Kernel: Don't use peridic interrupts for scheduling

Scheduler will now set a timer interrupt deadline for when it should get
interrupted next. This fixes a big issue I've had for a long time where
sleeping on an idle core would have to wait for the next periodic
interrupt to wake up. This could cause 1 ms sleeps to be close to 10 ms.

This currently only supports lapic timers and will still fallback to
periodic interrupts when running with PIC. I should add deadline support
to PIT/HPET but why would you run with APIC disabled ;)

BAN/include/BAN/Sort.h

@@ -67,11 +67,20 @@ namespace BAN::sort
 	template<typename It, typename Comp = less<it_value_type_t<It>>>
 	void quick_sort(It begin, It end, Comp comp = {})
 	{
-		if (distance(begin, end) <= 1)
-			return;
-		const auto [lt, gt] = detail::partition(begin, end, comp);
-		quick_sort(begin, lt, comp);
-		quick_sort(gt, end, comp);
+		while (distance(begin, end) > 1)
+		{
+			const auto [lt, gt] = detail::partition(begin, end, comp);
+			if (distance(begin, lt) < distance(gt, end))
+			{
+				quick_sort(begin, lt);
+				begin = gt;
+			}
+			else
+			{
+				quick_sort(gt, end);
+				end = lt;
+			}
+		}
 	}
 
 	template<typename It, typename Comp = less<it_value_type_t<It>>>
@@ -102,13 +111,28 @@ namespace BAN::sort
 		template<typename It, typename Comp>
 		void intro_sort_impl(It begin, It end, size_t max_depth, Comp comp)
 		{
-			if (distance(begin, end) <= 16)
-				return insertion_sort(begin, end, comp);
-			if (max_depth == 0)
-				return heap_sort(begin, end, comp);
-			const auto [lt, gt] = detail::partition(begin, end, comp);
-			intro_sort_impl(begin, lt, max_depth - 1, comp);
-			intro_sort_impl(gt, end, max_depth - 1, comp);
+			for (size_t i = 0; i < max_depth; i++)
+			{
+				const size_t len = distance(begin, end);
+				if (len <= 1)
+					return;
+				if (len <= 16)
+					return insertion_sort(begin, end, comp);
+
+				const auto [lt, gt] = detail::partition(begin, end, comp);
+				if (distance(begin, lt) < distance(gt, end))
+				{
+					intro_sort_impl(begin, lt, max_depth - 1, comp);
+					begin = gt;
+				}
+				else
+				{
+					intro_sort_impl(gt, end, max_depth - 1, comp);
+					end = lt;
+				}
+			}
+
+			heap_sort(begin, end, comp);
 		}
 
 	}

BAN/include/BAN/Span.h

@@ -100,3 +100,22 @@ namespace BAN
 	};
 
 }
+
+#include <BAN/Formatter.h>
+
+namespace BAN::Formatter
+{
+
+	template<typename F, typename T>
+	void print_argument(F putc, const Span<T>& span, const ValueFormat& format)
+	{
+		putc('[');
+		for (typename Span<T>::size_type i = 0; i < span.size(); i++)
+		{
+			if (i != 0) putc(',');
+			print_argument(putc, span[i], format);
+		}
+		putc(']');
+	}
+
+}

README.md

@@ -118,7 +118,7 @@ To change the bootloader you can set environment variable BANAN\_BOOTLOADER; sup
 
 To run with UEFI set environment variable BANAN\_UEFI\_BOOT=1. You will also have to set OVMF\_PATH to the correct OVMF (default */usr/share/ovmf/x64/OVMF.fd*).
 
-To build an image with no physical root filesystem, but an initrd set environment variable BANAN\_INITRD=1. This can be used when testing on hardware with unsupported USB controller.
+To build an image with no physical root filesystem, but an initrd set environment variable BANAN\_INITRD=1. This can be used when testing on hardware with unsupported USB controller. If BANAN\_INITRD is set to a value larger than 1, initrd will be gzip compressed.
 
 If you have corrupted your disk image or want to create new one, you can either manually delete *build/banan-os.img* and build system will automatically create you a new one or you can run the following command.
 ```sh

kernel/arch/i686/Signal.S

@@ -59,7 +59,7 @@ signal_trampoline:
 	addl $24, %esp
 
 	// restore sigmask
-	movl $79, %eax      // SYS_SIGPROCMASK
+	movl $78, %eax      // SYS_SIGPROCMASK
 	movl $3,  %ebx      // SIG_SETMASK
 	leal 72(%esp), %ecx // set
 	xorl %edx, %edx     // oset

kernel/arch/i686/Yield.S

@@ -12,7 +12,7 @@ asm_yield_trampoline:
 	pushl %ebp
 
 	pushl %esp
-	call scheduler_on_yield
+	call scheduler_on_yield_trampoline
 	addl $4, %esp
 
 	popl %ebp

kernel/arch/i686/interrupts.S

@@ -27,14 +27,10 @@
 
 isr_stub:
 	intr_header 12
-	movl %cr0, %eax; pushl %eax
-	movl %cr2, %eax; pushl %eax
-	movl %cr3, %eax; pushl %eax
-	movl %cr4, %eax; pushl %eax
 
-	movl 48(%esp), %edi // isr number
-	movl 52(%esp), %esi // error code
-	leal 56(%esp), %edx // interrupt stack ptr
+	movl 32(%esp), %edi // isr number
+	movl 36(%esp), %esi // error code
+	leal 40(%esp), %edx // interrupt stack ptr
 	movl %esp,     %ecx // register ptr
 
 	# stack frame for stack trace
@@ -52,7 +48,7 @@ isr_stub:
 	call cpp_isr_handler
 
 	movl %ebp, %esp
-	addl $24, %esp
+	addl $8, %esp
 
 	intr_footer 12
 	addl $8, %esp

kernel/arch/x86_64/Signal.S

@@ -61,7 +61,7 @@ signal_trampoline:
 	addq $40, %rsp
 
 	// restore sigmask
-	movq $79, %rdi       // SYS_SIGPROCMASK
+	movq $78, %rdi       // SYS_SIGPROCMASK
 	movq $3,  %rsi       // SIG_SETMASK
 	leaq 192(%rsp), %rdx // set
 	xorq %r10, %r10      // oset

kernel/arch/x86_64/Yield.S

@@ -15,7 +15,7 @@ asm_yield_trampoline:
 	pushq %r15
 
 	movq %rsp, %rdi
-	call scheduler_on_yield
+	call scheduler_on_yield_trampoline
 
 	popq %r15
 	popq %r14

kernel/arch/x86_64/interrupts.S

@@ -44,17 +44,12 @@
 
 isr_stub:
 	intr_header 24
-	movq %cr0, %rax; pushq %rax
-	movq %cr2, %rax; pushq %rax
-	movq %cr3, %rax; pushq %rax
-	movq %cr4, %rax; pushq %rax
 
-	movq 152(%rsp), %rdi	// isr number
-	movq 160(%rsp), %rsi	// error code
-	leaq 168(%rsp), %rdx	// interrupt stack ptr
-	movq %rsp, %rcx			// register ptr
+	movq 120(%rsp), %rdi // isr number
+	movq 128(%rsp), %rsi // error code
+	leaq 136(%rsp), %rdx // interrupt stack ptr
+	movq %rsp, %rcx      // register ptr
 	call cpp_isr_handler
-	addq $32, %rsp
 
 	intr_footer 24
 	addq $16, %rsp

kernel/include/kernel/API/SharedPage.h

@@ -8,6 +8,7 @@ namespace Kernel::API
 	enum SharedPageFeature : uint32_t
 	{
 		SPF_GETTIME = 1 << 0,
+		SPF_RDTSCP  = 1 << 1,
 	};
 
 	struct SharedPage
@@ -18,9 +19,8 @@ namespace Kernel::API
 
 		struct
 		{
-			uint8_t shift;
-			uint64_t mult;
-			uint64_t realtime_seconds;
+			uint64_t realtime_s;
+			uint32_t realtime_ns;
 		} gettime_shared;
 
 		struct
@@ -28,6 +28,8 @@ namespace Kernel::API
 			struct
 			{
 				uint32_t seq;
+				uint32_t mult;
+				int8_t shift;
 				uint64_t last_ns;
 				uint64_t last_tsc;
 			} gettime_local;

kernel/include/kernel/APIC.h

@@ -26,6 +26,7 @@ namespace Kernel
 		BAN::ErrorOr<uint8_t> reserve_gsi(uint32_t gsi);
 
 		void initialize_timer();
+		void set_timer_dealine(uint64_t ns);
 
 	private:
 		uint32_t read_from_local_apic(ptrdiff_t);

kernel/include/kernel/Audio/Controller.h

@@ -38,7 +38,7 @@ namespace Kernel
 
 		BAN::ErrorOr<size_t> write_impl(off_t, BAN::ConstByteSpan) override;
 
-		BAN::ErrorOr<long> ioctl_impl(int cmd, void* arg) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long request, void* arg) override;
 
 	protected:
 		ThreadBlocker m_sample_data_blocker;

kernel/include/kernel/CPUID.h

@@ -82,5 +82,8 @@ namespace CPUID
 	bool has_pat();
 	bool has_1gib_pages();
 	bool has_invariant_tsc();
+	bool has_rdtscp();
+	uint64_t get_tsc_frequency();
+	bool has_kvm_pvclock();
 
 }

kernel/include/kernel/Device/FramebufferDevice.h

@@ -36,7 +36,7 @@ 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;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long request, void* arg) override;
 
 		virtual bool can_read_impl() const override { return true; }
 		virtual bool can_write_impl() const override { return true; }

kernel/include/kernel/ELF.h

@@ -14,8 +14,8 @@ namespace Kernel::ELF
 			size_t size;
 		};
 
-		bool open_execfd;
 		vaddr_t entry_point;
+		BAN::Optional<vaddr_t> interp_base;
 		BAN::Optional<TLS> master_tls;
 		BAN::Vector<BAN::UniqPtr<MemoryRegion>> regions;
 	};

kernel/include/kernel/FS/Inode.h

@@ -150,7 +150,7 @@ namespace Kernel
 		bool has_error() const  { return has_error_impl(); }
 		bool has_hungup() const { return has_hungup_impl(); }
 
-		BAN::ErrorOr<long> ioctl(int request, void* arg);
+		BAN::ErrorOr<long> ioctl(unsigned long request, void* arg);
 
 		BAN::ErrorOr<void> add_epoll(class Epoll*);
 		void del_epoll(class Epoll*);
@@ -199,7 +199,7 @@ namespace Kernel
 		virtual bool has_error_impl() const = 0;
 		virtual bool has_hungup_impl() const = 0;
 
-		virtual BAN::ErrorOr<long> ioctl_impl(int, void*) { return BAN::Error::from_errno(ENOTSUP); }
+		virtual BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) { return BAN::Error::from_errno(ENOTSUP); }
 
 	protected:
 		// TODO: this is supposed to be const I guess?

kernel/include/kernel/Networking/E1000/Definitions.h

@@ -29,6 +29,7 @@ namespace Kernel
 		REG_TDLEN	= 0x3808,
 		REG_TDH		= 0x3810,
 		REG_TDT		= 0x3818,
+		REG_RXCSUM	= 0x5000,
 		REG_MTA		= 0x5200,
 	};
 
@@ -167,6 +168,15 @@ namespace Kernel
 		RCTL_BSIZE_16384	= (0b01 << 16) | RCTL_BSEX,
 	};
 
+	enum E1000_RXCSUM : uint32_t
+	{
+		RXSUM_IPOFLD  = 1 << 8,
+		RXSUM_TUOFLD  = 1 << 9,
+		RXSUM_CRCOFLD = 1 << 11,
+		RXSUM_IPPCE   = 1 << 12,
+		RXSUM_PCSD    = 1 << 13,
+	};
+
 	enum E1000_CMD : uint8_t
 	{
 		CMD_EOP		= 1 << 0,
@@ -178,6 +188,26 @@ namespace Kernel
 		CMD_IDE		= 1 << 7,
 	};
 
+	enum E1000_RX_STS : uint8_t
+	{
+		RX_STS_IPCS  = 1 << 6,
+		RX_STS_TCPCS = 1 << 5,
+		RX_STS_UDPCS = 1 << 5,
+		RX_STS_EOP   = 1 << 1,
+		RX_STS_DD    = 1 << 0,
+	};
+
+	enum E1000_RX_ERR : uint8_t
+	{
+		RX_ERR_RXE  = 1 << 7,
+		RX_ERR_IPE  = 1 << 6,
+		RX_ERR_TCPE = 1 << 5,
+		RX_ERR_CXE  = 1 << 4,
+		RX_ERR_SEQ  = 1 << 2,
+		RX_ERR_SE   = 1 << 1,
+		RX_ERR_CE   = 1 << 0,
+	};
+
 	struct e1000_rx_desc
 	{
 		uint64_t addr;

kernel/include/kernel/Networking/E1000/E1000.h

@@ -45,7 +45,7 @@ namespace Kernel
 		uint32_t read32(uint16_t reg);
 		void write32(uint16_t reg, uint32_t value);
 
-		BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload) override;
+		BAN::ErrorOr<void> send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers) override;
 
 		bool can_read_impl() const override { return false; }
 		bool can_write_impl() const override { return false; }
@@ -74,8 +74,8 @@ namespace Kernel
 		BAN::UniqPtr<DMARegion>	m_rx_descriptor_region;
 		BAN::UniqPtr<DMARegion>	m_tx_descriptor_region;
 
-		BAN::Atomic<uint32_t> m_tx_head1 { 0 };
-		BAN::Atomic<uint32_t> m_tx_head2 { 0 };
+		BAN::Atomic<uint32_t> m_tx_head   { 0 };
+		BAN::Atomic<uint32_t> m_tx_commit { 0 };
 
 		SpinLock m_rx_lock;
 		ThreadBlocker m_rx_blocker;

kernel/include/kernel/Networking/IPv4Layer.h

@@ -41,14 +41,14 @@ namespace Kernel
 
 		ARPTable& arp_table() { return *m_arp_table; }
 
-		BAN::ErrorOr<void> handle_ipv4_packet(NetworkInterface&, BAN::ConstByteSpan);
+		BAN::ErrorOr<void> handle_ipv4_packet(NetworkInterface&, BAN::ConstByteSpan, uint32_t validated_cksums);
 
 		virtual void unbind_socket(uint16_t port) override;
 		virtual BAN::ErrorOr<void> bind_socket_with_target(BAN::RefPtr<NetworkSocket>, const sockaddr* target_address, socklen_t target_address_len) override;
 		virtual BAN::ErrorOr<void> bind_socket_to_address(BAN::RefPtr<NetworkSocket>, const sockaddr* address, socklen_t address_len) override;
 		virtual BAN::ErrorOr<void> get_socket_address(BAN::RefPtr<NetworkSocket>, sockaddr* address, socklen_t* address_len) override;
 
-		virtual BAN::ErrorOr<size_t> sendto(NetworkSocket&, BAN::ConstByteSpan, const sockaddr*, socklen_t) override;
+		virtual BAN::ErrorOr<void> sendto(NetworkSocket&, BAN::ConstByteSpan, const sockaddr*, socklen_t) override;
 
 		virtual Socket::Domain domain() const override { return Socket::Domain::INET ;}
 		virtual size_t header_size() const override { return sizeof(IPv4Header); }

kernel/include/kernel/Networking/Loopback.h

@@ -27,7 +27,7 @@ namespace Kernel
 		{}
 		~LoopbackInterface();
 
-		BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload) override;
+		BAN::ErrorOr<void> send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers) override;
 
 		bool can_read_impl() const override { return false; }
 		bool can_write_impl() const override { return false; }
@@ -46,7 +46,7 @@ namespace Kernel
 		};
 
 	private:
-		Mutex m_buffer_lock;
+		SpinLock m_buffer_lock;
 		BAN::UniqPtr<VirtualRange> m_buffer;
 
 		uint32_t m_buffer_tail { 0 };

kernel/include/kernel/Networking/NetworkInterface.h

@@ -23,6 +23,13 @@ namespace Kernel
 		ARP = 0x0806,
 	};
 
+	enum NetworkChecksum : uint32_t
+	{
+		CKSUM_IPV4 = 1 << 0,
+		CKSUM_TCP  = 1 << 1,
+		CKSUM_UDP  = 1 << 2,
+	};
+
 	class NetworkInterface : public CharacterDevice
 	{
 		BAN_NON_COPYABLE(NetworkInterface);
@@ -59,14 +66,33 @@ namespace Kernel
 
 		virtual BAN::StringView name() const override { return m_name; }
 
-		BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::ConstByteSpan payload)
+		BAN::ErrorOr<void> send_with_ethernet_header(BAN::MACAddress dst_mac, EtherType ether_type, BAN::ConstByteSpan buffer)
 		{
-			return send_bytes(destination, protocol, { &payload, 1 });
+			BAN::ConstByteSpan buffer_array[1] { buffer };
+			return send_with_ethernet_header(dst_mac, ether_type, buffer_array);
 		}
-		virtual BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload) = 0;
+
+		template<size_t SIZE>
+		BAN::ErrorOr<void> send_with_ethernet_header(BAN::MACAddress dst_mac, EtherType ether_type, const BAN::ConstByteSpan (&buffer_array)[SIZE])
+		{
+			const auto ethernet_header = EthernetHeader {
+				.dst_mac = dst_mac,
+				.src_mac = get_mac_address(),
+				.ether_type = ether_type,
+			};
+
+			BAN::ConstByteSpan new_buffer_array[SIZE + 1];
+			new_buffer_array[0] = BAN::ConstByteSpan::from(ethernet_header);
+			for (size_t i = 0; i < SIZE; i++)
+				new_buffer_array[i + 1] = buffer_array[i];
+
+			return send_raw_bytes({ new_buffer_array, SIZE + 1 });
+		}
+
+		virtual BAN::ErrorOr<void> send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers) = 0;
 
 	private:
-		BAN::ErrorOr<long> ioctl_impl(int, void*) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) override;
 
 	private:
 		const Type m_type;

kernel/include/kernel/Networking/NetworkLayer.h

@@ -27,7 +27,7 @@ namespace Kernel
 		virtual BAN::ErrorOr<void> bind_socket_to_address(BAN::RefPtr<NetworkSocket>, const sockaddr* address, socklen_t address_len) = 0;
 		virtual BAN::ErrorOr<void> get_socket_address(BAN::RefPtr<NetworkSocket>, sockaddr* address, socklen_t* address_len) = 0;
 
-		virtual BAN::ErrorOr<size_t> sendto(NetworkSocket&, BAN::ConstByteSpan, const sockaddr*, socklen_t) = 0;
+		virtual BAN::ErrorOr<void> sendto(NetworkSocket&, BAN::ConstByteSpan, const sockaddr*, socklen_t) = 0;
 
 		virtual Socket::Domain domain() const = 0;
 		virtual size_t header_size() const = 0;

kernel/include/kernel/Networking/NetworkManager.h

@@ -27,7 +27,7 @@ namespace Kernel
 		BAN::ErrorOr<BAN::RefPtr<Socket>> create_socket(Socket::Domain, Socket::Type, mode_t, uid_t, gid_t);
 		BAN::ErrorOr<void> connect_sockets(Socket::Domain, BAN::RefPtr<Socket>, BAN::RefPtr<Socket>);
 
-		void on_receive(NetworkInterface&, BAN::ConstByteSpan);
+		void on_receive(NetworkInterface&, BAN::ConstByteSpan, uint32_t validated_cksums);
 
 	private:
 		NetworkManager() {}

kernel/include/kernel/Networking/NetworkSocket.h

@@ -35,7 +35,7 @@ namespace Kernel
 		virtual void get_protocol_header(BAN::ByteSpan header, BAN::ConstByteSpan payload, uint16_t dst_port, PseudoHeader) = 0;
 		virtual NetworkProtocol protocol() const = 0;
 
-		virtual void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len) = 0;
+		virtual void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len, uint32_t validated_cksums) = 0;
 
 		bool is_bound() const { return m_address_len >= static_cast<socklen_t>(sizeof(sa_family_t)) && m_address.ss_family != AF_UNSPEC; }
 		in_port_t bound_port() const
@@ -54,7 +54,7 @@ namespace Kernel
 	protected:
 		NetworkSocket(NetworkLayer&, const Socket::Info&);
 
-		virtual BAN::ErrorOr<long> ioctl_impl(int request, void* arg) override;
+		virtual BAN::ErrorOr<long> ioctl_impl(unsigned long request, void* arg) override;
 		virtual BAN::ErrorOr<void> getsockname_impl(sockaddr*, socklen_t*) override;
 		virtual BAN::ErrorOr<void> getpeername_impl(sockaddr*, socklen_t*) override = 0;
 

kernel/include/kernel/Networking/RTL8169/Definitions.h

@@ -7,25 +7,25 @@ namespace Kernel
 
 	enum RTL8169_IO_REGS : uint16_t
 	{
-		RTL8169_IO_IDR0   = 0x00,
-		RTL8169_IO_IDR1   = 0x01,
-		RTL8169_IO_IDR2   = 0x02,
-		RTL8169_IO_IDR3   = 0x03,
-		RTL8169_IO_IDR4   = 0x04,
-		RTL8169_IO_IDR5   = 0x05,
+		RTL8169_IO_IDR0    = 0x00,
+		RTL8169_IO_IDR1    = 0x01,
+		RTL8169_IO_IDR2    = 0x02,
+		RTL8169_IO_IDR3    = 0x03,
+		RTL8169_IO_IDR4    = 0x04,
+		RTL8169_IO_IDR5    = 0x05,
 
-		RTL8169_IO_TNPDS  = 0x20,
-		RTL8169_IO_CR     = 0x37,
-		RTL8169_IO_TPPoll = 0x38,
-		RTL8169_IO_IMR    = 0x3C,
-		RTL8169_IO_ISR    = 0x3E,
-		RTL8169_IO_TCR    = 0x40,
-		RTL8169_IO_RCR    = 0x44,
-		RTL8169_IO_9346CR = 0x50,
-		RTL8169_IO_PHYSts = 0x6C,
-		RTL8169_IO_RMS    = 0xDA,
-		RTL8169_IO_RDSAR  = 0xE4,
-		RTL8169_IO_MTPS   = 0xEC,
+		RTL8169_IO_TNPDS   = 0x20,
+		RTL8169_IO_CR      = 0x37,
+		RTL8169_IO_TPPoll  = 0x38,
+		RTL8169_IO_IMR     = 0x3C,
+		RTL8169_IO_ISR     = 0x3E,
+		RTL8169_IO_TCR     = 0x40,
+		RTL8169_IO_RCR     = 0x44,
+		RTL8169_IO_PHYSts  = 0x6C,
+		RTL8169_IO_RMS     = 0xDA,
+		RTL8169_IO_CPlusCR = 0xE0,
+		RTL8169_IO_RDSAR   = 0xE4,
+		RTL8169_IO_MTPS    = 0xEC,
 	};
 
 	enum RTL8169_CR : uint8_t

kernel/include/kernel/Networking/RTL8169/RTL8169.h

@@ -20,7 +20,7 @@ namespace Kernel
 		virtual bool link_up() override { return m_link_up; }
 		virtual int link_speed() override;
 
-		virtual size_t payload_mtu() const override { return 7436 - sizeof(EthernetHeader); }
+		virtual size_t payload_mtu() const override { return 0x1FF8 - sizeof(EthernetHeader); }
 
 		virtual void handle_irq() override;
 
@@ -33,7 +33,7 @@ namespace Kernel
 
 		BAN::ErrorOr<void> initialize();
 
-		virtual BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan>) override;
+		virtual BAN::ErrorOr<void> send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers) override;
 
 		virtual bool can_read_impl() const override { return false; }
 		virtual bool can_write_impl() const override { return false; }
@@ -64,14 +64,17 @@ namespace Kernel
 		BAN::UniqPtr<DMARegion>	m_rx_descriptor_region;
 		BAN::UniqPtr<DMARegion>	m_tx_descriptor_region;
 
-		SpinLock m_lock;
+		BAN::Atomic<bool> m_rx_thread_should_die { false };
+		BAN::Atomic<bool> m_rx_thread_is_dead { true };
 
-		bool m_thread_should_die { false };
-		BAN::Atomic<bool> m_thread_is_dead { true };
-		ThreadBlocker m_thread_blocker;
+		uint32_t m_rx_head { 0 };
+		SpinLock m_rx_lock;
+		ThreadBlocker m_rx_blocker;
 
-		uint32_t m_rx_current { 0 };
-		size_t m_tx_current { 0 };
+		BAN::Atomic<uint32_t> m_tx_head { 0 };
+		BAN::Atomic<uint32_t> m_tx_commit { 0 };
+		SpinLock m_tx_lock;
+		ThreadBlocker m_tx_blocker;
 
 		BAN::MACAddress	m_mac_address {};
 		BAN::Atomic<bool> m_link_up { false };

kernel/include/kernel/Networking/TCPSocket.h

@@ -66,9 +66,9 @@ namespace Kernel
 		BAN::ErrorOr<void> getsockopt_impl(int, int, void*, socklen_t*) override;
 		BAN::ErrorOr<void> setsockopt_impl(int, int, const void*, socklen_t) override;
 
-		BAN::ErrorOr<long> ioctl_impl(int, void*) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) override;
 
-		void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len) override;
+		void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len, uint32_t validated_cksums) override;
 
 		bool can_read_impl() const override;
 		bool can_write_impl() const override;

kernel/include/kernel/Networking/UDPSocket.h

@@ -31,7 +31,7 @@ namespace Kernel
 		void get_protocol_header(BAN::ByteSpan header, BAN::ConstByteSpan payload, uint16_t dst_port, PseudoHeader) override;
 
 	protected:
-		void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len) override;
+		void receive_packet(BAN::ConstByteSpan, const sockaddr* sender, socklen_t sender_len, uint32_t validated_cksums) override;
 
 		BAN::ErrorOr<void> connect_impl(const sockaddr*, socklen_t) override;
 		BAN::ErrorOr<void> bind_impl(const sockaddr* address, socklen_t address_len) override;
@@ -41,7 +41,7 @@ namespace Kernel
 		BAN::ErrorOr<void> getsockopt_impl(int, int, void*, socklen_t*) override;
 		BAN::ErrorOr<void> setsockopt_impl(int, int, const void*, socklen_t) override;
 
-		BAN::ErrorOr<long> ioctl_impl(int, void*) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) override;
 
 		bool can_read_impl() const override { return !m_packets.empty(); }
 		bool can_write_impl() const override { return true; }

kernel/include/kernel/Networking/UNIX/Socket.h

@@ -30,6 +30,7 @@ namespace Kernel
 		virtual BAN::ErrorOr<void> bind_impl(const sockaddr*, socklen_t) override;
 		virtual BAN::ErrorOr<size_t> recvmsg_impl(msghdr& message, int flags) override;
 		virtual BAN::ErrorOr<size_t> sendmsg_impl(const msghdr& message, int flags) override;
+		virtual BAN::ErrorOr<void> getsockname_impl(sockaddr*, socklen_t*) override;
 		virtual BAN::ErrorOr<void> getpeername_impl(sockaddr*, socklen_t*) override;
 		virtual BAN::ErrorOr<void> getsockopt_impl(int, int, void*, socklen_t*) override;
 		virtual BAN::ErrorOr<void> setsockopt_impl(int, int, const void*, socklen_t) override;

kernel/include/kernel/PCI.h

@@ -4,7 +4,6 @@
 #include <BAN/Vector.h>
 #include <kernel/ACPI/AML/Node.h>
 #include <kernel/Memory/Types.h>
-#include <kernel/Storage/StorageController.h>
 
 #include <sys/types.h>
 
@@ -179,7 +178,7 @@ namespace Kernel::PCI
 	private:
 		struct PCIeInfo
 		{
-			paddr_t bus_paddr[256];
+			paddr_t bus_paddr[256] {};
 		};
 
 		PCIManager() = default;

kernel/include/kernel/Process.h

@@ -44,10 +44,10 @@ namespace Kernel
 		~Process();
 		void cleanup_function(Thread*);
 
-		void register_to_scheduler();
+		BAN::ErrorOr<vaddr_t> setup_initial_process_stack(MemoryBackedRegion&, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv);
+
 		void exit(int status, int signal);
 
-		void add_thread(Thread*);
 		// returns true if thread was the last one
 		bool on_thread_exit(Thread&);
 
@@ -68,7 +68,6 @@ namespace Kernel
 		BAN::ErrorOr<long> sys_exec(const char* path, const char* const* argv, const char* const* envp);
 
 		BAN::ErrorOr<long> sys_wait(pid_t pid, int* stat_loc, int options);
-		BAN::ErrorOr<long> sys_sleep(int seconds);
 		BAN::ErrorOr<long> sys_nanosleep(const timespec* rqtp, timespec* rmtp);
 		BAN::ErrorOr<long> sys_setitimer(int which, const itimerval* value, itimerval* ovalue);
 
@@ -137,7 +136,7 @@ namespace Kernel
 		BAN::ErrorOr<long> sys_recvmsg(int socket, msghdr* message, int flags);
 		BAN::ErrorOr<long> sys_sendmsg(int socket, const msghdr* message, int flags);
 
-		BAN::ErrorOr<long> sys_ioctl(int fildes, int request, void* arg);
+		BAN::ErrorOr<long> sys_ioctl(int fildes, unsigned long request, void* arg);
 
 		BAN::ErrorOr<long> sys_pselect(sys_pselect_t* arguments);
 		BAN::ErrorOr<long> sys_ppoll(pollfd* fds, nfds_t nfds, const timespec* tmp_p, const sigset_t* sigmask);
@@ -210,12 +209,12 @@ namespace Kernel
 		BAN::ErrorOr<long> sys_set_gsbase(void*);
 		BAN::ErrorOr<long> sys_get_gsbase();
 
-		BAN::ErrorOr<long> sys_pthread_create(const pthread_attr_t* attr, void (*entry)(void*), void* arg);
-		BAN::ErrorOr<long> sys_pthread_exit(void* value);
-		BAN::ErrorOr<long> sys_pthread_join(pthread_t thread, void** value);
-		BAN::ErrorOr<long> sys_pthread_self();
-		BAN::ErrorOr<long> sys_pthread_kill(pthread_t thread, int signal);
-		BAN::ErrorOr<long> sys_pthread_detach(pthread_t thread);
+		BAN::ErrorOr<long> sys_thread_create(void (*entry)(void*), void* arg, void* stack_base, size_t stack_size);
+		BAN::ErrorOr<long> sys_thread_exit(void* value);
+		BAN::ErrorOr<long> sys_thread_join(pid_t tid, void** value);
+		BAN::ErrorOr<long> sys_thread_getid();
+		BAN::ErrorOr<long> sys_thread_kill(pid_t tid, int signal);
+		BAN::ErrorOr<long> sys_thread_detach(pid_t tid);
 
 		BAN::ErrorOr<long> sys_clock_gettime(clockid_t, timespec*);
 
@@ -229,11 +228,12 @@ namespace Kernel
 
 		vaddr_t shared_page_vaddr() const { return m_shared_page_vaddr; }
 
-		PageTable& page_table() { return m_page_table ? *m_page_table : PageTable::kernel(); }
+		PageTable& page_table() { return *m_page_table; }
 
 		size_t proc_meminfo(off_t offset, BAN::ByteSpan) const;
 		size_t proc_cmdline(off_t offset, BAN::ByteSpan) const;
 		size_t proc_environ(off_t offset, BAN::ByteSpan) const;
+		size_t proc_cputime(off_t offset, BAN::ByteSpan) const;
 		BAN::ErrorOr<BAN::String> proc_cwd() const;
 		BAN::ErrorOr<BAN::String> proc_exe() const;
 
@@ -364,13 +364,13 @@ namespace Kernel
 
 		BAN::Vector<Thread*> m_threads;
 
-		struct pthread_info_t
+		struct exited_thread_info_t
 		{
-			pthread_t thread;
+			pid_t tid;
 			void* value;
 		};
-		BAN::Vector<pthread_info_t> m_exited_pthreads;
-		ThreadBlocker m_pthread_exit_blocker;
+		BAN::Vector<exited_thread_info_t> m_exited_threads;
+		ThreadBlocker m_thread_exit_blocker;
 
 		uint64_t m_alarm_interval_ns { 0 };
 		uint64_t m_alarm_wake_time_ns { 0 };

kernel/include/kernel/Processor.h

@@ -124,7 +124,7 @@ namespace Kernel
 		static void yield();
 		static Scheduler& scheduler() { return *read_gs_sized<Scheduler*>(offsetof(Processor, m_scheduler)); }
 
-		static void initialize_tsc(uint8_t shift, uint64_t mult, uint64_t realtime_seconds);
+		static void initialize_tsc(uint64_t realtime_seconds);
 		static void update_tsc();
 		static uint64_t ns_since_boot_tsc();
 

kernel/include/kernel/Scheduler.h

@@ -23,7 +23,7 @@ namespace Kernel
 
 	public:
 		void add_thread_to_back(Node*);
-		void add_thread_with_wake_time(Node*);
+		bool add_thread_with_wake_time(Node*); // return true if node was inserted as the first element
 		template<typename F>
 		Node* remove_with_condition(F callback);
 		void remove_node(Node*);
@@ -60,7 +60,8 @@ namespace Kernel
 		void reschedule(YieldRegisters*);
 		void reschedule_if_idle();
 
-		void timer_interrupt();
+		void on_timer_interrupt();
+		void on_yield(YieldRegisters*);
 
 		static BAN::ErrorOr<void> bind_thread_to_processor(Thread*, ProcessorID);
 		// if thread is already bound, this will never fail
@@ -82,6 +83,7 @@ namespace Kernel
 		void update_most_loaded_node_queue(SchedulerQueue::Node*, SchedulerQueue* target_queue);
 		void remove_node_from_most_loaded(SchedulerQueue::Node*);
 
+		void update_wake_up_deadline();
 		void wake_up_sleeping_threads();
 
 		void do_load_balancing();

kernel/include/kernel/Storage/ATA/AHCI/Controller.h

@@ -10,12 +10,14 @@
 namespace Kernel
 {
 
-	class AHCIController final : public StorageController, public Interruptable
+	class AHCIController final : public BAN::RefCounted<AHCIController>, public Interruptable
 	{
 		BAN_NON_COPYABLE(AHCIController);
 		BAN_NON_MOVABLE(AHCIController);
 
 	public:
+		static BAN::ErrorOr<BAN::RefPtr<AHCIController>> create(PCI::Device&);
+
 		~AHCIController();
 
 		virtual void handle_irq() override;
@@ -27,6 +29,7 @@ namespace Kernel
 			: m_pci_device(pci_device)
 		{ }
 		BAN::ErrorOr<void> initialize();
+
 		BAN::Optional<AHCIPortType> check_port_type(volatile HBAPortMemorySpace&);
 
 	private:

kernel/include/kernel/Storage/ATA/ATABus.h

@@ -56,8 +56,6 @@ namespace Kernel
 
 		// Non-owning pointers
 		BAN::Vector<ATADevice*> m_devices;
-
-		friend class ATAController;
 	};
 
 }

kernel/include/kernel/Storage/ATA/ATAController.h

@@ -2,23 +2,22 @@
 
 #include <BAN/UniqPtr.h>
 #include <kernel/PCI.h>
-#include <kernel/Storage/StorageController.h>
 #include <kernel/Storage/ATA/ATABus.h>
 #include <kernel/Storage/ATA/ATADevice.h>
 
 namespace Kernel
 {
 
-	class ATAController : public StorageController
+	class ATAController : public BAN::RefCounted<ATAController>
 	{
 	public:
-		static BAN::ErrorOr<BAN::RefPtr<StorageController>> create(PCI::Device&);
-		virtual BAN::ErrorOr<void> initialize() override;
+		static BAN::ErrorOr<BAN::RefPtr<ATAController>> create(PCI::Device&);
 
 	private:
 		ATAController(PCI::Device& pci_device)
 			: m_pci_device(pci_device)
 		{ }
+		BAN::ErrorOr<void> initialize();
 
 	private:
 		PCI::Device& m_pci_device;

kernel/include/kernel/Storage/NVMe/Controller.h

@@ -10,13 +10,13 @@
 namespace Kernel
 {
 
-	class NVMeController final : public StorageController, public CharacterDevice
+	class NVMeController final : public CharacterDevice
 	{
 		BAN_NON_COPYABLE(NVMeController);
 		BAN_NON_MOVABLE(NVMeController);
 
 	public:
-		static BAN::ErrorOr<BAN::RefPtr<StorageController>> create(PCI::Device&);
+		static BAN::ErrorOr<BAN::RefPtr<NVMeController>> create(PCI::Device&);
 
 		NVMeQueue& io_queue() { return *m_io_queue; }
 
@@ -30,7 +30,7 @@ namespace Kernel
 
 	private:
 		NVMeController(PCI::Device& pci_device);
-		virtual BAN::ErrorOr<void> initialize() override;
+		BAN::ErrorOr<void> initialize();
 
 		BAN::ErrorOr<void> identify_controller();
 		BAN::ErrorOr<void> identify_namespaces();

kernel/include/kernel/Storage/StorageController.h

@@ -1,15 +0,0 @@
-#pragma once
-
-#include <BAN/RefPtr.h>
-
-namespace Kernel
-{
-
-	class StorageController : public BAN::RefCounted<StorageController>
-	{
-	public:
-		virtual ~StorageController() {}
-		virtual BAN::ErrorOr<void> initialize() = 0;
-	};
-
-}

kernel/include/kernel/Terminal/PseudoTerminal.h

@@ -32,7 +32,7 @@ namespace Kernel
 
 		void on_close(int) override;
 
-		BAN::ErrorOr<long> ioctl_impl(int, void*) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) override;
 
 	private:
 		PseudoTerminalMaster(BAN::UniqPtr<VirtualRange>, mode_t, uid_t, gid_t);

kernel/include/kernel/Terminal/TTY.h

@@ -68,7 +68,7 @@ namespace Kernel
 		virtual bool has_error_impl() const override { return false; }
 		virtual bool has_hungup_impl() const override { return false; }
 
-		virtual BAN::ErrorOr<long> ioctl_impl(int, void*) override;
+		virtual BAN::ErrorOr<long> ioctl_impl(unsigned long, void*) override;
 
 	private:
 		bool putchar(uint8_t ch);

kernel/include/kernel/Thread.h

@@ -41,22 +41,20 @@ namespace Kernel
 		// TODO: userspace stack size is hard limited, maybe make this dynamic?
 #if ARCH(x86_64)
 		static constexpr size_t userspace_stack_size { 32 << 20 };
+		static constexpr vaddr_t userspace_stack_base { 0x0000700000000000 };
 #elif ARCH(i686)
 		static constexpr size_t userspace_stack_size { 4 << 20 };
+		static constexpr vaddr_t userspace_stack_base { 0xB0000000 };
 #endif
 
 	public:
 		static BAN::ErrorOr<Thread*> create_kernel(entry_t, void*);
-		static BAN::ErrorOr<Thread*> create_userspace(Process*, PageTable&);
+		static BAN::ErrorOr<Thread*> create_userspace(Process*, PageTable&, vaddr_t userspace_stack_vaddr, size_t userspace_stack_size, vaddr_t entry_point, vaddr_t stack_pointer);
 		~Thread();
 
-		BAN::ErrorOr<Thread*> pthread_create(entry_t, void*);
-
 		BAN::ErrorOr<Thread*> clone(Process*, uintptr_t sp, uintptr_t ip);
 		void setup_process_cleanup();
 
-		BAN::ErrorOr<void> initialize_userspace(vaddr_t entry, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv);
-
 		// Returns true, if thread is going to trigger signal
 		bool is_interrupted_by_signal(bool skip_stop_and_cont = false) const;
 
@@ -77,15 +75,21 @@ namespace Kernel
 
 		// blocks current thread and returns either on unblock, eintr, spuriously or after timeout
 		// if mutex is not nullptr, it will be atomically freed before blocking and automatically locked on wake
-		BAN::ErrorOr<void> sleep_or_eintr_ns(uint64_t ns);
+		BAN::ErrorOr<void> sleep_or_eintr_for_ns(uint64_t timeout_ns);
+		BAN::ErrorOr<void> sleep_or_eintr_until_ns(uint64_t waketime_ns);
 		BAN::ErrorOr<void> block_or_eintr_indefinite(ThreadBlocker& thread_blocker, BaseMutex* mutex);
 		BAN::ErrorOr<void> block_or_eintr_or_timeout_ns(ThreadBlocker& thread_blocker, uint64_t timeout_ns, bool etimedout, BaseMutex* mutex);
 		BAN::ErrorOr<void> block_or_eintr_or_waketime_ns(ThreadBlocker& thread_blocker, uint64_t wake_time_ns, bool etimedout, BaseMutex* mutex);
 
-		BAN::ErrorOr<void> sleep_or_eintr_ms(uint64_t ms)
+		BAN::ErrorOr<void> sleep_or_eintr_for_ms(uint64_t timeout_ms)
 		{
-			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(ms, 1'000'000));
-			return sleep_or_eintr_ns(ms * 1'000'000);
+			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(timeout_ms, 1'000'000));
+			return sleep_or_eintr_for_ns(timeout_ms * 1'000'000);
+		}
+		BAN::ErrorOr<void> sleep_or_eintr_until_ms(uint64_t waketime_ms)
+		{
+			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(waketime_ms, 1'000'000));
+			return sleep_or_eintr_until_ns(waketime_ms * 1'000'000);
 		}
 		BAN::ErrorOr<void> block_or_eintr_or_timeout_ms(ThreadBlocker& thread_blocker, uint64_t timeout_ms, bool etimedout, BaseMutex* mutex)
 		{
@@ -106,8 +110,6 @@ namespace Kernel
 		vaddr_t kernel_stack_top() const	{ return m_kernel_stack->vaddr() + m_kernel_stack->size(); }
 		VirtualRange& kernel_stack() { return *m_kernel_stack; }
 
-		MemoryBackedRegion& userspace_stack() { ASSERT(is_userspace() && m_userspace_stack); return *m_userspace_stack; }
-
 		static Thread& current();
 		static pid_t current_tid();
 
@@ -151,8 +153,6 @@ namespace Kernel
 	private:
 		Thread(pid_t tid, Process*);
 
-		void setup_exec(vaddr_t ip, vaddr_t sp);
-
 		static void on_exit_trampoline(Thread*);
 		void on_exit();
 
@@ -174,7 +174,6 @@ namespace Kernel
 		BAN::UniqPtr<PageTable>    m_keep_alive_page_table;
 
 		BAN::UniqPtr<VirtualRange> m_kernel_stack;
-		MemoryBackedRegion*        m_userspace_stack      { nullptr };
 		const pid_t                m_tid                  { 0 };
 		State                      m_state                { State::NotStarted };
 		Process*                   m_process              { nullptr };
@@ -182,6 +181,9 @@ namespace Kernel
 		BAN::Atomic<bool>          m_is_detached          { false };
 		bool                       m_delete_process       { false };
 
+		vaddr_t                    m_userspace_stack_vaddr { 0 };
+		size_t                     m_userspace_stack_size  { 0 };
+
 		vaddr_t                    m_fsbase               { 0 };
 		vaddr_t                    m_gsbase               { 0 };
 

kernel/include/kernel/Timer/HPET.h

@@ -12,7 +12,7 @@ namespace Kernel
 	class HPET final : public Timer, public Interruptable
 	{
 	public:
-		static BAN::ErrorOr<BAN::UniqPtr<HPET>> create(bool force_pic);
+		static BAN::ErrorOr<BAN::UniqPtr<HPET>> create();
 		~HPET();
 
 		virtual uint64_t ms_since_boot() const override;
@@ -26,7 +26,7 @@ namespace Kernel
 
 	private:
 		HPET() = default;
-		BAN::ErrorOr<void> initialize(bool force_pic);
+		BAN::ErrorOr<void> initialize();
 
 		volatile HPETRegisters& registers();
 		const volatile HPETRegisters& registers() const;

kernel/include/kernel/Timer/Timer.h

@@ -2,6 +2,7 @@
 
 #include <BAN/UniqPtr.h>
 #include <BAN/Vector.h>
+#include <kernel/Memory/DMARegion.h>
 #include <kernel/Timer/RTC.h>
 
 #include <time.h>
@@ -31,7 +32,14 @@ namespace Kernel
 	class SystemTimer : public Timer
 	{
 	public:
-		static void initialize(bool force_pic);
+		struct TSCInfo
+		{
+			int8_t shift;
+			uint32_t mult;
+		};
+
+	public:
+		static void initialize();
 		static SystemTimer& get();
 		static bool is_initialized();
 
@@ -44,29 +52,60 @@ namespace Kernel
 		virtual bool pre_scheduler_sleep_needs_lock() const override;
 		virtual void pre_scheduler_sleep_ns(uint64_t) override;
 
-		void sleep_ms(uint64_t ms) const { ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(ms, 1'000'000)); return sleep_ns(ms * 1'000'000); }
-		void sleep_ns(uint64_t ns) const;
+		void sleep_for_ns(uint64_t timeout_ns) const;
+		void sleep_until_ns(uint64_t waketime_ns) const;
+
+		void sleep_for_ms(uint64_t timeout_ms) const
+		{
+			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(timeout_ms, 1'000'000));
+			return sleep_for_ns(timeout_ms * 1'000'000);
+		}
+		void sleep_until_ms(uint64_t waketime_ms) const
+		{
+			ASSERT(!BAN::Math::will_multiplication_overflow<uint64_t>(waketime_ms, 1'000'000));
+			return sleep_until_ns(waketime_ms * 1'000'000);
+		}
 
 		void dont_invoke_scheduler() { m_timer->m_should_invoke_scheduler = false; }
 
-		void update_tsc() const;
-		uint64_t ns_since_boot_no_tsc() const;
+		void update_tsc();
+		TSCInfo tsc_info() const;
+
+		uint64_t ns_since_boot_no_tsc() const { return m_timer->ns_since_boot(); }
 
 		timespec real_time() const;
 
 	private:
 		SystemTimer() = default;
 
-		void initialize_timers(bool force_pic);
+		void initialize_timers();
 
-		uint64_t get_tsc_frequency() const;
+		void initialize_invariant_tsc();
+		void initialize_pvclock();
 
 	private:
+		enum class TSCType
+		{
+			None,
+			Invariant,
+			PVClock,
+		};
+
 		uint64_t m_boot_time { 0 };
 		BAN::UniqPtr<RTC> m_rtc;
 		BAN::UniqPtr<Timer> m_timer;
-		bool m_has_invariant_tsc { false };
-		mutable uint32_t m_timer_ticks { 0 };
+
+		TSCType m_tsc_type = TSCType::None;
+
+		union {
+			struct {
+				int8_t shift;
+				uint32_t mult;
+			} invariant;
+		} m_tsc_info;
+
+		BAN::UniqPtr<DMARegion> m_tsc_page;
+		uint64_t m_tsc_update_ns { 0 };
 	};
 
 }

kernel/include/kernel/USB/HID/Joystick.h

@@ -34,7 +34,7 @@ namespace Kernel
 		BAN::ErrorOr<size_t> read_impl(off_t, BAN::ByteSpan) override;
 		bool can_read_impl() const override { return true; }
 
-		BAN::ErrorOr<long> ioctl_impl(int request, void* arg) override;
+		BAN::ErrorOr<long> ioctl_impl(unsigned long request, void* arg) override;
 
 	private:
 		USBJoystick(USBHIDDriver&);

kernel/include/kernel/USB/MassStorage/SCSIDevice.h

@@ -33,7 +33,7 @@ namespace Kernel
 		const uint64_t m_block_count;
 		const uint32_t m_block_size;
 
-		const char m_name[4];
+		char m_name[4];
 
 		friend class BAN::RefPtr<USBSCSIDevice>;
 	};

kernel/kernel/ACPI/ACPI.cpp

@@ -952,7 +952,7 @@ acpi_release_global_lock:
 			{
 				if (IO::inw(fadt().pm1a_cnt_blk) & PM1_CNT_SCI_EN)
 					break;
-				SystemTimer::get().sleep_ms(10);
+				SystemTimer::get().sleep_for_ms(10);
 			}
 
 			if (!(IO::inw(fadt().pm1a_cnt_blk) & PM1_CNT_SCI_EN))

kernel/kernel/ACPI/AML/Node.cpp

@@ -2146,9 +2146,9 @@ namespace Kernel::ACPI::AML
 
 		auto milliseconds = TRY(convert_node(TRY(parse_node(context)), ConvInteger, sizeof(uint64_t)));
 
-		const uint64_t wakeup_ms = SystemTimer::get().ms_since_boot() + milliseconds.as.integer.value;
-		for (uint64_t curr_ms = SystemTimer::get().ms_since_boot(); curr_ms < wakeup_ms; curr_ms = SystemTimer::get().ms_since_boot())
-			SystemTimer::get().sleep_ms(wakeup_ms - curr_ms);
+		const uint64_t waketime_ms = SystemTimer::get().ms_since_boot() + milliseconds.as.integer.value;
+		while (SystemTimer::get().ms_since_boot() < waketime_ms)
+			SystemTimer::get().sleep_until_ms(waketime_ms);
 
 		return {};
 	}
@@ -2392,7 +2392,7 @@ namespace Kernel::ACPI::AML
 		result.type = Node::Type::Integer;
 		result.as.integer.value = BAN::numeric_limits<uint64_t>::max();
 
-		const uint64_t wake_time_ms = (timeout_ms >= 0xFFFF)
+		const uint64_t waketime_ms = (timeout_ms >= 0xFFFF)
 			? BAN::numeric_limits<uint64_t>::max()
 			: SystemTimer::get().ms_since_boot() + timeout_ms;
 
@@ -2401,9 +2401,9 @@ namespace Kernel::ACPI::AML
 		{
 			if (sync_object->node.as.mutex->mutex.try_lock())
 				break;
-			if (SystemTimer::get().ms_since_boot() >= wake_time_ms)
+			if (SystemTimer::get().ms_since_boot() >= waketime_ms)
 				return result;
-			SystemTimer::get().sleep_ms(1);
+			Processor::yield();
 		}
 
 		if (sync_object->node.as.mutex->global_lock)

kernel/kernel/APIC.cpp

@@ -437,8 +437,27 @@ namespace Kernel
 
 		dprintln("CPU {}: lapic timer frequency: {} Hz", Kernel::Processor::current_id(), m_lapic_timer_frequency_hz);
 
-		write_to_local_apic(LAPIC_TIMER_LVT,         TimerModePeriodic | IRQ_TIMER);
-		write_to_local_apic(LAPIC_TIMER_INITIAL_REG, m_lapic_timer_frequency_hz / 2 / 100);
+		write_to_local_apic(LAPIC_TIMER_LVT,         TimerModeOneShot | IRQ_TIMER);
+		write_to_local_apic(LAPIC_TIMER_INITIAL_REG, 0);
+	}
+
+	void APIC::set_timer_dealine(uint64_t target_ns)
+	{
+		// TODO: don't hardcode divide by 2
+		const uint64_t effective_freq = m_lapic_timer_frequency_hz / 2;
+
+		const uint64_t delta_ns = [&]() -> uint64_t {
+			const uint64_t current_ns = SystemTimer::get().ns_since_boot();
+			if (current_ns >= target_ns)
+				return 1;
+			const uint64_t delta_ns = target_ns - current_ns;
+			if (BAN::Math::will_multiplication_overflow(delta_ns, effective_freq))
+				return BAN::numeric_limits<uint64_t>::max() / effective_freq;
+			return delta_ns;
+		}();
+
+		const uint64_t ticks = BAN::Math::div_round_up<uint64_t>(delta_ns * effective_freq, 1'000'000'000);
+		write_to_local_apic(LAPIC_TIMER_INITIAL_REG, BAN::Math::min<uint64_t>(ticks, BAN::numeric_limits<uint32_t>::max()));
 	}
 
 	uint32_t APIC::read_from_local_apic(ptrdiff_t offset)

kernel/kernel/Audio/Controller.cpp

@@ -77,9 +77,9 @@ namespace Kernel
 		return to_copy;
 	}
 
-	BAN::ErrorOr<long> AudioController::ioctl_impl(int cmd, void* arg)
+	BAN::ErrorOr<long> AudioController::ioctl_impl(unsigned long request, void* arg)
 	{
-		switch (cmd)
+		switch (request)
 		{
 			case SND_GET_CHANNELS:
 				*static_cast<uint32_t*>(arg) = get_channels();
@@ -92,7 +92,7 @@ namespace Kernel
 			{
 				SpinLockGuard _(m_spinlock);
 				*static_cast<uint32_t*>(arg) = m_sample_data->size();
-				if (cmd == SND_RESET_BUFFER)
+				if (request == SND_RESET_BUFFER)
 					m_sample_data->pop(m_sample_data->size());
 				return 0;
 			}
@@ -113,7 +113,7 @@ namespace Kernel
 				return 0;
 		}
 
-		return CharacterDevice::ioctl_impl(cmd, arg);
+		return CharacterDevice::ioctl_impl(request, arg);
 	}
 
 }

kernel/kernel/Audio/HDAudio/AudioFunctionGroup.cpp

@@ -448,7 +448,7 @@ namespace Kernel
 				continue;
 			const auto& amp = path[i]->output_amplifier.value();
 
-			const int32_t step_mdB = amp.step_size * 250;
+			const int32_t step_mdB = (amp.step_size + 1) * 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;

kernel/kernel/Audio/HDAudio/Controller.cpp

@@ -113,7 +113,7 @@ namespace Kernel
 		// 4.3 The software must wait at least 521 us (25 frames) after reading CRST as a 1
 		// before assuming that codecs have all made status change requests and have been
 		// registered by the controller
-		SystemTimer::get().sleep_ms(1);
+		SystemTimer::get().sleep_for_ns(521'000);
 
 		return {};
 	}

kernel/kernel/CPUID.cpp

@@ -85,6 +85,38 @@ namespace CPUID
 		return buffer[3] & (1 << 8);
 	}
 
+	bool has_rdtscp()
+	{
+		uint32_t buffer[4] {};
+		get_cpuid(0x80000000, buffer);
+		if (buffer[0] < 0x80000001)
+			return false;
+		get_cpuid(0x80000001, buffer);
+		return buffer[3] & (1 << 27);
+	}
+
+	uint64_t get_tsc_frequency()
+	{
+		uint32_t buffer[4];
+		get_cpuid(0x00, buffer);
+		if (buffer[0] < 0x15)
+			return 0;
+		get_cpuid(0x15, buffer);
+		if (buffer[0] == 0 || buffer[1] == 0 || buffer[2] == 0)
+			return 0;
+		return static_cast<uint64_t>(buffer[2]) * buffer[1] / buffer[0];
+	}
+
+	bool has_kvm_pvclock()
+	{
+		uint32_t buffer[4] {};
+		get_cpuid(0x40000000, buffer);
+		if (buffer[1] != 0x4B4D564B || buffer[2] != 0x564B4D56 || buffer[3] != 0x4D)
+			return false;
+		get_cpuid(0x40000001, buffer);
+		return buffer[0] & (1 << 3);
+	}
+
 	const char* feature_string_ecx(uint32_t feat)
 	{
 		switch (feat)

kernel/kernel/Device/FramebufferDevice.cpp

@@ -135,9 +135,9 @@ namespace Kernel
 		return bytes_to_copy;
 	}
 
-	BAN::ErrorOr<long> FramebufferDevice::ioctl_impl(int cmd, void* arg)
+	BAN::ErrorOr<long> FramebufferDevice::ioctl_impl(unsigned long request, void* arg)
 	{
-		switch (cmd)
+		switch (request)
 		{
 			case FB_MSYNC_RECTANGLE:
 			{
@@ -152,7 +152,7 @@ namespace Kernel
 			}
 		}
 
-		return CharacterDevice::ioctl(cmd, arg);
+		return CharacterDevice::ioctl_impl(request, arg);
 	}
 
 	uint32_t FramebufferDevice::get_pixel(uint32_t x, uint32_t y) const

kernel/kernel/ELF.cpp

@@ -8,7 +8,6 @@
 
 #include <ctype.h>
 #include <fcntl.h>
-#include <pthread.h>
 
 namespace Kernel::ELF
 {
@@ -291,7 +290,9 @@ namespace Kernel::ELF
 			TRY(memory_regions.emplace_back(BAN::move(region)));
 		}
 
-		result.open_execfd = !interpreter.empty();
+		if (!interpreter.empty())
+			result.interp_base = load_base_vaddr;
+
 		result.entry_point = load_base_vaddr + file_header.e_entry;
 		result.regions = BAN::move(memory_regions);
 		return BAN::move(result);

kernel/kernel/FS/DevFS/FileSystem.cpp

@@ -50,6 +50,7 @@ namespace Kernel
 			[](void* _devfs)
 			{
 				auto* devfs = static_cast<DevFileSystem*>(_devfs);
+				uint64_t next_update_ms = SystemTimer::get().ms_since_boot();
 				while (true)
 				{
 					{
@@ -57,7 +58,8 @@ namespace Kernel
 						for (auto& device : devfs->m_devices)
 							device->update();
 					}
-					SystemTimer::get().sleep_ms(10);
+					SystemTimer::get().sleep_until_ms(next_update_ms);
+					next_update_ms += 10;
 				}
 			}, s_instance
 		));

kernel/kernel/FS/Ext2/Inode.cpp

@@ -164,7 +164,7 @@ namespace Kernel
 				}
 				return block_from_indirect_block_no_lock(block, data_block_index, i + 1, allocate);
 			}
-			data_block_index -= indices_per_block;
+			data_block_index -= depth_block_count;
 			depth_block_count *= indices_per_block;
 		}
 
@@ -917,6 +917,8 @@ needs_new_block:
 		auto inode_location = TRY(m_fs.locate_inode(ino()));
 		auto block_buffer = TRY(m_fs.get_block_buffer());
 
+		// FIXME: race condition when syncing multiple inodes :)
+
 		TRY(m_fs.read_block(inode_location.block, block_buffer));
 		if (memcmp(block_buffer.data() + inode_location.offset, &inode, sizeof(Ext2::Inode)))
 		{

kernel/kernel/FS/Inode.cpp

@@ -316,7 +316,7 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<long> Inode::ioctl(int request, void* arg)
+	BAN::ErrorOr<long> Inode::ioctl(unsigned long request, void* arg)
 	{
 		auto ret = ioctl_impl(request, arg);
 		if (!ret.is_error() || ret.error().get_error_code() != ENOTSUP)

kernel/kernel/FS/ProcFS/Inode.cpp

@@ -16,12 +16,13 @@ namespace Kernel
 
 		TRY(inode->link_inode(*inode, "."_sv));
 		TRY(inode->link_inode(static_cast<TmpInode&>(*fs.root_inode()), ".."_sv));
-		TRY(inode->link_inode(*MUST(ProcROProcessInode::create_new(process, &Process::proc_meminfo, fs, 0400)), "meminfo"_sv));
+		TRY(inode->link_inode(*MUST(ProcROProcessInode::create_new(process, &Process::proc_meminfo, fs, 0444)), "meminfo"_sv));
 		TRY(inode->link_inode(*MUST(ProcROProcessInode::create_new(process, &Process::proc_cmdline, fs, 0444)), "cmdline"_sv));
 		TRY(inode->link_inode(*MUST(ProcROProcessInode::create_new(process, &Process::proc_environ, fs, 0400)), "environ"_sv));
+		TRY(inode->link_inode(*MUST(ProcROProcessInode::create_new(process, &Process::proc_cputime, fs, 0444)), "cputime"_sv));
 		TRY(inode->link_inode(*MUST(ProcSymlinkProcessInode::create_new(process, &Process::proc_cwd, fs, 0777)), "cwd"_sv));
 		TRY(inode->link_inode(*MUST(ProcSymlinkProcessInode::create_new(process, &Process::proc_exe, fs, 0777)), "exe"_sv));
-		TRY(inode->link_inode(*MUST(ProcFDDirectoryInode::create_new(process, fs, 0777)), "fd"_sv));
+		TRY(inode->link_inode(*MUST(ProcFDDirectoryInode::create_new(process, fs, 0500)), "fd"_sv));
 
 		return inode;
 	}
@@ -38,6 +39,7 @@ namespace Kernel
 		(void)TmpDirectoryInode::unlink_impl("meminfo"_sv);
 		(void)TmpDirectoryInode::unlink_impl("cmdline"_sv);
 		(void)TmpDirectoryInode::unlink_impl("environ"_sv);
+		(void)TmpDirectoryInode::unlink_impl("cputime"_sv);
 		(void)TmpDirectoryInode::unlink_impl("cwd"_sv);
 		(void)TmpDirectoryInode::unlink_impl("exe"_sv);
 		(void)TmpDirectoryInode::unlink_impl("fd"_sv);

kernel/kernel/FS/VirtualFileSystem.cpp

@@ -157,7 +157,7 @@ namespace Kernel
 			if (i == 4)
 				dwarnln("Could not find specified root device, waiting for it to get loaded...");
 
-			SystemTimer::get().sleep_ms(sleep_ms);
+			SystemTimer::get().sleep_for_ms(sleep_ms);
 		}
 
 		derrorln("Could not find root device '{}' after {} ms", root_path, timeout_ms);

kernel/kernel/IDT.cpp

@@ -24,11 +24,6 @@ namespace Kernel
 #if ARCH(x86_64)
 	struct Registers
 	{
-		uint64_t cr4;
-		uint64_t cr3;
-		uint64_t cr2;
-		uint64_t cr0;
-
 		uint64_t r15;
 		uint64_t r14;
 		uint64_t r13;
@@ -49,11 +44,6 @@ namespace Kernel
 #elif ARCH(i686)
 	struct Registers
 	{
-		uint32_t cr4;
-		uint32_t cr3;
-		uint32_t cr2;
-		uint32_t cr0;
-
 		uint32_t edi;
 		uint32_t esi;
 		uint32_t ebp;
@@ -226,8 +216,11 @@ namespace Kernel
 					return;
 				}
 
+				uintptr_t cr2;
+				asm volatile("mov %%cr2, %0" : "=r"(cr2));
+
 				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);
+				auto result = Process::current().allocate_page_for_demand_paging(cr2, page_fault_error.write, page_fault_error.instruction);
 				Processor::set_interrupt_state(InterruptState::Disabled);
 
 				if (result.is_error())
@@ -286,6 +279,15 @@ namespace Kernel
 			}
 		}
 
+		uintptr_t cr0, cr2, cr3, cr4;
+		asm volatile(
+			"mov %%cr0, %0;"
+			"mov %%cr2, %1;"
+			"mov %%cr3, %2;"
+			"mov %%cr4, %3;"
+			: "=r"(cr0), "=r"(cr2), "=r"(cr3), "=r"(cr4)
+		);
+
 		Debug::s_debug_lock.lock();
 
 		if (PageTable::current().get_page_flags(interrupt_stack->ip & PAGE_ADDR_MASK) & PageTable::Flags::Present)
@@ -323,7 +325,7 @@ namespace Kernel
 			regs->r8, regs->r9, regs->r10, regs->r11,
 			regs->r12, regs->r13, regs->r14, regs->r15,
 			interrupt_stack->ip, interrupt_stack->flags,
-			regs->cr0, regs->cr2, regs->cr3, regs->cr4
+			cr0, cr2, cr3, cr4
 		);
 #elif ARCH(i686)
 		dwarnln(
@@ -337,7 +339,7 @@ namespace Kernel
 			regs->eax, regs->ebx, regs->ecx, regs->edx,
 			interrupt_stack->sp, regs->ebp, regs->edi, regs->esi,
 			interrupt_stack->ip, interrupt_stack->flags,
-			regs->cr0, regs->cr2, regs->cr3, regs->cr4
+			cr0, cr2, cr3, cr4
 		);
 #endif
 		if (isr == ISR::PageFault)
@@ -385,11 +387,11 @@ namespace Kernel
 				break;
 			case ISR::PageFault:
 				signal_info.si_signo = SIGSEGV;
-				if (PageTable::current().get_page_flags(regs->cr2 & PAGE_ADDR_MASK) & PageTable::Flags::Present)
+				if (PageTable::current().get_page_flags(cr2 & PAGE_ADDR_MASK) & PageTable::Flags::Present)
 					signal_info.si_code = SEGV_ACCERR;
 				else
 					signal_info.si_code = SEGV_MAPERR;
-				signal_info.si_addr = reinterpret_cast<void*>(regs->cr2);
+				signal_info.si_addr = reinterpret_cast<void*>(cr2);
 				break;
 			default:
 				dwarnln("Unhandled exception");
@@ -433,7 +435,7 @@ namespace Kernel
 		if (Processor::current_is_bsp())
 			Process::update_alarm_queue();
 
-		Processor::scheduler().timer_interrupt();
+		Processor::scheduler().on_timer_interrupt();
 	}
 
 	extern "C" void cpp_irq_handler(uint32_t irq)

kernel/kernel/Input/PS2/Keyboard.cpp

@@ -151,16 +151,18 @@ namespace Kernel::Input
 		uint16_t toggle_mask = 0;
 		switch (dummy_event.key)
 		{
-			case Key::LeftShift:	modifier_mask = KeyModifier::LShift;	break;
-			case Key::RightShift:	modifier_mask = KeyModifier::RShift;	break;
-			case Key::LeftCtrl:		modifier_mask = KeyModifier::LCtrl;		break;
-			case Key::RightCtrl:	modifier_mask = KeyModifier::RCtrl;		break;
-			case Key::LeftAlt:		modifier_mask = KeyModifier::LAlt;		break;
-			case Key::RightAlt:		modifier_mask = KeyModifier::RAlt;		break;
+			case Key::LeftShift:  modifier_mask = KeyModifier::LShift; break;
+			case Key::RightShift: modifier_mask = KeyModifier::RShift; break;
+			case Key::LeftCtrl:   modifier_mask = KeyModifier::LCtrl;  break;
+			case Key::RightCtrl:  modifier_mask = KeyModifier::RCtrl;  break;
+			case Key::LeftAlt:    modifier_mask = KeyModifier::LAlt;   break;
+			case Key::RightAlt:   modifier_mask = KeyModifier::RAlt;   break;
+			case Key::LeftSuper:  modifier_mask = KeyModifier::LSuper; break;
+			case Key::RightSuper: modifier_mask = KeyModifier::RSuper; break;
 
-			case Key::ScrollLock:	toggle_mask = KeyModifier::ScrollLock;	break;
-			case Key::NumLock:		toggle_mask = KeyModifier::NumLock;		break;
-			case Key::CapsLock:		toggle_mask = KeyModifier::CapsLock;	break;
+			case Key::ScrollLock: toggle_mask = KeyModifier::ScrollLock; break;
+			case Key::NumLock:    toggle_mask = KeyModifier::NumLock;    break;
+			case Key::CapsLock:   toggle_mask = KeyModifier::CapsLock;   break;
 
 			default: break;
 		}

kernel/kernel/Input/PS2/Keymap.cpp

@@ -95,7 +95,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_normal[0x38]   = keycode_normal(4,  2);
 		m_scancode_to_keycode_normal[0x39]   = keycode_normal(4,  3);
 		m_scancode_to_keycode_extended[0x38] = keycode_normal(4,  4);
-		m_scancode_to_keycode_extended[0x1D] = keycode_normal(4,  5);
+		m_scancode_to_keycode_extended[0x5C] = keycode_normal(4,  5);
+		m_scancode_to_keycode_extended[0x5D] = keycode_normal(4,  6);
+		m_scancode_to_keycode_extended[0x1D] = keycode_normal(4,  7);
 
 		m_scancode_to_keycode_normal[0x45]   = keycode_numpad(0, 0);
 		m_scancode_to_keycode_extended[0x35] = keycode_numpad(0, 1);
@@ -208,7 +210,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_normal[0x11]   = keycode_normal(4,  2);
 		m_scancode_to_keycode_normal[0x29]   = keycode_normal(4,  3);
 		m_scancode_to_keycode_extended[0x11] = keycode_normal(4,  4);
-		m_scancode_to_keycode_extended[0x14] = keycode_normal(4,  5);
+		m_scancode_to_keycode_extended[0x27] = keycode_normal(4,  5);
+		m_scancode_to_keycode_extended[0x2F] = keycode_normal(4,  6);
+		m_scancode_to_keycode_extended[0x14] = keycode_normal(4,  7);
 
 		m_scancode_to_keycode_normal[0x77]   = keycode_numpad(0, 0);
 		m_scancode_to_keycode_extended[0x4A] = keycode_numpad(0, 1);
@@ -321,7 +325,9 @@ namespace Kernel::Input
 		m_scancode_to_keycode_normal[0x19] = keycode_normal(4,  2);
 		m_scancode_to_keycode_normal[0x29] = keycode_normal(4,  3);
 		m_scancode_to_keycode_normal[0x39] = keycode_normal(4,  4);
-		m_scancode_to_keycode_normal[0x58] = keycode_normal(4,  5);
+		m_scancode_to_keycode_normal[0x8C] = keycode_normal(4,  5);
+		m_scancode_to_keycode_normal[0x8D] = keycode_normal(4,  6);
+		m_scancode_to_keycode_normal[0x58] = keycode_normal(4,  7);
 
 		m_scancode_to_keycode_normal[0x76] = keycode_numpad(0, 0);
 		m_scancode_to_keycode_normal[0x4A] = keycode_numpad(0, 1);

kernel/kernel/Memory/SharedMemoryObject.cpp

@@ -112,8 +112,8 @@ namespace Kernel
 			paddr = Heap::get().take_free_page();
 			if (paddr == 0)
 				return BAN::Error::from_errno(ENOMEM);
-			PageTable::with_fast_page(paddr, [&] {
-				memset(PageTable::fast_page_as_ptr(), 0x00, PAGE_SIZE);
+			PageTable::with_per_cpu_fast_page(paddr, [&](void* addr) {
+				memset(addr, 0x00, PAGE_SIZE);
 			});
 			m_object->paddrs[(vaddr - m_vaddr) / PAGE_SIZE] = paddr;
 		}

kernel/kernel/Networking/ARPTable.cpp

@@ -12,8 +12,8 @@ namespace Kernel
 		Reply = 2,
 	};
 
-	static constexpr BAN::IPv4Address	s_broadcast_ipv4 { 0xFFFFFFFF };
-	static constexpr BAN::MACAddress	s_broadcast_mac {{ 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF }};
+	static constexpr BAN::IPv4Address s_broadcast_ipv4 { 0xFFFFFFFF };
+	static constexpr BAN::MACAddress  s_broadcast_mac  {{ 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF }};
 
 	BAN::ErrorOr<BAN::UniqPtr<ARPTable>> ARPTable::create()
 	{
@@ -50,18 +50,19 @@ namespace Kernel
 				return it->value;
 		}
 
-		ARPPacket arp_request;
-		arp_request.htype = 0x0001;
-		arp_request.ptype = EtherType::IPv4;
-		arp_request.hlen = 0x06;
-		arp_request.plen = 0x04;
-		arp_request.oper = ARPOperation::Request;
-		arp_request.sha = interface.get_mac_address();
-		arp_request.spa = interface.get_ipv4_address();
-		arp_request.tha = {{ 0, 0, 0, 0, 0, 0 }};
-		arp_request.tpa = ipv4_address;
+		const auto arp_request = ARPPacket {
+			.htype = 0x0001,
+			.ptype = EtherType::IPv4,
+			.hlen = 0x06,
+			.plen = 0x04,
+			.oper = ARPOperation::Request,
+			.sha = interface.get_mac_address(),
+			.spa = interface.get_ipv4_address(),
+			.tha = {{ 0, 0, 0, 0, 0, 0 }},
+			.tpa = ipv4_address,
+		};
 
-		TRY(interface.send_bytes(s_broadcast_mac, EtherType::ARP, BAN::ConstByteSpan::from(arp_request)));
+		TRY(interface.send_with_ethernet_header(s_broadcast_mac, EtherType::ARP, BAN::ConstByteSpan::from(arp_request)));
 
 		uint64_t timeout = SystemTimer::get().ms_since_boot() + 1'000;
 		while (SystemTimer::get().ms_since_boot() < timeout)
@@ -100,7 +101,7 @@ namespace Kernel
 			{
 				if (packet.tpa == interface.get_ipv4_address())
 				{
-					const ARPPacket arp_reply {
+					const auto arp_reply = ARPPacket {
 						.htype = 0x0001,
 						.ptype = EtherType::IPv4,
 						.hlen = 0x06,
@@ -111,7 +112,7 @@ namespace Kernel
 						.tha = packet.sha,
 						.tpa = packet.spa,
 					};
-					TRY(interface.send_bytes(packet.sha, EtherType::ARP, BAN::ConstByteSpan::from(arp_reply)));
+					TRY(interface.send_with_ethernet_header(packet.sha, EtherType::ARP, BAN::ConstByteSpan::from(arp_reply)));
 				}
 				break;
 			}

kernel/kernel/Networking/E1000/E1000.cpp

@@ -202,6 +202,8 @@ namespace Kernel
 		rctrl |= RCTL_BSIZE_8192;
 		write32(REG_RCTL, rctrl);
 
+		write32(REG_RXCSUM, RXSUM_IPOFLD | RXSUM_TUOFLD);
+
 		return {};
 	}
 
@@ -279,9 +281,13 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<void> E1000::send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload)
+	BAN::ErrorOr<void> E1000::send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers)
 	{
-		const uint32_t tx_current = m_tx_head1.fetch_add(1) % E1000_TX_DESCRIPTOR_COUNT;
+		const auto interrupt_state = Processor::get_interrupt_state();
+		Processor::set_interrupt_state(InterruptState::Disabled);
+
+		const uint32_t tx_current_nowrap = m_tx_head.fetch_add(1);
+		const uint32_t tx_current = tx_current_nowrap % E1000_TX_DESCRIPTOR_COUNT;
 
 		auto& descriptor = reinterpret_cast<volatile e1000_tx_desc*>(m_tx_descriptor_region->vaddr())[tx_current];
 		while (descriptor.status == 0)
@@ -289,15 +295,10 @@ namespace Kernel
 
 		auto* tx_buffer = reinterpret_cast<uint8_t*>(m_tx_buffer_region->vaddr() + E1000_TX_BUFFER_SIZE * tx_current);
 
-		auto& ethernet_header = *reinterpret_cast<EthernetHeader*>(tx_buffer);
-		ethernet_header.dst_mac = destination;
-		ethernet_header.src_mac = get_mac_address();
-		ethernet_header.ether_type = protocol;
-
-		size_t packet_size = sizeof(EthernetHeader);
-		for (const auto& buffer : payload)
+		size_t packet_size = 0;
+		for (const auto& buffer : buffers)
 		{
-			ASSERT(packet_size + buffer.size() < E1000_TX_BUFFER_SIZE);
+			ASSERT(packet_size + buffer.size() <= E1000_TX_BUFFER_SIZE);
 			memcpy(tx_buffer + packet_size, buffer.data(), buffer.size());
 			packet_size += buffer.size();
 		}
@@ -306,8 +307,12 @@ namespace Kernel
 		descriptor.status = 0;
 		descriptor.cmd = CMD_EOP | CMD_IFCS | CMD_RS;
 
-		if (tx_current == m_tx_head2.fetch_add(1) % E1000_TX_DESCRIPTOR_COUNT)
-			write32(REG_TDT, (tx_current + 1) % E1000_TX_DESCRIPTOR_COUNT);
+		while (tx_current_nowrap != m_tx_commit.load())
+			Processor::pause();
+		write32(REG_TDT, (tx_current + 1) % E1000_TX_DESCRIPTOR_COUNT);
+		m_tx_commit.add_fetch(1);
+
+		Processor::set_interrupt_state(interrupt_state);
 
 		dprintln_if(DEBUG_E1000, "sent {} bytes", packet_size);
 
@@ -316,35 +321,66 @@ namespace Kernel
 
 	void E1000::receive_thread()
 	{
-		SpinLockGuard _(m_rx_lock);
+		SpinLockGuard guard(m_rx_lock);
+
+		uint32_t rx_current = 0;
 
 		while (!m_thread_should_die)
 		{
+			uint32_t rx_tail = rx_current;
+
 			for (;;)
 			{
-				const uint32_t rx_current = (read32(REG_RDT0) + 1) % E1000_RX_DESCRIPTOR_COUNT;
-
 				auto& descriptor = reinterpret_cast<volatile e1000_rx_desc*>(m_rx_descriptor_region->vaddr())[rx_current];
-				if (!(descriptor.status & 1))
+
+				const auto status = descriptor.status;
+				if (!(status & RX_STS_DD))
 					break;
-				ASSERT(descriptor.length <= E1000_RX_BUFFER_SIZE);
 
-				dprintln_if(DEBUG_E1000, "got {} bytes", (uint16_t)descriptor.length);
+				const auto errors = descriptor.errors;
+				if (!(status & RX_STS_EOP))
+					dwarnln("multi descriptor packet??");
+				else if (errors & (RX_ERR_CE | RX_ERR_SE | RX_ERR_RXE))
+					dwarnln("descriptor error {2h}", errors);
+				else if ((status & RX_STS_IPCS) && (errors & RX_ERR_IPE))
+					dwarnln("IPv4 checksum error");
+				else if ((status & RX_STS_TCPCS) && (errors & RX_ERR_TCPE))
+					dwarnln("TCP checkum error");
+				else if ((status & RX_STS_UDPCS) && (errors & RX_ERR_TCPE))
+					dwarnln("UDP checksum error");
+				else
+				{
+					m_rx_lock.unlock(InterruptState::Enabled);
 
-				m_rx_lock.unlock(InterruptState::Enabled);
+					const uint32_t packet_length = descriptor.length;
+					ASSERT(packet_length <= E1000_RX_BUFFER_SIZE);
 
-				NetworkManager::get().on_receive(*this, BAN::ConstByteSpan {
-					reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + rx_current * E1000_RX_BUFFER_SIZE),
-					descriptor.length
-				});
+					dprintln_if(DEBUG_E1000, "got {} bytes", packet_length);
 
-				m_rx_lock.lock();
+					uint32_t validated_cksums = 0;
+					if ((status & RX_STS_IPCS) && !(errors & RX_ERR_IPE))
+						validated_cksums |= CKSUM_IPV4;
+					if ((status & RX_STS_TCPCS) && !(errors & RX_ERR_TCPE))
+						validated_cksums |= CKSUM_TCP;
+					if ((status & RX_STS_UDPCS) && !(errors & RX_ERR_TCPE))
+						validated_cksums |= CKSUM_UDP;
+
+					const uint8_t* packet_data = reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + rx_current * E1000_RX_BUFFER_SIZE);
+					NetworkManager::get().on_receive(*this, BAN::ConstByteSpan { packet_data, packet_length }, validated_cksums);
+
+					m_rx_lock.lock();
+				}
 
 				descriptor.status = 0;
-				write32(REG_RDT0, rx_current);
+
+				rx_tail = rx_current;
+				rx_current = (rx_current + 1) % E1000_RX_DESCRIPTOR_COUNT;
 			}
 
-			SpinLockAsMutex smutex(m_rx_lock, InterruptState::Enabled);
+			if (rx_current != rx_tail)
+				write32(REG_RDT0, rx_tail);
+
+			SpinLockGuardAsMutex smutex(guard);
 			m_rx_blocker.block_indefinite(&smutex);
 		}
 

kernel/kernel/Networking/IPv4Layer.cpp

@@ -180,7 +180,7 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<size_t> IPv4Layer::sendto(NetworkSocket& socket, BAN::ConstByteSpan payload, const sockaddr* address, socklen_t address_len)
+	BAN::ErrorOr<void> IPv4Layer::sendto(NetworkSocket& socket, BAN::ConstByteSpan payload, const sockaddr* address, socklen_t address_len)
 	{
 		if (address->sa_family != AF_INET)
 			return BAN::Error::from_errno(EINVAL);
@@ -224,23 +224,21 @@ namespace Kernel
 		};
 
 		uint8_t protocol_header_buffer[32];
-		ASSERT(socket.protocol_header_size() < sizeof(protocol_header_buffer));
-
 		auto protocol_header = BAN::ByteSpan::from(protocol_header_buffer).slice(0, socket.protocol_header_size());
 		socket.get_protocol_header(protocol_header, payload, dst_port, pseudo_header);
 
-		BAN::ConstByteSpan buffers[] {
+		const BAN::ConstByteSpan buffers[] {
 			BAN::ConstByteSpan::from(ipv4_header),
 			protocol_header,
 			payload,
 		};
 
-		TRY(interface->send_bytes(dst_mac, EtherType::IPv4, { buffers, sizeof(buffers) / sizeof(*buffers) }));
+		TRY(interface->send_with_ethernet_header(dst_mac, EtherType::IPv4, buffers));
 
-		return payload.size();
+		return {};
 	}
 
-	BAN::ErrorOr<void> IPv4Layer::handle_ipv4_packet(NetworkInterface& interface, BAN::ConstByteSpan packet)
+	BAN::ErrorOr<void> IPv4Layer::handle_ipv4_packet(NetworkInterface& interface, BAN::ConstByteSpan packet, uint32_t validated_cksums)
 	{
 		if (packet.size() < sizeof(IPv4Header))
 		{
@@ -249,7 +247,7 @@ namespace Kernel
 		}
 
 		auto& ipv4_header = packet.as<const IPv4Header>();
-		if (calculate_internet_checksum(BAN::ConstByteSpan::from(ipv4_header)) != 0)
+		if (!(validated_cksums & CKSUM_IPV4) && calculate_internet_checksum(BAN::ConstByteSpan::from(ipv4_header)) != 0)
 		{
 			dwarnln_if(DEBUG_IPV4, "IPv4 packet checksum failed");
 			return {};
@@ -284,34 +282,33 @@ namespace Kernel
 				{
 					case ICMPType::EchoRequest:
 					{
-						auto dst_mac = TRY(m_arp_table->get_mac_from_ipv4(interface, src_ipv4));
+						const auto dst_mac = TRY(m_arp_table->get_mac_from_ipv4(interface, src_ipv4));
 
-						auto send_ipv4_header = get_ipv4_header(
+						const auto send_ipv4_header = get_ipv4_header(
 							ipv4_data.size(),
 							interface.get_ipv4_address(),
 							src_ipv4,
 							NetworkProtocol::ICMP
 						);
 
-						ICMPHeader send_icmp_header {
+						auto send_icmp_header = ICMPHeader {
 							.type = ICMPType::EchoReply,
 							.code = icmp_header.code,
 							.checksum = 0,
 							.rest = icmp_header.rest,
 						};
 
-						auto send_payload = ipv4_data.slice(sizeof(ICMPHeader));
-
 						const BAN::ConstByteSpan send_buffers[] {
 							BAN::ConstByteSpan::from(send_ipv4_header),
 							BAN::ConstByteSpan::from(send_icmp_header),
-							send_payload
+							ipv4_data.slice(sizeof(ICMPHeader))
 						};
-						auto send_buffers_span = BAN::Span { send_buffers, sizeof(send_buffers) / sizeof(*send_buffers) };
 
-						send_icmp_header.checksum = calculate_internet_checksum(send_buffers_span.slice(1));
+						send_icmp_header.checksum = calculate_internet_checksum({
+							send_buffers + 1, sizeof(send_buffers) / sizeof(*send_buffers) - 1
+						});
 
-						TRY(interface.send_bytes(dst_mac, EtherType::IPv4, send_buffers_span));
+						TRY(interface.send_with_ethernet_header(dst_mac, EtherType::IPv4, send_buffers));
 
 						break;
 					}
@@ -389,7 +386,7 @@ namespace Kernel
 		sender.sin_family = AF_INET;
 		sender.sin_port = BAN::host_to_network_endian(src_port);
 		sender.sin_addr.s_addr = src_ipv4.raw;
-		bound_socket->receive_packet(ipv4_data, reinterpret_cast<const sockaddr*>(&sender), sizeof(sender));
+		bound_socket->receive_packet(ipv4_data, reinterpret_cast<const sockaddr*>(&sender), sizeof(sender), validated_cksums);
 
 		return {};
 	}

kernel/kernel/Networking/Loopback.cpp

@@ -1,4 +1,4 @@
-#include <kernel/Lock/LockGuard.h>
+#include <kernel/Lock/SpinLockAsMutex.h>
 #include <kernel/Networking/Loopback.h>
 #include <kernel/Networking/NetworkManager.h>
 
@@ -54,12 +54,15 @@ namespace Kernel
 			Processor::yield();
 	}
 
-	BAN::ErrorOr<void> LoopbackInterface::send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload)
+	BAN::ErrorOr<void> LoopbackInterface::send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers)
 	{
+		const auto interrupt_state = Processor::get_interrupt_state();
+		Processor::set_interrupt_state(InterruptState::Disabled);
+
 		auto& descriptor =
 			[&]() -> Descriptor&
 			{
-				LockGuard _(m_buffer_lock);
+				SpinLockGuard guard(m_buffer_lock);
 				for (;;)
 				{
 					auto& descriptor = m_descriptors[m_buffer_head];
@@ -69,34 +72,31 @@ namespace Kernel
 						descriptor.state = 1;
 						return descriptor;
 					}
-					m_thread_blocker.block_indefinite(&m_buffer_lock);
+					SpinLockGuardAsMutex smutex(guard);
+					m_thread_blocker.block_indefinite(&smutex);
 				}
 			}();
 
-		auto& ethernet_header = *reinterpret_cast<EthernetHeader*>(descriptor.addr);
-		ethernet_header.dst_mac = destination;
-		ethernet_header.src_mac = get_mac_address();
-		ethernet_header.ether_type = protocol;
-
-		size_t packet_size = sizeof(EthernetHeader);
-		for (const auto& buffer : payload)
+		size_t packet_size = 0;
+		for (const auto& buffer : buffers)
 		{
 			ASSERT(packet_size + buffer.size() <= buffer_size);
 			memcpy(descriptor.addr + packet_size, buffer.data(), buffer.size());
 			packet_size += buffer.size();
 		}
 
-		LockGuard _(m_buffer_lock);
+		m_buffer_lock.lock();
 		descriptor.size = packet_size;
 		descriptor.state = 2;
 		m_thread_blocker.unblock();
+		m_buffer_lock.unlock(interrupt_state);
 
 		return {};
 	}
 
 	void LoopbackInterface::receive_thread()
 	{
-		LockGuard _(m_buffer_lock);
+		SpinLockGuard guard(m_buffer_lock);
 
 		while (!m_thread_should_die)
 		{
@@ -107,12 +107,9 @@ namespace Kernel
 					break;
 				m_buffer_tail = (m_buffer_tail + 1) % buffer_count;
 
-				m_buffer_lock.unlock();
+				m_buffer_lock.unlock(InterruptState::Enabled);
 
-				NetworkManager::get().on_receive(*this, {
-					descriptor.addr,
-					descriptor.size,
-				});
+				NetworkManager::get().on_receive(*this, BAN::ConstByteSpan { descriptor.addr, descriptor.size }, 0);
 
 				m_buffer_lock.lock();
 
@@ -121,7 +118,8 @@ namespace Kernel
 				m_thread_blocker.unblock();
 			}
 
-			m_thread_blocker.block_indefinite(&m_buffer_lock);
+			SpinLockGuardAsMutex smutex(guard);
+			m_thread_blocker.block_indefinite(&smutex);
 		}
 
 		m_thread_is_dead = true;

kernel/kernel/Networking/NetworkInterface.cpp

@@ -45,7 +45,7 @@ namespace Kernel
 		}
 	}
 
-	BAN::ErrorOr<long> NetworkInterface::ioctl_impl(int request, void* arg)
+	BAN::ErrorOr<long> NetworkInterface::ioctl_impl(unsigned long request, void* arg)
 	{
 		if (arg == nullptr)
 		{

kernel/kernel/Networking/NetworkManager.cpp

@@ -148,7 +148,7 @@ namespace Kernel
 		return {};
 	}
 
-	void NetworkManager::on_receive(NetworkInterface& interface, BAN::ConstByteSpan packet)
+	void NetworkManager::on_receive(NetworkInterface& interface, BAN::ConstByteSpan packet, uint32_t validated_cksums)
 	{
 		if (packet.size() < sizeof(EthernetHeader))
 			return;
@@ -163,7 +163,7 @@ namespace Kernel
 					dwarnln("ARP: {}", ret.error());
 				break;
 			case EtherType::IPv4:
-				if (auto ret = m_ipv4_layer->handle_ipv4_packet(interface, packet_data); ret.is_error())
+				if (auto ret = m_ipv4_layer->handle_ipv4_packet(interface, packet_data, validated_cksums); ret.is_error())
 					dwarnln("IPv4; {}", ret.error());
 				break;
 			default:

kernel/kernel/Networking/NetworkSocket.cpp

@@ -108,7 +108,7 @@ namespace Kernel
 		m_address_len = addr_len;
 	}
 
-	BAN::ErrorOr<long> NetworkSocket::ioctl_impl(int request, void* arg)
+	BAN::ErrorOr<long> NetworkSocket::ioctl_impl(unsigned long request, void* arg)
 	{
 		switch (request)
 		{

kernel/kernel/Networking/RTL8169/RTL8169.cpp

@@ -47,30 +47,29 @@ namespace Kernel
 
 		TRY(reset());
 
+		// disable rx, tx, checksum offload
+		m_io_bar_region->write16(RTL8169_IO_CPlusCR, 0x0000);
+		m_io_bar_region->write8 (RTL8169_IO_CR,      0x00);
+
 		dprintln("  reset done");
 
 		for (size_t i = 0; i < 6; i++)
 			m_mac_address.address[i] = m_io_bar_region->read8(RTL8169_IO_IDR0 + i);
 		dprintln("  MAC {}", m_mac_address);
 
-		// unlock config registers
-		m_io_bar_region->write8(RTL8169_IO_9346CR, RTL8169_9346CR_MODE_CONFIG);
-
 		TRY(initialize_rx());
 		TRY(initialize_tx());
 		m_io_bar_region->write8(RTL8169_IO_CR, RTL8169_CR_RE | RTL8169_CR_TE);
 		dprintln("  descriptors initialized");
 
 		m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts;
-		if (m_link_up)
+		dprintln("  link status {}", link_up() ? "UP" : "DOWN");
+		if (link_up())
 			dprintln("  link speed {}", link_speed());
 
 		TRY(enable_interrupt());
 		dprintln("  interrupts enabled");
 
-		// lock config registers
-		m_io_bar_region->write8(RTL8169_IO_9346CR, RTL8169_9346CR_MODE_NORMAL);
-
 		auto* thread = TRY(Thread::create_kernel([](void* rtl8169_ptr) {
 			static_cast<RTL8169*>(rtl8169_ptr)->receive_thread();
 		}, this));
@@ -79,17 +78,17 @@ namespace Kernel
 			delete thread;
 			return ret.release_error();
 		}
-		m_thread_is_dead = false;
+		m_rx_thread_is_dead = false;
 
 		return {};
 	}
 
 	RTL8169::~RTL8169()
 	{
-		m_thread_should_die = true;
-		m_thread_blocker.unblock();
+		m_rx_thread_should_die = true;
+		m_rx_blocker.unblock();
 
-		while (!m_thread_is_dead)
+		while (!m_rx_thread_is_dead)
 			Processor::yield();
 	}
 
@@ -110,35 +109,33 @@ namespace Kernel
 		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)));
 
+		auto* rx_descriptors = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr());
 		for (size_t i = 0; i < m_rx_descriptor_count; i++)
 		{
 			const paddr_t rx_buffer_paddr = m_rx_buffer_region->paddr() + i * s_buffer_size;
-
-			uint32_t command = 0x1FF8 | RTL8169_DESC_CMD_OWN;
-			if (i == m_rx_descriptor_count - 1)
-				command |= RTL8169_DESC_CMD_EOR;
-
-			auto& rx_descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[i];
-			rx_descriptor.command     = command;
-			rx_descriptor.vlan        = 0;
-			rx_descriptor.buffer_low  = rx_buffer_paddr & 0xFFFFFFFF;
-			rx_descriptor.buffer_high = rx_buffer_paddr >> 32;
+			rx_descriptors[i].command     = 0x1FF8 | RTL8169_DESC_CMD_OWN;
+			rx_descriptors[i].vlan        = 0;
+			rx_descriptors[i].buffer_low  = rx_buffer_paddr & 0xFFFFFFFF;
+			rx_descriptors[i].buffer_high = rx_buffer_paddr >> 32;
 		}
+		rx_descriptors[m_rx_descriptor_count - 1].command |= RTL8169_DESC_CMD_EOR;
 
 		// configure rx descriptor addresses
-		m_io_bar_region->write32(RTL8169_IO_RDSAR + 0, m_rx_descriptor_region->paddr() & 0xFFFFFFFF);
 		m_io_bar_region->write32(RTL8169_IO_RDSAR + 4, m_rx_descriptor_region->paddr() >> 32);
+		m_io_bar_region->write32(RTL8169_IO_RDSAR + 0, m_rx_descriptor_region->paddr() & 0xFFFFFFFF);
 
-		// configure receibe control (no fifo threshold, max dma burst 1024, accept physical match, broadcast, multicast)
+		// configure receive control (no fifo threshold, max dma burst unlimited, broadcast, multicast, accept physical match)
 		m_io_bar_region->write32(RTL8169_IO_RCR,
-			RTL8169_RCR_RXFTH_NO | RTL8169_RCR_MXDMA_1024 | RTL8169_RCR_AB | RTL8169_RCR_AM | RTL8169_RCR_APM
+			RTL8169_RCR_RXFTH_NO | RTL8169_RCR_MXDMA_UNLIMITED | RTL8169_RCR_AB | RTL8169_RCR_AM | RTL8169_RCR_APM
 		);
 
-		m_io_bar_region->write32(0x44, 0b111111 | (0b111 << 8) | (0b111 << 13));
-
 		// configure max rx packet size
 		m_io_bar_region->write16(RTL8169_IO_RMS, RTL8169_RMS_MAX);
 
+		// enable ip/tcp/udp checksum offloading
+		// TODO: is this supported on all cards?
+		m_io_bar_region->write16(RTL8169_IO_CPlusCR, 1 << 5);
+
 		return {};
 	}
 
@@ -147,27 +144,23 @@ namespace Kernel
 		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)));
 
+		auto* tx_descriptors = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr());
 		for (size_t i = 0; i < m_tx_descriptor_count; i++)
 		{
 			const paddr_t tx_buffer_paddr = m_tx_buffer_region->paddr() + i * s_buffer_size;
-
-			uint32_t command = 0;
-			if (i == m_tx_descriptor_count - 1)
-				command |= RTL8169_DESC_CMD_EOR;
-
-			auto& tx_descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr())[i];
-			tx_descriptor.command     = command;
-			tx_descriptor.vlan        = 0;
-			tx_descriptor.buffer_low  = tx_buffer_paddr & 0xFFFFFFFF;
-			tx_descriptor.buffer_high = tx_buffer_paddr >> 32;
+			tx_descriptors[i].command     = 0;
+			tx_descriptors[i].vlan        = 0;
+			tx_descriptors[i].buffer_low  = tx_buffer_paddr & 0xFFFFFFFF;
+			tx_descriptors[i].buffer_high = tx_buffer_paddr >> 32;
 		}
+		tx_descriptors[m_tx_descriptor_count - 1].command |= RTL8169_DESC_CMD_EOR;
 
 		// configure tx descriptor addresses
-		m_io_bar_region->write32(RTL8169_IO_TNPDS + 0, m_tx_descriptor_region->paddr() & 0xFFFFFFFF);
 		m_io_bar_region->write32(RTL8169_IO_TNPDS + 4, m_tx_descriptor_region->paddr() >> 32);
+		m_io_bar_region->write32(RTL8169_IO_TNPDS + 0, m_tx_descriptor_region->paddr() & 0xFFFFFFFF);
 
-		// configure transmit control (standard ifg, max dma burst 1024)
-		m_io_bar_region->write32(RTL8169_IO_TCR, RTL8169_TCR_IFG_0 | RTL8169_TCR_MXDMA_1024);
+		// configure transmit control (standard ifg, max dma burst unlimited)
+		m_io_bar_region->write32(RTL8169_IO_TCR, RTL8169_TCR_IFG_0 | RTL8169_TCR_MXDMA_UNLIMITED);
 
 		// configure max tx packet size
 		m_io_bar_region->write8(RTL8169_IO_MTPS, RTL8169_MTPS_MAX);
@@ -181,14 +174,13 @@ namespace Kernel
 		m_pci_device.enable_interrupt(0, *this);
 
 		m_io_bar_region->write16(RTL8169_IO_IMR,
-			RTL8169_IR_ROK
-			| RTL8169_IR_RER
-			| RTL8169_IR_TOK
-			| RTL8169_IR_TER
-			| RTL8169_IR_RDU
-			| RTL8169_IR_LinkChg
-			| RTL8169_IR_FVOW
-			| RTL8169_IR_TDU
+			RTL8169_IR_ROK |
+			RTL8169_IR_RER |
+			RTL8169_IR_TOK |
+			RTL8169_IR_TER |
+			RTL8169_IR_RDU |
+			RTL8169_IR_LinkChg |
+			RTL8169_IR_FVOW
 		);
 		m_io_bar_region->write16(RTL8169_IO_ISR, 0xFFFF);
 
@@ -209,122 +201,156 @@ namespace Kernel
 		return 0;
 	}
 
-	BAN::ErrorOr<void> RTL8169::send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload)
+	BAN::ErrorOr<void> RTL8169::send_raw_bytes(BAN::Span<const BAN::ConstByteSpan> buffers)
 	{
 		if (!link_up())
 			return BAN::Error::from_errno(EADDRNOTAVAIL);
 
-		auto state = m_lock.lock();
+		const auto interrupt_state = Processor::get_interrupt_state();
+		Processor::set_interrupt_state(InterruptState::Disabled);
 
-		const uint32_t tx_current = m_tx_current;
-		m_tx_current = (m_tx_current + 1) % m_tx_descriptor_count;
+		const uint32_t tx_current_nowrap = m_tx_head.fetch_add(1);
+		const uint32_t tx_current = tx_current_nowrap % m_tx_descriptor_count;
 
 		auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr())[tx_current];
-		while (descriptor.command & RTL8169_DESC_CMD_OWN)
+		if (descriptor.command & RTL8169_DESC_CMD_OWN)
 		{
-			SpinLockAsMutex smutex(m_lock, state);
-			m_thread_blocker.block_indefinite(&smutex);
+			SpinLockGuard guard(m_tx_lock);
+			while (descriptor.command & RTL8169_DESC_CMD_OWN)
+			{
+				SpinLockGuardAsMutex smutex(guard);
+				m_tx_blocker.block_indefinite(&smutex);
+			}
 		}
 
-		m_lock.unlock(state);
-
 		auto* tx_buffer = reinterpret_cast<uint8_t*>(m_tx_buffer_region->vaddr() + tx_current * s_buffer_size);
 
 		// write packet
-		auto& ethernet_header = *reinterpret_cast<EthernetHeader*>(tx_buffer);
-		ethernet_header.dst_mac = destination;
-		ethernet_header.src_mac = get_mac_address();
-		ethernet_header.ether_type = protocol;
-
-		size_t packet_size = sizeof(EthernetHeader);
-		for (const auto& buffer : payload)
+		size_t packet_size = 0;
+		for (const auto& buffer : buffers)
 		{
+			ASSERT(packet_size + buffer.size() <= s_buffer_size);
 			memcpy(tx_buffer + packet_size, buffer.data(), buffer.size());
 			packet_size += buffer.size();
 		}
 
 		// give packet ownership to NIC
 		uint32_t command = packet_size | RTL8169_DESC_CMD_OWN | RTL8169_DESC_CMD_LS | RTL8169_DESC_CMD_FS;
-		if (tx_current >= m_tx_descriptor_count - 1)
+		if (tx_current == m_tx_descriptor_count - 1)
 			command |= RTL8169_DESC_CMD_EOR;
 		descriptor.command = command;
 
-		// notify NIC about new packet
-		m_io_bar_region->write8(RTL8169_IO_TPPoll, RTL8169_TPPoll_NPQ);
+		// ring tx queue doorbell
+		if (tx_current_nowrap == m_tx_commit.load())
+			m_io_bar_region->write8(RTL8169_IO_TPPoll, RTL8169_TPPoll_NPQ);
+		while (tx_current_nowrap != m_tx_commit.load())
+			Processor::pause();
+		m_tx_commit.add_fetch(1);
+
+		Processor::set_interrupt_state(interrupt_state);
 
 		return {};
 	}
 
 	void RTL8169::receive_thread()
 	{
-		SpinLockGuard _(m_lock);
+		SpinLockGuard rx_lock_guard(m_rx_lock);
 
-		while (!m_thread_should_die)
+		while (!m_rx_thread_should_die)
 		{
 			for (;;)
 			{
-				auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[m_rx_current];
+				auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[m_rx_head];
+
+				const auto command = descriptor.command;
 				if (descriptor.command & RTL8169_DESC_CMD_OWN)
 					break;
 
 				// packet buffer can only hold single packet, so we should not receive any multi-descriptor packets
-				ASSERT((descriptor.command & RTL8169_DESC_CMD_LS) && (descriptor.command & RTL8169_DESC_CMD_FS));
+				ASSERT((command & RTL8169_DESC_CMD_LS) && (command & RTL8169_DESC_CMD_FS));
 
-				const uint16_t packet_length = descriptor.command & 0x3FFF;
+				const uint8_t protocol = (command >> 17) & 3;
+				const uint16_t packet_length = command & 0x3FFF;
 				if (packet_length > s_buffer_size)
 					dwarnln("Got {} bytes to {} byte buffer", packet_length, s_buffer_size);
-				else if (descriptor.command & (1u << 21))
-					; // descriptor has an error
+				else if (command & (1u << 21))
+					dwarnln("descriptor error {4h}", command);
+				else if (protocol == 1 && (command & (1u << 14)))
+					dwarnln("TCP checksum error");
+				else if (protocol == 2 && (command & (1u << 15)))
+					dwarnln("UDP checksum error");
+				else if (protocol != 0 && (command & (1u << 16)))
+					dwarnln("IPv4 checksum error");
 				else
 				{
-					m_lock.unlock(InterruptState::Enabled);
+					m_rx_lock.unlock(InterruptState::Enabled);
 
-					NetworkManager::get().on_receive(*this, BAN::ConstByteSpan {
-						reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + m_rx_current * s_buffer_size),
-						packet_length
-					});
+					uint32_t validated_cksums;
+					switch (protocol)
+					{
+						case 0: validated_cksums = 0;                      break;
+						case 1: validated_cksums = CKSUM_IPV4 | CKSUM_TCP; break;
+						case 2: validated_cksums = CKSUM_IPV4 | CKSUM_UDP; break;
+						case 3: validated_cksums = CKSUM_IPV4;             break;
+					}
 
-					m_lock.lock();
+					const uint8_t* packet_data = reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + m_rx_head * s_buffer_size);
+					NetworkManager::get().on_receive(*this, BAN::ConstByteSpan { packet_data, packet_length }, validated_cksums);
+
+					m_rx_lock.lock();
 				}
 
-				m_rx_current = (m_rx_current + 1) % m_rx_descriptor_count;
+				uint32_t new_command = 0x1FF8 | RTL8169_DESC_CMD_OWN;
+				if (m_rx_head == m_rx_descriptor_count - 1)
+					new_command |= RTL8169_DESC_CMD_EOR;
+				descriptor.command = new_command;
 
-				descriptor.command = descriptor.command | RTL8169_DESC_CMD_OWN;
+				m_rx_head = (m_rx_head + 1) % m_rx_descriptor_count;
 			}
 
-			SpinLockAsMutex smutex(m_lock, InterruptState::Enabled);
-			m_thread_blocker.block_indefinite(&smutex);
+			SpinLockGuardAsMutex smutex(rx_lock_guard);
+			m_rx_blocker.block_indefinite(&smutex);
 		}
 
-		m_thread_is_dead = true;
+		m_rx_thread_is_dead = true;
 	}
 
 	void RTL8169::handle_irq()
 	{
-		const uint16_t interrupt_status = m_io_bar_region->read16(RTL8169_IO_ISR);
-		m_io_bar_region->write16(RTL8169_IO_ISR, interrupt_status);
-
-		if (interrupt_status & RTL8169_IR_LinkChg)
+		uint16_t isr;
+		while ((isr = m_io_bar_region->read16(RTL8169_IO_ISR)))
 		{
-			m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts;
-			dprintln("link status -> {}", m_link_up.load());
-		}
+			m_io_bar_region->write16(RTL8169_IO_ISR, isr);
 
-		if (interrupt_status & (RTL8169_IR_TOK | RTL8169_IR_ROK))
-		{
-			SpinLockGuard _(m_lock);
-			m_thread_blocker.unblock();
-		}
+			if (isr & RTL8169_IR_LinkChg)
+			{
+				m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts;
+				dprintln("link status {}", link_up() ? "UP" : "DOWN");
+				if (link_up())
+					dprintln("link speed {}", link_speed());
+			}
 
-		if (interrupt_status & RTL8169_IR_RER)
-			dwarnln("Rx error");
-		if (interrupt_status & RTL8169_IR_TER)
-			dwarnln("Tx error");
-		if (interrupt_status & RTL8169_IR_RDU)
-			dwarnln("Rx descriptor not available");
-		if (interrupt_status & RTL8169_IR_FVOW)
-			dwarnln("Rx FIFO overflow");
-		// dont log TDU is sent after each sent packet
+			if (isr & (RTL8169_IR_TER | RTL8169_IR_TOK))
+			{
+				SpinLockGuard _(m_tx_lock);
+				m_tx_blocker.unblock();
+			}
+
+			if (isr & (RTL8169_IR_RER | RTL8169_IR_ROK))
+			{
+				SpinLockGuard _(m_rx_lock);
+				m_rx_blocker.unblock();
+			}
+
+			if (isr & RTL8169_IR_RER)
+				dwarnln("Rx error");
+			if (isr & RTL8169_IR_TER)
+				dwarnln("Tx error");
+			if (isr & RTL8169_IR_RDU)
+				dwarnln("Rx descriptor not available");
+			if (isr & RTL8169_IR_FVOW)
+				dwarnln("Rx FIFO overflow");
+		}
 	}
 
 }

kernel/kernel/Networking/TCPSocket.cpp

@@ -402,7 +402,7 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<long> TCPSocket::ioctl_impl(int request, void* argument)
+	BAN::ErrorOr<long> TCPSocket::ioctl_impl(unsigned long request, void* argument)
 	{
 		switch (request)
 		{
@@ -572,7 +572,7 @@ namespace Kernel
 		dprintln_if(DEBUG_TCP, "  seq {}", (uint32_t)header.seq_number);
 	}
 
-	void TCPSocket::receive_packet(BAN::ConstByteSpan buffer, const sockaddr* sender, socklen_t sender_len)
+	void TCPSocket::receive_packet(BAN::ConstByteSpan packet, const sockaddr* sender, socklen_t sender_len, uint32_t validated_cksums)
 	{
 		if (m_state == State::Listen)
 		{
@@ -585,9 +585,10 @@ namespace Kernel
 				}();
 
 			if (socket)
-				return socket->receive_packet(buffer, sender, sender_len);
+				return socket->receive_packet(packet, sender, sender_len, validated_cksums);
 		}
 
+		if (!(validated_cksums & CKSUM_TCP))
 		{
 			uint16_t checksum = 0;
 
@@ -603,23 +604,23 @@ namespace Kernel
 					.src_ipv4 = BAN::IPv4Address(addr_in.sin_addr.s_addr),
 					.dst_ipv4 = interface->get_ipv4_address(),
 					.protocol = NetworkProtocol::TCP,
-					.length = buffer.size(),
+					.length = packet.size(),
 				};
 				const BAN::ConstByteSpan buffers[] {
 					BAN::ConstByteSpan::from(pseudo_header),
-					buffer
+					packet
 				};
 				checksum = calculate_internet_checksum({ buffers, sizeof(buffers) / sizeof(*buffers) });
 			}
 			else
 			{
-				dwarnln("No tcp checksum validation for socket family {}", sender->sa_family);
+				dwarnln("no TCP checksum validation for socket family {}", sender->sa_family);
 				return;
 			}
 
 			if (checksum != 0)
 			{
-				dprintln("Checksum does not match");
+				dwarnln("checksum does not match");
 				return;
 			}
 		}
@@ -628,7 +629,7 @@ namespace Kernel
 
 		const bool hungup_before = has_hungup_impl();
 
-		auto& header = buffer.as<const TCPHeader>();
+		const auto& header = packet.as<const TCPHeader>();
 
 		dprintln_if(DEBUG_TCP, "receiving {} {8b}", (uint8_t)m_state, header.flags);
 		dprintln_if(DEBUG_TCP, "  ack {}", (uint32_t)header.ack_number);
@@ -768,7 +769,7 @@ namespace Kernel
 			if (header.ack_number > m_send_window.current_ack)
 				m_send_window.current_ack = header.ack_number;
 
-			auto payload = buffer.slice(header.data_offset * sizeof(uint32_t));
+			auto payload = packet.slice(header.data_offset * sizeof(uint32_t));
 
 			if (header.seq_number < expected_seq)
 			{
@@ -967,6 +968,9 @@ namespace Kernel
 			if (m_last_sent_window_size == 0)
 				m_should_send_zero_window = false;
 
+			if (m_last_sent_window_size == m_recv_window.buffer->free())
+				m_should_send_window_update = false;
+
 			if (m_should_send_zero_window || m_should_send_window_update)
 			{
 				ASSERT(m_connection_info.has_value());

kernel/kernel/Networking/UDPSocket.cpp

@@ -57,21 +57,69 @@ namespace Kernel
 			header.checksum = 0xFFFF;
 	}
 
-	void UDPSocket::receive_packet(BAN::ConstByteSpan packet, const sockaddr* sender, socklen_t sender_len)
+	void UDPSocket::receive_packet(BAN::ConstByteSpan packet, const sockaddr* sender, socklen_t sender_len, uint32_t validated_cksums)
 	{
-		auto payload = packet.slice(sizeof(UDPHeader));
+		const auto header = packet.as<const UDPHeader>();
+
+		if (!(validated_cksums & CKSUM_UDP) && header.checksum)
+		{
+			uint16_t checksum = 0;
+
+			if (sender->sa_family == AF_INET)
+			{
+				auto interface_or_error = interface(sender, sender_len);
+				if (interface_or_error.is_error())
+					return;
+				auto interface = interface_or_error.release_value();
+
+				auto& addr_in = *reinterpret_cast<const sockaddr_in*>(sender);
+				const PseudoHeader pseudo_header {
+					.src_ipv4 = BAN::IPv4Address(addr_in.sin_addr.s_addr),
+					.dst_ipv4 = interface->get_ipv4_address(),
+					.protocol = NetworkProtocol::UDP,
+					.length = packet.size(),
+				};
+				const UDPHeader udp_header {
+					.src_port = header.src_port,
+					.dst_port = header.dst_port,
+					.length   = header.length,
+					.checksum = 0,
+				};
+				const BAN::ConstByteSpan buffers[] {
+					BAN::ConstByteSpan::from(pseudo_header),
+					BAN::ConstByteSpan::from(udp_header),
+					packet.slice(sizeof(UDPHeader))
+				};
+				checksum = calculate_internet_checksum({ buffers, sizeof(buffers) / sizeof(*buffers) });
+			}
+			else
+			{
+				dwarnln("no UDP checksum validation for socket family {}", sender->sa_family);
+				return;
+			}
+
+			const bool checksum_valid1 = (header.checksum == checksum);
+			const bool checksum_valid2 = (header.checksum == 0xFFFF && checksum == 0);
+			if (!checksum_valid1 && !checksum_valid2)
+			{
+				dwarnln("checksum does not match {4h}", checksum);
+				return;
+			}
+		}
 
 		SpinLockGuard _(m_packet_lock);
 
 		if (m_packets.full())
 		{
-			dprintln("Packet buffer full, dropping packet");
+			dwarnln("Packet buffer full, dropping packet");
 			return;
 		}
 
+		const auto payload = packet.slice(sizeof(UDPHeader));
+
 		if (m_packet_total_size + payload.size() > m_packet_buffer->size())
 		{
-			dprintln("Packet buffer full, dropping packet");
+			dwarnln("Packet buffer full, dropping packet");
 			return;
 		}
 
@@ -224,7 +272,9 @@ namespace Kernel
 			address_len = message.msg_namelen;
 		}
 
-		return TRY(m_network_layer.sendto(*this, buffer.span(), address, address_len));
+		TRY(m_network_layer.sendto(*this, buffer.span(), address, address_len));
+
+		return buffer.size();
 	}
 
 	BAN::ErrorOr<void> UDPSocket::getsockopt_impl(int level, int option, void* value, socklen_t* value_len)
@@ -286,7 +336,7 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<long> UDPSocket::ioctl_impl(int request, void* argument)
+	BAN::ErrorOr<long> UDPSocket::ioctl_impl(unsigned long request, void* argument)
 	{
 		switch (request)
 		{

kernel/kernel/Networking/UNIX/Socket.cpp

@@ -704,6 +704,25 @@ namespace Kernel
 		}
 	}
 
+	BAN::ErrorOr<void> UnixDomainSocket::getsockname_impl(sockaddr* address, socklen_t* address_len)
+	{
+		sockaddr_un sa_un {
+			.sun_family = AF_UNIX,
+			.sun_path = {},
+		};
+
+		{
+			LockGuard _(m_bind_mutex);
+			strcpy(sa_un.sun_path, m_bound_file.canonical_path.data());
+		}
+
+		const size_t to_copy = BAN::Math::min<socklen_t>(sizeof(sockaddr_un), *address_len);
+		memcpy(address, &sa_un, to_copy);
+		*address_len = to_copy;
+		return {};
+	}
+
+
 	BAN::ErrorOr<void> UnixDomainSocket::getpeername_impl(sockaddr* address, socklen_t* address_len)
 	{
 		if (!m_info.has<ConnectionInfo>())
@@ -713,20 +732,7 @@ namespace Kernel
 		if (!connection)
 			return BAN::Error::from_errno(ENOTCONN);
 
-		sockaddr_un sa_un {
-			.sun_family = AF_UNIX,
-			.sun_path = {},
-		};
-
-		{
-			LockGuard _(m_bind_mutex);
-			strcpy(sa_un.sun_path, connection->m_bound_file.canonical_path.data());
-		}
-
-		const size_t to_copy = BAN::Math::min<socklen_t>(sizeof(sockaddr_un), *address_len);
-		memcpy(address, &sa_un, to_copy);
-		*address_len = to_copy;
-		return {};
+		return connection->getsockname_impl(address, address_len);
 	}
 
 	BAN::ErrorOr<void> UnixDomainSocket::getsockopt_impl(int level, int option, void* value, socklen_t* value_len)
@@ -743,12 +749,24 @@ namespace Kernel
 			case SO_ERROR:
 				result = 0;
 				break;
+			case SO_KEEPALIVE:
+				result = 1;
+				break;
 			case SO_SNDBUF:
 				result = m_sndbuf;
 				break;
 			case SO_RCVBUF:
 				result = m_packet_buffer->size();
 				break;
+			case SO_TYPE:
+				switch (m_socket_type)
+				{
+					case Type::STREAM:    result = SOCK_STREAM;    break;
+					case Type::DGRAM:     result = SOCK_DGRAM;     break;
+					case Type::SEQPACKET: result = SOCK_SEQPACKET; break;
+					default: ASSERT_NOT_REACHED();
+				}
+				break;
 			default:
 				dwarnln("getsockopt(SOL_SOCKET, {})", option);
 				return BAN::Error::from_errno(ENOTSUP);

kernel/kernel/OpenFileDescriptorSet.cpp

@@ -401,11 +401,19 @@ namespace Kernel
 
 	BAN::ErrorOr<void> OpenFileDescriptorSet::close(int fd)
 	{
-		LockGuard _(m_mutex);
+		BAN::RefPtr<OpenFileDescription> open_file;
 
-		TRY(validate_fd(fd));
+		{
+			LockGuard _(m_mutex);
 
-		auto& open_file = m_open_files[fd];
+			TRY(validate_fd(fd));
+
+			open_file = m_open_files[fd];
+
+			m_open_files[fd]->file.inode->on_close(m_open_files[fd]->status_flags);
+			m_open_files[fd] = {};
+			remove_cloexec(fd);
+		}
 
 		if (auto& flock = open_file->flock; Thread::current().has_process() && flock.lockers.contains(Process::current().pid()))
 		{
@@ -429,10 +437,6 @@ namespace Kernel
 			}
 		}
 
-		open_file->file.inode->on_close(open_file->status_flags);
-		open_file = {};
-		remove_cloexec(fd);
-
 		return {};
 	}
 

kernel/kernel/PCI.cpp

@@ -235,11 +235,13 @@ namespace Kernel::PCI
 						switch (pci_device.subclass())
 						{
 							case 0x01:
-							case 0x05:
-							case 0x06:
 								if (auto res = ATAController::create(pci_device); res.is_error())
 									dprintln("ATA: {}", res.error());
 								break;
+							case 0x06:
+								if (auto res = AHCIController::create(pci_device); res.is_error())
+									dprintln("AHCI: {}", res.error());
+								break;
 							case 0x08:
 								if (auto res = NVMeController::create(pci_device); res.is_error())
 									dprintln("NVMe: {}", res.error());

kernel/kernel/Process.cpp

@@ -28,7 +28,6 @@
 #include <LibInput/KeyboardLayout.h>
 
 #include <fcntl.h>
-#include <pthread.h>
 #include <stdio.h>
 #include <sys/banan-os.h>
 #include <sys/eventfd.h>
@@ -98,25 +97,19 @@ namespace Kernel
 		return process;
 	}
 
-	void Process::register_to_scheduler()
-	{
-		// FIXME: Allow failing...
-		{
-			SpinLockGuard _(s_process_lock);
-			MUST(s_processes.push_back(this));
-		}
-		for (auto* thread : m_threads)
-			MUST(Processor::scheduler().add_thread(thread));
-	}
-
 	BAN::ErrorOr<Process*> Process::create_userspace(const Credentials& credentials, BAN::StringView path, BAN::Span<BAN::StringView> arguments)
 	{
 		auto* process = create_process(credentials, 0);
+		BAN::ScopeGuard process_deleter([process] {
+			process->m_mapped_regions.clear();
+			process->m_page_table.clear();
+			delete process;
+		});
 
 		process->m_working_directory = VirtualFileSystem::get().root_file();
 		process->m_root_file         = VirtualFileSystem::get().root_file();
 
-		process->m_page_table = BAN::UniqPtr<PageTable>::adopt(MUST(PageTable::create_userspace()));
+		process->m_page_table = BAN::UniqPtr<PageTable>::adopt(TRY(PageTable::create_userspace()));
 
 		TRY(process->m_cmdline.emplace_back());
 		TRY(process->m_cmdline.back().append(path));
@@ -126,14 +119,12 @@ namespace Kernel
 			TRY(process->m_cmdline.back().append(argument));
 		}
 
-		LockGuard _(process->m_process_lock);
-
 		auto executable_file = TRY(process->find_file(AT_FDCWD, path.data(), O_EXEC));
 		auto executable_inode = executable_file.inode;
 
 		auto executable = TRY(ELF::load_from_inode(process->m_root_file.inode, executable_inode, process->m_credentials, process->page_table()));
 		for (auto& region : executable.regions)
-			TRY(process->add_mapped_region(BAN::move(region)));
+			TRY(process->m_mapped_regions.push_back(BAN::move(region)));
 		executable.regions.clear();
 
 		TRY(process->m_executable.append(executable_file.canonical_path));
@@ -143,19 +134,7 @@ namespace Kernel
 		if (executable_inode->mode().mode & +Inode::Mode::ISGID)
 			process->m_credentials.set_egid(executable_inode->gid());
 
-		BAN::Vector<LibELF::AuxiliaryVector> auxiliary_vector;
-		TRY(auxiliary_vector.reserve(1 + executable.open_execfd));
-
-		if (executable.open_execfd)
-		{
-			const int execfd = TRY(process->m_open_file_descriptors.open(BAN::move(executable_file), O_RDONLY));
-			TRY(auxiliary_vector.push_back({
-				.a_type = LibELF::AT_EXECFD,
-				.a_un = { .a_val = static_cast<uint32_t>(execfd) },
-			}));
-		}
-
-		process->m_shared_page_vaddr = process->page_table().reserve_free_page(process->m_mapped_regions.back()->vaddr(), USERSPACE_END);
+		process->m_shared_page_vaddr = process->page_table().reserve_free_page(executable.interp_base.value_or(0x400000), USERSPACE_END);
 		if (process->m_shared_page_vaddr == 0)
 			return BAN::Error::from_errno(ENOMEM);
 		process->page_table().map_page_at(
@@ -164,31 +143,89 @@ namespace Kernel
 			PageTable::UserSupervisor | PageTable::Present
 		);
 
+		auto userspace_stack = TRY(MemoryBackedRegion::create(
+			process->page_table(),
+			Thread::userspace_stack_size,
+			{ Thread::userspace_stack_base, USERSPACE_END },
+			MemoryRegion::Type::PRIVATE,
+			PageTable::UserSupervisor | PageTable::ReadWrite | PageTable::Present,
+			O_RDWR
+		));
+
+		BAN::Vector<LibELF::AuxiliaryVector> auxiliary_vector;
+		TRY(auxiliary_vector.reserve(5 + 2 * executable.interp_base.has_value()));
+
+		if (executable.interp_base.has_value())
+		{
+			const int execfd = TRY(process->m_open_file_descriptors.open(BAN::move(executable_file), O_RDONLY));
+			TRY(auxiliary_vector.push_back({
+				.a_type = LibELF::AT_EXECFD,
+				.a_un = { .a_val = static_cast<uint32_t>(execfd) },
+			}));
+			TRY(auxiliary_vector.push_back({
+				.a_type = LibELF::AT_BASE,
+				.a_un = { .a_ptr = reinterpret_cast<void*>(executable.interp_base.value()) },
+			}));
+		}
+
+		TRY(auxiliary_vector.push_back({
+			.a_type = LibELF::AT_PAGESZ,
+			.a_un = { .a_val = PAGE_SIZE },
+		}));
+
 		TRY(auxiliary_vector.push_back({
 			.a_type = LibELF::AT_SHARED_PAGE,
 			.a_un = { .a_ptr = reinterpret_cast<void*>(process->m_shared_page_vaddr) },
 		}));
 
+		TRY(auxiliary_vector.push_back({
+			.a_type = LibELF::AT_STACK_BASE,
+			.a_un = { .a_ptr = reinterpret_cast<void*>(userspace_stack->vaddr()) },
+		}));
+
+		TRY(auxiliary_vector.push_back({
+			.a_type = LibELF::AT_STACK_SIZE,
+			.a_un = { .a_ptr = reinterpret_cast<void*>(userspace_stack->size()) },
+		}));
+
 		TRY(auxiliary_vector.push_back({
 			.a_type = LibELF::AT_NULL,
 			.a_un = { .a_val = 0 },
 		}));
 
-		BAN::Optional<vaddr_t> tls_addr;
-		if (executable.master_tls.has_value())
-		{
-			auto tls_result = TRY(process->initialize_thread_local_storage(process->page_table(), *executable.master_tls));
-			TRY(process->add_mapped_region(BAN::move(tls_result.region)));
-			tls_addr = tls_result.addr;
-		}
-
-		auto* thread = MUST(Thread::create_userspace(process, process->page_table()));
-		MUST(thread->initialize_userspace(
-			executable.entry_point,
+		const vaddr_t initial_stack_pointer = TRY(process->setup_initial_process_stack(
+			*userspace_stack,
 			process->m_cmdline.span(),
 			process->m_environ.span(),
 			auxiliary_vector.span()
 		));
+
+		const vaddr_t userspace_stack_vaddr = userspace_stack->vaddr();
+		const vaddr_t userspace_stack_size  = userspace_stack->size();
+		TRY(process->m_mapped_regions.push_back(BAN::move(userspace_stack)));
+
+		BAN::Optional<vaddr_t> tls_addr;
+		if (executable.master_tls.has_value())
+		{
+			auto tls_result = TRY(process->initialize_thread_local_storage(process->page_table(), *executable.master_tls));
+			TRY(process->m_mapped_regions.push_back(BAN::move(tls_result.region)));
+			tls_addr = tls_result.addr;
+		}
+
+		BAN::sort::sort(process->m_mapped_regions.begin(), process->m_mapped_regions.end(), [](const auto& a, const auto& b) {
+			return a->vaddr() < b->vaddr();
+		});
+
+		auto* thread = TRY(Thread::create_userspace(
+			process,
+			process->page_table(),
+			userspace_stack_vaddr,
+			userspace_stack_size,
+			executable.entry_point,
+			initial_stack_pointer
+		));
+		BAN::ScopeGuard thread_deleter([thread] { delete thread; });
+
 		if (tls_addr.has_value())
 		{
 #if ARCH(x86_64)
@@ -200,11 +237,118 @@ namespace Kernel
 #endif
 		}
 
-		process->add_thread(thread);
-		process->register_to_scheduler();
+		// NOTE: make sure the last two `MUST`s don't fail
+		TRY(process->m_threads.reserve(1));
+		TRY(Processor::scheduler().bind_thread_to_processor(thread, Processor::current_id()));
+
+		{
+			SpinLockGuard _(s_process_lock);
+			TRY(s_processes.push_back(process));
+		}
+
+		MUST(process->m_threads.push_back(thread));
+		MUST(Processor::scheduler().add_thread(thread));
+
+		process_deleter.disable();
+		thread_deleter.disable();
+
 		return process;
 	}
 
+	BAN::ErrorOr<vaddr_t> Process::setup_initial_process_stack(MemoryBackedRegion& stack_region, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv)
+	{
+		// System V ABI: Initial process stack
+
+		size_t needed_size = 0;
+
+		// argc
+		needed_size += sizeof(uintptr_t);
+
+		// argv
+		needed_size += (argv.size() + 1) * sizeof(uintptr_t);
+		for (auto arg : argv)
+			needed_size += arg.size() + 1;
+
+		// envp
+		needed_size += (envp.size() + 1) * sizeof(uintptr_t);
+		for (auto env : envp)
+			needed_size += env.size() + 1;
+
+		// auxv
+		needed_size += auxv.size() * sizeof(LibELF::AuxiliaryVector);
+
+		if (auto rem = needed_size % alignof(char*))
+			needed_size += alignof(char*) - rem;
+
+		if (needed_size > stack_region.size())
+			return BAN::Error::from_errno(ENOBUFS);
+
+		vaddr_t vaddr = stack_region.vaddr() + stack_region.size() - needed_size;
+
+		const size_t page_count = BAN::Math::div_round_up<size_t>(needed_size, PAGE_SIZE);
+		for (size_t i = 0; i < page_count; i++)
+			TRY(stack_region.allocate_page_containing(vaddr + i * PAGE_SIZE, true));
+
+		const auto stack_copy_buf =
+			[&stack_region](BAN::ConstByteSpan buffer, vaddr_t vaddr) -> void
+			{
+				ASSERT(vaddr >= stack_region.vaddr());
+				ASSERT(vaddr + buffer.size() <= stack_region.vaddr() + stack_region.size());
+				MUST(stack_region.copy_data_to_region(vaddr - stack_region.vaddr(), buffer.data(), buffer.size()));
+			};
+
+		const auto stack_push_buf =
+			[&stack_copy_buf, &vaddr](BAN::ConstByteSpan buffer) -> void
+			{
+				stack_copy_buf(buffer, vaddr);
+				vaddr += buffer.size();
+			};
+
+		const auto stack_push_uint =
+			[&stack_push_buf](uintptr_t value) -> void
+			{
+				stack_push_buf(BAN::ConstByteSpan::from(value));
+			};
+
+		const auto stack_push_str =
+			[&stack_push_buf](BAN::StringView string) -> void
+			{
+				const uint8_t* string_u8 = reinterpret_cast<const uint8_t*>(string.data());
+				stack_push_buf(BAN::ConstByteSpan(string_u8, string.size() + 1));
+			};
+
+		// argc
+		stack_push_uint(argv.size());
+
+		// argv
+		const vaddr_t argv_vaddr = vaddr;
+		vaddr += argv.size() * sizeof(uintptr_t);
+		stack_push_uint(0);
+
+		// envp
+		const vaddr_t envp_vaddr = vaddr;
+		vaddr += envp.size() * sizeof(uintptr_t);
+		stack_push_uint(0);
+
+		// auxv
+		for (auto aux : auxv)
+			stack_push_buf(BAN::ConstByteSpan::from(aux));
+
+		// information
+		for (size_t i = 0; i < argv.size(); i++)
+		{
+			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), argv_vaddr + i * sizeof(uintptr_t));
+			stack_push_str(argv[i]);
+		}
+		for (size_t i = 0; i < envp.size(); i++)
+		{
+			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), envp_vaddr + i * sizeof(uintptr_t));
+			stack_push_str(envp[i]);
+		}
+
+		return stack_region.vaddr() + stack_region.size() - needed_size;
+	}
+
 	Process::Process(const Credentials& credentials, pid_t pid, pid_t parent, pid_t sid, pid_t pgrp)
 		: m_credentials(credentials)
 		, m_open_file_descriptors(m_credentials)
@@ -223,7 +367,7 @@ namespace Kernel
 	Process::~Process()
 	{
 		ASSERT(m_threads.empty());
-		ASSERT(m_exited_pthreads.empty());
+		ASSERT(m_exited_threads.empty());
 		ASSERT(m_mapped_regions.empty());
 		ASSERT(!m_page_table);
 	}
@@ -245,12 +389,6 @@ namespace Kernel
 		return valid_pgrp;
 	}
 
-	void Process::add_thread(Thread* thread)
-	{
-		LockGuard _(m_process_lock);
-		MUST(m_threads.push_back(thread));
-	}
-
 	void Process::cleanup_function(Thread* thread)
 	{
 		{
@@ -271,7 +409,7 @@ namespace Kernel
 			}
 		}
 
-		m_exited_pthreads.clear();
+		m_exited_threads.clear();
 
 		ProcFileSystem::get().on_process_delete(*this);
 
@@ -283,21 +421,14 @@ namespace Kernel
 		// NOTE: We must unmap ranges while the page table is still alive
 		m_mapped_regions.clear();
 
+		// After we give our page table to the thread, we cannot get rescheduled
+		Processor::set_interrupt_state(InterruptState::Disabled);
 		thread->give_keep_alive_page_table(BAN::move(m_page_table));
 	}
 
 	bool Process::on_thread_exit(Thread& thread)
 	{
-		{
-			RWLockWRGuard _(m_memory_region_lock);
-
-			const size_t index = find_mapped_region(thread.userspace_stack().vaddr());
-			ASSERT(m_mapped_regions[index].ptr() == thread.m_userspace_stack);
-
-			m_mapped_regions.remove(index);
-
-			thread.m_userspace_stack = nullptr;
-		}
+		// TODO: if main thread exists, should we delete its stack?
 
 		LockGuard _(m_process_lock);
 
@@ -410,52 +541,41 @@ namespace Kernel
 			O_EXEC | O_RDWR
 		));
 
-		BAN::Vector<uint8_t> temp_buffer;
-		TRY(temp_buffer.resize(BAN::Math::min<size_t>(master_size, PAGE_SIZE)));
-
 		size_t bytes_copied = 0;
 		while (bytes_copied < master_size)
 		{
-			const size_t to_copy = BAN::Math::min(master_size - bytes_copied, temp_buffer.size());
+			uint8_t buffer[PAGE_SIZE];
+			const size_t to_copy = BAN::Math::min(master_size - bytes_copied, sizeof(buffer));
 
 			const vaddr_t vaddr = master_addr + bytes_copied;
 			const paddr_t paddr = page_table.physical_address_of(vaddr & PAGE_ADDR_MASK);
 			PageTable::with_fast_page(paddr, [&] {
-				memcpy(temp_buffer.data(), PageTable::fast_page_as_ptr(vaddr % PAGE_SIZE), to_copy);
+				memcpy(buffer, PageTable::fast_page_as_ptr(vaddr % PAGE_SIZE), to_copy);
 			});
 
-			TRY(region->copy_data_to_region(bytes_copied, temp_buffer.data(), to_copy));
+			TRY(region->copy_data_to_region(bytes_copied, buffer, to_copy));
 			bytes_copied += to_copy;
 		}
 
-		auto uthread = TRY(BAN::UniqPtr<struct uthread>::create());
-		*uthread = {
-			.self = reinterpret_cast<struct uthread*>(region->vaddr() + master_size),
-			.master_tls_addr = reinterpret_cast<void*>(master_addr),
-			.master_tls_size = master_size,
-			.master_tls_module_count = 1,
-			.dynamic_tls = nullptr,
-			.cleanup_stack = nullptr,
-			.id = 0,
-			.errno_ = 0,
-			.cancel_type = 0,
-			.cancel_state = 0,
-			.canceled = 0,
-			.specific_keys = {},
-			.specific_values = {},
-			.dtv = { 0, region->vaddr() }
-		};
+		uthread uthread {};
+		uthread.self = reinterpret_cast<struct uthread*>(region->vaddr() + master_size);
+		uthread.master_tls_addr = reinterpret_cast<void*>(master_addr),
+		uthread.master_tls_size = master_size,
+		uthread.master_tls_module_count = 1;
+		uthread.dynamic_tls = nullptr;
+		uthread.dtv[0] = 0;
+		uthread.dtv[1] = region->vaddr();
 
 		TRY(region->copy_data_to_region(
 			master_size,
-			reinterpret_cast<const uint8_t*>(uthread.ptr()),
+			reinterpret_cast<const uint8_t*>(&uthread),
 			sizeof(struct uthread)
 		));
 
-		TLSResult result;
-		result.addr = region->vaddr() + master_size;;
-		result.region = BAN::move(region);
-		return result;
+		return TLSResult {
+			.region = BAN::move(region),
+			.addr   = region->vaddr() + master_size,
+		};
 	}
 
 	BAN::ErrorOr<void> Process::add_mapped_region(BAN::UniqPtr<MemoryRegion>&& region)
@@ -574,6 +694,25 @@ namespace Kernel
 		return read_from_vec_of_str(m_environ, offset, buffer);
 	}
 
+	size_t Process::proc_cputime(off_t offset, BAN::ByteSpan buffer) const
+	{
+		const uint64_t cpu_time_ns = [this] {
+			uint64_t cpu_time_ns { 0 };
+			LockGuard _(m_process_lock);
+			for (auto* thread : m_threads)
+				cpu_time_ns += thread->cpu_time_ns();
+			return cpu_time_ns;
+		}();
+
+		auto data = MUST(BAN::String::formatted("{}", cpu_time_ns));
+		if (static_cast<size_t>(offset) >= data.size() + 1)
+			return 0;
+
+		const size_t to_copy = BAN::Math::min<size_t>(data.size() - offset + 1, buffer.size());
+		memcpy(buffer.data(), data.data(), to_copy);
+		return to_copy;
+	}
+
 	BAN::ErrorOr<BAN::String> Process::proc_cwd() const
 	{
 		LockGuard _(m_process_lock);
@@ -724,13 +863,30 @@ namespace Kernel
 		);
 
 		Process* forked = create_process(m_credentials, m_pid, m_sid, m_pgrp);
+		BAN::ScopeGuard process_deleter([forked] {
+			forked->m_page_table.clear();
+			delete forked;
+		});
+		forked->m_page_table = BAN::move(page_table);
+
+		Thread* thread = TRY(Thread::current().clone(forked, sp, ip));
+		BAN::ScopeGuard thread_deleter([thread] { delete thread; });
+
+		// NOTE: make sure the last two `MUST`s don't fail
+		TRY(forked->m_threads.reserve(1));
+		TRY(Processor::scheduler().bind_thread_to_processor(thread, Processor::current_id()));
+
+		{
+			SpinLockGuard _(s_process_lock);
+			TRY(s_processes.push_back(forked));
+		}
+
 		forked->m_controlling_terminal = m_controlling_terminal;
 		forked->m_working_directory = BAN::move(working_directory);
 		forked->m_root_file = BAN::move(root_file);
 		forked->m_cmdline = BAN::move(cmdline);
 		forked->m_environ = BAN::move(environ);
 		forked->m_executable = BAN::move(executable);
-		forked->m_page_table = BAN::move(page_table);
 		forked->m_shared_page_vaddr = BAN::move(shared_page_vaddr);
 		forked->m_open_file_descriptors = BAN::move(*open_file_descriptors);
 		forked->m_mapped_regions = BAN::move(mapped_regions);
@@ -742,10 +898,12 @@ namespace Kernel
 		child_exit_status->pgrp = forked->pgrp();
 
 		ASSERT(this == &Process::current());
-		// FIXME: this should be able to fail
-		Thread* thread = MUST(Thread::current().clone(forked, sp, ip));
-		forked->add_thread(thread);
-		forked->register_to_scheduler();
+
+		MUST(forked->m_threads.push_back(thread));
+		MUST(Processor::scheduler().add_thread(thread));
+
+		process_deleter.disable();
+		thread_deleter.disable();
 
 		return forked->pid();
 	}
@@ -801,8 +959,26 @@ namespace Kernel
 			BAN::String executable_path;
 			TRY(executable_path.append(executable_file.canonical_path));
 
+			const vaddr_t shared_page_vaddr = new_page_table->reserve_free_page(executable.interp_base.value_or(0x400000), USERSPACE_END);
+			if (shared_page_vaddr == 0)
+				return BAN::Error::from_errno(ENOMEM);
+			new_page_table->map_page_at(
+				Processor::shared_page_paddr(),
+				shared_page_vaddr,
+				PageTable::UserSupervisor | PageTable::Present
+			);
+
+			auto userspace_stack = TRY(MemoryBackedRegion::create(
+				*new_page_table,
+				Thread::userspace_stack_size,
+				{ Thread::userspace_stack_base, USERSPACE_END },
+				MemoryRegion::Type::PRIVATE,
+				PageTable::UserSupervisor | PageTable::ReadWrite | PageTable::Present,
+				O_RDWR
+			));
+
 			BAN::Vector<LibELF::AuxiliaryVector> auxiliary_vector;
-			TRY(auxiliary_vector.reserve(1 + executable.open_execfd));
+			TRY(auxiliary_vector.reserve(5 + 2 * executable.interp_base.has_value()));
 
 			BAN::ScopeGuard execfd_guard([this, &auxiliary_vector] {
 				if (auxiliary_vector.empty())
@@ -812,64 +988,94 @@ namespace Kernel
 				MUST(m_open_file_descriptors.close(auxiliary_vector.front().a_un.a_val));
 			});
 
-			if (executable.open_execfd)
+			if (executable.interp_base.has_value())
 			{
 				const int execfd = TRY(m_open_file_descriptors.open(BAN::move(executable_file), O_RDONLY));
 				TRY(auxiliary_vector.push_back({
 					.a_type = LibELF::AT_EXECFD,
 					.a_un = { .a_val = static_cast<uint32_t>(execfd) },
 				}));
+				TRY(auxiliary_vector.push_back({
+					.a_type = LibELF::AT_BASE,
+					.a_un = { .a_ptr = reinterpret_cast<void*>(executable.interp_base.value()) },
+				}));
 			}
-
-			const vaddr_t shared_page_vaddr = new_page_table->reserve_free_page(new_mapped_regions.back()->vaddr(), USERSPACE_END);
-			if (shared_page_vaddr == 0)
-				return BAN::Error::from_errno(ENOMEM);
-			new_page_table->map_page_at(
-				Processor::shared_page_paddr(),
-				shared_page_vaddr,
-				PageTable::UserSupervisor | PageTable::Present
-			);
+			TRY(auxiliary_vector.push_back({
+				.a_type = LibELF::AT_PAGESZ,
+				.a_un = { .a_val = PAGE_SIZE },
+			}));
 
 			TRY(auxiliary_vector.push_back({
 				.a_type = LibELF::AT_SHARED_PAGE,
 				.a_un = { .a_ptr = reinterpret_cast<void*>(shared_page_vaddr) },
 			}));
 
+			TRY(auxiliary_vector.push_back({
+				.a_type = LibELF::AT_STACK_BASE,
+				.a_un = { .a_ptr = reinterpret_cast<void*>(userspace_stack->vaddr()) },
+			}));
+
+			TRY(auxiliary_vector.push_back({
+				.a_type = LibELF::AT_STACK_SIZE,
+				.a_un = { .a_ptr = reinterpret_cast<void*>(userspace_stack->size()) },
+			}));
+
 			TRY(auxiliary_vector.push_back({
 				.a_type = LibELF::AT_NULL,
 				.a_un = { .a_val = 0 },
 			}));
 
-			// This is ugly but thread insterts userspace stack to process' memory region
-			BAN::swap(m_mapped_regions, new_mapped_regions);
-			auto new_thread_or_error = Thread::create_userspace(this, *new_page_table);
-			BAN::swap(m_mapped_regions, new_mapped_regions);
-
-			auto* new_thread = TRY(new_thread_or_error);
-			TRY(new_thread->initialize_userspace(
-				executable.entry_point,
+			const vaddr_t initial_stack_pointer = TRY(setup_initial_process_stack(
+				*userspace_stack,
 				str_argv.span(),
 				str_envp.span(),
 				auxiliary_vector.span()
 			));
 
+			const vaddr_t userspace_stack_vaddr = userspace_stack->vaddr();
+			const vaddr_t userspace_stack_size  = userspace_stack->size();
+			TRY(new_mapped_regions.emplace_back(BAN::move(userspace_stack)));
+
+			BAN::Optional<vaddr_t> tls_addr;
 			if (executable.master_tls.has_value())
 			{
 				auto tls_result = TRY(initialize_thread_local_storage(*new_page_table, *executable.master_tls));
 				TRY(new_mapped_regions.emplace_back(BAN::move(tls_result.region)));
-#if ARCH(x86_64)
-				new_thread->set_fsbase(tls_result.addr);
-#elif ARCH(i686)
-				new_thread->set_gsbase(tls_result.addr);
-#else
-#error
-#endif
+				tls_addr = tls_result.addr;
 			}
 
 			BAN::sort::sort(new_mapped_regions.begin(), new_mapped_regions.end(), [](auto& a, auto& b) {
 				return a->vaddr() < b->vaddr();
 			});
 
+			auto* new_thread = TRY(Thread::create_userspace(
+				this,
+				*new_page_table,
+				userspace_stack_vaddr,
+				userspace_stack_size,
+				executable.entry_point,
+				initial_stack_pointer
+			));
+			if (tls_addr.has_value())
+			{
+#if ARCH(x86_64)
+				new_thread->set_fsbase(tls_addr.value());
+#elif ARCH(i686)
+				new_thread->set_gsbase(tls_addr.value());
+#else
+#error
+#endif
+			}
+
+			// NOTE: bind new thread to this processor so it wont be rescheduled before end of this function
+			//       and so that adding the thread to the scheduler cannot fail
+			if (auto ret = Scheduler::bind_thread_to_processor(new_thread, Processor::current_id()); ret.is_error())
+			{
+				Processor::set_interrupt_state(InterruptState::Enabled);
+				delete new_thread;
+				return ret.release_error();
+			}
+
 			RWLockWRGuard wr_guard(m_memory_region_lock);
 
 			// NOTE: this is done before disabling interrupts and moving the threads as
@@ -880,13 +1086,6 @@ namespace Kernel
 			ASSERT(Processor::get_interrupt_state() == InterruptState::Enabled);
 			Processor::set_interrupt_state(InterruptState::Disabled);
 
-			// NOTE: bind new thread to this processor so it wont be rescheduled before end of this function
-			if (auto ret = Scheduler::bind_thread_to_processor(new_thread, Processor::current_id()); ret.is_error())
-			{
-				Processor::set_interrupt_state(InterruptState::Enabled);
-				return ret.release_error();
-			}
-
 			// after this point, everything is initialized and nothing can fail!
 
 			ASSERT(m_threads.size() == 1);
@@ -1009,52 +1208,45 @@ namespace Kernel
 		return child_pid;
 	}
 
-	BAN::ErrorOr<long> Process::sys_sleep(int seconds)
-	{
-		if (seconds == 0)
-			return 0;
-
-		const uint64_t wake_time_ms = SystemTimer::get().ms_since_boot() + (seconds * 1000);
-
-		while (!Thread::current().is_interrupted_by_signal())
-		{
-			const uint64_t current_ms = SystemTimer::get().ms_since_boot();
-			if (current_ms >= wake_time_ms)
-				break;
-			SystemTimer::get().sleep_ms(wake_time_ms - current_ms);
-		}
-
-		const uint64_t current_ms = SystemTimer::get().ms_since_boot();
-		if (current_ms < wake_time_ms)
-			return BAN::Math::div_round_up<long>(wake_time_ms - current_ms, 1000);
-		return 0;
-	}
-
 	BAN::ErrorOr<long> Process::sys_nanosleep(const timespec* user_rqtp, timespec* user_rmtp)
 	{
+		// take current timestamp ASAP
+		uint64_t current_ns = SystemTimer::get().ns_since_boot();
+
 		timespec rqtp;
 		TRY(read_from_user(user_rqtp, &rqtp, sizeof(timespec)));
 
-		if (rqtp.tv_nsec < 0 || rqtp.tv_nsec >= 1'000'000'000)
+		if (rqtp.tv_sec < 0 || rqtp.tv_nsec < 0 || rqtp.tv_nsec >= 1'000'000'000)
 			return BAN::Error::from_errno(EINVAL);
 
-		const uint64_t sleep_ns = (rqtp.tv_sec * 1'000'000'000) + rqtp.tv_nsec;
-		if (sleep_ns == 0)
-			return 0;
+		const uint64_t waketime_ns = [&rqtp, current_ns]() -> uint64_t {
+			if (BAN::Math::will_multiplication_overflow<uint64_t>(rqtp.tv_sec, 1'000'000'000))
+				return BAN::numeric_limits<uint64_t>::max();
+			if (BAN::Math::will_addition_overflow<uint64_t>(rqtp.tv_sec * 1'000'000'000, rqtp.tv_nsec))
+				return BAN::numeric_limits<uint64_t>::max();
+			if (BAN::Math::will_addition_overflow<uint64_t>(rqtp.tv_sec * 1'000'000'000 + rqtp.tv_nsec, current_ns))
+				return BAN::numeric_limits<uint64_t>::max();
+			return rqtp.tv_sec * 1'000'000'000 + rqtp.tv_nsec + current_ns;
+		}();
 
-		const uint64_t wake_time_ns = SystemTimer::get().ns_since_boot() + sleep_ns;
-		SystemTimer::get().sleep_ns(sleep_ns);
-
-		const uint64_t current_ns = SystemTimer::get().ns_since_boot();
-		if (current_ns < wake_time_ns)
+		// NOTE: we shouldn't get woken up unless the thread is interrupted but there is no harm in looping
+		while (current_ns < waketime_ns && !Thread::current().is_interrupted_by_signal())
+		{
+			SystemTimer::get().sleep_until_ns(waketime_ns);
+			current_ns = SystemTimer::get().ns_since_boot();
+		}
+
+		if (current_ns < waketime_ns)
 		{
-			const uint64_t remaining_ns = wake_time_ns - current_ns;
-			const timespec remaining_ts = {
-				.tv_sec = static_cast<time_t>(remaining_ns / 1'000'000'000),
-				.tv_nsec = static_cast<long>(remaining_ns % 1'000'000'000),
-			};
 			if (user_rmtp != nullptr)
+			{
+				const uint64_t remaining_ns = waketime_ns - current_ns;
+				const timespec remaining_ts = {
+					.tv_sec = static_cast<time_t>(remaining_ns / 1'000'000'000),
+					.tv_nsec = static_cast<long>(remaining_ns % 1'000'000'000),
+				};
 				TRY(write_to_user(user_rmtp, &remaining_ts, sizeof(timespec)));
+			}
 			return BAN::Error::from_errno(EINTR);
 		}
 
@@ -1851,7 +2043,7 @@ namespace Kernel
 		return TRY(m_open_file_descriptors.sendmsg(socket, message, flags));
 	}
 
-	BAN::ErrorOr<long> Process::sys_ioctl(int fildes, int request, void* arg)
+	BAN::ErrorOr<long> Process::sys_ioctl(int fildes, unsigned long request, void* arg)
 	{
 		auto inode = TRY(m_open_file_descriptors.inode_of(fildes));
 		return TRY(inode->ioctl(request, arg));
@@ -2663,9 +2855,9 @@ namespace Kernel
 		if (prot & PROT_READ)
 			flags |= PageTable::Flags::Present;
 		if (prot & PROT_WRITE)
-			flags |= PageTable::Flags::ReadWrite | PageTable::Flags::Present;
+			flags |= PageTable::Flags::Present | PageTable::Flags::ReadWrite;
 		if (prot & PROT_EXEC)
-			flags |= PageTable::Flags::Execute | PageTable::Flags::Execute;
+			flags |= PageTable::Flags::Present | PageTable::Flags::Execute;
 
 		if (flags == 0)
 			flags = PageTable::Flags::Reserved;
@@ -2964,6 +3156,9 @@ namespace Kernel
 
 	void Process::wait_while_stopped()
 	{
+		if (!m_stopped) [[likely]]
+			return;
+
 		for (;;)
 		{
 			while (Thread::current().will_exit_because_of_signal())
@@ -3354,41 +3549,63 @@ namespace Kernel
 		return Thread::current().get_gsbase();
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_create(const pthread_attr_t* user_attr, void (*entry)(void*), void* arg)
+	BAN::ErrorOr<long> Process::sys_thread_create(void (*entry)(void*), void* arg, void* stack_base, size_t stack_size)
 	{
-		if (user_attr != nullptr)
-			dwarnln("TODO: ignoring thread attr");
+		const vaddr_t stack_vaddr = reinterpret_cast<vaddr_t>(stack_base);
+		if (stack_vaddr % PAGE_SIZE || stack_size % PAGE_SIZE)
+			return BAN::Error::from_errno(EINVAL);
 
-		LockGuard _(m_process_lock);
+		auto* memory_region = TRY(validate_and_pin_pointer_access(stack_base, stack_size, true));
+		BAN::ScopeGuard _0([memory_region] { if (memory_region) memory_region->unpin(); });
 
-		auto* new_thread = TRY(Thread::current().pthread_create(entry, arg));
-		MUST(m_threads.push_back(new_thread));
-		MUST(Processor::scheduler().add_thread(new_thread));
+		const vaddr_t initial_stack_pointer = stack_vaddr + stack_size - sizeof(void*);
+		*reinterpret_cast<void**>(initial_stack_pointer) = arg;
 
-		return new_thread->tid();
-	}
+		auto* thread = TRY(Thread::create_userspace(
+			this,
+			page_table(),
+			stack_vaddr,
+			stack_size,
+			reinterpret_cast<vaddr_t>(entry),
+			initial_stack_pointer
+		));
+		thread->m_signal_block_mask = Thread::current().m_signal_block_mask;
 
-	BAN::ErrorOr<long> Process::sys_pthread_exit(void* value)
-	{
-		LockGuard _(m_process_lock);
+		LockGuard _1(m_process_lock);
 
-		if (m_threads.size() == 1)
-			return sys_exit(0);
-
-		auto& thread = Thread::current();
-		if (!thread.is_detached())
+		TRY(m_threads.push_back(thread));
+		if (auto ret = Processor::scheduler().add_thread(thread); ret.is_error())
 		{
-			TRY(m_exited_pthreads.emplace_back(thread.tid(), value));
-			m_pthread_exit_blocker.unblock();
+			m_threads.pop_back();
+			delete thread;
+			return ret.release_error();
 		}
 
-		m_process_lock.unlock();
-		thread.on_exit();
+		return thread->tid();
+	}
+
+	BAN::ErrorOr<long> Process::sys_thread_exit(void* value)
+	{
+		{
+			LockGuard _(m_process_lock);
+
+			if (m_threads.size() == 1)
+				return sys_exit(0);
+
+			auto& thread = Thread::current();
+			if (!thread.is_detached())
+			{
+				TRY(m_exited_threads.emplace_back(thread.tid(), value));
+				m_thread_exit_blocker.unblock();
+			}
+		}
+
+		Thread::current().on_exit();
 
 		ASSERT_NOT_REACHED();
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_join(pthread_t tid, void** user_value)
+	BAN::ErrorOr<long> Process::sys_thread_join(pid_t tid, void** user_value)
 	{
 		LockGuard _(m_process_lock);
 
@@ -3398,12 +3615,12 @@ namespace Kernel
 		const auto check_thread =
 			[&]() -> BAN::Optional<void*>
 			{
-				for (size_t i = 0; i < m_exited_pthreads.size(); i++)
+				for (size_t i = 0; i < m_exited_threads.size(); i++)
 				{
-					if (m_exited_pthreads[i].thread != tid)
+					if (m_exited_threads[i].tid != tid)
 						continue;
-					void* ret = m_exited_pthreads[i].value;
-					m_exited_pthreads.remove(i);
+					void* ret = m_exited_threads[i].value;
+					m_exited_threads.remove(i);
 					return ret;
 				}
 
@@ -3436,16 +3653,16 @@ namespace Kernel
 					return BAN::Error::from_errno(EINVAL);
 			}
 
-			TRY(Thread::current().block_or_eintr_indefinite(m_pthread_exit_blocker, &m_process_lock));
+			TRY(Thread::current().block_or_eintr_indefinite(m_thread_exit_blocker, &m_process_lock));
 		}
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_self()
+	BAN::ErrorOr<long> Process::sys_thread_getid()
 	{
 		return Thread::current().tid();
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_kill(pthread_t tid, int signal)
+	BAN::ErrorOr<long> Process::sys_thread_kill(pid_t tid, int signal)
 	{
 		if (signal != 0 && (signal < _SIGMIN || signal > _SIGMAX))
 			return BAN::Error::from_errno(EINVAL);
@@ -3475,7 +3692,7 @@ namespace Kernel
 		return BAN::Error::from_errno(ESRCH);
 	}
 
-	BAN::ErrorOr<long> Process::sys_pthread_detach(pthread_t tid)
+	BAN::ErrorOr<long> Process::sys_thread_detach(pid_t tid)
 	{
 		LockGuard _(m_process_lock);
 
@@ -3486,7 +3703,7 @@ namespace Kernel
 			if (thread->is_detached())
 				return BAN::Error::from_errno(EINVAL);
 			thread->detach();
-			m_pthread_exit_blocker.unblock();
+			m_thread_exit_blocker.unblock();
 			return 0;
 		}
 

kernel/kernel/Processor.cpp

@@ -1,3 +1,4 @@
+#include <kernel/CPUID.h>
 #include <kernel/InterruptController.h>
 #include <kernel/Memory/Heap.h>
 #include <kernel/Memory/kmalloc.h>
@@ -20,6 +21,8 @@ namespace Kernel
 	static constexpr uint32_t MSR_IA32_FMASK = 0xC0000084;
 #endif
 
+	static constexpr uint32_t MSR_IA32_TSC_AUX = 0xC0000103;
+
 	ProcessorID          Processor::s_bsp_id                     { PROCESSOR_NONE };
 	BAN::Atomic<uint8_t> Processor::s_processor_count            { 0 };
 	BAN::Atomic<bool>    Processor::s_is_smp_enabled             { false };
@@ -219,6 +222,9 @@ namespace Kernel
 		shared_page.gdt_cpu_offset = GDT::cpu_index_offset();
 		shared_page.features = 0;
 
+		if (CPUID::has_rdtscp())
+			shared_page.features |= API::SPF_RDTSCP;
+
 		ASSERT(Processor::count() + sizeof(Kernel::API::SharedPage) <= PAGE_SIZE);
 	}
 
@@ -289,13 +295,11 @@ namespace Kernel
 		}
 	}
 
-	void Processor::initialize_tsc(uint8_t shift, uint64_t mult, uint64_t realtime_seconds)
+	void Processor::initialize_tsc(uint64_t realtime_seconds)
 	{
 		auto& shared_page = Processor::shared_page();
-
-		shared_page.gettime_shared.shift = shift;
-		shared_page.gettime_shared.mult = mult;
-		shared_page.gettime_shared.realtime_seconds = realtime_seconds;
+		shared_page.gettime_shared.realtime_s = realtime_seconds;
+		shared_page.gettime_shared.realtime_ns = 0;
 
 		update_tsc();
 
@@ -322,23 +326,63 @@ namespace Kernel
 
 	void Processor::update_tsc()
 	{
-		auto& sgettime = shared_page().cpus[current_index()].gettime_local;
-		sgettime.seq = sgettime.seq + 1;
-		sgettime.last_ns = SystemTimer::get().ns_since_boot_no_tsc();
-		sgettime.last_tsc = __builtin_ia32_rdtsc();
-		sgettime.seq = sgettime.seq + 1;
+		auto& lgettime = shared_page().cpus[current_index()].gettime_local;
+
+		const auto seq = BAN::atomic_load(lgettime.seq, BAN::memory_order_relaxed);
+
+		BAN::atomic_store(lgettime.seq, seq + 1, BAN::memory_order_release);
+
+		if (lgettime.seq == 1)
+		{
+			const auto tsc_info = SystemTimer::get().tsc_info();
+			lgettime.shift = tsc_info.shift;
+			lgettime.mult  = tsc_info.mult;
+			lgettime.last_ns  = SystemTimer::get().ns_since_boot_no_tsc();
+			lgettime.last_tsc = __builtin_ia32_rdtsc();
+
+			if (CPUID::has_rdtscp())
+				asm volatile("wrmsr" :: "d"(0x00000000), "a"(current_index()), "c"(MSR_IA32_TSC_AUX));
+		}
+		else
+		{
+			const auto current_ns = SystemTimer::get().ns_since_boot_no_tsc();
+			const auto current_tsc = __builtin_ia32_rdtsc();
+
+			auto delta_ns = current_tsc - lgettime.last_tsc;
+			if (lgettime.shift >= 0)
+				delta_ns <<= lgettime.shift;
+			else
+				delta_ns >>= -lgettime.shift;
+			delta_ns = (delta_ns * lgettime.mult) >> 32;
+
+			lgettime.last_ns += delta_ns;
+			lgettime.last_tsc = current_tsc;
+
+			// scale mult by [-0.25%, 0.25%] to fix for clock drift
+			const auto error_ns = static_cast<int64_t>(current_ns) - static_cast<int64_t>(lgettime.last_ns);
+			const auto correction_ppm = BAN::Math::clamp<int64_t>(error_ns * 1'000'000 / 1'000'000'000, -100, 100);
+			const auto correction_delta = -lgettime.mult * correction_ppm / 1'000'000;
+			lgettime.mult += correction_delta;
+		}
+
+		BAN::atomic_store(lgettime.seq, seq + 2, BAN::memory_order_release);
 	}
 
 	uint64_t Processor::ns_since_boot_tsc()
 	{
 		const auto& shared_page = Processor::shared_page();
-		const auto& sgettime = shared_page.gettime_shared;
 		const auto& lgettime = shared_page.cpus[current_index()].gettime_local;
 
 		auto state = get_interrupt_state();
 		set_interrupt_state(InterruptState::Disabled);
 
-		const auto current_ns = lgettime.last_ns + (((__builtin_ia32_rdtsc() - lgettime.last_tsc) * sgettime.mult) >> sgettime.shift);
+		uint64_t current_ns = __builtin_ia32_rdtsc() - lgettime.last_tsc;
+		if (lgettime.shift >= 0)
+			current_ns <<= lgettime.shift;
+		else
+			current_ns >>= -lgettime.shift;
+		current_ns = (current_ns * lgettime.mult) >> 32;
+		current_ns += lgettime.last_ns;
 
 		set_interrupt_state(state);
 

kernel/kernel/Scheduler.cpp

@@ -1,5 +1,6 @@
 #include <BAN/Optional.h>
 #include <BAN/Sort.h>
+#include <kernel/APIC.h>
 #include <kernel/InterruptController.h>
 #include <kernel/Lock/Mutex.h>
 #include <kernel/Process.h>
@@ -45,12 +46,15 @@ namespace Kernel
 		m_tail = node;
 	}
 
-	void SchedulerQueue::add_thread_with_wake_time(Node* node)
+	bool SchedulerQueue::add_thread_with_wake_time(Node* node)
 	{
 		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
 
 		if (m_tail == nullptr || node->wake_time_ns >= m_tail->wake_time_ns)
-			return add_thread_to_back(node);
+		{
+			add_thread_to_back(node);
+			return node == m_head;
+		}
 
 		Node* next = m_head;
 		Node* prev = nullptr;
@@ -64,6 +68,8 @@ namespace Kernel
 		node->prev = prev;
 		(next ? next->prev : m_tail) = node;
 		(prev ? prev->next : m_head) = node;
+
+		return node == m_head;
 	}
 
 	template<typename F>
@@ -122,15 +128,9 @@ namespace Kernel
 		m_idle_thread = TRY(Thread::create_kernel([](void*) { asm volatile("1: hlt; jmp 1b"); }, nullptr));
 		ASSERT(m_idle_thread);
 
-		size_t processor_index = 0;
-		for (; processor_index < Processor::count(); processor_index++)
-			if (Processor::id_from_index(processor_index) == Processor::current_id())
-				break;
-		ASSERT(processor_index < Processor::count());
-
 		// each CPU does load balance at different times. This calulates the offset to other CPUs
-		m_last_load_balance_ns = s_load_balance_interval_ns * processor_index / Processor::count();
-		m_idle_ns              = -m_last_load_balance_ns;
+		m_last_load_balance_ns = s_load_balance_interval_ns * Processor::current_index() / Processor::count();
+		m_idle_ns              = m_last_load_balance_ns;
 
 		s_schedulers_initialized++;
 		while (s_schedulers_initialized < Processor::count())
@@ -139,8 +139,6 @@ namespace Kernel
 		if (Processor::count() > 1)
 			Processor::set_smp_enabled();
 
-		m_next_reschedule_ns = SystemTimer::get().ns_since_boot();
-
 		return {};
 	}
 
@@ -232,17 +230,14 @@ namespace Kernel
 					m_thread_count--;
 					break;
 				case Thread::State::Executing:
-				{
-					const uint64_t current_ns = SystemTimer::get().ns_since_boot();
 					m_current->thread->yield_registers() = *yield_registers;
-					m_current->time_used_ns += current_ns - m_current->last_start_ns;
+					m_current->time_used_ns += SystemTimer::get().ns_since_boot() - m_current->last_start_ns;
 					add_current_to_most_loaded(m_current->blocked ? &m_block_queue : &m_run_queue);
 					if (!m_current->blocked)
 						m_run_queue.add_thread_to_back(m_current);
-					else
-						m_block_queue.add_thread_with_wake_time(m_current);
+					else if (m_block_queue.add_thread_with_wake_time(m_current))
+						update_wake_up_deadline();
 					break;
-				}
 				case Thread::State::NotStarted:
 					ASSERT(!m_current->blocked);
 					m_current->time_used_ns = 0;
@@ -321,26 +316,50 @@ namespace Kernel
 		}
 	}
 
+	void Scheduler::update_wake_up_deadline()
+	{
+		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
+
+		auto& interrupt_controller = InterruptController::get();
+
+		// TODO: support timer deadlines on non-apic timers and abstract it :)
+		if (!interrupt_controller.is_using_apic())
+			return;
+
+		wake_up_sleeping_threads();
+
+		uint64_t deadline_ns = m_next_reschedule_ns;
+		if (!m_block_queue.empty())
+			deadline_ns = BAN::Math::min(deadline_ns, m_block_queue.front()->wake_time_ns);
+		if (Processor::is_smp_enabled())
+			deadline_ns = BAN::Math::min(deadline_ns, m_last_load_balance_ns + s_load_balance_interval_ns);
+
+		static_cast<APIC&>(interrupt_controller).set_timer_dealine(deadline_ns);
+	}
+
 	void Scheduler::reschedule_if_idle()
 	{
 		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
 
-		if (m_current != nullptr)
-			return;
-
-		if (m_run_queue.empty())
-			wake_up_sleeping_threads();
-
-		if (!m_run_queue.empty())
+		if (m_current == nullptr && !m_run_queue.empty())
 			Processor::yield();
 	}
 
-	extern "C" void scheduler_on_yield(YieldRegisters* yield_registers)
+	extern "C" void scheduler_on_yield_trampoline(YieldRegisters* yield_registers)
 	{
-		Processor::scheduler().reschedule(yield_registers);
+		Processor::scheduler().on_yield(yield_registers);
 	}
 
-	void Scheduler::timer_interrupt()
+	void Scheduler::on_yield(YieldRegisters* yield_registers)
+	{
+		reschedule(yield_registers);
+		m_next_reschedule_ns = !is_idle()
+			? SystemTimer::get().ns_since_boot() + s_reschedule_interval_ns
+			: BAN::numeric_limits<uint64_t>::max();
+		update_wake_up_deadline();
+	}
+
+	void Scheduler::on_timer_interrupt()
 	{
 		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
 
@@ -349,11 +368,8 @@ namespace Kernel
 
 		wake_up_sleeping_threads();
 
-		if (SystemTimer::get().ns_since_boot() >= m_next_reschedule_ns)
-		{
-			m_next_reschedule_ns += s_reschedule_interval_ns;
+		if (is_idle() || SystemTimer::get().ns_since_boot() >= m_next_reschedule_ns)
 			Processor::yield();
-		}
 	}
 
 	void Scheduler::unblock_thread(SchedulerQueue::Node* node)
@@ -394,8 +410,8 @@ namespace Kernel
 
 		if (!node->blocked)
 			m_run_queue.add_thread_to_back(node);
-		else
-			m_block_queue.add_thread_with_wake_time(node);
+		else if (m_block_queue.add_thread_with_wake_time(node))
+			update_wake_up_deadline();
 
 		if (auto* thread = node->thread; thread->is_userspace() && thread->has_process())
 			thread->update_processor_index_address();

kernel/kernel/Storage/ATA/AHCI/Controller.cpp

@@ -7,6 +7,16 @@
 namespace Kernel
 {
 
+	BAN::ErrorOr<BAN::RefPtr<AHCIController>> AHCIController::create(PCI::Device& pci_device)
+	{
+		auto* controller_ptr = new AHCIController(pci_device);
+		if (controller_ptr == nullptr)
+			return BAN::Error::from_errno(ENOMEM);
+		auto controller = BAN::RefPtr<AHCIController>::adopt(controller_ptr);
+		TRY(controller->initialize());
+		return controller;
+	}
+
 	BAN::ErrorOr<void> AHCIController::initialize()
 	{
 		m_abar = TRY(m_pci_device.allocate_bar_region(5));

kernel/kernel/Storage/ATA/ATABus.cpp

@@ -66,7 +66,7 @@ namespace Kernel
 
 	static void select_delay()
 	{
-		SystemTimer::get().sleep_ns(400);
+		SystemTimer::get().sleep_for_ns(400);
 	}
 
 	void ATABus::select_device(bool is_secondary)
@@ -87,7 +87,7 @@ namespace Kernel
 		io_write(ATA_PORT_LBA1, 0);
 		io_write(ATA_PORT_LBA2, 0);
 		io_write(ATA_PORT_COMMAND, ATA_COMMAND_IDENTIFY);
-		SystemTimer::get().sleep_ms(1);
+		SystemTimer::get().sleep_for_ms(1);
 
 		// No device on port
 		if (io_read(ATA_PORT_STATUS) == 0)
@@ -112,7 +112,7 @@ namespace Kernel
 			}
 
 			io_write(ATA_PORT_COMMAND, ATA_COMMAND_IDENTIFY_PACKET);
-			SystemTimer::get().sleep_ms(1);
+			SystemTimer::get().sleep_for_ms(1);
 		}
 
 		TRY(wait(true));

kernel/kernel/Storage/ATA/ATAController.cpp

@@ -9,29 +9,12 @@
 namespace Kernel
 {
 
-	BAN::ErrorOr<BAN::RefPtr<StorageController>> ATAController::create(PCI::Device& pci_device)
+	BAN::ErrorOr<BAN::RefPtr<ATAController>> ATAController::create(PCI::Device& pci_device)
 	{
-		StorageController* controller_ptr = nullptr;
-
-		switch (pci_device.subclass())
-		{
-			case 0x01:
-				controller_ptr = new ATAController(pci_device);
-				break;
-			case 0x05:
-				dwarnln("unsupported DMA ATA Controller");
-				return BAN::Error::from_errno(ENOTSUP);
-			case 0x06:
-				controller_ptr = new AHCIController(pci_device);
-				break;
-			default:
-				ASSERT_NOT_REACHED();
-		}
-
+		auto* controller_ptr = new ATAController(pci_device);
 		if (controller_ptr == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
-
-		auto controller = BAN::RefPtr<StorageController>::adopt(controller_ptr);
+		auto controller = BAN::RefPtr<ATAController>::adopt(controller_ptr);
 		TRY(controller->initialize());
 		return controller;
 	}

kernel/kernel/Storage/NVMe/Controller.cpp

@@ -16,12 +16,12 @@ namespace Kernel
 		return minor++;
 	}
 
-	BAN::ErrorOr<BAN::RefPtr<StorageController>> NVMeController::create(PCI::Device& pci_device)
+	BAN::ErrorOr<BAN::RefPtr<NVMeController>> NVMeController::create(PCI::Device& pci_device)
 	{
 		auto* controller_ptr = new NVMeController(pci_device);
 		if (controller_ptr == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
-		auto controller = BAN::RefPtr<StorageController>::adopt(controller_ptr);
+		auto controller = BAN::RefPtr<NVMeController>::adopt(controller_ptr);
 		TRY(controller->initialize());
 		return controller;
 	}
@@ -115,8 +115,6 @@ namespace Kernel
 
 		DevFileSystem::get().add_device(this);
 
-		StorageController::ref();
-
 		return {};
 	}
 

kernel/kernel/Terminal/PseudoTerminal.cpp

@@ -157,7 +157,7 @@ namespace Kernel
 			slave->m_write_blocker.unblock();
 	}
 
-	BAN::ErrorOr<long> PseudoTerminalMaster::ioctl_impl(int request, void* argument)
+	BAN::ErrorOr<long> PseudoTerminalMaster::ioctl_impl(unsigned long request, void* argument)
 	{
 		switch (request)
 		{
@@ -174,7 +174,7 @@ namespace Kernel
 				return BAN::Error::from_errno(ENODEV);
 		}
 
-		return CharacterDevice::ioctl(request, argument);
+		return CharacterDevice::ioctl_impl(request, argument);
 	}
 
 	PseudoTerminalSlave::PseudoTerminalSlave(BAN::String&& name, uint32_t number, mode_t mode, uid_t uid, gid_t gid)

kernel/kernel/Terminal/TTY.cpp

@@ -133,7 +133,7 @@ namespace Kernel
 		(void)Process::kill(-m_foreground_pgrp, SIGWINCH);
 	}
 
-	BAN::ErrorOr<long> TTY::ioctl_impl(int request, void* argument)
+	BAN::ErrorOr<long> TTY::ioctl_impl(unsigned long request, void* argument)
 	{
 		switch (request)
 		{
@@ -204,7 +204,7 @@ namespace Kernel
 			}
 		}
 
-		return CharacterDevice::ioctl(request, argument);
+		return CharacterDevice::ioctl_impl(request, argument);
 	}
 
 	void TTY::on_key_event(LibInput::RawKeyEvent event)
@@ -449,9 +449,14 @@ namespace Kernel
 	void TTY::putchar_current(uint8_t ch)
 	{
 		ASSERT(s_tty);
-		LockGuard _(s_tty->m_write_lock);
+
+		while (!s_tty->m_write_lock.try_lock())
+			Processor::pause();
+
 		s_tty->putchar(ch);
 		s_tty->after_write();
+
+		s_tty->m_write_lock.unlock();
 	}
 
 	bool TTY::is_initialized()

kernel/kernel/Terminal/VirtualTTY.cpp

@@ -137,25 +137,25 @@ namespace Kernel
 			case 7:  m_colors_inverted = true;  break;
 			case 27: m_colors_inverted = false; break;
 
-			case 30: case 31: case 32: case 33: case 34: case 35: case 36: case 37:
+			case 30 ... 37:
 				m_foreground = m_palette[value - 30];
 				break;
 			case 39:
 				m_foreground = m_palette[15];
 				break;
 
-			case 40: case 41: case 42: case 43: case 44: case 45: case 46: case 47:
+			case 40 ... 47:
 				m_background = m_palette[value - 40];
 				break;
 			case 49:
 				m_background = m_palette[0];
 				break;
 
-			case 90: case 91: case 92: case 93: case 94: case 95: case 96: case 97:
+			case 90 ... 97:
 				m_foreground = m_palette[value - 90 + 8];
 				break;
 
-			case 100: case 101: case 102: case 103: case 104: case 105: case 106: case 107:
+			case 100 ... 107:
 				m_background = m_palette[value - 100 + 8];
 				break;
 
@@ -204,9 +204,7 @@ namespace Kernel
 		ASSERT(m_write_lock.is_locked_by_current_thread());
 		switch (ch)
 		{
-			case '0': case '1': case '2': case '3': case '4':
-			case '5': case '6': case '7': case '8': case '9':
-			{
+			case '0' ... '9':
 				if (m_ansi_state.index >= m_ansi_state.max_nums)
 					dwarnln("Only {} arguments supported with ANSI codes", m_ansi_state.max_nums);
 				else
@@ -215,7 +213,6 @@ namespace Kernel
 					val = (val == -1) ? (ch - '0') : (val * 10 + ch - '0');
 				}
 				return;
-			}
 			case ';':
 				m_ansi_state.index = BAN::Math::min<size_t>(m_ansi_state.index + 1, m_ansi_state.max_nums);
 				return;

kernel/kernel/Thread.cpp

@@ -15,15 +15,9 @@
 namespace Kernel
 {
 
-	static_assert(SYS_SIGPROCMASK == 79, "this is hard coded in arch/*/Signal.S");
+	static_assert(SYS_SIGPROCMASK == 78, "this is hard coded in arch/*/Signal.S");
 	static_assert(SIG_SETMASK     ==  3, "this is hard coded in arch/*/Signal.S");
 
-#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();
 
@@ -198,7 +192,7 @@ namespace Kernel
 		return thread;
 	}
 
-	BAN::ErrorOr<Thread*> Thread::create_userspace(Process* process, PageTable& page_table)
+	BAN::ErrorOr<Thread*> Thread::create_userspace(Process* process, PageTable& page_table, vaddr_t userspace_stack_vaddr, size_t userspace_stack_size, vaddr_t entry_point, vaddr_t stack_pointer)
 	{
 		ASSERT(process);
 
@@ -212,22 +206,32 @@ namespace Kernel
 
 		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
 			page_table,
-			{ s_user_stack_addr_start, USERSPACE_END },
+			{ userspace_stack_base, USERSPACE_END },
 			kernel_stack_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			true
 		));
 
-		auto userspace_stack = TRY(MemoryBackedRegion::create(
-			page_table,
-			userspace_stack_size,
-			{ s_user_stack_addr_start, USERSPACE_END },
-			MemoryRegion::Type::PRIVATE,
-			PageTable::Flags::UserSupervisor | PageTable::Flags::ReadWrite | PageTable::Flags::Present,
-			O_RDWR
-		));
-		thread->m_userspace_stack = userspace_stack.ptr();
-		TRY(process->add_mapped_region(BAN::move(userspace_stack)));
+		thread->m_userspace_stack_vaddr = userspace_stack_vaddr;
+		thread->m_userspace_stack_size  = userspace_stack_size;
+
+		// Initialize stack for returning
+		PageTable::with_fast_page(thread->kernel_stack().paddr_of(thread->kernel_stack_top() - PAGE_SIZE), [=] {
+			uintptr_t cur_sp = PageTable::fast_page() + PAGE_SIZE;
+			write_to_stack(cur_sp, 0x20 | 3);
+			write_to_stack(cur_sp, stack_pointer);
+			write_to_stack(cur_sp, 0x202);
+#if ARCH(x86_64)
+			write_to_stack(cur_sp, 0x28 | 3);
+#elif ARCH(i686)
+			write_to_stack(cur_sp, 0x18 | 3);
+#endif
+			write_to_stack(cur_sp, entry_point);
+		});
+
+		thread->m_yield_registers = {};
+		thread->m_yield_registers.ip = reinterpret_cast<vaddr_t>(start_userspace_thread);
+		thread->m_yield_registers.sp = thread->kernel_stack_top() - 5 * sizeof(uintptr_t);
 
 		thread_deleter.disable();
 
@@ -311,26 +315,6 @@ namespace Kernel
 		Processor::gdt().set_cpu_index(Processor::current_index());
 	}
 
-	BAN::ErrorOr<Thread*> Thread::pthread_create(entry_t entry, void* arg)
-	{
-		auto* thread = TRY(create_userspace(m_process, m_process->page_table()));
-
-		if (Processor::get_current_sse_thread() == this)
-			save_sse();
-		memcpy(thread->m_sse_storage, m_sse_storage, sizeof(m_sse_storage));
-
-		TRY(thread->userspace_stack().copy_data_to_region(
-			thread->m_userspace_stack->size() - sizeof(void*),
-			reinterpret_cast<const uint8_t*>(&arg),
-			sizeof(void*)
-		));
-
-		const vaddr_t entry_addr = reinterpret_cast<vaddr_t>(entry);
-		thread->setup_exec(entry_addr, thread->userspace_stack().vaddr() + thread->userspace_stack().size() - sizeof(void*));
-
-		return thread;
-	}
-
 	BAN::ErrorOr<Thread*> Thread::clone(Process* new_process, uintptr_t sp, uintptr_t ip)
 	{
 		ASSERT(m_is_userspace);
@@ -345,7 +329,7 @@ namespace Kernel
 
 		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
 			new_process->page_table(),
-			{ s_user_stack_addr_start, USERSPACE_END },
+			{ userspace_stack_base, USERSPACE_END },
 			kernel_stack_size,
 			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
 			true
@@ -362,9 +346,8 @@ namespace Kernel
 			);
 		});
 
-		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());
-		ASSERT(thread->m_userspace_stack->vaddr() == m_userspace_stack->vaddr());
+		thread->m_userspace_stack_vaddr = m_userspace_stack_vaddr;
+		thread->m_userspace_stack_size  = m_userspace_stack_size;
 
 		thread->m_fsbase = m_fsbase;
 		thread->m_gsbase = m_gsbase;
@@ -385,151 +368,23 @@ namespace Kernel
 		return thread;
 	}
 
-	BAN::ErrorOr<void> Thread::initialize_userspace(vaddr_t entry, BAN::Span<BAN::String> argv, BAN::Span<BAN::String> envp, BAN::Span<LibELF::AuxiliaryVector> auxv)
-	{
-		// System V ABI: Initial process stack
-
-		ASSERT(m_is_userspace);
-		ASSERT(m_userspace_stack);
-
-		size_t needed_size = 0;
-
-		// argc
-		needed_size += sizeof(uintptr_t);
-
-		// argv
-		needed_size += (argv.size() + 1) * sizeof(uintptr_t);
-		for (auto arg : argv)
-			needed_size += arg.size() + 1;
-
-		// envp
-		needed_size += (envp.size() + 1) * sizeof(uintptr_t);
-		for (auto env : envp)
-			needed_size += env.size() + 1;
-
-		// auxv
-		needed_size += auxv.size() * sizeof(LibELF::AuxiliaryVector);
-
-		if (auto rem = needed_size % alignof(char*))
-			needed_size += alignof(char*) - rem;
-
-		if (needed_size > m_userspace_stack->size())
-			return BAN::Error::from_errno(ENOBUFS);
-
-		vaddr_t vaddr = userspace_stack().vaddr() + userspace_stack().size() - needed_size;
-
-		const size_t page_count = BAN::Math::div_round_up<size_t>(needed_size, PAGE_SIZE);
-		for (size_t i = 0; i < page_count; i++)
-			TRY(m_userspace_stack->allocate_page_containing(vaddr + i * PAGE_SIZE, true));
-
-		const auto stack_copy_buf =
-			[this](BAN::ConstByteSpan buffer, vaddr_t vaddr) -> void
-			{
-				ASSERT(vaddr >= m_userspace_stack->vaddr());
-				ASSERT(vaddr + buffer.size() <= m_userspace_stack->vaddr() + m_userspace_stack->size());
-				MUST(m_userspace_stack->copy_data_to_region(vaddr - m_userspace_stack->vaddr(), buffer.data(), buffer.size()));
-			};
-
-		const auto stack_push_buf =
-			[&stack_copy_buf, &vaddr](BAN::ConstByteSpan buffer) -> void
-			{
-				stack_copy_buf(buffer, vaddr);
-				vaddr += buffer.size();
-			};
-
-		const auto stack_push_uint =
-			[&stack_push_buf](uintptr_t value) -> void
-			{
-				stack_push_buf(BAN::ConstByteSpan::from(value));
-			};
-
-		const auto stack_push_str =
-			[&stack_push_buf](BAN::StringView string) -> void
-			{
-				const uint8_t* string_u8 = reinterpret_cast<const uint8_t*>(string.data());
-				stack_push_buf(BAN::ConstByteSpan(string_u8, string.size() + 1));
-			};
-
-		// argc
-		stack_push_uint(argv.size());
-
-		// argv
-		const vaddr_t argv_vaddr = vaddr;
-		vaddr += argv.size() * sizeof(uintptr_t);
-		stack_push_uint(0);
-
-		// envp
-		const vaddr_t envp_vaddr = vaddr;
-		vaddr += envp.size() * sizeof(uintptr_t);
-		stack_push_uint(0);
-
-		// auxv
-		for (auto aux : auxv)
-			stack_push_buf(BAN::ConstByteSpan::from(aux));
-
-		// information
-		for (size_t i = 0; i < argv.size(); i++)
-		{
-			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), argv_vaddr + i * sizeof(uintptr_t));
-			stack_push_str(argv[i]);
-		}
-		for (size_t i = 0; i < envp.size(); i++)
-		{
-			stack_copy_buf(BAN::ConstByteSpan::from(vaddr), envp_vaddr + i * sizeof(uintptr_t));
-			stack_push_str(envp[i]);
-		}
-
-		setup_exec(entry, m_userspace_stack->vaddr() + m_userspace_stack->size() - needed_size);
-
-		return {};
-	}
-
-	void Thread::setup_exec(vaddr_t ip, vaddr_t sp)
-	{
-		ASSERT(is_userspace());
-		m_state = State::NotStarted;
-
-		// Signal mask is inherited
-
-		// Initialize stack for returning
-		PageTable::with_fast_page(kernel_stack().paddr_of(kernel_stack_top() - PAGE_SIZE), [=] {
-			uintptr_t cur_sp = PageTable::fast_page() + PAGE_SIZE;
-			write_to_stack(cur_sp, 0x20 | 3);
-			write_to_stack(cur_sp, sp);
-			write_to_stack(cur_sp, 0x202);
-#if ARCH(x86_64)
-			write_to_stack(cur_sp, 0x28 | 3);
-#elif ARCH(i686)
-			write_to_stack(cur_sp, 0x18 | 3);
-#endif
-			write_to_stack(cur_sp, ip);
-		});
-
-		m_yield_registers = {};
-		m_yield_registers.ip = reinterpret_cast<vaddr_t>(start_userspace_thread);
-		m_yield_registers.sp = kernel_stack_top() - 5 * sizeof(uintptr_t);
-	}
-
 	void Thread::setup_process_cleanup()
 	{
 		ASSERT(Processor::get_interrupt_state() == InterruptState::Disabled);
 
+		static entry_t entry = [](void* process_ptr) {
+			auto* thread = &Thread::current();
+			auto* process = static_cast<Process*>(process_ptr);
+			ASSERT(thread->m_process == process);
+
+			process->cleanup_function(thread);
+
+			thread->m_delete_process = true;
+
+			// will call on thread exit after return
+		};
+
 		m_state = State::NotStarted;
-		static entry_t entry(
-			[](void* process_ptr)
-			{
-				auto* thread = &Thread::current();
-				auto* process = static_cast<Process*>(process_ptr);
-
-				ASSERT(thread->m_process == process);
-
-				process->cleanup_function(thread);
-
-				thread->m_delete_process = true;
-
-				// will call on thread exit after return
-			}
-		);
 
 		m_signal_pending_mask = 0;
 		m_signal_block_mask = ~0ull;
@@ -609,16 +464,15 @@ namespace Kernel
 			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++)
-			{
-				const uint64_t mask = 1ull << signal;
-				if ((full_pending_mask & mask) && !(m_signal_block_mask & mask))
-					break;
-			}
-			if (signal > _SIGMAX)
+			const uint64_t full_pending_mask = (m_signal_pending_mask | process_signal_pending_mask) & ~m_signal_block_mask;
+			if (full_pending_mask == 0) [[likely]]
 				return false;
 
+			for (signal = _SIGMIN; signal <= _SIGMAX; signal++)
+				if (full_pending_mask & (1ull << signal))
+					break;
+			ASSERT(signal <= _SIGMAX);
+
 			if (process_signal_pending_mask & (1ull << signal))
 				signal_info = m_process->m_signal_infos[signal];
 			else
@@ -864,11 +718,21 @@ namespace Kernel
 		return {};
 	}
 
-	BAN::ErrorOr<void> Thread::sleep_or_eintr_ns(uint64_t ns)
+	BAN::ErrorOr<void> Thread::sleep_or_eintr_for_ns(uint64_t timeout_ns)
 	{
 		if (is_interrupted_by_signal(true))
 			return BAN::Error::from_errno(EINTR);
-		SystemTimer::get().sleep_ns(ns);
+		SystemTimer::get().sleep_for_ns(timeout_ns);
+		if (is_interrupted_by_signal(true))
+			return BAN::Error::from_errno(EINTR);
+		return {};
+	}
+
+	BAN::ErrorOr<void> Thread::sleep_or_eintr_until_ns(uint64_t waketime_ns)
+	{
+		if (is_interrupted_by_signal(true))
+			return BAN::Error::from_errno(EINTR);
+		SystemTimer::get().sleep_until_ns(waketime_ns);
 		if (is_interrupted_by_signal(true))
 			return BAN::Error::from_errno(EINTR);
 		return {};

kernel/kernel/Timer/HPET.cpp

@@ -1,5 +1,6 @@
 #include <BAN/ScopeGuard.h>
 #include <kernel/ACPI/ACPI.h>
+#include <kernel/APIC.h>
 #include <kernel/IDT.h>
 #include <kernel/InterruptController.h>
 #include <kernel/Memory/PageTable.h>
@@ -112,13 +113,13 @@ namespace Kernel
 	static_assert(offsetof(HPETRegisters, timers[0])	== 0x100);
 	static_assert(offsetof(HPETRegisters, timers[1])	== 0x120);
 
-	BAN::ErrorOr<BAN::UniqPtr<HPET>> HPET::create(bool force_pic)
+	BAN::ErrorOr<BAN::UniqPtr<HPET>> HPET::create()
 	{
 		HPET* hpet_ptr = new HPET();
 		if (hpet_ptr == nullptr)
 			return BAN::Error::from_errno(ENOMEM);
 		auto hpet = BAN::UniqPtr<HPET>::adopt(hpet_ptr);
-		TRY(hpet->initialize(force_pic));
+		TRY(hpet->initialize());
 		return hpet;
 	}
 
@@ -129,7 +130,7 @@ namespace Kernel
 		m_mmio_base = 0;
 	}
 
-	BAN::ErrorOr<void> HPET::initialize(bool force_pic)
+	BAN::ErrorOr<void> HPET::initialize()
 	{
 		auto* header = static_cast<const ACPI::HPET*>(ACPI::ACPI::get().get_header("HPET"_sv, 0));
 		if (header == nullptr)
@@ -138,7 +139,7 @@ namespace Kernel
 		if (header->hardware_rev_id == 0)
 			return BAN::Error::from_errno(EINVAL);
 
-		if (force_pic && !header->legacy_replacement_irq_routing_cable)
+		if (!InterruptController::get().is_using_apic() && !header->legacy_replacement_irq_routing_cable)
 		{
 			dwarnln("HPET doesn't support legacy mapping");
 			return BAN::Error::from_errno(ENOTSUP);
@@ -151,18 +152,19 @@ namespace Kernel
 
 		auto& regs = registers();
 
+#if ARCH(x86_64)
 		m_is_64bit = regs.capabilities & COUNT_SIZE_CAP;
+#else
+		// spec: It is strongly recommended that 32-bit software only operate the timer in 32-bit mode.
+		m_is_64bit = false;
+#endif
 
 		// Disable and reset main counter
 		regs.configuration.low = regs.configuration.low & ~ENABLE_CNF;
 		regs.main_counter.high = 0;
 		regs.main_counter.low = 0;
 
-		// Enable legacy routing if available
-		if (regs.capabilities & LEG_RT_CAP)
-			regs.configuration.low = regs.configuration.low | LEG_RT_CNF;
-
-		uint32_t period_fs = regs.counter_clk_period;
+		const uint32_t period_fs = regs.counter_clk_period;
 		if (period_fs == 0 || period_fs > HPET_PERIOD_MAX)
 		{
 			dwarnln("HPET: Invalid counter period");
@@ -172,7 +174,7 @@ namespace Kernel
 		m_ticks_per_s = FS_PER_S / period_fs;
 		dprintln("HPET frequency {} Hz", m_ticks_per_s);
 
-		uint8_t last_timer = (regs.capabilities & NUM_TIM_CAP_MASK) >> NUM_TIM_CAP_SHIFT;
+		const uint8_t last_timer = (regs.capabilities & NUM_TIM_CAP_MASK) >> NUM_TIM_CAP_SHIFT;
 		dprintln("HPET has {} timers", last_timer + 1);
 
 		// Disable all timers
@@ -190,21 +192,54 @@ namespace Kernel
 		}
 
 		// enable interrupts
-		timer0.configuration  = timer0.configuration | Tn_INT_ENB_CNF;
-		// clear interrupt mask (set irq to 0)
-		timer0.configuration = timer0.configuration & ~Tn_INT_ROUTE_CNF_MASK;
+		timer0.configuration = timer0.configuration |  Tn_INT_ENB_CNF;
 		// edge triggered interrupts
 		timer0.configuration = timer0.configuration & ~Tn_INT_TYPE_CNF;
 		// periodic timer
-		timer0.configuration = timer0.configuration | Tn_TYPE_CNF;
+		timer0.configuration = timer0.configuration |  Tn_TYPE_CNF;
 		// disable 32 bit mode
 		timer0.configuration = timer0.configuration & ~Tn_32MODE_CNF;
 		// disable FSB interrupts
 		if (timer0.configuration & Tn_FSB_INT_DEL_CAP)
 			timer0.configuration = timer0.configuration & ~Tn_FSB_EN_CNF;
 
+		uint16_t irq;
+		if (header->legacy_replacement_irq_routing_cable)
+		{
+			TRY(InterruptController::get().reserve_irq(0));
+			irq = 0;
+
+			regs.configuration.low = regs.configuration.low | LEG_RT_CNF;
+			timer0.configuration = timer0.configuration & ~Tn_INT_ROUTE_CNF_MASK;
+		}
+		else
+		{
+			ASSERT(InterruptController::get().is_using_apic());
+			auto& apic = static_cast<APIC&>(InterruptController::get());
+
+			uint8_t gsi = 0;
+			for (; gsi < 32; gsi++)
+			{
+				if (!(timer0.int_route_cap & (1u << gsi)))
+					continue;
+				auto ret = apic.reserve_gsi(gsi);
+				if (ret.is_error())
+					continue;
+				irq = ret.value();
+				break;
+			}
+			if (gsi == 32)
+			{
+				dwarnln("Could not route any interrupt for HPET");
+				return BAN::Error::from_errno(EFAULT);
+			}
+
+			regs.configuration.low = regs.configuration.low & ~LEG_RT_CNF;
+			timer0.configuration = (timer0.configuration & ~Tn_INT_ROUTE_CNF_MASK) | (gsi << Tn_INT_ROUTE_CNF_SHIFT);
+		}
+
 		// set timer period to 1000 Hz
-		uint64_t ticks_per_ms = m_ticks_per_s / 1000;
+		const uint64_t ticks_per_ms = m_ticks_per_s / 1000;
 		timer0.configuration = timer0.configuration | Tn_VAL_SET_CNF;
 		timer0.comparator.low = ticks_per_ms;
 		if (timer0.configuration & Tn_SIZE_CAP)
@@ -221,9 +256,8 @@ namespace Kernel
 		// enable main counter
 		regs.configuration.low = regs.configuration.low | ENABLE_CNF;
 
-		TRY(InterruptController::get().reserve_irq(0));
-		set_irq(0);
-		InterruptController::get().enable_irq(0);
+		set_irq(irq);
+		InterruptController::get().enable_irq(irq);
 
 		return {};
 	}
@@ -240,65 +274,62 @@ namespace Kernel
 
 	uint64_t HPET::read_main_counter() const
 	{
-		auto& regs = registers();
+		const auto& regs = registers();
+
 		if (m_is_64bit)
 			return regs.main_counter.full;
 
 		SpinLockGuard _(m_lock);
-		uint32_t current_low = regs.main_counter.low;
-		uint32_t wraps = m_32bit_wraps;
-		if (current_low < (uint32_t)m_last_ticks)
-			wraps++;
-		return ((uint64_t)wraps << 32) | current_low;
+		const uint32_t current_low = regs.main_counter.low;
+		const uint32_t wraps = m_32bit_wraps + (current_low < static_cast<uint32_t>(m_last_ticks));
+		return (static_cast<uint64_t>(wraps) << 32) | current_low;
 	}
 
 	void HPET::handle_irq()
 	{
 		{
-			auto& regs = registers();
+			const auto& regs = registers();
 
 			SpinLockGuard _(m_lock);
 
-			uint64_t current_ticks;
 			if (m_is_64bit)
-				current_ticks = regs.main_counter.full;
+				m_last_ticks = regs.main_counter.full;
 			else
 			{
-				uint32_t current_low = regs.main_counter.low;
-				if (current_low < (uint32_t)m_last_ticks)
+				const uint32_t current_low = regs.main_counter.low;
+				if (current_low < static_cast<uint32_t>(m_last_ticks))
 					m_32bit_wraps++;
-				current_ticks = ((uint64_t)m_32bit_wraps << 32) | current_low;
+				m_last_ticks = (static_cast<uint64_t>(m_32bit_wraps) << 32) | current_low;
 			}
-			m_last_ticks = current_ticks;
 		}
 
 		SystemTimer::get().update_tsc();
 
 		if (should_invoke_scheduler())
-			Processor::scheduler().timer_interrupt();
+			Processor::scheduler().on_timer_interrupt();
 	}
 
 	uint64_t HPET::ms_since_boot() const
 	{
-		auto current = time_since_boot();
+		const auto current = time_since_boot();
 		return current.tv_sec * 1'000 + current.tv_nsec / 1'000'000;
 	}
 
 	uint64_t HPET::ns_since_boot() const
 	{
-		auto current = time_since_boot();
+		const auto current = time_since_boot();
 		return current.tv_sec * 1'000'000'000 + current.tv_nsec;
 	}
 
 	timespec HPET::time_since_boot() const
 	{
-		auto& regs = registers();
+		const auto& regs = registers();
 
-		uint64_t counter = read_main_counter();
-		uint64_t seconds			= counter / m_ticks_per_s;
-		uint64_t ticks_this_second	= counter % m_ticks_per_s;
+		const uint64_t counter           = read_main_counter();
+		const uint64_t seconds           = counter / m_ticks_per_s;
+		const uint64_t ticks_this_second = counter % m_ticks_per_s;
 
-		long ns_this_second = ticks_this_second * regs.counter_clk_period / FS_PER_NS;
+		const long ns_this_second = ticks_this_second * regs.counter_clk_period / FS_PER_NS;
 
 		return timespec {
 			.tv_sec = static_cast<time_t>(seconds),
@@ -308,7 +339,7 @@ namespace Kernel
 
 	void HPET::pre_scheduler_sleep_ns(uint64_t ns)
 	{
-		auto& regs = registers();
+		const auto& regs = registers();
 
 		const uint64_t target_ticks = BAN::Math::div_round_up<uint64_t>(ns * FS_PER_NS, regs.counter_clk_period);
 

kernel/kernel/Timer/PIT.cpp

@@ -61,7 +61,7 @@ namespace Kernel
 		SystemTimer::get().update_tsc();
 
 		if (should_invoke_scheduler())
-			Processor::scheduler().timer_interrupt();
+			Processor::scheduler().on_timer_interrupt();
 	}
 
 	uint64_t PIT::read_counter_ns() const

kernel/kernel/Timer/Timer.cpp

@@ -11,12 +11,24 @@ namespace Kernel
 
 	static SystemTimer* s_instance = nullptr;
 
-	void SystemTimer::initialize(bool force_pic)
+	struct pvclock_vcpu_time_info
+	{
+		uint32_t version;
+		uint32_t pad0;
+		uint64_t tsc_timestamp;
+		uint64_t system_time;
+		uint32_t tsc_to_system_mul;
+		int8_t tsc_shift;
+		uint8_t flags;
+		uint8_t pad[2];
+	};
+
+	void SystemTimer::initialize()
 	{
 		ASSERT(s_instance == nullptr);
 		auto* temp = new SystemTimer;
 		ASSERT(temp);
-		temp->initialize_timers(force_pic);
+		temp->initialize_timers();
 		s_instance = temp;
 	}
 
@@ -31,12 +43,12 @@ namespace Kernel
 		return !!s_instance;
 	}
 
-	void SystemTimer::initialize_timers(bool force_pic)
+	void SystemTimer::initialize_timers()
 	{
 		m_rtc = MUST(BAN::UniqPtr<RTC>::create());
 		m_boot_time = BAN::to_unix_time(m_rtc->get_current_time());
 
-		if (auto res = HPET::create(force_pic); res.is_error())
+		if (auto res = HPET::create(); res.is_error())
 			dwarnln("HPET: {}", res.error());
 		else
 		{
@@ -59,62 +71,100 @@ namespace Kernel
 
 	void SystemTimer::initialize_tsc()
 	{
-		if (!CPUID::has_invariant_tsc())
-		{
-			dwarnln("CPU does not have an invariant TSC");
-			return;
-		}
+		if (CPUID::has_kvm_pvclock())
+			return initialize_pvclock();
+		if (CPUID::has_invariant_tsc())
+			return initialize_invariant_tsc();
+		dwarnln("No supported TSC based timers available");
+	}
 
-		const uint64_t tsc_freq = get_tsc_frequency();
+	void SystemTimer::initialize_invariant_tsc()
+	{
+		const uint64_t tsc_freq = [this]() -> uint64_t {
+			if (const auto cpuid_freq = CPUID::get_tsc_frequency())
+				return cpuid_freq;
+
+			// take 5x 50 ms samples and use the median value
+
+			constexpr size_t tsc_sample_count = 5;
+			constexpr size_t tsc_sample_ns = 50'000'000;
+
+			uint64_t tsc_freq_samples[tsc_sample_count];
+			for (size_t i = 0; i < tsc_sample_count; i++)
+			{
+				const auto start_ns = m_timer->ns_since_boot();
+
+				const auto start_tsc = __builtin_ia32_rdtsc();
+				while (m_timer->ns_since_boot() < start_ns + tsc_sample_ns)
+					Processor::pause();
+				const auto stop_tsc = __builtin_ia32_rdtsc();
+
+				const auto stop_ns = m_timer->ns_since_boot();
+
+				const auto duration_ns = stop_ns - start_ns;
+				const auto count_tsc = stop_tsc - start_tsc;
+
+				tsc_freq_samples[i] = count_tsc * 1'000'000'000 / duration_ns;
+			}
+
+			BAN::sort::sort(tsc_freq_samples, tsc_freq_samples + tsc_sample_count);
+
+			return tsc_freq_samples[tsc_sample_count / 2];
+		}();
+
+		m_tsc_info = { .invariant = {
+			.shift = 0,
+			.mult  = static_cast<uint32_t>((1'000'000'000ull << 32) / tsc_freq),
+		}};
+		m_tsc_type = TSCType::Invariant;
+		Processor::initialize_tsc(m_boot_time);
 
 		dprintln("Initialized invariant TSC ({} Hz)", tsc_freq);
-
-		const uint8_t tsc_shift = 22;
-		const uint64_t tsc_mult = (static_cast<uint64_t>(1'000'000'000) << tsc_shift) / tsc_freq;
-		Processor::initialize_tsc(tsc_shift, tsc_mult, m_boot_time);
-
-		m_has_invariant_tsc = true;
 	}
 
-	uint64_t SystemTimer::get_tsc_frequency() const
+	static pvclock_vcpu_time_info read_pvclock_safe(vaddr_t pvclock_vaddr)
 	{
-		// take 5x 50 ms samples and use the median value
-
-		constexpr size_t tsc_sample_count = 5;
-		constexpr size_t tsc_sample_ns = 50'000'000;
-
-		uint64_t tsc_freq_samples[tsc_sample_count];
-		for (size_t i = 0; i < tsc_sample_count; i++)
+		for (;;)
 		{
-			const auto start_ns = m_timer->ns_since_boot();
+			const volatile auto& pvclock = *reinterpret_cast<const volatile pvclock_vcpu_time_info*>(pvclock_vaddr);
 
-			const auto start_tsc = __builtin_ia32_rdtsc();
-			while (m_timer->ns_since_boot() < start_ns + tsc_sample_ns)
-				Processor::pause();
-			const auto stop_tsc = __builtin_ia32_rdtsc();
+			const auto version = pvclock.version;
+			if (version & 1)
+				continue;
 
-			const auto stop_ns = m_timer->ns_since_boot();
+			pvclock_vcpu_time_info copy;
+			memcpy(&copy, const_cast<const pvclock_vcpu_time_info*>(&pvclock), sizeof(pvclock_vcpu_time_info));
 
-			const auto duration_ns = stop_ns - start_ns;
-			const auto count_tsc = stop_tsc - start_tsc;
-
-			tsc_freq_samples[i] = count_tsc * 1'000'000'000 / duration_ns;
+			if (pvclock.version == version)
+				return copy;
 		}
-
-		BAN::sort::sort(tsc_freq_samples, tsc_freq_samples + tsc_sample_count);
-
-		return tsc_freq_samples[tsc_sample_count / 2];
 	}
 
-	void SystemTimer::update_tsc() const
+	void SystemTimer::initialize_pvclock()
 	{
-		if (!m_has_invariant_tsc)
+		m_tsc_page = MUST(DMARegion::create(sizeof(pvclock_vcpu_time_info), PageTable::MemoryType::Normal));
+		memset(reinterpret_cast<void*>(m_tsc_page->vaddr()), 0, sizeof(pvclock_vcpu_time_info));
+
+		const uint32_t paddr_hi = m_tsc_page->paddr() >> 32;
+		const uint32_t paddr_lo = m_tsc_page->paddr() & 0xFFFFFFFF;
+		asm volatile("wrmsr" :: "d"(paddr_hi), "a"(paddr_lo | 1), "c"(0x4b564d01));
+
+		m_tsc_type = TSCType::PVClock;
+		Processor::initialize_tsc(m_boot_time);
+
+		dprintln("Initialized pvclock");
+	}
+
+	void SystemTimer::update_tsc()
+	{
+		if (m_tsc_type == TSCType::None)
 			return;
 
-		// only update every 100 ms
-		if (++m_timer_ticks < 100)
+		// only update once per second
+		const uint64_t current_ns = Processor::ns_since_boot_tsc();
+		if (current_ns < m_tsc_update_ns)
 			return;
-		m_timer_ticks = 0;
+		m_tsc_update_ns = current_ns + 1'000'000'000;
 
 		Processor::update_tsc();
 		Processor::broadcast_smp_message({
@@ -123,28 +173,44 @@ namespace Kernel
 		});
 	}
 
-	uint64_t SystemTimer::ns_since_boot_no_tsc() const
+	SystemTimer::TSCInfo SystemTimer::tsc_info() const
 	{
-		return m_timer->ns_since_boot();
+		switch (m_tsc_type)
+		{
+			case TSCType::None:
+				ASSERT_NOT_REACHED();
+			case TSCType::Invariant:
+				return {
+					.shift = m_tsc_info.invariant.shift,
+					.mult  = m_tsc_info.invariant.mult,
+				};
+			case TSCType::PVClock:
+				const auto pvclock = read_pvclock_safe(m_tsc_page->vaddr());
+				return {
+					.shift = pvclock.tsc_shift,
+					.mult  = pvclock.tsc_to_system_mul,
+				};
+		}
+		ASSERT_NOT_REACHED();
 	}
 
 	uint64_t SystemTimer::ms_since_boot() const
 	{
-		if (!m_has_invariant_tsc)
+		if (m_tsc_type == TSCType::None)
 			return m_timer->ms_since_boot();
 		return Processor::ns_since_boot_tsc() / 1'000'000;
 	}
 
 	uint64_t SystemTimer::ns_since_boot() const
 	{
-		if (!m_has_invariant_tsc)
+		if (m_tsc_type == TSCType::None)
 			return m_timer->ns_since_boot();
 		return Processor::ns_since_boot_tsc();
 	}
 
 	timespec SystemTimer::time_since_boot() const
 	{
-		if (!m_has_invariant_tsc)
+		if (m_tsc_type == TSCType::None)
 			return m_timer->time_since_boot();
 		const auto ns_since_boot = Processor::ns_since_boot_tsc();
 		return {
@@ -163,11 +229,18 @@ namespace Kernel
 		return m_timer->pre_scheduler_sleep_ns(ns);
 	}
 
-	void SystemTimer::sleep_ns(uint64_t ns) const
+	void SystemTimer::sleep_for_ns(uint64_t timeout_ns) const
 	{
-		if (ns == 0)
+		const uint64_t base_ns = ns_since_boot();
+		ASSERT(!BAN::Math::will_addition_overflow(base_ns, timeout_ns));
+		Processor::scheduler().block_current_thread(nullptr, base_ns + timeout_ns, nullptr);
+	}
+
+	void SystemTimer::sleep_until_ns(uint64_t waketime_ns) const
+	{
+		if (ns_since_boot() >= waketime_ns)
 			return;
-		Processor::scheduler().block_current_thread(nullptr, ns_since_boot() + ns, nullptr);
+		Processor::scheduler().block_current_thread(nullptr, waketime_ns, nullptr);
 	}
 
 	timespec SystemTimer::real_time() const

kernel/kernel/USB/HID/HIDDriver.cpp

@@ -253,8 +253,9 @@ namespace Kernel
 			BAN::Vector<USBHID::Report> outputs;
 			TRY(gather_collection_reports(collection, outputs, USBHID::Report::Type::Output));
 
-			if (collection.usage_page == 0x01)
+			switch (collection.usage_page)
 			{
+			case 0x01:
 				switch (collection.usage_id)
 				{
 					case 0x02:
@@ -273,6 +274,15 @@ namespace Kernel
 						dwarnln("Unsupported generic descript page usage 0x{2H}", collection.usage_id);
 						break;
 				}
+				break;
+			case 0x0C:
+				switch (collection.usage_id)
+				{
+					case 0x01:
+						report.device = TRY(BAN::RefPtr<USBKeyboard>::create(*this, BAN::move(outputs)));
+						dprintln("Initialized an USB Consumer Control");
+						break;
+				}
 			}
 
 			TRY(result.push_back(BAN::move(report)));

kernel/kernel/USB/HID/Joystick.cpp

@@ -185,7 +185,7 @@ namespace Kernel
 		return to_copy;
 	}
 
-	BAN::ErrorOr<long> USBJoystick::ioctl_impl(int request, void* arg)
+	BAN::ErrorOr<long> USBJoystick::ioctl_impl(unsigned long request, void* arg)
 	{
 		switch (request)
 		{
@@ -217,7 +217,7 @@ namespace Kernel
 				ASSERT_NOT_REACHED();
 		}
 
-		return USBHIDDevice::ioctl(request, arg);
+		return USBHIDDevice::ioctl_impl(request, arg);
 	}
 
 	void USBJoystick::update()

kernel/kernel/USB/HID/Keyboard.cpp

@@ -56,6 +56,8 @@ namespace Kernel
 		READ_MODIFIER(0xE4, KeyModifier::RCtrl);
 		READ_MODIFIER(0xE2, KeyModifier::LAlt);
 		READ_MODIFIER(0xE6, KeyModifier::RAlt);
+		READ_MODIFIER(0xE3, KeyModifier::LSuper);
+		READ_MODIFIER(0xE7, KeyModifier::RSuper);
 #undef READ_MODIFIER
 
 #define READ_TOGGLE(scancode, key_modifier)                                 \
@@ -135,15 +137,33 @@ namespace Kernel
 
 		ASSERT(m_keyboard_lock.current_processor_has_lock());
 
-		if (usage_page != 0x07)
+		if (state == 0)
+			return;
+
+		switch (usage_page)
 		{
-			dprintln_if(DEBUG_USB_KEYBOARD, "Unsupported keyboard usage page {2H}", usage_page);
-			return;
+			case 0x07:
+				if (usage >= 4 && usage < m_keyboard_state_temp.size())
+					m_keyboard_state_temp[usage] = true;
+				break;
+			case 0x0C:
+				switch (usage)
+				{
+					case 0xE9:
+						m_keyboard_state_temp[0x80] = true;
+						break;
+					case 0xEA:
+						m_keyboard_state_temp[0x81] = true;
+						break;
+					default:
+						dprintln_if(DEBUG_USB_KEYBOARD, "Unsupported consumer page usage {2H}", usage);
+						break;
+				}
+				break;
+			default:
+				dprintln_if(DEBUG_USB_KEYBOARD, "Unsupported keyboard usage page {2H}", usage_page);
+				break;
 		}
-		if (!state)
-			return;
-		if (usage >= 4 && usage < m_keyboard_state_temp.size())
-			m_keyboard_state_temp[usage] = state;
 	}
 
 	void USBKeyboard::update()
@@ -322,8 +342,10 @@ namespace Kernel
 		s_scancode_to_keycode[0xE3] = keycode_normal(4,  1);
 		s_scancode_to_keycode[0xE2] = keycode_normal(4,  2);
 		s_scancode_to_keycode[0x2C] = keycode_normal(4,  3);
-		s_scancode_to_keycode[0xE6] = keycode_normal(4,  5);
-		s_scancode_to_keycode[0xE4] = keycode_normal(4,  6);
+		s_scancode_to_keycode[0xE6] = keycode_normal(4,  4);
+		s_scancode_to_keycode[0xE7] = keycode_normal(4,  5);
+		s_scancode_to_keycode[0x65] = keycode_normal(5,  6);
+		s_scancode_to_keycode[0xE4] = keycode_normal(4,  7);
 
 		s_scancode_to_keycode[0x52] = keycode_normal(5,  0);
 		s_scancode_to_keycode[0x50] = keycode_normal(5,  1);

kernel/kernel/USB/MassStorage/SCSIDevice.cpp

@@ -132,9 +132,12 @@ namespace Kernel
 		, m_lun(lun)
 		, m_block_count(block_count)
 		, m_block_size(block_size)
-		, m_name { 's', 'd', (char)('a' + minor(m_rdev)), '\0' }
 	{
 		m_rdev = scsi_get_rdev();
+
+		strcpy(m_name, "sda");
+		m_name[2] += minor(m_rdev);
+
 	}
 
 	USBSCSIDevice::~USBSCSIDevice()

kernel/kernel/kernel.cpp

@@ -151,7 +151,7 @@ extern "C" void kernel_main(uint32_t boot_magic, uint32_t boot_info)
 	InterruptController::initialize(cmdline.force_pic);
 	dprintln("Interrupt controller initialized");
 
-	SystemTimer::initialize(cmdline.force_pic);
+	SystemTimer::initialize();
 	dprintln("Timers initialized");
 
 	DevFileSystem::initialize();

ports/.gitignore

@@ -2,3 +2,5 @@
 !*/patches/
 !*/build.sh
 .installed_ports
+config.sub
+*-banan_os-meson.txt

ports/SDL2_image/build.sh

@@ -3,7 +3,7 @@
 NAME='SDL2_image'
 VERSION='2.8.8'
 DOWNLOAD_URL="https://github.com/libsdl-org/SDL_image/releases/download/release-$VERSION/SDL2_image-$VERSION.tar.gz#2213b56fdaff2220d0e38c8e420cbe1a83c87374190cba8c70af2156097ce30a"
-DEPENDENCIES=('SDL2' 'libpng' 'libjpeg' 'libtiff' 'libwebp')
+DEPENDENCIES=('sdl2-compat' 'libpng' 'libjpeg-turbo' 'libtiff' 'libwebp')
 
 configure() {
 	cmake --fresh -S . -B build -G Ninja \

ports/SDL2_mixer/build.sh

@@ -3,7 +3,7 @@
 NAME='SDL2_mixer'
 VERSION='2.8.1'
 DOWNLOAD_URL="https://github.com/libsdl-org/SDL_mixer/releases/download/release-$VERSION/SDL2_mixer-$VERSION.tar.gz#cb760211b056bfe44f4a1e180cc7cb201137e4d1572f2002cc1be728efd22660"
-DEPENDENCIES=('SDL2' 'timidity')
+DEPENDENCIES=('sdl2-compat' 'timidity')
 
 configure() {
 	cmake --fresh -S . -B build -G Ninja \