		{
-		if (empty())
-			return end();
-		auto bucket_it = get_bucket_iterator(key);
-		for (auto it = bucket_it->begin(); it != bucket_it->end(); it++)
-			if (it->key == key)
-				return const_iterator(m_buckets.end(), bucket_it, it);
-		return end();
+			return find(key)->value;
 		}
 
-	template<typename Key, typename T, typename HASH>
-	bool HashMap<Key, T, HASH>::contains(const Key& key) const
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		const T& operator[](const U& key) const
+		{
+			return find(key)->value;
+		}
+
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		bool contains(const U& key) const
 		{
 			return find(key) != end();
 		}
 
-	template<typename Key, typename T, typename HASH>
-	bool HashMap<Key, T, HASH>::empty() const
+		size_type capacity() const
 		{
-		return m_size == 0;
+			return m_hash_set.capacity();
 		}
 
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::size_type HashMap<Key, T, HASH>::size() const
+		size_type size() const
 		{
-		return m_size;
+			return m_hash_set.size();
 		}
 
-	template<typename Key, typename T, typename HASH>
-	ErrorOr<void> HashMap<Key, T, HASH>::rebucket(size_type bucket_count)
+		bool empty() const
 		{
-		if (m_buckets.size() >= bucket_count)
-			return {};
-
-		size_type new_bucket_count = BAN::Math::max<size_type>(bucket_count, m_buckets.size() * 2);
-		Vector<LinkedList<Entry>> new_buckets;
-		TRY(new_buckets.resize(new_bucket_count));
-
-		for (auto& bucket : m_buckets)
-		{
-			for (auto it = bucket.begin(); it != bucket.end();)
-			{
-				size_type new_bucket_index = HASH()(it->key) % new_buckets.size();
-				it = bucket.move_element_to_other_linked_list(new_buckets[new_bucket_index], new_buckets[new_bucket_index].end(), it);
-			}
+			return m_hash_set.empty();
 		}
 
-		m_buckets = move(new_buckets);
-		return {};
-	}
-
-	template<typename Key, typename T, typename HASH>
-	LinkedList<typename HashMap<Key, T, HASH>::Entry>& HashMap<Key, T, HASH>::get_bucket(const Key& key)
-	{
-		return *get_bucket_iterator(key);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	const LinkedList<typename HashMap<Key, T, HASH>::Entry>& HashMap<Key, T, HASH>::get_bucket(const Key& key) const
-	{
-		return *get_bucket_iterator(key);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	Vector<LinkedList<typename HashMap<Key, T, HASH>::Entry>>::iterator HashMap<Key, T, HASH>::get_bucket_iterator(const Key& key)
-	{
-		ASSERT(!m_buckets.empty());
-		auto index = HASH()(key) % m_buckets.size();
-		return next(m_buckets.begin(), index);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	Vector<LinkedList<typename HashMap<Key, T, HASH>::Entry>>::const_iterator HashMap<Key, T, HASH>::get_bucket_iterator(const Key& key) const
-	{
-		ASSERT(!m_buckets.empty());
-		auto index = HASH()(key) % m_buckets.size();
-		return next(m_buckets.begin(), index);
-	}
+	private:
+		HashSet<Entry, EntryHash, EntryComp> m_hash_set;
+	};
 
 }

BAN/include/BAN/HashSet.h

@@ -2,198 +2,358 @@
 
 #include <BAN/Errors.h>
 #include <BAN/Hash.h>
-#include <BAN/Iterators.h>
-#include <BAN/LinkedList.h>
 #include <BAN/Math.h>
 #include <BAN/Move.h>
-#include <BAN/Vector.h>
+#include <BAN/New.h>
 
 namespace BAN
 {
 
-	template<typename T, typename HASH = hash<T>>
+	template<typename HashSet, typename Bucket, typename T>
+	class HashSetIterator
+	{
+	public:
+		HashSetIterator() = default;
+
+		const T& operator*() const
+		{
+			ASSERT(m_bucket);
+			return *m_bucket->element();
+		}
+
+		const T* operator->() const
+		{
+			ASSERT(m_bucket);
+			return m_bucket->element();
+		}
+
+		HashSetIterator& operator++()
+		{
+			ASSERT(m_bucket);
+			m_bucket++;
+			skip_to_valid_bucket();
+			return *this;
+		}
+		HashSetIterator operator++(int)
+		{
+			auto temp = *this;
+			++(*this);
+			return temp;
+		}
+
+		bool operator==(HashSetIterator other) const
+		{
+			return m_bucket == other.m_bucket;
+		}
+		bool operator!=(HashSetIterator other) const
+		{
+			return m_bucket != other.m_bucket;
+		}
+
+	private:
+		explicit HashSetIterator(Bucket* bucket)
+			: m_bucket(bucket)
+		{
+			if (m_bucket != nullptr)
+				skip_to_valid_bucket();
+		}
+
+		void skip_to_valid_bucket()
+		{
+			while (m_bucket->state != Bucket::USED && !m_bucket->end)
+				m_bucket++;
+			if (m_bucket->end)
+				m_bucket = nullptr;
+		}
+
+	private:
+		Bucket* m_bucket { nullptr };
+		friend HashSet;
+	};
+
+	namespace detail
+	{
+		template<typename T, typename U, typename HASH, typename COMP>
+		concept HashSetFindable = requires(const U& a, const T& b) { COMP()(a, b); HASH()(b); };
+	}
+
+	template<typename T, typename HASH = BAN::hash<T>, typename COMP = BAN::equal<T>>
 	class HashSet
 	{
+	private:
+		struct Bucket
+		{
+			static constexpr uint8_t UNUSED  = 0;
+			static constexpr uint8_t USED    = 1;
+			static constexpr uint8_t REMOVED = 2;
+
+			alignas(T) uint8_t storage[sizeof(T)];
+			hash_t hash;
+			uint8_t state       : 2;
+			uint8_t chain_start : 1;
+			uint8_t end         : 1;
+
+			T* element() { return reinterpret_cast<T*>(storage); }
+			const T* element() const { return reinterpret_cast<const T*>(storage); }
+		};
+
 	public:
 		using value_type     = T;
 		using size_type      = size_t;
-		using iterator = IteratorDouble<T, Vector, LinkedList, HashSet>;
-		using const_iterator = ConstIteratorDouble<T, Vector, LinkedList, HashSet>;
+		using iterator       = HashSetIterator<HashSet,       Bucket,       T>;
+		using const_iterator = HashSetIterator<HashSet, const Bucket, const T>;
 
 	public:
 		HashSet() = default;
-		HashSet(const HashSet&);
-		HashSet(HashSet&&);
+		~HashSet() { clear(); }
 
-		HashSet& operator=(const HashSet&);
-		HashSet& operator=(HashSet&&);
-
-		ErrorOr<void> insert(const T&);
-		ErrorOr<void> insert(T&&);
-		void remove(const T&);
-		void clear();
-
-		ErrorOr<void> reserve(size_type);
-
-		iterator begin() { return iterator(m_buckets.end(), m_buckets.begin()); }
-		iterator end()   { return iterator(m_buckets.end(), m_buckets.end()); }
-		const_iterator begin() const { return const_iterator(m_buckets.end(), m_buckets.begin()); }
-		const_iterator end() const   { return const_iterator(m_buckets.end(), m_buckets.end()); }
-
-		bool contains(const T&) const;
-
-		size_type size() const;
-		bool empty() const;
-
-	private:
-		ErrorOr<void> rebucket(size_type);
-		LinkedList<T>& get_bucket(const T&);
-		const LinkedList<T>& get_bucket(const T&) const;
-
-	private:
-		Vector<LinkedList<T>> m_buckets;
-		size_type m_size = 0;
-	};
-
-	template<typename T, typename HASH>
-	HashSet<T, HASH>::HashSet(const HashSet& other)
-		: m_buckets(other.m_buckets)
-		, m_size(other.m_size)
-	{
-	}
-
-	template<typename T, typename HASH>
-	HashSet<T, HASH>::HashSet(HashSet&& other)
-		: m_buckets(move(other.m_buckets))
-		, m_size(other.m_size)
-	{
-		other.clear();
-	}
-
-	template<typename T, typename HASH>
-	HashSet<T, HASH>& HashSet<T, HASH>::operator=(const HashSet& other)
+		HashSet(const HashSet& other) { *this = other; }
+		HashSet& operator=(const HashSet& other)
 		{
 			clear();
+
+			MUST(reserve(other.size()));
+			for (auto& bucket : other)
+				MUST(insert(bucket));
+
+			return *this;
+		}
+
+		HashSet(HashSet&& other) { *this = BAN::move(other); }
+		HashSet& operator=(HashSet&& other)
+		{
+			clear();
+
 			m_buckets  = other.m_buckets;
+			m_capacity = other.m_capacity;
 			m_size     = other.m_size;
+			m_removed  = other.m_removed;
+
+			other.m_buckets  = nullptr;
+			other.m_capacity = 0;
+			other.m_size     = 0;
+			other.m_removed  = 0;
+
 			return *this;
 		}
 
-	template<typename T, typename HASH>
-	HashSet<T, HASH>& HashSet<T, HASH>::operator=(HashSet&& other)
+		iterator begin()             { return iterator(m_buckets); }
+		iterator end()               { return iterator(nullptr); }
+		const_iterator begin() const { return const_iterator(m_buckets); }
+		const_iterator end() const   { return const_iterator(nullptr); }
+
+		ErrorOr<iterator> insert(const T& value)
 		{
-		clear();
-		m_buckets = move(other.m_buckets);
-		m_size = other.m_size;
-		other.clear();
-		return *this;
+			return insert(T(value));
 		}
 
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::insert(const T& key)
+		ErrorOr<iterator> insert(T&& value)
 		{
-		return insert(move(T(key)));
+			if (should_rehash_with_size(m_size + 1))
+				TRY(rehash(m_size * 2));
+			return insert_impl(BAN::move(value), HASH()(value));
 		}
 
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::insert(T&& key)
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		void remove(const U& value)
 		{
-		if (!empty() && get_bucket(key).contains(key))
-			return {};
-
-		TRY(rebucket(m_size + 1));
-		TRY(get_bucket(key).push_back(move(key)));
-		m_size++;
-		return {};
+			if (auto it = find(value); it != end())
+				remove(it);
 		}
 
-	template<typename T, typename HASH>
-	void HashSet<T, HASH>::remove(const T& key)
+		iterator remove(iterator it)
 		{
-		if (empty()) return;
-		auto& bucket = get_bucket(key);
-		for (auto it = bucket.begin(); it != bucket.end(); it++)
-		{
-			if (*it == key)
-			{
-				bucket.remove(it);
+			auto& bucket = *it.m_bucket;
+			bucket.element()->~T();
+			bucket.state = Bucket::REMOVED;
 			m_size--;
-				break;
-			}
-		}
+			m_removed++;
+			return iterator(&bucket);
 		}
 
-	template<typename T, typename HASH>
-	void HashSet<T, HASH>::clear()
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		iterator find(const U& value)
 		{
-		m_buckets.clear();
+			return iterator(const_cast<Bucket*>(find_impl(value).m_bucket));
+		}
+
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		const_iterator find(const U& value) const
+		{
+			return find_impl(value);
+		}
+
+		void clear()
+		{
+			if (m_buckets == nullptr)
+				return;
+
+			for (size_type i = 0; i < m_capacity; i++)
+				if (m_buckets[i].state == Bucket::USED)
+					m_buckets[i].element()->~T();
+
+			BAN::deallocator(m_buckets);
+			m_buckets  = nullptr;
+			m_capacity = 0;
 			m_size     = 0;
+			m_removed  = 0;
 		}
 
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::reserve(size_type size)
+		ErrorOr<void> reserve(size_type size)
 		{
-		TRY(rebucket(size));
+			if (should_rehash_with_size(size))
+				TRY(rehash(size * 2));
 			return {};
 		}
 
-	template<typename T, typename HASH>
-	bool HashSet<T, HASH>::contains(const T& key) const
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		bool contains(const U& value) const
 		{
-		if (empty()) return false;
-		return get_bucket(key).contains(key);
+			return find(value) != end();
 		}
 
-	template<typename T, typename HASH>
-	typename HashSet<T, HASH>::size_type HashSet<T, HASH>::size() const
+		size_type capacity() const
+		{
+			return m_capacity;
+		}
+
+		size_type size() const
 		{
 			return m_size;
 		}
 
-	template<typename T, typename HASH>
-	bool HashSet<T, HASH>::empty() const
+		bool empty() const
 		{
 			return m_size == 0;
 		}
 
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::rebucket(size_type bucket_count)
+	private:
+		ErrorOr<void> rehash(size_type new_capacity)
 		{
-		if (m_buckets.size() >= bucket_count)
-			return {};
+			new_capacity = BAN::Math::max<size_t>(16, BAN::Math::max(new_capacity, m_size + 1));
+			new_capacity = BAN::Math::round_up_to_power_of_two(new_capacity);
 
-		size_type new_bucket_count = Math::max<size_type>(bucket_count, m_buckets.size() * 2);
-		Vector<LinkedList<T>> new_buckets;
-		if (new_buckets.resize(new_bucket_count).is_error())
-			return Error::from_errno(ENOMEM);
+			void* new_buckets = BAN::allocator((new_capacity + 1) * sizeof(Bucket));
+			if (new_buckets == nullptr)
+				return BAN::Error::from_errno(ENOMEM);
+			memset(new_buckets, 0, (new_capacity + 1) * sizeof(Bucket));
 
-		for (auto& bucket : m_buckets)
+			Bucket* old_buckets = m_buckets;
+			const size_type old_capacity = m_capacity;
+
+			m_buckets  = static_cast<Bucket*>(new_buckets);
+			m_capacity = new_capacity;
+			m_size     = 0;
+			m_removed  = 0;
+
+			for (size_type i = 0; i < old_capacity; i++)
 			{
-			for (auto it = bucket.begin(); it != bucket.end();)
-			{
-				size_type new_bucket_index = HASH()(*it) % new_buckets.size();
-				it = bucket.move_element_to_other_linked_list(new_buckets[new_bucket_index], new_buckets[new_bucket_index].end(), it);
-			}
+				auto& old_bucket = old_buckets[i];
+				if (old_bucket.state != Bucket::USED)
+					continue;
+				insert_impl(BAN::move(*old_bucket.element()), old_bucket.hash);
+				old_bucket.element()->~T();
 			}
 
-		m_buckets = move(new_buckets);
+			m_buckets[m_capacity].end = true;
+
+			BAN::deallocator(old_buckets);
+
 			return {};
 		}
 
-	template<typename T, typename HASH>
-	LinkedList<T>& HashSet<T, HASH>::get_bucket(const T& key)
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		const_iterator find_impl(const U& value) const
 		{
-		ASSERT(!m_buckets.empty());
-		size_type index = HASH()(key) % m_buckets.size();
-		return m_buckets[index];
+			if (m_capacity == 0)
+				return end();
+
+			bool first = true;
+			const hash_t orig_hash = HASH()(value);
+			for (auto hash = orig_hash;; hash = get_next_hash_in_chain(hash, orig_hash), first = false)
+			{
+				auto& bucket = m_buckets[hash & (m_capacity - 1)];
+				if (bucket.state == Bucket::USED && bucket.hash == orig_hash && COMP()(*bucket.element(), value))
+					return const_iterator(&bucket);
+				if (bucket.state == Bucket::UNUSED)
+					return end();
+				if (!first && bucket.chain_start)
+					return end();
+			}
 		}
 
-	template<typename T, typename HASH>
-	const LinkedList<T>& HashSet<T, HASH>::get_bucket(const T& key) const
+		iterator insert_impl(T&& value, hash_t orig_hash)
 		{
-		ASSERT(!m_buckets.empty());
-		size_type index = HASH()(key) % m_buckets.size();
-		return m_buckets[index];
+			ASSERT(!should_rehash_with_size(m_size + 1));
+
+			Bucket* target = nullptr;
+
+			bool first = true;
+			for (auto hash = orig_hash;; hash = get_next_hash_in_chain(hash, orig_hash), first = false)
+			{
+				auto& bucket = m_buckets[hash & (m_capacity - 1)];
+
+				if (!first)
+					bucket.chain_start = false;
+
+				if (bucket.state == Bucket::USED)
+				{
+					if (bucket.hash != orig_hash || !COMP()(*bucket.element(), value))
+						continue;
+					target = &bucket;
+					break;
 				}
 
+				if (target == nullptr)
+					target = &bucket;
+
+				if (bucket.state == Bucket::UNUSED)
+					break;
+			}
+
+			switch (target->state)
+			{
+				case Bucket::USED:
+					target->element()->~T();
+					break;
+				case Bucket::REMOVED:
+					m_removed--;
+					[[fallthrough]];
+				case Bucket::UNUSED:
+					m_size++;
+					break;
+			}
+
+			target->chain_start = first && target->state == Bucket::UNUSED;
+			target->hash = orig_hash;
+			target->state = Bucket::USED;
+
+			new (target->element()) T(BAN::move(value));
+
+			return iterator(target);
+		}
+
+		bool should_rehash_with_size(size_type size) const
+		{
+			if (m_capacity < 16)
+				return true;
+			if (size + m_removed > m_capacity / 4 * 3)
+				return true;
+			return false;
+		}
+
+		hash_t get_next_hash_in_chain(hash_t prev_hash, hash_t orig_hash) const
+		{
+			// TODO: does this even provide better performance than `return prev_hash + 1`
+			//       when using "good" hash functions
+			return prev_hash * 1103515245 + (orig_hash | 1);
+		}
+
+	private:
+		Bucket* m_buckets { nullptr };
+		size_type m_capacity { 0 };
+		size_type m_size     { 0 };
+		size_type m_removed  { 0 };
+	};
+
 }

BAN/include/BAN/Heap.h

@@ -0,0 +1,89 @@
+#pragma once
+
+#include <BAN/Iterators.h>
+#include <BAN/Swap.h>
+#include <BAN/Traits.h>
+
+namespace BAN
+{
+
+	namespace detail
+	{
+
+		template<typename It, typename Comp>
+		void heapify_up(It begin, size_t index, Comp comp)
+		{
+			size_t parent = (index - 1) / 2;
+			while (parent < index)
+			{
+				if (comp(*(begin + index), *(begin + parent)))
+					break;
+				swap(*(begin + parent), *(begin + index));
+				index = parent;
+				parent = (index - 1) / 2;
+			}
+		}
+
+		template<typename It, typename Comp>
+		void heapify_down(It begin, size_t index, size_t len, Comp comp)
+		{
+			for (;;)
+			{
+				const size_t lchild = 2 * index + 1;
+				const size_t rchild = 2 * index + 2;
+
+				size_t child = 0;
+				if (lchild < len && !comp(*(begin + lchild), *(begin + index)))
+				{
+					if (rchild < len && !comp(*(begin + rchild), *(begin + lchild)))
+						child = rchild;
+					else
+						child = lchild;
+				}
+				else if (rchild < len && !comp(*(begin + rchild), *(begin + index)))
+					child = rchild;
+				else
+					break;
+
+				swap(*(begin + child), *(begin + index));
+				index = child;
+			}
+		}
+
+	}
+
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
+	void make_heap(It begin, It end, Comp comp = {})
+	{
+		const size_t len = distance(begin, end);
+		if (len <= 1)
+			return;
+
+		size_t index = (len - 2) / 2;
+		while (index < len)
+			detail::heapify_down(begin, index--, len, comp);
+	}
+
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
+	void push_heap(It begin, It end, Comp comp = {})
+	{
+		const size_t len = distance(begin, end);
+		detail::heapify_up(begin, len - 1, comp);
+	}
+
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
+	void pop_heap(It begin, It end, Comp comp = {})
+	{
+		const size_t len = distance(begin, end);
+		swap(*begin, *(begin + len - 1));
+		detail::heapify_down(begin, 0, len - 1, comp);
+	}
+
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
+	void sort_heap(It begin, It end, Comp comp = {})
+	{
+		while (begin != end)
+			pop_heap(begin, end--, comp);
+	}
+
+}

BAN/include/BAN/IPv4.h

@@ -9,6 +9,10 @@ namespace BAN
 
 	struct IPv4Address
 	{
+		constexpr IPv4Address()
+			: IPv4Address(0)
+		{ }
+
 		constexpr IPv4Address(uint32_t u32_address)
 		{
 			raw = u32_address;

BAN/include/BAN/MacroUtils.h

@@ -0,0 +1,46 @@
+#pragma once
+
+#define _ban_count_args_impl(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, ...) _9
+#define _ban_count_args(...) _ban_count_args_impl(__VA_ARGS__ __VA_OPT__(,) 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
+
+#define _ban_concat_impl(a, b) a##b
+#define _ban_concat(a, b) _ban_concat_impl(a, b)
+
+#define _ban_stringify_impl(x) #x
+#define _ban_stringify(x) _ban_stringify_impl(x)
+
+#define _ban_fe_0(f)
+#define _ban_fe_1(f, _0)                                 f(0, _0)
+#define _ban_fe_2(f, _0, _1)                             f(0, _0) f(1, _1)
+#define _ban_fe_3(f, _0, _1, _2)                         f(0, _0) f(1, _1) f(2, _2)
+#define _ban_fe_4(f, _0, _1, _2, _3)                     f(0, _0) f(1, _1) f(2, _2) f(3, _3)
+#define _ban_fe_5(f, _0, _1, _2, _3, _4)                 f(0, _0) f(1, _1) f(2, _2) f(3, _3) f(4, _4)
+#define _ban_fe_6(f, _0, _1, _2, _3, _4, _5)             f(0, _0) f(1, _1) f(2, _2) f(3, _3) f(4, _4) f(5, _5)
+#define _ban_fe_7(f, _0, _1, _2, _3, _4, _5, _6)         f(0, _0) f(1, _1) f(2, _2) f(3, _3) f(4, _4) f(5, _5) f(6, _6)
+#define _ban_fe_8(f, _0, _1, _2, _3, _4, _5, _6, _7)     f(0, _0) f(1, _1) f(2, _2) f(3, _3) f(4, _4) f(5, _5) f(6, _6) f(7, _7)
+#define _ban_fe_9(f, _0, _1, _2, _3, _4, _5, _6, _7, _8) f(0, _0) f(1, _1) f(2, _2) f(3, _3) f(4, _4) f(5, _5) f(6, _6) f(7, _7) f(8, _8)
+#define _ban_for_each(f, ...) _ban_concat(_ban_fe_, _ban_count_args(__VA_ARGS__))(f __VA_OPT__(,) __VA_ARGS__)
+
+#define _ban_fe_comma_0(f)
+#define _ban_fe_comma_1(f, _0)                                 f(0, _0)
+#define _ban_fe_comma_2(f, _0, _1)                             f(0, _0), f(1, _1)
+#define _ban_fe_comma_3(f, _0, _1, _2)                         f(0, _0), f(1, _1), f(2, _2)
+#define _ban_fe_comma_4(f, _0, _1, _2, _3)                     f(0, _0), f(1, _1), f(2, _2), f(3, _3)
+#define _ban_fe_comma_5(f, _0, _1, _2, _3, _4)                 f(0, _0), f(1, _1), f(2, _2), f(3, _3), f(4, _4)
+#define _ban_fe_comma_6(f, _0, _1, _2, _3, _4, _5)             f(0, _0), f(1, _1), f(2, _2), f(3, _3), f(4, _4), f(5, _5)
+#define _ban_fe_comma_7(f, _0, _1, _2, _3, _4, _5, _6)         f(0, _0), f(1, _1), f(2, _2), f(3, _3), f(4, _4), f(5, _5), f(6, _6)
+#define _ban_fe_comma_8(f, _0, _1, _2, _3, _4, _5, _6, _7)     f(0, _0), f(1, _1), f(2, _2), f(3, _3), f(4, _4), f(5, _5), f(6, _6), f(7, _7)
+#define _ban_fe_comma_9(f, _0, _1, _2, _3, _4, _5, _6, _7, _8) f(0, _0), f(1, _1), f(2, _2), f(3, _3), f(4, _4), f(5, _5), f(6, _6), f(7, _7), f(8, _8)
+#define _ban_for_each_comma(f, ...) _ban_concat(_ban_fe_comma_, _ban_count_args(__VA_ARGS__))(f __VA_OPT__(,) __VA_ARGS__)
+
+#define _ban_get_0(a0, ...)                                     a0
+#define _ban_get_1(a0, a1, ...)                                 a1
+#define _ban_get_2(a0, a1, a2, ...)                             a2
+#define _ban_get_3(a0, a1, a2, a3, ...)                         a3
+#define _ban_get_4(a0, a1, a2, a3, a4, ...)                     a4
+#define _ban_get_5(a0, a1, a2, a3, a4, a5, ...)                 a5
+#define _ban_get_6(a0, a1, a2, a3, a4, a5, a6, ...)             a6
+#define _ban_get_7(a0, a1, a2, a3, a4, a5, a6, a7, ...)         a7
+#define _ban_get_8(a0, a1, a2, a3, a4, a5, a6, a7, a8, ...)     a8
+#define _ban_get_9(a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, ...) a9
+#define _ban_get(n, ...) _ban_concat(_ban_get_, n)(__VA_ARGS__)

BAN/include/BAN/Math.h

@@ -36,12 +36,11 @@ namespace BAN::Math
 	template<integral T>
 	inline constexpr T gcd(T a, T b)
 	{
-		T t;
 		while (b)
 		{
-			t = b;
+			T temp = b;
 			b = a % b;
-			a = t;
+			a = temp;
 		}
 		return a;
 	}
@@ -66,25 +65,36 @@ namespace BAN::Math
 		return (x & (x - 1)) == 0;
 	}
 
-	template<BAN::integral T>
-	static constexpr bool will_multiplication_overflow(T a, T b)
+	template<integral T> requires(sizeof(T) <= 8)
+	inline constexpr T round_up_to_power_of_two(T x)
 	{
-		if (a == 0 || b == 0)
-			return false;
-		if ((a > 0) == (b > 0))
-			return a > BAN::numeric_limits<T>::max() / b;
-		else
-			return a < BAN::numeric_limits<T>::min() / b;
+		x--;
+		x |= x >> 1;
+		x |= x >> 2;
+		x |= x >> 4;
+		if constexpr(sizeof(T) >= 2)
+			x |= x >> 8;
+		if constexpr(sizeof(T) >= 4)
+			x |= x >> 16;
+		if constexpr(sizeof(T) >= 8)
+			x |= x >> 32;
+		return x + 1;
 	}
 
-	template<BAN::integral T>
-	static constexpr bool will_addition_overflow(T a, T b)
+	template<integral T>
+	__attribute__((always_inline))
+	inline constexpr bool will_multiplication_overflow(T a, T b)
 	{
-		if (a > 0 && b > 0)
-			return a > BAN::numeric_limits<T>::max() - b;
-		if (a < 0 && b < 0)
-			return a < BAN::numeric_limits<T>::min() - b;
-		return false;
+		T dummy;
+		return __builtin_mul_overflow(a, b, &dummy);
+	}
+
+	template<integral T>
+	__attribute__((always_inline))
+	inline constexpr bool will_addition_overflow(T a, T b)
+	{
+		T dummy;
+		return __builtin_add_overflow(a, b, &dummy);
 	}
 
 	template<typename T>
@@ -98,6 +108,19 @@ namespace BAN::Math
 		return sizeof(T) * 8 - __builtin_clzll(x) - 1;
 	}
 
+// This is ugly but my clangd does not like including
+// intrinsic headers at all
+#if !defined(__SSE__) || !defined(__SSE2__)
+#pragma GCC push_options
+#ifndef __SSE__
+#pragma GCC target("sse")
+#endif
+#ifndef __SSE2__
+#pragma GCC target("sse2")
+#endif
+#define BAN_MATH_POP_OPTIONS
+#endif
+
 	template<floating_point T>
 	inline constexpr T floor(T x)
 	{
@@ -159,7 +182,23 @@ namespace BAN::Math
 			"jne 1b;"
 			: "+t"(a)
 			: "u"(b)
-			: "ax"
+			: "ax", "cc"
+		);
+		return a;
+	}
+
+	template<floating_point T>
+	inline constexpr T remainder(T a, T b)
+	{
+		asm(
+			"1:"
+			"fprem1;"
+			"fnstsw %%ax;"
+			"testb $4, %%ah;"
+			"jne 1b;"
+			: "+t"(a)
+			: "u"(b)
+			: "ax", "cc"
 		);
 		return a;
 	}
@@ -167,7 +206,7 @@ namespace BAN::Math
 	template<floating_point T>
 	static T modf(T x, T* iptr)
 	{
-		const T frac = BAN::Math::fmod<T>(x, 1);
+		const T frac = BAN::Math::fmod<T>(x, (T)1.0);
 		*iptr = x - frac;
 		return frac;
 	}
@@ -175,15 +214,15 @@ namespace BAN::Math
 	template<floating_point T>
 	inline constexpr T frexp(T num, int* exp)
 	{
-		if (num == 0.0)
+		if (num == (T)0.0)
 		{
 			*exp = 0;
-			return 0.0;
+			return (T)0.0;
 		}
 
-		T _exp;
-		asm("fxtract" : "+t"(num), "=u"(_exp));
-		*exp = (int)_exp + 1;
+		T e;
+		asm("fxtract" : "+t"(num), "=u"(e));
+		*exp = (int)e + 1;
 		return num / (T)2.0;
 	}
 
@@ -251,6 +290,7 @@ namespace BAN::Math
 			"fstp %%st(1);"
 			: "+t"(x)
 		);
+
 		return x;
 	}
 
@@ -263,18 +303,9 @@ namespace BAN::Math
 	template<floating_point T>
 	inline constexpr T pow(T x, T y)
 	{
-		asm(
-			"fyl2x;"
-			"fld1;"
-			"fld %%st(1);"
-			"fprem;"
-			"f2xm1;"
-			"faddp;"
-			"fscale;"
-			: "+t"(x), "+u"(y)
-		);
-
-		return x;
+		if (x == (T)0.0)
+			return (T)0.0;
+		return exp2<T>(y * log2<T>(x));
 	}
 
 	template<floating_point T>
@@ -309,17 +340,28 @@ namespace BAN::Math
 
 	template<floating_point T>
 	inline constexpr T sqrt(T x)
+	{
+		if constexpr(BAN::is_same_v<T, float>)
+		{
+			using v4sf = float __attribute__((vector_size(16)));
+			return __builtin_ia32_sqrtss((v4sf) { x, 0.0f, 0.0f, 0.0f })[0];
+		}
+		else if constexpr(BAN::is_same_v<T, double>)
+		{
+			using v2df = double __attribute__((vector_size(16)));
+			return __builtin_ia32_sqrtsd((v2df) { x, 0.0 })[0];
+		}
+		else if constexpr(BAN::is_same_v<T, long double>)
 		{
 			asm("fsqrt" : "+t"(x));
 			return x;
 		}
+	}
 
 	template<floating_point T>
 	inline constexpr T cbrt(T value)
 	{
-		if (value == 0.0)
-			return 0.0;
-		return pow<T>(value, 1.0 / 3.0);
+		return pow<T>(value, (T)1.0 / (T)3.0);
 	}
 
 	template<floating_point T>
@@ -346,30 +388,21 @@ namespace BAN::Math
 	inline constexpr T tan(T x)
 	{
 		T one, ret;
-		asm(
-			"fptan"
-			: "=t"(one), "=u"(ret)
-			: "0"(x)
-		);
+		asm("fptan" : "=t"(one), "=u"(ret) : "0"(x));
 		return ret;
 	}
 
 	template<floating_point T>
 	inline constexpr T atan2(T y, T x)
 	{
-		asm(
-			"fpatan"
-			: "+t"(x)
-			: "u"(y)
-			: "st(1)"
-		);
+		asm("fpatan" : "+t"(x) : "u"(y) : "st(1)");
 		return x;
 	}
 
 	template<floating_point T>
 	inline constexpr T atan(T x)
 	{
-		return atan2<T>(x, 1.0);
+		return atan2<T>(x, (T)1.0);
 	}
 
 	template<floating_point T>
@@ -378,10 +411,10 @@ namespace BAN::Math
 		if (x == (T)0.0)
 			return (T)0.0;
 		if (x == (T)1.0)
-			return  numbers::pi_v<T> / (T)2.0;
+			return +numbers::pi_v<T> / (T)2.0;
 		if (x == (T)-1.0)
 			return -numbers::pi_v<T> / (T)2.0;
-		return (T)2.0 * atan<T>(x / (T(1.0) + sqrt<T>((T)1.0 - x * x)));
+		return (T)2.0 * atan<T>(x / ((T)1.0 + sqrt<T>((T)1.0 - x * x)));
 	}
 
 	template<floating_point T>
@@ -411,7 +444,7 @@ namespace BAN::Math
 	template<floating_point T>
 	inline constexpr T tanh(T x)
 	{
-		const T exp_px = exp<T>(x);
+		const T exp_px = exp<T>(+x);
 		const T exp_nx = exp<T>(-x);
 		return (exp_px - exp_nx) / (exp_px + exp_nx);
 	}
@@ -440,4 +473,9 @@ namespace BAN::Math
 		return sqrt<T>(x * x + y * y);
 	}
 
+#ifdef BAN_MATH_POP_OPTIONS
+#undef BAN_MATH_POP_OPTIONS
+#pragma GCC pop_options
+#endif
+
 }

BAN/include/BAN/New.h

@@ -9,10 +9,12 @@
 namespace BAN
 {
 	#if defined(__is_kernel)
-		static constexpr void*(&allocator)(size_t) = kmalloc;
-		static constexpr void(&deallocator)(void*) = kfree;
+		static constexpr void*(*allocator)(size_t) = kmalloc;
+		static constexpr void*(*reallocator)(void*, size_t) = nullptr;
+		static constexpr void(*deallocator)(void*) = kfree;
 	#else
-		static constexpr void*(&allocator)(size_t) = malloc;
-		static constexpr void(&deallocator)(void*) = free;
+		static constexpr void*(*allocator)(size_t) = malloc;
+		static constexpr void*(*reallocator)(void*, size_t) = realloc;
+		static constexpr void(*deallocator)(void*) = free;
 	#endif
 }

BAN/include/BAN/PriorityQueue.h

@@ -0,0 +1,64 @@
+#pragma once
+
+#include "BAN/Errors.h"
+#include <BAN/Vector.h>
+#include <BAN/Heap.h>
+
+namespace BAN
+{
+
+	template<typename T, typename Comp = less<T>>
+	class PriorityQueue
+	{
+	public:
+		PriorityQueue() = default;
+		PriorityQueue(Comp comp)
+			: m_comp(comp)
+		{ }
+
+		ErrorOr<void> push(const T& value)
+		{
+			TRY(m_data.push_back(value));
+			push_heap(m_data.begin(), m_data.end());
+			return {};
+		}
+
+		ErrorOr<void> push(T&& value)
+		{
+			TRY(m_data.push_back(move(value)));
+			push_heap(m_data.begin(), m_data.end());
+			return {};
+		}
+
+		template<typename... Args>
+		ErrorOr<void> emplace(Args&&... args) requires is_constructible_v<T, Args...>
+		{
+			TRY(m_data.emplace_back(forward<Args>(args)...));
+			push_heap(m_data.begin(), m_data.end());
+			return {};
+		}
+
+		void pop()
+		{
+			pop_heap(m_data.begin(), m_data.end());
+			m_data.pop_back();
+		}
+
+		BAN::ErrorOr<void> reserve(Vector<T>::size_type size)
+		{
+			return m_data.reserve(size);
+		}
+
+		T& top() { return m_data.front(); }
+		const T& top() const { return m_data.front(); }
+
+		bool empty() const { return m_data.empty(); }
+		Vector<T>::size_type size() const { return m_data.size(); }
+		Vector<T>::size_type capacity() const { return m_data.capacity(); }
+
+	private:
+		Comp m_comp;
+		Vector<T> m_data;
+	};
+
+}

BAN/include/BAN/RefPtr.h

@@ -2,9 +2,9 @@
 
 #include <BAN/Atomic.h>
 #include <BAN/Errors.h>
+#include <BAN/Hash.h>
 #include <BAN/Move.h>
 #include <BAN/NoCopyMove.h>
-#include <stdint.h>
 
 namespace BAN
 {
@@ -129,14 +129,9 @@ namespace BAN
 			return *this;
 		}
 
-		T* ptr() { ASSERT(!empty()); return m_pointer; }
-		const T* ptr() const { ASSERT(!empty()); return m_pointer; }
-
-		T& operator*() { return *ptr(); }
-		const T& operator*() const { return *ptr(); }
-
-		T* operator->() { return ptr(); }
-		const T* operator->() const { return ptr(); }
+		T* ptr() const { return m_pointer; }
+		T& operator*() const { ASSERT(!empty()); return *ptr(); }
+		T* operator->() const { ASSERT(!empty()); return ptr(); }
 
 		bool operator==(RefPtr other) const { return m_pointer == other.m_pointer; }
 		bool operator!=(RefPtr other) const { return m_pointer != other.m_pointer; }
@@ -158,4 +153,13 @@ namespace BAN
 		friend class RefPtr;
 	};
 
+	template<typename T>
+	struct hash<RefPtr<T>>
+	{
+		constexpr hash_t operator()(const RefPtr<T>& ptr) const
+		{
+			return hash<T*>()(ptr.ptr());
+		}
+	};
+
 }

BAN/include/BAN/Sort.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <BAN/Heap.h>
 #include <BAN/Math.h>
 #include <BAN/Swap.h>
 #include <BAN/Traits.h>
@@ -8,7 +9,7 @@
 namespace BAN::sort
 {
 
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void exchange_sort(It begin, It end, Comp comp = {})
 	{
 		for (It lhs = begin; lhs != end; ++lhs)
@@ -20,46 +21,67 @@ namespace BAN::sort
 	namespace detail
 	{
 
+		template<typename It>
+		struct partition_pair
+		{
+			It lt;
+			It gt;
+		};
+
 		template<typename It, typename Comp>
-		It partition(It begin, It end, Comp comp)
+		partition_pair<It> partition(It begin, It end, Comp comp)
 		{
-			It pivot = prev(end, 1);
+			It pivot = next(begin, distance(begin, end) / 2);
 
-			It it1 = begin;
-			for (It it2 = begin; it2 != pivot; ++it2)
+			It lt = begin;
+			It eq = begin;
+			It gt = end;
+
+			while (eq != gt)
 			{
-				if (comp(*it2, *pivot))
+				if (comp(*eq, *pivot))
 				{
-					swap(*it1, *it2);
-					++it1;
+					swap(*eq, *lt);
+					if (pivot == lt)
+						pivot = eq;
+					++lt;
+					++eq;
+				}
+				else if (comp(*pivot, *eq))
+				{
+					--gt;
+					swap(*eq, *gt);
+					if (pivot == gt)
+						pivot = eq;
+				}
+				else
+				{
+					++eq;
 				}
 			}
 
-			swap(*it1, *pivot);
-
-			return it1;
+			return { lt, gt };
+		}
 	}
 
-	}
-
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void quick_sort(It begin, It end, Comp comp = {})
 	{
 		if (distance(begin, end) <= 1)
 			return;
-		It mid = detail::partition(begin, end, comp);
-		quick_sort(begin, mid, comp);
-		quick_sort(++mid, end, comp);
+		const auto [lt, gt] = detail::partition(begin, end, comp);
+		quick_sort(begin, lt, comp);
+		quick_sort(gt, end, comp);
 	}
 
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void insertion_sort(It begin, It end, Comp comp = {})
 	{
 		if (distance(begin, end) <= 1)
 			return;
 		for (It it1 = next(begin, 1); it1 != end; ++it1)
 		{
-			typename It::value_type x = move(*it1);
+			auto x = move(*it1);
 			It it2 = it1;
 			for (; it2 != begin && comp(x, *prev(it2, 1)); --it2)
 				*it2 = move(*prev(it2, 1));
@@ -67,83 +89,7 @@ namespace BAN::sort
 		}
 	}
 
-	namespace detail
-	{
-
-		template<typename It, typename Comp>
-		void push_heap(It begin, size_t hole_index, size_t top_index, typename It::value_type value, Comp comp)
-		{
-			size_t parent = (hole_index - 1) / 2;
-			while (hole_index > top_index && comp(*next(begin, parent), value))
-			{
-				*next(begin, hole_index) = move(*next(begin, parent));
-				hole_index = parent;
-				parent = (hole_index - 1) / 2;
-			}
-			*next(begin, hole_index) = move(value);
-		}
-
-		template<typename It, typename Comp>
-		void adjust_heap(It begin, size_t hole_index, size_t len, typename It::value_type value, Comp comp)
-		{
-			const size_t top_index = hole_index;
-			size_t child = hole_index;
-			while (child < (len - 1) / 2)
-			{
-				child = 2 * (child + 1);
-				if (comp(*next(begin, child), *next(begin, child - 1)))
-					child--;
-				*next(begin, hole_index) = move(*next(begin, child));
-				hole_index = child;
-			}
-			if (len % 2 == 0 && child == (len - 2) / 2)
-			{
-				child = 2 * (child + 1);
-				*next(begin, hole_index) = move(*next(begin, child - 1));
-				hole_index = child - 1;
-			}
-			push_heap(begin, hole_index, top_index, move(value), comp);
-		}
-
-	}
-
-	template<typename It, typename Comp = less<typename It::value_type>>
-	void make_heap(It begin, It end, Comp comp = {})
-	{
-		const size_t len = distance(begin, end);
-		if (len <= 1)
-			return;
-
-		size_t parent = (len - 2) / 2;
-		while (true)
-		{
-			detail::adjust_heap(begin, parent, len, move(*next(begin, parent)), comp);
-
-			if (parent == 0)
-				break;
-
-			parent--;
-		}
-	}
-
-	template<typename It, typename Comp = less<typename It::value_type>>
-	void sort_heap(It begin, It end, Comp comp = {})
-	{
-		const size_t len = distance(begin, end);
-		if (len <= 1)
-			return;
-
-		size_t last = len;
-		while (last > 1)
-		{
-			last--;
-			typename It::value_type x = move(*next(begin, last));
-			*next(begin, last) = move(*begin);
-			detail::adjust_heap(begin, 0, last, move(x), comp);
-		}
-	}
-
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void heap_sort(It begin, It end, Comp comp = {})
 	{
 		make_heap(begin, end, comp);
@@ -160,14 +106,14 @@ namespace BAN::sort
 				return insertion_sort(begin, end, comp);
 			if (max_depth == 0)
 				return heap_sort(begin, end, comp);
-			It mid = detail::partition(begin, end, comp);
-			intro_sort_impl(begin, mid, max_depth - 1, comp);
-			intro_sort_impl(++mid, end, max_depth - 1, comp);
+			const auto [lt, gt] = detail::partition(begin, end, comp);
+			intro_sort_impl(begin, lt, max_depth - 1, comp);
+			intro_sort_impl(gt, end, max_depth - 1, comp);
 		}
 
 	}
 
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void intro_sort(It begin, It end, Comp comp = {})
 	{
 		const size_t len = distance(begin, end);
@@ -190,28 +136,21 @@ namespace BAN::sort
 	}
 
 	template<typename It, size_t radix = 256>
-	requires is_unsigned_v<typename It::value_type> && (radix > 0 && (radix & (radix - 1)) == 0)
-	BAN::ErrorOr<void> radix_sort(It begin, It end)
+	requires is_unsigned_v<it_value_type_t<It>> && (radix > 0 && (radix & (radix - 1)) == 0)
+	void radix_sort(It begin, It end, BAN::Span<it_value_type_t<It>> storage)
 	{
-		using value_type = typename It::value_type;
-
 		const size_t len = distance(begin, end);
 		if (len <= 1)
-			return {};
+			return;
 
-		Vector<value_type> temp;
-		TRY(temp.resize(len));
-
-		Vector<size_t> counts;
-		TRY(counts.resize(radix));
+		ASSERT(storage.size() >= len);
 
 		constexpr size_t mask  = radix - 1;
 		constexpr size_t shift = detail::lsb_index(radix);
 
-		for (size_t s = 0; s < sizeof(value_type) * 8; s += shift)
+		for (size_t s = 0; s < sizeof(it_value_type_t<It>) * 8; s += shift)
 		{
-			for (auto& cnt : counts)
-				cnt = 0;
+			size_t counts[radix] {};
 			for (It it = begin; it != end; ++it)
 				counts[(*it >> s) & mask]++;
 
@@ -221,17 +160,32 @@ namespace BAN::sort
 			for (It it = end; it != begin;)
 			{
 				--it;
-				temp[--counts[(*it >> s) & mask]] = *it;
+				storage[--counts[(*it >> s) & mask]] = *it;
 			}
 
-			for (size_t j = 0; j < temp.size(); j++)
-				*next(begin, j) = temp[j];
+			It it = begin;
+			for (size_t j = 0; j < storage.size(); j++, ++it)
+				*it = storage[j];
 		}
+	}
+
+	template<typename It, size_t radix = 256>
+	requires is_unsigned_v<it_value_type_t<It>> && (radix > 0 && (radix & (radix - 1)) == 0)
+	BAN::ErrorOr<void> radix_sort(It begin, It end)
+	{
+		const size_t len = distance(begin, end);
+		if (len <= 1)
+			return {};
+
+		Vector<it_value_type_t<It>> temp;
+		TRY(temp.resize(len));
+
+		radix_sort(begin, end, temp.span());
 
 		return {};
 	}
 
-	template<typename It, typename Comp = less<typename It::value_type>>
+	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void sort(It begin, It end, Comp comp = {})
 	{
 		return intro_sort(begin, end, comp);

BAN/include/BAN/Span.h

@@ -69,7 +69,6 @@ namespace BAN
 
 		value_type* data() const
 		{
-			ASSERT(m_data);
 			return m_data;
 		}
 
@@ -84,7 +83,6 @@ namespace BAN
 
 		Span slice(size_type start, size_type length = ~size_type(0)) const
 		{
-			ASSERT(m_data);
 			ASSERT(start <= m_size);
 			if (length == ~size_type(0))
 				length = m_size - start;

BAN/include/BAN/String.h

@@ -14,6 +14,7 @@ namespace BAN
 	{
 	public:
 		using size_type = size_t;
+		using value_type = char;
 		using iterator = IteratorSimple<char, String>;
 		using const_iterator = ConstIteratorSimple<char, String>;
 		static constexpr size_type sso_capacity = 15;
@@ -352,10 +353,9 @@ namespace BAN::Formatter
 {
 
 	template<typename F>
-	void print_argument(F putc, const String& string, const ValueFormat&)
+	void print_argument(F putc, const String& string, const ValueFormat& format)
 	{
-		for (String::size_type i = 0; i < string.size(); i++)
-			putc(string[i]);
+		print_argument(putc, string.sv(), format);
 	}
 
 }

BAN/include/BAN/StringView.h

@@ -14,6 +14,7 @@ namespace BAN
 	{
 	public:
 		using size_type = size_t;
+		using value_type = char;
 		using const_iterator = ConstIteratorSimple<char, StringView>;
 
 	public:
@@ -21,7 +22,8 @@ namespace BAN
 		constexpr StringView(const char* string, size_type len = -1)
 		{
 			if (len == size_type(-1))
-				len = strlen(string);
+				for (len = 0; string[len];)
+					len++;
 			m_data = string;
 			m_size = len;
 		}
@@ -245,10 +247,12 @@ namespace BAN::Formatter
 {
 
 	template<typename F>
-	void print_argument(F putc, const StringView& sv, const ValueFormat&)
+	void print_argument(F putc, const StringView& sv, const ValueFormat& format)
 	{
 		for (StringView::size_type i = 0; i < sv.size(); i++)
 			putc(sv[i]);
+		for (int i = sv.size(); i < format.fill; i++)
+			putc(' ');
 	}
 
 }

BAN/include/BAN/Traits.h

@@ -61,6 +61,9 @@ namespace BAN
 	template<typename T> struct is_move_constructible { static constexpr bool value = is_constructible_v<T, T&&>; };
 	template<typename T> inline constexpr bool is_move_constructible_v = is_move_constructible<T>::value;
 
+	template<typename T> struct is_trivially_copyable { static constexpr bool value = __is_trivially_copyable(T); };
+	template<typename T> inline constexpr bool is_trivially_copyable_v = is_trivially_copyable<T>::value;
+
 	template<typename T> struct is_integral { static constexpr bool value = requires (T t, T* p, void (*f)(T)) { reinterpret_cast<T>(t); f(0); p + t; }; };
 	template<typename T> inline constexpr bool is_integral_v = is_integral<T>::value;
 	template<typename T> concept integral = is_integral_v<T>;
@@ -139,6 +142,10 @@ namespace BAN
 	template<typename T> using make_signed_t = typename make_signed<T>::type;
 #undef __BAN_TRAITS_MAKE_SIGNED_CV
 
+	template<typename T> struct it_value_type { using value_type = T::value_type; };
+	template<typename T> struct it_value_type<T*> { using value_type = T; };
+	template<typename T> using it_value_type_t = typename it_value_type<T>::value_type;
+
 	template<typename T> struct less	{ constexpr bool operator()(const T& lhs, const T& rhs) const { return lhs < rhs; } };
 	template<typename T> struct equal	{ constexpr bool operator()(const T& lhs, const T& rhs) const { return lhs == rhs; } };
 	template<typename T> struct greater	{ constexpr bool operator()(const T& lhs, const T& rhs) const { return lhs > rhs; } };

BAN/include/BAN/UTF8.h

@@ -18,7 +18,7 @@ namespace BAN::UTF8
 			return 3;
 		if ((first_byte & 0xF8) == 0xF0)
 			return 4;
-		return 0;
+		return UTF8::invalid;
 	}
 
 	template<typename T> requires (sizeof(T) == 1)

BAN/include/BAN/UniqPtr.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <BAN/Errors.h>
+#include <BAN/Hash.h>
 #include <BAN/NoCopyMove.h>
 
 namespace BAN
@@ -53,32 +54,12 @@ namespace BAN
 			return *this;
 		}
 
-		T& operator*()
-		{
-			ASSERT(m_pointer);
-			return *m_pointer;
-		}
+		T* ptr() const { return m_pointer; }
+		T& operator*() const { ASSERT(!empty()); return *ptr(); }
+		T* operator->() const { ASSERT(!empty()); return ptr(); }
 
-		const T& operator*() const
-		{
-			ASSERT(m_pointer);
-			return *m_pointer;
-		}
-
-		T* operator->()
-		{
-			ASSERT(m_pointer);
-			return m_pointer;
-		}
-
-		const T* operator->() const
-		{
-			ASSERT(m_pointer);
-			return m_pointer;
-		}
-
-		T* ptr() { return m_pointer; }
-		const T* ptr() const { return m_pointer; }
+		bool empty() const { return m_pointer == nullptr; }
+		explicit operator bool() const { return m_pointer; }
 
 		void clear()
 		{
@@ -87,8 +68,6 @@ namespace BAN
 			m_pointer = nullptr;
 		}
 
-		operator bool() const { return m_pointer != nullptr; }
-
 	private:
 		T* m_pointer = nullptr;
 
@@ -96,4 +75,13 @@ namespace BAN
 		friend class UniqPtr;
 	};
 
+	template<typename T>
+	struct hash<UniqPtr<T>>
+	{
+		constexpr hash_t operator()(const UniqPtr<T>& ptr) const
+		{
+			return hash<T*>()(ptr.ptr());
+		}
+	};
+
 }

BAN/include/BAN/Variant.h

@@ -126,6 +126,7 @@ namespace BAN
 		Variant(Variant&& other)
 			: m_index(other.m_index)
 		{
+			if (other.has_value())
 				detail::move_construct<Ts...>(other.m_index, other.m_storage, m_storage);
 			other.clear();
 		}
@@ -133,6 +134,7 @@ namespace BAN
 		Variant(const Variant& other)
 			: m_index(other.m_index)
 		{
+			if (other.has_value())
 				detail::copy_construct<Ts...>(other.m_index, other.m_storage, m_storage);
 		}
 
@@ -157,11 +159,12 @@ namespace BAN
 
 		Variant& operator=(Variant&& other)
 		{
-			if (m_index == other.m_index)
+			if (m_index == other.m_index && m_index != invalid_index())
 				detail::move_assign<Ts...>(m_index, other.m_storage, m_storage);
 			else
 			{
 				clear();
+				if (other.has_value())
 					detail::move_construct<Ts...>(other.m_index, other.m_storage, m_storage);
 				m_index = other.m_index;
 			}
@@ -171,11 +174,12 @@ namespace BAN
 
 		Variant& operator=(const Variant& other)
 		{
-			if (m_index == other.m_index)
+			if (m_index == other.m_index && m_index != invalid_index())
 				detail::copy_assign<Ts...>(m_index, other.m_storage, m_storage);
 			else
 			{
 				clear();
+				if (other.has_value())
 					detail::copy_construct<Ts...>(other.m_index, other.m_storage, m_storage);
 				m_index = other.m_index;
 			}

BAN/include/BAN/Vector.h

@@ -56,7 +56,7 @@ namespace BAN
 		bool contains(const T&) const;
 
 		Span<T> span() { return Span(m_data, m_size); }
-		const Span<T> span() const { return Span(m_data, m_size); }
+		Span<const T> span() const { return Span(m_data, m_size); }
 
 		const T& operator[](size_type) const;
 		T& operator[](size_type);
@@ -381,10 +381,35 @@ namespace BAN
 	template<typename T>
 	ErrorOr<void> Vector<T>::ensure_capacity(size_type size)
 	{
+		static_assert(alignof(T) <= alignof(max_align_t), "over aligned types not supported");
+
 		if (m_capacity >= size)
 			return {};
-		size_type new_cap = BAN::Math::max<size_type>(size, m_capacity * 2);
-		T* new_data = (T*)BAN::allocator(new_cap * sizeof(T));
+
+		const size_type new_cap = BAN::Math::max<size_type>(size, m_capacity * 2);
+
+		if constexpr (BAN::is_trivially_copyable_v<T>)
+		{
+			if constexpr (BAN::reallocator)
+			{
+				auto* new_data = static_cast<T*>(BAN::reallocator(m_data, new_cap * sizeof(T)));
+				if (new_data == nullptr)
+					return Error::from_errno(ENOMEM);
+				m_data = new_data;
+			}
+			else
+			{
+				auto* new_data = static_cast<T*>(BAN::allocator(new_cap * sizeof(T)));
+				if (new_data == nullptr)
+					return Error::from_errno(ENOMEM);
+				memcpy(new_data, m_data, m_size * sizeof(T));
+				BAN::deallocator(m_data);
+				m_data = new_data;
+			}
+		}
+		else
+		{
+			auto* new_data = static_cast<T*>(BAN::allocator(new_cap * sizeof(T)));
 			if (new_data == nullptr)
 				return Error::from_errno(ENOMEM);
 			for (size_type i = 0; i < m_size; i++)
@@ -394,6 +419,8 @@ namespace BAN
 			}
 			BAN::deallocator(m_data);
 			m_data = new_data;
+		}
+
 		m_capacity = new_cap;
 		return {};
 	}

BAN/include/BAN/WeakPtr.h

@@ -20,7 +20,7 @@ namespace BAN
 	class WeakLink : public RefCounted<WeakLink<T>>
 	{
 	public:
-		RefPtr<T> try_lock()
+		RefPtr<T> try_lock() const
 		{
 #if __is_kernel
 			Kernel::SpinLockGuard _(m_weak_lock);
@@ -44,7 +44,7 @@ namespace BAN
 	private:
 		T* m_ptr;
 #if __is_kernel
-		Kernel::SpinLock m_weak_lock;
+		mutable Kernel::SpinLock m_weak_lock;
 #endif
 		friend class RefPtr<WeakLink<T>>;
 	};
@@ -99,7 +99,7 @@ namespace BAN
 			return *this;
 		}
 
-		RefPtr<T> lock()
+		RefPtr<T> lock() const
 		{
 			if (m_link)
 				return m_link->try_lock();

CMakeLists.txt

@@ -29,12 +29,12 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 # include headers of ${library} to ${target}
 function(banan_include_headers target library)
-	target_include_directories(${target} PRIVATE $<TARGET_PROPERTY:${library},SOURCE_DIR>/include)
+	target_include_directories(${target} PUBLIC $<TARGET_PROPERTY:${library},SOURCE_DIR>/include)
 endfunction()
 
 # include headers and link ${library} to ${target}
 function(banan_link_library target library)
-	target_link_libraries(${target} PRIVATE ${library})
+	target_link_libraries(${target} PUBLIC ${library})
 	banan_include_headers(${target} ${library})
 endfunction()
 

README.md

@@ -23,7 +23,7 @@ If you want to try out DOOM, you should first enter the GUI environment using th
 - [x] Basic graphical environment
   - [x] Terminal emulator
   - [x] Status bar
-  - [ ] Program launcher
+  - [x] Program launcher
   - [ ] Some nice apps
 - [x] ELF dynamic linking
 - [x] copy-on-write memory
@@ -38,10 +38,14 @@ If you want to try out DOOM, you should first enter the GUI environment using th
 - [x] PS2 keyboard (all scancode sets)
 - [x] PS2 mouse
 - [x] USB
+  - [x] xHCI
+  - [ ] EHCI
+  - [ ] OHCI
+  - [ ] UHCI
   - [x] Keyboard
   - [x] Mouse
   - [x] Mass storage
-  - [ ] Hubs
+  - [x] Hubs
   - [ ] ...
 - [ ] virtio devices (network, storage)
 
@@ -114,6 +118,8 @@ To change the bootloader you can set environment variable BANAN\_BOOTLOADER; sup
 
 To run with UEFI set environment variable BANAN\_UEFI\_BOOT=1. You will also have to set OVMF\_PATH to the correct OVMF (default */usr/share/ovmf/x64/OVMF.fd*).
 
+To build an image with no physical root filesystem, but an initrd set environment variable BANAN\_INITRD=1. This can be used when testing on hardware with unsupported USB controller.
+
 If you have corrupted your disk image or want to create new one, you can either manually delete *build/banan-os.img* and build system will automatically create you a new one or you can run the following command.
 ```sh
 ./bos image-full

kernel/CMakeLists.txt

@@ -1,11 +1,17 @@
 set(KERNEL_SOURCES
 	font/prefs.psf.o
 	kernel/ACPI/ACPI.cpp
-	kernel/ACPI/BatterySystem.cpp
 	kernel/ACPI/AML/Namespace.cpp
 	kernel/ACPI/AML/Node.cpp
 	kernel/ACPI/AML/OpRegion.cpp
+	kernel/ACPI/BatterySystem.cpp
+	kernel/ACPI/EmbeddedController.cpp
 	kernel/APIC.cpp
+	kernel/Audio/AC97/Controller.cpp
+	kernel/Audio/Controller.cpp
+	kernel/Audio/HDAudio/AudioFunctionGroup.cpp
+	kernel/Audio/HDAudio/Controller.cpp
+	kernel/Banos.cpp
 	kernel/BootInfo.cpp
 	kernel/CPUID.cpp
 	kernel/Credentials.cpp
@@ -17,8 +23,10 @@ set(KERNEL_SOURCES
 	kernel/Device/RandomDevice.cpp
 	kernel/Device/ZeroDevice.cpp
 	kernel/ELF.cpp
+	kernel/Epoll.cpp
 	kernel/Errors.cpp
 	kernel/FS/DevFS/FileSystem.cpp
+	kernel/FS/EventFD.cpp
 	kernel/FS/Ext2/FileSystem.cpp
 	kernel/FS/Ext2/Inode.cpp
 	kernel/FS/FAT/FileSystem.cpp
@@ -30,6 +38,7 @@ set(KERNEL_SOURCES
 	kernel/FS/ProcFS/Inode.cpp
 	kernel/FS/TmpFS/FileSystem.cpp
 	kernel/FS/TmpFS/Inode.cpp
+	kernel/FS/USTARModule.cpp
 	kernel/FS/VirtualFileSystem.cpp
 	kernel/GDT.cpp
 	kernel/IDT.cpp
@@ -42,6 +51,8 @@ set(KERNEL_SOURCES
 	kernel/Interruptable.cpp
 	kernel/InterruptController.cpp
 	kernel/kernel.cpp
+	kernel/Lock/SpinLock.cpp
+	kernel/Memory/ByteRingBuffer.cpp
 	kernel/Memory/DMARegion.cpp
 	kernel/Memory/FileBackedRegion.cpp
 	kernel/Memory/Heap.cpp
@@ -55,6 +66,7 @@ set(KERNEL_SOURCES
 	kernel/Networking/E1000/E1000.cpp
 	kernel/Networking/E1000/E1000E.cpp
 	kernel/Networking/IPv4Layer.cpp
+	kernel/Networking/Loopback.cpp
 	kernel/Networking/NetworkInterface.cpp
 	kernel/Networking/NetworkLayer.cpp
 	kernel/Networking/NetworkManager.cpp
@@ -71,7 +83,6 @@ set(KERNEL_SOURCES
 	kernel/Processor.cpp
 	kernel/Random.cpp
 	kernel/Scheduler.cpp
-	kernel/ThreadBlocker.cpp
 	kernel/SSP.cpp
 	kernel/Storage/ATA/AHCI/Controller.cpp
 	kernel/Storage/ATA/AHCI/Device.cpp
@@ -89,9 +100,12 @@ set(KERNEL_SOURCES
 	kernel/Terminal/FramebufferTerminal.cpp
 	kernel/Terminal/PseudoTerminal.cpp
 	kernel/Terminal/Serial.cpp
+	kernel/Terminal/TerminalDriver.cpp
+	kernel/Terminal/TextModeTerminal.cpp
 	kernel/Terminal/TTY.cpp
 	kernel/Terminal/VirtualTTY.cpp
 	kernel/Thread.cpp
+	kernel/ThreadBlocker.cpp
 	kernel/Timer/HPET.cpp
 	kernel/Timer/PIT.cpp
 	kernel/Timer/RTC.cpp
@@ -99,6 +113,7 @@ set(KERNEL_SOURCES
 	kernel/USB/Controller.cpp
 	kernel/USB/Device.cpp
 	kernel/USB/HID/HIDDriver.cpp
+	kernel/USB/HID/Joystick.cpp
 	kernel/USB/HID/Keyboard.cpp
 	kernel/USB/HID/Mouse.cpp
 	kernel/USB/Hub/HubDriver.cpp
@@ -107,6 +122,7 @@ set(KERNEL_SOURCES
 	kernel/USB/USBManager.cpp
 	kernel/USB/XHCI/Controller.cpp
 	kernel/USB/XHCI/Device.cpp
+	kernel/UserCopy.cpp
 	icxxabi.cpp
 )
 
@@ -125,6 +141,8 @@ if("${BANAN_ARCH}" STREQUAL "x86_64")
 		arch/x86_64/Signal.S
 		arch/x86_64/Syscall.S
 		arch/x86_64/Thread.S
+		arch/x86_64/User.S
+		arch/x86_64/Yield.S
 	)
 elseif("${BANAN_ARCH}" STREQUAL "i686")
 	set(KERNEL_SOURCES
@@ -135,6 +153,8 @@ elseif("${BANAN_ARCH}" STREQUAL "i686")
 		arch/i686/Signal.S
 		arch/i686/Syscall.S
 		arch/i686/Thread.S
+		arch/i686/User.S
+		arch/i686/Yield.S
 	)
 else()
 	message(FATAL_ERROR "unsupported architecure ${BANAN_ARCH}")
@@ -150,10 +170,13 @@ set(BAN_SOURCES
 set(KLIBC_SOURCES
 	klibc/ctype.cpp
 	klibc/string.cpp
+	klibc/arch/${BANAN_ARCH}/string.S
+)
 
-	# Ehhh don't do this but for now libc uses the same stuff kernel can use
-	# This won't work after libc starts using sse implemetations tho
-	../userspace/libraries/LibC/arch/${BANAN_ARCH}/string.S
+set(LIBDEFLATE_SOURCE
+	../userspace/libraries/LibDEFLATE/Compressor.cpp
+	../userspace/libraries/LibDEFLATE/Decompressor.cpp
+	../userspace/libraries/LibDEFLATE/HuffmanTree.cpp
 )
 
 set(LIBFONT_SOURCES
@@ -166,18 +189,25 @@ set(LIBINPUT_SOURCE
 	../userspace/libraries/LibInput/KeyEvent.cpp
 )
 
+set(LIBQR_SOURCE
+	../userspace/libraries/LibQR/QRCode.cpp
+)
+
 set(KERNEL_SOURCES
 	${KERNEL_SOURCES}
 	${BAN_SOURCES}
 	${KLIBC_SOURCES}
+	${LIBDEFLATE_SOURCE}
 	${LIBFONT_SOURCES}
 	${LIBINPUT_SOURCE}
+	${LIBQR_SOURCE}
 )
 
 add_executable(kernel ${KERNEL_SOURCES})
 
 target_compile_definitions(kernel PRIVATE __is_kernel)
 target_compile_definitions(kernel PRIVATE __arch=${BANAN_ARCH})
+target_compile_definitions(kernel PRIVATE LIBDEFLATE_AVOID_STACK=1)
 
 target_compile_options(kernel PRIVATE
 	-O2 -g
@@ -231,9 +261,11 @@ add_custom_command(
 
 banan_include_headers(kernel ban)
 banan_include_headers(kernel libc)
-banan_include_headers(kernel libfont)
+banan_include_headers(kernel libdeflate)
 banan_include_headers(kernel libelf)
+banan_include_headers(kernel libfont)
 banan_include_headers(kernel libinput)
+banan_include_headers(kernel libqr)
 
 banan_install_headers(kernel)
 set_target_properties(kernel PROPERTIES OUTPUT_NAME banan-os.kernel)

kernel/arch/i686/PageTable.cpp

@@ -1,7 +1,7 @@
 #include <kernel/BootInfo.h>
 #include <kernel/CPUID.h>
 #include <kernel/Lock/SpinLock.h>
-#include <kernel/Memory/kmalloc.h>
+#include <kernel/Memory/Heap.h>
 #include <kernel/Memory/PageTable.h>
 
 extern uint8_t g_kernel_start[];
@@ -16,11 +16,16 @@ extern uint8_t g_kernel_writable_end[];
 extern uint8_t g_userspace_start[];
 extern uint8_t g_userspace_end[];
 
+extern uint64_t g_boot_fast_page_pt[];
+
 namespace Kernel
 {
 
 	SpinLock PageTable::s_fast_page_lock;
 
+	constexpr uint64_t s_page_flag_mask = 0x8000000000000FFF;
+	constexpr uint64_t s_page_addr_mask = ~s_page_flag_mask;
+
 	static PageTable* s_kernel = nullptr;
 	static bool s_has_nxe = false;
 	static bool s_has_pge = false;
@@ -28,6 +33,46 @@ namespace Kernel
 
 	static paddr_t s_global_pdpte = 0;
 
+	static uint64_t* s_fast_page_pt { nullptr };
+
+	alignas(PAGE_SIZE) static uint64_t s_fast_page_pt_storage[512] {};
+
+	static paddr_t allocate_zeroed_page_aligned_page()
+	{
+		const paddr_t paddr = Heap::get().take_free_page();
+		ASSERT(paddr);
+
+		PageTable::with_fast_page(paddr, [] {
+			memset(PageTable::fast_page_as_ptr(), 0, PAGE_SIZE);
+		});
+
+		return paddr;
+	}
+
+	static void unallocate_page(paddr_t paddr)
+	{
+		Heap::get().release_page(paddr);
+	}
+
+	static uint64_t read_entry_from_table(paddr_t paddr, uint16_t entry)
+	{
+		uint64_t result;
+		PageTable::with_fast_page(paddr & s_page_addr_mask, [&result, entry] {
+			result = PageTable::fast_page_as_sized<uint64_t>(entry);
+		});
+		return result;
+	}
+
+	static uint64_t write_entry_to_table(paddr_t paddr, uint16_t entry, uint64_t value)
+	{
+		uint64_t old_value;
+		PageTable::with_fast_page(paddr & s_page_addr_mask, [&old_value, entry, value] {
+			old_value = PageTable::fast_page_as_sized<uint64_t>(entry);
+			PageTable::fast_page_as_sized<uint64_t>(entry) = value;
+		});
+		return old_value;
+	}
+
 	static inline PageTable::flags_t parse_flags(uint64_t entry)
 	{
 		using Flags = PageTable::Flags;
@@ -46,31 +91,22 @@ namespace Kernel
 		return result;
 	}
 
-	void PageTable::initialize_pre_heap()
+	void PageTable::initialize_fast_page()
+	{
+		s_fast_page_pt = g_boot_fast_page_pt;
+	}
+
+	static void detect_cpu_features()
 	{
 		if (CPUID::has_nxe())
 			s_has_nxe = true;
-
 		if (CPUID::has_pge())
 			s_has_pge = true;
-
 		if (CPUID::has_pat())
 			s_has_pat = true;
-
-		ASSERT(s_kernel == nullptr);
-		s_kernel = new PageTable();
-		ASSERT(s_kernel);
-
-		s_kernel->initialize_kernel();
-		s_kernel->initial_load();
 	}
 
-	void PageTable::initialize_post_heap()
-	{
-		// NOTE: this is no-op as our 32 bit target does not use hhdm
-	}
-
-	void PageTable::initial_load()
+	void PageTable::enable_cpu_features()
 	{
 		if (s_has_nxe)
 		{
@@ -111,8 +147,49 @@ namespace Kernel
 			"movl %%eax, %%cr0;"
 			::: "rax"
 		);
+	}
 
-		load();
+	void PageTable::initialize_and_load()
+	{
+		detect_cpu_features();
+		enable_cpu_features();
+
+		ASSERT(s_kernel == nullptr);
+		s_kernel = new PageTable();
+		ASSERT(s_kernel);
+
+		s_kernel->m_highest_paging_struct = allocate_zeroed_page_aligned_page();
+		s_kernel->map_kernel_memory();
+
+		s_global_pdpte = read_entry_from_table(s_kernel->m_highest_paging_struct, 3);
+
+		// update fast page pt
+		{
+			constexpr vaddr_t vaddr = fast_page();
+			constexpr uint16_t pdpte = (vaddr >> 30) & 0x1FF;
+			constexpr uint16_t pde   = (vaddr >> 21) & 0x1FF;
+
+			const auto get_or_allocate_entry =
+				[](paddr_t table_paddr, uint16_t entry, uint64_t flags)
+				{
+					const uint64_t value = read_entry_from_table(table_paddr, entry);
+					if (value & Flags::Present)
+						return value & s_page_addr_mask;
+
+					const paddr_t paddr = allocate_zeroed_page_aligned_page();
+					write_entry_to_table(table_paddr, entry, paddr | flags);
+					return paddr;
+				};
+
+			const paddr_t pdpt = s_kernel->m_highest_paging_struct;
+			const paddr_t pd = get_or_allocate_entry(pdpt, pdpte, Flags::Present);
+
+			const paddr_t entry_paddr = reinterpret_cast<uintptr_t>(&s_fast_page_pt_storage) - KERNEL_OFFSET + g_boot_info.kernel_paddr;
+			write_entry_to_table(pd, pde, entry_paddr | PageTable::Flags::ReadWrite | PageTable::Flags::Present);
+			s_fast_page_pt = s_fast_page_pt_storage;
+		}
+
+		s_kernel->load();
 	}
 
 	PageTable& PageTable::kernel()
@@ -126,121 +203,89 @@ namespace Kernel
 		return true;
 	}
 
-	static uint64_t* allocate_zeroed_page_aligned_page()
+	void PageTable::map_kernel_memory()
 	{
-		void* page = kmalloc(PAGE_SIZE, PAGE_SIZE, true);
-		ASSERT(page);
-		memset(page, 0, PAGE_SIZE);
-		return (uint64_t*)page;
-	}
-
-	template<typename T>
-	static paddr_t V2P(const T vaddr)
-	{
-		return (vaddr_t)vaddr - KERNEL_OFFSET + g_boot_info.kernel_paddr;
-	}
-
-	template<typename T>
-	static vaddr_t P2V(const T paddr)
-	{
-		return (paddr_t)paddr - g_boot_info.kernel_paddr + KERNEL_OFFSET;
-	}
-
-	void PageTable::initialize_kernel()
-	{
-		ASSERT(s_global_pdpte == 0);
-		s_global_pdpte = V2P(allocate_zeroed_page_aligned_page());
-
-		map_kernel_memory();
-
-		prepare_fast_page();
-
 		// Map (phys_kernel_start -> phys_kernel_end) to (virt_kernel_start -> virt_kernel_end)
-		ASSERT((vaddr_t)g_kernel_start % PAGE_SIZE == 0);
+		const vaddr_t kernel_start = reinterpret_cast<vaddr_t>(g_kernel_start);
 		map_range_at(
-			V2P(g_kernel_start),
-			(vaddr_t)g_kernel_start,
+			kernel_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
+			kernel_start,
 			g_kernel_end - g_kernel_start,
 			Flags::Present
 		);
 
 		// Map executable kernel memory as executable
+		const vaddr_t kernel_execute_start = reinterpret_cast<vaddr_t>(g_kernel_execute_start);
 		map_range_at(
-			V2P(g_kernel_execute_start),
-			(vaddr_t)g_kernel_execute_start,
+			kernel_execute_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
+			kernel_execute_start,
 			g_kernel_execute_end - g_kernel_execute_start,
 			Flags::Execute | Flags::Present
 		);
 
 		// Map writable kernel memory as writable
+		const vaddr_t kernel_writable_start = reinterpret_cast<vaddr_t>(g_kernel_writable_start);
 		map_range_at(
-			V2P(g_kernel_writable_start),
-			(vaddr_t)g_kernel_writable_start,
+			kernel_writable_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
+			kernel_writable_start,
 			g_kernel_writable_end - g_kernel_writable_start,
 			Flags::ReadWrite | Flags::Present
 		);
 
 		// Map userspace memory
+		const vaddr_t kernel_userspace_start = reinterpret_cast<vaddr_t>(g_userspace_start);
 		map_range_at(
-			V2P(g_userspace_start),
-			(vaddr_t)g_userspace_start,
+			kernel_userspace_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
+			kernel_userspace_start,
 			g_userspace_end - g_userspace_start,
 			Flags::Execute | Flags::UserSupervisor | Flags::Present
 		);
 	}
 
-	void PageTable::prepare_fast_page()
-	{
-		constexpr uint64_t pdpte = (fast_page() >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (fast_page() >> 21) & 0x1FF;
-
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-		ASSERT(pdpt[pdpte] & Flags::Present);
-
-		uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte]) & PAGE_ADDR_MASK);
-		ASSERT(!(pd[pde] & Flags::Present));
-		pd[pde] = V2P(allocate_zeroed_page_aligned_page()) | Flags::ReadWrite | Flags::Present;
-	}
-
 	void PageTable::map_fast_page(paddr_t paddr)
 	{
-		ASSERT(s_kernel);
-		ASSERT(paddr);
-
-		ASSERT(s_fast_page_lock.current_processor_has_lock());
-
-		constexpr uint64_t pdpte = (fast_page() >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (fast_page() >> 21) & 0x1FF;
-		constexpr uint64_t pte   = (fast_page() >> 12) & 0x1FF;
-
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(s_kernel->m_highest_paging_struct));
-		uint64_t* pd   = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
-		uint64_t* pt   = reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK));
-
-		ASSERT(!(pt[pte] & Flags::Present));
-		pt[pte] = paddr | Flags::ReadWrite | Flags::Present;
-
-		invalidate(fast_page(), false);
+		map_fast_page(0, paddr);
 	}
 
 	void PageTable::unmap_fast_page()
 	{
-		ASSERT(s_kernel);
+		unmap_fast_page(0);
+	}
 
+	void* PageTable::map_fast_page(size_t index, paddr_t paddr)
+	{
+		ASSERT(paddr && paddr % PAGE_SIZE == 0);
+
+		ASSERT(index < 512);
+		ASSERT(s_fast_page_pt);
+
+		if (index < reserved_fast_pages)
 			ASSERT(s_fast_page_lock.current_processor_has_lock());
+		else
+			ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
 
-		constexpr uint64_t pdpte = (fast_page() >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (fast_page() >> 21) & 0x1FF;
-		constexpr uint64_t pte   = (fast_page() >> 12) & 0x1FF;
+		ASSERT(!(s_fast_page_pt[index] & Flags::Present));
+		s_fast_page_pt[index] = paddr | Flags::ReadWrite | Flags::Present;
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(s_kernel->m_highest_paging_struct));
-		uint64_t* pd   = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
-		uint64_t* pt   = reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK));
+		void* address = reinterpret_cast<void*>(fast_page() + index * PAGE_SIZE);
+		asm volatile("invlpg (%0)" :: "r"(address));
+		return address;
+	}
 
-		ASSERT(pt[pte] & Flags::Present);
-		pt[pte] = 0;
+	void PageTable::unmap_fast_page(size_t index)
+	{
+		ASSERT(index < 512);
+		ASSERT(s_fast_page_pt);
 
-		invalidate(fast_page(), false);
+		if (index < reserved_fast_pages)
+			ASSERT(s_fast_page_lock.current_processor_has_lock());
+		else
+			ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
+
+		ASSERT((s_fast_page_pt[index] & Flags::Present));
+		s_fast_page_pt[index] = 0;
+
+		asm volatile("invlpg (%0)" :: "r"(fast_page() + index * PAGE_SIZE));
 	}
 
 	BAN::ErrorOr<PageTable*> PageTable::create_userspace()
@@ -249,25 +294,19 @@ namespace Kernel
 		PageTable* page_table = new PageTable;
 		if (page_table == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
-		page_table->map_kernel_memory();
-		return page_table;
-	}
 
-	void PageTable::map_kernel_memory()
-	{
-		ASSERT(s_kernel);
-		ASSERT(s_global_pdpte);
+		page_table->m_highest_paging_struct = allocate_zeroed_page_aligned_page();
 
-		ASSERT(m_highest_paging_struct == 0);
-		m_highest_paging_struct = V2P(kmalloc(32, 32, true));
-		ASSERT(m_highest_paging_struct);
-
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
+		PageTable::with_fast_page(page_table->m_highest_paging_struct, [] {
+			uint64_t* pdpt = &PageTable::fast_page_as<uint64_t>();
 			pdpt[0] = 0;
 			pdpt[1] = 0;
 			pdpt[2] = 0;
 			pdpt[3] = s_global_pdpte | Flags::Present;
 			static_assert(KERNEL_OFFSET == 0xC0000000);
+		});
+
+		return page_table;
 	}
 
 	PageTable::~PageTable()
@@ -275,37 +314,43 @@ namespace Kernel
 		if (m_highest_paging_struct == 0)
 			return;
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-
+		const uint64_t pdpt = m_highest_paging_struct;
 		for (uint32_t pdpte = 0; pdpte < 3; pdpte++)
 		{
-			if (!(pdpt[pdpte] & Flags::Present))
+			const uint64_t pd = read_entry_from_table(pdpt, pdpte);
+			if (!(pd & Flags::Present))
 				continue;
-			uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
 			for (uint32_t pde = 0; pde < 512; pde++)
 			{
-				if (!(pd[pde] & Flags::Present))
+				const uint64_t pt = read_entry_from_table(pd, pde);
+				if (!(pt & Flags::Present))
 					continue;
-				kfree(reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK)));
+				unallocate_page(pt & s_page_addr_mask);
 			}
-			kfree(pd);
+			unallocate_page(pd & s_page_addr_mask);
 		}
-		kfree(pdpt);
+		unallocate_page(m_highest_paging_struct);
 	}
 
 	void PageTable::load()
 	{
 		SpinLockGuard _(m_lock);
 		ASSERT(m_highest_paging_struct < 0x100000000);
-		const uint32_t pdpt_lo = m_highest_paging_struct;
-		asm volatile("movl %0, %%cr3" :: "r"(pdpt_lo));
+		asm volatile("movl %0, %%cr3" :: "r"(static_cast<uint32_t>(m_highest_paging_struct)));
 		Processor::set_current_page_table(this);
 	}
 
-	void PageTable::invalidate(vaddr_t vaddr, bool send_smp_message)
+	void PageTable::invalidate_range(vaddr_t vaddr, size_t pages, bool send_smp_message)
 	{
 		ASSERT(vaddr % PAGE_SIZE == 0);
-		asm volatile("invlpg (%0)" :: "r"(vaddr) : "memory");
+
+		const bool is_userspace = (vaddr < KERNEL_OFFSET);
+		if (is_userspace && this != &PageTable::current())
+			;
+		else if (pages >= full_tlb_flush_threshold)
+			invalidate_full_address_space(!is_userspace);
+		else for (size_t i = 0; i < pages; i++)
+			asm volatile("invlpg (%0)" :: "r"(vaddr + i * PAGE_SIZE));
 
 		if (send_smp_message)
 		{
@@ -313,13 +358,42 @@ namespace Kernel
 				.type = Processor::SMPMessage::Type::FlushTLB,
 				.flush_tlb = {
 					.vaddr      = vaddr,
-					.page_count = 1
+					.page_count = pages,
+					.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
 				}
 			});
 		}
 	}
 
-	void PageTable::unmap_page(vaddr_t vaddr, bool send_smp_message)
+	void PageTable::invalidate_full_address_space(bool global)
+	{
+		if (!global || !s_has_pge)
+		{
+			asm volatile(
+				"movl %%cr3, %%eax;"
+				"movl %%eax, %%cr3;"
+				::: "eax"
+			);
+		}
+		else
+		{
+			asm volatile(
+				"movl %%cr4, %%eax;"
+
+				"andl $~0x80, %%eax;"
+				"movl %%eax, %%cr4;"
+
+				"movl %%cr3, %%ecx;"
+				"movl %%ecx, %%cr3;"
+
+				"orl $0x80, %%eax;"
+				"movl %%eax, %%cr4;"
+				::: "eax", "ecx"
+			);
+		}
+	}
+
+	void PageTable::unmap_page(vaddr_t vaddr, bool invalidate)
 	{
 		ASSERT(vaddr);
 		ASSERT(vaddr % PAGE_SIZE == 0);
@@ -338,12 +412,14 @@ namespace Kernel
 		if (is_page_free(vaddr))
 			Kernel::panic("trying to unmap unmapped page 0x{H}", vaddr);
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-		uint64_t* pd   = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
-		uint64_t* pt   = reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK));
+		const uint64_t pdpt = m_highest_paging_struct;
+		const uint64_t pd = read_entry_from_table(pdpt, pdpte);
+		const uint64_t pt = read_entry_from_table(pd, pde);
 
-		pt[pte] = 0;
-		invalidate(vaddr, send_smp_message);
+		const uint64_t old_entry = write_entry_to_table(pt, pte, 0);
+
+		if (invalidate && (old_entry & s_page_addr_mask))
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::unmap_range(vaddr_t vaddr, size_t size)
@@ -355,17 +431,10 @@ namespace Kernel
 		SpinLockGuard _(m_lock);
 		for (vaddr_t page = 0; page < page_count; page++)
 			unmap_page(vaddr + page * PAGE_SIZE, false);
-
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = page_count
-			}
-		});
+		invalidate_range(vaddr, page_count, true);
 	}
 
-	void PageTable::map_page_at(paddr_t paddr, vaddr_t vaddr, flags_t flags, MemoryType memory_type, bool send_smp_message)
+	void PageTable::map_page_at(paddr_t paddr, vaddr_t vaddr, flags_t flags, MemoryType memory_type, bool invalidate)
 	{
 		ASSERT(vaddr);
 		ASSERT(vaddr != fast_page());
@@ -400,25 +469,32 @@ namespace Kernel
 
 		SpinLockGuard _(m_lock);
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-		if (!(pdpt[pdpte] & Flags::Present))
-			pdpt[pdpte] = V2P(allocate_zeroed_page_aligned_page()) | Flags::Present;
+		const uint64_t pdpt = m_highest_paging_struct;
 
-		uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
-		if ((pd[pde] & uwr_flags) != uwr_flags)
+		uint64_t pd = read_entry_from_table(pdpt, pdpte);
+		if (!(pd & Flags::Present))
 		{
-			if (!(pd[pde] & Flags::Present))
-				pd[pde] = V2P(allocate_zeroed_page_aligned_page());
-			pd[pde] |= uwr_flags;
+			pd = allocate_zeroed_page_aligned_page();
+			pd |= Flags::Present;
+			write_entry_to_table(pdpt, pdpte, pd);
+		}
+
+		uint64_t pt = read_entry_from_table(pd, pde);
+		if ((pt & uwr_flags) != uwr_flags)
+		{
+			if (!(pt & Flags::Present))
+				pt = allocate_zeroed_page_aligned_page();
+			pt |= uwr_flags;
+			write_entry_to_table(pd, pde, pt);
 		}
 
 		if (!(flags & Flags::Present))
 			uwr_flags &= ~Flags::Present;
 
-		uint64_t* pt = reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK));
-		pt[pte] = paddr | uwr_flags | extra_flags;
+		const uint64_t old_entry = write_entry_to_table(pt, pte, paddr | uwr_flags | extra_flags);
 
-		invalidate(vaddr, send_smp_message);
+		if (invalidate && (old_entry & s_page_addr_mask))
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::map_range_at(paddr_t paddr, vaddr_t vaddr, size_t size, flags_t flags, MemoryType memory_type)
@@ -432,14 +508,54 @@ namespace Kernel
 		SpinLockGuard _(m_lock);
 		for (size_t page = 0; page < page_count; page++)
 			map_page_at(paddr + page * PAGE_SIZE, vaddr + page * PAGE_SIZE, flags, memory_type, false);
-
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = page_count
+		invalidate_range(vaddr, page_count, true);
 	}
-		});
+
+	void PageTable::remove_writable_from_range(vaddr_t vaddr, size_t size)
+	{
+		ASSERT(vaddr);
+		ASSERT(vaddr % PAGE_SIZE == 0);
+
+		uint32_t pdpte = (vaddr >> 30) & 0x1FF;
+		uint32_t pde   = (vaddr >> 21) & 0x1FF;
+		uint32_t pte   = (vaddr >> 12) & 0x1FF;
+
+		const uint32_t e_pdpte = ((vaddr + size - 1) >> 30) & 0x1FF;
+		const uint32_t e_pde   = ((vaddr + size - 1) >> 21) & 0x1FF;
+		const uint32_t e_pte   = ((vaddr + size - 1) >> 12) & 0x1FF;
+
+		SpinLockGuard _0(m_lock);
+
+		SpinLockGuard _1(s_fast_page_lock);
+
+		const uint64_t* pdpt = static_cast<uint64_t*>(map_fast_page(0, m_highest_paging_struct));
+		for (; pdpte <= e_pdpte; pdpte++)
+		{
+			if (!(pdpt[pdpte] & Flags::Present))
+				continue;
+			const uint64_t* pd = static_cast<uint64_t*>(map_fast_page(1, pdpt[pdpte] & s_page_addr_mask));
+			for (; pde < 512; pde++)
+			{
+				if (pdpte == e_pdpte && pde > e_pde)
+					break;
+				if (!(pd[pde] & Flags::ReadWrite))
+					continue;
+				uint64_t* pt = static_cast<uint64_t*>(map_fast_page(2, pd[pde] & s_page_addr_mask));
+				for (; pte < 512; pte++)
+				{
+					if (pdpte == e_pdpte && pde == e_pde && pte > e_pte)
+						break;
+					pt[pte] &= ~static_cast<uint64_t>(Flags::ReadWrite);
+				}
+				unmap_fast_page(2);
+				pte = 0;
+			}
+			unmap_fast_page(1);
+			pde = 0;
+		}
+		unmap_fast_page(0);
+
+		invalidate_range(vaddr, size / PAGE_SIZE, true);
 	}
 
 	uint64_t PageTable::get_page_data(vaddr_t vaddr) const
@@ -452,19 +568,17 @@ namespace Kernel
 
 		SpinLockGuard _(m_lock);
 
-		uint64_t* pdpt = (uint64_t*)P2V(m_highest_paging_struct);
-		if (!(pdpt[pdpte] & Flags::Present))
+		const uint64_t pdpt = m_highest_paging_struct;
+
+		const uint64_t pd = read_entry_from_table(pdpt, pdpte);
+		if (!(pd & Flags::Present))
 			return 0;
 
-		uint64_t* pd = (uint64_t*)P2V(pdpt[pdpte] & PAGE_ADDR_MASK);
-		if (!(pd[pde] & Flags::Present))
+		const uint64_t pt = read_entry_from_table(pd, pde);
+		if (!(pt & Flags::Present))
 			return 0;
 
-		uint64_t* pt = (uint64_t*)P2V(pd[pde] & PAGE_ADDR_MASK);
-		if (!(pt[pte] & Flags::Used))
-			return 0;
-
-		return pt[pte];
+		return read_entry_from_table(pt, pte);
 	}
 
 	PageTable::flags_t PageTable::get_page_flags(vaddr_t vaddr) const
@@ -474,8 +588,7 @@ namespace Kernel
 
 	paddr_t PageTable::physical_address_of(vaddr_t vaddr) const
 	{
-		uint64_t page_data = get_page_data(vaddr);
-		return (page_data & PAGE_ADDR_MASK) & ~(1ull << 63);
+		return get_page_data(vaddr) & s_page_addr_mask;
 	}
 
 	bool PageTable::is_page_free(vaddr_t vaddr) const
@@ -496,28 +609,24 @@ namespace Kernel
 		return true;
 	}
 
-	bool PageTable::reserve_page(vaddr_t vaddr, bool only_free)
+	void PageTable::reserve_page(vaddr_t vaddr)
 	{
-		SpinLockGuard _(m_lock);
 		ASSERT(vaddr % PAGE_SIZE == 0);
-		if (only_free && !is_page_free(vaddr))
-			return false;
-		map_page_at(0, vaddr, Flags::Reserved);
-		return true;
+		SpinLockGuard _(m_lock);
+		ASSERT(is_page_free(vaddr));
+		map_page_at(0, vaddr, Flags::Reserved, MemoryType::Normal, false);
 	}
 
-	bool PageTable::reserve_range(vaddr_t vaddr, size_t bytes, bool only_free)
+	void PageTable::reserve_range(vaddr_t vaddr, size_t bytes)
 	{
 		if (size_t rem = bytes % PAGE_SIZE)
 			bytes += PAGE_SIZE - rem;
 		ASSERT(vaddr % PAGE_SIZE == 0);
 
 		SpinLockGuard _(m_lock);
-		if (only_free && !is_range_free(vaddr, bytes))
-			return false;
+		ASSERT(is_range_free(vaddr, bytes));
 		for (size_t offset = 0; offset < bytes; offset += PAGE_SIZE)
 			reserve_page(vaddr + offset);
-		return true;
 	}
 
 	vaddr_t PageTable::reserve_free_page(vaddr_t first_address, vaddr_t last_address)
@@ -529,61 +638,73 @@ namespace Kernel
 		if (size_t rem = last_address % PAGE_SIZE)
 			last_address -= rem;
 
-		const uint32_t s_pdpte = (first_address >> 30) & 0x1FF;
-		const uint32_t s_pde   = (first_address >> 21) & 0x1FF;
-		const uint32_t s_pte   = (first_address >> 12) & 0x1FF;
+		uint32_t pdpte = (first_address >> 30) & 0x1FF;
+		uint32_t pde   = (first_address >> 21) & 0x1FF;
+		uint32_t pte   = (first_address >> 12) & 0x1FF;
 
-		const uint32_t e_pdpte = (last_address >> 30) & 0x1FF;
-		const uint32_t e_pde   = (last_address >> 21) & 0x1FF;
-		const uint32_t e_pte   = (last_address >> 12) & 0x1FF;
+		const uint32_t e_pdpte = ((last_address - 1) >> 30) & 0x1FF;
+		const uint32_t e_pde   = ((last_address - 1) >> 21) & 0x1FF;
+		const uint32_t e_pte   = ((last_address - 1) >> 12) & 0x1FF;
 
 		SpinLockGuard _(m_lock);
 
+		auto state = s_fast_page_lock.lock();
+
 		// Try to find free page that can be mapped without
 		// allocations (page table with unused entries)
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-		for (uint32_t pdpte = s_pdpte; pdpte < 4; pdpte++)
+		const uint64_t* pdpt = static_cast<uint64_t*>(map_fast_page(0, m_highest_paging_struct));
+		for (; pdpte <= e_pdpte; pdpte++)
 		{
-			if (pdpte > e_pdpte)
-				break;
 			if (!(pdpt[pdpte] & Flags::Present))
 				continue;
-			uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
-			for (uint32_t pde = s_pde; pde < 512; pde++)
+			const uint64_t* pd = static_cast<uint64_t*>(map_fast_page(1, pdpt[pdpte] & s_page_addr_mask));
+			for (; pde < 512; pde++)
 			{
 				if (pdpte == e_pdpte && pde > e_pde)
 					break;
 				if (!(pd[pde] & Flags::Present))
 					continue;
-				uint64_t* pt = (uint64_t*)P2V(pd[pde] & PAGE_ADDR_MASK);
-				for (uint32_t pte = s_pte; pte < 512; pte++)
+				const uint64_t* pt = static_cast<uint64_t*>(map_fast_page(2, pd[pde] & s_page_addr_mask));
+				for (; pte < 512; pte++)
 				{
-					if (pdpte == e_pdpte && pde == e_pde && pte >= e_pte)
+					if (pdpte == e_pdpte && pde == e_pde && pte > e_pte)
 						break;
-					if (!(pt[pte] & Flags::Used))
-					{
+					if (pt[pte] & Flags::Used)
+						continue;
+
+					unmap_fast_page(2);
+					unmap_fast_page(1);
+					unmap_fast_page(0);
+					s_fast_page_lock.unlock(state);
+
 					vaddr_t vaddr = 0;
 					vaddr |= (vaddr_t)pdpte << 30;
 					vaddr |= (vaddr_t)pde   << 21;
 					vaddr |= (vaddr_t)pte   << 12;
-						ASSERT(reserve_page(vaddr));
+					reserve_page(vaddr);
 					return vaddr;
 				}
+				unmap_fast_page(2);
+				pte = 0;
 			}
+			unmap_fast_page(1);
+			pde = 0;
 		}
-		}
+		unmap_fast_page(0);
+
+		s_fast_page_lock.unlock(state);
 
 		// Find any free page
 		for (vaddr_t vaddr = first_address; vaddr < last_address; vaddr += PAGE_SIZE)
 		{
 			if (is_page_free(vaddr))
 			{
-				ASSERT(reserve_page(vaddr));
+				reserve_page(vaddr);
 				return vaddr;
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	vaddr_t PageTable::reserve_free_contiguous_pages(size_t page_count, vaddr_t first_address, vaddr_t last_address)
@@ -611,12 +732,12 @@ namespace Kernel
 			}
 			if (valid)
 			{
-				ASSERT(reserve_range(vaddr, page_count * PAGE_SIZE));
+				reserve_range(vaddr, page_count * PAGE_SIZE);
 				return vaddr;
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	static void dump_range(vaddr_t start, vaddr_t end, PageTable::flags_t flags)
@@ -634,12 +755,14 @@ namespace Kernel
 
 	void PageTable::debug_dump()
 	{
-		SpinLockGuard _(m_lock);
+		SpinLockGuard _0(m_lock);
 
 		flags_t flags = 0;
 		vaddr_t start = 0;
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
+		SpinLockGuard _1(s_fast_page_lock);
+
+		const uint64_t* pdpt = static_cast<uint64_t*>(map_fast_page(0, m_highest_paging_struct));
 		for (uint32_t pdpte = 0; pdpte < 4; pdpte++)
 		{
 			if (!(pdpt[pdpte] & Flags::Present))
@@ -648,7 +771,7 @@ namespace Kernel
 				start = 0;
 				continue;
 			}
-			uint64_t* pd = (uint64_t*)P2V(pdpt[pdpte] & PAGE_ADDR_MASK);
+			const uint64_t* pd = static_cast<uint64_t*>(map_fast_page(1, pdpt[pdpte] & s_page_addr_mask));
 			for (uint64_t pde = 0; pde < 512; pde++)
 			{
 				if (!(pd[pde] & Flags::Present))
@@ -657,7 +780,7 @@ namespace Kernel
 					start = 0;
 					continue;
 				}
-				uint64_t* pt = (uint64_t*)P2V(pd[pde] & PAGE_ADDR_MASK);
+				const uint64_t* pt = static_cast<uint64_t*>(map_fast_page(2, pd[pde] & s_page_addr_mask));
 				for (uint64_t pte = 0; pte < 512; pte++)
 				{
 					if (parse_flags(pt[pte]) != flags)
@@ -675,8 +798,11 @@ namespace Kernel
 						start = (pdpte << 30) | (pde << 21) | (pte << 12);
 					}
 				}
+				unmap_fast_page(2);
 			}
+			unmap_fast_page(1);
 		}
+		unmap_fast_page(0);
 	}
 
 }

kernel/arch/i686/Signal.S

@@ -1,35 +1,86 @@
 .section .userspace, "ax"
 
 // stack contains
-//    return address
-//    signal number
-//    signal handler
+//    (4 bytes)  return address (on return stack)
+//    (4 bytes)  return stack
+//    (4 bytes)  return rflags
+//    (8 bytes)  restore sigmask
+//    (36 bytes) siginfo_t
+//    (4 bytes)  signal number
+//    (4 bytes)  signal handler
 
 .global signal_trampoline
 signal_trampoline:
+	pushl %esi // gregs
+	pushl %edi
+	pushl %edx
+	pushl %ecx
+	pushl %ebx
+	pushl %eax
 	pushl %ebp
-	movl %esp, %ebp
 
-	pusha
+	movl 84(%esp), %eax
+	pushl %eax; addl $4, (%esp)
+	pushl (%eax)
 
-	movl 40(%esp), %edi
-	movl 36(%esp), %eax
+	// FIXME: populate these
+	xorl %eax, %eax
+	pushl %eax // stack
+	pushl %eax
+	pushl %eax
+	pushl %eax // sigset
+	pushl %eax
+	pushl %eax // link
+
+	movl    %esp,  %edx // ucontext
+	leal 68(%esp), %esi // siginfo
+	movl 64(%esp), %edi // signal number
+	movl 60(%esp), %eax // handlers
 
 	// align stack to 16 bytes
-	movl %esp, %ebx
-	andl $0x0F, %ebx
-	subl %ebx, %esp
+	movl %esp, %ebp
+	andl $-16, %esp
 
-	subl $12, %esp
+	subl $512, %esp
+	fxsave (%esp)
+
+	subl $4, %esp
+	pushl %edx
+	pushl %esi
 	pushl %edi
 	call *%eax
 	addl $16, %esp
 
-	// restore stack
-	addl %ebx, %esp
-	popa
+	fxrstor (%esp)
+	addl $512, %esp
 
-	leave
+	// restore stack
+	movl %ebp, %esp
+	addl $24, %esp
+
+	// restore sigmask
+	movl $79, %eax      // SYS_SIGPROCMASK
+	movl $3,  %ebx      // SIG_SETMASK
+	leal 72(%esp), %ecx // set
+	xorl %edx, %edx     // oset
+	int $0xF0
+
+	// restore registers
 	addl $8, %esp
+	popl %ebp
+	popl %eax
+	popl %ebx
+	popl %ecx
+	popl %edx
+	popl %edi
+	popl %esi
+
+	// skip handler, number, siginfo_t, sigmask
+	addl $52, %esp
+
+	// restore flags
+	popf
+
+	movl (%esp), %esp
 
 	ret

kernel/arch/i686/Syscall.S

@@ -1,12 +1,6 @@
 // arguments in EAX, EBX, ECX, EDX, ESI, EDI
 .global asm_syscall_handler
 asm_syscall_handler:
-	# save segment registers
-	pushw %ds
-	pushw %es
-	pushw %fs
-	pushw %gs
-
 	# save general purpose registers
 	pushl %ebx
 	pushl %ecx
@@ -18,13 +12,10 @@ asm_syscall_handler:
 
 	# align stack
 	movl %esp, %ebp
-	subl $15, %esp
-	andl $0xFFFFFFF0, %esp
+	andl $-16, %esp
 
 	# push arguments
-	subl $4, %esp
-	pushl %ebp
-	addl $32, (%esp)
+	subl $8, %esp
 	pushl %edi
 	pushl %esi
 	pushl %edx
@@ -44,6 +35,15 @@ asm_syscall_handler:
 
 	movl %ebp, %esp
 
+	# restore userspace segments
+	movw $(0x20 | 3), %bx
+	movw %bx, %ds
+	movw %bx, %es
+	movw $(0x30 | 3), %bx
+	movw %bx, %fs
+	movw $(0x38 | 3), %bx
+	movw %bx, %gs
+
 	# restore general purpose registers
 	popl %ebp
 	popl %esi
@@ -52,12 +52,6 @@ asm_syscall_handler:
 	popl %ecx
 	popl %ebx
 
-	# restore segment registers
-	popw %gs
-	popw %fs
-	popw %es
-	popw %ds
-
 	iret
 
 .global sys_fork_trampoline
@@ -69,7 +63,7 @@ sys_fork_trampoline:
 
 	call read_ip
 	testl %eax, %eax
-	jz .reload_stack
+	jz .done
 
 	movl %esp, %ebx
 
@@ -85,9 +79,3 @@ sys_fork_trampoline:
 	popl %ebx
 	popl %ebp
 	ret
-
- .reload_stack:
-	call get_thread_start_sp
-	movl %eax, %esp
-	xorl %eax, %eax
-	jmp .done

kernel/arch/i686/Thread.S

@@ -7,9 +7,6 @@ read_ip:
 # void start_kernel_thread()
 .global start_kernel_thread
 start_kernel_thread:
-	call get_thread_start_sp
-	movl %eax, %esp
-
 	# STACK LAYOUT
 	#   on_exit arg
 	#   on_exit func
@@ -31,25 +28,15 @@ start_kernel_thread:
 	subl $12, %esp
 	pushl %edi
 	call *%esi
-	addl $16, %esp
-
 
 .global start_userspace_thread
 start_userspace_thread:
-	call load_thread_sse
-
-	call get_thread_start_sp
-	movl %eax, %esp
-
-	# ds, es = user data
 	movw $(0x20 | 3), %bx
 	movw %bx, %ds
 	movw %bx, %es
-	# gs = thread local
 	movw $(0x30 | 3), %bx
-	movw %bx, %gs
-	# fs = 0
-	xorw %bx, %bx
 	movw %bx, %fs
+	movw $(0x38 | 3), %bx
+	movw %bx, %gs
 
 	iret

kernel/arch/i686/User.S

@@ -0,0 +1,54 @@
+# bool safe_user_memcpy(void*, const void*, size_t)
+.global safe_user_memcpy
+.global safe_user_memcpy_end
+.global safe_user_memcpy_fault
+safe_user_memcpy:
+	xorl %eax, %eax
+	xchgl 4(%esp), %edi
+	xchgl 8(%esp), %esi
+	movl 12(%esp), %ecx
+	movl %edi, %edx
+	rep movsb
+	movl 4(%esp), %edi
+	movl 8(%esp), %esi
+	incl %eax
+safe_user_memcpy_fault:
+	ret
+safe_user_memcpy_end:
+
+# bool safe_user_strncpy(void*, const void*, size_t)
+.global safe_user_strncpy
+.global safe_user_strncpy_end
+.global safe_user_strncpy_fault
+safe_user_strncpy:
+	xchgl 4(%esp), %edi
+	xchgl 8(%esp), %esi
+	movl 12(%esp), %ecx
+
+	testl %ecx, %ecx
+	jz safe_user_strncpy_fault
+
+ .safe_user_strncpy_loop:
+	movb (%esi), %al
+	movb %al, (%edi)
+	testb %al, %al
+	jz .safe_user_strncpy_done
+
+	incl %edi
+	incl %esi
+	decl %ecx
+	jnz .safe_user_strncpy_loop
+
+ safe_user_strncpy_fault:
+	xorl %eax, %eax
+	jmp .safe_user_strncpy_return
+
+ .safe_user_strncpy_done:
+	movl $1, %eax
+
+ .safe_user_strncpy_return:
+	movl 4(%esp), %edi
+	movl 8(%esp), %esi
+	ret
+
+safe_user_strncpy_end:

kernel/arch/i686/Yield.S

@@ -0,0 +1,25 @@
+.global asm_yield_trampoline
+asm_yield_trampoline:
+	leal 4(%esp), %ecx
+	movl 4(%esp), %esp
+
+	pushl -4(%ecx)
+	pushl %ecx
+	pushl %eax
+	pushl %ebx
+	pushl %esi
+	pushl %edi
+	pushl %ebp
+
+	pushl %esp
+	call scheduler_on_yield
+	addl $4, %esp
+
+	popl %ebp
+	popl %edi
+	popl %esi
+	popl %ebx
+	popl %eax
+	movl 4(%esp), %ecx
+	movl 0(%esp), %esp
+	jmp *%ecx

kernel/arch/i686/boot.S

@@ -11,9 +11,28 @@
 
 .code32
 
-# multiboot2 header
+// video mode info, page align modules
+.set multiboot_flags, (1 << 2) | (1 << 0)
+
 .section .multiboot, "aw"
-	.align 8
+multiboot_start:
+	.long 0x1BADB002
+	.long multiboot_flags
+	.long -(0x1BADB002 + multiboot_flags)
+
+	.long 0
+	.long 0
+	.long 0
+	.long 0
+	.long 0
+
+	.long 0
+	.long FB_WIDTH
+	.long FB_HEIGHT
+	.long FB_BPP
+multiboot_end:
+
+.section .multiboot2, "aw"
 multiboot2_start:
 	.long 0xE85250D6
 	.long 0
@@ -36,6 +55,12 @@ multiboot2_start:
 	.long 12
 	.long V2P(_start)
 
+	# page align modules
+	.align 8
+	.short 6
+	.short 0
+	.long 8
+
 	.align 8
 	.short 0
 	.short 0
@@ -43,7 +68,6 @@ multiboot2_start:
 multiboot2_end:
 
 .section .bananboot, "aw"
-	.align 8
 bananboot_start:
 	.long 0xBABAB007
 	.long -(0xBABAB007 + FB_WIDTH + FB_HEIGHT + FB_BPP)
@@ -53,10 +77,10 @@ bananboot_start:
 bananboot_end:
 
 .section .bss, "aw", @nobits
-	.align 4096
-	boot_stack_bottom:
+    .global g_boot_stack_top
+	g_boot_stack_bottom:
 		.skip 4096 * 4
-	boot_stack_top:
+	g_boot_stack_top:
 
 	.global g_kernel_cmdline
 	g_kernel_cmdline:
@@ -74,8 +98,7 @@ bananboot_end:
 boot_pdpt:
 	.long V2P(boot_pd) + (PG_PRESENT)
 	.long 0
-	.quad 0
-	.quad 0
+	.skip 2 * 8
 	.long V2P(boot_pd) + (PG_PRESENT)
 	.long 0
 .align 4096
@@ -88,13 +111,16 @@ boot_pd:
 	.endr
 boot_pts:
 	.set i, 0
-	.rept 512
+	.rept 511
 		.rept 512
 			.long i + (PG_READ_WRITE | PG_PRESENT)
 			.long 0
 			.set i, i + 0x1000
 		.endr
 	.endr
+.global g_boot_fast_page_pt
+g_boot_fast_page_pt:
+	.skip 512 * 8
 
 boot_gdt:
 	.quad 0x0000000000000000 # null descriptor
@@ -161,6 +187,13 @@ enable_sse:
 	movl %eax, %cr4
 	ret
 
+enable_tsc:
+	# allow userspace to use RDTSC
+	movl %cr4, %ecx
+	andl $0xFFFFFFFB, %ecx
+	movl %ecx, %cr4
+	ret
+
 initialize_paging:
 	# enable PAE
 	movl %cr4, %ecx
@@ -188,7 +221,7 @@ _start:
 	movl %ebx, V2P(bootloader_info)
 
 	# load boot stack
-	movl $V2P(boot_stack_top), %esp
+	movl $V2P(g_boot_stack_top), %esp
 
 	# load boot GDT
 	lgdt V2P(boot_gdtr)
@@ -203,10 +236,11 @@ gdt_flush:
 	# do processor initialization
 	call check_requirements
 	call enable_sse
+	call enable_tsc
 	call initialize_paging
 
 	# load higher half stack pointer
-	movl $boot_stack_top, %esp
+	movl $g_boot_stack_top, %esp
 
 	# jump to higher half
 	leal higher_half, %ecx
@@ -242,7 +276,7 @@ system_halt:
 	jmp 1b
 
 
-#define AP_V2P(vaddr) ((vaddr) - ap_trampoline + 0xF000)
+#define AP_REL(vaddr) ((vaddr) - ap_trampoline + 0xF000)
 
 .section .ap_init, "ax"
 
@@ -252,21 +286,27 @@ ap_trampoline:
 	jmp 1f
 
 .align 8
-ap_stack_ptr:
+ap_stack_paddr:
+	.skip 4
+ap_stack_vaddr:
+	.skip 4
+ap_prepare_paging:
+	.skip 4
+ap_page_table:
+	.skip 4
+ap_ready:
 	.skip 4
-ap_stack_loaded:
-	.skip 1
 
 1:	cli; cld
-	ljmpl $0x00, $AP_V2P(ap_cs_clear)
+	ljmpl $0x00, $AP_REL(ap_cs_clear)
 
 ap_cs_clear:
 	# load ap gdt and enter protected mode
-	lgdt AP_V2P(ap_gdtr)
+	lgdt AP_REL(ap_gdtr)
 	movl %cr0, %eax
 	orb $1, %al
 	movl %eax, %cr0
-	ljmpl $0x08, $AP_V2P(ap_protected_mode)
+	ljmpl $0x08, $AP_REL(ap_protected_mode)
 
 .code32
 ap_protected_mode:
@@ -275,32 +315,36 @@ ap_protected_mode:
 	movw %ax, %ss
 	movw %ax, %es
 
-	movl AP_V2P(ap_stack_ptr), %esp
-	movb $1, AP_V2P(ap_stack_loaded)
+	movl AP_REL(ap_stack_paddr), %esp
 
 	leal V2P(enable_sse),        %ecx; call *%ecx
+	leal V2P(enable_tsc),        %ecx; call *%ecx
 	leal V2P(initialize_paging), %ecx; call *%ecx
 
 	# load boot gdt and enter long mode
 	lgdt V2P(boot_gdtr)
-	ljmpl $0x08, $AP_V2P(ap_flush_gdt)
+	ljmpl $0x08, $AP_REL(ap_flush_gdt)
 
 ap_flush_gdt:
-	# move stack pointer to higher half
-	movl %esp, %esp
-	addl $KERNEL_OFFSET, %esp
-
-	# jump to higher half
-	leal ap_higher_half, %ecx
+	movl $ap_higher_half, %ecx
 	jmp *%ecx
 
 ap_higher_half:
+	movl AP_REL(ap_prepare_paging), %eax
+	call *%eax
+
+	# load AP's initial values
+	movl AP_REL(ap_stack_vaddr), %esp
+	movl AP_REL(ap_page_table), %eax
+	movl $1, AP_REL(ap_ready)
+	movl %eax, %cr3
+
 	# clear rbp for stacktrace
 	xorl %ebp, %ebp
 
 1:	pause
 	cmpb $0, g_ap_startup_done
-	jz 1b
+	je 1b
 
 	lock incb g_ap_running_count
 

kernel/arch/i686/interrupts.S

@@ -1,66 +1,67 @@
-.macro push_userspace
-	pushw %gs
-	pushw %fs
-	pushw %es
-	pushw %ds
+.macro intr_header, n
 	pushal
-.endm
-
-.macro load_kernel_segments
+	testb $3, \n+8*4(%esp)
+	jz 1f
 	movw $0x10, %ax
 	movw %ax, %ds
 	movw %ax, %es
 	movw %ax, %fs
-
 	movw $0x28, %ax
 	movw %ax, %gs
+1:	cld
 .endm
 
-.macro pop_userspace
-	popal
-	popw %ds
-	popw %es
-	popw %fs
-	popw %gs
+.macro intr_footer, n
+	testb $3, \n+8*4(%esp)
+	jz 1f
+	call cpp_check_signal
+	movw $(0x20 | 3), %bx
+	movw %bx, %ds
+	movw %bx, %es
+	movw $(0x30 | 3), %bx
+	movw %bx, %fs
+	movw $(0x38 | 3), %bx
+	movw %bx, %gs
+1:	popal
 .endm
 
 isr_stub:
-	push_userspace
-	load_kernel_segments
-	cld
-
+	intr_header 12
 	movl %cr0, %eax; pushl %eax
 	movl %cr2, %eax; pushl %eax
 	movl %cr3, %eax; pushl %eax
 	movl %cr4, %eax; pushl %eax
 
-	movl %esp,     %eax // register ptr
-	leal 64(%esp), %ebx // interrupt stack ptr
-	movl 60(%esp), %ecx // error code
-	movl 56(%esp), %edx // isr number
+	movl 48(%esp), %edi // isr number
+	movl 52(%esp), %esi // error code
+	leal 56(%esp), %edx // interrupt stack ptr
+	movl %esp,     %ecx // register ptr
+
+	# stack frame for stack trace
+	leal 56(%esp), %eax
+	pushl (%eax)
+	pushl %ebp
 
 	movl %esp, %ebp
 	andl $-16, %esp
 
-	pushl %eax
-	pushl %ebx
 	pushl %ecx
 	pushl %edx
+	pushl %esi
+	pushl %edi
 	call cpp_isr_handler
 
 	movl %ebp, %esp
-	addl $16, %esp
+	addl $24, %esp
 
-	pop_userspace
+	intr_footer 12
 	addl $8, %esp
 	iret
 
 irq_stub:
-	push_userspace
-	load_kernel_segments
-	cld
+	intr_header 12
 
-	movl 40(%esp), %edi # interrupt number
+	movl 32(%esp), %edi # interrupt number
 
 	movl %esp, %ebp
 	andl $-16, %esp
@@ -71,37 +72,13 @@ irq_stub:
 
 	movl %ebp, %esp
 
-	pop_userspace
+	intr_footer 12
 	addl $8, %esp
 	iret
 
-.global asm_yield_handler
-asm_yield_handler:
-	# This can only be called from kernel, so no segment saving is needed
-	pushal
-	cld
-
-	leal 32(%esp), %edi # interrupt stack ptr
-	movl %esp,     %esi # interrupt registers ptr
-
-	movl %esp, %ebp
-	andl $-16, %esp
-
-	subl $8, %esp
-	pushl %esi
-	pushl %edi
-	call cpp_yield_handler
-
-	movl %ebp, %esp
-
-	popal
-	iret
-
 .global asm_ipi_handler
 asm_ipi_handler:
-	push_userspace
-	load_kernel_segments
-	cld
+	intr_header 4
 
 	movl %esp, %ebp
 	andl $-16, %esp
@@ -110,14 +87,12 @@ asm_ipi_handler:
 
 	movl %ebp, %esp
 
-	pop_userspace
+	intr_footer 4
 	iret
 
 .global asm_timer_handler
 asm_timer_handler:
-	push_userspace
-	load_kernel_segments
-	cld
+	intr_header 4
 
 	movl %esp, %ebp
 	andl $-16, %esp
@@ -126,7 +101,7 @@ asm_timer_handler:
 
 	movl %ebp, %esp
 
-	pop_userspace
+	intr_footer 4
 	iret
 
 .macro isr n

kernel/arch/i686/linker.ld

@@ -11,20 +11,21 @@ SECTIONS
 	{
 		g_kernel_execute_start = .;
 		*(.multiboot)
+		*(.multiboot2)
 		*(.bananboot)
 		*(.text.*)
 	}
+	.ap_init ALIGN(4K) : AT(ADDR(.ap_init) - KERNEL_OFFSET)
+	{
+		g_ap_init_addr = .;
+		*(.ap_init)
+		g_kernel_execute_end = .;
+	}
 	.userspace ALIGN(4K) : AT(ADDR(.userspace) - KERNEL_OFFSET)
 	{
 		g_userspace_start = .;
 		*(.userspace)
 		g_userspace_end = .;
-		g_kernel_execute_end = .;
-	}
-	.ap_init ALIGN(4K) : AT(ADDR(.ap_init) - KERNEL_OFFSET)
-	{
-		g_ap_init_addr = .;
-		*(.ap_init)
 	}
 	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
 	{

kernel/arch/x86_64/Signal.S

@@ -1,57 +1,97 @@
 .section .userspace, "ax"
 
 // stack contains
-//    return address
-//    signal number
-//    signal handler
+//    (8 bytes)  return address (on return stack)
+//    (8 bytes)  return stack
+//    (8 bytes)  return rflags
+//    (8 bytes)  restore sigmask
+//    (56 bytes) siginfo_t
+//    (8 bytes)  signal number
+//    (8 bytes)  signal handler
 
 .global signal_trampoline
 signal_trampoline:
-	pushq %rax
-	pushq %rbx
-	pushq %rcx
-	pushq %rdx
-	pushq %rbp
-	pushq %rdi
-	pushq %rsi
-	pushq %r8
-	pushq %r9
-	pushq %r10
-	pushq %r11
-	pushq %r12
-	pushq %r13
+	pushq %r15 // gregs
 	pushq %r14
-	pushq %r15
+	pushq %r13
+	pushq %r12
+	pushq %r11
+	pushq %r10
+	pushq %r9
+	pushq %r8
+	pushq %rsi
+	pushq %rdi
+	pushq %rdx
+	pushq %rcx
+	pushq %rbx
+	pushq %rax
+	pushq %rbp
 
-	movq 128(%rsp), %rdi
-	movq 120(%rsp), %rax
+	movq 208(%rsp), %rax
+	pushq %rax; addq $(128 + 8), (%rsp)
+	pushq (%rax)
+
+	// FIXME: populate these
+	xorq %rax, %rax
+	pushq %rax // stack
+	pushq %rax
+	pushq %rax
+	pushq %rax // sigset
+	pushq %rax // link
+
+	movq  %rsp,     %rdx // ucontext
+	leaq 192(%rsp), %rsi // siginfo
+	movq 184(%rsp), %rdi // signal number
+	movq 176(%rsp), %rax // handler
 
 	// align stack to 16 bytes
-	movq %rsp, %rbx
-	andq $0x0F, %rbx
-	subq %rbx, %rsp
+	movq %rsp, %rbp
+	andq $-16, %rsp
+
+	subq $512, %rsp
+	fxsave64 (%rsp)
 
 	call *%rax
 
-	// restore stack
-	addq %rbx, %rsp
-	popq %r15
-	popq %r14
-	popq %r13
-	popq %r12
-	popq %r11
-	popq %r10
-	popq %r9
-	popq %r8
-	popq %rsi
-	popq %rdi
-	popq %rbp
-	popq %rdx
-	popq %rcx
-	popq %rbx
-	popq %rax
+	fxrstor64 (%rsp)
+	addq $512, %rsp
 
+	// restore stack
+	movq %rbp, %rsp
+	addq $40, %rsp
+
+	// restore sigmask
+	movq $79, %rdi       // SYS_SIGPROCMASK
+	movq $3,  %rsi       // SIG_SETMASK
+	leaq 192(%rsp), %rdx // set
+	xorq %r10, %r10      // oset
+	syscall
+
+	// restore registers
 	addq $16, %rsp
+	popq %rbp
+	popq %rax
+	popq %rbx
+	popq %rcx
+	popq %rdx
+	popq %rdi
+	popq %rsi
+	popq %r8
+	popq %r9
+	popq %r10
+	popq %r11
+	popq %r12
+	popq %r13
+	popq %r14
+	popq %r15
+
+	// skip handler, number, siginfo_t, sigmask
+	addq $80, %rsp
+
+	// restore flags
+	popfq
+
+	movq (%rsp), %rsp
 
 	// return over red-zone
 	ret $128

kernel/arch/x86_64/Syscall.S

@@ -1,49 +1,26 @@
-// arguments in RAX, RBX, RCX, RDX, RSI, RDI
-// System V ABI: RDI, RSI, RDX, RCX, R8, R9
 .global asm_syscall_handler
 asm_syscall_handler:
-	pushq %rbx
-	pushq %rcx
-	pushq %rdx
-	pushq %rdi
-	pushq %rsi
-	pushq %rbp
-	pushq %r8
-	pushq %r9
-	pushq %r10
+	swapgs
+
+	movq %rsp, %rax
+	movq %gs:8, %rsp
+
+	pushq $(0x20 | 3)
+	pushq %rax
 	pushq %r11
-	pushq %r12
-	pushq %r13
-	pushq %r14
-	pushq %r15
-	cld
+	pushq $(0x28 | 3)
+	pushq %rcx
+	subq $8, %rsp
 
-	movq %rsi, %r8
-	movq %rdi, %r9
-	movq %rax, %rdi
-	movq %rbx, %rsi
-	xchgq %rcx, %rdx
-	leaq 112(%rsp), %rbx
-	pushq %rbx
+	movq %r10, %rcx
 	call cpp_syscall_handler
-	addq $8, %rsp
 
-	popq %r15
-	popq %r14
-	popq %r13
-	popq %r12
-	popq %r11
-	popq %r10
-	popq %r9
-	popq %r8
-	popq %rbp
-	popq %rsi
-	popq %rdi
-	popq %rdx
-	popq %rcx
-	popq %rbx
-	iretq
+	movq  8(%rsp), %rcx
+	movq 24(%rsp), %r11
+	movq 32(%rsp), %rsp
 
+	swapgs
+	sysretq
 
 .global sys_fork_trampoline
 sys_fork_trampoline:
@@ -56,7 +33,7 @@ sys_fork_trampoline:
 
 	call read_ip
 	testq %rax, %rax
-	je .reload_stack
+	jz .done
 
 	movq %rax, %rsi
 	movq %rsp, %rdi
@@ -70,9 +47,3 @@ sys_fork_trampoline:
 	popq %rbp
 	popq %rbx
 	ret
-
- .reload_stack:
-	call get_thread_start_sp
-	movq %rax, %rsp
-	xorq %rax, %rax
-	jmp .done

kernel/arch/x86_64/Thread.S

@@ -7,9 +7,6 @@ read_ip:
 # void start_kernel_thread()
 .global start_kernel_thread
 start_kernel_thread:
-	call get_thread_start_sp
-	movq %rax, %rsp
-
 	# STACK LAYOUT
 	#   on_exit arg
 	#   on_exit func
@@ -27,9 +24,5 @@ start_kernel_thread:
 
 .global start_userspace_thread
 start_userspace_thread:
-	call load_thread_sse
-
-	call get_thread_start_sp
-	movq %rax, %rsp
-
+	swapgs
 	iretq

kernel/arch/x86_64/User.S

@@ -0,0 +1,87 @@
+# bool safe_user_memcpy(void*, const void*, size_t)
+.global safe_user_memcpy
+.global safe_user_memcpy_end
+.global safe_user_memcpy_fault
+safe_user_memcpy:
+	xorq %rax, %rax
+	movq %rdx, %rcx
+	rep movsb
+	incq %rax
+safe_user_memcpy_fault:
+	ret
+safe_user_memcpy_end:
+
+# bool safe_user_strncpy(void*, const void*, size_t)
+.global safe_user_strncpy
+.global safe_user_strncpy_end
+.global safe_user_strncpy_fault
+safe_user_strncpy:
+	movq %rdx, %rcx
+	testq %rcx, %rcx
+	jz safe_user_strncpy_fault
+
+ .safe_user_strncpy_align_loop:
+	testb $0x7, %sil
+	jz .safe_user_strncpy_align_done
+
+	movb (%rsi), %al
+	movb %al, (%rdi)
+	testb %al, %al
+	jz .safe_user_strncpy_done
+
+	incq %rdi
+	incq %rsi
+	decq %rcx
+	jnz .safe_user_strncpy_align_loop
+	jmp safe_user_strncpy_fault
+
+ .safe_user_strncpy_align_done:
+	movq $0x0101010101010101, %r8
+	movq $0x8080808080808080, %r9
+
+ .safe_user_strncpy_qword_loop:
+	cmpq $8, %rcx
+	jb .safe_user_strncpy_qword_done
+
+	movq (%rsi), %rax
+	movq %rax, %r10
+	movq %rax, %r11
+
+	# https://graphics.stanford.edu/~seander/bithacks.html#ZeroInWord
+	subq %r8, %r10
+	notq %r11
+	andq %r11, %r10
+	andq %r9, %r10
+	jnz .safe_user_strncpy_byte_loop
+
+	movq %rax, (%rdi)
+
+	addq $8, %rdi
+	addq $8, %rsi
+	subq $8, %rcx
+	jnz .safe_user_strncpy_qword_loop
+	jmp safe_user_strncpy_fault
+
+ .safe_user_strncpy_qword_done:
+	testq %rcx, %rcx
+	jz safe_user_strncpy_fault
+
+ .safe_user_strncpy_byte_loop:
+	movb (%rsi), %al
+	movb %al, (%rdi)
+	testb %al, %al
+	jz .safe_user_strncpy_done
+
+	incq %rdi
+	incq %rsi
+	decq %rcx
+	jnz .safe_user_strncpy_byte_loop
+
+ safe_user_strncpy_fault:
+	xorq %rax, %rax
+	ret
+
+ .safe_user_strncpy_done:
+	movb $1, %al
+	ret
+safe_user_strncpy_end:

kernel/arch/x86_64/Yield.S

@@ -0,0 +1,29 @@
+.global asm_yield_trampoline
+asm_yield_trampoline:
+	leaq 8(%rsp), %rcx
+	movq %rdi, %rsp
+
+	subq $8, %rsp
+	pushq -8(%rcx)
+	pushq %rcx
+	pushq %rax
+	pushq %rbx
+	pushq %rbp
+	pushq %r12
+	pushq %r13
+	pushq %r14
+	pushq %r15
+
+	movq %rsp, %rdi
+	call scheduler_on_yield
+
+	popq %r15
+	popq %r14
+	popq %r13
+	popq %r12
+	popq %rbp
+	popq %rbx
+	popq %rax
+	movq 8(%rsp), %rcx
+	movq 0(%rsp), %rsp
+	jmp *%rcx

kernel/arch/x86_64/boot.S

@@ -11,9 +11,28 @@
 
 .code32
 
-# multiboot2 header
+// custom addresses, video mode info, page align modules
+.set multiboot_flags, (1 << 16) | (1 << 2) | (1 << 0)
+
 .section .multiboot, "aw"
-	.align 8
+multiboot_start:
+	.long 0x1BADB002
+	.long multiboot_flags
+	.long -(0x1BADB002 + multiboot_flags)
+
+	.long V2P(multiboot_start)
+	.long V2P(g_kernel_start)
+	.long V2P(g_kernel_bss_start)
+	.long V2P(g_kernel_end)
+	.long V2P(_start)
+
+	.long 0
+	.long FB_WIDTH
+	.long FB_HEIGHT
+	.long FB_BPP
+multiboot_end:
+
+.section .multiboot2, "aw"
 multiboot2_start:
 	.long 0xE85250D6
 	.long 0
@@ -36,6 +55,12 @@ multiboot2_start:
 	.long 12
 	.long V2P(_start)
 
+	# page align modules
+	.align 8
+	.short 6
+	.short 0
+	.long 8
+
 	.align 8
 	.short 0
 	.short 0
@@ -43,7 +68,6 @@ multiboot2_start:
 multiboot2_end:
 
 .section .bananboot, "aw"
-	.align 8
 bananboot_start:
 	.long 0xBABAB007
 	.long -(0xBABAB007 + FB_WIDTH + FB_HEIGHT + FB_BPP)
@@ -53,9 +77,10 @@ bananboot_start:
 bananboot_end:
 
 .section .bss, "aw", @nobits
-	boot_stack_bottom:
-		.skip 4096 * 64
-	boot_stack_top:
+    .global g_boot_stack_top
+	g_boot_stack_bottom:
+		.skip 4096 * 4
+	g_boot_stack_top:
 
 	.global g_kernel_cmdline
 	g_kernel_cmdline:
@@ -72,27 +97,25 @@ bananboot_end:
 .align 4096
 boot_pml4:
 	.quad V2P(boot_pdpt_lo) + (PG_READ_WRITE | PG_PRESENT)
-	.rept 510
-		.quad 0
-	.endr
+	.skip 510 * 8
 	.quad V2P(boot_pdpt_hi) + (PG_READ_WRITE | PG_PRESENT)
 boot_pdpt_lo:
 	.quad V2P(boot_pd) + (PG_READ_WRITE | PG_PRESENT)
-	.rept 511
-		.quad 0
-	.endr
+	.skip 511 * 8
 boot_pdpt_hi:
-	.rept 510
-		.quad 0
-	.endr
+	.skip 510 * 8
 	.quad V2P(boot_pd) + (PG_READ_WRITE | PG_PRESENT)
-	.quad 0
+	.skip 8
 boot_pd:
 	.set i, 0
-	.rept 512
+	.rept 511
 		.quad i + (PG_PAGE_SIZE | PG_READ_WRITE | PG_PRESENT)
 		.set i, i + 0x200000
 	.endr
+	.quad V2P(g_boot_fast_page_pt) + (PG_READ_WRITE | PG_PRESENT)
+.global g_boot_fast_page_pt
+g_boot_fast_page_pt:
+	.skip 512 * 8
 
 boot_gdt:
 	.quad 0x0000000000000000 # null descriptor
@@ -155,6 +178,13 @@ enable_sse:
 	movl %eax, %cr4
 	ret
 
+enable_tsc:
+	# allow userspace to use RDTSC
+	movl %cr4, %ecx
+	andl $0xFFFFFFFB, %ecx
+	movl %ecx, %cr4
+	ret
+
 initialize_paging:
 	# enable PAE
 	movl %cr4, %ecx
@@ -187,10 +217,11 @@ _start:
 	movl %eax, V2P(bootloader_magic)
 	movl %ebx, V2P(bootloader_info)
 
-	movl $V2P(boot_stack_top), %esp
+	movl $V2P(g_boot_stack_top), %esp
 
 	call check_requirements
 	call enable_sse
+	call enable_tsc
 	call initialize_paging
 
 	# flush gdt and jump to 64 bit
@@ -240,7 +271,7 @@ system_halt:
 	jmp 1b
 
 
-#define AP_V2P(vaddr) ((vaddr) - ap_trampoline + 0xF000)
+#define AP_REL(vaddr) ((vaddr) - ap_trampoline + 0xF000)
 
 .section .ap_init, "ax"
 
@@ -250,21 +281,27 @@ ap_trampoline:
 	jmp 1f
 
 .align 8
-ap_stack_ptr:
-	.skip 4
-ap_stack_loaded:
-	.skip 1
+ap_stack_paddr:
+	.skip 8
+ap_stack_vaddr:
+	.skip 8
+ap_prepare_paging:
+	.skip 8
+ap_page_table:
+	.skip 8
+ap_ready:
+	.skip 8
 
 1:	cli; cld
-	ljmpl $0x00, $AP_V2P(ap_cs_clear)
+	ljmpl $0x00, $AP_REL(ap_cs_clear)
 
 ap_cs_clear:
 	# load ap gdt and enter protected mode
-	lgdt AP_V2P(ap_gdtr)
+	lgdt AP_REL(ap_gdtr)
 	movl %cr0, %eax
 	orb $1, %al
 	movl %eax, %cr0
-	ljmpl $0x08, $AP_V2P(ap_protected_mode)
+	ljmpl $0x08, $AP_REL(ap_protected_mode)
 
 .code32
 ap_protected_mode:
@@ -273,36 +310,42 @@ ap_protected_mode:
 	movw %ax, %ss
 	movw %ax, %es
 
-	movl AP_V2P(ap_stack_ptr), %esp
-	movb $1, AP_V2P(ap_stack_loaded)
+	movl AP_REL(ap_stack_paddr), %esp
 
 	leal V2P(enable_sse),        %ecx; call *%ecx
+	leal V2P(enable_tsc),        %ecx; call *%ecx
 	leal V2P(initialize_paging), %ecx; call *%ecx
 
 	# load boot gdt and enter long mode
 	lgdt V2P(boot_gdtr)
-	ljmpl $0x08, $AP_V2P(ap_long_mode)
+	ljmpl $0x08, $AP_REL(ap_long_mode)
 
 .code64
 ap_long_mode:
-	# move stack pointer to higher half
-	movl %esp, %esp
-	addq $KERNEL_OFFSET, %rsp
+	movq $ap_higher_half, %rax
+	jmp *%rax
+
+ap_higher_half:
+	movq AP_REL(ap_prepare_paging), %rax
+	call *%rax
+
+	# load AP's initial values
+	movq AP_REL(ap_stack_vaddr), %rsp
+	movq AP_REL(ap_page_table), %rax
+	movq $1, AP_REL(ap_ready)
+	movq %rax, %cr3
 
 	# clear rbp for stacktrace
 	xorq %rbp, %rbp
 
-	xorb %al, %al
 1:	pause
-	cmpb %al, g_ap_startup_done
-	jz 1b
+	cmpb $0, g_ap_startup_done
+	je 1b
 
 	lock incb g_ap_running_count
 
-	# jump to ap_main in higher half
-	movabsq $ap_main, %rcx
-	call *%rcx
-	jmp V2P(system_halt)
+	call ap_main
+	jmp system_halt
 
 ap_gdt:
 	.quad 0x0000000000000000 # null descriptor

kernel/arch/x86_64/interrupts.S

@@ -1,4 +1,4 @@
-.macro pushaq
+.macro intr_header, n
 	pushq %rax
 	pushq %rcx
 	pushq %rdx
@@ -14,10 +14,18 @@
 	pushq %r13
 	pushq %r14
 	pushq %r15
+	testb $3, \n+15*8(%rsp)
+	jz 1f
+	swapgs
+1:	cld
 .endm
 
-.macro popaq
-	popq %r15
+.macro intr_footer, n
+	testb $3, \n+15*8(%rsp)
+	jz 1f
+	call cpp_check_signal
+	swapgs
+1:	popq %r15
 	popq %r14
 	popq %r13
 	popq %r12
@@ -35,8 +43,7 @@
 .endm
 
 isr_stub:
-	pushaq
-	cld
+	intr_header 24
 	movq %cr0, %rax; pushq %rax
 	movq %cr2, %rax; pushq %rax
 	movq %cr3, %rax; pushq %rax
@@ -49,43 +56,33 @@ isr_stub:
 	call cpp_isr_handler
 	addq $32, %rsp
 
-	popaq
+	intr_footer 24
 	addq $16, %rsp
 	iretq
 
 irq_stub:
-	pushaq
-	cld
+	intr_header 24
+	xorq %rbp, %rbp
 	movq 120(%rsp), %rdi	# irq number
 	call cpp_irq_handler
-	popaq
+	intr_footer 24
 	addq $16, %rsp
 	iretq
 
-.global asm_yield_handler
-asm_yield_handler:
-	pushaq
-	cld
-	leaq 120(%rsp), %rdi # interrupt stack ptr
-	movq     %rsp,  %rsi # interrupt register ptr
-	call cpp_yield_handler
-	popaq
-	iretq
-
 .global asm_ipi_handler
 asm_ipi_handler:
-	pushaq
-	cld
+	intr_header 8
+	xorq %rbp, %rbp
 	call cpp_ipi_handler
-	popaq
+	intr_footer 8
 	iretq
 
 .global asm_timer_handler
 asm_timer_handler:
-	pushaq
-	cld
+	intr_header 8
+	xorq %rbp, %rbp
 	call cpp_timer_handler
-	popaq
+	intr_footer 8
 	iretq
 
 .macro isr n

kernel/arch/x86_64/linker.ld

@@ -11,20 +11,21 @@ SECTIONS
 	{
 		g_kernel_execute_start = .;
 		*(.multiboot)
+		*(.multiboot2)
 		*(.bananboot)
 		*(.text.*)
 	}
+	.ap_init ALIGN(4K) : AT(ADDR(.ap_init) - KERNEL_OFFSET)
+	{
+		g_ap_init_addr = .;
+		*(.ap_init)
+		g_kernel_execute_end = .;
+	}
 	.userspace ALIGN(4K) : AT(ADDR(.userspace) - KERNEL_OFFSET)
 	{
 		g_userspace_start = .;
 		*(.userspace)
 		g_userspace_end = .;
-		g_kernel_execute_end = .;
-	}
-	.ap_init ALIGN(4K) : AT(ADDR(.ap_init) - KERNEL_OFFSET)
-	{
-		g_ap_init_addr = .;
-		*(.ap_init)
 	}
 	.rodata ALIGN(4K) : AT(ADDR(.rodata) - KERNEL_OFFSET)
 	{
@@ -40,9 +41,21 @@ SECTIONS
 	{
 		g_kernel_writable_start = .;
 		*(.data)
+
+		. = ALIGN(8);
+		g_drv_builtin_begin = .;
+		KEEP(*(.banos-driver))
+		g_drv_builtin_end = .;
+		. = ALIGN(8);
+		g_banos_export = .;
+		KEEP(*(.banos-export))
+		g_banos_export_end = .;
 	}
+
+
 	.bss ALIGN(4K) : AT(ADDR(.bss) - KERNEL_OFFSET)
 	{
+		g_kernel_bss_start = .;
 		*(COMMON)
 		*(.bss)
 		g_kernel_writable_end = .;

kernel/include/banos/driver.h

@@ -0,0 +1,27 @@
+#pragma once
+// Copyright (c) 2026 Dcraftbg
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+// THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+#include "version.h"
+#include "revision.h"
+
+#define BANOS_DRIVER_REVISION_CURRENT 0
+
+typedef struct Banos_Driver Banos_Driver;
+struct Banos_Driver {
+    unsigned long driver_size;
+    banos_version_t minimal_banos_version;
+    const char* name;
+    const char* license;
+    banos_version_t version;
+
+    // NOTE: checkout BANOS_DRIVER_INSTANCE_SIZE.
+    // You may use this instance data for anything you wish to store.
+    // If you need more than that just allocate it on the heap or
+    // globally if you add the proper verification of having your driver run only
+    // within a single instance
+    int (*init)(Banos_Driver* drv);
+    int (*uninit)(Banos_Driver* drv);
+};
+#define BANOS_DRIVER_API static __attribute__((section(".banos-driver"), used, aligned(8)))

kernel/include/banos/export.h

@@ -0,0 +1,15 @@
+#pragma once
+// Copyright (c) 2026 Dcraftbg
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+// THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+typedef struct Banos_Symbol {
+    const char* name;
+    void* arg;
+} Banos_Symbol;
+#define BANOS_EXPORT_SYMBOL(symname, str, ptr) \
+    static __attribute__((section(".banos-export"), used, aligned(8))) Banos_Symbol __symbol_##symname = {\
+        .name = str, \
+        .arg = ptr \
+    };
+#define BANOS_EXPORT(name) BANOS_EXPORT_SYMBOL(name, #name, (void*)&name)

kernel/include/banos/print.h

@@ -0,0 +1,13 @@
+#pragma once
+// Copyright (c) 2026 Dcraftbg
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+// THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void banos_dprintln(const char* str);
+#ifdef __cplusplus
+}
+#endif

kernel/include/banos/revision.h

@@ -0,0 +1,2 @@
+#pragma once
+typedef unsigned long banos_revision_t;

kernel/include/banos/version.h

@@ -0,0 +1,28 @@
+#pragma once
+// Copyright (c) 2026 Dcraftbg
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+// THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+// [ 8 bit major ] [ 8 minor ] [ 16 patch]
+typedef unsigned int banos_version_t;
+#define BANOS_VERSION_MAJOR_SHIFT 24
+#define BANOS_VERSION_MINOR_SHIFT 16
+#define BANOS_VERSION_PATCH_SHIFT 0
+
+#define BANOS_VERSION_MAJOR_MASK 0xFF
+#define BANOS_VERSION_MINOR_MASK 0xFF
+#define BANOS_VERSION_PATCH_MASK 0xFFFF
+
+#define BANOS_VERSION_MAKE(major, minor, patch) \
+    (banos_version_t)( \
+        (((major) & BANOS_VERSION_MAJOR_MASK) << BANOS_VERSION_MAJOR_SHIFT) | \
+        (((minor) & BANOS_VERSION_MINOR_MASK) << BANOS_VERSION_MINOR_SHIFT) | \
+        (((patch) & BANOS_VERSION_PATCH_MASK) << BANOS_VERSION_PATCH_SHIFT) \
+    )
+
+#define BANOS_VERSION_CURRENT BANOS_VERSION_MAKE(0, 0, 1)
+
+#define BANOS_VERSION_GET_MAJOR(v) (((v) >> BANOS_VERSION_MAJOR_SHIFT) & BANOS_VERSION_MAJOR_MASK)
+#define BANOS_VERSION_GET_MINOR(v) (((v) >> BANOS_VERSION_MINOR_SHIFT) & BANOS_VERSION_MINOR_MASK)
+#define BANOS_VERSION_GET_PATCH(v) (((v) >> BANOS_VERSION_PATCH_SHIFT) & BANOS_VERSION_PATCH_MASK)

kernel/include/kernel/ACPI/ACPI.h

@@ -2,6 +2,7 @@
 
 #include <BAN/Vector.h>
 #include <kernel/ACPI/AML/Namespace.h>
+#include <kernel/ACPI/EmbeddedController.h>
 #include <kernel/ACPI/Headers.h>
 #include <kernel/Memory/Types.h>
 #include <kernel/ThreadBlocker.h>
@@ -35,8 +36,12 @@ namespace Kernel::ACPI
 		BAN::ErrorOr<void> poweroff();
 		BAN::ErrorOr<void> reset();
 
+		BAN::ErrorOr<void> register_gpe_handler(uint8_t gpe, void (*callback)(void*), void* argument);
+
 		void handle_irq() override;
 
+		BAN::Span<BAN::UniqPtr<EmbeddedController>> embedded_controllers() { return m_embedded_controllers.span(); }
+
 	private:
 		ACPI() = default;
 		BAN::ErrorOr<void> initialize_impl();
@@ -50,6 +55,12 @@ namespace Kernel::ACPI
 
 		BAN::ErrorOr<void> route_interrupt_link_device(const AML::Scope& device, uint64_t& routed_irq_mask);
 
+		BAN::ErrorOr<void> initialize_embedded_controller(const AML::Scope& embedded_controller);
+		BAN::ErrorOr<void> initialize_embedded_controllers();
+
+		BAN::Optional<GAS> find_gpe_block(size_t index);
+		bool enable_gpe(uint8_t gpe);
+
 	private:
 		paddr_t m_header_table_paddr = 0;
 		vaddr_t m_header_table_vaddr = 0;
@@ -68,8 +79,23 @@ namespace Kernel::ACPI
 
 		ThreadBlocker m_event_thread_blocker;
 
+		BAN::Vector<BAN::UniqPtr<EmbeddedController>> m_embedded_controllers;
+
+		struct GPEHandler
+		{
+			bool has_callback { false };
+			union {
+				AML::Reference* method;
+				struct
+				{
+					void (*callback)(void*);
+					void* argument;
+				};
+			};
+		};
+		bool m_has_any_gpes { false };
 		AML::Scope m_gpe_scope;
-		BAN::Array<AML::Reference*, 0xFF> m_gpe_methods { nullptr };
+		BAN::Array<GPEHandler, 0xFF> m_gpe_methods;
 
 		bool m_hardware_reduced { false };
 		AML::Namespace* m_namespace { nullptr };

kernel/include/kernel/ACPI/AML/Namespace.h

@@ -50,6 +50,7 @@ namespace Kernel::ACPI::AML
 		BAN::ErrorOr<void> for_each_child(const Scope&, const BAN::Function<BAN::Iteration(const Scope&, Reference*)>&);
 
 		BAN::ErrorOr<BAN::Vector<Scope>> find_device_with_eisa_id(BAN::StringView eisa_id);
+		BAN::ErrorOr<BAN::Vector<Scope>> find_device_with_eisa_id(BAN::Span<BAN::StringView> eisa_ids);
 
 	private:
 		BAN::ErrorOr<Scope> resolve_path(const Scope& scope, const NameString& name_string);

kernel/include/kernel/ACPI/AML/Node.h

@@ -89,6 +89,9 @@ namespace Kernel::ACPI::AML
 		GAS::AddressSpaceID address_space;
 		uint64_t offset;
 		uint64_t length;
+		alignas(Scope) uint8_t scope_storage[sizeof(Scope)];
+		Scope& scope() { return *reinterpret_cast<Scope*>(scope_storage); }
+		const Scope& scope() const { return *reinterpret_cast<const Scope*>(scope_storage); }
 	};
 
 	struct FieldUnit

kernel/include/kernel/ACPI/EmbeddedController.h

@@ -0,0 +1,69 @@
+#pragma once
+
+#include <BAN/Atomic.h>
+#include <BAN/UniqPtr.h>
+
+#include <kernel/ACPI/AML/Scope.h>
+#include <kernel/Lock/Mutex.h>
+#include <kernel/Thread.h>
+
+namespace Kernel::ACPI
+{
+
+	class EmbeddedController
+	{
+	public:
+		static BAN::ErrorOr<BAN::UniqPtr<EmbeddedController>> create(AML::Scope&& scope, uint16_t command_port, uint16_t data_port, BAN::Optional<uint8_t> gpe);
+		~EmbeddedController();
+
+		BAN::ErrorOr<uint8_t> read_byte(uint8_t offset);
+		BAN::ErrorOr<void> write_byte(uint8_t offset, uint8_t value);
+
+		const AML::Scope& scope() const { return m_scope; }
+
+	private:
+		EmbeddedController(AML::Scope&& scope, uint16_t command_port, uint16_t data_port, bool has_gpe)
+			: m_scope(BAN::move(scope))
+			, m_command_port(command_port)
+			, m_data_port(data_port)
+			, m_has_gpe(has_gpe)
+		{ }
+
+	private:
+		void wait_status_bit(uint8_t bit, uint8_t value);
+
+		uint8_t read_one(uint16_t port);
+		void write_one(uint16_t port, uint8_t value);
+
+		static void handle_gpe_wrapper(void*);
+		void handle_gpe();
+
+		BAN::ErrorOr<void> call_query_method(uint8_t notification);
+
+		void thread_task();
+
+		struct Command
+		{
+			uint8_t command;
+			BAN::Optional<uint8_t> data1;
+			BAN::Optional<uint8_t> data2;
+			uint8_t* response;
+			BAN::Atomic<bool> done;
+		};
+		BAN::ErrorOr<void> send_command(Command& command);
+
+	private:
+		const AML::Scope m_scope;
+		const uint16_t m_command_port;
+		const uint16_t m_data_port;
+		const bool m_has_gpe;
+
+		Mutex m_mutex;
+		ThreadBlocker m_thread_blocker;
+
+		BAN::Optional<Command*> m_queued_command;
+
+		Thread* m_thread { nullptr };
+	};
+
+}

kernel/include/kernel/ACPI/Resource.h

@@ -0,0 +1,179 @@
+#pragma once
+
+#include <BAN/ByteSpan.h>
+#include <BAN/Debug.h>
+#include <BAN/Optional.h>
+
+namespace Kernel::ACPI
+{
+
+	struct ResourceData
+	{
+		enum class Type
+		{
+			IRQ,
+			DMA,
+			IOPort,
+			FixedIOPort,
+			FixedDMA,
+
+			// TODO: large stuff the stuff :)
+		};
+
+		union {
+			struct {
+				uint16_t irq_mask;
+				union {
+					struct {
+						uint8_t edge_triggered : 1;
+						uint8_t                : 2;
+						uint8_t active_low     : 1;
+						uint8_t shared         : 1;
+						uint8_t wake_capable   : 1;
+						uint8_t                : 2;
+					};
+					uint8_t raw;
+				};
+			} irq;
+			struct {
+				uint8_t channel_mask;
+				union {
+					struct {
+						uint8_t type          : 2; // 0: 8 bit, 1: 8 and 16 bit, 2: 16 bit only
+						uint8_t bus_master    : 1;
+						uint8_t               : 2;
+						uint8_t channel_speed : 2; // 0: compatibility, 1: type A, 2: type B, 3: type F
+						uint8_t               : 1;
+					};
+					uint8_t raw;
+				};
+			} dma;
+			struct {
+				uint16_t range_min_base;
+				uint16_t range_max_base;
+				uint8_t base_alignment;
+				uint8_t range_length;
+			} io_port;
+			struct {
+				uint16_t range_base;
+				uint8_t range_length;
+			} fixed_io_port;
+			struct {
+				uint16_t request_line;
+				uint16_t channel;
+				uint8_t transfer_width; // 0: 8 bit, 1: 16 bit, 2: 32 bit, 3: 64 bit, 4: 128 bit
+			} fixed_dma;
+		} as;
+		Type type;
+	};
+
+	class ResourceParser
+	{
+	public:
+		ResourceParser(BAN::ConstByteSpan buffer)
+			: m_buffer(buffer)
+		{}
+
+		BAN::Optional<ResourceData> get_next()
+		{
+			for (;;)
+			{
+				if (m_buffer.empty())
+					return {};
+
+				if (m_buffer[0] & 0x80)
+				{
+					dprintln("Skipping large resource 0x{2H}", m_buffer[0] & 0x7F);
+					const uint16_t length = (m_buffer[2] << 8) | m_buffer[1];
+					if (m_buffer.size() < static_cast<size_t>(3 + length))
+						return {};
+					m_buffer = m_buffer.slice(3 + length);
+					continue;
+				}
+
+				const uint8_t length = m_buffer[0] & 0x07;
+				if (m_buffer.size() < static_cast<size_t>(1 + length))
+					return {};
+
+				BAN::Optional<ResourceData> result;
+				switch ((m_buffer[0] >> 3) & 0x0F)
+				{
+					case 0x04:
+						if (length < 2)
+							break;
+						result = ResourceData {
+							.as = { .irq = {
+								.irq_mask = static_cast<uint16_t>((m_buffer[2] << 8) | m_buffer[1]),
+								.raw = (length >= 3) ? m_buffer[3] : static_cast<uint8_t>(1),
+							}},
+							.type = ResourceData::Type::IRQ,
+						};
+						break;
+					case 0x05:
+						if (length < 2)
+							break;
+						result = ResourceData {
+							.as = { .dma = {
+								.channel_mask = m_buffer[1],
+								.raw = m_buffer[2],
+							}},
+							.type = ResourceData::Type::DMA,
+						};
+						break;
+					case 0x08:
+						if (length < 7)
+							break;
+						result = ResourceData {
+							.as = { .io_port = {
+								.range_min_base = static_cast<uint16_t>(((m_buffer[3] << 8) | m_buffer[2]) & ((m_buffer[1] & 1) ? 0xFFFF : 0x03FF)),
+								.range_max_base = static_cast<uint16_t>(((m_buffer[5] << 8) | m_buffer[4]) & ((m_buffer[1] & 1) ? 0xFFFF : 0x03FF)),
+								.base_alignment = m_buffer[6],
+								.range_length = m_buffer[7],
+							}},
+							.type = ResourceData::Type::IOPort,
+						};
+						break;
+					case 0x09:
+						if (length < 3)
+							break;
+						result = ResourceData {
+							.as = { .fixed_io_port = {
+								.range_base = static_cast<uint16_t>(((m_buffer[2] << 8) | m_buffer[1]) & 0x03FF),
+								.range_length = m_buffer[3],
+							}},
+							.type = ResourceData::Type::FixedIOPort,
+						};
+						break;
+					case 0x0A:
+						if (length < 5)
+							break;
+						result = ResourceData {
+							.as = { .fixed_dma = {
+								.request_line = static_cast<uint16_t>((m_buffer[2] << 8) | m_buffer[1]),
+								.channel      = static_cast<uint16_t>((m_buffer[4] << 8) | m_buffer[3]),
+								.transfer_width = m_buffer[5],
+							}},
+							.type = ResourceData::Type::FixedDMA,
+						};
+						break;
+					case 0x0F:
+						// End tag
+						return {};
+					case 0x06:
+					case 0x07:
+					case 0x0E:
+						dprintln("Skipping short resource 0x{2H}", (m_buffer[0] >> 3) & 0x0F);
+						break;
+				}
+
+				m_buffer = m_buffer.slice(1 + length);
+				if (result.has_value())
+					return result.release_value();
+			}
+		}
+
+	private:
+		BAN::ConstByteSpan m_buffer;
+	};
+
+}

kernel/include/kernel/API/SharedPage.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#include <stdint.h>
+
+namespace Kernel::API
+{
+
+	enum SharedPageFeature : uint32_t
+	{
+		SPF_GETTIME = 1 << 0,
+	};
+
+	struct SharedPage
+	{
+		uint16_t gdt_cpu_offset;
+
+		uint32_t features;
+
+		struct
+		{
+			uint8_t shift;
+			uint64_t mult;
+			uint64_t realtime_seconds;
+		} gettime_shared;
+
+		struct
+		{
+			struct
+			{
+				uint32_t seq;
+				uint64_t last_ns;
+				uint64_t last_tsc;
+			} gettime_local;
+		} cpus[];
+	};
+
+}

kernel/include/kernel/API/Syscall.h

@@ -0,0 +1,34 @@
+#pragma once
+
+#include <BAN/MacroUtils.h>
+
+#if defined(__x86_64__)
+	#define _kas_instruction "syscall"
+	#define _kas_result rax
+	#define _kas_arguments rdi, rsi, rdx, r10, r8, r9
+	#define _kas_globbers rcx, rdx, rdi, rsi, r8, r9, r10, r11
+#elif defined(__i686__)
+	#define _kas_instruction "int $0xF0"
+	#define _kas_result eax
+	#define _kas_arguments eax, ebx, ecx, edx, esi, edi
+	#define _kas_globbers
+#endif
+
+#define _kas_argument_var(index, value) register long _kas_a##index asm(_ban_stringify(_ban_get(index, _kas_arguments))) = (long)(value);
+#define _kas_dummy_var(index, value) register long _kas_d##index asm(#value);
+#define _kas_input(index, _) "r"(_kas_a##index)
+#define _kas_output(index, _) , "=r"(_kas_d##index)
+#define _kas_globber(_, value) #value
+
+#define _kas_syscall(...) ({ \
+		register long _kas_ret asm(_ban_stringify(_kas_result)); \
+		_ban_for_each(_kas_argument_var, __VA_ARGS__) \
+		_ban_for_each(_kas_dummy_var, _kas_globbers) \
+		asm volatile( \
+			_kas_instruction \
+			: "=r"(_kas_ret) _ban_for_each(_kas_output, _kas_globbers) \
+			: _ban_for_each_comma(_kas_input, __VA_ARGS__) \
+			: "cc", "memory"); \
+		(void)_kas_a0; /* require 1 argument */ \
+		_kas_ret; \
+	})

kernel/include/kernel/Audio/AC97/Controller.h

@@ -0,0 +1,59 @@
+#pragma once
+
+#include <kernel/Audio/Controller.h>
+#include <kernel/Memory/DMARegion.h>
+
+namespace Kernel
+{
+
+	class AC97AudioController final : public AudioController, public Interruptable
+	{
+	public:
+		static BAN::ErrorOr<void> create(PCI::Device& pci_device);
+
+		void handle_irq() override;
+
+	protected:
+		void handle_new_data() override;
+
+		uint32_t get_channels() const override { return 2; }
+		uint32_t get_sample_rate() const override { return 48000; }
+
+		uint32_t get_total_pins() const override { return 1; }
+		uint32_t get_current_pin() const override { return 0; }
+		BAN::ErrorOr<void> set_current_pin(uint32_t pin) override { if (pin != 0) return BAN::Error::from_errno(EINVAL); return {}; }
+
+		BAN::ErrorOr<void> set_volume_mdB(int32_t) override;
+
+	private:
+		AC97AudioController(PCI::Device& pci_device)
+			: m_pci_device(pci_device)
+		{ }
+
+		uint32_t get_volume_data() const;
+
+		BAN::ErrorOr<void> initialize();
+		BAN::ErrorOr<void> initialize_bld();
+		BAN::ErrorOr<void> initialize_interrupts();
+
+		bool queue_samples_to_bld();
+
+	private:
+		static constexpr size_t m_bdl_entries = 32;
+		static constexpr size_t m_samples_per_entry = 0x1000;
+
+		// We only store samples in 2 BDL entries at a time to reduce the amount of samples queued.
+		// This is to reduce latency as you cannot remove data already passed to the BDLs
+		static constexpr size_t m_used_bdl_entries = 2;
+
+		PCI::Device& m_pci_device;
+		BAN::UniqPtr<PCI::BarRegion> m_mixer;
+		BAN::UniqPtr<PCI::BarRegion> m_bus_master;
+
+		BAN::UniqPtr<DMARegion> m_bdl_region;
+
+		uint32_t m_bdl_tail { 0 };
+		uint32_t m_bdl_head { 0 };
+	};
+
+}

kernel/include/kernel/Audio/AC97/Definitions.h

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <stdint.h>
+
+namespace Kernel::AC97
+{
+
+	struct BufferDescriptorListEntry
+	{
+		uint32_t address;
+		uint16_t samples;
+		uint16_t flags; // bit 14: last entry, bit 15: IOC
+	};
+
+}

kernel/include/kernel/Audio/Controller.h

@@ -0,0 +1,56 @@
+#pragma once
+
+#include <kernel/Device/Device.h>
+#include <kernel/Memory/ByteRingBuffer.h>
+#include <kernel/PCI.h>
+
+#include <sys/ioctl.h>
+
+namespace Kernel
+{
+
+	class AudioController : public CharacterDevice
+	{
+	public:
+		static BAN::ErrorOr<void> create(PCI::Device& pci_device);
+
+		BAN::StringView name() const override { return m_name; }
+
+	protected:
+		AudioController();
+		BAN::ErrorOr<void> initialize();
+
+		virtual void handle_new_data() = 0;
+
+		virtual uint32_t get_channels() const = 0;
+		virtual uint32_t get_sample_rate() const = 0;
+
+		virtual uint32_t get_total_pins() const = 0;
+		virtual uint32_t get_current_pin() const = 0;
+		virtual BAN::ErrorOr<void> set_current_pin(uint32_t) = 0;
+
+		virtual BAN::ErrorOr<void> set_volume_mdB(int32_t) = 0;
+
+		bool can_read_impl() const override { return false; }
+		bool can_write_impl() const override { SpinLockGuard _(m_spinlock); return !m_sample_data->full(); }
+		bool has_error_impl() const override { return false; }
+		bool has_hungup_impl() const override { return false; }
+
+		BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
+
+		BAN::ErrorOr<long> ioctl_impl(int cmd, void* arg) override;
+
+	protected:
+		ThreadBlocker m_sample_data_blocker;
+		mutable SpinLock m_spinlock;
+
+		static constexpr size_t m_sample_data_capacity = 1 << 20;
+		BAN::UniqPtr<ByteRingBuffer> m_sample_data;
+
+		snd_volume_info m_volume_info {};
+
+	private:
+		char m_name[10] {};
+	};
+
+}

kernel/include/kernel/Audio/HDAudio/AudioFunctionGroup.h

@@ -0,0 +1,80 @@
+#pragma once
+
+#include <kernel/Audio/Controller.h>
+#include <kernel/Audio/HDAudio/Controller.h>
+
+namespace Kernel
+{
+
+	class HDAudioController;
+
+	class HDAudioFunctionGroup final : public AudioController
+	{
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<HDAudioFunctionGroup>> create(BAN::RefPtr<HDAudioController>, uint8_t cid, HDAudio::AFGNode&&);
+
+		void on_stream_interrupt(uint8_t stream_index);
+
+	protected:
+		// FIXME: allow setting these :D
+		uint32_t get_channels() const override { return 2; }
+		uint32_t get_sample_rate() const override { return 48000; }
+
+		uint32_t get_total_pins() const override;
+		uint32_t get_current_pin() const override;
+		BAN::ErrorOr<void> set_current_pin(uint32_t) override;
+
+		BAN::ErrorOr<void> set_volume_mdB(int32_t) override;
+
+		void handle_new_data() override;
+
+	private:
+		HDAudioFunctionGroup(BAN::RefPtr<HDAudioController> controller, uint8_t cid, HDAudio::AFGNode&& afg_node)
+			: m_controller(controller)
+			, m_afg_node(BAN::move(afg_node))
+			, m_cid(cid)
+		{ }
+		~HDAudioFunctionGroup();
+
+		BAN::ErrorOr<void> initialize();
+		BAN::ErrorOr<void> initialize_stream();
+		BAN::ErrorOr<void> initialize_output();
+
+		BAN::ErrorOr<void> enable_output_path(uint8_t index);
+		BAN::ErrorOr<void> disable_output_path(uint8_t index);
+
+		BAN::ErrorOr<void> reset_stream();
+
+		BAN::ErrorOr<void> recurse_output_paths(const HDAudio::AFGWidget& widget, BAN::Vector<const HDAudio::AFGWidget*>& path);
+
+		uint16_t get_format_data() const;
+
+		size_t bdl_offset() const;
+
+		void queue_bdl_data();
+
+	private:
+		// use 6 512 sample BDL entries
+		// each entry is ~10.7 ms at 48 kHz
+		// -> total buffered audio is 64 ms
+		static constexpr size_t m_bdl_entry_sample_frames = 512;
+		static constexpr size_t m_bdl_entry_count         = 6;
+
+		BAN::RefPtr<HDAudioController> m_controller;
+		const HDAudio::AFGNode m_afg_node;
+		const uint8_t m_cid;
+
+		BAN::Vector<BAN::Vector<const HDAudio::AFGWidget*>> m_output_paths;
+		BAN::Vector<const HDAudio::AFGWidget*> m_output_pins;
+		size_t m_output_path_index { SIZE_MAX };
+
+		uint8_t m_stream_id    { 0xFF };
+		uint8_t m_stream_index { 0xFF };
+		BAN::UniqPtr<DMARegion> m_bdl_region;
+
+		size_t m_bdl_head { 0 };
+		size_t m_bdl_tail { 0 };
+		bool m_stream_running { false };
+	};
+
+}

kernel/include/kernel/Audio/HDAudio/Controller.h

@@ -0,0 +1,77 @@
+#pragma once
+
+#include <kernel/Audio/Controller.h>
+#include <kernel/Audio/HDAudio/Definitions.h>
+#include <kernel/Memory/DMARegion.h>
+
+namespace Kernel
+{
+
+	class HDAudioController : public Interruptable, public BAN::RefCounted<HDAudioController>
+	{
+	public:
+		static BAN::ErrorOr<void> create(PCI::Device& pci_device);
+
+		BAN::ErrorOr<uint32_t> send_command(HDAudio::CORBEntry);
+
+		uint8_t get_stream_index(HDAudio::StreamType type, uint8_t index) const;
+
+		BAN::ErrorOr<uint8_t> allocate_stream_id();
+		void deallocate_stream_id(uint8_t id);
+
+		BAN::ErrorOr<uint8_t> allocate_stream(HDAudio::StreamType type, void* afg);
+		void deallocate_stream(uint8_t index);
+
+		PCI::BarRegion& bar0() { return *m_bar0; }
+
+		bool is_64bit() const { return m_is64bit; }
+
+		void handle_irq() override;
+
+	private:
+		HDAudioController(PCI::Device& pci_device)
+			: m_pci_device(pci_device)
+		{ }
+
+		BAN::ErrorOr<void> initialize();
+		BAN::ErrorOr<void> initialize_ring_buffers();
+
+		BAN::ErrorOr<void> reset_controller();
+
+		BAN::ErrorOr<HDAudio::Codec> initialize_codec(uint8_t codec);
+		BAN::ErrorOr<HDAudio::AFGNode> initialize_node(uint8_t codec, uint8_t node);
+		BAN::ErrorOr<HDAudio::AFGWidget> initialize_widget(uint8_t codec, uint8_t node);
+
+	private:
+		struct RingBuffer
+		{
+			vaddr_t vaddr;
+			uint32_t index;
+			uint32_t size;
+		};
+
+	private:
+		PCI::Device& m_pci_device;
+		BAN::UniqPtr<PCI::BarRegion> m_bar0;
+		bool m_is64bit { false };
+
+		bool m_use_immediate_command { false };
+
+		uint8_t m_output_streams { 0 };
+		uint8_t m_input_streams { 0 };
+		uint8_t m_bidir_streams { 0 };
+		void* m_allocated_streams[30] {};
+
+		// NOTE: stream ids are from 1 to 15
+		uint16_t m_allocated_stream_ids { 0 };
+
+		Mutex m_command_mutex;
+		SpinLock m_rb_lock;
+		ThreadBlocker m_rb_blocker;
+
+		RingBuffer m_corb;
+		RingBuffer m_rirb;
+		BAN::UniqPtr<DMARegion> m_ring_buffer_region;
+	};
+
+}

kernel/include/kernel/Audio/HDAudio/Definitions.h

@@ -0,0 +1,90 @@
+#pragma once
+
+#include <BAN/Vector.h>
+
+namespace Kernel::HDAudio
+{
+
+	struct CORBEntry
+	{
+		union {
+			struct {
+				uint32_t data          : 8;
+				uint32_t command       : 12;
+				uint32_t node_index    : 8;
+				uint32_t codec_address : 4;
+			};
+			uint32_t raw;
+		};
+	};
+	static_assert(sizeof(CORBEntry) == sizeof(uint32_t));
+
+	struct BDLEntry
+	{
+		paddr_t address;
+		uint32_t length;
+		uint32_t ioc;
+	};
+	static_assert(sizeof(BDLEntry) == 16);
+
+	struct AFGWidget
+	{
+		enum class Type
+		{
+			OutputConverter,
+			InputConverter,
+			Mixer,
+			Selector,
+			PinComplex,
+			Power,
+			VolumeKnob,
+			BeepGenerator,
+		};
+
+		Type type;
+		uint8_t id;
+
+		union
+		{
+			struct
+			{
+				bool input;
+				bool output;
+				bool display; // HDMI or DP
+				uint32_t config;
+			} pin_complex;
+		};
+
+		struct Amplifier
+		{
+			uint8_t offset;
+			uint8_t num_steps;
+			uint8_t step_size;
+			bool mute;
+		};
+
+		BAN::Optional<Amplifier> output_amplifier;
+
+		BAN::Vector<uint16_t> connections;
+	};
+
+	struct AFGNode
+	{
+		uint8_t id;
+		BAN::Vector<AFGWidget> widgets;
+	};
+
+	struct Codec
+	{
+		uint8_t id;
+		BAN::Vector<AFGNode> nodes;
+	};
+
+	enum class StreamType
+	{
+		Input,
+		Output,
+		Bidirectional,
+	};
+
+}

kernel/include/kernel/Audio/HDAudio/Registers.h

@@ -0,0 +1,50 @@
+#pragma once
+
+#include <stdint.h>
+
+namespace Kernel::HDAudio
+{
+
+	enum Regs : uint8_t
+	{
+		GCAP = 0x00,
+		VMIN = 0x02,
+		VMAJ = 0x03,
+		GCTL = 0x08,
+		STATESTS = 0x0E,
+
+		INTCTL = 0x20,
+		INTSTS = 0x24,
+
+		CORBLBASE = 0x40,
+		CORBUBASE = 0x44,
+		CORBWP    = 0x48,
+		CORBRP    = 0x4A,
+		CORBCTL   = 0x4C,
+		CORBSTS   = 0x4D,
+		CORBSIZE  = 0x4E,
+
+		RIRBLBASE = 0x50,
+		RIRBUBASE = 0x54,
+		RIRBWP    = 0x58,
+		RINTCNT   = 0x5A,
+		RIRBCTL   = 0x5C,
+		RIRBSTS   = 0x5D,
+		RIRBSIZE  = 0x5E,
+
+		ICOI = 0x60,
+		ICII = 0x64,
+		ICIS = 0x68,
+
+		SDCTL   = 0x00,
+		SDSTS   = 0x03,
+		SDLPIB  = 0x04,
+		SDCBL   = 0x08,
+		SDLVI   = 0x0C,
+		SDFIFOD = 0x10,
+		SDFMT   = 0x12,
+		SDBDPL  = 0x18,
+		SDBDPU  = 0x1C,
+	};
+
+}

kernel/include/kernel/BananBootloader.h

@@ -4,6 +4,7 @@
 
 #define BANAN_BOOTLOADER_MAGIC	0xD3C60CFF
 #define BANAN_BOOTLOADER_FB_RGB		1
+#define BANAN_BOOTLOADER_FB_TEXT	2
 
 struct BananBootFramebufferInfo
 {

kernel/include/kernel/Banos.h

@@ -0,0 +1,10 @@
+#pragma once
+#include <BAN/Vector.h>
+#include <BAN/StringView.h>
+typedef struct Banos_Symbol Banos_Symbol;
+namespace Banos {
+	void* resolve_symbol(const char* name);
+	void import_symbols(Banos_Symbol* symbols, size_t count);
+	void initialize_initial_drivers(void);
+	BAN::ErrorOr<size_t> load_driver_from_image(const char* u_image);
+}

kernel/include/kernel/BootInfo.h

@@ -15,6 +15,7 @@ namespace Kernel
 			None,
 			Unknown,
 			RGB,
+			Text,
 		};
 
 		paddr_t  address;
@@ -40,6 +41,12 @@ namespace Kernel
 		Type     type;
 	};
 
+	struct BootModule
+	{
+		paddr_t start;
+		uint64_t size;
+	};
+
 	struct BootInfo
 	{
 		BAN::String     command_line;
@@ -47,6 +54,7 @@ namespace Kernel
 		RSDP            rsdp         {};
 		paddr_t         kernel_paddr {};
 
+		BAN::Vector<BootModule> modules;
 		BAN::Vector<MemoryMapEntry>	memory_map_entries;
 	};
 

kernel/include/kernel/CPUID.h

@@ -81,5 +81,6 @@ namespace CPUID
 	bool has_pge();
 	bool has_pat();
 	bool has_1gib_pages();
+	bool has_invariant_tsc();
 
 }

kernel/include/kernel/Credentials.h

@@ -35,10 +35,8 @@ namespace Kernel
 
 		bool has_egid(gid_t) const;
 
-		BAN::ErrorOr<void> initialize_supplementary_groups();
-
-	private:
-		BAN::ErrorOr<BAN::String> find_username() const;
+		BAN::Span<const gid_t> groups() const { return m_supplementary.span(); }
+		BAN::ErrorOr<void> set_groups(BAN::Span<const gid_t> groups);
 
 	private:
 		uid_t m_ruid, m_euid, m_suid;

kernel/include/kernel/Debug.h

@@ -49,6 +49,8 @@
 
 #define DEBUG_VTTY 1
 
+#define DEBUG_DEVFS 0
+
 #define DEBUG_PCI 0
 #define DEBUG_SCHEDULER 0
 #define DEBUG_PS2 1
@@ -70,10 +72,15 @@
 #define DEBUG_USB_MOUSE 0
 #define DEBUG_USB_MASS_STORAGE 0
 
+#define DEBUG_HDAUDIO 0
+
 
 namespace Debug
 {
 	void dump_stack_trace();
+	void dump_stack_trace(uintptr_t ip, uintptr_t bp);
+	void dump_qr_code();
+
 	void putchar(char);
 	void print_prefix(const char*, int);
 

kernel/include/kernel/Device/DebugDevice.h

@@ -3,20 +3,19 @@
 namespace Kernel
 {
 
-	class DebugDevice : public CharacterDevice
+	class DebugDevice final : public CharacterDevice
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<DebugDevice>> create(mode_t, uid_t, gid_t);
 
-		virtual dev_t rdev() const override { return m_rdev; }
-
 		virtual BAN::StringView name() const override { return "debug"_sv; }
 
 	protected:
 		DebugDevice(mode_t mode, uid_t uid, gid_t gid, dev_t rdev)
 			: CharacterDevice(mode, uid, gid)
-			, m_rdev(rdev)
-		{ }
+		{
+			m_rdev = rdev;
+		}
 
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override { return 0; }
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan buffer) override;
@@ -24,9 +23,7 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
-
-	private:
-		const dev_t m_rdev;
+		virtual bool has_hungup_impl() const override { return false; }
 	};
 
 }

kernel/include/kernel/Device/Device.h

@@ -12,20 +12,20 @@ namespace Kernel
 		virtual ~Device() = default;
 		virtual void update() {}
 
-		virtual bool is_device() const override { return true; }
-		virtual bool is_partition() const { return false; }
-		virtual bool is_storage_device() const { return false; }
-
-		virtual BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> mmap_region(PageTable&, off_t offset, size_t len, AddressRange, MemoryRegion::Type, PageTable::flags_t) { (void)offset; (void)len; return BAN::Error::from_errno(EINVAL); }
-
-		virtual dev_t rdev() const override = 0;
+		virtual BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> mmap_region(PageTable&, off_t offset, size_t len, AddressRange, MemoryRegion::Type, PageTable::flags_t, int status_flags)
+		{
+			(void)offset; (void)len; (void)status_flags;
+			return BAN::Error::from_errno(ENOTSUP);
+		}
 
 		virtual BAN::StringView name() const = 0;
 
 	protected:
 		Device(mode_t, uid_t, gid_t);
 
-		virtual BAN::ErrorOr<void> fsync_impl() final override { return BAN::Error::from_errno(EINVAL); }
+	private:
+		BAN::ErrorOr<void> sync_inode(SyncType) final override { return {}; }
+		BAN::ErrorOr<void> sync_data()          final override { return {}; }
 	};
 
 	class BlockDevice : public Device, public BAN::Weakable<BlockDevice>
@@ -41,7 +41,7 @@ namespace Kernel
 		BlockDevice(mode_t mode, uid_t uid, gid_t gid)
 			: Device(mode, uid, gid)
 		{
-			m_inode_info.mode |= Inode::Mode::IFBLK;
+			m_mode |= Inode::Mode::IFBLK;
 		}
 	};
 
@@ -51,7 +51,7 @@ namespace Kernel
 		CharacterDevice(mode_t mode, uid_t uid, gid_t gid)
 			: Device(mode, uid, gid)
 		{
-			m_inode_info.mode |= Inode::Mode::IFCHR;
+			m_mode |= Inode::Mode::IFCHR;
 		}
 	};
 

kernel/include/kernel/Device/DeviceNumbers.h

@@ -16,11 +16,14 @@ namespace Kernel
 		Debug,
 		Keyboard,
 		Mouse,
+		Joystick,
 		SCSI,
 		NVMeController,
 		NVMeNamespace,
 		Ethernet,
+		Loopback,
 		TmpFS,
+		AudioController,
 	};
 
 }

kernel/include/kernel/Device/FramebufferDevice.h

@@ -10,12 +10,15 @@ namespace Kernel
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<FramebufferDevice>> create_from_boot_framebuffer();
+		static BAN::RefPtr<FramebufferDevice> boot_framebuffer();
 		~FramebufferDevice();
 
 		uint32_t width() const { return m_width; }
 		uint32_t height() const { return m_height; }
 
+		uint32_t get_pixel(uint32_t x, uint32_t y) const;
 		void set_pixel(uint32_t x, uint32_t y, uint32_t rgb);
+		void fill(uint32_t rgb);
 
 		// positive rows -> empty pixels on bottom
 		// negative rows -> empty pixels on top
@@ -25,18 +28,20 @@ namespace Kernel
 		void sync_pixels_linear(uint32_t first_pixel, uint32_t pixel_count);
 		void sync_pixels_rectangle(uint32_t top_right_x, uint32_t top_right_y, uint32_t width, uint32_t height);
 
-		virtual BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> mmap_region(PageTable&, off_t offset, size_t len, AddressRange, MemoryRegion::Type, PageTable::flags_t) override;
+		virtual BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> mmap_region(PageTable&, off_t offset, size_t len, AddressRange, MemoryRegion::Type, PageTable::flags_t, int status_flags) override;
 
-		virtual dev_t rdev() const override { return m_rdev; }
 		virtual BAN::StringView name() const override { return m_name.sv(); }
 
 	protected:
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
 
+		BAN::ErrorOr<long> ioctl_impl(int cmd, void* arg) override;
+
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		FramebufferDevice(mode_t mode, uid_t uid, gid_t gid, dev_t rdev, paddr_t paddr, uint32_t width, uint32_t height, uint32_t pitch, uint8_t bpp);
@@ -44,7 +49,6 @@ namespace Kernel
 
 	private:
 		const BAN::String m_name;
-		const dev_t m_rdev;
 
 		vaddr_t			m_video_memory_vaddr { 0 };
 		const paddr_t	m_video_memory_paddr;

kernel/include/kernel/Device/NullDevice.h

@@ -5,20 +5,19 @@
 namespace Kernel
 {
 
-	class NullDevice : public CharacterDevice
+	class NullDevice final : public CharacterDevice
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<NullDevice>> create(mode_t, uid_t, gid_t);
 
-		virtual dev_t rdev() const override { return m_rdev; }
-
 		virtual BAN::StringView name() const override { return "null"_sv; }
 
 	protected:
 		NullDevice(mode_t mode, uid_t uid, gid_t gid, dev_t rdev)
 			: CharacterDevice(mode, uid, gid)
-			, m_rdev(rdev)
-		{ }
+		{
+			m_rdev = rdev;
+		}
 
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override { return 0; }
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan buffer) override { return buffer.size(); };
@@ -26,9 +25,8 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
-	private:
-		const dev_t m_rdev;
 	};
 
 }

kernel/include/kernel/Device/RandomDevice.h

@@ -3,20 +3,19 @@
 namespace Kernel
 {
 
-	class RandomDevice : public CharacterDevice
+	class RandomDevice final : public CharacterDevice
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<RandomDevice>> create(mode_t, uid_t, gid_t);
 
-		virtual dev_t rdev() const override { return m_rdev; }
-
 		virtual BAN::StringView name() const override { return "random"_sv; }
 
 	protected:
 		RandomDevice(mode_t mode, uid_t uid, gid_t gid, dev_t rdev)
 			: CharacterDevice(mode, uid, gid)
-			, m_rdev(rdev)
-		{ }
+		{
+			m_rdev = rdev;
+		}
 
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan buffer) override { return buffer.size(); };
@@ -24,9 +23,7 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
-
-	private:
-		const dev_t m_rdev;
+		virtual bool has_hungup_impl() const override { return false; }
 	};
 
 }

kernel/include/kernel/Device/ZeroDevice.h

@@ -3,20 +3,19 @@
 namespace Kernel
 {
 
-	class ZeroDevice : public CharacterDevice
+	class ZeroDevice final : public CharacterDevice
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<ZeroDevice>> create(mode_t, uid_t, gid_t);
 
-		virtual dev_t rdev() const override { return m_rdev; }
-
 		virtual BAN::StringView name() const override { return "zero"_sv; }
 
 	protected:
 		ZeroDevice(mode_t mode, uid_t uid, gid_t gid, dev_t rdev)
 			: CharacterDevice(mode, uid, gid)
-			, m_rdev(rdev)
-		{ }
+		{
+			m_rdev = rdev;
+		}
 
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan buffer) override { return buffer.size(); };
@@ -24,9 +23,7 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
-
-	private:
-		const dev_t m_rdev;
+		virtual bool has_hungup_impl() const override { return false; }
 	};
 
 }

kernel/include/kernel/ELF.h

@@ -20,6 +20,6 @@ namespace Kernel::ELF
 		BAN::Vector<BAN::UniqPtr<MemoryRegion>> regions;
 	};
 
-	BAN::ErrorOr<LoadResult> load_from_inode(BAN::RefPtr<Inode>, const Credentials&, PageTable&);
+	BAN::ErrorOr<LoadResult> load_from_inode(BAN::RefPtr<Inode> root, BAN::RefPtr<Inode> inode, const Credentials&, PageTable&);
 
 }

kernel/include/kernel/Epoll.h

@@ -0,0 +1,86 @@
+#pragma once
+
+#include <BAN/HashMap.h>
+#include <kernel/FS/Inode.h>
+#include <kernel/Lock/Mutex.h>
+
+#include <sys/epoll.h>
+
+namespace Kernel
+{
+
+	class Epoll final : public Inode
+	{
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<Epoll>> create();
+		~Epoll();
+
+		BAN::ErrorOr<void> ctl(int op, int fd, BAN::RefPtr<Inode> inode, epoll_event event);
+		BAN::ErrorOr<size_t> wait(BAN::Span<epoll_event> events, uint64_t waketime_ns);
+
+		void notify(BAN::RefPtr<Inode> inode, uint32_t event);
+
+	private:
+		Epoll();
+
+		const FileSystem* filesystem() const override { return nullptr; }
+
+		bool can_read_impl() const override { return false; }
+		bool can_write_impl() const override { return false; }
+		bool has_error_impl() const override { return false; }
+		bool has_hungup_impl() const override { return false; }
+
+		BAN::ErrorOr<void> sync_inode(SyncType) override { return {}; }
+		BAN::ErrorOr<void> sync_data() override { return {}; }
+
+	private:
+		struct ListenEventList
+		{
+			ListenEventList() = default;
+
+			ListenEventList(const ListenEventList&) = delete;
+			ListenEventList& operator=(const ListenEventList&) = delete;
+
+			ListenEventList(ListenEventList&& other)
+				: events(BAN::move(other.events))
+			{}
+			ListenEventList& operator=(ListenEventList&& other)
+			{
+				events = BAN::move(other.events);
+				return *this;
+			}
+
+			BAN::HashMap<int, epoll_event> events;
+
+			bool has_fd(int fd) const
+			{
+				return events.contains(fd);
+			}
+
+			bool empty() const
+			{
+				return events.empty();
+			}
+
+			BAN::ErrorOr<void> add_fd(int fd, epoll_event event)
+			{
+				TRY(events.insert(fd, event));
+				return {};
+			}
+
+			void remove_fd(int fd)
+			{
+				events.remove(fd);
+			}
+		};
+
+	private:
+		Mutex m_mutex;
+		ThreadBlocker m_thread_blocker;
+		SpinLock m_ready_lock;
+		BAN::HashMap<BAN::RefPtr<Inode>, uint32_t> m_ready_events;
+		BAN::HashMap<BAN::RefPtr<Inode>, uint32_t> m_processing_events;
+		BAN::HashMap<BAN::RefPtr<Inode>, ListenEventList> m_listening_events;
+	};
+
+}

kernel/include/kernel/FS/DevFS/FileSystem.h

@@ -22,6 +22,7 @@ namespace Kernel
 
 		void add_inode(BAN::StringView path, BAN::RefPtr<TmpInode>);
 
+		void initiate_disk_cache_drop();
 		void initiate_sync(bool should_block);
 
 	private:
@@ -37,6 +38,10 @@ namespace Kernel
 		ThreadBlocker m_sync_done;
 		ThreadBlocker m_sync_thread_blocker;
 		volatile bool m_should_sync { false };
+
+		SpinLock m_disk_cache_lock;
+		ThreadBlocker m_disk_cache_thread_blocker;
+		BAN::Atomic<bool> m_should_drop_disk_cache { false };
 	};
 
 }

kernel/include/kernel/FS/EventFD.h

@@ -0,0 +1,38 @@
+#pragma once
+
+#include <kernel/FS/Inode.h>
+#include <kernel/Lock/Mutex.h>
+
+namespace Kernel
+{
+
+	class EventFD final : public Inode
+	{
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<Inode>> create(uint64_t initval, bool semaphore);
+
+	private:
+		EventFD(uint64_t initval, bool is_semaphore);
+
+		const FileSystem* filesystem() const override { return nullptr; }
+
+		BAN::ErrorOr<void> sync_inode(SyncType) override { return {}; }
+		BAN::ErrorOr<void> sync_data() override { return {}; }
+
+		BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
+		BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
+
+		bool can_read_impl() const override { return m_value > 0; }
+		bool can_write_impl() const override { return m_value < UINT64_MAX - 1; }
+		bool has_error_impl() const override { return false; }
+		bool has_hungup_impl() const override { return false; }
+
+	private:
+		const bool m_is_semaphore;
+		BAN::Atomic<uint64_t> m_value;
+
+		Mutex m_mutex;
+		ThreadBlocker m_thread_blocker;
+	};
+
+}

kernel/include/kernel/FS/Ext2/Definitions.h

@@ -79,6 +79,12 @@ namespace Kernel::Ext2
 		uint8_t __reserved[12];
 	};
 
+	struct InodeBlocks {
+		uint32_t block[15];
+	};
+	struct Osd2 {
+		uint32_t osd2[3];
+	};
 	struct Inode
 	{
 		uint16_t mode;
@@ -93,12 +99,12 @@ namespace Kernel::Ext2
 		uint32_t blocks;
 		uint32_t flags;
 		uint32_t osd1;
-		uint32_t block[15];
+		InodeBlocks block;
 		uint32_t generation;
 		uint32_t file_acl;
 		uint32_t dir_acl;
 		uint32_t faddr;
-		uint32_t osd2[3];
+		Osd2 osd2;
 	};
 
 	struct LinkedDirectoryEntry

kernel/include/kernel/FS/Ext2/FileSystem.h

@@ -26,18 +26,32 @@ namespace Kernel
 		class BlockBufferWrapper
 		{
 			BAN_NON_COPYABLE(BlockBufferWrapper);
-			BAN_NON_MOVABLE(BlockBufferWrapper);
 
 		public:
-			BlockBufferWrapper(BAN::Span<uint8_t> buffer, bool& used)
+			BlockBufferWrapper(BAN::Span<uint8_t> buffer, void (*callback)(void*, const uint8_t*), void* argument)
 				: m_buffer(buffer)
-				, m_used(used)
-			{
-				ASSERT(m_used);
-			}
+				, m_callback(callback)
+				, m_argument(argument)
+			{ }
+			BlockBufferWrapper(BlockBufferWrapper&& other) { *this = BAN::move(other); }
 			~BlockBufferWrapper()
 			{
-				m_used = false;
+				if (m_callback == nullptr)
+					return;
+				m_callback(m_argument, m_buffer.data());
+			}
+
+			BlockBufferWrapper& operator=(BlockBufferWrapper&& other)
+			{
+				this->m_buffer = other.m_buffer;
+				this->m_callback = other.m_callback;
+				this->m_argument = other.m_argument;
+
+				other.m_buffer = {};
+				other.m_callback = nullptr;
+				other.m_argument = nullptr;
+
+				return *this;
 			}
 
 			size_t size() const { return m_buffer.size(); }
@@ -53,7 +67,8 @@ namespace Kernel
 
 		private:
 			BAN::Span<uint8_t> m_buffer;
-			bool& m_used;
+			void (*m_callback)(void*, const uint8_t*);
+			void* m_argument;
 		};
 
 	public:
@@ -62,8 +77,6 @@ namespace Kernel
 
 		virtual BAN::RefPtr<Inode> root_inode() override { return m_root_inode; }
 
-		virtual dev_t dev() const override { return m_block_device->rdev(); };
-
 	private:
 		Ext2FS(BAN::RefPtr<BlockDevice> block_device)
 			: m_block_device(block_device)
@@ -81,12 +94,13 @@ namespace Kernel
 		BAN::ErrorOr<void> sync_superblock();
 		BAN::ErrorOr<void> sync_block(uint32_t block);
 
-		BlockBufferWrapper get_block_buffer();
+		BAN::ErrorOr<BlockBufferWrapper> get_block_buffer();
 
 		BAN::ErrorOr<uint32_t> reserve_free_block(uint32_t primary_bgd);
 		BAN::ErrorOr<void> release_block(uint32_t block);
 
-		BAN::HashMap<ino_t, BAN::RefPtr<Ext2Inode>>& inode_cache() { return m_inode_cache; }
+		BAN::ErrorOr<BAN::RefPtr<Ext2Inode>> open_inode(ino_t);
+		void remove_from_cache(ino_t);
 
 		const Ext2::Superblock& superblock() const { return m_superblock; }
 
@@ -104,10 +118,13 @@ namespace Kernel
 		{
 		public:
 			BlockBufferManager() = default;
-			BlockBufferWrapper get_buffer();
+			BAN::ErrorOr<BlockBufferWrapper> get_buffer();
 
 			BAN::ErrorOr<void> initialize(size_t block_size);
 
+		private:
+			void destroy_callback(const uint8_t* buffer_ptr);
+
 		private:
 			struct BlockBuffer
 			{
@@ -115,8 +132,20 @@ namespace Kernel
 				bool used { false };
 			};
 
+			struct ThreadInfo
+			{
+				pid_t tid { 0 };
+				size_t buffers { 0 };
+			};
+
 		private:
-			BAN::Array<BlockBuffer, 10> m_buffers;
+			static constexpr size_t max_threads = 8;
+			static constexpr size_t max_buffers_per_thread = 6;
+
+			Mutex m_buffer_mutex;
+			ThreadBlocker m_buffer_blocker;
+			BAN::Array<BlockBuffer, max_threads * max_buffers_per_thread> m_buffers;
+			BAN::Array<ThreadInfo, max_threads> m_thread_infos;
 		};
 
 	private:
@@ -127,6 +156,7 @@ namespace Kernel
 		BAN::RefPtr<Inode> m_root_inode;
 		BAN::Vector<uint32_t> m_superblock_backups;
 
+		Mutex m_inode_cache_lock;
 		BAN::HashMap<ino_t, BAN::RefPtr<Ext2Inode>> m_inode_cache;
 
 		BlockBufferManager m_buffer_manager;

kernel/include/kernel/FS/Ext2/Inode.h

@@ -4,6 +4,7 @@
 #include <BAN/StringView.h>
 #include <kernel/FS/Ext2/Definitions.h>
 #include <kernel/FS/Inode.h>
+#include <kernel/Lock/RWLock.h>
 
 namespace Kernel
 {
@@ -15,28 +16,18 @@ namespace Kernel
 	public:
 		~Ext2Inode();
 
-		virtual ino_t ino() const override { return m_ino; };
-		virtual Mode mode() const override { return { m_inode.mode }; }
-		virtual nlink_t nlink() const override { return m_inode.links_count; }
-		virtual uid_t uid() const override { return m_inode.uid; }
-		virtual gid_t gid() const override { return m_inode.gid; }
-		virtual off_t size() const override { return m_inode.size; }
-		virtual timespec atime() const override { return timespec { .tv_sec = m_inode.atime, .tv_nsec = 0 }; }
-		virtual timespec mtime() const override { return timespec { .tv_sec = m_inode.mtime, .tv_nsec = 0 }; }
-		virtual timespec ctime() const override { return timespec { .tv_sec = m_inode.ctime, .tv_nsec = 0 }; }
-		virtual blksize_t blksize() const override;
-		virtual blkcnt_t blocks() const override;
-		virtual dev_t dev() const override { return 0; }
-		virtual dev_t rdev() const override { return 0; }
-
 		virtual const FileSystem* filesystem() const override;
 
-	protected:
+	private:
+		virtual BAN::ErrorOr<void> sync_inode(SyncType) override;
+		virtual BAN::ErrorOr<void> sync_data() override;
+
 		virtual BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_impl(BAN::StringView) override;
 		virtual BAN::ErrorOr<size_t> list_next_inodes_impl(off_t, struct dirent*, size_t) override;
 		virtual BAN::ErrorOr<void> create_file_impl(BAN::StringView, mode_t, uid_t, gid_t) override;
 		virtual BAN::ErrorOr<void> create_directory_impl(BAN::StringView, mode_t, uid_t, gid_t) override;
 		virtual BAN::ErrorOr<void> link_inode_impl(BAN::StringView, BAN::RefPtr<Inode>) override;
+		virtual BAN::ErrorOr<void> rename_inode_impl(BAN::RefPtr<Inode>, BAN::StringView, BAN::StringView) override;
 		virtual BAN::ErrorOr<void> unlink_impl(BAN::StringView) override;
 
 		virtual BAN::ErrorOr<BAN::String> link_target_impl() override;
@@ -45,46 +36,103 @@ namespace Kernel
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
 		virtual BAN::ErrorOr<void> truncate_impl(size_t) override;
-		virtual BAN::ErrorOr<void> chmod_impl(mode_t) override;
-		virtual BAN::ErrorOr<void> fsync_impl() override;
 
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
+		uint32_t block_group() const;
+
 		// Returns maximum number of data blocks in use
 		// NOTE: the inode might have more blocks than what this suggests if it has been shrinked
 		uint32_t max_used_data_block_count() const { return size() / blksize(); }
 
-		BAN::ErrorOr<BAN::Optional<uint32_t>> block_from_indirect_block(uint32_t block, uint32_t index, uint32_t depth);
-		BAN::ErrorOr<BAN::Optional<uint32_t>> fs_block_of_data_block_index(uint32_t data_block_index);
+		BAN::ErrorOr<void> sync_inode_no_lock();
 
-		BAN::ErrorOr<void> link_inode_to_directory(Ext2Inode&, BAN::StringView name);
-		BAN::ErrorOr<bool> is_directory_empty();
+		BAN::ErrorOr<bool> is_directory_empty_no_lock();
+		BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_no_lock(BAN::StringView);
 
-		BAN::ErrorOr<void> cleanup_indirect_block(uint32_t block, uint32_t depth);
-		BAN::ErrorOr<void> cleanup_default_links();
-		BAN::ErrorOr<void> cleanup_from_fs();
+		/* needs write end of the lock when allocate is true*/
+		BAN::ErrorOr<BAN::Optional<uint32_t>> block_from_indirect_block_no_lock(uint32_t block, uint32_t index, uint32_t depth, bool allocate);
+		BAN::ErrorOr<BAN::Optional<uint32_t>> fs_block_of_data_block_index_no_lock(uint32_t data_block_index, bool allocate);
 
-		BAN::ErrorOr<uint32_t> allocate_new_block_to_indirect_block(uint32_t& block, uint32_t index, uint32_t depth);
-		BAN::ErrorOr<uint32_t> allocate_new_block(uint32_t data_block_index);
-		BAN::ErrorOr<void> sync();
+		/* needs write end of the lock */
+		BAN::ErrorOr<void> link_inode_to_directory_no_lock(Ext2Inode&, BAN::StringView name);
+		BAN::ErrorOr<void> remove_inode_from_directory_no_lock(BAN::StringView name, bool cleanup_directory);
 
-		uint32_t block_group() const;
+		/* needs write end of the lock */
+		BAN::ErrorOr<void> cleanup_indirect_block_no_lock(uint32_t block, uint32_t depth);
+		BAN::ErrorOr<void> cleanup_default_links_no_lock();
+		BAN::ErrorOr<void> cleanup_data_blocks_no_lock();
+		BAN::ErrorOr<void> cleanup_from_fs_no_lock();
 
 	private:
-		Ext2Inode(Ext2FS& fs, Ext2::Inode inode, uint32_t ino)
-			: m_fs(fs)
-			, m_inode(inode)
-			, m_ino(ino)
+		Ext2Inode(Ext2FS& fs, Ext2::Inode inode, uint32_t ino);
+
+		BAN::Optional<uint32_t> block_cache_find(uint32_t block, uint32_t index) const;
+		void block_cache_remove(uint32_t block, uint32_t index);
+		void block_cache_add(uint32_t block, uint32_t index, uint32_t target);
+
+		BAN::RefPtr<Inode> dir_cache_find(BAN::StringView) const;
+		void dir_cache_remove(BAN::StringView);
+		void dir_cache_add(BAN::StringView, BAN::RefPtr<Inode>);
+
+	private:
+		struct ScopedSync
+		{
+			ScopedSync(Ext2Inode& inode)
+				: inode(inode)
 			{ }
-		static BAN::ErrorOr<BAN::RefPtr<Ext2Inode>> create(Ext2FS&, uint32_t);
+
+			~ScopedSync()
+			{
+				// TODO: there was some memcmp smarty pants stuff here.
+				// How do we wanna approach it?
+				if (auto ret = inode.sync_inode_no_lock(); ret.is_error())
+					dwarnln("failed to sync inode: {}", ret.error());
+			}
+
+			Ext2Inode& inode;
+		};
 
 	private:
 		Ext2FS& m_fs;
-		Ext2::Inode m_inode;
-		const uint32_t m_ino;
+		RWLock m_lock;
+
+		Ext2::InodeBlocks m_ext2_blocks;
+		// NOTE: some fields from the original disk inode
+		// that we do not use, but we keep for serialise.
+		const uint32_t m_og_dtime;
+		const uint32_t m_og_flags;
+		const uint32_t m_og_osd1;
+		const uint32_t m_og_generation;
+		const uint32_t m_og_file_acl;
+		const uint32_t m_og_dir_acl;
+		const uint32_t m_og_faddr;
+		const Ext2::Osd2 m_og_osd2;
+
+		struct BlockCacheEntry
+		{
+			mutable uint32_t freq;
+			uint32_t block;
+			uint32_t index;
+			uint32_t target;
+		};
+		mutable SpinLock m_block_cache_lock;
+		BAN::Array<BlockCacheEntry, 8> m_block_cache;
+
+		struct DirCacheEntry
+		{
+			mutable size_t freq { 0 };
+			BAN::RefPtr<Inode> inode;
+			size_t name_len { 0 };
+			char   name[256];
+		};
+		static constexpr size_t dir_cache_size = 32;
+		mutable RWLock m_dir_cache_lock;
+		BAN::Vector<DirCacheEntry> m_dir_cache;
 
 		friend class Ext2FS;
 		friend class BAN::RefPtr<Ext2Inode>;

kernel/include/kernel/FS/FAT/FileSystem.h

@@ -37,8 +37,6 @@ namespace Kernel
 
 		virtual BAN::RefPtr<Inode> root_inode() override { return m_root_inode; }
 
-		virtual dev_t dev() const override { return m_block_device->rdev(); };
-
 		BAN::ErrorOr<BAN::RefPtr<FATInode>> open_inode(BAN::RefPtr<FATInode> parent, const FAT::DirectoryEntry& entry, uint32_t cluster_index, uint32_t entry_index);
 		BAN::ErrorOr<void> inode_read_cluster(BAN::RefPtr<FATInode>, size_t index, BAN::ByteSpan buffer);
 		blksize_t inode_block_size(BAN::RefPtr<const FATInode>) const;

kernel/include/kernel/FS/FAT/Inode.h

@@ -15,25 +15,14 @@ namespace Kernel
 	class FATInode final : public Inode, public BAN::Weakable<FATInode>
 	{
 	public:
-		virtual ino_t ino() const override { return m_ino; };
-		virtual Mode mode() const override { return Mode { ((m_entry.attr & FAT::FileAttr::DIRECTORY) ? Mode::IFDIR : Mode::IFREG) | 0777 }; }
-		virtual nlink_t nlink() const override { return 1; }
-		virtual uid_t uid() const override { return 0; }
-		virtual gid_t gid() const override { return 0; }
-		virtual off_t size() const override { return m_entry.file_size; }
-		virtual timespec atime() const override;
-		virtual timespec mtime() const override;
-		virtual timespec ctime() const override;
-		virtual blksize_t blksize() const override;
-		virtual blkcnt_t blocks() const override { return m_block_count; }
-		virtual dev_t dev() const override { return 0; }
-		virtual dev_t rdev() const override { return 0; }
-
 		virtual const FileSystem* filesystem() const override;
 
 		const FAT::DirectoryEntry& entry() const { return m_entry; }
 
-	protected:
+	private:
+		virtual BAN::ErrorOr<void> sync_inode(SyncType) override { return {}; }
+		virtual BAN::ErrorOr<void> sync_data() override { return {}; }
+
 		virtual BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_impl(BAN::StringView) override;
 		virtual BAN::ErrorOr<size_t> list_next_inodes_impl(off_t, struct dirent*, size_t) override;
 		//virtual BAN::ErrorOr<void> create_file_impl(BAN::StringView, mode_t, uid_t, gid_t) override;
@@ -43,20 +32,15 @@ namespace Kernel
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		//virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
 		//virtual BAN::ErrorOr<void> truncate_impl(size_t) override;
-		//virtual BAN::ErrorOr<void> chmod_impl(mode_t) override;
-		virtual BAN::ErrorOr<void> fsync_impl() override { return {}; }
 
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
-		FATInode(FATFS& fs, const FAT::DirectoryEntry& entry, ino_t ino, uint32_t block_count)
-			: m_fs(fs)
-			, m_entry(entry)
-			, m_ino(ino)
-			, m_block_count(block_count)
-		{ }
+		FATInode(FATFS& fs, const FAT::DirectoryEntry& entry, ino_t ino, uint32_t block_count);
+
 		~FATInode()	{}
 
 		BAN::ErrorOr<void> for_each_directory_entry(BAN::ConstByteSpan, BAN::Function<BAN::Iteration(const FAT::DirectoryEntry&)>);
@@ -65,7 +49,6 @@ namespace Kernel
 	private:
 		FATFS& m_fs;
 		FAT::DirectoryEntry m_entry;
-		const ino_t m_ino;
 		uint32_t m_block_count;
 
 		friend class Ext2FS;

kernel/include/kernel/FS/FileSystem.h

@@ -26,8 +26,6 @@ namespace Kernel
 		static BAN::ErrorOr<BAN::RefPtr<FileSystem>> from_block_device(BAN::RefPtr<BlockDevice>);
 
 		virtual BAN::RefPtr<Inode> root_inode() = 0;
-
-		virtual dev_t dev() const = 0;
 	};
 
 }

kernel/include/kernel/FS/Inode.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <BAN/ByteSpan.h>
+#include <BAN/LinkedList.h>
 #include <BAN/RefPtr.h>
 #include <BAN/String.h>
 #include <BAN/StringView.h>
@@ -9,7 +10,6 @@
 
 #include <kernel/Credentials.h>
 #include <kernel/Debug.h>
-#include <kernel/Lock/Mutex.h>
 
 #include <dirent.h>
 #include <sys/socket.h>
@@ -19,10 +19,9 @@
 namespace Kernel
 {
 
-	class FileSystem;
-
 	class FileBackedRegion;
-	class SharedFileData;
+	class FileSystem;
+	struct SharedFileData;
 
 	class Inode : public BAN::RefCounted<Inode>
 	{
@@ -63,6 +62,24 @@ namespace Kernel
 			mode_t mode;
 		};
 
+		enum InodeKind : uint8_t
+		{
+			DEVICE    = 0x01,
+			EPOLL     = 0x02,
+			PIPE      = 0x04,
+			TTY       = 0x08,
+			PARTITION = 0x10,
+			STORAGE   = 0x20,
+		};
+
+		enum class SyncType
+		{
+			General,
+			Mode,
+			UidGid,
+			Times,
+		};
+
 	public:
 		virtual ~Inode() {}
 
@@ -70,23 +87,26 @@ namespace Kernel
 
 		bool operator==(const Inode& other) const { return dev() == other.dev() && ino() == other.ino(); }
 
-		virtual ino_t ino() const = 0;
-		virtual Mode mode() const = 0;
-		virtual nlink_t nlink() const = 0;
-		virtual uid_t uid() const = 0;
-		virtual gid_t gid() const = 0;
-		virtual off_t size() const = 0;
-		virtual timespec atime() const = 0;
-		virtual timespec mtime() const = 0;
-		virtual timespec ctime() const = 0;
-		virtual blksize_t blksize() const = 0;
-		virtual blkcnt_t blocks() const = 0;
-		virtual dev_t dev() const = 0;
-		virtual dev_t rdev() const = 0;
+		ino_t ino() const { return m_ino; }
+		Mode mode() const { return Mode(m_mode); }
+		nlink_t nlink() const { return m_nlink; }
+		uid_t uid() const { return m_uid; }
+		gid_t gid() const { return m_gid; }
+		off_t size() const { return m_size; }
+		timespec atime() const { return m_atime; }
+		timespec mtime() const { return m_mtime; }
+		timespec ctime() const { return m_ctime; }
+		blksize_t blksize() const { return m_blksize; }
+		blkcnt_t blocks() const { return m_blocks; }
+		dev_t dev() const { return m_dev; }
+		dev_t rdev() const { return m_rdev; }
 
-		virtual bool is_device() const { return false; }
-		virtual bool is_pipe() const { return false; }
-		virtual bool is_tty() const { return false; }
+		bool is_device()         const { return m_kind & InodeKind::DEVICE;    }
+		bool is_epoll()          const { return m_kind & InodeKind::EPOLL;     }
+		bool is_pipe()           const { return m_kind & InodeKind::PIPE;      }
+		bool is_tty()            const { return m_kind & InodeKind::TTY;       }
+		bool is_partition()      const { return m_kind & InodeKind::PARTITION; }
+		bool is_storage_device() const { return m_kind & InodeKind::STORAGE;   }
 
 		virtual const FileSystem* filesystem() const = 0;
 
@@ -96,6 +116,7 @@ namespace Kernel
 		BAN::ErrorOr<void> create_file(BAN::StringView, mode_t, uid_t, gid_t);
 		BAN::ErrorOr<void> create_directory(BAN::StringView, mode_t, uid_t, gid_t);
 		BAN::ErrorOr<void> link_inode(BAN::StringView, BAN::RefPtr<Inode>);
+		BAN::ErrorOr<void> rename_inode(BAN::RefPtr<Inode>, BAN::StringView, BAN::StringView);
 		BAN::ErrorOr<void> unlink(BAN::StringView);
 
 		// Link API
@@ -107,9 +128,12 @@ namespace Kernel
 		BAN::ErrorOr<void> bind(const sockaddr* address, socklen_t address_len);
 		BAN::ErrorOr<void> connect(const sockaddr* address, socklen_t address_len);
 		BAN::ErrorOr<void> listen(int backlog);
-		BAN::ErrorOr<size_t> sendto(BAN::ConstByteSpan message, const sockaddr* address, socklen_t address_len);
-		BAN::ErrorOr<size_t> recvfrom(BAN::ByteSpan buffer, sockaddr* address, socklen_t* address_len);
+		BAN::ErrorOr<size_t> sendmsg(const msghdr& message, int flags);
+		BAN::ErrorOr<size_t> recvmsg(msghdr& message, int flags);
 		BAN::ErrorOr<void> getsockname(sockaddr* address, socklen_t* address_len);
+		BAN::ErrorOr<void> getpeername(sockaddr* address, socklen_t* address_len);
+		BAN::ErrorOr<void> getsockopt(int level, int option, void* value, socklen_t* value_len);
+		BAN::ErrorOr<void> setsockopt(int level, int option, const void* value, socklen_t value_len);
 
 		// General API
 		BAN::ErrorOr<size_t> read(off_t, BAN::ByteSpan buffer);
@@ -117,15 +141,27 @@ namespace Kernel
 		BAN::ErrorOr<void> truncate(size_t);
 		BAN::ErrorOr<void> chmod(mode_t);
 		BAN::ErrorOr<void> chown(uid_t, gid_t);
+		BAN::ErrorOr<void> utimens(const timespec[2]);
 		BAN::ErrorOr<void> fsync();
 
 		// Select/Non blocking API
-		bool can_read() const;
-		bool can_write() const;
-		bool has_error() const;
+		bool can_read() const   { return can_read_impl(); }
+		bool can_write() const  { return can_write_impl(); }
+		bool has_error() const  { return has_error_impl(); }
+		bool has_hungup() const { return has_hungup_impl(); }
 
 		BAN::ErrorOr<long> ioctl(int request, void* arg);
 
+		BAN::ErrorOr<void> add_epoll(class Epoll*);
+		void del_epoll(class Epoll*);
+		void epoll_notify(uint32_t event);
+
+		virtual void on_close(int status_flags) { (void)status_flags; }
+		virtual void on_clone(int status_flags) { (void)status_flags; }
+
+		virtual BAN::ErrorOr<void> sync_inode(SyncType) = 0;
+		virtual BAN::ErrorOr<void> sync_data() = 0;
+
 	protected:
 		// Directory API
 		virtual BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_impl(BAN::StringView)							{ return BAN::Error::from_errno(ENOTSUP); }
@@ -133,6 +169,7 @@ namespace Kernel
 		virtual BAN::ErrorOr<void> create_file_impl(BAN::StringView, mode_t, uid_t, gid_t)					{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> create_directory_impl(BAN::StringView, mode_t, uid_t, gid_t)				{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> link_inode_impl(BAN::StringView, BAN::RefPtr<Inode>)						{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<void> rename_inode_impl(BAN::RefPtr<Inode>, BAN::StringView, BAN::StringView)	{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> unlink_impl(BAN::StringView)												{ return BAN::Error::from_errno(ENOTSUP); }
 
 		// Link API
@@ -144,33 +181,58 @@ namespace Kernel
 		virtual BAN::ErrorOr<void> connect_impl(const sockaddr*, socklen_t)							{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> listen_impl(int)													{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> bind_impl(const sockaddr*, socklen_t)							{ return BAN::Error::from_errno(ENOTSUP); }
-		virtual BAN::ErrorOr<size_t> sendto_impl(BAN::ConstByteSpan, const sockaddr*, socklen_t)	{ return BAN::Error::from_errno(ENOTSUP); }
-		virtual BAN::ErrorOr<size_t> recvfrom_impl(BAN::ByteSpan, sockaddr*, socklen_t*)			{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<size_t> recvmsg_impl(msghdr&, int)										{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<size_t> sendmsg_impl(const msghdr&, int)								{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> getsockname_impl(sockaddr*, socklen_t*)							{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<void> getpeername_impl(sockaddr*, socklen_t*)							{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<void> getsockopt_impl(int, int, void*, socklen_t*)						{ return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<void> setsockopt_impl(int, int, const void*, socklen_t)				{ return BAN::Error::from_errno(ENOTSUP); }
 
 		// General API
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan)		{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan)	{ return BAN::Error::from_errno(ENOTSUP); }
 		virtual BAN::ErrorOr<void> truncate_impl(size_t)					{ return BAN::Error::from_errno(ENOTSUP); }
-		virtual BAN::ErrorOr<void> chmod_impl(mode_t)						{ return BAN::Error::from_errno(ENOTSUP); }
-		virtual BAN::ErrorOr<void> chown_impl(uid_t, gid_t)					{ return BAN::Error::from_errno(ENOTSUP); }
-		virtual BAN::ErrorOr<void> fsync_impl() = 0;
 
 		// Select/Non blocking API
 		virtual bool can_read_impl() const = 0;
 		virtual bool can_write_impl() const = 0;
 		virtual bool has_error_impl() const = 0;
+		virtual bool has_hungup_impl() const = 0;
 
 		virtual BAN::ErrorOr<long> ioctl_impl(int, void*) { return BAN::Error::from_errno(ENOTSUP); }
 
 	protected:
-		mutable PriorityMutex m_mutex;
+		// TODO: this is supposed to be const I guess?
+		// But the thing is I would have to refactor a big chunk of the codebase
+		// to add it as a parameter to Inode() soooooo yeah no, not doing that rn.
+		uint8_t m_kind = 0;
+
+		BAN::Atomic<ino_t>     m_ino;
+		BAN::Atomic<mode_t>    m_mode;
+		BAN::Atomic<nlink_t>   m_nlink;
+		BAN::Atomic<uid_t>     m_uid;
+		BAN::Atomic<gid_t>     m_gid;
+		BAN::Atomic<off_t>     m_size;
+		// TODO: make these guys atomic :)
+		timespec m_atime;
+		timespec m_mtime;
+		timespec m_ctime;
+		BAN::Atomic<blksize_t> m_blksize;
+		BAN::Atomic<blkcnt_t>  m_blocks;
+		BAN::Atomic<dev_t>     m_dev;
+		BAN::Atomic<dev_t>     m_rdev;
 
 	private:
+		SpinLock m_shared_region_lock;
 		BAN::WeakPtr<SharedFileData> m_shared_region;
+
+		SpinLock m_epoll_lock;
+		BAN::LinkedList<class Epoll*> m_epolls;
+
+		friend class Epoll;
 		friend class FileBackedRegion;
 		friend class OpenFileDescriptorSet;
-		friend class SharedFileData;
+		friend struct SharedFileData;
 		friend class TTY;
 	};
 

kernel/include/kernel/FS/Pipe.h

@@ -1,62 +1,53 @@
 #pragma once
 
-#include <BAN/Array.h>
 #include <kernel/FS/Inode.h>
+#include <kernel/Lock/Mutex.h>
+#include <kernel/Memory/ByteRingBuffer.h>
 #include <kernel/ThreadBlocker.h>
 
+#include <sys/stat.h>
+
 namespace Kernel
 {
 
-	class Pipe : public Inode
+	class Pipe final : public Inode, public BAN::Weakable<Pipe>
 	{
 	public:
-		static BAN::ErrorOr<BAN::RefPtr<Inode>> create(const Credentials&);
+		static BAN::ErrorOr<BAN::RefPtr<Inode>> open(BAN::RefPtr<Inode>, int status_flags);
+		static BAN::ErrorOr<BAN::RefPtr<Inode>> create(uid_t, gid_t);
+		~Pipe();
 
-		virtual bool is_pipe() const override { return true; }
-		void clone_writing();
-		void close_writing();
-
-		virtual ino_t ino() const override { return 0; } // FIXME
-		virtual Mode mode() const override { return { Mode::IFIFO | Mode::IRUSR | Mode::IWUSR }; }
-		virtual nlink_t nlink() const override { return 1; }
-		virtual uid_t uid() const override { return m_uid; }
-		virtual gid_t gid() const override { return m_gid; }
-		virtual off_t size() const override { return 0; }
-		virtual timespec atime() const override { return m_atime; }
-		virtual timespec mtime() const override { return m_mtime; }
-		virtual timespec ctime() const override { return m_ctime; }
-		virtual blksize_t blksize() const override { return 4096; }
-		virtual blkcnt_t blocks() const override { return 0; }
-		virtual dev_t dev() const override { return 0; } // FIXME
-		virtual dev_t rdev() const override { return 0; } // FIXME
+		void on_close(int status_flags) override;
+		void on_clone(int status_flags) override;
 
 		virtual const FileSystem* filesystem() const override { return nullptr; }
 
-	protected:
+	private:
+		virtual BAN::ErrorOr<void> sync_inode(SyncType) override;
+		virtual BAN::ErrorOr<void> sync_data() override;
+
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
-		virtual BAN::ErrorOr<void> fsync_impl() final override { return {}; }
+		virtual BAN::ErrorOr<void> truncate_impl(size_t) override;
 
-		virtual bool can_read_impl() const override { return m_buffer_size > 0; }
+		virtual bool can_read_impl() const override { return !m_buffer->empty(); }
 		virtual bool can_write_impl() const override { return true; }
-		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_error_impl() const override { return m_reading_count == 0; }
+		virtual bool has_hungup_impl() const override { return m_writing_count == 0; }
 
 	private:
-		Pipe(const Credentials&);
+		Pipe(const struct stat&);
 
 	private:
-		const uid_t m_uid;
-		const gid_t m_gid;
-		timespec m_atime {};
-		timespec m_mtime {};
-		timespec m_ctime {};
+		Mutex m_mutex;
 		ThreadBlocker m_thread_blocker;
 
-		BAN::Array<uint8_t, PAGE_SIZE> m_buffer;
-		BAN::Atomic<size_t> m_buffer_size { 0 };
-		size_t m_buffer_tail { 0 };
+		BAN::UniqPtr<ByteRingBuffer> m_buffer;
 
-		BAN::Atomic<uint32_t> m_writing_count { 1 };
+		BAN::Atomic<uint32_t> m_writing_count { 0 };
+		BAN::Atomic<uint32_t> m_reading_count { 0 };
+
+		BAN::RefPtr<Inode> m_named_inode;
 	};
 
 }

kernel/include/kernel/FS/ProcFS/FileSystem.h

@@ -13,6 +13,8 @@ namespace Kernel
 		static void initialize();
 		static ProcFileSystem& get();
 
+		void post_scheduler_initialize();
+
 		BAN::ErrorOr<void> on_process_create(Process&);
 		void on_process_delete(Process&);
 

kernel/include/kernel/FS/ProcFS/Inode.h

@@ -9,34 +9,27 @@ namespace Kernel
 
 	class ProcPidInode final : public TmpDirectoryInode
 	{
+		// FIXME: dynamically update ruid/rgid.
+		// Possibly just have a magic uid/gid of -1 or something
+		// which means use current process ID
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<ProcPidInode>> create_new(Process&, TmpFileSystem&, mode_t);
 		~ProcPidInode() = default;
 
-		virtual uid_t uid() const override { return m_process.credentials().ruid(); }
-		virtual gid_t gid() const override { return m_process.credentials().rgid(); }
-
 		void cleanup();
-
 	protected:
 		virtual BAN::ErrorOr<void> unlink_impl(BAN::StringView) override { return BAN::Error::from_errno(EPERM); }
 
 	private:
 		ProcPidInode(Process&, TmpFileSystem&, const TmpInodeInfo&);
-
-	private:
-		Process& m_process;
 	};
 
 	class ProcROProcessInode final : public TmpInode
 	{
+		//FIXME: dynamically update ruid/rgid
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<ProcROProcessInode>> create_new(Process&, size_t (Process::*callback)(off_t, BAN::ByteSpan) const, TmpFileSystem&, mode_t);
 		~ProcROProcessInode() = default;
-
-		virtual uid_t uid() const override { return m_process.credentials().ruid(); }
-		virtual gid_t gid() const override { return m_process.credentials().rgid(); }
-
 	protected:
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 
@@ -47,6 +40,7 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		ProcROProcessInode(Process&, size_t (Process::*)(off_t, BAN::ByteSpan) const, TmpFileSystem&, const TmpInodeInfo&);
@@ -56,10 +50,33 @@ namespace Kernel
 		size_t (Process::*m_callback)(off_t, BAN::ByteSpan) const;
 	};
 
+	class ProcSymlinkProcessInode final : public TmpInode
+	{
+		//FIXME: dynamically update ruid/rgid
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<ProcSymlinkProcessInode>> create_new(Process& process, BAN::ErrorOr<BAN::String> (Process::*callback)() const, TmpFileSystem&, mode_t);
+		~ProcSymlinkProcessInode() = default;
+
+	protected:
+		virtual BAN::ErrorOr<BAN::String> link_target_impl() override;
+
+		virtual bool can_read_impl() const override { return false; }
+		virtual bool can_write_impl() const override { return false; }
+		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
+
+	private:
+		ProcSymlinkProcessInode(Process& process, BAN::ErrorOr<BAN::String> (Process::*)() const, TmpFileSystem&, const TmpInodeInfo&);
+
+	private:
+		Process& m_process;
+		BAN::ErrorOr<BAN::String> (Process::*m_callback)() const;
+	};
+
 	class ProcROInode final : public TmpInode
 	{
 	public:
-		static BAN::ErrorOr<BAN::RefPtr<ProcROInode>> create_new(size_t (*callback)(off_t, BAN::ByteSpan), TmpFileSystem&, mode_t, uid_t, gid_t);
+		static BAN::ErrorOr<BAN::RefPtr<ProcROInode>> create_new(BAN::ErrorOr<size_t> (*callback)(off_t, BAN::ByteSpan, void*), TmpFileSystem&, void*, mode_t, uid_t, gid_t);
 		~ProcROInode() = default;
 
 	protected:
@@ -72,12 +89,64 @@ namespace Kernel
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
-		ProcROInode(size_t (*callback)(off_t, BAN::ByteSpan), TmpFileSystem&, const TmpInodeInfo&);
+		ProcROInode(BAN::ErrorOr<size_t> (*callback)(off_t, BAN::ByteSpan, void*), TmpFileSystem&, void*, const TmpInodeInfo&);
 
 	private:
-		size_t (*m_callback)(off_t, BAN::ByteSpan);
+		BAN::ErrorOr<size_t> (*m_callback)(off_t, BAN::ByteSpan, void*);
+		void* m_argument;
+	};
+
+	class ProcSymlinkInode final : public TmpInode
+	{
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<ProcSymlinkInode>> create_new(BAN::ErrorOr<BAN::String> (*)(void*), void (*)(void*), void* data, TmpFileSystem&, mode_t, uid_t, gid_t);
+		~ProcSymlinkInode();
+
+	protected:
+		virtual BAN::ErrorOr<BAN::String> link_target_impl() override;
+
+		virtual bool can_read_impl() const override { return false; }
+		virtual bool can_write_impl() const override { return false; }
+		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
+
+	private:
+		ProcSymlinkInode(BAN::ErrorOr<BAN::String> (*callback)(void*), void (*destructor)(void*), void* data, TmpFileSystem&, const TmpInodeInfo&);
+
+	private:
+		BAN::ErrorOr<BAN::String> (*m_callback)(void*);
+		void (*m_destructor)(void*);
+		void* m_data;
+	};
+
+	class ProcFDDirectoryInode final : public TmpInode
+	{
+		//FIXME: dynamically update ruid/rgid
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<ProcFDDirectoryInode>> create_new(Process&, TmpFileSystem&, mode_t);
+		~ProcFDDirectoryInode() = default;
+	protected:
+		virtual BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_impl(BAN::StringView) override;
+		virtual BAN::ErrorOr<size_t> list_next_inodes_impl(off_t, struct dirent*, size_t) override;
+		virtual BAN::ErrorOr<void> create_file_impl(BAN::StringView, mode_t, uid_t, gid_t) override                 { return BAN::Error::from_errno(EPERM); }
+		virtual BAN::ErrorOr<void> create_directory_impl(BAN::StringView, mode_t, uid_t, gid_t) override            { return BAN::Error::from_errno(EPERM); }
+		virtual BAN::ErrorOr<void> link_inode_impl(BAN::StringView, BAN::RefPtr<Inode>) override                    { return BAN::Error::from_errno(EPERM); }
+		virtual BAN::ErrorOr<void> rename_inode_impl(BAN::RefPtr<Inode>, BAN::StringView, BAN::StringView) override { return BAN::Error::from_errno(EPERM); }
+		virtual BAN::ErrorOr<void> unlink_impl(BAN::StringView) override                                            { return BAN::Error::from_errno(EPERM); }
+
+		virtual bool can_read_impl() const override { return false; }
+		virtual bool can_write_impl() const override { return false; }
+		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
+
+	private:
+		ProcFDDirectoryInode(Process&, TmpFileSystem&, const TmpInodeInfo&);
+
+	private:
+		Process& m_process;
 	};
 
 }

kernel/include/kernel/FS/Socket.h

@@ -30,33 +30,19 @@ namespace Kernel
 		};
 
 	public:
-		ino_t		ino()		const final override { ASSERT_NOT_REACHED(); }
-		Mode		mode()		const final override { return Mode(m_info.mode); }
-		nlink_t		nlink()		const final override { ASSERT_NOT_REACHED(); }
-		uid_t		uid()		const final override { return m_info.uid; }
-		gid_t		gid()		const final override { return m_info.gid; }
-		off_t		size()		const final override { ASSERT_NOT_REACHED(); }
-		timespec	atime()		const final override { ASSERT_NOT_REACHED(); }
-		timespec	mtime()		const final override { ASSERT_NOT_REACHED(); }
-		timespec	ctime()		const final override { ASSERT_NOT_REACHED(); }
-		blksize_t	blksize()	const final override { ASSERT_NOT_REACHED(); }
-		blkcnt_t	blocks()	const final override { ASSERT_NOT_REACHED(); }
-		dev_t		dev()		const final override { ASSERT_NOT_REACHED(); }
-		dev_t		rdev()		const final override { ASSERT_NOT_REACHED(); }
-
 		const FileSystem* filesystem() const final override { return nullptr; }
 
 	protected:
 		Socket(const Info& info)
-			: m_info(info)
-		{}
-
-		BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan buffer) override { return recvfrom_impl(buffer, nullptr, nullptr); }
-		BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan buffer) override { return sendto_impl(buffer, nullptr, 0); }
-		BAN::ErrorOr<void> fsync_impl() final override { return {}; }
+		{
+			m_mode = info.mode;
+			m_uid = info.uid;
+			m_gid = info.gid;
+		}
 
 	private:
-		const Info m_info;
+		BAN::ErrorOr<void> sync_inode(SyncType) final override { return {}; }
+		BAN::ErrorOr<void> sync_data() final override { return {}; }
 	};
 
 }

kernel/include/kernel/FS/TmpFS/Definitions.h

@@ -11,6 +11,28 @@
 namespace Kernel
 {
 
+	struct TmpBlocks
+	{
+#if ARCH(x86_64)
+		// 2x direct blocks
+		// 1x singly indirect
+		// 1x doubly indirect
+		// 1x triply indirect
+		BAN::Array<size_t, 5> block;
+		static constexpr size_t direct_block_count = 2;
+#elif ARCH(i686)
+		uint32_t __padding;
+		// 5x direct blocks
+		// 1x singly indirect
+		// 1x doubly indirect
+		// 1x triply indirect
+		BAN::Array<size_t, 16> block;
+		static constexpr size_t direct_block_count = 13;
+#else
+	#error
+#endif
+	};
+
 	struct TmpInodeInfo
 	{
 		mode_t		mode	{ 0 };
@@ -22,27 +44,10 @@ namespace Kernel
 		nlink_t		nlink	{ 0 };
 		size_t		size	{ 0 };
 		blkcnt_t	blocks	{ 0 };
+		TmpBlocks   tmp_blocks;
 
-#if ARCH(x86_64)
-		// 2x direct blocks
-		// 1x singly indirect
-		// 1x doubly indirect
-		// 1x triply indirect
-		BAN::Array<paddr_t, 5> block;
-		static constexpr size_t direct_block_count = 2;
-#elif ARCH(i686)
-		uint32_t __padding;
-		// 5x direct blocks
-		// 1x singly indirect
-		// 1x doubly indirect
-		// 1x triply indirect
-		BAN::Array<paddr_t, 8> block;
-		static constexpr size_t direct_block_count = 5;
-#else
-	#error
-#endif
 		static constexpr size_t max_size =
-			direct_block_count * PAGE_SIZE +
+			TmpBlocks::direct_block_count * PAGE_SIZE +
 			(PAGE_SIZE / sizeof(paddr_t)) * PAGE_SIZE +
 			(PAGE_SIZE / sizeof(paddr_t)) * (PAGE_SIZE / sizeof(paddr_t)) * PAGE_SIZE +
 			(PAGE_SIZE / sizeof(paddr_t)) * (PAGE_SIZE / sizeof(paddr_t)) * (PAGE_SIZE / sizeof(paddr_t)) * PAGE_SIZE;

kernel/include/kernel/FS/TmpFS/FileSystem.h

@@ -58,7 +58,7 @@ namespace Kernel
 
 		virtual BAN::RefPtr<Inode> root_inode() override { return m_root_inode; }
 
-		virtual dev_t dev() const override { return m_rdev; }
+		dev_t rdev() const { return m_rdev; }
 
 		BAN::ErrorOr<BAN::RefPtr<TmpInode>> open_inode(ino_t ino);
 
@@ -118,16 +118,8 @@ namespace Kernel
 
 	private:
 		InodeLocation find_inode(ino_t ino);
-
 		paddr_t find_block(size_t index);
 
-		template<TmpFuncs::for_each_indirect_paddr_allocating_callback F>
-		BAN::ErrorOr<void> for_each_indirect_paddr_allocating(PageInfo page_info, F callback, size_t depth);
-		template<TmpFuncs::for_each_indirect_paddr_allocating_callback F>
-		BAN::ErrorOr<BAN::Iteration> for_each_indirect_paddr_allocating_internal(PageInfo page_info, F callback, size_t depth);
-
-		paddr_t find_indirect(PageInfo root, size_t index, size_t depth);
-
 	private:
 		const dev_t m_rdev;
 
@@ -146,14 +138,14 @@ namespace Kernel
 		static constexpr size_t max_data_pages =
 			(PAGE_SIZE / sizeof(PageInfo)) *
 			(PAGE_SIZE / sizeof(PageInfo)) *
-			(PAGE_SIZE / sizeof(PageInfo));
+			(PAGE_SIZE / sizeof(PageInfo) - 1);
 
 		// We store inodes in pages with double indirection.
 		// With 64-bit pointers we can store 512^2 pages of inodes
 		// which should be enough for now.
 		// In future this should be dynamically calculated based on maximum
 		// number of pages for this file system.
-		PageInfo m_inode_pages;
+		PageInfo m_inode_pages {};
 		static constexpr size_t first_inode = 1;
 		static constexpr size_t max_inodes =
 			(PAGE_SIZE / sizeof(PageInfo)) *

kernel/include/kernel/FS/TmpFS/Inode.h

@@ -4,6 +4,7 @@
 #include <BAN/Optional.h>
 #include <kernel/FS/Inode.h>
 #include <kernel/FS/TmpFS/Definitions.h>
+#include <kernel/Lock/Mutex.h>
 
 namespace Kernel
 {
@@ -23,43 +24,37 @@ namespace Kernel
 
 	class TmpInode : public Inode
 	{
-	public:
-		virtual ino_t		ino()		const override { return m_ino; }
-		virtual Mode		mode()		const override { return Mode(m_inode_info.mode); }
-		virtual nlink_t		nlink()		const override { return m_inode_info.nlink; }
-		virtual uid_t		uid()		const override { return m_inode_info.uid; }
-		virtual gid_t		gid()		const override { return m_inode_info.gid; }
-		virtual off_t		size()		const override { return m_inode_info.size; }
-		virtual timespec	atime()		const override { return m_inode_info.atime; }
-		virtual timespec	mtime()		const override { return m_inode_info.mtime; }
-		virtual timespec	ctime()		const override { return m_inode_info.ctime; }
-		virtual blksize_t	blksize()	const override { return PAGE_SIZE; }
-		virtual blkcnt_t	blocks()	const override { return m_inode_info.blocks; }
-		virtual dev_t		dev()		const override;
-		virtual dev_t		rdev()		const override { return 0; }
-
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<TmpInode>> create_from_existing(TmpFileSystem&, ino_t, const TmpInodeInfo&);
-		~TmpInode();
+		virtual ~TmpInode();
 
 		virtual const FileSystem* filesystem() const override;
 
 	protected:
 		TmpInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
 
-		virtual BAN::ErrorOr<void> fsync_impl() override { return {}; }
+		void write_inode_to_fs();
+		virtual BAN::ErrorOr<void> prepare_unlink_no_lock() { return {}; };
 
-		void sync();
 		void free_all_blocks();
-		virtual BAN::ErrorOr<void> prepare_unlink() { return {}; };
+		void free_indirect_blocks_no_lock(size_t block, uint32_t depth);
 
 		BAN::Optional<size_t> block_index(size_t data_block_index);
+		BAN::Optional<size_t> block_index_from_indirect_no_lock(size_t block, size_t index, uint32_t depth);
+
 		BAN::ErrorOr<size_t> block_index_with_allocation(size_t data_block_index);
+		BAN::ErrorOr<size_t> block_index_from_indirect_with_allocation_no_lock(size_t& block, size_t index, uint32_t depth);
+
+	private:
+		BAN::ErrorOr<void> sync_inode(SyncType) override;
+		BAN::ErrorOr<void> sync_data() override;
 
 	protected:
 		TmpFileSystem& m_fs;
-		TmpInodeInfo m_inode_info;
-		const ino_t m_ino;
+		TmpBlocks m_tmp_blocks;
+
+		// TODO: try to reduce locking or replace this with rwlock(?)
+		Mutex m_lock;
 
 		// has to be able to increase link count
 		friend class TmpDirectoryInode;
@@ -71,15 +66,15 @@ namespace Kernel
 		static BAN::ErrorOr<BAN::RefPtr<TmpFileInode>> create_new(TmpFileSystem&, mode_t, uid_t, gid_t);
 		~TmpFileInode();
 
-	protected:
+	private:
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
 		virtual BAN::ErrorOr<void> truncate_impl(size_t) override;
-		virtual BAN::ErrorOr<void> chmod_impl(mode_t) override;
 
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return true; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		TmpFileInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
@@ -87,21 +82,44 @@ namespace Kernel
 		friend class TmpInode;
 	};
 
+	// NOTE: this is just a dummy, when opening a fifo a pipe is created
+	class TmpFIFOInode : public TmpInode
+	{
+	public:
+		static BAN::ErrorOr<BAN::RefPtr<TmpFIFOInode>> create_new(TmpFileSystem&, mode_t, uid_t, gid_t);
+		~TmpFIFOInode();
+
+	private:
+		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override { return BAN::Error::from_errno(ENODEV); }
+		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override { return BAN::Error::from_errno(ENODEV); }
+		virtual BAN::ErrorOr<void> truncate_impl(size_t) override { return BAN::Error::from_errno(ENODEV); }
+
+		virtual bool can_read_impl() const override { return false; }
+		virtual bool can_write_impl() const override { return false; }
+		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
+
+	private:
+		TmpFIFOInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
+
+		friend class TmpInode;
+	};
+
 	class TmpSocketInode : public TmpInode
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<TmpSocketInode>> create_new(TmpFileSystem&, mode_t, uid_t, gid_t);
 		~TmpSocketInode();
 
-	protected:
+	private:
 		virtual BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override { return BAN::Error::from_errno(ENODEV); }
 		virtual BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override { return BAN::Error::from_errno(ENODEV); }
 		virtual BAN::ErrorOr<void> truncate_impl(size_t) override { return BAN::Error::from_errno(ENODEV); }
-		virtual BAN::ErrorOr<void> chmod_impl(mode_t) override;
 
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		TmpSocketInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
@@ -115,14 +133,14 @@ namespace Kernel
 		static BAN::ErrorOr<BAN::RefPtr<TmpSymlinkInode>> create_new(TmpFileSystem&, mode_t, uid_t, gid_t, BAN::StringView target);
 		~TmpSymlinkInode();
 
-		BAN::ErrorOr<void> set_link_target(BAN::StringView);
-
-	protected:
-		virtual BAN::ErrorOr<BAN::String> link_target_impl() override;
+	private:
+		BAN::ErrorOr<BAN::String> link_target_impl() override;
+		BAN::ErrorOr<void> set_link_target_impl(BAN::StringView) override;
 
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		TmpSymlinkInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
@@ -141,23 +159,28 @@ namespace Kernel
 	protected:
 		TmpDirectoryInode(TmpFileSystem&, ino_t, const TmpInodeInfo&);
 
-		virtual BAN::ErrorOr<void> prepare_unlink() override;
+		virtual BAN::ErrorOr<void> prepare_unlink_no_lock() override;
 
 	protected:
 		virtual BAN::ErrorOr<BAN::RefPtr<Inode>> find_inode_impl(BAN::StringView) override final;
 		virtual BAN::ErrorOr<size_t> list_next_inodes_impl(off_t, struct dirent*, size_t) override final;
 		virtual BAN::ErrorOr<void> create_file_impl(BAN::StringView, mode_t, uid_t, gid_t) override final;
 		virtual BAN::ErrorOr<void> create_directory_impl(BAN::StringView, mode_t, uid_t, gid_t) override final;
+		virtual BAN::ErrorOr<void> link_inode_impl(BAN::StringView, BAN::RefPtr<Inode>) override final;
+		virtual BAN::ErrorOr<void> rename_inode_impl(BAN::RefPtr<Inode>, BAN::StringView, BAN::StringView) override final;
 		virtual BAN::ErrorOr<void> unlink_impl(BAN::StringView) override;
 
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return false; }
 		virtual bool has_error_impl() const override { return false; }
+		virtual bool has_hungup_impl() const override { return false; }
 
 	private:
 		template<TmpFuncs::for_each_valid_entry_callback F>
 		void for_each_valid_entry(F callback);
 
+		BAN::ErrorOr<void> unlink_inode(BAN::StringView, bool cleanup);
+
 		friend class TmpInode;
 	};
 

kernel/include/kernel/FS/USTARModule.h

@@ -0,0 +1,11 @@
+#pragma once
+
+#include <kernel/BootInfo.h>
+#include <kernel/FS/Inode.h>
+
+namespace Kernel
+{
+
+	BAN::ErrorOr<bool> unpack_boot_module_into_directory(BAN::RefPtr<Inode>, const BootModule&);
+
+}

kernel/include/kernel/FS/VirtualFileSystem.h

@@ -29,9 +29,6 @@ namespace Kernel
 
 		virtual BAN::RefPtr<Inode> root_inode() override { return m_root_fs->root_inode(); }
 
-		// FIXME:
-		virtual dev_t dev() const override { return 0; }
-
 		BAN::ErrorOr<void> mount(const Credentials&, BAN::StringView, BAN::StringView);
 		BAN::ErrorOr<void> mount(const Credentials&, BAN::RefPtr<FileSystem>, BAN::StringView);
 
@@ -74,13 +71,13 @@ namespace Kernel
 
 		File root_file()
 		{
-			return File(root_inode(), "/"_sv);
+			return File { root_inode(), "/"_sv };
 		}
 
-		BAN::ErrorOr<File> file_from_relative_path(const File& parent, const Credentials&, BAN::StringView, int);
-		BAN::ErrorOr<File> file_from_absolute_path(const Credentials& credentials, BAN::StringView path, int flags)
+		BAN::ErrorOr<File> file_from_relative_path(BAN::RefPtr<Inode> root_inode, const File& parent, const Credentials&, BAN::StringView, int);
+		BAN::ErrorOr<File> file_from_absolute_path(BAN::RefPtr<Inode> root_inode, const Credentials& credentials, BAN::StringView path, int flags)
 		{
-			return file_from_relative_path(root_file(), credentials, path, flags);
+			return file_from_relative_path(root_inode, File { root_inode, "/"_sv }, credentials, path, flags);
 		}
 
 	private:
@@ -95,8 +92,9 @@ namespace Kernel
 		MountPoint* mount_from_root_inode(BAN::RefPtr<Inode>);
 
 	private:
-		Mutex					m_mutex;
 		BAN::RefPtr<FileSystem>	m_root_fs;
+
+		Mutex					m_mount_point_lock;
 		BAN::Vector<MountPoint>	m_mount_points;
 
 		friend class BAN::RefPtr<VirtualFileSystem>;

kernel/include/kernel/GDT.h

@@ -129,9 +129,16 @@ namespace Kernel
 		}
 
 #if ARCH(i686)
-		void set_tls(uintptr_t addr);
+		void set_fsbase(uintptr_t addr);
+		void set_gsbase(uintptr_t addr);
 #endif
 
+		static uint16_t cpu_index_offset() { return m_cpu_index_offset; }
+		void set_cpu_index(uint8_t index)
+		{
+			write_entry(m_cpu_index_offset, 0, index, 0xF2, 0x4);
+		}
+
 	private:
 		GDT() = default;
 
@@ -150,11 +157,13 @@ namespace Kernel
 
 	private:
 #if ARCH(x86_64)
-		BAN::Array<SegmentDescriptor, 7> m_gdt; // null, kernel code, kernel data, user code, user data, tss low, tss high
-		static constexpr uint16_t m_tss_offset = 0x28;
-#elif ARCH(i686)
-		BAN::Array<SegmentDescriptor, 8> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, tls, tss
+		BAN::Array<SegmentDescriptor, 9> m_gdt; // null, kernel code, kernel data, user code (32 bit), user data, user code (64 bit), cpu-index, tss low, tss high
+		static constexpr uint16_t m_cpu_index_offset = 0x30;
 		static constexpr uint16_t m_tss_offset = 0x38;
+#elif ARCH(i686)
+		BAN::Array<SegmentDescriptor, 10> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, fsbase, gsbase, cpu-index, tss
+		static constexpr uint16_t m_cpu_index_offset = 0x40;
+		static constexpr uint16_t m_tss_offset = 0x48;
 #endif
 		TaskStateSegment m_tss;
 		const GDTR m_gdtr {

kernel/include/kernel/IDT.h

@@ -18,10 +18,12 @@ namespace Kernel
 
 	constexpr uint8_t IRQ_VECTOR_BASE   = 0x20;
 	constexpr uint8_t IRQ_MSI_BASE      = 0x80;
-	constexpr uint8_t IRQ_SYSCALL       = 0xF0;
-	constexpr uint8_t IRQ_YIELD         = 0xF1;
-	constexpr uint8_t IRQ_IPI           = 0xF2;
-	constexpr uint8_t IRQ_TIMER         = 0xF3;
+	constexpr uint8_t IRQ_MSI_END       = 0xF0;
+#if ARCH(i686)
+	constexpr uint8_t IRQ_SYSCALL       = 0xF0; // hard coded in kernel/API/Syscall.h
+#endif
+	constexpr uint8_t IRQ_IPI           = 0xF1;
+	constexpr uint8_t IRQ_TIMER         = 0xF2;
 
 #if ARCH(x86_64)
 	struct GateDescriptor
@@ -75,7 +77,7 @@ namespace Kernel
 	private:
 		IDT() = default;
 
-		void register_interrupt_handler(uint8_t index, void (*handler)());
+		void register_interrupt_handler(uint8_t index, void (*handler)(), uint8_t ist = 0);
 		void register_syscall_handler(uint8_t index, void (*handler)());
 
 	private: