Kernel: Fix ByteRingBuffer->back()

No need to swap bytes of every 16 bit word in the packet, we can just do
one swap at the return

BAN/include/BAN/HashMap.h

@@ -1,49 +1,154 @@
 #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
+		{
+			constexpr bool operator()(const Key& a)
+			{
+				return HASH()(a);
+			}
+			constexpr bool operator()(const Entry& a)
+			{
+				return HASH()(a.key);
+			}
+		};
+
+		struct EntryComp
+		{
+			constexpr bool operator()(const Entry& a, const Key& 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 size_type      = size_t;
+		using key_type       = Key;
+		using value_type     = T;
+		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 +162,100 @@ 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...>;
+		ErrorOr<iterator> emplace_or_assign(Key&& key, Args&&... args) requires is_constructible_v<T, Args...>
+		{
+			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);
+			}
 
-		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()); }
+			auto it = TRY(m_hash_set.insert(Entry { BAN::move(key), T(BAN::forward<Args>(args)...) }));
+			return iterator(it);
+		}
 
-		ErrorOr<void> reserve(size_type);
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		void remove(const U& key)
+		{
+			if (auto it = find(key); it != end())
+				remove(it);
+		}
 
-		void remove(const Key&);
-		void remove(iterator it);
-		void clear();
+		iterator remove(iterator it)
+		{
+			return iterator(m_hash_set.remove(it.m_iterator));
+		}
 
-		T& operator[](const Key&);
-		const T& operator[](const Key&) const;
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		iterator find(const U& key)
+		{
+			return iterator(m_hash_set.find(key));
+		}
 
-		iterator find(const Key& key);
-		const_iterator find(const Key& key) const;
-		bool contains(const Key&) const;
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		const_iterator find(const U& key) const
+		{
+			return const_iterator(m_hash_set.find(key));
+		}
 
-		bool empty() const;
-		size_type size() const;
+		void clear()
+		{
+			m_hash_set.clear();
+		}
+
+		ErrorOr<void> reserve(size_type size)
+		{
+			return m_hash_set.reserve(size);
+		}
+
+		template<detail::HashMapFindable<Key, HASH, COMP> U>
+		T& operator[](const U& key)
+		{
+			return find(key)->value;
+		}
+
+		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();
+		}
+
+		size_type capacity() const
+		{
+			return m_hash_set.capacity();
+		}
+
+		size_type size() const
+		{
+			return m_hash_set.size();
+		}
+
+		bool empty() const
+		{
+			return m_hash_set.empty();
+		}
 
 	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;
+		HashSet<Entry, EntryHash, EntryComp> m_hash_set;
 	};
 
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::HashMap(const HashMap<Key, T, HASH>& other)
-	{
-		*this = other;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::HashMap(HashMap<Key, T, HASH>&& other)
-	{
-		*this = move(other);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	HashMap<Key, T, HASH>::~HashMap()
-	{
-		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())
-			remove(it);
-	}
-
-	template<typename Key, typename T, typename HASH>
-	void HashMap<Key, T, HASH>::remove(iterator it)
-	{
-		it.outer_current()->remove(it.inner_current());
-		m_size--;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	void HashMap<Key, T, HASH>::clear()
-	{
-		m_buckets.clear();
-		m_size = 0;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	T& HashMap<Key, T, HASH>::operator[](const Key& key)
-	{
-		ASSERT(!empty());
-		auto& bucket = get_bucket(key);
-		for (Entry& entry : bucket)
-			if (entry.key == key)
-				return entry.value;
-		ASSERT_NOT_REACHED();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	const T& HashMap<Key, T, HASH>::operator[](const Key& key) const
-	{
-		ASSERT(!empty());
-		const auto& bucket = get_bucket(key);
-		for (const Entry& entry : bucket)
-			if (entry.key == key)
-				return entry.value;
-		ASSERT_NOT_REACHED();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::iterator HashMap<Key, T, HASH>::find(const Key& 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 iterator(m_buckets.end(), bucket_it, it);
-		return end();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::const_iterator HashMap<Key, T, HASH>::find(const Key& key) const
-	{
-		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();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	bool HashMap<Key, T, HASH>::contains(const Key& key) const
-	{
-		return find(key) != end();
-	}
-
-	template<typename Key, typename T, typename HASH>
-	bool HashMap<Key, T, HASH>::empty() const
-	{
-		return m_size == 0;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	typename HashMap<Key, T, HASH>::size_type HashMap<Key, T, HASH>::size() const
-	{
-		return m_size;
-	}
-
-	template<typename Key, typename T, typename HASH>
-	ErrorOr<void> HashMap<Key, T, HASH>::rebucket(size_type bucket_count)
-	{
-		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);
-			}
-		}
-
-		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);
-	}
-
 }

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>>
-	class HashSet
+	template<typename HashSet, typename Bucket, typename T>
+	class HashSetIterator
 	{
 	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>;
+		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       = 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&&);
+		HashSet(const HashSet& other) { *this = other; }
+		HashSet& operator=(const HashSet& other)
+		{
+			clear();
 
-		ErrorOr<void> insert(const T&);
-		ErrorOr<void> insert(T&&);
-		void remove(const T&);
-		void clear();
+			MUST(reserve(other.size()));
+			for (auto& bucket : other)
+				MUST(insert(bucket));
 
-		ErrorOr<void> reserve(size_type);
+			return *this;
+		}
 
-		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()); }
+		HashSet(HashSet&& other) { *this = BAN::move(other); }
+		HashSet& operator=(HashSet&& other)
+		{
+			clear();
 
-		bool contains(const T&) const;
+			m_buckets  = other.m_buckets;
+			m_capacity = other.m_capacity;
+			m_size     = other.m_size;
+			m_removed  = other.m_removed;
 
-		size_type size() const;
-		bool empty() const;
+			other.m_buckets  = nullptr;
+			other.m_capacity = 0;
+			other.m_size     = 0;
+			other.m_removed  = 0;
+
+			return *this;
+		}
+
+		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)
+		{
+			return insert(T(value));
+		}
+
+		ErrorOr<iterator> insert(T&& value)
+		{
+			if (should_rehash_with_size(m_size + 1))
+				TRY(rehash(m_size * 2));
+			return insert_impl(BAN::move(value), HASH()(value));
+		}
+
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		void remove(const U& value)
+		{
+			if (auto it = find(value); it != end())
+				remove(it);
+		}
+
+		iterator remove(iterator it)
+		{
+			auto& bucket = *it.m_bucket;
+			bucket.element()->~T();
+			bucket.state = Bucket::REMOVED;
+			m_size--;
+			m_removed++;
+			return iterator(&bucket);
+		}
+
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		iterator find(const U& value)
+		{
+			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;
+		}
+
+		ErrorOr<void> reserve(size_type size)
+		{
+			if (should_rehash_with_size(size))
+				TRY(rehash(size * 2));
+			return {};
+		}
+
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		bool contains(const U& value) const
+		{
+			return find(value) != end();
+		}
+
+		size_type capacity() const
+		{
+			return m_capacity;
+		}
+
+		size_type size() const
+		{
+			return m_size;
+		}
+
+		bool empty() const
+		{
+			return m_size == 0;
+		}
 
 	private:
-		ErrorOr<void> rebucket(size_type);
-		LinkedList<T>& get_bucket(const T&);
-		const LinkedList<T>& get_bucket(const T&) const;
+		ErrorOr<void> rehash(size_type new_capacity)
+		{
+			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);
+
+			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));
+
+			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++)
+			{
+				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[m_capacity].end = true;
+
+			BAN::deallocator(old_buckets);
+
+			return {};
+		}
+
+		template<detail::HashSetFindable<T, HASH, COMP> U>
+		const_iterator find_impl(const U& value) const
+		{
+			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();
+			}
+		}
+
+		iterator insert_impl(T&& value, hash_t orig_hash)
+		{
+			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:
-		Vector<LinkedList<T>> m_buckets;
-		size_type m_size = 0;
+		Bucket* m_buckets { nullptr };
+		size_type m_capacity { 0 };
+		size_type m_size     { 0 };
+		size_type m_removed  { 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)
-	{
-		clear();
-		m_buckets = other.m_buckets;
-		m_size = other.m_size;
-		return *this;
-	}
-
-	template<typename T, typename HASH>
-	HashSet<T, HASH>& HashSet<T, HASH>::operator=(HashSet&& other)
-	{
-		clear();
-		m_buckets = move(other.m_buckets);
-		m_size = other.m_size;
-		other.clear();
-		return *this;
-	}
-
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::insert(const T& key)
-	{
-		return insert(move(T(key)));
-	}
-
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::insert(T&& key)
-	{
-		if (!empty() && get_bucket(key).contains(key))
-			return {};
-
-		TRY(rebucket(m_size + 1));
-		TRY(get_bucket(key).push_back(move(key)));
-		m_size++;
-		return {};
-	}
-
-	template<typename T, typename HASH>
-	void HashSet<T, HASH>::remove(const T& key)
-	{
-		if (empty()) return;
-		auto& bucket = get_bucket(key);
-		for (auto it = bucket.begin(); it != bucket.end(); it++)
-		{
-			if (*it == key)
-			{
-				bucket.remove(it);
-				m_size--;
-				break;
-			}
-		}
-	}
-
-	template<typename T, typename HASH>
-	void HashSet<T, HASH>::clear()
-	{
-		m_buckets.clear();
-		m_size = 0;
-	}
-
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::reserve(size_type size)
-	{
-		TRY(rebucket(size));
-		return {};
-	}
-
-	template<typename T, typename HASH>
-	bool HashSet<T, HASH>::contains(const T& key) const
-	{
-		if (empty()) return false;
-		return get_bucket(key).contains(key);
-	}
-
-	template<typename T, typename HASH>
-	typename HashSet<T, HASH>::size_type HashSet<T, HASH>::size() const
-	{
-		return m_size;
-	}
-
-	template<typename T, typename HASH>
-	bool HashSet<T, HASH>::empty() const
-	{
-		return m_size == 0;
-	}
-
-	template<typename T, typename HASH>
-	ErrorOr<void> HashSet<T, HASH>::rebucket(size_type bucket_count)
-	{
-		if (m_buckets.size() >= bucket_count)
-			return {};
-
-		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);
-
-		for (auto& bucket : m_buckets)
-		{
-			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);
-			}
-		}
-
-		m_buckets = move(new_buckets);
-		return {};
-	}
-
-	template<typename T, typename HASH>
-	LinkedList<T>& HashSet<T, HASH>::get_bucket(const T& key)
-	{
-		ASSERT(!m_buckets.empty());
-		size_type index = HASH()(key) % m_buckets.size();
-		return m_buckets[index];
-	}
-
-	template<typename T, typename HASH>
-	const LinkedList<T>& HashSet<T, HASH>::get_bucket(const T& key) const
-	{
-		ASSERT(!m_buckets.empty());
-		size_type index = HASH()(key) % m_buckets.size();
-		return m_buckets[index];
-	}
-
 }

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

@@ -6,19 +6,6 @@
 
 #include <float.h>
 
-// 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
-
 namespace BAN::Math
 {
 
@@ -49,12 +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;
 	}
@@ -79,25 +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>
@@ -111,6 +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)
 	{
@@ -172,7 +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;
 	}
@@ -447,9 +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/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/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());
+		}
+	};
+
 }

kernel/CMakeLists.txt

@@ -139,6 +139,7 @@ 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")
@@ -150,6 +151,7 @@ elseif("${BANAN_ARCH}" STREQUAL "i686")
 		arch/i686/Signal.S
 		arch/i686/Syscall.S
 		arch/i686/Thread.S
+		arch/i686/User.S
 		arch/i686/Yield.S
 	)
 else()
@@ -166,10 +168,7 @@ set(BAN_SOURCES
 set(KLIBC_SOURCES
 	klibc/ctype.cpp
 	klibc/string.cpp
-
-	# Ehhh don't do this but for now libc uses the same stuff kernel can use
-	# This won't work after libc starts using sse implemetations tho
-	../userspace/libraries/LibC/arch/${BANAN_ARCH}/string.S
+	klibc/arch/${BANAN_ARCH}/string.S
 )
 
 set(LIBDEFLATE_SOURCE

kernel/arch/i686/PageTable.cpp

@@ -16,11 +16,18 @@ 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 bool s_is_initialized = false;
+
 	static PageTable* s_kernel = nullptr;
 	static bool s_has_nxe = false;
 	static bool s_has_pge = false;
@@ -28,6 +35,28 @@ namespace Kernel
 
 	static paddr_t s_global_pdpte = 0;
 
+	static uint64_t* s_fast_page_pt { nullptr };
+
+	static uint64_t* allocate_zeroed_page_aligned_page()
+	{
+		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 uint64_t* P2V(const T paddr)
+	{
+		return reinterpret_cast<uint64_t*>(reinterpret_cast<paddr_t>(paddr) - g_boot_info.kernel_paddr + KERNEL_OFFSET);
+	}
+
 	static inline PageTable::flags_t parse_flags(uint64_t entry)
 	{
 		using Flags = PageTable::Flags;
@@ -46,31 +75,22 @@ 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 +131,56 @@ 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);
+
+		auto* pdpt = allocate_zeroed_page_aligned_page();
+		ASSERT(pdpt);
+
+		s_kernel->m_highest_paging_struct = V2P(pdpt);
+		s_kernel->map_kernel_memory();
+
+		PageTable::with_fast_page(s_kernel->m_highest_paging_struct, [] {
+			s_global_pdpte = PageTable::fast_page_as_sized<paddr_t>(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)
+				{
+					uint64_t* table = P2V(table_paddr);
+
+					if (!(table[entry] & Flags::Present))
+					{
+						auto* vaddr = allocate_zeroed_page_aligned_page();
+						ASSERT(vaddr);
+						table[entry] = V2P(vaddr);
+					}
+
+					table[entry] |= flags;
+
+					return table[entry] & s_page_addr_mask;
+				};
+
+			const paddr_t pdpt = s_kernel->m_highest_paging_struct;
+			const paddr_t pd = get_or_allocate_entry(pdpt, pdpte, Flags::Present);
+			s_fast_page_pt = P2V(get_or_allocate_entry(pd, pde, Flags::ReadWrite | Flags::Present));
+		}
+
+		s_kernel->load();
 	}
 
 	PageTable& PageTable::kernel()
@@ -126,40 +194,12 @@ 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);
 		map_range_at(
 			V2P(g_kernel_start),
-			(vaddr_t)g_kernel_start,
+			reinterpret_cast<vaddr_t>(g_kernel_start),
 			g_kernel_end - g_kernel_start,
 			Flags::Present
 		);
@@ -167,7 +207,7 @@ namespace Kernel
 		// Map executable kernel memory as executable
 		map_range_at(
 			V2P(g_kernel_execute_start),
-			(vaddr_t)g_kernel_execute_start,
+			reinterpret_cast<vaddr_t>(g_kernel_execute_start),
 			g_kernel_execute_end - g_kernel_execute_start,
 			Flags::Execute | Flags::Present
 		);
@@ -175,7 +215,7 @@ namespace Kernel
 		// Map writable kernel memory as writable
 		map_range_at(
 			V2P(g_kernel_writable_start),
-			(vaddr_t)g_kernel_writable_start,
+			reinterpret_cast<vaddr_t>(g_kernel_writable_start),
 			g_kernel_writable_end - g_kernel_writable_start,
 			Flags::ReadWrite | Flags::Present
 		);
@@ -183,65 +223,34 @@ namespace Kernel
 		// Map userspace memory
 		map_range_at(
 			V2P(g_userspace_start),
-			(vaddr_t)g_userspace_start,
+			reinterpret_cast<vaddr_t>(g_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(paddr % PAGE_SIZE == 0);
+		ASSERT(paddr && paddr % PAGE_SIZE == 0);
 
+		ASSERT(s_fast_page_pt);
 		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;
+		ASSERT(!(*s_fast_page_pt & Flags::Present));
+		s_fast_page_pt[0] = 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));
-
-		ASSERT(!(pt[pte] & Flags::Present));
-		pt[pte] = paddr | Flags::ReadWrite | Flags::Present;
-
-		asm volatile("invlpg (%0)" :: "r"(fast_page()) : "memory");
+		asm volatile("invlpg (%0)" :: "r"(fast_page()));
 	}
 
 	void PageTable::unmap_fast_page()
 	{
-		ASSERT(s_kernel);
-
+		ASSERT(s_fast_page_pt);
 		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;
+		ASSERT((*s_fast_page_pt & Flags::Present));
+		s_fast_page_pt[0] = 0;
 
-		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] = 0;
-
-		asm volatile("invlpg (%0)" :: "r"(fast_page()) : "memory");
+		asm volatile("invlpg (%0)" :: "r"(fast_page()));
 	}
 
 	BAN::ErrorOr<PageTable*> PageTable::create_userspace()
@@ -250,25 +259,23 @@ 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);
+		uint64_t* pdpt = allocate_zeroed_page_aligned_page();
+		if (pdpt == nullptr)
+		{
+			delete page_table;
+			return BAN::Error::from_errno(ENOMEM);
+		}
 
-		ASSERT(m_highest_paging_struct == 0);
-		m_highest_paging_struct = V2P(kmalloc(32, 32, true));
-		ASSERT(m_highest_paging_struct);
+		page_table->m_highest_paging_struct = V2P(pdpt);
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
 		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()
@@ -276,18 +283,17 @@ namespace Kernel
 		if (m_highest_paging_struct == 0)
 			return;
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
-
+		uint64_t* pdpt = P2V(m_highest_paging_struct);
 		for (uint32_t pdpte = 0; pdpte < 3; pdpte++)
 		{
 			if (!(pdpt[pdpte] & Flags::Present))
 				continue;
-			uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
+			uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
 			for (uint32_t pde = 0; pde < 512; pde++)
 			{
 				if (!(pd[pde] & Flags::Present))
 					continue;
-				kfree(reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK)));
+				kfree(P2V(pd[pde] & s_page_addr_mask));
 			}
 			kfree(pd);
 		}
@@ -298,15 +304,43 @@ namespace Kernel
 	{
 		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 <= 32 || !s_is_initialized)
+		{
+			for (size_t i = 0; i < pages; i++, vaddr += PAGE_SIZE)
+				asm volatile("invlpg (%0)" :: "r"(vaddr));
+		}
+		else if (is_userspace || !s_has_pge)
+		{
+			asm volatile("movl %0, %%cr3" :: "r"(static_cast<uint32_t>(m_highest_paging_struct)));
+		}
+		else
+		{
+			asm volatile(
+				"movl %%cr4, %%eax;"
+
+				"andl $~0x80, %%eax;"
+				"movl %%eax, %%cr4;"
+
+				"movl %0, %%cr3;"
+
+				"orl $0x80, %%eax;"
+				"movl %%eax, %%cr4;"
+				:
+				: "r"(static_cast<uint32_t>(m_highest_paging_struct))
+				: "eax"
+			);
+		}
 
 		if (send_smp_message)
 		{
@@ -314,14 +348,14 @@ 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::unmap_page(vaddr_t vaddr, bool invalidate)
 	{
 		ASSERT(vaddr);
 		ASSERT(vaddr % PAGE_SIZE == 0);
@@ -340,16 +374,16 @@ 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));
+		uint64_t* pdpt = P2V(m_highest_paging_struct);
+		uint64_t* pd   = P2V(pdpt[pdpte] & s_page_addr_mask);
+		uint64_t* pt   = P2V(pd[pde] & s_page_addr_mask);
 
-		const paddr_t old_paddr = pt[pte] & PAGE_ADDR_MASK;
+		const paddr_t old_paddr = pt[pte] & s_page_addr_mask;
 
 		pt[pte] = 0;
 
-		if (old_paddr != 0)
-			invalidate(vaddr, send_smp_message);
+		if (invalidate && old_paddr != 0)
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::unmap_range(vaddr_t vaddr, size_t size)
@@ -361,18 +395,10 @@ 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,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
-			}
-		});
+		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());
@@ -407,11 +433,11 @@ namespace Kernel
 
 		SpinLockGuard _(m_lock);
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
+		uint64_t* pdpt = P2V(m_highest_paging_struct);
 		if (!(pdpt[pdpte] & Flags::Present))
 			pdpt[pdpte] = V2P(allocate_zeroed_page_aligned_page()) | Flags::Present;
 
-		uint64_t* pd = reinterpret_cast<uint64_t*>(P2V(pdpt[pdpte] & PAGE_ADDR_MASK));
+		uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
 		if ((pd[pde] & uwr_flags) != uwr_flags)
 		{
 			if (!(pd[pde] & Flags::Present))
@@ -422,14 +448,14 @@ namespace Kernel
 		if (!(flags & Flags::Present))
 			uwr_flags &= ~Flags::Present;
 
-		uint64_t* pt = reinterpret_cast<uint64_t*>(P2V(pd[pde] & PAGE_ADDR_MASK));
+		uint64_t* pt = P2V(pd[pde] & s_page_addr_mask);
 
-		const paddr_t old_paddr = pt[pte] & PAGE_ADDR_MASK;
+		const paddr_t old_paddr = pt[pte] & s_page_addr_mask;
 
 		pt[pte] = paddr | uwr_flags | extra_flags;
 
-		if (old_paddr != 0)
-			invalidate(vaddr, send_smp_message);
+		if (invalidate && old_paddr != 0)
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::map_range_at(paddr_t paddr, vaddr_t vaddr, size_t size, flags_t flags, MemoryType memory_type)
@@ -443,15 +469,49 @@ 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);
+		invalidate_range(vaddr, page_count, true);
+	}
 
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = page_count,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
+	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 _(m_lock);
+
+		const uint64_t* pdpt = P2V(m_highest_paging_struct);
+		for (; pdpte <= e_pdpte; pdpte++)
+		{
+			if (!(pdpt[pdpte] & Flags::Present))
+				continue;
+			const uint64_t* pd = P2V(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 = P2V(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);
+				}
+				pte = 0;
 			}
-		});
+			pde = 0;
+		}
+
+		invalidate_range(vaddr, size / PAGE_SIZE, true);
 	}
 
 	uint64_t PageTable::get_page_data(vaddr_t vaddr) const
@@ -464,15 +524,15 @@ namespace Kernel
 
 		SpinLockGuard _(m_lock);
 
-		uint64_t* pdpt = (uint64_t*)P2V(m_highest_paging_struct);
+		const uint64_t* pdpt = P2V(m_highest_paging_struct);
 		if (!(pdpt[pdpte] & Flags::Present))
 			return 0;
 
-		uint64_t* pd = (uint64_t*)P2V(pdpt[pdpte] & PAGE_ADDR_MASK);
+		const uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
 		if (!(pd[pde] & Flags::Present))
 			return 0;
 
-		uint64_t* pt = (uint64_t*)P2V(pd[pde] & PAGE_ADDR_MASK);
+		const uint64_t* pt = P2V(pd[pde] & s_page_addr_mask);
 		if (!(pt[pte] & Flags::Used))
 			return 0;
 
@@ -486,8 +546,7 @@ 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
@@ -529,14 +588,8 @@ namespace Kernel
 			return false;
 		for (size_t offset = 0; offset < bytes; offset += PAGE_SIZE)
 			reserve_page(vaddr + offset, true, false);
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = bytes / PAGE_SIZE,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
-			}
-		});
+		invalidate_range(vaddr, bytes / PAGE_SIZE, true);
+
 		return true;
 	}
 
@@ -549,48 +602,47 @@ 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);
 
 		// 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 = P2V(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 = P2V(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 = P2V(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))
-					{
-						vaddr_t vaddr = 0;
-						vaddr |= (vaddr_t)pdpte << 30;
-						vaddr |= (vaddr_t)pde   << 21;
-						vaddr |= (vaddr_t)pte   << 12;
-						ASSERT(reserve_page(vaddr));
-						return vaddr;
-					}
+					if (pt[pte] & Flags::Used)
+						continue;
+					vaddr_t vaddr = 0;
+					vaddr |= (vaddr_t)pdpte << 30;
+					vaddr |= (vaddr_t)pde   << 21;
+					vaddr |= (vaddr_t)pte   << 12;
+					ASSERT(reserve_page(vaddr));
+					return vaddr;
 				}
+				pte = 0;
 			}
+			pde = 0;
 		}
 
 		// Find any free page
@@ -603,7 +655,7 @@ namespace Kernel
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	vaddr_t PageTable::reserve_free_contiguous_pages(size_t page_count, vaddr_t first_address, vaddr_t last_address)
@@ -636,7 +688,7 @@ namespace Kernel
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	static void dump_range(vaddr_t start, vaddr_t end, PageTable::flags_t flags)
@@ -659,7 +711,7 @@ namespace Kernel
 		flags_t flags = 0;
 		vaddr_t start = 0;
 
-		uint64_t* pdpt = reinterpret_cast<uint64_t*>(P2V(m_highest_paging_struct));
+		const uint64_t* pdpt = P2V(m_highest_paging_struct);
 		for (uint32_t pdpte = 0; pdpte < 4; pdpte++)
 		{
 			if (!(pdpt[pdpte] & Flags::Present))
@@ -668,7 +720,7 @@ namespace Kernel
 				start = 0;
 				continue;
 			}
-			uint64_t* pd = (uint64_t*)P2V(pdpt[pdpte] & PAGE_ADDR_MASK);
+			const uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
 			for (uint64_t pde = 0; pde < 512; pde++)
 			{
 				if (!(pd[pde] & Flags::Present))
@@ -677,7 +729,7 @@ namespace Kernel
 					start = 0;
 					continue;
 				}
-				uint64_t* pt = (uint64_t*)P2V(pd[pde] & PAGE_ADDR_MASK);
+				const uint64_t* pt = P2V(pd[pde] & s_page_addr_mask);
 				for (uint64_t pte = 0; pte < 512; pte++)
 				{
 					if (parse_flags(pt[pte]) != flags)

kernel/arch/i686/Signal.S

@@ -1,12 +1,13 @@
 .section .userspace, "ax"
 
 // stack contains
-//    return address
-//    return stack
-//    return rflags
-//    siginfo_t
-//    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:
@@ -18,6 +19,10 @@ signal_trampoline:
 	pushl %eax
 	pushl %ebp
 
+	movl 84(%esp), %eax
+	pushl %eax; addl $4, (%esp)
+	pushl (%eax)
+
 	// FIXME: populate these
 	xorl %eax, %eax
 	pushl %eax // stack
@@ -28,9 +33,9 @@ signal_trampoline:
 	pushl %eax // link
 
 	movl    %esp,  %edx // ucontext
-	leal 60(%esp), %esi // siginfo
-	movl 56(%esp), %edi // signal number
-	movl 52(%esp), %eax // handlers
+	leal 68(%esp), %esi // siginfo
+	movl 64(%esp), %edi // signal number
+	movl 60(%esp), %eax // handlers
 
 	// align stack to 16 bytes
 	movl %esp, %ebp
@@ -53,7 +58,15 @@ signal_trampoline:
 	movl %ebp, %esp
 	addl $24, %esp
 
+	// restore sigmask
+	movl $83, %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
@@ -62,8 +75,8 @@ signal_trampoline:
 	popl %edi
 	popl %esi
 
-	// skip handler, number, siginfo_t
-	addl $44, %esp
+	// skip handler, number, siginfo_t, sigmask
+	addl $52, %esp
 
 	// restore flags
 	popf

kernel/arch/i686/Syscall.S

@@ -63,7 +63,7 @@ sys_fork_trampoline:
 
 	call read_ip
 	testl %eax, %eax
-	jz .reload_stack
+	jz .done
 
 	movl %esp, %ebx
 
@@ -79,9 +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
@@ -34,9 +31,6 @@ start_kernel_thread:
 
 .global start_userspace_thread
 start_userspace_thread:
-	call get_thread_start_sp
-	movl %eax, %esp
-
 	movw $(0x20 | 3), %bx
 	movw %bx, %ds
 	movw %bx, %es

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/boot.S

@@ -11,13 +11,14 @@
 
 .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 (1 << 2) # page align modules
-	.long -(0x1BADB002 + (1 << 2))
+	.long multiboot_flags
+	.long -(0x1BADB002 + multiboot_flags)
 
 	.long 0
 	.long 0
@@ -30,7 +31,8 @@ multiboot_start:
 	.long FB_HEIGHT
 	.long FB_BPP
 multiboot_end:
-	.align 8
+
+.section .multiboot2, "aw"
 multiboot2_start:
 	.long 0xE85250D6
 	.long 0
@@ -66,7 +68,6 @@ multiboot2_start:
 multiboot2_end:
 
 .section .bananboot, "aw"
-	.align 8
 bananboot_start:
 	.long 0xBABAB007
 	.long -(0xBABAB007 + FB_WIDTH + FB_HEIGHT + FB_BPP)
@@ -97,8 +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
@@ -111,13 +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
@@ -273,7 +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"
 
@@ -283,21 +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:
@@ -306,8 +315,7 @@ 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
@@ -315,24 +323,28 @@ ap_protected_mode:
 
 	# 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

@@ -1,20 +1,20 @@
-.macro maybe_load_kernel_segments, n
-	testb $3, \n(%esp)
-	jz 1f; jnp 1f
-
+.macro intr_header, n
+	pushal
+	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:
+1:	cld
 .endm
 
-.macro maybe_load_userspace_segments, n
-	testb $3, \n(%esp)
-	jz 1f; jnp 1f
-
+.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
@@ -22,14 +22,11 @@
 	movw %bx, %fs
 	movw $(0x38 | 3), %bx
 	movw %bx, %gs
-1:
+1:	popal
 .endm
 
 isr_stub:
-	pushal
-	maybe_load_kernel_segments 44
-	cld
-
+	intr_header 12
 	movl %cr0, %eax; pushl %eax
 	movl %cr2, %eax; pushl %eax
 	movl %cr3, %eax; pushl %eax
@@ -57,15 +54,12 @@ isr_stub:
 	movl %ebp, %esp
 	addl $24, %esp
 
-	maybe_load_userspace_segments 44
-	popal
+	intr_footer 12
 	addl $8, %esp
 	iret
 
 irq_stub:
-	pushal
-	maybe_load_kernel_segments 44
-	cld
+	intr_header 12
 
 	movl 32(%esp), %edi # interrupt number
 
@@ -78,16 +72,13 @@ irq_stub:
 
 	movl %ebp, %esp
 
-	maybe_load_userspace_segments 44
-	popal
+	intr_footer 12
 	addl $8, %esp
 	iret
 
 .global asm_ipi_handler
 asm_ipi_handler:
-	pushal
-	maybe_load_kernel_segments 36
-	cld
+	intr_header 4
 
 	movl %esp, %ebp
 	andl $-16, %esp
@@ -96,15 +87,12 @@ asm_ipi_handler:
 
 	movl %ebp, %esp
 
-	maybe_load_userspace_segments 36
-	popal
+	intr_footer 4
 	iret
 
 .global asm_timer_handler
 asm_timer_handler:
-	pushal
-	maybe_load_kernel_segments 36
-	cld
+	intr_header 4
 
 	movl %esp, %ebp
 	andl $-16, %esp
@@ -113,8 +101,7 @@ asm_timer_handler:
 
 	movl %ebp, %esp
 
-	maybe_load_userspace_segments 36
-	popal
+	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/PageTable.cpp

@@ -2,7 +2,6 @@
 #include <kernel/CPUID.h>
 #include <kernel/Lock/SpinLock.h>
 #include <kernel/Memory/Heap.h>
-#include <kernel/Memory/kmalloc.h>
 #include <kernel/Memory/PageTable.h>
 
 extern uint8_t g_kernel_start[];
@@ -17,13 +16,15 @@ 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;
 
 	static constexpr vaddr_t s_hhdm_offset = 0xFFFF800000000000;
-	static bool s_is_hddm_initialized = false;
+	static bool s_is_initialized = false;
 
 	constexpr uint64_t s_page_flag_mask = 0x8000000000000FFF;
 	constexpr uint64_t s_page_addr_mask = ~s_page_flag_mask;
@@ -35,6 +36,8 @@ namespace Kernel
 
 	static paddr_t s_global_pml4_entries[512] { 0 };
 
+	static uint64_t* s_fast_page_pt { nullptr };
+
 	static constexpr inline bool is_canonical(uintptr_t addr)
 	{
 		constexpr uintptr_t mask = 0xFFFF800000000000;
@@ -54,68 +57,27 @@ namespace Kernel
 		return addr;
 	}
 
-	struct FuncsKmalloc
+	static paddr_t allocate_zeroed_page_aligned_page()
 	{
-		static paddr_t allocate_zeroed_page_aligned_page()
-		{
-			void* page = kmalloc(PAGE_SIZE, PAGE_SIZE, true);
-			ASSERT(page);
-			memset(page, 0, PAGE_SIZE);
-			return kmalloc_paddr_of(reinterpret_cast<vaddr_t>(page)).value();
-		}
+		const paddr_t paddr = Heap::get().take_free_page();
+		ASSERT(paddr);
+		memset(reinterpret_cast<void*>(paddr + s_hhdm_offset), 0, PAGE_SIZE);
+		return paddr;
+	}
 
-		static void unallocate_page(paddr_t paddr)
-		{
-			kfree(reinterpret_cast<void*>(kmalloc_vaddr_of(paddr).value()));
-		}
-
-		static paddr_t V2P(vaddr_t vaddr)
-		{
-			return vaddr - KERNEL_OFFSET + g_boot_info.kernel_paddr;
-		}
-
-		static uint64_t* P2V(paddr_t paddr)
-		{
-			return reinterpret_cast<uint64_t*>(paddr - g_boot_info.kernel_paddr + KERNEL_OFFSET);
-		}
-	};
-
-	struct FuncsHHDM
+	static void unallocate_page(paddr_t paddr)
 	{
-		static paddr_t allocate_zeroed_page_aligned_page()
-		{
-			const paddr_t paddr = Heap::get().take_free_page();
-			ASSERT(paddr);
-			memset(reinterpret_cast<void*>(paddr + s_hhdm_offset), 0, PAGE_SIZE);
-			return paddr;
-		}
+		Heap::get().release_page(paddr);
+	}
 
-		static void unallocate_page(paddr_t paddr)
-		{
-			Heap::get().release_page(paddr);
-		}
+	static uint64_t* P2V(paddr_t paddr)
+	{
+		ASSERT(paddr != 0);
+		ASSERT(!BAN::Math::will_addition_overflow(paddr, s_hhdm_offset));
+		return reinterpret_cast<uint64_t*>(paddr + s_hhdm_offset);
+	}
 
-		static paddr_t V2P(vaddr_t vaddr)
-		{
-			ASSERT(vaddr >= s_hhdm_offset);
-			ASSERT(vaddr < KERNEL_OFFSET);
-			return vaddr - s_hhdm_offset;
-		}
-
-		static uint64_t* P2V(paddr_t paddr)
-		{
-			ASSERT(paddr != 0);
-			ASSERT(!BAN::Math::will_addition_overflow(paddr, s_hhdm_offset));
-			return reinterpret_cast<uint64_t*>(paddr + s_hhdm_offset);
-		}
-	};
-
-	static paddr_t   (*allocate_zeroed_page_aligned_page)() = &FuncsKmalloc::allocate_zeroed_page_aligned_page;
-	static void      (*unallocate_page)(paddr_t)            = &FuncsKmalloc::unallocate_page;
-	static paddr_t   (*V2P)(vaddr_t)                        = &FuncsKmalloc::V2P;
-	static uint64_t* (*P2V)(paddr_t)                        = &FuncsKmalloc::P2V;
-
-	static inline PageTable::flags_t parse_flags(uint64_t entry)
+	static PageTable::flags_t parse_flags(uint64_t entry)
 	{
 		using Flags = PageTable::Flags;
 
@@ -137,7 +99,7 @@ namespace Kernel
 	//   0: 4 KiB
 	//   1: 2 MiB
 	//   2: 1 GiB
-	static void init_map_hhdm_page(paddr_t pml4, paddr_t paddr, uint8_t page_size)
+	static void map_hhdm_page(paddr_t pml4, paddr_t paddr, uint8_t page_size)
 	{
 		ASSERT(0 <= page_size && page_size <= 2);
 
@@ -184,7 +146,7 @@ namespace Kernel
 		const uint64_t noexec_flag = s_has_nxe ? (static_cast<uint64_t>(1) << 63) : 0;
 
 		const paddr_t pdpt = get_or_allocate_entry(pml4, pml4e, noexec_flag);
-		s_global_pml4_entries[pml4e] = pdpt | hhdm_flags;
+		s_global_pml4_entries[pml4e] = pdpt | hhdm_flags | noexec_flag;
 
 		paddr_t lowest_paddr = pdpt;
 		uint16_t lowest_entry = pdpte;
@@ -207,23 +169,11 @@ namespace Kernel
 		});
 	}
 
-	static void init_map_hhdm(paddr_t pml4)
+	static void initialize_hhdm(paddr_t pml4)
 	{
 		for (const auto& entry : g_boot_info.memory_map_entries)
 		{
-			bool should_map = false;
-			switch (entry.type)
-			{
-				case MemoryMapEntry::Type::Available:
-					should_map = true;
-					break;
-				case MemoryMapEntry::Type::ACPIReclaim:
-				case MemoryMapEntry::Type::ACPINVS:
-				case MemoryMapEntry::Type::Reserved:
-					should_map = false;
-					break;
-			}
-			if (!should_map)
+			if (entry.type != MemoryMapEntry::Type::Available)
 				continue;
 
 			constexpr size_t one_gib = 1024 * 1024 * 1024;
@@ -235,156 +185,39 @@ namespace Kernel
 			{
 				if (s_has_gib && paddr % one_gib == 0 && paddr + one_gib <= entry_end)
 				{
-					init_map_hhdm_page(pml4, paddr, 2);
+					map_hhdm_page(pml4, paddr, 2);
 					paddr += one_gib;
 				}
 				else if (paddr % two_mib == 0 && paddr + two_mib <= entry_end)
 				{
-					init_map_hhdm_page(pml4, paddr, 1);
+					map_hhdm_page(pml4, paddr, 1);
 					paddr += two_mib;
 				}
 				else
 				{
-					init_map_hhdm_page(pml4, paddr, 0);
+					map_hhdm_page(pml4, paddr, 0);
 					paddr += PAGE_SIZE;
 				}
 			}
 		}
 	}
 
-	static paddr_t copy_page_from_kmalloc_to_heap(paddr_t kmalloc_paddr)
+	void PageTable::initialize_fast_page()
 	{
-		const paddr_t heap_paddr = Heap::get().take_free_page();
-		ASSERT(heap_paddr);
-
-		const vaddr_t kmalloc_vaddr = kmalloc_vaddr_of(kmalloc_paddr).value();
-
-		PageTable::with_fast_page(heap_paddr, [kmalloc_vaddr] {
-			memcpy(PageTable::fast_page_as_ptr(), reinterpret_cast<void*>(kmalloc_vaddr), PAGE_SIZE);
-		});
-
-		return heap_paddr;
+		s_fast_page_pt = g_boot_fast_page_pt;
 	}
 
-	static void copy_paging_structure_to_heap(uint64_t* old_table, uint64_t* new_table, int depth)
-	{
-		if (depth == 0)
-			return;
-
-		constexpr uint64_t page_flag_mask = 0x8000000000000FFF;
-		constexpr uint64_t page_addr_mask = ~page_flag_mask;
-
-		for (uint16_t index = 0; index < 512; index++)
-		{
-			const uint64_t old_entry = old_table[index];
-			if (old_entry == 0)
-			{
-				new_table[index] = 0;
-				continue;
-			}
-
-			const paddr_t old_paddr = old_entry & page_addr_mask;
-			const paddr_t new_paddr = copy_page_from_kmalloc_to_heap(old_paddr);
-			new_table[index] = new_paddr | (old_entry & page_flag_mask);
-
-			uint64_t* next_old_table = reinterpret_cast<uint64_t*>(old_paddr + s_hhdm_offset);
-			uint64_t* next_new_table = reinterpret_cast<uint64_t*>(new_paddr + s_hhdm_offset);
-			copy_paging_structure_to_heap(next_old_table, next_new_table, depth - 1);
-		}
-	}
-
-	static void free_kmalloc_paging_structure(uint64_t* table, int depth)
-	{
-		if (depth == 0)
-			return;
-
-		constexpr uint64_t page_flag_mask = 0x8000000000000FFF;
-		constexpr uint64_t page_addr_mask = ~page_flag_mask;
-
-		for (uint16_t index = 0; index < 512; index++)
-		{
-			const uint64_t entry = table[index];
-			if (entry == 0)
-				continue;
-
-			const paddr_t paddr = entry & page_addr_mask;
-
-			uint64_t* next_table = reinterpret_cast<uint64_t*>(paddr + s_hhdm_offset);
-			free_kmalloc_paging_structure(next_table, depth - 1);
-
-			kfree(reinterpret_cast<void*>(kmalloc_vaddr_of(paddr).value()));
-		}
-	}
-
-	void PageTable::initialize_pre_heap()
+	static void detect_cpu_features()
 	{
 		if (CPUID::has_nxe())
 			s_has_nxe = true;
-
 		if (CPUID::has_pge())
 			s_has_pge = true;
-
 		if (CPUID::has_1gib_pages())
 			s_has_gib = true;
-
-		ASSERT(s_kernel == nullptr);
-		s_kernel = new PageTable();
-		ASSERT(s_kernel);
-
-		s_kernel->m_highest_paging_struct = allocate_zeroed_page_aligned_page();
-		s_kernel->prepare_fast_page();
-		s_kernel->initialize_kernel();
-
-		for (auto pml4e : s_global_pml4_entries)
-			ASSERT(pml4e == 0);
-		const uint64_t* pml4 = P2V(s_kernel->m_highest_paging_struct);
-		s_global_pml4_entries[511] = pml4[511];
 	}
 
-	void PageTable::initialize_post_heap()
-	{
-		ASSERT(s_kernel);
-
-		init_map_hhdm(s_kernel->m_highest_paging_struct);
-
-		const paddr_t old_pml4_paddr = s_kernel->m_highest_paging_struct;
-		const paddr_t new_pml4_paddr = copy_page_from_kmalloc_to_heap(old_pml4_paddr);
-
-		uint64_t* old_pml4 = reinterpret_cast<uint64_t*>(kmalloc_vaddr_of(old_pml4_paddr).value());
-		uint64_t* new_pml4 = reinterpret_cast<uint64_t*>(new_pml4_paddr + s_hhdm_offset);
-
-		const paddr_t old_pdpt_paddr = old_pml4[511] & s_page_addr_mask;
-		const paddr_t new_pdpt_paddr = Heap::get().take_free_page();
-		ASSERT(new_pdpt_paddr);
-
-		uint64_t* old_pdpt = reinterpret_cast<uint64_t*>(old_pdpt_paddr + s_hhdm_offset);
-		uint64_t* new_pdpt = reinterpret_cast<uint64_t*>(new_pdpt_paddr + s_hhdm_offset);
-		copy_paging_structure_to_heap(old_pdpt, new_pdpt, 2);
-
-		new_pml4[511] = new_pdpt_paddr | (old_pml4[511] & s_page_flag_mask);
-		s_global_pml4_entries[511] = new_pml4[511];
-
-		s_kernel->m_highest_paging_struct = new_pml4_paddr;
-		s_kernel->load();
-
-		free_kmalloc_paging_structure(old_pdpt, 2);
-		kfree(reinterpret_cast<void*>(kmalloc_vaddr_of(old_pdpt_paddr).value()));
-		kfree(reinterpret_cast<void*>(kmalloc_vaddr_of(old_pml4_paddr).value()));
-
-		allocate_zeroed_page_aligned_page = &FuncsHHDM::allocate_zeroed_page_aligned_page;
-		unallocate_page                   = &FuncsHHDM::unallocate_page;
-		V2P                               = &FuncsHHDM::V2P;
-		P2V                               = &FuncsHHDM::P2V;
-
-		s_is_hddm_initialized = true;
-
-		// This is a hack to unmap fast page. fast page pt is copied
-		// while it is mapped, so we need to manually unmap it
-		SpinLockGuard _(s_fast_page_lock);
-		unmap_fast_page();
-	}
-
-	void PageTable::initial_load()
+	void PageTable::enable_cpu_features()
 	{
 		if (s_has_nxe)
 		{
@@ -423,8 +256,63 @@ namespace Kernel
 			"movq %%rax, %%cr0;"
 			::: "rax"
 		);
+	}
 
-		load();
+	void PageTable::initialize_and_load()
+	{
+		detect_cpu_features();
+		enable_cpu_features();
+
+		const paddr_t boot_pml4_paddr = ({
+			paddr_t paddr;
+			asm volatile("movq %%cr3, %0" : "=r"(paddr));
+			paddr;
+		});
+
+		initialize_hhdm(boot_pml4_paddr);
+
+		ASSERT(s_kernel == nullptr);
+		s_kernel = new PageTable();
+		ASSERT(s_kernel != nullptr);
+
+		s_kernel->m_highest_paging_struct = allocate_zeroed_page_aligned_page();
+		ASSERT(s_kernel->m_highest_paging_struct);
+
+		uint64_t* pml4 = P2V(s_kernel->m_highest_paging_struct);
+		memcpy(pml4, s_global_pml4_entries, sizeof(s_global_pml4_entries));
+		s_kernel->map_kernel_memory();
+		s_global_pml4_entries[511] = pml4[511];
+
+		// update fast page pt
+		{
+			constexpr vaddr_t uc_vaddr = uncanonicalize(fast_page());
+			constexpr uint16_t pml4e = (uc_vaddr >> 39) & 0x1FF;
+			constexpr uint16_t pdpte = (uc_vaddr >> 30) & 0x1FF;
+			constexpr uint16_t pde   = (uc_vaddr >> 21) & 0x1FF;
+
+			const auto get_or_allocate_entry =
+				[](paddr_t table_paddr, uint16_t entry, uint64_t flags)
+				{
+					uint64_t* table = P2V(table_paddr);
+
+					if (!(table[entry] & Flags::Present))
+					{
+						table[entry] = allocate_zeroed_page_aligned_page();
+						ASSERT(table[entry]);
+					}
+
+					table[entry] |= flags;
+
+					return table[entry] & s_page_addr_mask;
+				};
+
+			const paddr_t pml4 = s_kernel->m_highest_paging_struct;
+			const paddr_t pdpt = get_or_allocate_entry(pml4, pml4e, Flags::ReadWrite | Flags::Present);
+			const paddr_t pd   = get_or_allocate_entry(pdpt, pdpte, Flags::ReadWrite | Flags::Present);
+			s_fast_page_pt = P2V(get_or_allocate_entry(pd,   pde,   Flags::ReadWrite | Flags::Present));
+		}
+
+		s_kernel->load();
 	}
 
 	PageTable& PageTable::kernel()
@@ -440,12 +328,12 @@ namespace Kernel
 		return true;
 	}
 
-	void PageTable::initialize_kernel()
+	void PageTable::map_kernel_memory()
 	{
 		// Map (phys_kernel_start -> phys_kernel_end) to (virt_kernel_start -> virt_kernel_end)
 		const vaddr_t kernel_start = reinterpret_cast<vaddr_t>(g_kernel_start);
 		map_range_at(
-			V2P(kernel_start),
+			kernel_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
 			kernel_start,
 			g_kernel_end - g_kernel_start,
 			Flags::Present
@@ -454,7 +342,7 @@ namespace Kernel
 		// Map executable kernel memory as executable
 		const vaddr_t kernel_execute_start = reinterpret_cast<vaddr_t>(g_kernel_execute_start);
 		map_range_at(
-			V2P(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
@@ -463,7 +351,7 @@ namespace Kernel
 		// Map writable kernel memory as writable
 		const vaddr_t kernel_writable_start = reinterpret_cast<vaddr_t>(g_kernel_writable_start);
 		map_range_at(
-			V2P(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
@@ -472,109 +360,58 @@ namespace Kernel
 		// Map userspace memory
 		const vaddr_t userspace_start = reinterpret_cast<vaddr_t>(g_userspace_start);
 		map_range_at(
-			V2P(userspace_start),
+			userspace_start - KERNEL_OFFSET + g_boot_info.kernel_paddr,
 			userspace_start,
 			g_userspace_end - g_userspace_start,
 			Flags::Execute | Flags::UserSupervisor | Flags::Present
 		);
 	}
 
-	void PageTable::prepare_fast_page()
-	{
-		constexpr vaddr_t uc_vaddr = uncanonicalize(fast_page());
-		constexpr uint64_t pml4e = (uc_vaddr >> 39) & 0x1FF;
-		constexpr uint64_t pdpte = (uc_vaddr >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (uc_vaddr >> 21) & 0x1FF;
-
-		uint64_t* pml4 = P2V(m_highest_paging_struct);
-		ASSERT(!(pml4[pml4e] & Flags::Present));
-		pml4[pml4e] = allocate_zeroed_page_aligned_page() | Flags::ReadWrite | Flags::Present;
-
-		uint64_t* pdpt = P2V(pml4[pml4e] & s_page_addr_mask);
-		ASSERT(!(pdpt[pdpte] & Flags::Present));
-		pdpt[pdpte] = allocate_zeroed_page_aligned_page() | Flags::ReadWrite | Flags::Present;
-
-		uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
-		ASSERT(!(pd[pde] & Flags::Present));
-		pd[pde] = allocate_zeroed_page_aligned_page() | Flags::ReadWrite | Flags::Present;
-	}
-
 	void PageTable::map_fast_page(paddr_t paddr)
 	{
-		ASSERT(s_kernel);
-		ASSERT(paddr);
-		ASSERT(paddr % PAGE_SIZE == 0);
+		ASSERT(paddr && paddr % PAGE_SIZE == 0);
 
+		ASSERT(s_fast_page_pt);
 		ASSERT(s_fast_page_lock.current_processor_has_lock());
 
-		constexpr vaddr_t uc_vaddr = uncanonicalize(fast_page());
-		constexpr uint64_t pml4e = (uc_vaddr >> 39) & 0x1FF;
-		constexpr uint64_t pdpte = (uc_vaddr >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (uc_vaddr >> 21) & 0x1FF;
-		constexpr uint64_t pte   = (uc_vaddr >> 12) & 0x1FF;
+		ASSERT(!(*s_fast_page_pt & Flags::Present));
+		s_fast_page_pt[0] = paddr | Flags::ReadWrite | Flags::Present;
 
-		const uint64_t* pml4 = P2V(s_kernel->m_highest_paging_struct);
-		const uint64_t* pdpt = P2V(pml4[pml4e] & s_page_addr_mask);
-		const uint64_t* pd   = P2V(pdpt[pdpte] & s_page_addr_mask);
-		      uint64_t* pt   = P2V(pd[pde] & s_page_addr_mask);
-
-		ASSERT(!(pt[pte] & Flags::Present));
-		pt[pte] = paddr | Flags::ReadWrite | Flags::Present;
-
-		asm volatile("invlpg (%0)" :: "r"(fast_page()) : "memory");
+		asm volatile("invlpg (%0)" :: "r"(fast_page()));
 	}
 
 	void PageTable::unmap_fast_page()
 	{
-		ASSERT(s_kernel);
-
+		ASSERT(s_fast_page_pt);
 		ASSERT(s_fast_page_lock.current_processor_has_lock());
 
-		constexpr vaddr_t uc_vaddr = uncanonicalize(fast_page());
-		constexpr uint64_t pml4e = (uc_vaddr >> 39) & 0x1FF;
-		constexpr uint64_t pdpte = (uc_vaddr >> 30) & 0x1FF;
-		constexpr uint64_t pde   = (uc_vaddr >> 21) & 0x1FF;
-		constexpr uint64_t pte   = (uc_vaddr >> 12) & 0x1FF;
+		ASSERT((*s_fast_page_pt & Flags::Present));
+		s_fast_page_pt[0] = 0;
 
-		const uint64_t* pml4 = P2V(s_kernel->m_highest_paging_struct);
-		const uint64_t* pdpt = P2V(pml4[pml4e] & s_page_addr_mask);
-		const uint64_t* pd   = P2V(pdpt[pdpte] & s_page_addr_mask);
-		      uint64_t* pt   = P2V(pd[pde] & s_page_addr_mask);
-
-		ASSERT(pt[pte] & Flags::Present);
-		pt[pte] = 0;
-
-		asm volatile("invlpg (%0)" :: "r"(fast_page()) : "memory");
+		asm volatile("invlpg (%0)" :: "r"(fast_page()));
 	}
 
 	BAN::ErrorOr<PageTable*> PageTable::create_userspace()
 	{
 		SpinLockGuard _(s_kernel->m_lock);
+
 		PageTable* page_table = new PageTable;
 		if (page_table == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
-		page_table->map_kernel_memory();
+
+		page_table->m_highest_paging_struct = allocate_zeroed_page_aligned_page();
+		if (page_table->m_highest_paging_struct == 0)
+		{
+			delete page_table;
+			return BAN::Error::from_errno(ENOMEM);
+		}
+
+		uint64_t* pml4 = P2V(page_table->m_highest_paging_struct);
+		memcpy(pml4, s_global_pml4_entries, sizeof(s_global_pml4_entries));
+
 		return page_table;
 	}
 
-	void PageTable::map_kernel_memory()
-	{
-		ASSERT(s_kernel);
-		ASSERT(s_global_pml4_entries[511]);
-		ASSERT(m_highest_paging_struct == 0);
-		m_highest_paging_struct = allocate_zeroed_page_aligned_page();
-
-		PageTable::with_fast_page(m_highest_paging_struct, [] {
-			for (size_t i = 0; i < 512; i++)
-			{
-				if (s_global_pml4_entries[i] == 0)
-					continue;
-				ASSERT(i >= 256);
-				PageTable::fast_page_as_sized<uint64_t>(i) = s_global_pml4_entries[i];
-			}
-		});
-	}
-
 	PageTable::~PageTable()
 	{
 		if (m_highest_paging_struct == 0)
@@ -612,10 +449,39 @@ namespace Kernel
 		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 <= 32 || !s_is_initialized)
+		{
+			for (size_t i = 0; i < pages; i++, vaddr += PAGE_SIZE)
+				asm volatile("invlpg (%0)" :: "r"(vaddr));
+		}
+		else if (is_userspace || !s_has_pge)
+		{
+			asm volatile("movq %0, %%cr3" :: "r"(m_highest_paging_struct));
+		}
+		else
+		{
+			asm volatile(
+				"movq %%cr4, %%rax;"
+
+				"andq $~0x80, %%rax;"
+				"movq %%rax, %%cr4;"
+
+				"movq %0, %%cr3;"
+
+				"orq $0x80, %%rax;"
+				"movq %%rax, %%cr4;"
+				:
+				: "r"(m_highest_paging_struct)
+				: "rax"
+			);
+		}
 
 		if (send_smp_message)
 		{
@@ -623,14 +489,14 @@ 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::unmap_page(vaddr_t vaddr, bool invalidate)
 	{
 		ASSERT(vaddr);
 		ASSERT(vaddr != fast_page());
@@ -663,31 +529,23 @@ namespace Kernel
 
 		pt[pte] = 0;
 
-		if (old_paddr != 0)
-			invalidate(vaddr, send_smp_message);
+		if (invalidate && old_paddr != 0)
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::unmap_range(vaddr_t vaddr, size_t size)
 	{
 		ASSERT(vaddr % PAGE_SIZE == 0);
 
-		size_t page_count = range_page_count(vaddr, size);
+		const size_t page_count = range_page_count(vaddr, size);
 
 		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,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
-			}
-		});
+		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());
@@ -752,8 +610,8 @@ namespace Kernel
 
 		pt[pte] = paddr | uwr_flags | extra_flags;
 
-		if (old_paddr != 0)
-			invalidate(vaddr, send_smp_message);
+		if (invalidate && old_paddr != 0)
+			invalidate_page(vaddr, true);
 	}
 
 	void PageTable::map_range_at(paddr_t paddr, vaddr_t vaddr, size_t size, flags_t flags, MemoryType memory_type)
@@ -769,15 +627,66 @@ 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);
+		invalidate_range(vaddr, page_count, true);
+	}
 
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = page_count,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
+	void PageTable::remove_writable_from_range(vaddr_t vaddr, size_t size)
+	{
+		ASSERT(vaddr);
+		ASSERT(vaddr % PAGE_SIZE == 0);
+
+		ASSERT(is_canonical(vaddr));
+		ASSERT(is_canonical(vaddr + size - 1));
+
+		const vaddr_t uc_vaddr_start = uncanonicalize(vaddr);
+		const vaddr_t uc_vaddr_end   = uncanonicalize(vaddr + size - 1);
+
+		uint16_t pml4e = (uc_vaddr_start >> 39) & 0x1FF;
+		uint16_t pdpte = (uc_vaddr_start >> 30) & 0x1FF;
+		uint16_t pde   = (uc_vaddr_start >> 21) & 0x1FF;
+		uint16_t pte   = (uc_vaddr_start >> 12) & 0x1FF;
+
+		const uint16_t e_pml4e = (uc_vaddr_end >> 39) & 0x1FF;
+		const uint16_t e_pdpte = (uc_vaddr_end >> 30) & 0x1FF;
+		const uint16_t e_pde   = (uc_vaddr_end >> 21) & 0x1FF;
+		const uint16_t e_pte   = (uc_vaddr_end >> 12) & 0x1FF;
+
+		SpinLockGuard _(m_lock);
+
+		const uint64_t* pml4 = P2V(m_highest_paging_struct);
+		for (; pml4e <= e_pml4e; pml4e++)
+		{
+			if (!(pml4[pml4e] & Flags::ReadWrite))
+				continue;
+			const uint64_t* pdpt = P2V(pml4[pml4e] & s_page_addr_mask);
+			for (; pdpte < 512; pdpte++)
+			{
+				if (pml4e == e_pml4e && pdpte > e_pdpte)
+					break;
+				if (!(pdpt[pdpte] & Flags::ReadWrite))
+					continue;
+				const uint64_t* pd = P2V(pdpt[pdpte] & s_page_addr_mask);
+				for (; pde < 512; pde++)
+				{
+					if (pml4e == e_pml4e && pdpte == e_pdpte && pde > e_pde)
+						break;
+					if (!(pd[pde] & Flags::ReadWrite))
+						continue;
+					uint64_t* pt = P2V(pd[pde] & s_page_addr_mask);
+					for (; pte < 512; pte++)
+					{
+						if (pml4e == e_pml4e && pdpte == e_pdpte && pde == e_pde && pte > e_pte)
+							break;
+						pt[pte] &= ~static_cast<uint64_t>(Flags::ReadWrite);
+					}
+					pte = 0;
+				}
+				pde = 0;
 			}
-		});
+			pdpte = 0;
+		}
+
+		invalidate_range(vaddr, size / PAGE_SIZE, true);
 	}
 
 	uint64_t PageTable::get_page_data(vaddr_t vaddr) const
@@ -824,13 +733,13 @@ namespace Kernel
 		return page_data & s_page_addr_mask;
 	}
 
-	bool PageTable::reserve_page(vaddr_t vaddr, bool only_free, bool send_smp_message)
+	bool PageTable::reserve_page(vaddr_t vaddr, bool only_free, bool invalidate)
 	{
 		SpinLockGuard _(m_lock);
 		ASSERT(vaddr % PAGE_SIZE == 0);
 		if (only_free && !is_page_free(vaddr))
 			return false;
-		map_page_at(0, vaddr, Flags::Reserved, MemoryType::Normal, send_smp_message);
+		map_page_at(0, vaddr, Flags::Reserved, MemoryType::Normal, invalidate);
 		return true;
 	}
 
@@ -845,14 +754,7 @@ namespace Kernel
 			return false;
 		for (size_t offset = 0; offset < bytes; offset += PAGE_SIZE)
 			reserve_page(vaddr + offset, true, false);
-		Processor::broadcast_smp_message({
-			.type = Processor::SMPMessage::Type::FlushTLB,
-			.flush_tlb = {
-				.vaddr      = vaddr,
-				.page_count = bytes / PAGE_SIZE,
-				.page_table = vaddr < KERNEL_OFFSET ? this : nullptr,
-			}
-		});
+		invalidate_range(vaddr, bytes / PAGE_SIZE, true);
 		return true;
 	}
 
@@ -866,9 +768,9 @@ namespace Kernel
 			last_address -= rem;
 
 		ASSERT(is_canonical(first_address));
-		ASSERT(is_canonical(last_address));
+		ASSERT(is_canonical(last_address - 1));
 		const vaddr_t uc_vaddr_start = uncanonicalize(first_address);
-		const vaddr_t uc_vaddr_end   = uncanonicalize(last_address);
+		const vaddr_t uc_vaddr_end   = uncanonicalize(last_address - 1);
 
 		uint16_t pml4e = (uc_vaddr_start >> 39) & 0x1FF;
 		uint16_t pdpte = (uc_vaddr_start >> 30) & 0x1FF;
@@ -885,10 +787,8 @@ namespace Kernel
 		// Try to find free page that can be mapped without
 		// allocations (page table with unused entries)
 		const uint64_t* pml4 = P2V(m_highest_paging_struct);
-		for (; pml4e < 512; pml4e++)
+		for (; pml4e <= e_pml4e; pml4e++)
 		{
-			if (pml4e > e_pml4e)
-				break;
 			if (!(pml4[pml4e] & Flags::Present))
 				continue;
 			const uint64_t* pdpt = P2V(pml4[pml4e] & s_page_addr_mask);
@@ -908,22 +808,24 @@ namespace Kernel
 					const uint64_t* pt = P2V(pd[pde] & s_page_addr_mask);
 					for (; pte < 512; pte++)
 					{
-						if (pml4e == e_pml4e && pdpte == e_pdpte && pde == e_pde && pte >= e_pte)
+						if (pml4e == e_pml4e && pdpte == e_pdpte && pde == e_pde && pte > e_pte)
 							break;
-						if (!(pt[pte] & Flags::Used))
-						{
-							vaddr_t vaddr = 0;
-							vaddr |= static_cast<uint64_t>(pml4e) << 39;
-							vaddr |= static_cast<uint64_t>(pdpte) << 30;
-							vaddr |= static_cast<uint64_t>(pde)   << 21;
-							vaddr |= static_cast<uint64_t>(pte)   << 12;
-							vaddr = canonicalize(vaddr);
-							ASSERT(reserve_page(vaddr));
-							return vaddr;
-						}
+						if (pt[pte] & Flags::Used)
+							continue;
+						vaddr_t vaddr = 0;
+						vaddr |= static_cast<uint64_t>(pml4e) << 39;
+						vaddr |= static_cast<uint64_t>(pdpte) << 30;
+						vaddr |= static_cast<uint64_t>(pde)   << 21;
+						vaddr |= static_cast<uint64_t>(pte)   << 12;
+						vaddr = canonicalize(vaddr);
+						ASSERT(reserve_page(vaddr));
+						return vaddr;
 					}
+					pte = 0;
 				}
+				pde = 0;
 			}
+			pdpte = 0;
 		}
 
 		for (vaddr_t uc_vaddr = uc_vaddr_start; uc_vaddr < uc_vaddr_end; uc_vaddr += PAGE_SIZE)
@@ -935,7 +837,7 @@ namespace Kernel
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	vaddr_t PageTable::reserve_free_contiguous_pages(size_t page_count, vaddr_t first_address, vaddr_t last_address)
@@ -948,7 +850,7 @@ namespace Kernel
 			last_address -= rem;
 
 		ASSERT(is_canonical(first_address));
-		ASSERT(is_canonical(last_address));
+		ASSERT(is_canonical(last_address - 1));
 
 		SpinLockGuard _(m_lock);
 
@@ -977,7 +879,7 @@ namespace Kernel
 			}
 		}
 
-		ASSERT_NOT_REACHED();
+		return 0;
 	}
 
 	bool PageTable::is_page_free(vaddr_t page) const

kernel/arch/x86_64/Signal.S

@@ -1,12 +1,13 @@
 .section .userspace, "ax"
 
 // stack contains
-//    return address
-//    return stack
-//    return rflags
-//    siginfo_t
-//    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:
@@ -26,6 +27,10 @@ signal_trampoline:
 	pushq %rax
 	pushq %rbp
 
+	movq 208(%rsp), %rax
+	pushq %rax; addq $(128 + 8), (%rsp)
+	pushq (%rax)
+
 	// FIXME: populate these
 	xorq %rax, %rax
 	pushq %rax // stack
@@ -35,9 +40,9 @@ signal_trampoline:
 	pushq %rax // link
 
 	movq  %rsp,     %rdx // ucontext
-	leaq 176(%rsp), %rsi // siginfo
-	movq 168(%rsp), %rdi // signal number
-	movq 160(%rsp), %rax // handler
+	leaq 192(%rsp), %rsi // siginfo
+	movq 184(%rsp), %rdi // signal number
+	movq 176(%rsp), %rax // handler
 
 	// align stack to 16 bytes
 	movq %rsp, %rbp
@@ -55,7 +60,15 @@ signal_trampoline:
 	movq %rbp, %rsp
 	addq $40, %rsp
 
+	// restore sigmask
+	movq $83, %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
@@ -72,13 +85,13 @@ signal_trampoline:
 	popq %r14
 	popq %r15
 
-	// skip handler, number, siginfo_t
-	addq $72, %rsp
+	// skip handler, number, siginfo_t, sigmask
+	addq $80, %rsp
 
 	// restore flags
 	popfq
 
 	movq (%rsp), %rsp
 
-	// return over red-zone and siginfo_t
+	// return over red-zone
 	ret $128

kernel/arch/x86_64/Syscall.S

@@ -33,7 +33,7 @@ sys_fork_trampoline:
 
 	call read_ip
 	testq %rax, %rax
-	je .done
+	jz .done
 
 	movq %rax, %rsi
 	movq %rsp, %rdi

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 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/boot.S

@@ -11,26 +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 (1 << 2) # page align modules
-	.long -(0x1BADB002 + (1 << 2))
+	.long multiboot_flags
+	.long -(0x1BADB002 + multiboot_flags)
 
-	.long 0
-	.long 0
-	.long 0
-	.long 0
-	.long 0
+	.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:
-	.align 8
+
+.section .multiboot2, "aw"
 multiboot2_start:
 	.long 0xE85250D6
 	.long 0
@@ -66,7 +68,6 @@ multiboot2_start:
 multiboot2_end:
 
 .section .bananboot, "aw"
-	.align 8
 bananboot_start:
 	.long 0xBABAB007
 	.long -(0xBABAB007 + FB_WIDTH + FB_HEIGHT + FB_BPP)
@@ -96,27 +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
@@ -272,7 +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"
 
@@ -282,21 +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:
@@ -305,8 +310,7 @@ 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
@@ -314,28 +318,34 @@ ap_protected_mode:
 
 	# 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

@@ -1,12 +1,4 @@
-.macro swapgs_if_necessary, n
-	testb $3, \n(%rsp)
-	jz 1f; jnp 1f
-	swapgs
-1:
-.endm
-
-.macro pushaq, n
-	swapgs_if_necessary \n
+.macro intr_header, n
 	pushq %rax
 	pushq %rcx
 	pushq %rdx
@@ -22,10 +14,18 @@
 	pushq %r13
 	pushq %r14
 	pushq %r15
+	testb $3, \n+15*8(%rsp)
+	jz 1f
+	swapgs
+1:	cld
 .endm
 
-.macro popaq, n
-	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
@@ -40,12 +40,10 @@
 	popq %rdx
 	popq %rcx
 	popq %rax
-	swapgs_if_necessary \n
 .endm
 
 isr_stub:
-	pushaq 24
-	cld
+	intr_header 24
 	movq %cr0, %rax; pushq %rax
 	movq %cr2, %rax; pushq %rax
 	movq %cr3, %rax; pushq %rax
@@ -58,33 +56,33 @@ isr_stub:
 	call cpp_isr_handler
 	addq $32, %rsp
 
-	popaq 24
+	intr_footer 24
 	addq $16, %rsp
 	iretq
 
 irq_stub:
-	pushaq 24
-	cld
+	intr_header 24
+	xorq %rbp, %rbp
 	movq 120(%rsp), %rdi	# irq number
 	call cpp_irq_handler
-	popaq 24
+	intr_footer 24
 	addq $16, %rsp
 	iretq
 
 .global asm_ipi_handler
 asm_ipi_handler:
-	pushaq 8
-	cld
+	intr_header 8
+	xorq %rbp, %rbp
 	call cpp_ipi_handler
-	popaq 8
+	intr_footer 8
 	iretq
 
 .global asm_timer_handler
 asm_timer_handler:
-	pushaq 8
-	cld
+	intr_header 8
+	xorq %rbp, %rbp
 	call cpp_timer_handler
-	popaq 8
+	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)
 	{
@@ -43,6 +44,7 @@ SECTIONS
 	}
 	.bss ALIGN(4K) : AT(ADDR(.bss) - KERNEL_OFFSET)
 	{
+		g_kernel_bss_start = .;
 		*(COMMON)
 		*(.bss)
 		g_kernel_writable_end = .;

kernel/include/kernel/API/SharedPage.h

@@ -12,7 +12,7 @@ namespace Kernel::API
 
 	struct SharedPage
 	{
-		uint8_t __sequence[0x100];
+		uint16_t gdt_cpu_offset;
 
 		uint32_t features;
 

kernel/include/kernel/API/Syscall.h

@@ -1,44 +1,34 @@
 #pragma once
 
-#include <kernel/Attributes.h>
-#include <kernel/IDT.h>
-#include <stdint.h>
-#include <sys/syscall.h>
+#include <BAN/MacroUtils.h>
 
-namespace Kernel
-{
-
-	ALWAYS_INLINE long syscall(int syscall, uintptr_t arg1 = 0, uintptr_t arg2 = 0, uintptr_t arg3 = 0, uintptr_t arg4 = 0, uintptr_t arg5 = 0)
-	{
-		long ret;
-#if ARCH(x86_64)
-		register uintptr_t r10 asm("r10") = arg3;
-		register uintptr_t r8  asm( "r8") = arg4;
-		register uintptr_t r9  asm( "r9") = arg5;
-		asm volatile(
-			"syscall"
-			: "=a"(ret)
-			, "+D"(syscall)
-			, "+S"(arg1)
-			, "+d"(arg2)
-			, "+r"(r10)
-			, "+r"(r8)
-			, "+r"(r9)
-			:: "rcx", "r11", "memory");
-#elif ARCH(i686)
-		asm volatile(
-			"int %[irq]"
-			: "=a"(ret)
-			: [irq]"i"(static_cast<int>(IRQ_SYSCALL)) // WTF GCC 15
-			, "a"(syscall)
-			, "b"(arg1)
-			, "c"(arg2)
-			, "d"(arg3)
-			, "S"(arg4)
-			, "D"(arg5)
-			: "memory");
+#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
-		return ret;
-	}
 
-}
+#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

@@ -6,7 +6,7 @@
 namespace Kernel
 {
 
-	class AC97AudioController : public AudioController, public Interruptable
+	class AC97AudioController final : public AudioController, public Interruptable
 	{
 	public:
 		static BAN::ErrorOr<void> create(PCI::Device& pci_device);
@@ -23,11 +23,15 @@ namespace Kernel
 		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();

kernel/include/kernel/Audio/Controller.h

@@ -1,8 +1,11 @@
 #pragma once
 
 #include <kernel/Device/Device.h>
+#include <kernel/Memory/ByteRingBuffer.h>
 #include <kernel/PCI.h>
 
+#include <sys/ioctl.h>
+
 namespace Kernel
 {
 
@@ -16,6 +19,7 @@ namespace Kernel
 
 	protected:
 		AudioController();
+		BAN::ErrorOr<void> initialize();
 
 		virtual void handle_new_data() = 0;
 
@@ -26,8 +30,10 @@ namespace Kernel
 		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_size < m_sample_data_capacity; }
+		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; }
 
@@ -40,9 +46,9 @@ namespace Kernel
 		mutable SpinLock m_spinlock;
 
 		static constexpr size_t m_sample_data_capacity = 1 << 20;
-		uint8_t m_sample_data[m_sample_data_capacity];
-		size_t m_sample_data_head { 0 };
-		size_t m_sample_data_size { 0 };
+		BAN::UniqPtr<ByteRingBuffer> m_sample_data;
+
+		snd_volume_info m_volume_info {};
 
 	private:
 		const dev_t m_rdev;

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

@@ -8,7 +8,7 @@ namespace Kernel
 
 	class HDAudioController;
 
-	class HDAudioFunctionGroup : public AudioController
+	class HDAudioFunctionGroup final : public AudioController
 	{
 	public:
 		static BAN::ErrorOr<BAN::RefPtr<HDAudioFunctionGroup>> create(BAN::RefPtr<HDAudioController>, uint8_t cid, HDAudio::AFGNode&&);
@@ -24,6 +24,8 @@ namespace Kernel
 		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:
@@ -46,15 +48,12 @@ namespace Kernel
 		BAN::ErrorOr<void> recurse_output_paths(const HDAudio::AFGWidget& widget, BAN::Vector<const HDAudio::AFGWidget*>& path);
 
 		uint16_t get_format_data() const;
-		uint16_t get_volume_data() const;
 
 		size_t bdl_offset() const;
 
 		void queue_bdl_data();
 
 	private:
-		static constexpr size_t m_max_path_length = 16;
-
 		// use 6 512 sample BDL entries
 		// each entry is ~10.7 ms at 48 kHz
 		// -> total buffered audio is 64 ms
@@ -66,6 +65,7 @@ namespace Kernel
 		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 };

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

@@ -50,9 +50,21 @@ namespace Kernel::HDAudio
 			{
 				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;
 	};
 

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

@@ -11,6 +11,7 @@ namespace Kernel::HDAudio
 		VMIN = 0x02,
 		VMAJ = 0x03,
 		GCTL = 0x08,
+		STATESTS = 0x0E,
 
 		INTCTL = 0x20,
 		INTSTS = 0x24,

kernel/include/kernel/BootInfo.h

@@ -44,7 +44,7 @@ namespace Kernel
 	struct BootModule
 	{
 		paddr_t start;
-		size_t size;
+		uint64_t size;
 	};
 
 	struct BootInfo

kernel/include/kernel/Device/FramebufferDevice.h

@@ -37,6 +37,8 @@ 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;
 
+		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; }

kernel/include/kernel/Epoll.h

@@ -1,12 +1,8 @@
 #pragma once
 
-#include <BAN/Array.h>
-#include <BAN/CircularQueue.h>
 #include <BAN/HashMap.h>
-#include <BAN/HashSet.h>
 #include <kernel/FS/Inode.h>
 
-#include <limits.h>
 #include <sys/epoll.h>
 
 namespace Kernel
@@ -53,57 +49,52 @@ namespace Kernel
 		BAN::ErrorOr<void> fsync_impl() override { return {}; }
 
 	private:
-		struct InodeRefPtrHash
-		{
-			BAN::hash_t operator()(const BAN::RefPtr<Inode>& inode)
-			{
-				return BAN::hash<const Inode*>()(inode.ptr());
-			}
-		};
-
 		struct ListenEventList
 		{
-			BAN::Array<epoll_event, OPEN_MAX> events;
-			uint32_t bitmap[(OPEN_MAX + 31) / 32] {};
+			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
 			{
-				// For some reason having (fd < 0 || ...) makes GCC 15.1.0
-				// think bitmap access can be out of bounds...
-				if (static_cast<size_t>(fd) >= events.size())
-					return false;
-				return bitmap[fd / 32] & (1u << (fd % 32));
+				return events.contains(fd);
 			}
 
 			bool empty() const
 			{
-				for (auto val : bitmap)
-					if (val != 0)
-						return false;
-				return true;
+				return events.empty();
 			}
 
-			void add_fd(int fd, epoll_event event)
+			BAN::ErrorOr<void> add_fd(int fd, epoll_event event)
 			{
-				ASSERT(!has_fd(fd));
-				bitmap[fd / 32] |= (1u << (fd % 32));
-				events[fd] = event;
+				TRY(events.insert(fd, event));
+				return {};
 			}
 
 			void remove_fd(int fd)
 			{
-				ASSERT(has_fd(fd));
-				bitmap[fd / 32] &= ~(1u << (fd % 32));
-				events[fd] = {};
+				events.remove(fd);
 			}
 		};
 
 	private:
 		ThreadBlocker m_thread_blocker;
 		SpinLock m_ready_lock;
-		BAN::HashMap<BAN::RefPtr<Inode>, uint32_t,        InodeRefPtrHash> m_ready_events;
-		BAN::HashMap<BAN::RefPtr<Inode>, uint32_t,        InodeRefPtrHash> m_processing_events;
-		BAN::HashMap<BAN::RefPtr<Inode>, ListenEventList, InodeRefPtrHash> m_listening_events;
+		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/USTARModule.h

@@ -1,12 +1,11 @@
 #pragma once
 
 #include <kernel/BootInfo.h>
-#include <kernel/FS/FileSystem.h>
+#include <kernel/FS/Inode.h>
 
 namespace Kernel
 {
 
-	bool is_ustar_boot_module(const BootModule&);
-	BAN::ErrorOr<void> unpack_boot_module_into_filesystem(BAN::RefPtr<FileSystem>, const BootModule&);
+	BAN::ErrorOr<bool> unpack_boot_module_into_directory(BAN::RefPtr<Inode>, const BootModule&);
 
 }

kernel/include/kernel/GDT.h

@@ -133,6 +133,12 @@ namespace Kernel
 		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;
 
@@ -151,11 +157,13 @@ namespace Kernel
 
 	private:
 #if ARCH(x86_64)
-		BAN::Array<SegmentDescriptor, 8> m_gdt; // null, kernel code, kernel data, user code (32 bit), user data, user code (64 bit), tss low, tss high
-		static constexpr uint16_t m_tss_offset = 0x30;
+		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, 9> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, fsbase, gsbase, tss
-		static constexpr uint16_t m_tss_offset = 0x40;
+		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

@@ -20,7 +20,7 @@ namespace Kernel
 	constexpr uint8_t IRQ_MSI_BASE      = 0x80;
 	constexpr uint8_t IRQ_MSI_END       = 0xF0;
 #if ARCH(i686)
-	constexpr uint8_t IRQ_SYSCALL       = 0xF0;
+	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;

kernel/include/kernel/Memory/ByteRingBuffer.h

@@ -51,7 +51,7 @@ namespace Kernel
 		uint8_t back() const
 		{
 			ASSERT(!empty());
-			return reinterpret_cast<const uint8_t*>(m_vaddr)[m_tail + m_size];
+			return reinterpret_cast<const uint8_t*>(m_vaddr)[m_tail + m_size - 1];
 		}
 
 		bool empty() const { return m_size == 0; }

kernel/include/kernel/Memory/DMARegion.h

@@ -8,7 +8,7 @@ namespace Kernel
 	class DMARegion
 	{
 	public:
-		static BAN::ErrorOr<BAN::UniqPtr<DMARegion>> create(size_t size);
+		static BAN::ErrorOr<BAN::UniqPtr<DMARegion>> create(size_t size, PageTable::MemoryType type = PageTable::MemoryType::Uncached);
 		~DMARegion();
 
 		size_t size() const { return m_size; }

kernel/include/kernel/Memory/Heap.h

@@ -18,6 +18,8 @@ namespace Kernel
 		static void initialize();
 		static Heap& get();
 
+		void release_boot_modules();
+
 		paddr_t take_free_page();
 		void release_page(paddr_t);
 

kernel/include/kernel/Memory/MemoryBackedRegion.h

@@ -28,6 +28,20 @@ namespace Kernel
 
 	private:
 		MemoryBackedRegion(PageTable&, size_t size, Type, PageTable::flags_t, int status_flags);
+
+	private:
+		struct PhysicalPage
+		{
+			PhysicalPage(paddr_t paddr)
+				: paddr(paddr)
+			{ }
+			~PhysicalPage();
+
+			BAN::Atomic<uint32_t> ref_count { 1 };
+			const paddr_t paddr;
+		};
+		BAN::Vector<PhysicalPage*> m_physical_pages;
+		Mutex m_mutex;
 	};
 
 }

kernel/include/kernel/Memory/MemoryRegion.h

@@ -9,12 +9,6 @@
 namespace Kernel
 {
 
-	struct AddressRange
-	{
-		vaddr_t start;
-		vaddr_t end;
-	};
-
 	class MemoryRegion
 	{
 		BAN_NON_COPYABLE(MemoryRegion);

kernel/include/kernel/Memory/PageTable.h

@@ -46,13 +46,22 @@ namespace Kernel
 		};
 
 	public:
-		static void initialize_pre_heap();
-		static void initialize_post_heap();
+		static void initialize_fast_page();
+		static void initialize_and_load();
+
+		static void enable_cpu_features();
 
 		static PageTable& kernel();
 		static PageTable& current() { return *reinterpret_cast<PageTable*>(Processor::get_current_page_table()); }
 
-		static constexpr vaddr_t fast_page() { return KERNEL_OFFSET; }
+		static constexpr vaddr_t fast_page()
+		{
+#if ARCH(x86_64)
+			return 0xffffffffbfe00000;
+#elif ARCH(i686)
+			return 0xffe00000;
+#endif
+		}
 
 		template<with_fast_page_callback F>
 		static void with_fast_page(paddr_t paddr, F callback)
@@ -100,40 +109,42 @@ namespace Kernel
 		static BAN::ErrorOr<PageTable*> create_userspace();
 		~PageTable();
 
-		void unmap_page(vaddr_t, bool send_smp_message = true);
+		void unmap_page(vaddr_t, bool invalidate = true);
 		void unmap_range(vaddr_t, size_t bytes);
 
-		void map_page_at(paddr_t, vaddr_t, flags_t, MemoryType = MemoryType::Normal, bool send_smp_message = true);
+		void map_page_at(paddr_t, vaddr_t, flags_t, MemoryType = MemoryType::Normal, bool invalidate = true);
 		void map_range_at(paddr_t, vaddr_t, size_t bytes, flags_t, MemoryType = MemoryType::Normal);
 
+		void remove_writable_from_range(vaddr_t, size_t);
+
 		paddr_t physical_address_of(vaddr_t) const;
 		flags_t get_page_flags(vaddr_t) const;
 
 		bool is_page_free(vaddr_t) const;
 		bool is_range_free(vaddr_t, size_t bytes) const;
 
-		bool reserve_page(vaddr_t, bool only_free = true, bool send_smp_message = true);
+		bool reserve_page(vaddr_t, bool only_free = true, bool invalidate = true);
 		bool reserve_range(vaddr_t, size_t bytes, bool only_free = true);
 
 		vaddr_t reserve_free_page(vaddr_t first_address, vaddr_t last_address = UINTPTR_MAX);
 		vaddr_t reserve_free_contiguous_pages(size_t page_count, vaddr_t first_address, vaddr_t last_address = UINTPTR_MAX);
 
 		void load();
-		void initial_load();
+
+		void invalidate_page(vaddr_t addr, bool send_smp_message) { invalidate_range(addr, 1, send_smp_message); }
+		void invalidate_range(vaddr_t addr, size_t pages, bool send_smp_message);
 
 		InterruptState lock() const { return m_lock.lock(); }
 		void unlock(InterruptState state) const { m_lock.unlock(state); }
 
+		paddr_t paddr() const { return m_highest_paging_struct; }
+
 		void debug_dump();
 
 	private:
 		PageTable() = default;
 		uint64_t get_page_data(vaddr_t) const;
-		void initialize_kernel();
 		void map_kernel_memory();
-		void prepare_fast_page();
-
-		void invalidate(vaddr_t, bool send_smp_message);
 
 		static void map_fast_page(paddr_t);
 		static void unmap_fast_page();

kernel/include/kernel/Memory/PhysicalRange.h

@@ -10,7 +10,7 @@ namespace Kernel
 	class PhysicalRange
 	{
 	public:
-		PhysicalRange(paddr_t, size_t);
+		PhysicalRange(paddr_t, uint64_t);
 
 		paddr_t reserve_page();
 		void release_page(paddr_t);

kernel/include/kernel/Memory/Types.h

@@ -4,7 +4,7 @@
 
 #if ARCH(x86_64)
 	#define KERNEL_OFFSET 0xFFFFFFFF80000000
-	#define USERSPACE_END 0xFFFF800000000000
+	#define USERSPACE_END 0x800000000000
 #elif ARCH(i686)
 	#define KERNEL_OFFSET 0xC0000000
 	#define USERSPACE_END 0xC0000000
@@ -23,4 +23,10 @@ namespace Kernel
 	using vaddr_t = uintptr_t;
 	using paddr_t = uint64_t;
 
+	struct AddressRange
+	{
+		vaddr_t start;
+		vaddr_t end;
+	};
+
 }

kernel/include/kernel/Memory/VirtualRange.h

@@ -14,44 +14,27 @@ namespace Kernel
 		BAN_NON_MOVABLE(VirtualRange);
 
 	public:
-		// Create virtual range to fixed virtual address
-		static BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> create_to_vaddr(PageTable&, vaddr_t, size_t, PageTable::flags_t flags, bool preallocate_pages, bool add_guard_pages);
-		// Create virtual range to virtual address range
-		static BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> create_to_vaddr_range(PageTable&, vaddr_t vaddr_start, vaddr_t vaddr_end, size_t, PageTable::flags_t flags, bool preallocate_pages, bool add_guard_pages);
+		static BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> create_to_vaddr_range(PageTable&, AddressRange address_range, size_t, PageTable::flags_t flags, bool add_guard_pages);
 		~VirtualRange();
 
-		BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> clone(PageTable&);
-
 		vaddr_t vaddr() const { return m_vaddr + (m_has_guard_pages ? PAGE_SIZE : 0); }
 		size_t size() const { return m_size - (m_has_guard_pages ? 2 * PAGE_SIZE : 0); }
 		PageTable::flags_t flags() const { return m_flags; }
 
-		paddr_t paddr_of(vaddr_t vaddr) const
-		{
-			ASSERT(vaddr % PAGE_SIZE == 0);
-			const size_t index = (vaddr - this->vaddr()) / PAGE_SIZE;
-			ASSERT(index < m_paddrs.size());
-			const paddr_t paddr = m_paddrs[index];
-			ASSERT(paddr);
-			return paddr;
-		}
+		paddr_t paddr_of(vaddr_t vaddr) const { return m_page_table.physical_address_of(vaddr & PAGE_ADDR_MASK); }
 
 		bool contains(vaddr_t address) const { return vaddr() <= address && address < vaddr() + size(); }
 
-		BAN::ErrorOr<bool> allocate_page_for_demand_paging(vaddr_t address);
-
 	private:
-		VirtualRange(PageTable&, bool preallocated, bool has_guard_pages, vaddr_t, size_t, PageTable::flags_t);
+		VirtualRange(PageTable&, bool has_guard_pages, vaddr_t, size_t, PageTable::flags_t);
 		BAN::ErrorOr<void> initialize();
 
 	private:
 		PageTable&               m_page_table;
-		const bool               m_preallocated;
 		const bool               m_has_guard_pages;
 		const vaddr_t            m_vaddr;
 		const size_t             m_size;
 		const PageTable::flags_t m_flags;
-		BAN::Vector<paddr_t>     m_paddrs;
 		SpinLock                 m_lock;
 
 		friend class BAN::UniqPtr<VirtualRange>;

kernel/include/kernel/Memory/kmalloc.h

@@ -13,4 +13,3 @@ void* kmalloc(size_t size, size_t align, bool force_identity_map = false);
 void kfree(void*);
 
 BAN::Optional<Kernel::paddr_t> kmalloc_paddr_of(Kernel::vaddr_t);
-BAN::Optional<Kernel::vaddr_t> kmalloc_vaddr_of(Kernel::paddr_t);

kernel/include/kernel/OpenFileDescriptorSet.h

@@ -44,6 +44,10 @@ namespace Kernel
 		void close_all();
 		void close_cloexec();
 
+		bool is_cloexec(int fd);
+		void add_cloexec(int fd);
+		void remove_cloexec(int fd);
+
 		BAN::ErrorOr<void> flock(int fd, int op);
 
 		BAN::ErrorOr<size_t> read(int fd, BAN::ByteSpan);
@@ -84,27 +88,6 @@ namespace Kernel
 			friend class BAN::RefPtr<OpenFileDescription>;
 		};
 
-		struct OpenFile
-		{
-			OpenFile() = default;
-			OpenFile(BAN::RefPtr<OpenFileDescription> description, int descriptor_flags)
-				: description(BAN::move(description))
-				, descriptor_flags(descriptor_flags)
-			{ }
-
-			BAN::RefPtr<Inode> inode() const { ASSERT(description); return description->file.inode; }
-			BAN::StringView    path()  const { ASSERT(description); return description->file.canonical_path.sv(); }
-
-			int& status_flags() { ASSERT(description); return description->status_flags; }
-			const int& status_flags() const { ASSERT(description); return description->status_flags; }
-
-			off_t& offset() { ASSERT(description); return description->offset; }
-			const off_t& offset() const { ASSERT(description); return description->offset; }
-
-			BAN::RefPtr<OpenFileDescription> description;
-			int descriptor_flags { 0 };
-		};
-
 		BAN::ErrorOr<void> validate_fd(int) const;
 		BAN::ErrorOr<int> get_free_fd() const;
 		BAN::ErrorOr<void> get_free_fd_pair(int fds[2]) const;
@@ -139,7 +122,8 @@ namespace Kernel
 		const Credentials& m_credentials;
 		mutable Mutex m_mutex;
 
-		BAN::Array<OpenFile, OPEN_MAX> m_open_files;
+		BAN::Array<BAN::RefPtr<OpenFileDescription>, OPEN_MAX> m_open_files;
+		BAN::Array<uint32_t, (OPEN_MAX + 31) / 32> m_cloexec_files {};
 	};
 
 }

kernel/include/kernel/Processor.h

@@ -2,6 +2,7 @@
 
 #include <BAN/Atomic.h>
 #include <BAN/ForwardList.h>
+#include <BAN/Math.h>
 
 #include <kernel/API/SharedPage.h>
 #include <kernel/Arch.h>
@@ -58,6 +59,8 @@ namespace Kernel
 		static Processor& create(ProcessorID id);
 		static Processor& initialize();
 
+		void allocate_stack();
+
 		static ProcessorID current_id() { return read_gs_sized<ProcessorID>(offsetof(Processor, m_id)); }
 		static uint8_t current_index()  { return read_gs_sized<uint8_t>(offsetof(Processor, m_index)); }
 		static ProcessorID id_from_index(size_t index);
@@ -100,11 +103,8 @@ namespace Kernel
 				handle_smp_messages();
 		}
 
-		static uintptr_t current_stack_bottom() { return read_gs_sized<uintptr_t>(offsetof(Processor, m_stack)); }
-		static uintptr_t current_stack_top()	{ return current_stack_bottom() + s_stack_size; }
-
-		uintptr_t stack_bottom() const	{ return reinterpret_cast<uintptr_t>(m_stack); }
-		uintptr_t stack_top() const		{ return stack_bottom() + s_stack_size; }
+		vaddr_t stack_top_vaddr() const { return m_stack_vaddr + s_stack_size; }
+		paddr_t stack_top_paddr() const { return m_stack_paddr + s_stack_size; }
 
 		static void set_thread_syscall_stack(vaddr_t vaddr) { write_gs_sized<vaddr_t>(offsetof(Processor, m_thread_syscall_stack), vaddr); }
 
@@ -140,11 +140,7 @@ namespace Kernel
 		static void disable_sse()
 		{
 			uintptr_t dummy;
-#if ARCH(x86_64)
-			asm volatile("movq %%cr0, %0; orq $0x08, %0; movq %0, %%cr0" : "=r"(dummy));
-#elif ARCH(i686)
-			asm volatile("movl %%cr0, %0; orl $0x08, %0; movl %0, %%cr0" : "=r"(dummy));
-#endif
+			asm volatile("mov %%cr0, %0; or $0x08, %0; mov %0, %%cr0" : "=r"(dummy));
 		}
 
 		static void enable_sse()
@@ -169,35 +165,17 @@ namespace Kernel
 		}
 
 		template<typename T>
-		static T read_gs_sized(uintptr_t offset) requires(sizeof(T) <= 8)
+		static T read_gs_sized(uintptr_t offset) requires(sizeof(T) <= 8 && BAN::Math::is_power_of_two(sizeof(T)))
 		{
-#define __ASM_INPUT(operation) asm volatile(operation " %%gs:%a[offset], %[result]" : [result]"=r"(result) : [offset]"ir"(offset))
-			T result;
-			if constexpr(sizeof(T) == 8)
-				__ASM_INPUT("movq");
-			if constexpr(sizeof(T) == 4)
-				__ASM_INPUT("movl");
-			if constexpr(sizeof(T) == 2)
-				__ASM_INPUT("movw");
-			if constexpr(sizeof(T) == 1)
-				__ASM_INPUT("movb");
-			return result;
-#undef __ASM_INPUT
+			T value;
+			asm volatile("mov %%gs:%a[offset], %[value]" : [value]"=r"(value) : [offset]"ir"(offset));
+			return value;
 		}
 
 		template<typename T>
-		static void write_gs_sized(uintptr_t offset, T value) requires(sizeof(T) <= 8)
+		static void write_gs_sized(uintptr_t offset, T value) requires(sizeof(T) <= 8 && BAN::Math::is_power_of_two(sizeof(T)))
 		{
-#define __ASM_INPUT(operation) asm volatile(operation " %[value], %%gs:%a[offset]" :: [value]"r"(value), [offset]"ir"(offset) : "memory")
-			if constexpr(sizeof(T) == 8)
-				__ASM_INPUT("movq");
-			if constexpr(sizeof(T) == 4)
-				__ASM_INPUT("movl");
-			if constexpr(sizeof(T) == 2)
-				__ASM_INPUT("movw");
-			if constexpr(sizeof(T) == 1)
-				__ASM_INPUT("movb");
-#undef __ASM_INPUT
+			asm volatile("mov %[value], %%gs:%a[offset]" :: [value]"r"(value), [offset]"ir"(offset) : "memory");
 		}
 
 	private:
@@ -215,8 +193,9 @@ namespace Kernel
 
 		Thread* m_sse_thread { nullptr };
 
-		static constexpr size_t s_stack_size { 4096 };
-		void* m_stack { nullptr };
+		static constexpr size_t s_stack_size { PAGE_SIZE };
+		vaddr_t m_stack_vaddr { 0 };
+		paddr_t m_stack_paddr { 0 };
 
 		GDT* m_gdt { nullptr };
 		IDT* m_idt { nullptr };

kernel/include/kernel/Scheduler.h

@@ -101,7 +101,7 @@ namespace Kernel
 		InterruptStack* m_interrupt_stack { nullptr };
 		InterruptRegisters* m_interrupt_registers { nullptr };
 
-		uint64_t m_last_reschedule_ns { 0 };
+		uint64_t m_next_reschedule_ns { 0 };
 		uint64_t m_last_load_balance_ns { 0 };
 
 		struct ThreadInfo

kernel/include/kernel/SchedulerQueueNode.h

@@ -22,8 +22,9 @@ namespace Kernel
 
 		uint64_t wake_time_ns { static_cast<uint64_t>(-1) };
 
-		SpinLock blocker_lock;
-		ThreadBlocker* blocker { nullptr };
+		BAN::Atomic<ThreadBlocker*> blocker { nullptr };
+		SchedulerQueueNode* block_chain_prev { nullptr };
+		SchedulerQueueNode* block_chain_next { nullptr };
 
 		ProcessorID processor_id { PROCESSOR_NONE };
 		bool blocked { false };

kernel/include/kernel/Thread.h

@@ -38,8 +38,12 @@ namespace Kernel
 		//        stack overflows on some machines with 8 page stack
 		static constexpr size_t kernel_stack_size    { PAGE_SIZE * 16 };
 
-		// TODO: userspace stack is hard limited to 32 MiB, maybe make this dynamic?
+		// TODO: userspace stack size is hard limited, maybe make this dynamic?
+#if ARCH(x86_64)
 		static constexpr size_t userspace_stack_size { 32 << 20 };
+#elif ARCH(i686)
+		static constexpr size_t userspace_stack_size { 4 << 20 };
+#endif
 
 	public:
 		static BAN::ErrorOr<Thread*> create_kernel(entry_t, void*);
@@ -56,12 +60,10 @@ namespace Kernel
 		// Returns true, if thread is going to trigger signal
 		bool is_interrupted_by_signal(bool skip_stop_and_cont = false) const;
 
-		// Returns true if pending signal can be added to thread
-		bool can_add_signal_to_execute() const;
-		bool will_execute_signal() const;
 		// Returns true if handled signal had SA_RESTART
-		bool handle_signal(int signal = 0, const siginfo_t& signal_info = {});
-		void add_signal(int signal, const siginfo_t& info);
+		bool handle_signal_if_interrupted();
+		bool handle_signal(int signal, const siginfo_t&);
+		void add_signal(int signal, const siginfo_t&);
 		void set_suspend_signal_mask(uint64_t sigmask);
 
 		static bool is_stopping_signal(int signal);
@@ -153,6 +155,16 @@ namespace Kernel
 
 		bool currently_on_alternate_stack() const;
 
+		struct signal_handle_info_t
+		{
+			vaddr_t handler;
+			vaddr_t stack_top;
+			uint64_t restore_sigmask;
+			bool has_sa_restart;
+		};
+		signal_handle_info_t remove_signal_and_get_info(int signal);
+		void handle_signal_impl(int signal, const siginfo_t&, const signal_handle_info_t&);
+
 	private:
 		// NOTE: this is the first member to force it being last destructed
 		//       {kernel,userspace}_stack has to be destroyed before page table

kernel/include/kernel/ThreadBlocker.h

@@ -15,7 +15,6 @@ namespace Kernel
 		void block_with_wake_time_ns(uint64_t wake_time_ns, BaseMutex*);
 		void unblock();
 
-
 		void block_with_timeout_ms(uint64_t timeout_ms, BaseMutex* mutex)
 		{
 			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(timeout_ms, 1'000'000));
@@ -29,14 +28,12 @@ namespace Kernel
 
 	private:
 		void add_thread_to_block_queue(SchedulerQueue::Node*);
-		void remove_blocked_thread(SchedulerQueue::Node*);
+		void remove_thread_from_block_queue(SchedulerQueue::Node*);
 
 	private:
+		SchedulerQueue::Node* m_block_chain { nullptr };
 		SpinLock m_lock;
 
-		SchedulerQueue::Node* m_block_chain[32] {};
-		size_t m_block_chain_length { 0 };
-
 		friend class Scheduler;
 	};
 

kernel/kernel/ACPI/ACPI.cpp

@@ -531,11 +531,8 @@ acpi_release_global_lock:
 			return BAN::Error::from_errno(EFAULT);
 		}
 
-		if (!s5_node.as.package->elements[0].resolved || !s5_node.as.package->elements[1].resolved)
-		{
-			dwarnln("TODO: lazy evaluate package \\_S5 elements");
-			return BAN::Error::from_errno(ENOTSUP);
-		}
+		TRY(AML::resolve_package_element(s5_node.as.package->elements[0], true));
+		TRY(AML::resolve_package_element(s5_node.as.package->elements[1], true));
 
 		auto slp_typa_node = TRY(AML::convert_node(TRY(s5_node.as.package->elements[0].value.node->copy()), AML::ConvInteger, sizeof(uint64_t)));
 		auto slp_typb_node = TRY(AML::convert_node(TRY(s5_node.as.package->elements[1].value.node->copy()), AML::ConvInteger, sizeof(uint64_t)));

kernel/kernel/APIC.cpp

@@ -293,9 +293,25 @@ namespace Kernel
 			dprintln("Trying to enable processor (lapic id {})", processor.apic_id);
 
 			auto& proc = Kernel::Processor::create(ProcessorID(processor.apic_id));
+			proc.allocate_stack();
+
+			struct ap_init_info_t
+			{
+				uintptr_t stack_paddr;
+				uintptr_t stack_vaddr;
+				uintptr_t prepare_paging;
+				uintptr_t page_table;
+				uintptr_t ready;
+			};
+
 			PageTable::with_fast_page(ap_init_paddr, [&] {
-				PageTable::fast_page_as_sized<uint32_t>(2) = kmalloc_paddr_of(proc.stack_top()).value();
-				PageTable::fast_page_as_sized<uint8_t>(13) = 0;
+				PageTable::fast_page_as<ap_init_info_t>(8) = {
+					.stack_paddr    = static_cast<uintptr_t>(proc.stack_top_paddr()),
+					.stack_vaddr    = proc.stack_top_vaddr(),
+					.prepare_paging = reinterpret_cast<uintptr_t>(&PageTable::enable_cpu_features),
+					.page_table     = static_cast<uintptr_t>(PageTable::kernel().paddr()),
+					.ready          = 0,
+				};
 			});
 
 			write_to_local_apic(LAPIC_ERROR_REG, 0x00);
@@ -334,12 +350,9 @@ namespace Kernel
 
 			// give processor upto 100 * 100 us + 200 us to boot
 			PageTable::with_fast_page(ap_init_paddr, [&] {
-				for (int i = 0; i < 100; i++)
-				{
-					if (__atomic_load_n(&PageTable::fast_page_as_sized<uint8_t>(13), __ATOMIC_SEQ_CST))
+				for (int i = 0; i < 100; i++, udelay(100))
+					if (__atomic_load_n(&PageTable::fast_page_as<ap_init_info_t>(8).ready, __ATOMIC_SEQ_CST))
 						break;
-					udelay(100);
-				}
 			});
 
 			initialized_aps++;

kernel/kernel/Audio/AC97/Controller.cpp

@@ -118,6 +118,8 @@ namespace Kernel
 
 	BAN::ErrorOr<void> AC97AudioController::initialize()
 	{
+		TRY(AudioController::initialize());
+
 		m_pci_device.enable_bus_mastering();
 
 		m_mixer = TRY(m_pci_device.allocate_bar_region(0));
@@ -135,8 +137,27 @@ namespace Kernel
 		// Reset mixer to default values
 		m_mixer->write16(AudioMixerRegister::Reset, 0);
 
-		// Master volume 100%, no mute
-		m_mixer->write16(AudioMixerRegister::MasterVolume, 0x0000);
+		// Master volumes
+		m_mixer->write16(AudioMixerRegister::MasterVolume, 0x2020);
+		if (m_mixer->read16(AudioMixerRegister::MasterVolume) == 0x2020)
+		{
+			m_volume_info = {
+				.min_mdB = -94500,
+				.max_mdB = 0,
+				.step_mdB = 1500,
+				.mdB = 0,
+			};
+		}
+		else
+		{
+			m_volume_info = {
+				.min_mdB = -46500,
+				.max_mdB = 0,
+				.step_mdB = 1500,
+				.mdB = 0,
+			};
+		}
+		m_mixer->write16(AudioMixerRegister::MasterVolume, get_volume_data());
 
 		// PCM output volume left/right +0 db, no mute
 		m_mixer->write16(AudioMixerRegister::PCMOutVolume, 0x0808);
@@ -185,6 +206,19 @@ namespace Kernel
 		return {};
 	}
 
+	uint32_t AC97AudioController::get_volume_data() const
+	{
+		const uint32_t steps = (-m_volume_info.mdB + m_volume_info.step_mdB / 2) / m_volume_info.step_mdB;
+		return (steps << 8) | steps;
+	}
+
+	BAN::ErrorOr<void> AC97AudioController::set_volume_mdB(int32_t mdB)
+	{
+		m_volume_info.mdB = BAN::Math::clamp(mdB, m_volume_info.min_mdB, m_volume_info.max_mdB);
+		m_mixer->write16(AudioMixerRegister::MasterVolume, get_volume_data());
+		return {};
+	}
+
 	void AC97AudioController::handle_new_data()
 	{
 		ASSERT(m_spinlock.current_processor_has_lock());
@@ -203,23 +237,22 @@ namespace Kernel
 			if (next_bld_head == m_bdl_tail)
 				break;
 
-			const size_t sample_data_tail = (m_sample_data_head + m_sample_data_capacity - m_sample_data_size) % m_sample_data_capacity;
-
-			const size_t max_memcpy = BAN::Math::min(m_sample_data_size, m_sample_data_capacity - sample_data_tail);
-			const size_t samples = BAN::Math::min(max_memcpy / 2, m_samples_per_entry);
-			if (samples == 0)
+			const size_t sample_frames = BAN::Math::min(m_sample_data->size() / get_channels() / sizeof(uint16_t), m_samples_per_entry / get_channels());
+			if (sample_frames == 0)
 				break;
 
+			const size_t copy_total_bytes = sample_frames * get_channels() * sizeof(uint16_t);
+
 			auto& entry = reinterpret_cast<AC97::BufferDescriptorListEntry*>(m_bdl_region->vaddr())[m_bdl_head];
-			entry.samples = samples;
+			entry.samples = sample_frames * get_channels();
 			entry.flags = (1 << 15);
 			memcpy(
 				reinterpret_cast<void*>(m_bdl_region->paddr_to_vaddr(entry.address)),
-				&m_sample_data[sample_data_tail],
-				samples * 2
+				m_sample_data->get_data().data(),
+				copy_total_bytes
 			);
 
-			m_sample_data_size -= samples * 2;
+			m_sample_data->pop(copy_total_bytes);
 
 			lvi = m_bdl_head;
 			m_bdl_head = next_bld_head;

kernel/kernel/Audio/Controller.cpp

@@ -53,34 +53,28 @@ namespace Kernel
 		return {};
 	}
 
+	BAN::ErrorOr<void> AudioController::initialize()
+	{
+		m_sample_data = TRY(ByteRingBuffer::create(m_sample_data_capacity));
+		return {};
+	}
+
 	BAN::ErrorOr<size_t> AudioController::write_impl(off_t, BAN::ConstByteSpan buffer)
 	{
 		SpinLockGuard lock_guard(m_spinlock);
 
-		while (m_sample_data_size >= m_sample_data_capacity)
+		while (m_sample_data->full())
 		{
 			SpinLockGuardAsMutex smutex(lock_guard);
 			TRY(Thread::current().block_or_eintr_indefinite(m_sample_data_blocker, &smutex));
 		}
 
-		size_t nwritten = 0;
-		while (nwritten < buffer.size())
-		{
-			if (m_sample_data_size >= m_sample_data_capacity)
-				break;
-
-			const size_t max_memcpy = BAN::Math::min(m_sample_data_capacity - m_sample_data_size, m_sample_data_capacity - m_sample_data_head);
-			const size_t to_copy = BAN::Math::min(buffer.size() - nwritten, max_memcpy);
-			memcpy(m_sample_data + m_sample_data_head, buffer.data() + nwritten, to_copy);
-
-			nwritten += to_copy;
-			m_sample_data_head = (m_sample_data_head + to_copy) % m_sample_data_capacity;
-			m_sample_data_size += to_copy;
-		}
+		const size_t to_copy = BAN::Math::min(buffer.size(), m_sample_data->free());
+		m_sample_data->push(buffer.slice(0, to_copy));
 
 		handle_new_data();
 
-		return nwritten;
+		return to_copy;
 	}
 
 	BAN::ErrorOr<long> AudioController::ioctl_impl(int cmd, void* arg)
@@ -97,9 +91,9 @@ namespace Kernel
 			case SND_GET_BUFFERSZ:
 			{
 				SpinLockGuard _(m_spinlock);
-				*static_cast<uint32_t*>(arg) = m_sample_data_size;
+				*static_cast<uint32_t*>(arg) = m_sample_data->size();
 				if (cmd == SND_RESET_BUFFER)
-					m_sample_data_size = 0;
+					m_sample_data->pop(m_sample_data->size());
 				return 0;
 			}
 			case SND_GET_TOTAL_PINS:
@@ -111,6 +105,12 @@ namespace Kernel
 			case SND_SET_PIN:
 				TRY(set_current_pin(*static_cast<uint32_t*>(arg)));
 				return 0;
+			case SND_GET_VOLUME_INFO:
+				*static_cast<snd_volume_info*>(arg) = m_volume_info;
+				return 0;
+			case SND_SET_VOLUME_MDB:
+				TRY(set_volume_mdB(*static_cast<int32_t*>(arg)));
+				return 0;
 		}
 
 		return CharacterDevice::ioctl_impl(cmd, arg);

kernel/kernel/Audio/HDAudio/AudioFunctionGroup.cpp

@@ -2,6 +2,8 @@
 #include <kernel/Audio/HDAudio/Registers.h>
 #include <kernel/FS/DevFS/FileSystem.h>
 
+#include <BAN/Sort.h>
+
 namespace Kernel
 {
 
@@ -25,6 +27,8 @@ namespace Kernel
 
 	BAN::ErrorOr<void> HDAudioFunctionGroup::initialize()
 	{
+		TRY(AudioController::initialize());
+
 		if constexpr(DEBUG_HDAUDIO)
 		{
 			const auto widget_to_string =
@@ -50,19 +54,13 @@ namespace Kernel
 			{
 				if (widget.type == HDAudio::AFGWidget::Type::PinComplex)
 				{
-					const uint32_t config = TRY(m_controller->send_command({
-						.data = 0x00,
-						.command = 0xF1C,
-						.node_index = widget.id,
-						.codec_address = m_cid,
-					}));
-
-					dprintln("  widget {}: {} ({}, {}), {32b}",
+					dprintln("  widget {}: {} ({}, {}, {}), {32b}",
 						widget.id,
 						widget_to_string(widget.type),
 						(int)widget.pin_complex.output,
 						(int)widget.pin_complex.input,
-						config
+						(int)widget.pin_complex.display,
+						widget.pin_complex.config
 					);
 				}
 				else
@@ -79,59 +77,49 @@ namespace Kernel
 		}
 
 		TRY(initialize_stream());
-		TRY(initialize_output());
+
+		if (auto ret = initialize_output(); ret.is_error())
+		{
+			// No usable pins, not really an error
+			if (ret.error().get_error_code() == ENODEV)
+				return {};
+			return ret.release_error();
+		}
+
 		DevFileSystem::get().add_device(this);
+
 		return {};
 	}
 
 	uint32_t HDAudioFunctionGroup::get_total_pins() const
 	{
-		uint32_t count = 0;
-		for (const auto& widget : m_afg_node.widgets)
-			if (widget.type == HDAudio::AFGWidget::Type::PinComplex && widget.pin_complex.output)
-				count++;
-		return count;
+		return m_output_pins.size();
 	}
 
 	uint32_t HDAudioFunctionGroup::get_current_pin() const
 	{
 		const auto current_id = m_output_paths[m_output_path_index].front()->id;
-
-		uint32_t pin = 0;
-		for (const auto& widget : m_afg_node.widgets)
-		{
-			if (widget.type != HDAudio::AFGWidget::Type::PinComplex || !widget.pin_complex.output)
-				continue;
-			if (widget.id == current_id)
-				return pin;
-			pin++;
-		}
-
+		for (size_t i = 0; i < m_output_pins.size(); i++)
+			if (m_output_pins[i]->id == current_id)
+				return i;
 		ASSERT_NOT_REACHED();
 	}
 
 	BAN::ErrorOr<void> HDAudioFunctionGroup::set_current_pin(uint32_t pin)
 	{
-		uint32_t pin_id = 0;
-		for (const auto& widget : m_afg_node.widgets)
-		{
-			if (widget.type != HDAudio::AFGWidget::Type::PinComplex || !widget.pin_complex.output)
-				continue;
-			if (pin-- > 0)
-				continue;
-			pin_id = widget.id;
-			break;
-		}
+		if (pin >= m_output_pins.size())
+			return BAN::Error::from_errno(EINVAL);
 
 		if (auto ret = disable_output_path(m_output_path_index); ret.is_error())
 			dwarnln("failed to disable old output path {}", ret.error());
 
+		const uint32_t pin_id = m_output_pins[pin]->id;
 		for (size_t i = 0; i < m_output_paths.size(); i++)
 		{
 			if (m_output_paths[i].front()->id != pin_id)
 				continue;
 
-			if (auto ret = enable_output_path(i); !ret.is_error())
+			if (auto ret = enable_output_path(i); ret.is_error())
 			{
 				if (ret.error().get_error_code() == ENOTSUP)
 					continue;
@@ -139,7 +127,6 @@ namespace Kernel
 				return ret.release_error();
 			}
 
-			dprintln("set output widget to {}", pin_id);
 			m_output_path_index = i;
 			return {};
 		}
@@ -149,6 +136,37 @@ namespace Kernel
 		return BAN::Error::from_errno(ENOTSUP);
 	}
 
+	BAN::ErrorOr<void> HDAudioFunctionGroup::set_volume_mdB(int32_t mdB)
+	{
+		mdB = BAN::Math::clamp(mdB, m_volume_info.min_mdB, m_volume_info.max_mdB);
+
+		const auto& path = m_output_paths[m_output_path_index];
+		for (size_t i = 0; i < path.size(); i++)
+		{
+			if (!path[i]->output_amplifier.has_value())
+				continue;
+
+			const int32_t step_round = (mdB >= 0)
+				? +m_volume_info.step_mdB / 2
+				: -m_volume_info.step_mdB / 2;
+			const uint32_t step = (mdB + step_round) / m_volume_info.step_mdB + path[i]->output_amplifier->offset;
+			const uint32_t volume = 0b1'0'1'1'0000'0'0000000 | step;
+
+			TRY(m_controller->send_command({
+				.data = static_cast<uint8_t>(volume & 0xFF),
+				.command = static_cast<uint16_t>(0x300 | (volume >> 8)),
+				.node_index = path[i]->id,
+				.codec_address = m_cid,
+			}));
+
+			break;
+		}
+
+		m_volume_info.mdB = mdB;
+
+		return {};
+	}
+
 	size_t HDAudioFunctionGroup::bdl_offset() const
 	{
 		const size_t bdl_entry_bytes = m_bdl_entry_sample_frames * get_channels() * sizeof(uint16_t);
@@ -230,18 +248,39 @@ namespace Kernel
 
 	BAN::ErrorOr<void> HDAudioFunctionGroup::initialize_output()
 	{
-		BAN::Vector<const HDAudio::AFGWidget*> path;
-		TRY(path.reserve(m_max_path_length));
-
 		for (const auto& widget : m_afg_node.widgets)
 		{
 			if (widget.type != HDAudio::AFGWidget::Type::PinComplex || !widget.pin_complex.output)
 				continue;
+
+			// no physical connection
+			if ((widget.pin_complex.config >> 30) == 0b01)
+				continue;
+
+			// needs a GPU
+			if (widget.pin_complex.display)
+				continue;
+
+			BAN::Vector<const HDAudio::AFGWidget*> path;
 			TRY(path.push_back(&widget));
 			TRY(recurse_output_paths(widget, path));
-			path.pop_back();
+
+			if (!m_output_paths.empty() && m_output_paths.back().front()->id == widget.id)
+				TRY(m_output_pins.push_back(&widget));
 		}
 
+		if (m_output_pins.empty())
+			return BAN::Error::from_errno(ENODEV);
+
+		// prefer short paths
+		BAN::sort::sort(m_output_paths.begin(), m_output_paths.end(),
+			[](const auto& a, const auto& b) {
+				if (a.front()->id != b.front()->id)
+					return a.front()->id < b.front()->id;
+				return a.size() < b.size();
+			}
+		);
+
 		dprintln_if(DEBUG_HDAUDIO, "found {} paths from output to DAC", m_output_paths.size());
 
 		// select first supported path
@@ -283,13 +322,6 @@ namespace Kernel
 		return 0b0'0'000'000'0'001'0001;
 	}
 
-	uint16_t HDAudioFunctionGroup::get_volume_data() const
-	{
-		// TODO: don't hardcode this
-		// left and right output, no mute, max gain
-		return 0b1'0'1'1'0000'0'1111111;
-	}
-
 	BAN::ErrorOr<void> HDAudioFunctionGroup::enable_output_path(uint8_t index)
 	{
 		ASSERT(index < m_output_paths.size());
@@ -310,7 +342,6 @@ namespace Kernel
 		}
 
 		const auto format = get_format_data();
-		const auto volume = get_volume_data();
 
 		for (size_t i = 0; i < path.size(); i++)
 		{
@@ -339,13 +370,17 @@ namespace Kernel
 				}));
 			}
 
-			// set volume
-			TRY(m_controller->send_command({
-				.data = static_cast<uint8_t>(volume & 0xFF),
-				.command = static_cast<uint16_t>(0x300 | (volume >> 8)),
-				.node_index = path[i]->id,
-				.codec_address = m_cid,
-			}));
+			// set volume to 0 dB, no mute
+			if (path[i]->output_amplifier.has_value())
+			{
+				const uint32_t volume = 0b1'0'1'1'0000'0'0000000 | path[i]->output_amplifier->offset;
+				TRY(m_controller->send_command({
+					.data = static_cast<uint8_t>(volume & 0xFF),
+					.command = static_cast<uint16_t>(0x300 | (volume >> 8)),
+					.node_index = path[i]->id,
+					.codec_address = m_cid,
+				}));
+			}
 
 			switch (path[i]->type)
 			{
@@ -390,6 +425,41 @@ namespace Kernel
 			}
 		}
 
+		// update volume info to this path
+		m_volume_info.min_mdB = 0;
+		m_volume_info.max_mdB = 0;
+		m_volume_info.step_mdB = 0;
+		for (size_t i = 0; i < path.size(); i++)
+		{
+			if (!path[i]->output_amplifier.has_value())
+				continue;
+			const auto& amp = path[i]->output_amplifier.value();
+
+			const int32_t step_mdB = amp.step_size * 250;
+			m_volume_info.step_mdB = step_mdB;
+			m_volume_info.min_mdB = -amp.offset * step_mdB;
+			m_volume_info.max_mdB = (amp.num_steps - amp.offset) * step_mdB;
+			m_volume_info.mdB = BAN::Math::clamp(m_volume_info.mdB, m_volume_info.min_mdB, m_volume_info.max_mdB);
+
+			const int32_t step_round = (m_volume_info.mdB >= 0)
+				? +step_mdB / 2
+				: -step_mdB / 2;
+			const uint32_t step = (m_volume_info.mdB + step_round) / step_mdB + amp.offset;
+			const uint32_t volume = 0b1'0'1'1'0000'0'0000000 | step;
+
+			TRY(m_controller->send_command({
+				.data = static_cast<uint8_t>(volume & 0xFF),
+				.command = static_cast<uint16_t>(0x300 | (volume >> 8)),
+				.node_index = path[i]->id,
+				.codec_address = m_cid,
+			}));
+
+			break;
+		}
+
+		if (m_volume_info.min_mdB == 0 && m_volume_info.max_mdB == 0)
+			m_volume_info.mdB = 0;
+
 		return {};
 	}
 
@@ -442,10 +512,6 @@ namespace Kernel
 
 	BAN::ErrorOr<void> HDAudioFunctionGroup::recurse_output_paths(const HDAudio::AFGWidget& widget, BAN::Vector<const HDAudio::AFGWidget*>& path)
 	{
-		// cycle "detection"
-		if (path.size() >= m_max_path_length)
-			return {};
-
 		// we've reached a DAC
 		if (widget.type == HDAudio::AFGWidget::Type::OutputConverter)
 		{
@@ -462,9 +528,18 @@ namespace Kernel
 		{
 			if (!widget.connections.contains(connection.id))
 				continue;
+
+			// cycle detection
+			for (const auto* w : path)
+				if (w == &connection)
+					goto already_visited;
+
 			TRY(path.push_back(&connection));
 			TRY(recurse_output_paths(connection, path));
 			path.pop_back();
+
+		already_visited:
+			continue;
 		}
 
 		return {};
@@ -483,30 +558,18 @@ namespace Kernel
 
 		while ((m_bdl_head + 1) % m_bdl_entry_count != m_bdl_tail)
 		{
-			const size_t sample_data_tail = (m_sample_data_head + m_sample_data_capacity - m_sample_data_size) % m_sample_data_capacity;
-
-			const size_t sample_frames = BAN::Math::min(m_sample_data_size / get_channels() / sizeof(uint16_t), m_bdl_entry_sample_frames);
+			const size_t sample_frames = BAN::Math::min(m_sample_data->size() / get_channels() / sizeof(uint16_t), m_bdl_entry_sample_frames);
 			if (sample_frames == 0)
 				break;
 
 			const size_t copy_total_bytes = sample_frames * get_channels() * sizeof(uint16_t);
-			const size_t copy_before_wrap = BAN::Math::min(copy_total_bytes, m_sample_data_capacity - sample_data_tail);
 
 			memcpy(
 				reinterpret_cast<void*>(m_bdl_region->vaddr() + m_bdl_head * bdl_entry_bytes),
-				&m_sample_data[sample_data_tail],
-				copy_before_wrap
+				m_sample_data->get_data().data(),
+				copy_total_bytes
 			);
 
-			if (copy_before_wrap < copy_total_bytes)
-			{
-				memcpy(
-					reinterpret_cast<void*>(m_bdl_region->vaddr() + m_bdl_head * bdl_entry_bytes + copy_before_wrap),
-					&m_sample_data[0],
-					copy_total_bytes - copy_before_wrap
-				);
-			}
-
 			if (copy_total_bytes < bdl_entry_bytes)
 			{
 				memset(
@@ -516,8 +579,7 @@ namespace Kernel
 				);
 			}
 
-			m_sample_data_size -= copy_total_bytes;
-
+			m_sample_data->pop(copy_total_bytes);
 			m_bdl_head = (m_bdl_head + 1) % m_bdl_entry_count;
 		}
 

kernel/kernel/Audio/HDAudio/Controller.cpp

@@ -65,8 +65,12 @@ namespace Kernel
 		m_pci_device.enable_interrupt(0, *this);
 		m_bar0->write32(Regs::INTCTL, UINT32_MAX);
 
-		for (uint8_t codec_id = 0; codec_id < 0x10; codec_id++)
+		const uint16_t state_sts = m_bar0->read16(Regs::STATESTS);
+		for (uint8_t codec_id = 0; codec_id < 15; codec_id++)
 		{
+			if (!(state_sts & (1 << codec_id)))
+				continue;
+
 			auto codec_or_error = initialize_codec(codec_id);
 			if (codec_or_error.is_error())
 				continue;
@@ -307,8 +311,26 @@ namespace Kernel
 		if (result.type == AFGWidget::Type::PinComplex)
 		{
 			const uint32_t cap = send_command_or_zero(0xF00, 0x0C);
-			result.pin_complex.output = !!(cap & (1 << 4));
-			result.pin_complex.input  = !!(cap & (1 << 5));
+			result.pin_complex = {
+				.input   = !!(cap & (1 << 5)),
+				.output  = !!(cap & (1 << 4)),
+				.display = !!(cap & ((1 << 7) | (1 << 24))),
+				.config  = send_command_or_zero(0xF1C, 0x00),
+			};
+		}
+
+		if (const uint32_t out_amp_cap = send_command_or_zero(0xF00, 0x12))
+		{
+			const uint8_t offset    = (out_amp_cap >>  0) & 0x7F;
+			const uint8_t num_steps = (out_amp_cap >>  8) & 0x7F;
+			const uint8_t step_size = (out_amp_cap >> 16) & 0x7F;
+			const bool    mute      = (out_amp_cap >> 31);
+			result.output_amplifier = HDAudio::AFGWidget::Amplifier {
+				.offset = offset,
+				.num_steps = num_steps,
+				.step_size = step_size,
+				.mute = mute,
+			};
 		}
 
 		const uint8_t connection_info = send_command_or_zero(0xF00, 0x0E);

kernel/kernel/Debug.cpp

@@ -7,6 +7,8 @@
 #include <kernel/Terminal/TTY.h>
 #include <kernel/Timer/Timer.h>
 
+#include <BAN/ScopeGuard.h>
+
 #include <LibDEFLATE/Compressor.h>
 #include <LibQR/QRCode.h>
 
@@ -14,6 +16,13 @@
 
 bool g_disable_debug = false;
 
+extern "C" bool safe_user_memcpy(void*, const void*, size_t);
+
+namespace Kernel
+{
+	extern bool g_safe_user_alloc_nonexisting;
+}
+
 namespace Debug
 {
 
@@ -45,48 +54,25 @@ namespace Debug
 
 		SpinLockGuard _(s_debug_lock);
 
-		const stackframe* frame = reinterpret_cast<const stackframe*>(bp);
-
-		void* first_ip = frame->ip;
-		void* last_ip = 0;
-		bool first = true;
+		const bool temp = g_safe_user_alloc_nonexisting;
+		g_safe_user_alloc_nonexisting = false;
+		BAN::ScopeGuard alloc_restore([temp] { g_safe_user_alloc_nonexisting = temp; });
 
 		BAN::Formatter::print(Debug::putchar, "\e[36mStack trace:\r\n");
 
 		if (ip != 0)
 			BAN::Formatter::print(Debug::putchar, "    {}\r\n", reinterpret_cast<void*>(ip));
 
-		while (frame)
+		stackframe frame;
+		if (!safe_user_memcpy(&frame, reinterpret_cast<void*>(bp), sizeof(stackframe)))
+			;
+		else for (size_t depth = 0; depth < 64; depth++)
 		{
-			if (!PageTable::is_valid_pointer((vaddr_t)frame))
-			{
-				derrorln("invalid pointer {H}", (vaddr_t)frame);
+			BAN::Formatter::print(Debug::putchar, "    {}\r\n", reinterpret_cast<void*>(frame.ip));
+			if (frame.bp == nullptr || !safe_user_memcpy(&frame, frame.bp, sizeof(stackframe)))
 				break;
-			}
-
-			if (PageTable::current().is_page_free((vaddr_t)frame & PAGE_ADDR_MASK))
-			{
-				derrorln("    {} not mapped", frame);
-				break;
-			}
-
-			BAN::Formatter::print(Debug::putchar, "    {}\r\n", (void*)frame->ip);
-
-			if (!first && frame->ip == first_ip)
-			{
-				derrorln("looping kernel panic :(");
-				break;
-			}
-			else if (!first && frame->ip == last_ip)
-			{
-				derrorln("repeating stack trace");
-				break;
-			}
-
-			last_ip = frame->ip;
-			frame = frame->bp;
-			first = false;
 		}
+
 		BAN::Formatter::print(Debug::putchar, "\e[m");
 	}
 

kernel/kernel/Device/FramebufferDevice.cpp

@@ -6,6 +6,7 @@
 #include <kernel/Terminal/FramebufferTerminal.h>
 
 #include <sys/framebuffer.h>
+#include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <sys/sysmacros.h>
 
@@ -83,10 +84,10 @@ namespace Kernel
 
 		m_video_buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET, UINTPTR_MAX,
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			BAN::Math::div_round_up<size_t>(m_width * m_height * (BANAN_FB_BPP / 8), PAGE_SIZE) * PAGE_SIZE,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, false
+			false
 		));
 
 		return {};
@@ -133,6 +134,26 @@ namespace Kernel
 		return bytes_to_copy;
 	}
 
+	BAN::ErrorOr<long> FramebufferDevice::ioctl_impl(int cmd, void* arg)
+	{
+		switch (cmd)
+		{
+			case FB_MSYNC_RECTANGLE:
+			{
+				auto& rectangle = *static_cast<fb_msync_region*>(arg);
+				sync_pixels_rectangle(
+					rectangle.min_x,
+					rectangle.min_y,
+					rectangle.max_x - rectangle.min_x,
+					rectangle.max_y - rectangle.min_y
+				);
+				return 0;
+			}
+		}
+
+		return CharacterDevice::ioctl(cmd, arg);
+	}
+
 	uint32_t FramebufferDevice::get_pixel(uint32_t x, uint32_t y) const
 	{
 		ASSERT(x < m_width && y < m_height);

kernel/kernel/Epoll.cpp

@@ -16,7 +16,7 @@ namespace Kernel
 
 	Epoll::~Epoll()
 	{
-		for (auto [inode, _] : m_listening_events)
+		for (auto& [inode, _] : m_listening_events)
 			inode->del_epoll(this);
 	}
 
@@ -44,7 +44,7 @@ namespace Kernel
 
 				if (!contains_inode)
 					TRY(inode->add_epoll(this));
-				it->value.add_fd(fd, event);
+				TRY(it->value.add_fd(fd, event));
 
 				SpinLockGuard _(m_ready_lock);
 				auto ready_it = m_ready_events.find(inode);
@@ -144,9 +144,8 @@ namespace Kernel
 
 					{
 						uint32_t listen_mask = EPOLLHUP | EPOLLERR;
-						for (size_t fd = 0; fd < listen.events.size(); fd++)
-							if (listen.has_fd(fd))
-								listen_mask |= listen.events[fd].events;
+						for (const auto& [_, events] : listen.events)
+							listen_mask |= events.events;
 						events &= listen_mask;
 					}
 
@@ -171,11 +170,10 @@ namespace Kernel
 
 #undef REMOVE_IT
 
-					for (size_t fd = 0; fd < listen.events.size() && event_count < event_span.size(); fd++)
+					for (auto& [_, listen_event] : listen.events)
 					{
-						if (!listen.has_fd(fd))
-							continue;
-						auto& listen_event = listen.events[fd];
+						if (event_count >= event_span.size())
+							break;
 
 						const auto new_events = (listen_event.events | EPOLLHUP | EPOLLERR) & events;
 						if (new_events == 0)

kernel/kernel/FS/USTARModule.cpp

@@ -1,94 +1,262 @@
 #include <BAN/ScopeGuard.h>
 #include <kernel/FS/USTARModule.h>
+#include <kernel/Timer/Timer.h>
+#include <LibDEFLATE/Decompressor.h>
 
 #include <tar.h>
 
 namespace Kernel
 {
 
-	bool is_ustar_boot_module(const BootModule& module)
+	class DataSource
 	{
-		if (module.start % PAGE_SIZE)
+	public:
+		DataSource() = default;
+		virtual ~DataSource() = default;
+
+		size_t data_size() const
 		{
-			dprintln("ignoring non-page-aligned module");
+			return m_data_size;
+		}
+
+		BAN::ConstByteSpan data()
+		{
+			return { m_data_buffer, m_data_size };
+		}
+
+		void pop_data(size_t size)
+		{
+			ASSERT(size <= m_data_size);
+			if (size > 0 && size < m_data_size)
+				memmove(m_data_buffer, m_data_buffer + size, m_data_size - size);
+			m_data_size -= size;
+			m_bytes_produced += size;
+		}
+
+		virtual BAN::ErrorOr<bool> produce_data() = 0;
+
+		uint64_t bytes_produced() const
+		{
+			return m_bytes_produced;
+		}
+
+		virtual uint64_t bytes_consumed() const = 0;
+
+	protected:
+		uint8_t m_data_buffer[4096];
+		size_t m_data_size { 0 };
+
+	private:
+		uint64_t m_bytes_produced { 0 };
+	};
+
+	class DataSourceRaw final : public DataSource
+	{
+	public:
+		DataSourceRaw(const BootModule& module)
+			: m_module(module)
+		{ }
+
+		BAN::ErrorOr<bool> produce_data() override
+		{
+			if (m_offset >= m_module.size || m_data_size >= sizeof(m_data_buffer))
+				return false;
+
+			while (m_offset < m_module.size && m_data_size < sizeof(m_data_buffer))
+			{
+				const size_t to_copy = BAN::Math::min(
+					sizeof(m_data_buffer) - m_data_size,
+					PAGE_SIZE - (m_offset % PAGE_SIZE)
+				);
+				PageTable::with_fast_page((m_module.start + m_offset) & PAGE_ADDR_MASK, [&] {
+					memcpy(m_data_buffer + m_data_size, PageTable::fast_page_as_ptr(m_offset % PAGE_SIZE), to_copy);
+				});
+				m_data_size += to_copy;
+				m_offset += to_copy;
+			}
+
+			return true;
+		}
+
+		uint64_t bytes_consumed() const override
+		{
+			return bytes_produced();
+		}
+
+	private:
+		const BootModule& m_module;
+		size_t m_offset { 0 };
+	};
+
+	class DataSourceGZip final : public DataSource
+	{
+	public:
+		DataSourceGZip(BAN::UniqPtr<DataSource>&& data_source)
+			: m_data_source(BAN::move(data_source))
+			, m_decompressor(LibDEFLATE::StreamType::GZip)
+		{ }
+
+		BAN::ErrorOr<bool> produce_data() override
+		{
+			if (m_is_done)
+				return false;
+
+			bool did_produce_data { false };
+			for (;;)
+			{
+				TRY(m_data_source->produce_data());
+
+				size_t input_consumed, output_produced;
+				const auto status = TRY(m_decompressor.decompress(
+					m_data_source->data(),
+					input_consumed,
+					{ m_data_buffer + m_data_size, sizeof(m_data_buffer) - m_data_size },
+					output_produced
+				));
+
+				m_data_source->pop_data(input_consumed);
+				m_data_size += output_produced;
+
+				if (output_produced)
+					did_produce_data = true;
+
+				switch (status)
+				{
+					using DecompStatus = LibDEFLATE::Decompressor::Status;
+					case DecompStatus::Done:
+						m_is_done = true;
+						return did_produce_data;
+					case DecompStatus::NeedMoreInput:
+						break;
+					case DecompStatus::NeedMoreOutput:
+						return did_produce_data;
+				}
+			}
+		}
+
+		uint64_t bytes_consumed() const override
+		{
+			return m_data_source->bytes_consumed();
+		}
+
+	private:
+		BAN::UniqPtr<DataSource> m_data_source;
+		LibDEFLATE::Decompressor m_decompressor;
+		bool m_is_done { false };
+	};
+
+	static BAN::ErrorOr<void> unpack_boot_module_into_directory(BAN::RefPtr<Inode> root_inode, DataSource& data_source);
+
+	BAN::ErrorOr<bool> unpack_boot_module_into_directory(BAN::RefPtr<Inode> root_inode, const BootModule& module)
+	{
+		ASSERT(root_inode->mode().ifdir());
+
+		BAN::UniqPtr<DataSource> data_source = TRY(BAN::UniqPtr<DataSourceRaw>::create(module));
+
+		bool is_compressed = false;
+
+		TRY(data_source->produce_data());
+		if (data_source->data_size() >= 2 && memcmp(&data_source->data()[0], "\x1F\x8B", 2) == 0)
+		{
+			data_source = TRY(BAN::UniqPtr<DataSourceGZip>::create(BAN::move(data_source)));
+			is_compressed = true;
+		}
+
+		TRY(data_source->produce_data());
+		if (data_source->data_size() < 512 || memcmp(&data_source->data()[257], "ustar", 5) != 0)
+		{
+			dwarnln("Unrecognized initrd format");
 			return false;
 		}
 
-		if (module.size < 512)
-			return false;
+		const auto module_size_kib = module.size / 1024;
+		dprintln("unpacking {}.{3} MiB{} initrd",
+			module_size_kib / 1024, (module_size_kib % 1024) * 1000 / 1024,
+			is_compressed ? " compressed" : ""
+		);
 
-		bool has_ustar_signature;
-		PageTable::with_fast_page(module.start, [&] {
-			has_ustar_signature = memcmp(PageTable::fast_page_as_ptr(257), "ustar", 5) == 0;
-		});
+		const auto unpack_ms1 = SystemTimer::get().ms_since_boot();
+		TRY(unpack_boot_module_into_directory(root_inode, *data_source));
+		const auto unpack_ms2 = SystemTimer::get().ms_since_boot();
 
-		return has_ustar_signature;
+		const auto duration_ms = unpack_ms2 - unpack_ms1;
+		dprintln("unpacking {}.{3} MiB{} initrd took {}.{3} s",
+			module_size_kib / 1024, (module_size_kib % 1024) * 1000 / 1024,
+			is_compressed ? " compressed" : "",
+			duration_ms / 1000, duration_ms % 1000
+		);
+
+		if (is_compressed)
+		{
+			const auto uncompressed_kib = data_source->bytes_produced() / 1024;
+			dprintln("uncompressed size {}.{3} MiB",
+				uncompressed_kib / 1024, (uncompressed_kib % 1024) * 1000 / 1024
+			);
+		}
+
+		return true;
 	}
 
-	BAN::ErrorOr<void> unpack_boot_module_into_filesystem(BAN::RefPtr<FileSystem> filesystem, const BootModule& module)
+	BAN::ErrorOr<void> unpack_boot_module_into_directory(BAN::RefPtr<Inode> root_inode, DataSource& data_source)
 	{
-		ASSERT(is_ustar_boot_module(module));
-
-		auto root_inode = filesystem->root_inode();
-
-		uint8_t* temp_page = static_cast<uint8_t*>(kmalloc(PAGE_SIZE));
-		if (temp_page == nullptr)
-			return BAN::Error::from_errno(ENOMEM);
-		BAN::ScopeGuard _([temp_page] { kfree(temp_page); });
-
 		BAN::String next_file_name;
 		BAN::String next_link_name;
 
-		size_t offset = 0;
-		while (offset + 512 <= module.size)
+		constexpr uint32_t print_interval_ms = 1000;
+		auto next_print_ms = SystemTimer::get().ms_since_boot() + print_interval_ms;
+
+		while (TRY(data_source.produce_data()), data_source.data_size() >= 512)
 		{
-			size_t file_size = 0;
-			mode_t file_mode = 0;
-			uid_t file_uid = 0;
-			gid_t file_gid = 0;
-			uint8_t file_type = 0;
-			char file_path[100 + 1 + 155 + 1] {};
-
-			PageTable::with_fast_page((module.start + offset) & PAGE_ADDR_MASK, [&] {
-				const size_t page_off = offset % PAGE_SIZE;
-
-				const auto parse_octal =
-					[page_off](size_t offset, size_t length) -> size_t
-					{
-						size_t result = 0;
-						for (size_t i = 0; i < length; i++)
-						{
-							const char ch = PageTable::fast_page_as<char>(page_off + offset + i);
-							if (ch == '\0')
-								break;
-							result = (result * 8) + (ch - '0');
-						}
-						return result;
-					};
-
-				if (memcmp(PageTable::fast_page_as_ptr(page_off + 257), "ustar", 5)) {
-					file_size = SIZE_MAX;
-					return;
+			if (SystemTimer::get().ms_since_boot() >= next_print_ms)
+			{
+				const auto kib_consumed = data_source.bytes_consumed() / 1024;
+				const auto kib_produced = data_source.bytes_produced() / 1024;
+				if (kib_consumed == kib_produced)
+				{
+					dprintln(" ... {}.{3} MiB",
+						kib_consumed / 1024, (kib_consumed % 1024) * 1000 / 1024
+					);
 				}
+				else
+				{
+					dprintln(" ... {}.{3} MiB ({}.{3} MiB)",
+						kib_consumed / 1024, (kib_consumed % 1024) * 1000 / 1024,
+						kib_produced / 1024, (kib_produced % 1024) * 1000 / 1024
+					);
+				}
+				next_print_ms = SystemTimer::get().ms_since_boot() + print_interval_ms;
+			}
 
-				memcpy(file_path, PageTable::fast_page_as_ptr(page_off + 345), 155);
-				const size_t prefix_len = strlen(file_path);
-				file_path[prefix_len] = '/';
-				memcpy(file_path + prefix_len + 1, PageTable::fast_page_as_ptr(page_off), 100);
+			const auto parse_octal =
+				[&data_source](size_t offset, size_t length) -> size_t
+				{
+					size_t result = 0;
+					for (size_t i = 0; i < length; i++)
+					{
+						const char ch = data_source.data()[offset + i];
+						if (ch == '\0')
+							break;
+						result = (result * 8) + (ch - '0');
+					}
+					return result;
+				};
 
-				file_mode = parse_octal(100, 8);
-				file_uid  = parse_octal(108, 8);
-				file_gid  = parse_octal(116, 8);
-				file_size = parse_octal(124, 12);
-				file_type = PageTable::fast_page_as<char>(page_off + 156);
-			});
-
-			if (file_size == SIZE_MAX)
-				break;
-			if (offset + 512 + file_size > module.size)
+			if (memcmp(&data_source.data()[257], "ustar", 5) != 0)
 				break;
 
-			auto parent_inode = filesystem->root_inode();
+			char file_path[100 + 1 + 155 + 1];
+			memcpy(file_path, &data_source.data()[345], 155);
+			const size_t prefix_len = strlen(file_path);
+			file_path[prefix_len] = '/';
+			memcpy(file_path + prefix_len + 1, &data_source.data()[0], 100);
+
+			mode_t        file_mode = parse_octal(100, 8);
+			const uid_t   file_uid  = parse_octal(108, 8);
+			const gid_t   file_gid  = parse_octal(116, 8);
+			const size_t  file_size = parse_octal(124, 12);
+			const uint8_t file_type = data_source.data()[156];
+
+			auto parent_inode = root_inode;
 
 			auto file_path_parts = TRY(BAN::StringView(next_file_name.empty() ? file_path : next_file_name.sv()).split('/'));
 			for (size_t i = 0; i < file_path_parts.size() - 1; i++)
@@ -111,27 +279,33 @@ namespace Kernel
 
 			auto file_name_sv = file_path_parts.back();
 
+			bool did_consume_data = false;
+
 			if (file_type == 'L' || file_type == 'K')
 			{
-				auto& target = (file_type == 'L') ? next_file_name : next_link_name;
-				TRY(target.resize(file_size));
+				auto& target_str = (file_type == 'L') ? next_file_name : next_link_name;
+				TRY(target_str.resize(file_size));
+
+				data_source.pop_data(512);
 
 				size_t nwritten = 0;
 				while (nwritten < file_size)
 				{
-					const paddr_t paddr = module.start + offset + 512 + nwritten;
-					PageTable::with_fast_page(paddr & PAGE_ADDR_MASK, [&] {
-						memcpy(temp_page, PageTable::fast_page_as_ptr(), PAGE_SIZE);
-					});
+					TRY(data_source.produce_data());
+					if (data_source.data_size() == 0)
+						return {};
 
-					const size_t page_off = paddr % PAGE_SIZE;
-					const size_t to_write = BAN::Math::min(file_size - nwritten, PAGE_SIZE - page_off);
-					memcpy(target.data() + nwritten, temp_page + page_off, to_write);
-					nwritten += to_write;
+					const size_t to_copy = BAN::Math::min(data_source.data_size(), file_size - nwritten);
+					memcpy(target_str.data() + nwritten, data_source.data().data(), to_copy);
+					nwritten += to_copy;
+
+					data_source.pop_data(to_copy);
 				}
 
-				while (!target.empty() && target.back() == '\0')
-					target.pop_back();
+				did_consume_data = true;
+
+				while (!target_str.empty() && target_str.back() == '\0')
+					target_str.pop_back();
 			}
 			else if (file_type == DIRTYPE)
 			{
@@ -149,14 +323,11 @@ namespace Kernel
 					link_name = next_link_name.sv();
 				else
 				{
-					const paddr_t paddr = module.start + offset;
-					PageTable::with_fast_page(paddr & PAGE_ADDR_MASK, [&] {
-						memcpy(link_buffer, PageTable::fast_page_as_ptr((paddr % PAGE_SIZE) + 157), 100);
-					});
+					memcpy(link_buffer, &data_source.data()[157], 100);
 					link_name = link_buffer;
 				}
 
-				auto target_inode = filesystem->root_inode();
+				auto target_inode = root_inode;
 
 				auto link_path_parts = TRY(link_name.split('/'));
 				for (const auto part : link_path_parts)
@@ -188,10 +359,7 @@ namespace Kernel
 						link_name = next_link_name.sv();
 					else
 					{
-						const paddr_t paddr = module.start + offset;
-						PageTable::with_fast_page(paddr & PAGE_ADDR_MASK, [&] {
-							memcpy(link_buffer, PageTable::fast_page_as_ptr((paddr % PAGE_SIZE) + 157), 100);
-						});
+						memcpy(link_buffer, &data_source.data()[157], 100);
 						link_name = link_buffer;
 					}
 
@@ -203,26 +371,26 @@ namespace Kernel
 			{
 				if (auto ret = parent_inode->create_file(file_name_sv, file_mode, file_uid, file_gid); ret.is_error())
 					dwarnln("failed to create file '{}': {}", file_name_sv, ret.error());
-				else
+				else if (file_size)
 				{
-					if (file_size)
+					auto inode = TRY(parent_inode->find_inode(file_name_sv));
+
+					data_source.pop_data(512);
+
+					size_t nwritten = 0;
+					while (nwritten < file_size)
 					{
-						auto inode = TRY(parent_inode->find_inode(file_name_sv));
+						TRY(data_source.produce_data());
+						ASSERT(data_source.data_size() > 0); // what to do?
 
-						size_t nwritten = 0;
-						while (nwritten < file_size)
-						{
-							const paddr_t paddr = module.start + offset + 512 + nwritten;
-							PageTable::with_fast_page(paddr & PAGE_ADDR_MASK, [&] {
-								memcpy(temp_page, PageTable::fast_page_as_ptr(), PAGE_SIZE);
-							});
+						const size_t to_write = BAN::Math::min(file_size - nwritten, data_source.data_size());
+						TRY(inode->write(nwritten, data_source.data().slice(0, to_write)));
+						nwritten += to_write;
 
-							const size_t page_off = paddr % PAGE_SIZE;
-							const size_t to_write = BAN::Math::min(file_size - nwritten, PAGE_SIZE - page_off);
-							TRY(inode->write(nwritten, { temp_page + page_off, to_write }));
-							nwritten += to_write;
-						}
+						data_source.pop_data(to_write);
 					}
+
+					did_consume_data = true;
 				}
 			}
 
@@ -232,9 +400,27 @@ namespace Kernel
 				next_link_name.clear();
 			}
 
-			offset += 512 + file_size;
-			if (auto rem = offset % 512)
-				offset += 512 - rem;
+			if (!did_consume_data)
+			{
+				data_source.pop_data(512);
+
+				size_t consumed = 0;
+				while (consumed < file_size)
+				{
+					TRY(data_source.produce_data());
+					if (data_source.data_size() == 0)
+						return {};
+					data_source.pop_data(BAN::Math::min(file_size - consumed, data_source.data_size()));
+				}
+			}
+
+			if (const auto rem = file_size % 512)
+			{
+				TRY(data_source.produce_data());
+				if (data_source.data_size() < rem)
+					return {};
+				data_source.pop_data(512 - rem);
+			}
 		}
 
 		return {};

kernel/kernel/FS/VirtualFileSystem.cpp

@@ -61,18 +61,22 @@ namespace Kernel
 		if (filesystem_or_error.is_error())
 			panic("Failed to create fallback filesystem: {}", filesystem_or_error.error());
 
-		dprintln("Loading fallback filesystem from {} modules", g_boot_info.modules.size());
+		dprintln("Trying to load fallback filesystem from {} modules", g_boot_info.modules.size());
 
 		auto filesystem = BAN::RefPtr<FileSystem>::adopt(filesystem_or_error.release_value());
 
+		bool loaded_initrd = false;
 		for (const auto& module : g_boot_info.modules)
 		{
-			if (!is_ustar_boot_module(module))
-				continue;
-			if (auto ret = unpack_boot_module_into_filesystem(filesystem, module); ret.is_error())
+			if (auto ret = unpack_boot_module_into_directory(filesystem->root_inode(), module); ret.is_error())
 				dwarnln("Failed to unpack boot module: {}", ret.error());
+			else
+				loaded_initrd |= ret.value();
 		}
 
+		if (!loaded_initrd)
+			panic("Could not load initrd from any boot module :(");
+
 		return filesystem;
 	}
 

kernel/kernel/GDT.cpp

@@ -32,6 +32,8 @@ namespace Kernel
 		gdt->write_entry(0x38, 0x00000000, 0x00000, 0xF2, 0xC); // gsbase
 #endif
 
+		gdt->write_entry(m_cpu_index_offset, 0, 0, 0, 0);
+
 		gdt->write_tss();
 
 		return gdt;

kernel/kernel/IDT.cpp

@@ -164,6 +164,35 @@ namespace Kernel
 		"Unkown Exception 0x1F",
 	};
 
+	extern "C" uint8_t safe_user_memcpy[];
+	extern "C" uint8_t safe_user_memcpy_end[];
+	extern "C" uint8_t safe_user_memcpy_fault[];
+
+	extern "C" uint8_t safe_user_strncpy[];
+	extern "C" uint8_t safe_user_strncpy_end[];
+	extern "C" uint8_t safe_user_strncpy_fault[];
+
+	struct safe_user_page_fault
+	{
+		const uint8_t* ip_start;
+		const uint8_t* ip_end;
+		const uint8_t* ip_fault;
+	};
+	static constexpr safe_user_page_fault s_safe_user_page_faults[] {
+		{
+			.ip_start = safe_user_memcpy,
+			.ip_end   = safe_user_memcpy_end,
+			.ip_fault = safe_user_memcpy_fault,
+		},
+		{
+			.ip_start = safe_user_strncpy,
+			.ip_end   = safe_user_strncpy_end,
+			.ip_fault = safe_user_strncpy_fault,
+		},
+	};
+
+	bool g_safe_user_alloc_nonexisting { true };
+
 	extern "C" void cpp_isr_handler(uint32_t isr, uint32_t error, InterruptStack* interrupt_stack, const Registers* regs)
 	{
 		if (g_paniced)
@@ -187,6 +216,16 @@ namespace Kernel
 				PageFaultError page_fault_error;
 				page_fault_error.raw = error;
 
+				const uint8_t* ip = reinterpret_cast<const uint8_t*>(interrupt_stack->ip);
+
+				if (!g_safe_user_alloc_nonexisting) for (const auto& safe_user : s_safe_user_page_faults)
+				{
+					if (ip < safe_user.ip_start || ip >= safe_user.ip_end)
+						continue;
+					interrupt_stack->ip = reinterpret_cast<vaddr_t>(safe_user.ip_fault);
+					return;
+				}
+
 				Processor::set_interrupt_state(InterruptState::Enabled);
 				auto result = Process::current().allocate_page_for_demand_paging(regs->cr2, page_fault_error.write, page_fault_error.instruction);
 				Processor::set_interrupt_state(InterruptState::Disabled);
@@ -194,13 +233,27 @@ namespace Kernel
 				if (result.is_error())
 				{
 					dwarnln("Demand paging: {}", result.error());
+
+					// TODO: this is too strict, we should maybe do SIGBUS and
+					//       SIGKILL only on recursive exceptions
+					Processor::set_interrupt_state(InterruptState::Enabled);
 					Thread::current().handle_signal(SIGKILL, {});
+					Processor::set_interrupt_state(InterruptState::Disabled);
+
 					return;
 				}
 
 				if (result.value())
 					return;
 
+				if (g_safe_user_alloc_nonexisting) for (const auto& safe_user : s_safe_user_page_faults)
+				{
+					if (ip < safe_user.ip_start || ip >= safe_user.ip_end)
+						continue;
+					interrupt_stack->ip = reinterpret_cast<vaddr_t>(safe_user.ip_fault);
+					return;
+				}
+
 				break;
 			}
 			case ISR::DeviceNotAvailable:
@@ -240,7 +293,7 @@ namespace Kernel
 
 		const char* process_name = (tid && Thread::current().has_process())
 			? Process::current().name()
-			: nullptr;
+			: "";
 
 #if ARCH(x86_64)
 		dwarnln(
@@ -285,7 +338,7 @@ namespace Kernel
 
 		Debug::s_debug_lock.unlock(InterruptState::Disabled);
 
-		if (tid && Thread::current().is_userspace())
+		if (tid && GDT::is_user_segment(interrupt_stack->cs))
 		{
 			// TODO: Confirm and fix the exception to signal mappings
 
@@ -316,6 +369,7 @@ namespace Kernel
 			case ISR::InvalidOpcode:
 				signal_info.si_signo = SIGILL;
 				signal_info.si_code = ILL_ILLOPC;
+				signal_info.si_addr = reinterpret_cast<void*>(interrupt_stack->ip);
 				break;
 			case ISR::PageFault:
 				signal_info.si_signo = SIGSEGV;
@@ -323,6 +377,7 @@ namespace Kernel
 					signal_info.si_code = SEGV_ACCERR;
 				else
 					signal_info.si_code = SEGV_MAPERR;
+				signal_info.si_addr = reinterpret_cast<void*>(regs->cr2);
 				break;
 			default:
 				dwarnln("Unhandled exception");
@@ -330,7 +385,9 @@ namespace Kernel
 				break;
 			}
 
+			Processor::set_interrupt_state(InterruptState::Enabled);
 			Thread::current().handle_signal(signal_info.si_signo, signal_info);
+			Processor::set_interrupt_state(InterruptState::Disabled);
 		}
 		else
 		{
@@ -365,10 +422,6 @@ namespace Kernel
 			Process::update_alarm_queue();
 
 		Processor::scheduler().timer_interrupt();
-
-		auto& current_thread = Thread::current();
-		if (current_thread.can_add_signal_to_execute())
-			current_thread.handle_signal();
 	}
 
 	extern "C" void cpp_irq_handler(uint32_t irq)
@@ -392,15 +445,18 @@ namespace Kernel
 		else
 			dprintln("no handler for irq 0x{2H}", irq);
 
-		auto& current_thread = Thread::current();
-		if (current_thread.can_add_signal_to_execute())
-			current_thread.handle_signal();
-
 		Processor::scheduler().reschedule_if_idle();
 
 		ASSERT(Thread::current().state() != Thread::State::Terminated);
 	}
 
+	extern "C" void cpp_check_signal()
+	{
+		Processor::set_interrupt_state(InterruptState::Enabled);
+		Thread::current().handle_signal_if_interrupted();
+		Processor::set_interrupt_state(InterruptState::Disabled);
+	}
+
 	void IDT::register_interrupt_handler(uint8_t index, void (*handler)(), uint8_t ist)
 	{
 		auto& desc = m_idt[index];
@@ -440,7 +496,6 @@ namespace Kernel
 	IRQ_LIST_X
 #undef X
 
-	extern "C" void asm_yield_handler();
 	extern "C" void asm_ipi_handler();
 	extern "C" void asm_timer_handler();
 #if ARCH(i686)

kernel/kernel/Input/PS2/Keymap.cpp

@@ -136,6 +136,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_extended[0x49] = keycode_function(18);
 		m_scancode_to_keycode_extended[0x51] = keycode_function(19);
 		m_scancode_to_keycode_normal[0x46]   = keycode_function(20);
+		m_scancode_to_keycode_extended[0x20] = keycode_function(21);
+		m_scancode_to_keycode_extended[0x2E] = keycode_function(22);
+		m_scancode_to_keycode_extended[0x30] = keycode_function(23);
 
 		// Arrow keys
 		m_scancode_to_keycode_extended[0x48] = keycode_normal(5, 0);
@@ -246,6 +249,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_extended[0x7D] = keycode_function(18);
 		m_scancode_to_keycode_extended[0x7A] = keycode_function(19);
 		m_scancode_to_keycode_normal[0x7E]   = keycode_function(20);
+		m_scancode_to_keycode_extended[0x23] = keycode_function(21);
+		m_scancode_to_keycode_extended[0x21] = keycode_function(22);
+		m_scancode_to_keycode_extended[0x32] = keycode_function(23);
 
 		// Arrow keys
 		m_scancode_to_keycode_extended[0x75] = keycode_normal(5, 0);
@@ -356,6 +362,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_normal[0x6F] = keycode_function(18);
 		m_scancode_to_keycode_normal[0x6D] = keycode_function(19);
 		m_scancode_to_keycode_normal[0x5F] = keycode_function(20);
+		m_scancode_to_keycode_normal[0x9C] = keycode_function(21);
+		m_scancode_to_keycode_normal[0x9D] = keycode_function(22);
+		m_scancode_to_keycode_normal[0x95] = keycode_function(23);
 
 		// Arrow keys
 		m_scancode_to_keycode_normal[0x63] = keycode_normal(5, 0);

kernel/kernel/Memory/DMARegion.cpp

@@ -5,7 +5,7 @@
 namespace Kernel
 {
 
-	BAN::ErrorOr<BAN::UniqPtr<DMARegion>> DMARegion::create(size_t size)
+	BAN::ErrorOr<BAN::UniqPtr<DMARegion>> DMARegion::create(size_t size, PageTable::MemoryType type)
 	{
 		size_t needed_pages = BAN::Math::div_round_up<size_t>(size, PAGE_SIZE);
 
@@ -26,7 +26,7 @@ namespace Kernel
 		vaddr_guard.disable();
 		paddr_guard.disable();
 
-		PageTable::kernel().map_range_at(paddr, vaddr, size, PageTable::Flags::ReadWrite | PageTable::Flags::Present, PageTable::MemoryType::Uncached);
+		PageTable::kernel().map_range_at(paddr, vaddr, size, PageTable::Flags::ReadWrite | PageTable::Flags::Present, type);
 
 		return BAN::UniqPtr<DMARegion>::adopt(region_ptr);
 	}

kernel/kernel/Memory/FileBackedRegion.cpp

@@ -67,8 +67,12 @@ namespace Kernel
 		ASSERT(success);
 
 		for (size_t i = 0; i < pages.size(); i++)
-			if (pages[i])
-				sync(i);
+		{
+			if (pages[i] == 0)
+				continue;
+			sync(i);
+			Heap::get().release_page(pages[i]);
+		}
 
 		mutex.unlock();
 	}

kernel/kernel/Memory/Heap.cpp

@@ -2,11 +2,62 @@
 #include <kernel/Memory/Heap.h>
 #include <kernel/Memory/PageTable.h>
 
+#include <BAN/Sort.h>
+
 extern uint8_t g_kernel_end[];
 
 namespace Kernel
 {
 
+	struct ReservedRegion
+	{
+		paddr_t paddr;
+		uint64_t size;
+	};
+
+	static BAN::Vector<ReservedRegion> get_reserved_regions()
+	{
+		BAN::Vector<ReservedRegion> reserved_regions;
+		MUST(reserved_regions.reserve(2 + g_boot_info.modules.size()));
+		MUST(reserved_regions.emplace_back(0, 0x100000));
+		MUST(reserved_regions.emplace_back(g_boot_info.kernel_paddr, reinterpret_cast<size_t>(g_kernel_end - KERNEL_OFFSET)));
+		for (const auto& module : g_boot_info.modules)
+			MUST(reserved_regions.emplace_back(module.start, module.size));
+
+		// page align regions
+		for (auto& region : reserved_regions)
+		{
+			const auto rem = region.paddr % PAGE_SIZE;
+			region.paddr -= rem;
+			region.size += rem;
+
+			if (const auto rem = region.size % PAGE_SIZE)
+				region.size += PAGE_SIZE - rem;
+		}
+
+		// sort regions
+		BAN::sort::sort(reserved_regions.begin(), reserved_regions.end(),
+			[](const auto& lhs, const auto& rhs) -> bool {
+				if (lhs.paddr == rhs.paddr)
+					return lhs.size < rhs.size;
+				return lhs.paddr < rhs.paddr;
+			}
+		);
+
+		// combine overlapping regions
+		for (size_t i = 1; i < reserved_regions.size(); i++)
+		{
+			auto& prev = reserved_regions[i - 1];
+			auto& curr = reserved_regions[i - 0];
+			if (prev.paddr > curr.paddr + curr.size || curr.paddr > prev.paddr + prev.size)
+				continue;
+			prev.size = BAN::Math::max(prev.size, curr.paddr + curr.size - prev.paddr);
+			reserved_regions.remove(i--);
+		}
+
+		return reserved_regions;
+	}
+
 	static Heap* s_instance = nullptr;
 
 	void Heap::initialize()
@@ -28,6 +79,7 @@ namespace Kernel
 		if (g_boot_info.memory_map_entries.empty())
 			panic("Bootloader did not provide a memory map");
 
+		auto reserved_regions = get_reserved_regions();
 		for (const auto& entry : g_boot_info.memory_map_entries)
 		{
 			const char* entry_type_string = nullptr;
@@ -58,33 +110,71 @@ namespace Kernel
 			if (entry.type != MemoryMapEntry::Type::Available)
 				continue;
 
-			// FIXME: only reserve kernel area and modules, not everything from 0 -> kernel end
-			paddr_t start = entry.address;
-			if (start < (vaddr_t)g_kernel_end - KERNEL_OFFSET + g_boot_info.kernel_paddr)
-				start = (vaddr_t)g_kernel_end - KERNEL_OFFSET + g_boot_info.kernel_paddr;
-			for (const auto& module : g_boot_info.modules)
-				if (start < module.start + module.size)
-					start = module.start + module.size;
+			paddr_t e_start = entry.address;
+			if (auto rem = e_start % PAGE_SIZE)
+				e_start = PAGE_SIZE - rem;
+
+			paddr_t e_end = entry.address + entry.length;
+			if (auto rem = e_end % PAGE_SIZE)
+				e_end -= rem;
+
+			for (const auto& reserved_region : reserved_regions)
+			{
+				const paddr_t r_start = reserved_region.paddr;
+				const paddr_t r_end   = reserved_region.paddr + reserved_region.size;
+				if (r_end < e_start)
+					continue;
+				if (r_start > e_end)
+					break;
+
+				const paddr_t end = BAN::Math::max(e_start, r_start);
+				if (e_start + 2 * PAGE_SIZE <= end)
+					MUST(m_physical_ranges.emplace_back(e_start, end - e_start));
+
+				e_start = BAN::Math::max(e_start, BAN::Math::min(e_end, r_end));
+			}
+
+			if (e_start + 2 * PAGE_SIZE <= e_end)
+				MUST(m_physical_ranges.emplace_back(e_start, e_end - e_start));
+		}
+
+		uint64_t total_kibi_bytes = 0;
+		for (auto& range : m_physical_ranges)
+		{
+			const uint64_t kibi_bytes = range.usable_memory() / 1024;
+			dprintln("RAM {8H}->{8H} ({}.{3} MiB)", range.start(), range.end(), kibi_bytes / 1024, kibi_bytes % 1024);
+			total_kibi_bytes += kibi_bytes;
+		}
+		dprintln("Total RAM {}.{3} MiB", total_kibi_bytes / 1024, total_kibi_bytes % 1024);
+	}
+
+	void Heap::release_boot_modules()
+	{
+		const auto modules = BAN::move(g_boot_info.modules);
+
+		uint64_t kibi_bytes = 0;
+		for (const auto& module : modules)
+		{
+			vaddr_t start = module.start;
 			if (auto rem = start % PAGE_SIZE)
 				start += PAGE_SIZE - rem;
 
-			paddr_t end = entry.address + entry.length;
+			vaddr_t end = module.start + module.size;
 			if (auto rem = end % PAGE_SIZE)
 				end -= rem;
 
-			// Physical pages needs atleast 2 pages
-			if (end > start + PAGE_SIZE)
-				MUST(m_physical_ranges.emplace_back(start, end - start));
+			const size_t size = end - start;
+			if (size < 2 * PAGE_SIZE)
+				continue;
+
+			SpinLockGuard _(m_lock);
+			MUST(m_physical_ranges.emplace_back(start, size));
+
+			kibi_bytes += m_physical_ranges.back().usable_memory() / 1024;
 		}
 
-		size_t total = 0;
-		for (auto& range : m_physical_ranges)
-		{
-			size_t bytes = range.usable_memory();
-			dprintln("RAM {8H}->{8H} ({}.{} MB)", range.start(), range.end(), bytes / (1 << 20), bytes % (1 << 20) * 1000 / (1 << 20));
-			total += bytes;
-		}
-		dprintln("Total RAM {}.{} MB", total / (1 << 20), total % (1 << 20) * 1000 / (1 << 20));
+		if (kibi_bytes)
+			dprintln("Released {}.{3} MiB of RAM from boot modules", kibi_bytes / 1024, kibi_bytes % 1024);
 	}
 
 	paddr_t Heap::take_free_page()

kernel/kernel/Memory/MemoryBackedRegion.cpp

@@ -1,5 +1,6 @@
 #include <kernel/Memory/Heap.h>
 #include <kernel/Memory/MemoryBackedRegion.h>
+#include <kernel/Lock/LockGuard.h>
 
 namespace Kernel
 {
@@ -14,6 +15,9 @@ namespace Kernel
 			return BAN::Error::from_errno(ENOMEM);
 		auto region = BAN::UniqPtr<MemoryBackedRegion>::adopt(region_ptr);
 
+		const size_t page_count = (size + PAGE_SIZE - 1) / PAGE_SIZE;
+		TRY(region->m_physical_pages.resize(page_count, nullptr));
+
 		TRY(region->initialize(address_range));
 
 		return region;
@@ -28,38 +32,75 @@ namespace Kernel
 	{
 		ASSERT(m_type == Type::PRIVATE);
 
-		size_t needed_pages = BAN::Math::div_round_up<size_t>(m_size, PAGE_SIZE);
-		for (size_t i = 0; i < needed_pages; i++)
-		{
-			paddr_t paddr = m_page_table.physical_address_of(m_vaddr + i * PAGE_SIZE);
-			if (paddr != 0)
-				Heap::get().release_page(paddr);
-		}
+		for (auto* page : m_physical_pages)
+			if (page && --page->ref_count == 0)
+				delete page;
+	}
+
+	MemoryBackedRegion::PhysicalPage::~PhysicalPage()
+	{
+		Heap::get().release_page(paddr);
 	}
 
 	BAN::ErrorOr<bool> MemoryBackedRegion::allocate_page_containing_impl(vaddr_t address, bool wants_write)
 	{
 		ASSERT(m_type == Type::PRIVATE);
-
 		ASSERT(contains(address));
-		(void)wants_write;
 
-		// Check if address is already mapped
-		vaddr_t vaddr = address & PAGE_ADDR_MASK;
-		if (m_page_table.physical_address_of(vaddr) != 0)
-			return false;
+		const vaddr_t vaddr = address & PAGE_ADDR_MASK;
 
-		// Map new physcial page to address
-		paddr_t paddr = Heap::get().take_free_page();
+		LockGuard _(m_mutex);
+
+		auto& physical_page = m_physical_pages[(vaddr - m_vaddr) / PAGE_SIZE];
+
+		if (physical_page == nullptr)
+		{
+			const paddr_t paddr = Heap::get().take_free_page();
+			if (paddr == 0)
+				return BAN::Error::from_errno(ENOMEM);
+
+			physical_page = new PhysicalPage(paddr);
+			if (physical_page == nullptr)
+				return BAN::Error::from_errno(ENOMEM);
+
+			m_page_table.map_page_at(paddr, vaddr, m_flags);
+			PageTable::with_fast_page(paddr, [] {
+				memset(PageTable::fast_page_as_ptr(), 0x00, PAGE_SIZE);
+			});
+
+			return true;
+		}
+
+		if (auto is_only_ref = (physical_page->ref_count == 1); is_only_ref || !wants_write)
+		{
+			auto flags = m_flags;
+			if (!is_only_ref)
+				flags &= ~PageTable::ReadWrite;
+
+			m_page_table.map_page_at(physical_page->paddr, vaddr, flags);
+
+			return true;
+		}
+
+		const paddr_t paddr = Heap::get().take_free_page();
 		if (paddr == 0)
 			return BAN::Error::from_errno(ENOMEM);
+
+		auto* new_physical_page = new PhysicalPage(paddr);
+		if (new_physical_page == nullptr)
+			return BAN::Error::from_errno(ENOMEM);
+
 		m_page_table.map_page_at(paddr, vaddr, m_flags);
 
-		// Zero out the new page
-		PageTable::with_fast_page(paddr, [&] {
-			memset(PageTable::fast_page_as_ptr(), 0x00, PAGE_SIZE);
+		ASSERT(&m_page_table == &PageTable::current());
+		PageTable::with_fast_page(physical_page->paddr, [vaddr] {
+			memcpy(reinterpret_cast<void*>(vaddr), PageTable::fast_page_as_ptr(), PAGE_SIZE);
 		});
 
+		if (--physical_page->ref_count == 0)
+			delete physical_page;
+		physical_page = new_physical_page;
+
 		return true;
 	}
 
@@ -67,16 +108,20 @@ namespace Kernel
 	{
 		ASSERT(&PageTable::current() == &m_page_table);
 
+		LockGuard _(m_mutex);
+
 		const size_t aligned_size = (m_size + PAGE_SIZE - 1) & PAGE_ADDR_MASK;
 		auto result = TRY(MemoryBackedRegion::create(new_page_table, m_size, { .start = m_vaddr, .end = m_vaddr + aligned_size }, m_type, m_flags, m_status_flags));
 
-		for (size_t offset = 0; offset < m_size; offset += PAGE_SIZE)
+		if (writable())
+			m_page_table.remove_writable_from_range(m_vaddr, m_size);
+
+		for (size_t i = 0; i < m_physical_pages.size(); i++)
 		{
-			paddr_t paddr = m_page_table.physical_address_of(m_vaddr + offset);
-			if (paddr == 0)
+			if (m_physical_pages[i] == nullptr)
 				continue;
-			const size_t to_copy = BAN::Math::min<size_t>(PAGE_SIZE, m_size - offset);
-			TRY(result->copy_data_to_region(offset, (const uint8_t*)(m_vaddr + offset), to_copy));
+			result->m_physical_pages[i] = m_physical_pages[i];
+			result->m_physical_pages[i]->ref_count++;
 		}
 
 		return BAN::UniqPtr<MemoryRegion>(BAN::move(result));
@@ -87,20 +132,35 @@ namespace Kernel
 		ASSERT(offset && offset < m_size);
 		ASSERT(offset % PAGE_SIZE == 0);
 
-		auto* new_region = new MemoryBackedRegion(m_page_table, m_size - offset, m_type, m_flags, m_status_flags);
-		if (new_region == nullptr)
+		LockGuard _(m_mutex);
+
+		auto* new_region_ptr = new MemoryBackedRegion(m_page_table, m_size - offset, m_type, m_flags, m_status_flags);
+		if (new_region_ptr == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
+		auto new_region = BAN::UniqPtr<MemoryBackedRegion>::adopt(new_region_ptr);
+
 		new_region->m_vaddr = m_vaddr + offset;
 
+		const size_t moved_pages = (m_size - offset + PAGE_SIZE - 1) / PAGE_SIZE;
+		TRY(new_region->m_physical_pages.resize(moved_pages));
+
+		const size_t remaining_pages = m_physical_pages.size() - moved_pages;
+
+		for (size_t i = 0; i < moved_pages; i++)
+			new_region->m_physical_pages[i] = m_physical_pages[remaining_pages + i];
+		MUST(m_physical_pages.resize(remaining_pages));
+
 		m_size = offset;
 
-		return BAN::UniqPtr<MemoryRegion>::adopt(new_region);
+		return BAN::UniqPtr<MemoryRegion>(BAN::move(new_region));
 	}
 
 	BAN::ErrorOr<void> MemoryBackedRegion::copy_data_to_region(size_t offset_into_region, const uint8_t* buffer, size_t buffer_size)
 	{
 		ASSERT(offset_into_region + buffer_size <= m_size);
 
+		LockGuard _(m_mutex);
+
 		size_t written = 0;
 		while (written < buffer_size)
 		{
@@ -108,18 +168,18 @@ namespace Kernel
 			vaddr_t page_offset = write_vaddr % PAGE_SIZE;
 			size_t bytes = BAN::Math::min<size_t>(buffer_size - written, PAGE_SIZE - page_offset);
 
-			paddr_t paddr = m_page_table.physical_address_of(write_vaddr & PAGE_ADDR_MASK);
-			if (paddr == 0)
+			if (!(m_page_table.get_page_flags(write_vaddr & PAGE_ADDR_MASK) & PageTable::ReadWrite))
 			{
-				if (!TRY(allocate_page_containing(write_vaddr, false)))
+				if (!TRY(allocate_page_containing(write_vaddr, true)))
 				{
 					dwarnln("Could not allocate page for data copying");
 					return BAN::Error::from_errno(EFAULT);
 				}
-				paddr = m_page_table.physical_address_of(write_vaddr & PAGE_ADDR_MASK);
-				ASSERT(paddr);
 			}
 
+			const paddr_t paddr = m_page_table.physical_address_of(write_vaddr & PAGE_ADDR_MASK);
+			ASSERT(paddr);
+
 			PageTable::with_fast_page(paddr, [&] {
 				memcpy(PageTable::fast_page_as_ptr(page_offset), (void*)(buffer + written), bytes);
 			});

kernel/kernel/Memory/MemoryRegion.cpp

@@ -22,8 +22,10 @@ namespace Kernel
 
 	BAN::ErrorOr<void> MemoryRegion::initialize(AddressRange address_range)
 	{
-		size_t needed_pages = BAN::Math::div_round_up<size_t>(m_size, PAGE_SIZE);
-		m_vaddr = m_page_table.reserve_free_contiguous_pages(needed_pages, address_range.start);
+		if (auto rem = address_range.end % PAGE_SIZE)
+			address_range.end += PAGE_SIZE - rem;
+		const size_t needed_pages = BAN::Math::div_round_up<size_t>(m_size, PAGE_SIZE);
+		m_vaddr = m_page_table.reserve_free_contiguous_pages(needed_pages, address_range.start, address_range.end);
 		if (m_vaddr == 0)
 			return BAN::Error::from_errno(ENOMEM);
 		if (m_vaddr + m_size > address_range.end)

kernel/kernel/Memory/PhysicalRange.cpp

@@ -10,7 +10,7 @@ namespace Kernel
 
 	static constexpr size_t bits_per_page = PAGE_SIZE * 8;
 
-	PhysicalRange::PhysicalRange(paddr_t paddr, size_t size)
+	PhysicalRange::PhysicalRange(paddr_t paddr, uint64_t size)
 		: m_paddr(paddr)
 		, m_page_count(size / PAGE_SIZE)
 		, m_free_pages(m_page_count)

kernel/kernel/Memory/VirtualRange.cpp

@@ -1,67 +1,47 @@
+#include <BAN/ScopeGuard.h>
 #include <kernel/Memory/Heap.h>
 #include <kernel/Memory/VirtualRange.h>
 
 namespace Kernel
 {
 
-	BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> VirtualRange::create_to_vaddr(PageTable& page_table, vaddr_t vaddr, size_t size, PageTable::flags_t flags, bool preallocate_pages, bool add_guard_pages)
-	{
-		ASSERT(size % PAGE_SIZE == 0);
-		ASSERT(vaddr % PAGE_SIZE == 0);
-		ASSERT(vaddr > 0);
-
-		if (add_guard_pages)
-		{
-			vaddr -= PAGE_SIZE;
-			size += 2 * PAGE_SIZE;
-		}
-
-		auto result = TRY(BAN::UniqPtr<VirtualRange>::create(page_table, preallocate_pages, add_guard_pages, vaddr, size, flags));
-		ASSERT(page_table.reserve_range(vaddr, size));
-		TRY(result->initialize());
-
-		return result;
-	}
-
-	BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> VirtualRange::create_to_vaddr_range(PageTable& page_table, vaddr_t vaddr_start, vaddr_t vaddr_end, size_t size, PageTable::flags_t flags, bool preallocate_pages, bool add_guard_pages)
+	BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> VirtualRange::create_to_vaddr_range(PageTable& page_table, AddressRange address_range, size_t size, PageTable::flags_t flags, bool add_guard_pages)
 	{
 		if (add_guard_pages)
 			size += 2 * PAGE_SIZE;
 
 		ASSERT(size % PAGE_SIZE == 0);
-		ASSERT(vaddr_start > 0);
-		ASSERT(vaddr_start + size <= vaddr_end);
 
 		// Align vaddr range to page boundaries
-		if (size_t rem = vaddr_start % PAGE_SIZE)
-			vaddr_start += PAGE_SIZE - rem;
-		if (size_t rem = vaddr_end % PAGE_SIZE)
-			vaddr_end -= rem;
-		ASSERT(vaddr_start < vaddr_end);
-		ASSERT(vaddr_end - vaddr_start + 1 >= size / PAGE_SIZE);
+		if (const size_t rem = address_range.start % PAGE_SIZE)
+			address_range.start += PAGE_SIZE - rem;
+		if (const size_t rem = address_range.end % PAGE_SIZE)
+			address_range.end -= rem;
 
-		const vaddr_t vaddr = page_table.reserve_free_contiguous_pages(size / PAGE_SIZE, vaddr_start, vaddr_end);
+		const vaddr_t vaddr = page_table.reserve_free_contiguous_pages(size / PAGE_SIZE, address_range.start, address_range.end);
 		if (vaddr == 0)
 			return BAN::Error::from_errno(ENOMEM);
-		ASSERT(vaddr >= vaddr_start);
-		ASSERT(vaddr + size <= vaddr_end);
 
-		auto result_or_error = BAN::UniqPtr<VirtualRange>::create(page_table, preallocate_pages, add_guard_pages, vaddr, size, flags);
-		if (result_or_error.is_error())
-		{
+		BAN::ScopeGuard vaddr_cleaner([&page_table, vaddr, size] {
 			page_table.unmap_range(vaddr, size);
-			return result_or_error.release_error();
-		}
+		});
 
-		auto result = result_or_error.release_value();
+		auto result = TRY(BAN::UniqPtr<VirtualRange>::create(
+			page_table,
+			add_guard_pages,
+			vaddr,
+			size,
+			flags)
+		);
 		TRY(result->initialize());
 
+		vaddr_cleaner.disable();
+
 		return result;
 	}
 
-	VirtualRange::VirtualRange(PageTable& page_table, bool preallocated, bool has_guard_pages, vaddr_t vaddr, size_t size, PageTable::flags_t flags)
+	VirtualRange::VirtualRange(PageTable& page_table, bool has_guard_pages, vaddr_t vaddr, size_t size, PageTable::flags_t flags)
 		: m_page_table(page_table)
-		, m_preallocated(preallocated)
 		, m_has_guard_pages(has_guard_pages)
 		, m_vaddr(vaddr)
 		, m_size(size)
@@ -71,100 +51,26 @@ namespace Kernel
 	VirtualRange::~VirtualRange()
 	{
 		ASSERT(m_vaddr);
-		m_page_table.unmap_range(m_vaddr, m_size);
-
-		for (paddr_t paddr : m_paddrs)
-			if (paddr != 0)
+		for (size_t off = 0; off < size(); off += PAGE_SIZE)
+			if (const auto paddr = m_page_table.physical_address_of(vaddr() + off))
 				Heap::get().release_page(paddr);
+		m_page_table.unmap_range(m_vaddr, m_size);
 	}
 
 	BAN::ErrorOr<void> VirtualRange::initialize()
 	{
-		TRY(m_paddrs.resize(size() / PAGE_SIZE, 0));
-
-		if (!m_preallocated)
-			return {};
-
 		const size_t page_count = size() / PAGE_SIZE;
 		for (size_t i = 0; i < page_count; i++)
 		{
-			m_paddrs[i] = Heap::get().take_free_page();
-			if (m_paddrs[i] == 0)
+			const auto paddr = Heap::get().take_free_page();
+			if (paddr == 0)
 				return BAN::Error::from_errno(ENOMEM);
-			m_page_table.map_page_at(m_paddrs[i], vaddr() + i * PAGE_SIZE, m_flags);
+			PageTable::with_fast_page(paddr, [] {
+				memset(PageTable::fast_page_as_ptr(), 0, PAGE_SIZE);
+			});
+			m_page_table.map_page_at(paddr, vaddr() + i * PAGE_SIZE, m_flags);
 		}
-
-		if (&PageTable::current() == &m_page_table || &PageTable::kernel() == &m_page_table)
-			memset(reinterpret_cast<void*>(vaddr()), 0, size());
-		else
-		{
-			const size_t page_count = size() / PAGE_SIZE;
-			for (size_t i = 0; i < page_count; i++)
-			{
-				PageTable::with_fast_page(m_paddrs[i], [&] {
-					memset(PageTable::fast_page_as_ptr(), 0, PAGE_SIZE);
-				});
-			}
-		}
-
 		return {};
 	}
 
-	BAN::ErrorOr<BAN::UniqPtr<VirtualRange>> VirtualRange::clone(PageTable& page_table)
-	{
-		ASSERT(&PageTable::current() == &m_page_table);
-		ASSERT(&m_page_table != &page_table);
-
-		SpinLockGuard _(m_lock);
-
-		auto result = TRY(create_to_vaddr(page_table, vaddr(), size(), m_flags, m_preallocated, m_has_guard_pages));
-
-		const size_t page_count = size() / PAGE_SIZE;
-		for (size_t i = 0; i < page_count; i++)
-		{
-			if (m_paddrs[i] == 0)
-				continue;
-			if (!result->m_preallocated)
-			{
-				result->m_paddrs[i] = Heap::get().take_free_page();
-				if (result->m_paddrs[i] == 0)
-					return BAN::Error::from_errno(ENOMEM);
-				result->m_page_table.map_page_at(result->m_paddrs[i], vaddr() + i * PAGE_SIZE, m_flags);
-			}
-
-			PageTable::with_fast_page(result->m_paddrs[i], [&] {
-				memcpy(PageTable::fast_page_as_ptr(), reinterpret_cast<void*>(vaddr() + i * PAGE_SIZE), PAGE_SIZE);
-			});
-		}
-
-		return result;
-	}
-
-	BAN::ErrorOr<bool> VirtualRange::allocate_page_for_demand_paging(vaddr_t vaddr)
-	{
-		ASSERT(contains(vaddr));
-		vaddr &= PAGE_ADDR_MASK;
-
-		if (m_preallocated)
-			return false;
-
-		SpinLockGuard _(m_lock);
-
-		const size_t index = (vaddr - this->vaddr()) / PAGE_SIZE;
-		if (m_paddrs[index])
-			return false;
-
-		m_paddrs[index] = Heap::get().take_free_page();
-		if (m_paddrs[index] == 0)
-			return BAN::Error::from_errno(ENOMEM);
-
-		PageTable::with_fast_page(m_paddrs[index], []{
-			memset(PageTable::fast_page_as_ptr(), 0x00, PAGE_SIZE);
-		});
-
-		m_page_table.map_page_at(m_paddrs[index], vaddr, m_flags);
-
-		return true;
-	}
-
 }

kernel/kernel/Memory/kmalloc.cpp

@@ -436,12 +436,3 @@ BAN::Optional<Kernel::paddr_t> kmalloc_paddr_of(Kernel::vaddr_t vaddr)
 		return {};
 	return vaddr - KERNEL_OFFSET + g_boot_info.kernel_paddr;
 }
-
-BAN::Optional<Kernel::vaddr_t> kmalloc_vaddr_of(Kernel::paddr_t paddr)
-{
-	using namespace Kernel;
-	const vaddr_t vaddr = paddr + KERNEL_OFFSET - g_boot_info.kernel_paddr;
-	if (!is_kmalloc_vaddr(vaddr))
-		return {};
-	return vaddr;
-}

kernel/kernel/Networking/E1000/E1000.cpp

@@ -173,7 +173,7 @@ namespace Kernel
 
 	BAN::ErrorOr<void> E1000::initialize_rx()
 	{
-		m_rx_buffer_region = TRY(DMARegion::create(E1000_RX_BUFFER_SIZE * E1000_RX_DESCRIPTOR_COUNT));
+		m_rx_buffer_region = TRY(DMARegion::create(E1000_RX_BUFFER_SIZE * E1000_RX_DESCRIPTOR_COUNT, PageTable::MemoryType::Normal));
 		m_rx_descriptor_region = TRY(DMARegion::create(sizeof(e1000_rx_desc) * E1000_RX_DESCRIPTOR_COUNT));
 
 		auto* rx_descriptors = reinterpret_cast<volatile e1000_rx_desc*>(m_rx_descriptor_region->vaddr());
@@ -207,7 +207,7 @@ namespace Kernel
 
 	BAN::ErrorOr<void> E1000::initialize_tx()
 	{
-		m_tx_buffer_region = TRY(DMARegion::create(E1000_TX_BUFFER_SIZE * E1000_TX_DESCRIPTOR_COUNT));
+		m_tx_buffer_region = TRY(DMARegion::create(E1000_TX_BUFFER_SIZE * E1000_TX_DESCRIPTOR_COUNT, PageTable::MemoryType::Normal));
 		m_tx_descriptor_region = TRY(DMARegion::create(sizeof(e1000_tx_desc) * E1000_TX_DESCRIPTOR_COUNT));
 
 		auto* tx_descriptors = reinterpret_cast<volatile e1000_tx_desc*>(m_tx_descriptor_region->vaddr());
@@ -304,7 +304,7 @@ namespace Kernel
 		// FIXME: there isnt really any reason to wait for transmission
 		write32(REG_TDT, (tx_current + 1) % E1000_TX_DESCRIPTOR_COUNT);
 		while (descriptor.status == 0)
-			continue;
+			Processor::pause();
 
 		dprintln_if(DEBUG_E1000, "sent {} bytes", packet_size);
 

kernel/kernel/Networking/Loopback.cpp

@@ -14,11 +14,10 @@ namespace Kernel
 
 		loopback->m_buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET,
-			BAN::numeric_limits<vaddr_t>::max(),
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			buffer_size * buffer_count,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, false
+			false
 		));
 
 		auto* thread = TRY(Thread::create_kernel([](void* loopback_ptr) {

kernel/kernel/Networking/NetworkLayer.cpp

@@ -18,16 +18,20 @@ namespace Kernel
 
 			const uint16_t* buffer_u16 = reinterpret_cast<const uint16_t*>(buffer.data());
 			for (size_t j = 0; j < buffer.size() / 2; j++)
-				checksum += BAN::host_to_network_endian(buffer_u16[j]);
-			if (buffer.size() % 2 == 0)
-				continue;
-			ASSERT(i == buffers.size() - 1);
-			checksum += buffer[buffer.size() - 1] << 8;
+				checksum += buffer_u16[j];
+
+			if (buffer.size() % 2)
+			{
+				// NOTE: we only allow last buffer to be odd-length
+				ASSERT(i == buffers.size() - 1);
+				checksum += buffer[buffer.size() - 1];
+			}
 		}
 
 		while (checksum >> 16)
-			checksum = (checksum >> 16) + (checksum & 0xFFFF);
-		return ~(uint16_t)checksum;
+			checksum = (checksum & 0xFFFF) + (checksum >> 16);
+
+		return BAN::host_to_network_endian<uint16_t>(~checksum);
 	}
 
 }

kernel/kernel/Networking/RTL8169/RTL8169.cpp

@@ -107,7 +107,7 @@ namespace Kernel
 
 	BAN::ErrorOr<void> RTL8169::initialize_rx()
 	{
-		m_rx_buffer_region = TRY(DMARegion::create(m_rx_descriptor_count * s_buffer_size));
+		m_rx_buffer_region = TRY(DMARegion::create(m_rx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal));
 		m_rx_descriptor_region = TRY(DMARegion::create(m_rx_descriptor_count * sizeof(RTL8169Descriptor)));
 
 		for (size_t i = 0; i < m_rx_descriptor_count; i++)
@@ -144,7 +144,7 @@ namespace Kernel
 
 	BAN::ErrorOr<void> RTL8169::initialize_tx()
 	{
-		m_tx_buffer_region = TRY(DMARegion::create(m_tx_descriptor_count * s_buffer_size));
+		m_tx_buffer_region = TRY(DMARegion::create(m_tx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal));
 		m_tx_descriptor_region = TRY(DMARegion::create(m_tx_descriptor_count * sizeof(RTL8169Descriptor)));
 
 		for (size_t i = 0; i < m_tx_descriptor_count; i++)

kernel/kernel/Networking/UDPSocket.cpp

@@ -14,11 +14,10 @@ namespace Kernel
 		auto socket = TRY(BAN::RefPtr<UDPSocket>::create(network_layer, info));
 		socket->m_packet_buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET,
-			~(uintptr_t)0,
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			packet_buffer_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, false
+			false
 		));
 		return socket;
 	}

kernel/kernel/Networking/UNIX/Socket.cpp

@@ -14,15 +14,7 @@
 namespace Kernel
 {
 
-	struct UnixSocketHash
-	{
-		BAN::hash_t operator()(const BAN::RefPtr<Inode>& socket)
-		{
-			return BAN::hash<const Inode*>{}(socket.ptr());
-		}
-	};
-
-	static BAN::HashMap<BAN::RefPtr<Inode>, BAN::WeakPtr<UnixDomainSocket>, UnixSocketHash> s_bound_sockets;
+	static BAN::HashMap<BAN::RefPtr<Inode>, BAN::WeakPtr<UnixDomainSocket>> s_bound_sockets;
 	static Mutex s_bound_socket_lock;
 
 	static constexpr size_t s_packet_buffer_size = 0x10000;
@@ -50,11 +42,10 @@ namespace Kernel
 		auto socket = TRY(BAN::RefPtr<UnixDomainSocket>::create(socket_type, info));
 		socket->m_packet_buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET,
-			~(uintptr_t)0,
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			s_packet_buffer_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, false
+			false
 		));
 		return socket;
 	}

kernel/kernel/OpenFileDescriptorSet.cpp

@@ -51,17 +51,19 @@ namespace Kernel
 
 		close_all();
 
-		for (int fd = 0; fd < (int)other.m_open_files.size(); fd++)
+		for (int fd = 0; fd < static_cast<int>(other.m_open_files.size()); fd++)
 		{
 			if (other.validate_fd(fd).is_error())
 				continue;
 
 			auto& open_file = m_open_files[fd];
-			open_file.description      = other.m_open_files[fd].description;
-			open_file.descriptor_flags = other.m_open_files[fd].descriptor_flags;
-			open_file.inode()->on_clone(open_file.status_flags());
+			open_file = other.m_open_files[fd];
+			open_file->file.inode->on_clone(open_file->status_flags);
 		}
 
+		for (size_t i = 0; i < m_cloexec_files.size(); i++)
+			m_cloexec_files[i] = other.m_cloexec_files[i];
+
 		return {};
 	}
 
@@ -81,11 +83,15 @@ namespace Kernel
 		if ((flags & O_TRUNC) && (flags & O_WRONLY) && file.inode->mode().ifreg())
 			TRY(file.inode->truncate(0));
 
-		LockGuard _(m_mutex);
 		constexpr int status_mask = O_APPEND | O_DSYNC | O_NONBLOCK | O_RSYNC | O_SYNC | O_ACCMODE;
-		int fd = TRY(get_free_fd());
-		m_open_files[fd].description = TRY(BAN::RefPtr<OpenFileDescription>::create(BAN::move(file), 0, flags & status_mask));
-		m_open_files[fd].descriptor_flags = flags & O_CLOEXEC;
+
+		LockGuard _(m_mutex);
+
+		const int fd = TRY(get_free_fd());
+		m_open_files[fd] = TRY(BAN::RefPtr<OpenFileDescription>::create(BAN::move(file), 0, flags & status_mask));
+		if (flags & O_CLOEXEC)
+			add_cloexec(fd);
+
 		return fd;
 	}
 
@@ -99,7 +105,7 @@ namespace Kernel
 		Socket::Domain domain;
 		Socket::Type type;
 		int status_flags;
-		int descriptor_flags;
+		bool cloexec;
 	};
 
 	static BAN::ErrorOr<SocketInfo> parse_socket_info(int domain, int type, int protocol)
@@ -124,11 +130,11 @@ namespace Kernel
 		}
 
 		info.status_flags = 0;
-		info.descriptor_flags = 0;
+		info.cloexec = false;
 		if (type & SOCK_NONBLOCK)
 			info.status_flags |= O_NONBLOCK;
 		if (type & SOCK_CLOEXEC)
-			info.descriptor_flags |= O_CLOEXEC;
+			info.cloexec = true;
 		type &= ~(SOCK_NONBLOCK | SOCK_CLOEXEC);
 
 		switch (type)
@@ -171,9 +177,12 @@ namespace Kernel
 			socket_sv = "<udp socket>";
 
 		LockGuard _(m_mutex);
-		int fd = TRY(get_free_fd());
-		m_open_files[fd].description = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket, socket_sv), 0, O_RDWR | sock_info.status_flags));
-		m_open_files[fd].descriptor_flags = sock_info.descriptor_flags;
+
+		const int fd = TRY(get_free_fd());
+		m_open_files[fd] = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket, socket_sv), 0, O_RDWR | sock_info.status_flags));
+		if (sock_info.cloexec)
+			add_cloexec(fd);
+
 		return fd;
 	}
 
@@ -188,10 +197,14 @@ namespace Kernel
 		LockGuard _(m_mutex);
 
 		TRY(get_free_fd_pair(socket_vector));
-		m_open_files[socket_vector[0]].description = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket1, "<socketpair>"_sv), 0, O_RDWR | sock_info.status_flags));
-		m_open_files[socket_vector[0]].descriptor_flags = sock_info.descriptor_flags;
-		m_open_files[socket_vector[1]].description = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket2, "<socketpair>"_sv), 0, O_RDWR | sock_info.status_flags));
-		m_open_files[socket_vector[1]].descriptor_flags = sock_info.descriptor_flags;
+		m_open_files[socket_vector[0]] = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket1, "<socketpair>"_sv), 0, O_RDWR | sock_info.status_flags));
+		m_open_files[socket_vector[1]] = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(socket2, "<socketpair>"_sv), 0, O_RDWR | sock_info.status_flags));
+		if (sock_info.cloexec)
+		{
+			add_cloexec(socket_vector[0]);
+			add_cloexec(socket_vector[1]);
+		}
+
 		return {};
 	}
 
@@ -202,10 +215,11 @@ namespace Kernel
 		TRY(get_free_fd_pair(fds));
 
 		auto pipe = TRY(Pipe::create(m_credentials));
-		m_open_files[fds[0]].description = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(pipe, "<pipe rd>"_sv), 0, O_RDONLY));
-		m_open_files[fds[0]].descriptor_flags = 0;
-		m_open_files[fds[1]].description = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(pipe, "<pipe wr>"_sv), 0, O_WRONLY));
-		m_open_files[fds[1]].descriptor_flags = 0;
+		m_open_files[fds[0]] = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(pipe, "<pipe rd>"_sv), 0, O_RDONLY));
+		m_open_files[fds[1]] = TRY(BAN::RefPtr<OpenFileDescription>::create(VirtualFileSystem::File(pipe, "<pipe wr>"_sv), 0, O_WRONLY));
+
+		ASSERT(!is_cloexec(fds[0]));
+		ASSERT(!is_cloexec(fds[1]));
 
 		return {};
 	}
@@ -224,9 +238,10 @@ namespace Kernel
 		(void)close(fildes2);
 
 		auto& open_file = m_open_files[fildes2];
-		open_file.description = m_open_files[fildes].description;
-		open_file.descriptor_flags = 0;
-		open_file.inode()->on_clone(open_file.status_flags());
+		open_file = m_open_files[fildes];
+		open_file->file.inode->on_clone(open_file->status_flags);
+
+		ASSERT(!is_cloexec(fildes2));
 
 		return fildes2;
 	}
@@ -246,18 +261,16 @@ namespace Kernel
 
 				TRY(validate_fd(fd));
 
-				const auto& open_file = m_open_files[fd];
-
-				inode = open_file.inode();
+				inode = m_open_files[fd]->file.inode;
 
 				switch (flock.l_whence)
 				{
 					case SEEK_SET:
 						break;
 					case SEEK_CUR:
-						if (BAN::Math::will_addition_overflow(flock.l_start, open_file.offset()))
+						if (BAN::Math::will_addition_overflow(flock.l_start, m_open_files[fd]->offset))
 							return BAN::Error::from_errno(EOVERFLOW);
-						flock.l_start += m_open_files[fd].offset();
+						flock.l_start += m_open_files[fd]->offset;
 						break;
 					case SEEK_END:
 						if (BAN::Math::will_addition_overflow(flock.l_start, inode->size()))
@@ -307,26 +320,27 @@ namespace Kernel
 				const int new_fd = TRY(get_free_fd());
 
 				auto& open_file = m_open_files[new_fd];
-				open_file.description = m_open_files[fd].description;
-				open_file.descriptor_flags = (cmd == F_DUPFD_CLOEXEC) ? O_CLOEXEC : 0;
-				open_file.inode()->on_clone(open_file.status_flags());
+				open_file = m_open_files[fd];
+				open_file->file.inode->on_clone(open_file->status_flags);
+				if (cmd == F_DUPFD_CLOEXEC)
+					add_cloexec(new_fd);
 
 				return new_fd;
 			}
 			case F_GETFD:
-				return m_open_files[fd].descriptor_flags;
+				return is_cloexec(fd) ? O_CLOEXEC : 0;
 			case F_SETFD:
 				if (extra & FD_CLOEXEC)
-					m_open_files[fd].descriptor_flags |= O_CLOEXEC;
+					add_cloexec(fd);
 				else
-					m_open_files[fd].descriptor_flags &= ~O_CLOEXEC;
+					remove_cloexec(fd);
 				return 0;
 			case F_GETFL:
-				return m_open_files[fd].status_flags();
+				return m_open_files[fd]->status_flags;
 			case F_SETFL:
 				extra &= O_APPEND | O_DSYNC | O_NONBLOCK | O_RSYNC | O_SYNC;
-				m_open_files[fd].status_flags() &= O_ACCMODE;
-				m_open_files[fd].status_flags() |= extra;
+				m_open_files[fd]->status_flags &= O_ACCMODE;
+				m_open_files[fd]->status_flags |= extra;
 				return 0;
 			default:
 				break;
@@ -349,10 +363,10 @@ namespace Kernel
 				base_offset = 0;
 				break;
 			case SEEK_CUR:
-				base_offset = m_open_files[fd].offset();
+				base_offset = m_open_files[fd]->offset;
 				break;
 			case SEEK_END:
-				base_offset = m_open_files[fd].inode()->size();
+				base_offset = m_open_files[fd]->file.inode->size();
 				break;
 			default:
 				return BAN::Error::from_errno(EINVAL);
@@ -362,7 +376,7 @@ namespace Kernel
 		if (new_offset < 0)
 			return BAN::Error::from_errno(EINVAL);
 
-		m_open_files[fd].offset() = new_offset;
+		m_open_files[fd]->offset = new_offset;
 
 		return new_offset;
 	}
@@ -371,14 +385,14 @@ namespace Kernel
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return m_open_files[fd].offset();
+		return m_open_files[fd]->offset;
 	}
 
 	BAN::ErrorOr<void> OpenFileDescriptorSet::truncate(int fd, off_t length)
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return m_open_files[fd].inode()->truncate(length);
+		return m_open_files[fd]->file.inode->truncate(length);
 	}
 
 	BAN::ErrorOr<void> OpenFileDescriptorSet::close(int fd)
@@ -389,7 +403,7 @@ namespace Kernel
 
 		auto& open_file = m_open_files[fd];
 
-		if (auto& flock = open_file.description->flock; Thread::current().has_process() && flock.lockers.contains(Process::current().pid()))
+		if (auto& flock = open_file->flock; Thread::current().has_process() && flock.lockers.contains(Process::current().pid()))
 		{
 			flock.lockers.remove(Process::current().pid());
 			if (flock.lockers.empty())
@@ -401,7 +415,7 @@ namespace Kernel
 
 		{
 			LockGuard _(s_fcntl_lock_mutex);
-			if (auto it = s_fcntl_locks.find(open_file.inode()); it != s_fcntl_locks.end())
+			if (auto it = s_fcntl_locks.find(open_file->file.inode); it != s_fcntl_locks.end())
 			{
 				auto& flocks = it->value;
 				for (size_t i = 0; i < flocks.size(); i++)
@@ -411,9 +425,9 @@ namespace Kernel
 			}
 		}
 
-		open_file.inode()->on_close(open_file.status_flags());
-		open_file.description.clear();
-		open_file.descriptor_flags = 0;
+		open_file->file.inode->on_close(open_file->status_flags);
+		open_file = {};
+		remove_cloexec(fd);
 
 		return {};
 	}
@@ -421,20 +435,31 @@ namespace Kernel
 	void OpenFileDescriptorSet::close_all()
 	{
 		LockGuard _(m_mutex);
-		for (int fd = 0; fd < (int)m_open_files.size(); fd++)
+		for (int fd = 0; fd < static_cast<int>(m_open_files.size()); fd++)
 			(void)close(fd);
 	}
 
 	void OpenFileDescriptorSet::close_cloexec()
 	{
 		LockGuard _(m_mutex);
-		for (int fd = 0; fd < (int)m_open_files.size(); fd++)
-		{
-			if (validate_fd(fd).is_error())
-				continue;
-			if (m_open_files[fd].descriptor_flags & O_CLOEXEC)
+		for (int fd = 0; fd < static_cast<int>(m_open_files.size()); fd++)
+			if (is_cloexec(fd))
 				(void)close(fd);
-		}
+	}
+
+	bool OpenFileDescriptorSet::is_cloexec(int fd)
+	{
+		return m_cloexec_files[fd / 32] & (1u << (fd % 32));
+	}
+
+	void OpenFileDescriptorSet::add_cloexec(int fd)
+	{
+		m_cloexec_files[fd / 32] |= 1u << (fd % 32);
+	}
+
+	void OpenFileDescriptorSet::remove_cloexec(int fd)
+	{
+		m_cloexec_files[fd / 32] &= ~(1u << (fd % 32));
 	}
 
 	static BAN::ErrorOr<void> fcntl_lock(BAN::RefPtr<Inode> inode, int cmd, struct flock& flock)
@@ -588,7 +613,7 @@ namespace Kernel
 		{
 			TRY(validate_fd(fd));
 
-			auto& flock = m_open_files[fd].description->flock;
+			auto& flock = m_open_files[fd]->flock;
 			switch (op & ~LOCK_NB)
 			{
 				case LOCK_UN:
@@ -645,11 +670,11 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			TRY(validate_fd(fd));
 			auto& open_file = m_open_files[fd];
-			if (!(open_file.status_flags() & O_RDONLY))
+			if (!(open_file->status_flags & O_RDONLY))
 				return BAN::Error::from_errno(EBADF);
-			inode = open_file.inode();
-			is_nonblock = !!(open_file.status_flags() & O_NONBLOCK);
-			offset = open_file.offset();
+			inode = open_file->file.inode;
+			is_nonblock = !!(open_file->status_flags & O_NONBLOCK);
+			offset = open_file->offset;
 		}
 
 		if (inode->mode().ifsock())
@@ -685,7 +710,7 @@ namespace Kernel
 		LockGuard _(m_mutex);
 		// NOTE: race condition with offset, its UB per POSIX
 		if (!validate_fd(fd).is_error())
-			m_open_files[fd].offset() = offset + nread;
+			m_open_files[fd]->offset = offset + nread;
 		return nread;
 	}
 
@@ -699,11 +724,11 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			TRY(validate_fd(fd));
 			auto& open_file = m_open_files[fd];
-			if (!(open_file.status_flags() & O_WRONLY))
+			if (!(open_file->status_flags & O_WRONLY))
 				return BAN::Error::from_errno(EBADF);
-			inode = open_file.inode();
-			is_nonblock = !!(open_file.status_flags() & O_NONBLOCK);
-			offset = (open_file.status_flags() & O_APPEND) ? inode->size() : open_file.offset();
+			inode = open_file->file.inode;
+			is_nonblock = !!(open_file->status_flags & O_NONBLOCK);
+			offset = (open_file->status_flags & O_APPEND) ? inode->size() : open_file->offset;
 		}
 
 		if (inode->mode().ifsock())
@@ -742,7 +767,7 @@ namespace Kernel
 		LockGuard _(m_mutex);
 		// NOTE: race condition with offset, its UB per POSIX
 		if (!validate_fd(fd).is_error())
-			m_open_files[fd].offset() = offset + nwrite;
+			m_open_files[fd]->offset = offset + nwrite;
 		return nwrite;
 	}
 
@@ -755,10 +780,10 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			TRY(validate_fd(fd));
 			auto& open_file = m_open_files[fd];
-			if (!(open_file.status_flags() & O_RDONLY))
+			if (!(open_file->status_flags & O_RDONLY))
 				return BAN::Error::from_errno(EACCES);
-			inode = open_file.inode();
-			offset = open_file.offset();
+			inode = open_file->file.inode;
+			offset = open_file->offset;
 		}
 
 		for (;;)
@@ -773,7 +798,7 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			// NOTE: race condition with offset, its UB per POSIX
 			if (!validate_fd(fd).is_error())
-				m_open_files[fd].offset() = offset;
+				m_open_files[fd]->offset = offset;
 			return ret;
 		}
 	}
@@ -787,10 +812,10 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			TRY(validate_fd(fd));
 			auto& open_file = m_open_files[fd];
-			if (!open_file.inode()->mode().ifsock())
+			if (!open_file->file.inode->mode().ifsock())
 				return BAN::Error::from_errno(ENOTSOCK);
-			inode = open_file.inode();
-			is_nonblock = !!(open_file.status_flags() & O_NONBLOCK);
+			inode = open_file->file.inode;
+			is_nonblock = !!(open_file->status_flags & O_NONBLOCK);
 		}
 
 		LockGuard _(inode->m_mutex);
@@ -808,10 +833,10 @@ namespace Kernel
 			LockGuard _(m_mutex);
 			TRY(validate_fd(fd));
 			auto& open_file = m_open_files[fd];
-			if (!open_file.inode()->mode().ifsock())
+			if (!open_file->file.inode->mode().ifsock())
 				return BAN::Error::from_errno(ENOTSOCK);
-			inode = open_file.inode();
-			is_nonblock = !!(open_file.status_flags() & O_NONBLOCK);
+			inode = open_file->file.inode;
+			is_nonblock = !!(open_file->status_flags & O_NONBLOCK);
 		}
 
 		LockGuard _(inode->m_mutex);
@@ -830,7 +855,7 @@ namespace Kernel
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return TRY(m_open_files[fd].description->file.clone());
+		return TRY(m_open_files[fd]->file.clone());
 	}
 
 	BAN::ErrorOr<BAN::String> OpenFileDescriptorSet::path_of(int fd) const
@@ -838,7 +863,7 @@ namespace Kernel
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
 		BAN::String path;
-		TRY(path.append(m_open_files[fd].path()));
+		TRY(path.append(m_open_files[fd]->file.canonical_path));
 		return path;
 	}
 
@@ -846,14 +871,14 @@ namespace Kernel
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return m_open_files[fd].inode();
+		return m_open_files[fd]->file.inode;
 	}
 
 	BAN::ErrorOr<int> OpenFileDescriptorSet::status_flags_of(int fd) const
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return m_open_files[fd].status_flags();
+		return m_open_files[fd]->status_flags;
 	}
 
 	BAN::ErrorOr<void> OpenFileDescriptorSet::validate_fd(int fd) const
@@ -861,7 +886,7 @@ namespace Kernel
 		LockGuard _(m_mutex);
 		if (fd < 0 || fd >= (int)m_open_files.size())
 			return BAN::Error::from_errno(EBADF);
-		if (!m_open_files[fd].description)
+		if (!m_open_files[fd])
 			return BAN::Error::from_errno(EBADF);
 		return {};
 	}
@@ -870,7 +895,7 @@ namespace Kernel
 	{
 		LockGuard _(m_mutex);
 		for (int fd = 0; fd < (int)m_open_files.size(); fd++)
-			if (!m_open_files[fd].description)
+			if (!m_open_files[fd])
 				return fd;
 		return BAN::Error::from_errno(EMFILE);
 	}
@@ -881,7 +906,7 @@ namespace Kernel
 		size_t found = 0;
 		for (int fd = 0; fd < (int)m_open_files.size(); fd++)
 		{
-			if (!m_open_files[fd].description)
+			if (!m_open_files[fd])
 				fds[found++] = fd;
 			if (found == 2)
 				return {};
@@ -929,7 +954,7 @@ namespace Kernel
 	{
 		LockGuard _(m_mutex);
 		TRY(validate_fd(fd));
-		return FDWrapper { m_open_files[fd].description };
+		return FDWrapper { m_open_files[fd] };
 	}
 
 	size_t OpenFileDescriptorSet::open_all_fd_wrappers(BAN::Span<FDWrapper> fd_wrappers)
@@ -943,8 +968,7 @@ namespace Kernel
 				return i;
 
 			const int fd = fd_or_error.release_value();
-			m_open_files[fd].description = BAN::move(fd_wrappers[i].m_description);
-			m_open_files[fd].descriptor_flags = 0;
+			m_open_files[fd] = BAN::move(fd_wrappers[i].m_description);
 			fd_wrappers[i].m_fd = fd;
 		}
 

kernel/kernel/Process.cpp

@@ -34,6 +34,8 @@
 #include <sys/sysmacros.h>
 #include <sys/wait.h>
 
+#pragma GCC diagnostic ignored "-Wstack-usage="
+
 namespace Kernel
 {
 
@@ -303,6 +305,8 @@ namespace Kernel
 
 	void Process::exit(int status, int signal)
 	{
+		ASSERT(Processor::get_interrupt_state() == InterruptState::Enabled);
+
 		bool expected = false;
 		if (!m_is_exiting.compare_exchange(expected, true))
 		{
@@ -310,9 +314,6 @@ namespace Kernel
 			ASSERT_NOT_REACHED();
 		}
 
-		const auto state = Processor::get_interrupt_state();
-		Processor::set_interrupt_state(InterruptState::Enabled);
-
 		if (m_parent)
 		{
 			Process* parent_process = nullptr;
@@ -369,8 +370,6 @@ namespace Kernel
 		while (m_threads.size() > 1)
 			Processor::yield();
 
-		Processor::set_interrupt_state(state);
-
 		Thread::current().on_exit();
 
 		ASSERT_NOT_REACHED();
@@ -2951,7 +2950,7 @@ namespace Kernel
 		for (;;)
 		{
 			while (Thread::current().will_exit_because_of_signal())
-				Thread::current().handle_signal();
+				Thread::current().handle_signal_if_interrupted();
 
 			SpinLockGuard guard(m_signal_lock);
 			if (!m_stopped)
@@ -3050,7 +3049,7 @@ namespace Kernel
 			TRY(read_from_user(user_act, &new_act, sizeof(struct sigaction)));
 
 		{
-			SpinLockGuard signal_lock_guard(m_signal_lock);
+			SpinLockGuard _(m_signal_lock);
 			old_act = m_signal_handlers[signal];
 			if (user_act != nullptr)
 				m_signal_handlers[signal] = new_act;
@@ -3064,7 +3063,14 @@ namespace Kernel
 
 	BAN::ErrorOr<long> Process::sys_sigpending(sigset_t* user_sigset)
 	{
-		const sigset_t sigset = (signal_pending_mask() | Thread::current().m_signal_pending_mask) & Thread::current().m_signal_block_mask;
+		sigset_t sigset;
+
+		{
+			auto& thread = Thread::current();
+			SpinLockGuard _(thread.m_signal_lock);
+			sigset = (signal_pending_mask() | thread.m_signal_pending_mask) & thread.m_signal_block_mask;
+		}
+
 		TRY(write_to_user(user_sigset, &sigset, sizeof(sigset_t)));
 		return 0;
 	}
@@ -3073,34 +3079,36 @@ namespace Kernel
 	{
 		LockGuard _(m_process_lock);
 
-		if (user_oset != nullptr)
-		{
-			const sigset_t current = Thread::current().m_signal_block_mask;
-			TRY(write_to_user(user_oset, &current, sizeof(sigset_t)));
-		}
+		auto& thread = Thread::current();
+
+		const sigset_t old_sigset = thread.m_signal_block_mask;
 
 		if (user_set != nullptr)
 		{
-			sigset_t set;
-			TRY(read_from_user(user_set, &set, sizeof(sigset_t)));
+			sigset_t mask;
+			TRY(read_from_user(user_set, &mask, sizeof(sigset_t)));
+			mask &= ~((1ull << SIGKILL) | (1ull << SIGSTOP));
 
-			const sigset_t mask = set & ~(SIGKILL | SIGSTOP);
+			SpinLockGuard _(thread.m_signal_lock);
 			switch (how)
 			{
 				case SIG_BLOCK:
-					Thread::current().m_signal_block_mask |= mask;
-					break;
-				case SIG_SETMASK:
-					Thread::current().m_signal_block_mask  = mask;
+					thread.m_signal_block_mask |= mask;
 					break;
 				case SIG_UNBLOCK:
-					Thread::current().m_signal_block_mask &= ~mask;
+					thread.m_signal_block_mask &= ~mask;
+					break;
+				case SIG_SETMASK:
+					thread.m_signal_block_mask = mask;
 					break;
 				default:
 					return BAN::Error::from_errno(EINVAL);
 			}
 		}
 
+		if (user_oset != nullptr)
+			TRY(write_to_user(user_oset, &old_sigset, sizeof(sigset_t)));
+
 		return 0;
 	}
 
@@ -3112,7 +3120,7 @@ namespace Kernel
 		LockGuard _(m_process_lock);
 
 		auto& thread = Thread::current();
-		thread.set_suspend_signal_mask(set & ~(SIGKILL | SIGSTOP));
+		thread.set_suspend_signal_mask(set & ~((1ull << SIGKILL) | (1ull << SIGSTOP)));
 
 		// FIXME: i *think* here is a race condition as kill doesnt hold process lock
 		while (!thread.is_interrupted_by_signal())
@@ -3207,6 +3215,9 @@ namespace Kernel
 				return BAN::Error::from_errno(ENOSYS);
 		}
 
+		if (op == FUTEX_WAIT && BAN::atomic_load(*addr) != val)
+			return BAN::Error::from_errno(EAGAIN);
+
 		uint64_t wake_time_ns = BAN::numeric_limits<uint64_t>::max();
 
 		if (user_abstime != nullptr)
@@ -3228,9 +3239,6 @@ namespace Kernel
 			}
 		}
 
-		if (op == FUTEX_WAIT && BAN::atomic_load(*addr) != val)
-			return BAN::Error::from_errno(EAGAIN);
-
 		futex_t* futex;
 
 		{
@@ -3305,7 +3313,8 @@ namespace Kernel
 
 	BAN::ErrorOr<long> Process::sys_set_fsbase(void* addr)
 	{
-		Thread::current().set_fsbase(reinterpret_cast<vaddr_t>(addr));
+		auto& thread = Thread::current();
+		thread.set_fsbase(reinterpret_cast<vaddr_t>(addr));
 		Processor::load_fsbase();
 		return 0;
 	}
@@ -3317,7 +3326,8 @@ namespace Kernel
 
 	BAN::ErrorOr<long> Process::sys_set_gsbase(void* addr)
 	{
-		Thread::current().set_gsbase(reinterpret_cast<vaddr_t>(addr));
+		auto& thread = Thread::current();
+		thread.set_gsbase(reinterpret_cast<vaddr_t>(addr));
 		Processor::load_gsbase();
 		return 0;
 	}
@@ -3345,9 +3355,8 @@ namespace Kernel
 	{
 		LockGuard _(m_process_lock);
 
-		// main thread cannot call pthread_exit
-		if (&Thread::current() == m_threads.front())
-			return BAN::Error::from_errno(EINVAL);
+		if (m_threads.size() == 1)
+			return sys_exit(0);
 
 		TRY(m_exited_pthreads.emplace_back(Thread::current().tid(), value));
 
@@ -3387,7 +3396,8 @@ namespace Kernel
 		{
 			if (auto ret = check_thread(); ret.has_value())
 			{
-				TRY(write_to_user(user_value, &ret.value(), sizeof(void*)));
+				if (user_value != nullptr)
+					TRY(write_to_user(user_value, &ret.value(), sizeof(void*)));
 				return 0;
 			}
 
@@ -3887,237 +3897,45 @@ namespace Kernel
 		return region->allocate_page_containing(address, wants_write);
 	}
 
-	// TODO: The following 3 functions could be simplified into one generic helper function
+	extern "C" bool safe_user_memcpy(void*, const void*, size_t);
+	extern "C" bool safe_user_strncpy(void*, const void*, size_t);
+
+	static inline bool is_valid_user_address(const void* user_addr, size_t size)
+	{
+		const vaddr_t user_vaddr = reinterpret_cast<vaddr_t>(user_addr);
+		if (BAN::Math::will_addition_overflow<vaddr_t>(user_vaddr, size))
+			return false;
+		if (user_vaddr + size > USERSPACE_END)
+			return false;
+		return true;
+	}
 
 	BAN::ErrorOr<void> Process::read_from_user(const void* user_addr, void* out, size_t size)
 	{
-		const vaddr_t user_vaddr = reinterpret_cast<vaddr_t>(user_addr);
-
-		auto* out_u8 = static_cast<uint8_t*>(out);
-		size_t ncopied = 0;
-
-		{
-			RWLockRDGuard _(m_memory_region_lock);
-
-			const size_t first_index = find_mapped_region(user_vaddr);
-			for (size_t i = first_index; ncopied < size && i < m_mapped_regions.size(); i++)
-			{
-				auto& region = m_mapped_regions[i];
-				if (!region->contains(user_vaddr + ncopied))
-					return BAN::Error::from_errno(EFAULT);
-
-				const size_t ncopy = BAN::Math::min<size_t>(
-					(region->vaddr() + region->size()) - (user_vaddr + ncopied),
-					size - ncopied
-				);
-
-				const size_t page_count = range_page_count(user_vaddr + ncopied, ncopy);
-				const vaddr_t page_base = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-				for (size_t p = 0; p < page_count; p++)
-				{
-					const auto flags = PageTable::UserSupervisor | PageTable::Present;
-					if ((m_page_table->get_page_flags(page_base + p * PAGE_SIZE) & flags) != flags)
-						goto read_from_user_with_allocation;
-				}
-
-				memcpy(out_u8 + ncopied, reinterpret_cast<void*>(user_vaddr + ncopied), ncopy);
-				ncopied += ncopy;
-			}
-
-			if (ncopied >= size)
-				return {};
-			if (ncopied > 0)
-				return BAN::Error::from_errno(EFAULT);
-		}
-
-	read_from_user_with_allocation:
-		RWLockWRGuard _(m_memory_region_lock);
-
-		const size_t first_index = find_mapped_region(user_vaddr + ncopied);
-		for (size_t i = first_index; ncopied < size && i < m_mapped_regions.size(); i++)
-		{
-			auto& region = m_mapped_regions[i];
-			if (!region->contains(user_vaddr + ncopied))
-				return BAN::Error::from_errno(EFAULT);
-
-			const size_t ncopy = BAN::Math::min<size_t>(
-				(region->vaddr() + region->size()) - (user_vaddr + ncopied),
-				size - ncopied
-			);
-
-			const size_t page_count = range_page_count(user_vaddr + ncopied, ncopy);
-			const vaddr_t page_base = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-			for (size_t p = 0; p < page_count; p++)
-			{
-				const auto flags = PageTable::UserSupervisor | PageTable::Present;
-				if ((m_page_table->get_page_flags(page_base + p * PAGE_SIZE) & flags) == flags)
-					continue;
-				if (!TRY(region->allocate_page_containing(page_base + p * PAGE_SIZE, false)))
-					return BAN::Error::from_errno(EFAULT);
-			}
-
-			memcpy(out_u8 + ncopied, reinterpret_cast<void*>(user_vaddr + ncopied), ncopy);
-			ncopied += ncopy;
-		}
-
-		if (ncopied >= size)
-			return {};
-		return BAN::Error::from_errno(EFAULT);
+		if (!is_valid_user_address(user_addr, size))
+			return BAN::Error::from_errno(EFAULT);
+		if (!safe_user_memcpy(out, user_addr, size))
+			return BAN::Error::from_errno(EFAULT);
+		return {};
 	}
 
 	BAN::ErrorOr<void> Process::read_string_from_user(const char* user_addr, char* out, size_t max_size)
 	{
-		const vaddr_t user_vaddr = reinterpret_cast<vaddr_t>(user_addr);
-
-		size_t ncopied = 0;
-
-		{
-			RWLockRDGuard _(m_memory_region_lock);
-
-			const size_t first_index = find_mapped_region(user_vaddr);
-			for (size_t i = first_index; ncopied < max_size && i < m_mapped_regions.size(); i++)
-			{
-				auto& region = m_mapped_regions[i];
-				if (!region->contains(user_vaddr + ncopied))
-					return BAN::Error::from_errno(EFAULT);
-
-				vaddr_t last_page = 0;
-
-				for (; ncopied < max_size; ncopied++)
-				{
-					const vaddr_t curr_page = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-					if (curr_page != last_page)
-					{
-						const auto flags = PageTable::UserSupervisor | PageTable::Present;
-						if ((m_page_table->get_page_flags(curr_page) & flags) != flags)
-							goto read_string_from_user_with_allocation;
-					}
-
-					out[ncopied] = user_addr[ncopied];
-					if (out[ncopied] == '\0')
-						return {};
-
-					last_page = curr_page;
-				}
-			}
-
-			if (ncopied >= max_size)
-				return BAN::Error::from_errno(ENAMETOOLONG);
-			if (ncopied > 0)
-				return BAN::Error::from_errno(EFAULT);
-		}
-
-	read_string_from_user_with_allocation:
-		RWLockWRGuard _(m_memory_region_lock);
-
-		const size_t first_index = find_mapped_region(user_vaddr + ncopied);
-		for (size_t i = first_index; ncopied < max_size && i < m_mapped_regions.size(); i++)
-		{
-			auto& region = m_mapped_regions[i];
-			if (!region->contains(user_vaddr + ncopied))
-				return BAN::Error::from_errno(EFAULT);
-
-			vaddr_t last_page = 0;
-
-			for (; ncopied < max_size; ncopied++)
-			{
-				const vaddr_t curr_page = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-				if (curr_page != last_page)
-				{
-					const auto flags = PageTable::UserSupervisor | PageTable::Present;
-					if ((m_page_table->get_page_flags(curr_page) & flags) == flags)
-						;
-					else if (!TRY(region->allocate_page_containing(curr_page, false)))
-						return BAN::Error::from_errno(EFAULT);
-				}
-
-				out[ncopied] = user_addr[ncopied];
-				if (out[ncopied] == '\0')
-					return {};
-
-				last_page = curr_page;
-			}
-		}
-
-		if (ncopied >= max_size)
-			return BAN::Error::from_errno(ENAMETOOLONG);
-		return BAN::Error::from_errno(EFAULT);
+		max_size = BAN::Math::min<size_t>(max_size, USERSPACE_END - reinterpret_cast<vaddr_t>(user_addr));
+		if (!is_valid_user_address(user_addr, max_size))
+			return BAN::Error::from_errno(EFAULT);
+		if (!safe_user_strncpy(out, user_addr, max_size))
+			return BAN::Error::from_errno(EFAULT);
+		return {};
 	}
 
 	BAN::ErrorOr<void> Process::write_to_user(void* user_addr, const void* in, size_t size)
 	{
-		const vaddr_t user_vaddr = reinterpret_cast<vaddr_t>(user_addr);
-
-		const auto* in_u8 = static_cast<const uint8_t*>(in);
-		size_t ncopied = 0;
-
-		{
-			RWLockRDGuard _(m_memory_region_lock);
-
-			const size_t first_index = find_mapped_region(user_vaddr);
-			for (size_t i = first_index; ncopied < size && i < m_mapped_regions.size(); i++)
-			{
-				auto& region = m_mapped_regions[i];
-				if (!region->contains(user_vaddr + ncopied))
-					return BAN::Error::from_errno(EFAULT);
-
-				const size_t ncopy = BAN::Math::min<size_t>(
-					(region->vaddr() + region->size()) - (user_vaddr + ncopied),
-					size - ncopied
-				);
-
-				const size_t page_count = range_page_count(user_vaddr + ncopied, ncopy);
-				const vaddr_t page_base = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-				for (size_t i = 0; i < page_count; i++)
-				{
-					const auto flags = PageTable::UserSupervisor | PageTable::ReadWrite | PageTable::Present;
-					if ((m_page_table->get_page_flags(page_base + i * PAGE_SIZE) & flags) != flags)
-						goto write_to_user_with_allocation;
-				}
-
-				memcpy(reinterpret_cast<void*>(user_vaddr + ncopied), in_u8 + ncopied, ncopy);
-				ncopied += ncopy;
-			}
-
-			if (ncopied >= size)
-				return {};
-			if (ncopied > 0)
-				return BAN::Error::from_errno(EFAULT);
-		}
-
-	write_to_user_with_allocation:
-		RWLockWRGuard _(m_memory_region_lock);
-
-		const size_t first_index = find_mapped_region(user_vaddr + ncopied);
-		for (size_t i = first_index; ncopied < size && i < m_mapped_regions.size(); i++)
-		{
-			auto& region = m_mapped_regions[i];
-			if (!region->contains(user_vaddr + ncopied))
-				return BAN::Error::from_errno(EFAULT);
-
-			const size_t ncopy = BAN::Math::min<size_t>(
-				(region->vaddr() + region->size()) - (user_vaddr + ncopied),
-				size - ncopied
-			);
-
-			const size_t page_count = range_page_count(user_vaddr + ncopied, ncopy);
-			const vaddr_t page_base = (user_vaddr + ncopied) & PAGE_ADDR_MASK;
-			for (size_t p = 0; p < page_count; p++)
-			{
-				const auto flags = PageTable::UserSupervisor | PageTable::ReadWrite | PageTable::Present;
-				if ((m_page_table->get_page_flags(page_base + p * PAGE_SIZE) & flags) == flags)
-					continue;
-				if (!TRY(region->allocate_page_containing(page_base + p * PAGE_SIZE, true)))
-					return BAN::Error::from_errno(EFAULT);
-			}
-
-			memcpy(reinterpret_cast<void*>(user_vaddr + ncopied), in_u8 + ncopied, ncopy);
-			ncopied += ncopy;
-		}
-
-		if (ncopied >= size)
-			return {};
-		return BAN::Error::from_errno(EFAULT);
+		if (!is_valid_user_address(user_addr, size))
+			return BAN::Error::from_errno(EFAULT);
+		if (!safe_user_memcpy(user_addr, in, size))
+			return BAN::Error::from_errno(EFAULT);
+		return {};
 	}
 
 	BAN::ErrorOr<MemoryRegion*> Process::validate_and_pin_pointer_access(const void* ptr, size_t size, bool needs_write)

kernel/kernel/Processor.cpp

@@ -73,9 +73,6 @@ namespace Kernel
 		ASSERT(processor.m_id == PROCESSOR_NONE);
 		processor.m_id = id;
 
-		processor.m_stack = kmalloc(s_stack_size, 4096, true);
-		ASSERT(processor.m_stack);
-
 		processor.m_gdt = GDT::create(&processor);
 		ASSERT(processor.m_gdt);
 
@@ -157,6 +154,21 @@ namespace Kernel
 		return processor;
 	}
 
+	// NOTE: I don't like this being a separate function but we need heap and page tables for this :)
+	void Processor::allocate_stack()
+	{
+		ASSERT(m_stack_paddr == 0);
+		ASSERT(m_stack_vaddr == 0);
+
+		m_stack_paddr = Heap::get().take_free_page();
+		ASSERT(m_stack_paddr);
+
+		m_stack_vaddr = PageTable::kernel().reserve_free_page(KERNEL_OFFSET);
+		ASSERT(m_stack_vaddr);
+
+		PageTable::kernel().map_page_at(m_stack_paddr, m_stack_vaddr, PageTable::ReadWrite | PageTable::Present);
+	}
+
 	void Processor::initialize_smp()
 	{
 		const auto processor_id = current_id();
@@ -205,8 +217,7 @@ namespace Kernel
 		memset(reinterpret_cast<void*>(s_shared_page_vaddr), 0, PAGE_SIZE);
 
 		auto& shared_page = *reinterpret_cast<volatile API::SharedPage*>(s_shared_page_vaddr);
-		for (size_t i = 0; i <= 0xFF; i++)
-			shared_page.__sequence[i] = i;
+		shared_page.gdt_cpu_offset = GDT::cpu_index_offset();
 		shared_page.features = 0;
 
 		ASSERT(Processor::count() + sizeof(Kernel::API::SharedPage) <= PAGE_SIZE);
@@ -377,8 +388,7 @@ namespace Kernel
 				case SMPMessage::Type::FlushTLB:
 					if (message->flush_tlb.page_table && message->flush_tlb.page_table != processor.m_current_page_table)
 						break;
-					for (size_t i = 0; i < message->flush_tlb.page_count; i++)
-						asm volatile("invlpg (%0)" :: "r"(message->flush_tlb.vaddr + i * PAGE_SIZE) : "memory");
+					PageTable::current().invalidate_range(message->flush_tlb.vaddr, message->flush_tlb.page_count, false);
 					break;
 				case SMPMessage::Type::NewThread:
 					processor.m_scheduler->add_thread(message->new_thread);
@@ -566,7 +576,7 @@ namespace Kernel
 		if (!scheduler().is_idle())
 			Thread::current().set_cpu_time_stop();
 
-		asm_yield_trampoline(Processor::current_stack_top());
+		asm_yield_trampoline(processor_info.stack_top_vaddr());
 
 		processor_info.m_start_ns = SystemTimer::get().ns_since_boot();
 

kernel/kernel/Scheduler.cpp

@@ -139,6 +139,8 @@ namespace Kernel
 		if (Processor::count() > 1)
 			Processor::set_smp_enabled();
 
+		m_next_reschedule_ns = SystemTimer::get().ns_since_boot();
+
 		return {};
 	}
 
@@ -305,15 +307,14 @@ namespace Kernel
 
 	void Scheduler::wake_up_sleeping_threads()
 	{
+		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
+
 		const uint64_t current_ns = SystemTimer::get().ns_since_boot();
 		while (!m_block_queue.empty() && current_ns >= m_block_queue.front()->wake_time_ns)
 		{
 			auto* node = m_block_queue.pop_front();
-			{
-				SpinLockGuard _(node->blocker_lock);
-				if (node->blocker)
-					node->blocker->remove_blocked_thread(node);
-			}
+			if (auto* blocker = node->blocker.load())
+				blocker->remove_thread_from_block_queue(node);
 			node->blocked = false;
 			update_most_loaded_node_queue(node, &m_run_queue);
 			m_run_queue.add_thread_to_back(node);
@@ -348,13 +349,10 @@ namespace Kernel
 
 		wake_up_sleeping_threads();
 
+		if (SystemTimer::get().ns_since_boot() >= m_next_reschedule_ns)
 		{
-			const uint64_t current_ns = SystemTimer::get().ns_since_boot();
-			if (current_ns >= m_last_reschedule_ns + s_reschedule_interval_ns)
-			{
-				m_last_reschedule_ns = current_ns;
-				Processor::yield();
-			}
+			m_next_reschedule_ns += s_reschedule_interval_ns;
+			Processor::yield();
 		}
 	}
 
@@ -369,11 +367,8 @@ namespace Kernel
 				return;
 			if (node != m_current)
 				m_block_queue.remove_node(node);
-			{
-				SpinLockGuard _(node->blocker_lock);
-				if (node->blocker)
-					node->blocker->remove_blocked_thread(node);
-			}
+			if (auto* blocker = node->blocker.load())
+				blocker->remove_thread_from_block_queue(node);
 			node->blocked = false;
 			if (node != m_current)
 				m_run_queue.add_thread_to_back(node);
@@ -666,11 +661,8 @@ namespace Kernel
 		m_current->blocked = true;
 		m_current->wake_time_ns = wake_time_ns;
 
-		{
-			SpinLockGuard _(m_current->blocker_lock);
-			if (blocker)
-				blocker->add_thread_to_block_queue(m_current);
-		}
+		if (blocker)
+			blocker->add_thread_to_block_queue(m_current);
 
 		update_most_loaded_node_queue(m_current, &m_block_queue);
 

kernel/kernel/Syscall.cpp

@@ -7,6 +7,7 @@
 #include <kernel/Timer/Timer.h>
 
 #include <termios.h>
+#include <sys/syscall.h>
 
 #define DUMP_ALL_SYSCALLS 0
 #define DUMP_LONG_SYSCALLS 0
@@ -94,13 +95,11 @@ namespace Kernel
 
 		Process::current().wait_while_stopped();
 
-		Processor::set_interrupt_state(InterruptState::Disabled);
+		if (Thread::current().handle_signal_if_interrupted())
+			if (ret.is_error() && ret.error().get_error_code() == EINTR && is_restartable_syscall(syscall))
+				ret = BAN::Error::from_errno(ERESTART);
 
-		auto& current_thread = Thread::current();
-		if (current_thread.can_add_signal_to_execute())
-			if (current_thread.handle_signal())
-				if (ret.is_error() && ret.error().get_error_code() == EINTR && is_restartable_syscall(syscall))
-					ret = BAN::Error::from_errno(ERESTART);
+		Processor::set_interrupt_state(InterruptState::Disabled);
 
 		ASSERT(Kernel::Thread::current().state() == Kernel::Thread::State::Executing);
 

kernel/kernel/Terminal/PseudoTerminal.cpp

@@ -42,10 +42,10 @@ namespace Kernel
 
 		auto pts_master_buffer = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET, static_cast<vaddr_t>(-1),
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			16 * PAGE_SIZE,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, false
+			false
 		));
 		auto pts_master = TRY(BAN::RefPtr<PseudoTerminalMaster>::create(BAN::move(pts_master_buffer), mode, uid, gid));
 		DevFileSystem::get().remove_from_cache(pts_master);

kernel/kernel/Thread.cpp

@@ -13,6 +13,12 @@
 namespace Kernel
 {
 
+#if ARCH(x86_64)
+	static constexpr vaddr_t s_user_stack_addr_start = 0x0000700000000000;
+#elif ARCH(i686)
+	static constexpr vaddr_t s_user_stack_addr_start = 0xB0000000;
+#endif
+
 	extern "C" [[noreturn]] void start_kernel_thread();
 	extern "C" [[noreturn]] void start_userspace_thread();
 
@@ -25,11 +31,6 @@ namespace Kernel
 		*(uintptr_t*)rsp = (uintptr_t)value;
 	}
 
-	extern "C" uintptr_t get_thread_start_sp()
-	{
-		return Thread::current().yield_registers().sp;
-	}
-
 	static pid_t s_next_tid = 1;
 
 	alignas(16) static uint8_t s_default_sse_storage[512];
@@ -170,11 +171,10 @@ namespace Kernel
 		// Initialize stack and registers
 		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
 			PageTable::kernel(),
-			KERNEL_OFFSET,
-			~(uintptr_t)0,
+			{ KERNEL_OFFSET, UINTPTR_MAX },
 			kernel_stack_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, true
+			true
 		));
 
 		// Initialize stack for returning
@@ -205,24 +205,18 @@ namespace Kernel
 
 		thread->m_is_userspace = true;
 
-#if ARCH(x86_64)
-		static constexpr vaddr_t stack_addr_start = 0x0000700000000000;
-#elif ARCH(i686)
-		static constexpr vaddr_t stack_addr_start = 0xB0000000;
-#endif
-
 		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
 			page_table,
-			stack_addr_start, USERSPACE_END,
+			{ s_user_stack_addr_start, USERSPACE_END },
 			kernel_stack_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			true, true
+			true
 		));
 
 		auto userspace_stack = TRY(MemoryBackedRegion::create(
 			page_table,
 			userspace_stack_size,
-			{ stack_addr_start, USERSPACE_END },
+			{ s_user_stack_addr_start, USERSPACE_END },
 			MemoryRegion::Type::PRIVATE,
 			PageTable::Flags::UserSupervisor | PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			O_RDWR
@@ -309,14 +303,7 @@ namespace Kernel
 	{
 		if (!is_userspace() || !has_process())
 			return;
-
-		const vaddr_t vaddr = process().shared_page_vaddr() + Processor::current_index();
-
-#if ARCH(x86_64)
-		set_gsbase(vaddr);
-#elif ARCH(i686)
-		set_fsbase(vaddr);
-#endif
+		Processor::gdt().set_cpu_index(Processor::current_index());
 	}
 
 	BAN::ErrorOr<Thread*> Thread::pthread_create(entry_t entry, void* arg)
@@ -351,7 +338,24 @@ namespace Kernel
 
 		thread->m_is_userspace = true;
 
-		thread->m_kernel_stack = TRY(m_kernel_stack->clone(new_process->page_table()));
+		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
+			new_process->page_table(),
+			{ s_user_stack_addr_start, USERSPACE_END },
+			kernel_stack_size,
+			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
+			true
+		));
+
+		// NOTE: copy [sp, stack_end] so fork return works
+		PageTable::with_fast_page(thread->m_kernel_stack->paddr_of(thread->kernel_stack_top() - PAGE_SIZE), [&] {
+			const size_t ncopy = kernel_stack_top() - sp;
+			ASSERT(ncopy <= PAGE_SIZE);
+			memcpy(
+				PageTable::fast_page_as_ptr(PAGE_SIZE - ncopy),
+				reinterpret_cast<void*>(sp),
+				ncopy
+			);
+		});
 
 		const auto stack_index = new_process->find_mapped_region(m_userspace_stack->vaddr());
 		thread->m_userspace_stack = static_cast<MemoryBackedRegion*>(new_process->m_mapped_regions[stack_index].ptr());
@@ -572,19 +576,6 @@ namespace Kernel
 		return false;
 	}
 
-	bool Thread::can_add_signal_to_execute() const
-	{
-		return is_interrupted_by_signal() && m_mutex_count == 0;
-	}
-
-	bool Thread::will_execute_signal() const
-	{
-		if (!is_userspace() || m_state != State::Executing)
-			return false;
-		auto& interrupt_stack = *reinterpret_cast<InterruptStack*>(kernel_stack_top() - sizeof(InterruptStack));
-		return interrupt_stack.ip == (uintptr_t)signal_trampoline;
-	}
-
 	bool Thread::will_exit_because_of_signal() const
 	{
 		const uint64_t full_pending_mask = m_signal_pending_mask | process().signal_pending_mask();
@@ -596,21 +587,17 @@ namespace Kernel
 		return false;
 	}
 
-	bool Thread::handle_signal(int signal, const siginfo_t& _signal_info)
+	bool Thread::handle_signal_if_interrupted()
 	{
-		ASSERT(&Thread::current() == this);
-		ASSERT(is_userspace());
+		int signal;
+		siginfo_t signal_info;
+		signal_handle_info_t handle_info;
 
-		auto state = m_signal_lock.lock();
-
-		auto& interrupt_stack = *reinterpret_cast<InterruptStack*>(kernel_stack_top() - sizeof(InterruptStack));
-		ASSERT(GDT::is_user_segment(interrupt_stack.cs));
-
-		auto signal_info = _signal_info;
-
-		if (signal == 0)
 		{
-			const uint64_t process_signal_pending_mask = process().signal_pending_mask();
+			SpinLockGuard _1(m_signal_lock);
+			SpinLockGuard _2(m_process->m_signal_lock);
+
+			const uint64_t process_signal_pending_mask = m_process->m_signal_pending_mask;
 			const uint64_t full_pending_mask = m_signal_pending_mask | process_signal_pending_mask;
 			for (signal = _SIGMIN; signal <= _SIGMAX; signal++)
 			{
@@ -618,42 +605,90 @@ namespace Kernel
 				if ((full_pending_mask & mask) && !(m_signal_block_mask & mask))
 					break;
 			}
-			ASSERT(signal <= _SIGMAX);
+			if (signal > _SIGMAX)
+				return false;
+
 			if (process_signal_pending_mask & (1ull << signal))
-				signal_info = process().m_signal_infos[signal];
+				signal_info = m_process->m_signal_infos[signal];
 			else
 				signal_info = m_signal_infos[signal];
-		}
-		else
-		{
-			ASSERT(signal >= _SIGMIN);
-			ASSERT(signal <= _SIGMAX);
+			signal_info.si_signo = signal;
+
+			handle_info = remove_signal_and_get_info(signal);
 		}
 
-		vaddr_t signal_handler;
-		bool has_sa_restart;
+		handle_signal_impl(signal, signal_info, handle_info);
+
+		return handle_info.has_sa_restart;
+	}
+
+	bool Thread::handle_signal(int signal, const siginfo_t& signal_info)
+	{
+		ASSERT(&Thread::current() == this);
+		ASSERT(is_userspace());
+
+		ASSERT(signal >= _SIGMIN);
+		ASSERT(signal <= _SIGMAX);
+
+		signal_handle_info_t handle_info;
+
+		// If this signal is blocked or ignored, terminate the process
+		bool terminate_process = false;
+
+		{
+			SpinLockGuard _1(m_signal_lock);
+			SpinLockGuard _2(m_process->m_signal_lock);
+
+			if (m_signal_block_mask & (1ull << signal))
+				terminate_process = true;
+
+			handle_info = remove_signal_and_get_info(signal);
+
+			if (handle_info.handler == reinterpret_cast<vaddr_t>(SIG_IGN))
+				terminate_process = true;
+		}
+
+		if (terminate_process)
+		{
+			process().exit(128 + signal, signal | 0x80);
+			ASSERT_NOT_REACHED();
+		}
+
+		handle_signal_impl(signal, signal_info, handle_info);
+
+		return handle_info.has_sa_restart;
+	}
+
+	Thread::signal_handle_info_t Thread::remove_signal_and_get_info(int signal)
+	{
+		ASSERT(m_signal_lock.current_processor_has_lock());
+		ASSERT(m_process->m_signal_lock.current_processor_has_lock());
+
+		ASSERT(signal >= _SIGMIN);
+		ASSERT(signal <= _SIGMAX);
+
+		const uint64_t restore_sigmask = m_signal_block_mask;
+
+		auto& handler = m_process->m_signal_handlers[signal];
+
+		const vaddr_t signal_handler = (handler.sa_flags & SA_SIGINFO)
+			? reinterpret_cast<vaddr_t>(handler.sa_sigaction)
+			: reinterpret_cast<vaddr_t>(handler.sa_handler);
+		const bool has_sa_restart = !!(handler.sa_flags & SA_RESTART);
+
 		vaddr_t signal_stack_top = 0;
+		if (m_signal_alt_stack.ss_flags != SS_DISABLE && (handler.sa_flags & SA_ONSTACK) && !currently_on_alternate_stack())
+			signal_stack_top = reinterpret_cast<vaddr_t>(m_signal_alt_stack.ss_sp) + m_signal_alt_stack.ss_size;
 
-		{
-			SpinLockGuard _(m_process->m_signal_lock);
+		m_signal_block_mask |= handler.sa_mask;
+		if (!(handler.sa_flags & SA_NODEFER))
+			m_signal_block_mask |= 1ull << signal;
 
-			auto& handler = m_process->m_signal_handlers[signal];
-
-			signal_handler = (handler.sa_flags & SA_SIGINFO)
-				? reinterpret_cast<vaddr_t>(handler.sa_sigaction)
-				: reinterpret_cast<vaddr_t>(handler.sa_handler);
-			has_sa_restart = !!(handler.sa_flags & SA_RESTART);
-
-			const auto& alt_stack = m_signal_alt_stack;
-			if (alt_stack.ss_flags != SS_DISABLE && (handler.sa_flags & SA_ONSTACK) && !currently_on_alternate_stack())
-				signal_stack_top = reinterpret_cast<vaddr_t>(alt_stack.ss_sp) + alt_stack.ss_size;
-
-			if (handler.sa_flags & SA_RESETHAND)
-				handler = { .sa_handler = SIG_DFL, .sa_mask = 0, .sa_flags = 0 };
-		}
+		if (handler.sa_flags & SA_RESETHAND)
+			handler = { .sa_handler = SIG_DFL, .sa_mask = 0, .sa_flags = 0 };
 
 		m_signal_pending_mask &= ~(1ull << signal);
-		process().remove_pending_signal(signal);
+		m_process->m_signal_pending_mask &= ~(1ull << signal);
 
 		if (m_signal_suspend_mask.has_value())
 		{
@@ -661,66 +696,54 @@ namespace Kernel
 			m_signal_suspend_mask.clear();
 		}
 
-		m_signal_lock.unlock(state);
+		return {
+			.handler = signal_handler,
+			.stack_top = signal_stack_top,
+			.restore_sigmask = restore_sigmask,
+			.has_sa_restart = has_sa_restart,
+		};
+	}
 
-		if (signal_handler == (vaddr_t)SIG_IGN)
+	void Thread::handle_signal_impl(int signal, const siginfo_t& signal_info, const signal_handle_info_t& handle_info)
+	{
+		ASSERT(this == &Thread::current());
+		ASSERT(is_userspace());
+		ASSERT(Processor::get_interrupt_state() == InterruptState::Enabled);
+
+		auto& interrupt_stack = *reinterpret_cast<InterruptStack*>(kernel_stack_top() - sizeof(InterruptStack));
+		ASSERT(GDT::is_user_segment(interrupt_stack.cs));
+
+		if (handle_info.handler == reinterpret_cast<vaddr_t>(SIG_IGN))
 			;
-		else if (signal_handler != (vaddr_t)SIG_DFL)
+		else if (handle_info.handler != reinterpret_cast<vaddr_t>(SIG_DFL))
 		{
 			// call userspace signal handlers
 #if ARCH(x86_64)
 			interrupt_stack.sp -= 128; // skip possible red-zone
 #endif
 
-			{
-				// Make sure stack is allocated
-
-				vaddr_t pages[3] {};
-				size_t page_count { 0 };
-
-				if (signal_stack_top == 0)
+			const auto write_to_stack =
+				[&]<typename T>(uintptr_t& sp, const T& value)
 				{
-					pages[0] = (interrupt_stack.sp - 1 * sizeof(uintptr_t)                    ) & PAGE_ADDR_MASK;
-					pages[1] = (interrupt_stack.sp - 5 * sizeof(uintptr_t) - sizeof(siginfo_t)) & PAGE_ADDR_MASK;
-					page_count = 2;
-				}
-				else
-				{
-					pages[0] = (interrupt_stack.sp - 1 * sizeof(uintptr_t)                    ) & PAGE_ADDR_MASK;
-					pages[2] = (signal_stack_top   - 1 * sizeof(uintptr_t)                    ) & PAGE_ADDR_MASK;
-					pages[1] = (signal_stack_top   - 4 * sizeof(uintptr_t) - sizeof(siginfo_t)) & PAGE_ADDR_MASK;
-					page_count = 3;
-				}
-
-				for (size_t i = 0; i < page_count; i++)
-				{
-					if (m_process->page_table().get_page_flags(pages[i]) & PageTable::Flags::Present)
-						continue;
-					Processor::set_interrupt_state(InterruptState::Enabled);
-					if (auto ret = m_process->allocate_page_for_demand_paging(pages[i], true, false); ret.is_error() || !ret.value())
-						m_process->exit(128 + SIGSEGV, SIGSEGV);
-					Processor::set_interrupt_state(InterruptState::Disabled);
-				}
-			}
+					static_assert(sizeof(T) >= sizeof(uintptr_t));
+					sp -= sizeof(T);
+					if (m_process->write_to_user(reinterpret_cast<void*>(sp), &value, sizeof(T)).is_error())
+						m_process->exit(128 + SIGSEGV, SIGSEGV | 0x80);
+				};
 
 			write_to_stack(interrupt_stack.sp, interrupt_stack.ip);
 			const vaddr_t old_stack = interrupt_stack.sp;
-			if (signal_stack_top)
-				interrupt_stack.sp = signal_stack_top;
+			if (handle_info.stack_top)
+				interrupt_stack.sp = handle_info.stack_top;
 			write_to_stack(interrupt_stack.sp, old_stack);
 			write_to_stack(interrupt_stack.sp, interrupt_stack.flags);
+			write_to_stack(interrupt_stack.sp, handle_info.restore_sigmask);
 
-			{
-				signal_info.si_signo = signal;
-				signal_info.si_addr = reinterpret_cast<void*>(interrupt_stack.ip);
+			ASSERT(signal_info.si_signo == signal);
+			write_to_stack(interrupt_stack.sp, signal_info);
 
-				interrupt_stack.sp -= sizeof(siginfo_t);
-				*reinterpret_cast<siginfo_t*>(interrupt_stack.sp) = signal_info;
-				static_assert(sizeof(siginfo_t) % sizeof(uintptr_t) == 0);
-			}
-
-			write_to_stack(interrupt_stack.sp, signal);
-			write_to_stack(interrupt_stack.sp, signal_handler);
+			write_to_stack(interrupt_stack.sp, static_cast<uintptr_t>(signal));
+			write_to_stack(interrupt_stack.sp, handle_info.handler);
 			interrupt_stack.ip = (uintptr_t)signal_trampoline;
 		}
 		else
@@ -752,8 +775,6 @@ namespace Kernel
 				panic("Executing unhandled signal {}", signal);
 			}
 		}
-
-		return has_sa_restart;
 	}
 
 	void Thread::add_signal(int signal, const siginfo_t& info)

kernel/kernel/ThreadBlocker.cpp

@@ -23,60 +23,64 @@ namespace Kernel
 
 	void ThreadBlocker::unblock()
 	{
-		decltype(m_block_chain) temp_block_chain;
-		size_t temp_block_chain_length { 0 };
+		SpinLockGuard _(m_lock);
 
+		for (auto* node = m_block_chain; node;)
 		{
-			SpinLockGuard _(m_lock);
-			for (size_t i = 0; i < m_block_chain_length; i++)
-				temp_block_chain[i] = m_block_chain[i];
-			temp_block_chain_length = m_block_chain_length;
-			m_block_chain_length = 0;
+			auto* next = node->block_chain_next;
+
+			ASSERT(node->blocked);
+			ASSERT(node->blocker == this);
+
+			node->blocker.store(nullptr);
+			node->block_chain_prev = nullptr;
+			node->block_chain_next = nullptr;
+
+			Processor::scheduler().unblock_thread(node);
+
+			node = next;
 		}
 
-		for (size_t i = 0; i < temp_block_chain_length; i++)
-			Processor::scheduler().unblock_thread(temp_block_chain[i]);
+		m_block_chain = nullptr;
 	}
 
 	void ThreadBlocker::add_thread_to_block_queue(SchedulerQueue::Node* node)
 	{
-		ASSERT(node->blocker_lock.current_processor_has_lock());
-
 		SpinLockGuard _(m_lock);
 
-		ASSERT(m_block_chain_length < sizeof(m_block_chain) / sizeof(m_block_chain[0]));
-
-		ASSERT(node);
 		ASSERT(node->blocked);
 		ASSERT(node->blocker == nullptr);
 
-		for (size_t i = 0 ; i < m_block_chain_length; i++)
-			ASSERT(m_block_chain[i] != node);
-		m_block_chain[m_block_chain_length++] = node;
+		node->blocker.store(this);
+		node->block_chain_prev = nullptr;
+		node->block_chain_next = m_block_chain;
 
-		node->blocker = this;
+		if (m_block_chain)
+			m_block_chain->block_chain_prev = node;
+		m_block_chain = node;
 	}
 
-	void ThreadBlocker::remove_blocked_thread(SchedulerQueue::Node* node)
+	void ThreadBlocker::remove_thread_from_block_queue(SchedulerQueue::Node* node)
 	{
-		ASSERT(node->blocker_lock.current_processor_has_lock());
-
 		SpinLockGuard _(m_lock);
 
-		ASSERT(node);
+		// NOTE: this is possible if we got here while another
+		//       core was doing an unblock on this blocker
+		if (node->blocker.load() != this)
+			return;
 		ASSERT(node->blocked);
-		ASSERT(node->blocker == this);
 
-		for (size_t i = 0 ; i < m_block_chain_length; i++)
-		{
-			if (m_block_chain[i] != node)
-				continue;
-			for (size_t j = i + 1; j < m_block_chain_length; j++)
-				m_block_chain[j - 1] = m_block_chain[j];
-			m_block_chain_length--;
-		}
+		if (node->block_chain_prev)
+			node->block_chain_prev->block_chain_next = node->block_chain_next;
+		if (node->block_chain_next)
+			node->block_chain_next->block_chain_prev = node->block_chain_prev;
 
-		node->blocker = nullptr;
+		if (node == m_block_chain)
+			m_block_chain = node->block_chain_next;
+
+		node->blocker.store(nullptr);
+		node->block_chain_prev = nullptr;
+		node->block_chain_next = nullptr;
 	}
 
 }

kernel/kernel/USB/HID/Keyboard.cpp

@@ -342,6 +342,9 @@ namespace Kernel
 		s_scancode_to_keycode[0x4B] = keycode_function(18);
 		s_scancode_to_keycode[0x4E] = keycode_function(19);
 		s_scancode_to_keycode[0x47] = keycode_function(20);
+		s_scancode_to_keycode[0x7F] = keycode_function(21);
+		s_scancode_to_keycode[0x81] = keycode_function(22);
+		s_scancode_to_keycode[0x80] = keycode_function(23);
 
 		s_scancode_to_keycode[0x53] = keycode_numpad(0, 0);
 		s_scancode_to_keycode[0x54] = keycode_numpad(0, 1);

kernel/kernel/kernel.cpp

@@ -126,19 +126,20 @@ extern "C" void kernel_main(uint32_t boot_magic, uint32_t boot_info)
 	parse_boot_info(boot_magic, boot_info);
 	dprintln("boot info parsed");
 
-	Processor::create(PROCESSOR_NONE);
+	auto& processor = Processor::create(PROCESSOR_NONE);
 	Processor::initialize();
 	dprintln("BSP initialized");
 
-	PageTable::initialize_pre_heap();
-	PageTable::kernel().initial_load();
-	dprintln("PageTable stage1 initialized");
+	PageTable::initialize_fast_page();
+	dprintln("fast page initialized");
 
 	Heap::initialize();
-	dprintln("Heap initialzed");
+	dprintln("Heap initialized");
 
-	PageTable::initialize_post_heap();
-	dprintln("PageTable stage2 initialized");
+	PageTable::initialize_and_load();
+	dprintln("PageTable initialized");
+
+	processor.allocate_stack();
 
 	parse_command_line();
 	dprintln("command line parsed, root='{}', console='{}'", cmdline.root, cmdline.console);
@@ -255,8 +256,8 @@ static void init2(void*)
 	VirtualFileSystem::initialize(cmdline.root);
 	dprintln("VFS initialized");
 
-	// FIXME: release memory used by modules. If modules are used
-	//        they are already loaded in here
+	// NOTE: All modules should be loaded
+	Heap::get().release_boot_modules();
 
 	TTY::initialize_devices();
 
@@ -270,7 +271,7 @@ extern "C" void ap_main()
 	using namespace Kernel;
 
 	Processor::initialize();
-	PageTable::kernel().initial_load();
+
 	InterruptController::get().enable();
 
 	Processor::wait_until_processors_ready();

kernel/klibc/arch/i686/string.S

@@ -0,0 +1,47 @@
+.align 16
+.global memcpy
+memcpy:
+	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
+	movl %edx, %eax
+	ret
+
+.align 16
+.global memmove
+memmove:
+	xchgl 4(%esp), %edi
+	xchgl 8(%esp), %esi
+	movl 12(%esp), %ecx
+	movl %edi, %edx
+	cmpl %edi, %esi
+	jb .memmove_slow
+	rep movsb
+ .memmove_done:
+	movl 4(%esp), %edi
+	movl 8(%esp), %esi
+	movl %edx, %eax
+	ret
+ .memmove_slow:
+	leal -1(%edi, %ecx), %edi
+	leal -1(%esi, %ecx), %esi
+	std
+	rep movsb
+	cld
+	jmp .memmove_done
+
+.align 16
+.global memset
+memset:
+	xchgl 4(%esp), %edi
+	movl  8(%esp), %eax
+	movl 12(%esp), %ecx
+	movl %edi, %edx
+	rep stosb
+	movl 4(%esp), %edi
+	movl %edx, %eax
+	ret

kernel/klibc/arch/x86_64/string.S

@@ -0,0 +1,31 @@
+.align 16
+.global memcpy
+memcpy:
+	movq %rdi, %rax
+	movq %rdx, %rcx
+	rep movsb
+	ret
+
+.align 16
+.global memmove
+memmove:
+	cmpq %rdi, %rsi
+	jae memcpy
+	movq %rdi, %rax
+	leaq -1(%rdi, %rdx), %rdi
+	leaq -1(%rsi, %rdx), %rsi
+	movq %rdx, %rcx
+	std
+	rep movsb
+	cld
+	ret
+
+.align 16
+.global memset
+memset:
+	movq %rdi, %r8
+	movb %sil, %al
+	movq %rdx, %rcx
+	rep stosb
+	movq %r8, %rax
+	ret

ports/SDL2/patches/0001-add-banan_os-support.patch

@@ -1,6 +1,6 @@
 diff -ruN SDL2-2.32.8/cmake/sdlplatform.cmake SDL2-2.32.8-banan_os/cmake/sdlplatform.cmake
 --- SDL2-2.32.8/cmake/sdlplatform.cmake	2024-08-14 13:35:43.000000000 +0300
-+++ SDL2-2.32.8-banan_os/cmake/sdlplatform.cmake	2026-01-07 19:04:34.332166371 +0200
++++ SDL2-2.32.8-banan_os/cmake/sdlplatform.cmake	2026-04-03 04:34:27.256800208 +0300
 @@ -28,6 +28,8 @@
        set(SDL_CMAKE_PLATFORM AIX)
      elseif(CMAKE_SYSTEM_NAME MATCHES "Minix.*")
@@ -12,7 +12,7 @@ diff -ruN SDL2-2.32.8/cmake/sdlplatform.cmake SDL2-2.32.8-banan_os/cmake/sdlplat
      endif()
 diff -ruN SDL2-2.32.8/CMakeLists.txt SDL2-2.32.8-banan_os/CMakeLists.txt
 --- SDL2-2.32.8/CMakeLists.txt	2025-06-03 02:00:39.000000000 +0300
-+++ SDL2-2.32.8-banan_os/CMakeLists.txt	2026-01-07 19:04:34.343116295 +0200
++++ SDL2-2.32.8-banan_os/CMakeLists.txt	2026-04-03 04:34:27.257159543 +0300
 @@ -14,7 +14,7 @@
    set(SDL2_SUBPROJECT ON)
  endif()
@@ -98,7 +98,7 @@ diff -ruN SDL2-2.32.8/CMakeLists.txt SDL2-2.32.8-banan_os/CMakeLists.txt
      file(GLOB MISC_SOURCES ${SDL2_SOURCE_DIR}/src/misc/riscos/*.c)
 diff -ruN SDL2-2.32.8/include/SDL_config.h.cmake SDL2-2.32.8-banan_os/include/SDL_config.h.cmake
 --- SDL2-2.32.8/include/SDL_config.h.cmake	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/include/SDL_config.h.cmake	2026-01-07 19:04:34.358682129 +0200
++++ SDL2-2.32.8-banan_os/include/SDL_config.h.cmake	2026-04-03 04:34:27.257563019 +0300
 @@ -307,6 +307,7 @@
  #cmakedefine SDL_AUDIO_DRIVER_FUSIONSOUND @SDL_AUDIO_DRIVER_FUSIONSOUND@
  #cmakedefine SDL_AUDIO_DRIVER_FUSIONSOUND_DYNAMIC @SDL_AUDIO_DRIVER_FUSIONSOUND_DYNAMIC@
@@ -125,7 +125,7 @@ diff -ruN SDL2-2.32.8/include/SDL_config.h.cmake SDL2-2.32.8-banan_os/include/SD
  #cmakedefine SDL_VIDEO_DRIVER_DIRECTFB @SDL_VIDEO_DRIVER_DIRECTFB@
 diff -ruN SDL2-2.32.8/include/SDL_platform.h SDL2-2.32.8-banan_os/include/SDL_platform.h
 --- SDL2-2.32.8/include/SDL_platform.h	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/include/SDL_platform.h	2026-01-07 19:04:34.370086235 +0200
++++ SDL2-2.32.8-banan_os/include/SDL_platform.h	2026-04-03 04:34:27.257711782 +0300
 @@ -36,6 +36,10 @@
  #undef __HAIKU__
  #define __HAIKU__   1
@@ -139,8 +139,8 @@ diff -ruN SDL2-2.32.8/include/SDL_platform.h SDL2-2.32.8-banan_os/include/SDL_pl
  #define __BSDI__    1
 diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.cpp
 --- SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.cpp	2026-01-07 19:04:34.370691623 +0200
-@@ -0,0 +1,150 @@
++++ SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.cpp	2026-04-03 16:06:27.541819001 +0300
+@@ -0,0 +1,134 @@
 +/*
 +  Simple DirectMedia Layer
 +  Copyright (C) 1997-2019 Sam Lantinga <slouken@libsdl.org>
@@ -154,11 +154,11 @@ diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp SDL2-2.32.8-bana
 +  freely, subject to the following restrictions:
 +
 +  1. The origin of this software must not be misrepresented; you must not
-+     claim that you wrote the original software. If you use this software
-+     in a product, an acknowledgment in the product documentation would be
-+     appreciated but is not required.
++	 claim that you wrote the original software. If you use this software
++	 in a product, an acknowledgment in the product documentation would be
++	 appreciated but is not required.
 +  2. Altered source versions must be plainly marked as such, and must not be
-+     misrepresented as being the original software.
++	 misrepresented as being the original software.
 +  3. This notice may not be removed or altered from any source distribution.
 +*/
 +#include "../../SDL_internal.h"
@@ -201,8 +201,8 @@ diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp SDL2-2.32.8-bana
 +	DUMP_FUNCTION();
 +
 +	// TODO: try to accept already existing spec
-+	_this->spec.freq = 44100;
-+	_this->spec.format = AUDIO_S16LSB;
++	_this->spec.freq = 48000;
++	_this->spec.format = AUDIO_F32LSB;
 +	_this->spec.channels = 2;
 +	_this->spec.samples = 2048;
 +	SDL_CalculateAudioSpec(&_this->spec);
@@ -229,38 +229,22 @@ diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp SDL2-2.32.8-bana
 +
 +	const bool should_play = SDL_AtomicGet(&_this->enabled) && !SDL_AtomicGet(&_this->paused);
 +	_this->hidden->audio.set_paused(!should_play);
-+    if (!should_play)
++	if (!should_play)
 +	{
 +		usleep(100);
 +		return;
 +	}
 +
-+	static_assert(BAN::is_same_v<LibAudio::AudioBuffer::sample_t, double>);
-+
-+    const auto convert_sample = [](int16_t input) {
-+        return (static_cast<double>(input) - BAN::numeric_limits<int16_t>::min()) / BAN::numeric_limits<int16_t>::max() * 2.0 - 1.0;
-+    };
-+
-+	const size_t input_samples = _this->spec.size / sizeof(int16_t);
-+	size_t samples_queued = 0;
-+
-+	const int16_t* mixbuf = static_cast<const int16_t*>(_this->hidden->mixbuf);
-+	while (samples_queued < input_samples)
++	auto sample_span = BAN::Span(
++		static_cast<const float*>(_this->hidden->mixbuf),
++		_this->spec.size / sizeof(float)
++	);
++	while (!sample_span.empty())
 +	{
-+		const size_t to_convert = BAN::Math::min(_this->hidden->conversion.size(), input_samples - samples_queued);
-+		for (size_t i = 0; i < to_convert; i++)
-+			_this->hidden->conversion[i] = convert_sample(mixbuf[samples_queued + i]);
-+
-+		auto sample_span = _this->hidden->conversion.span();
-+		while (!sample_span.empty())
-+		{
-+			const size_t queued = _this->hidden->audio.queue_samples(sample_span);
-+			if (queued == 0)
-+				usleep(100);
-+			sample_span = sample_span.slice(queued);
-+		}
-+
-+		samples_queued += to_convert;
++		const size_t queued = _this->hidden->audio.queue_samples(sample_span);
++		if (queued == 0)
++			usleep(100);
++		sample_span = sample_span.slice(queued);
 +	}
 +}
 +
@@ -268,33 +252,33 @@ diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.cpp SDL2-2.32.8-bana
 +{
 +	DUMP_FUNCTION();
 +
-+    return static_cast<Uint8*>(_this->hidden->mixbuf);
++	return static_cast<Uint8*>(_this->hidden->mixbuf);
 +}
 +
 +static SDL_bool BANANOS_Init(SDL_AudioDriverImpl* impl)
 +{
-+    impl->OpenDevice = BANANOS_OpenDevice;
-+    impl->CloseDevice = BANANOS_CloseDevice;
-+    impl->PlayDevice = BANANOS_PlayDevice;
++	impl->OpenDevice = BANANOS_OpenDevice;
++	impl->CloseDevice = BANANOS_CloseDevice;
++	impl->PlayDevice = BANANOS_PlayDevice;
 +	impl->GetDeviceBuf = BANANOS_GetDeviceBuf;
 +
-+    impl->ProvidesOwnCallbackThread  = SDL_FALSE;
-+    impl->HasCaptureSupport          = SDL_FALSE;
-+    impl->OnlyHasDefaultOutputDevice = SDL_TRUE;
-+    impl->SupportsNonPow2Samples     = SDL_TRUE;
++	impl->ProvidesOwnCallbackThread  = SDL_FALSE;
++	impl->HasCaptureSupport          = SDL_FALSE;
++	impl->OnlyHasDefaultOutputDevice = SDL_TRUE;
++	impl->SupportsNonPow2Samples     = SDL_TRUE;
 +
-+    return SDL_TRUE;
++	return SDL_TRUE;
 +}
 +
 +AudioBootStrap BANANOSAUDIO_bootstrap = {
-+    "banan-os", "banan-os AudioServer", BANANOS_Init, SDL_FALSE
++	"banan-os", "banan-os AudioServer", BANANOS_Init, SDL_FALSE
 +};
 +
 +#endif
 diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.h SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.h
 --- SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.h	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.h	2026-01-07 19:04:34.370883199 +0200
-@@ -0,0 +1,34 @@
++++ SDL2-2.32.8-banan_os/src/audio/banan_os/SDL_banan_os_audio.h	2026-04-03 15:56:58.603415021 +0300
+@@ -0,0 +1,33 @@
 +/*
 +  Simple DirectMedia Layer
 +  Copyright (C) 1997-2019 Sam Lantinga <slouken@libsdl.org>
@@ -327,11 +311,10 @@ diff -ruN SDL2-2.32.8/src/audio/banan_os/SDL_banan_os_audio.h SDL2-2.32.8-banan_
 +struct SDL_PrivateAudioData {
 +	LibAudio::Audio audio;
 +	void* mixbuf { nullptr };
-+	BAN::Array<LibAudio::AudioBuffer::sample_t, 4096> conversion;
 +};
 diff -ruN SDL2-2.32.8/src/audio/SDL_audio.c SDL2-2.32.8-banan_os/src/audio/SDL_audio.c
 --- SDL2-2.32.8/src/audio/SDL_audio.c	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/audio/SDL_audio.c	2026-01-07 19:04:34.371410923 +0200
++++ SDL2-2.32.8-banan_os/src/audio/SDL_audio.c	2026-04-03 04:34:27.258080476 +0300
 @@ -87,6 +87,9 @@
  #ifdef SDL_AUDIO_DRIVER_HAIKU
      &HAIKUAUDIO_bootstrap,
@@ -344,7 +327,7 @@ diff -ruN SDL2-2.32.8/src/audio/SDL_audio.c SDL2-2.32.8-banan_os/src/audio/SDL_a
  #endif
 diff -ruN SDL2-2.32.8/src/audio/SDL_sysaudio.h SDL2-2.32.8-banan_os/src/audio/SDL_sysaudio.h
 --- SDL2-2.32.8/src/audio/SDL_sysaudio.h	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/audio/SDL_sysaudio.h	2026-01-07 19:04:34.372150756 +0200
++++ SDL2-2.32.8-banan_os/src/audio/SDL_sysaudio.h	2026-04-03 04:34:27.258278128 +0300
 @@ -196,6 +196,7 @@
  extern AudioBootStrap WINMM_bootstrap;
  extern AudioBootStrap PAUDIO_bootstrap;
@@ -355,7 +338,7 @@ diff -ruN SDL2-2.32.8/src/audio/SDL_sysaudio.h SDL2-2.32.8-banan_os/src/audio/SD
  extern AudioBootStrap DUMMYAUDIO_bootstrap;
 diff -ruN SDL2-2.32.8/src/joystick/banan_os/SDL_banan_os_joystick.cpp SDL2-2.32.8-banan_os/src/joystick/banan_os/SDL_banan_os_joystick.cpp
 --- SDL2-2.32.8/src/joystick/banan_os/SDL_banan_os_joystick.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/joystick/banan_os/SDL_banan_os_joystick.cpp	2026-01-07 19:07:12.677617077 +0200
++++ SDL2-2.32.8-banan_os/src/joystick/banan_os/SDL_banan_os_joystick.cpp	2026-04-03 04:34:27.258447075 +0300
 @@ -0,0 +1,296 @@
 +/*
 +Simple DirectMedia Layer
@@ -655,7 +638,7 @@ diff -ruN SDL2-2.32.8/src/joystick/banan_os/SDL_banan_os_joystick.cpp SDL2-2.32.
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/joystick/SDL_joystick.c SDL2-2.32.8-banan_os/src/joystick/SDL_joystick.c
 --- SDL2-2.32.8/src/joystick/SDL_joystick.c	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/joystick/SDL_joystick.c	2026-01-07 19:04:34.373890653 +0200
++++ SDL2-2.32.8-banan_os/src/joystick/SDL_joystick.c	2026-04-03 04:34:27.258656321 +0300
 @@ -85,6 +85,9 @@
  #ifdef SDL_JOYSTICK_HAIKU
      &SDL_HAIKU_JoystickDriver,
@@ -668,7 +651,7 @@ diff -ruN SDL2-2.32.8/src/joystick/SDL_joystick.c SDL2-2.32.8-banan_os/src/joyst
  #endif
 diff -ruN SDL2-2.32.8/src/joystick/SDL_sysjoystick.h SDL2-2.32.8-banan_os/src/joystick/SDL_sysjoystick.h
 --- SDL2-2.32.8/src/joystick/SDL_sysjoystick.h	2025-01-01 17:47:53.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/joystick/SDL_sysjoystick.h	2026-01-07 19:04:34.374337431 +0200
++++ SDL2-2.32.8-banan_os/src/joystick/SDL_sysjoystick.h	2026-04-03 04:34:27.259001479 +0300
 @@ -235,6 +235,7 @@
  
  /* The available joystick drivers */
@@ -679,7 +662,7 @@ diff -ruN SDL2-2.32.8/src/joystick/SDL_sysjoystick.h SDL2-2.32.8-banan_os/src/jo
  extern SDL_JoystickDriver SDL_DUMMY_JoystickDriver;
 diff -ruN SDL2-2.32.8/src/misc/banan_os/SDL_sysurl.cpp SDL2-2.32.8-banan_os/src/misc/banan_os/SDL_sysurl.cpp
 --- SDL2-2.32.8/src/misc/banan_os/SDL_sysurl.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/misc/banan_os/SDL_sysurl.cpp	2026-01-07 19:04:34.379748697 +0200
++++ SDL2-2.32.8-banan_os/src/misc/banan_os/SDL_sysurl.cpp	2026-04-03 04:34:27.259173778 +0300
 @@ -0,0 +1,30 @@
 +/*
 +  Simple DirectMedia Layer
@@ -713,7 +696,7 @@ diff -ruN SDL2-2.32.8/src/misc/banan_os/SDL_sysurl.cpp SDL2-2.32.8-banan_os/src/
 +
 diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.cpp SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.cpp
 --- SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.cpp	2026-01-07 19:04:34.379995308 +0200
++++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.cpp	2026-04-03 04:34:27.259271557 +0300
 @@ -0,0 +1,51 @@
 +/*
 +  Simple DirectMedia Layer
@@ -768,7 +751,7 @@ diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.cpp SDL2-2.32.8-
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.h SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.h
 --- SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.h	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.h	2026-01-07 19:04:34.380137576 +0200
++++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_clipboard.h	2026-04-03 04:34:27.259318700 +0300
 @@ -0,0 +1,43 @@
 +/*
 +  Simple DirectMedia Layer
@@ -815,7 +798,7 @@ diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_clipboard.h SDL2-2.32.8-ba
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.cpp SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.cpp
 --- SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.cpp	2026-01-07 19:04:34.380308339 +0200
++++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.cpp	2026-04-03 04:34:27.259375481 +0300
 @@ -0,0 +1,60 @@
 +/*
 +  Simple DirectMedia Layer
@@ -879,7 +862,7 @@ diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.cpp SDL2-2.32.
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.h SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.h
 --- SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.h	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.h	2026-01-07 19:04:34.380550899 +0200
++++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_message_box.h	2026-04-03 04:34:27.259437082 +0300
 @@ -0,0 +1,45 @@
 +/*
 +  Simple DirectMedia Layer
@@ -928,7 +911,7 @@ diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_message_box.h SDL2-2.32.8-
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_video.cpp SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_video.cpp
 --- SDL2-2.32.8/src/video/banan_os/SDL_banan_os_video.cpp	1970-01-01 02:00:00.000000000 +0200
-+++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_video.cpp	2026-01-07 19:04:34.380720824 +0200
++++ SDL2-2.32.8-banan_os/src/video/banan_os/SDL_banan_os_video.cpp	2026-04-03 04:34:27.259503082 +0300
 @@ -0,0 +1,729 @@
 +/*
 +  Simple DirectMedia Layer
@@ -1661,7 +1644,7 @@ diff -ruN SDL2-2.32.8/src/video/banan_os/SDL_banan_os_video.cpp SDL2-2.32.8-bana
 +/* vi: set ts=4 sw=4 expandtab: */
 diff -ruN SDL2-2.32.8/src/video/SDL_sysvideo.h SDL2-2.32.8-banan_os/src/video/SDL_sysvideo.h
 --- SDL2-2.32.8/src/video/SDL_sysvideo.h	2025-05-20 00:24:41.000000000 +0300
-+++ SDL2-2.32.8-banan_os/src/video/SDL_sysvideo.h	2026-01-07 19:04:34.381316574 +0200
++++ SDL2-2.32.8-banan_os/src/video/SDL_sysvideo.h	2026-04-03 04:34:27.259743826 +0300
 @@ -462,6 +462,7 @@
  extern VideoBootStrap WINDOWS_bootstrap;
  extern VideoBootStrap WINRT_bootstrap;
@@ -1672,7 +1655,7 @@ diff -ruN SDL2-2.32.8/src/video/SDL_sysvideo.h SDL2-2.32.8-banan_os/src/video/SD
  extern VideoBootStrap Android_bootstrap;
 diff -ruN SDL2-2.32.8/src/video/SDL_video.c SDL2-2.32.8-banan_os/src/video/SDL_video.c
 --- SDL2-2.32.8/src/video/SDL_video.c	2025-05-20 00:24:41.000000000 +0300
-+++ SDL2-2.32.8-banan_os/src/video/SDL_video.c	2026-01-07 19:04:34.398132645 +0200
++++ SDL2-2.32.8-banan_os/src/video/SDL_video.c	2026-04-03 04:34:27.260110007 +0300
 @@ -96,6 +96,9 @@
  #ifdef SDL_VIDEO_DRIVER_HAIKU
      &HAIKU_bootstrap,

ports/bzip2/build.sh

@@ -0,0 +1,36 @@
+#!/bin/bash ../install.sh
+
+NAME='bzip2'
+VERSION='1.0.8'
+DOWNLOAD_URL="https://sourceware.org/pub/bzip2/bzip2-$VERSION.tar.gz#ab5a03176ee106d3f0fa90e381da478ddae405918153cca248e682cd0c4a2269"
+
+configure() {
+	:
+}
+
+build() {
+	make -j$(nproc) -f Makefile-libbz2_so CC="$CC" || exit 1
+}
+
+install() {
+	cp -v libbz2.so.$VERSION $BANAN_SYSROOT/usr/lib/ || exit 1
+	ln -svf libbz2.so.$VERSION $BANAN_SYSROOT/usr/lib/libbz2.so || exit 1
+	ln -svf libbz2.so.$VERSION $BANAN_SYSROOT/usr/lib/libbz2.so.1 || exit 1
+	ln -svf libbz2.so.$VERSION $BANAN_SYSROOT/usr/lib/libbz2.so.1.0 || exit 1
+
+	cp -v bzlib.h $BANAN_SYSROOT/usr/include/ || exit 1
+
+	cat > $BANAN_SYSROOT/usr/lib/pkgconfig/bzip2.pc << EOF
+prefix=/usr
+exec_prefix=\${prefix}
+bindir=\${exec_prefix}/bin
+libdir=\${exec_prefix}/lib
+includedir=\${prefix}/include
+
+Name: bzip2
+Description: A file compression library
+Version: $VERSION
+Libs: -L\${libdir} -lbz2
+Cflags: -I\${includedir}
+EOF
+}

ports/ca-certificates/build.sh

@@ -1,8 +1,8 @@
 #!/bin/bash ../install.sh
 
 NAME='ca-certificates'
-VERSION='2025-12-02'
-DOWNLOAD_URL="https://curl.se/ca/cacert-$VERSION.pem#f1407d974c5ed87d544bd931a278232e13925177e239fca370619aba63c757b4"
+VERSION='2026.03.19'
+DOWNLOAD_URL="https://curl.se/ca/cacert-${VERSION//./-}.pem#b6e66569cc3d438dd5abe514d0df50005d570bfc96c14dca8f768d020cb96171"
 
 configure() {
 	:
@@ -13,7 +13,10 @@ build() {
 }
 
 install() {
-	mkdir -p "$BANAN_SYSROOT/etc/ssl/certs"
-	cp -v "../cacert-$VERSION.pem" "$BANAN_SYSROOT/etc/ssl/certs/ca-certificates.crt"
-	ln -svf "certs/ca-certificates.crt" "$BANAN_SYSROOT/etc/ssl/cert.pem"
+	rm -rf "$BANAN_SYSROOT/etc/cacert/extracted"
+	mkdir -p "$BANAN_SYSROOT/etc/cacert/extracted"
+
+	cp -vf "../cacert-${VERSION//./-}.pem" "$BANAN_SYSROOT/etc/cacert/cacert.pem"
+	awk '/-----BEGIN CERTIFICATE-----/ {c=1;n++} c {print > sprintf("cert%03d.pem", n)} /-----END CERTIFICATE-----/ {c=0}' "../cacert-${VERSION//./-}.pem"
+	mv cert*.pem "$BANAN_SYSROOT/etc/cacert/extracted/"
 }

ports/curl/build.sh

@@ -3,7 +3,7 @@
 NAME='curl'
 VERSION='8.17.0'
 DOWNLOAD_URL="https://curl.se/download/curl-$VERSION.tar.xz#955f6e729ad6b3566260e8fef68620e76ba3c31acf0a18524416a185acf77992"
-DEPENDENCIES=('ca-certificates' 'openssl' 'zlib' 'zstd')
+DEPENDENCIES=('openssl' 'zlib' 'zstd')
 CONFIG_SUB=('config.sub')
 CONFIGURE_OPTIONS=(
 	'--disable-threaded-resolver'
@@ -16,6 +16,6 @@ CONFIGURE_OPTIONS=(
 	'--with-zlib'
 	'--with-zstd'
 	'--without-libpsl'
-	'--with-ca-bundle=/etc/ssl/certs/ca-certificates.crt'
-	'--without-ca-path'
+	'--with-ca-path=/etc/ssl/certs'
+	'--with-ca-bundle=/etc/ssl/certs/ca-bundle.crt'
 )

ports/libarchive/build.sh

@@ -0,0 +1,23 @@
+#!/bin/bash ../install.sh
+
+NAME='libarchive'
+VERSION='3.8.6'
+DOWNLOAD_URL="https://github.com/libarchive/libarchive/releases/download/v$VERSION/libarchive-$VERSION.tar.xz#8ac57c1f5e99550948d1fe755c806d26026e71827da228f36bef24527e372e6f"
+DEPENDENCIES=('zlib' 'zstd' 'bzip2' 'xz')
+
+configure() {
+	cmake --fresh -B build -S . -G Ninja  \
+		--toolchain="$BANAN_TOOLCHAIN_DIR/Toolchain.txt" \
+		-DCMAKE_INSTALL_PREFIX=/usr \
+		-DCMAKE_BUILD_TYPE=Release \
+		-DENABLE_TEST=OFF \
+		|| exit 1
+}
+
+build() {
+	cmake --build build || exit 1
+}
+
+install() {
+	cmake --install build || exit 1
+}

ports/openssl/build.sh

@@ -3,9 +3,24 @@
 NAME='openssl'
 VERSION='3.6.0'
 DOWNLOAD_URL="https://github.com/openssl/openssl/releases/download/openssl-$VERSION/openssl-$VERSION.tar.gz#b6a5f44b7eb69e3fa35dbf15524405b44837a481d43d81daddde3ff21fcbb8e9"
-DEPENDENCIES=('zlib')
+DEPENDENCIES=('ca-certificates' 'zlib')
 MAKE_INSTALL_TARGETS=('install_sw' 'install_ssldirs')
 
 configure() {
 	./Configure --prefix=/usr --openssldir=/etc/ssl -DOPENSSL_USE_IPV6=0 no-asm no-tests banan_os-generic threads zlib
 }
+
+post_install() {
+	rm -f "$BANAN_SYSROOT/etc/ssl/certs"/*
+
+	ln -svf "../cacert/cacert.pem" "$BANAN_SYSROOT/etc/ssl/cert.pem"
+	ln -svf "../../cacert/cacert.pem" "$BANAN_SYSROOT/etc/ssl/certs/ca-certificates.crt"
+	ln -svf "../../cacert/cacert.pem" "$BANAN_SYSROOT/etc/ssl/certs/ca-bundle.crt"
+
+	openssl rehash "$BANAN_SYSROOT/etc/cacert/extracted"
+	find "$BANAN_SYSROOT/etc/cacert/extracted" -type l -print0 |
+	while IFS= read -r -d '' link; do
+		ln -s "../../cacert/extracted/$(readlink "$link")" "$BANAN_SYSROOT/etc/ssl/certs/${link##*/}"
+		rm "$link"
+	done
+}

ports/xbanan/build.sh

@@ -2,8 +2,8 @@
 
 NAME='xbanan'
 VERSION='git'
-DOWNLOAD_URL="https://git.bananymous.com/Bananymous/xbanan.git#b228ef13c41adff2738acaeda5db804ebf493bfd"
-DEPENDENCIES=('mesa' 'libX11' 'xorgproto')
+DOWNLOAD_URL="https://git.bananymous.com/Bananymous/xbanan.git#b2c642f03d2e498e9d6acd55cc89a5e76c220811"
+DEPENDENCIES=('xorgproto')
 
 configure() {
 	cmake --fresh -B build -S . -G Ninja \