banan-os/kernel/kernel/Thread.cpp

#include <BAN/Errors.h>
#include <BAN/ScopeGuard.h>
#include <kernel/GDT.h>
#include <kernel/InterruptController.h>
#include <kernel/InterruptStack.h>
#include <kernel/Memory/kmalloc.h>
#include <kernel/Memory/MemoryBackedRegion.h>
#include <kernel/Process.h>
#include <kernel/Scheduler.h>
#include <kernel/Thread.h>
#include <kernel/Timer/Timer.h>
#include <kernel/UserCopy.h>
#include <sys/syscall.h>

namespace Kernel
{

	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");

	extern "C" [[noreturn]] void start_kernel_thread();
	extern "C" [[noreturn]] void start_userspace_thread();

	extern "C" void signal_trampoline();

	template<typename T>
	static void write_to_stack(uintptr_t& rsp, const T& value) requires(sizeof(T) <= sizeof(uintptr_t))
	{
		rsp -= sizeof(uintptr_t);
		*(uintptr_t*)rsp = (uintptr_t)value;
	}

	static pid_t s_next_tid = 1;

	alignas(16) static uint8_t s_default_sse_storage[512];
	static BAN::Atomic<bool> s_default_sse_storage_initialized = false;

	static void initialize_default_sse_storage()
	{
		static BAN::Atomic<bool> is_initializing { false };
		bool expected { false };
		if (!is_initializing.compare_exchange(expected, true))
		{
			while (!s_default_sse_storage_initialized)
				__builtin_ia32_pause();
			asm volatile("" ::: "memory");
			return;
		}

		const auto state = Processor::get_interrupt_state();
		Processor::set_interrupt_state(InterruptState::Disabled);

		Processor::enable_sse();

		const uint32_t mxcsr = 0x1F80;
		asm volatile(
			"finit;"
			"ldmxcsr %[mxcsr];"
#if ARCH(x86_64)
			"fxsave64 %[storage];"
#elif ARCH(i686)
			"fxsave %[storage];"
#else
#error
#endif
			: [storage]"=m"(s_default_sse_storage)
			: [mxcsr]"m"(mxcsr)
		);

		Processor::disable_sse();

		Processor::set_interrupt_state(state);

		s_default_sse_storage_initialized = true;
	}

	bool Thread::is_stopping_signal(int signal)
	{
		switch(signal)
		{
			case SIGSTOP:
			case SIGTSTP:
			case SIGTTIN:
			case SIGTTOU:
				return true;
			default:
				return false;
		}
	}

	bool Thread::is_continuing_signal(int signal)
	{
		switch(signal)
		{
			case SIGCONT:
				return true;
			default:
				return false;
		}
	}

	bool Thread::is_terminating_signal(int signal)
	{
		switch (signal)
		{
			case SIGALRM:
			case SIGHUP:
			case SIGINT:
			case SIGKILL:
			case SIGPIPE:
			case SIGTERM:
			case SIGUSR1:
			case SIGUSR2:
			case SIGPOLL:
			case SIGPROF:
			case SIGVTALRM:
				return true;
			default:
				return false;
		}
	}

	bool Thread::is_abnormal_terminating_signal(int signal)
	{
		switch (signal)
		{
			case SIGABRT:
			case SIGBUS:
			case SIGFPE:
			case SIGILL:
			case SIGQUIT:
			case SIGSEGV:
			case SIGSYS:
			case SIGTRAP:
			case SIGXCPU:
			case SIGXFSZ:
				return true;
			default:
				return false;
		}
	}

	bool Thread::is_realtime_signal(int signal)
	{
		return SIGRTMIN <= signal && signal <= SIGRTMAX;
	}

	static bool is_default_ignored_signal(int signal)
	{
		switch (signal)
		{
			case SIGCHLD:
			case SIGURG:
			case SIGWINCH:
			case SIGCANCEL:
				return true;
			default:
				return Thread::is_realtime_signal(signal);
		}
	}

	BAN::ErrorOr<Thread*> Thread::create_kernel(entry_t entry, void* data)
	{
		// Create the thread object
		Thread* thread = new Thread(s_next_tid++, nullptr);
		if (thread == nullptr)
			return BAN::Error::from_errno(ENOMEM);
		BAN::ScopeGuard thread_deleter([thread] { delete thread; });

		// Initialize stack and registers
		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
			PageTable::kernel(),
			{ KERNEL_OFFSET, UINTPTR_MAX },
			kernel_stack_size,
			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
			true
		));

		// Initialize stack for returning
		uintptr_t sp = thread->kernel_stack_top();
		write_to_stack(sp, thread);
		write_to_stack(sp, &Thread::on_exit_trampoline);
		write_to_stack(sp, data);
		write_to_stack(sp, entry);

		thread->m_yield_registers = {};
		thread->m_yield_registers.ip = reinterpret_cast<vaddr_t>(start_kernel_thread);
		thread->m_yield_registers.sp = sp;

		thread_deleter.disable();

		return thread;
	}

	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);

		// Create the thread object
		Thread* thread = new Thread(s_next_tid++, process);
		if (thread == nullptr)
			return BAN::Error::from_errno(ENOMEM);
		BAN::ScopeGuard thread_deleter([thread] { delete thread; });

		thread->m_is_userspace = true;

		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
			page_table,
			{ userspace_stack_base, USERSPACE_END },
			kernel_stack_size,
			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
			true
		));

		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();

		return thread;
	}

	Thread::Thread(pid_t tid, Process* process)
		: m_tid(tid), m_process(process)
	{
		if (!s_default_sse_storage_initialized)
			initialize_default_sse_storage();
		memcpy(m_sse_storage, s_default_sse_storage, sizeof(m_sse_storage));
	}

	Thread& Thread::current()
	{
		return Processor::scheduler().current_thread();
	}

	pid_t Thread::current_tid()
	{
		if (Processor::count() == 0)
			return 0;
		return Processor::scheduler().current_tid();
	}

	Process& Thread::process()
	{
		ASSERT(m_process);
		return *m_process;
	}

	const Process& Thread::process() const
	{
		ASSERT(m_process);
		return *m_process;
	}

	Thread::~Thread()
	{
		if (Processor::get_current_sse_thread() == this)
		{
			Processor::set_current_sse_thread(nullptr);
			Processor::disable_sse();
		}

		if (m_delete_process)
		{
			ASSERT(m_process);
			delete m_process;
		}
	}

	uint64_t Thread::cpu_time_ns() const
	{
		SpinLockGuard _(m_cpu_time_lock);
		if (m_cpu_time_start_ns == UINT64_MAX)
			return m_cpu_time_ns;
		return m_cpu_time_ns + (SystemTimer::get().ns_since_boot() - m_cpu_time_start_ns);
	}

	void Thread::set_cpu_time_start()
	{
		SpinLockGuard _(m_cpu_time_lock);
		ASSERT(m_cpu_time_start_ns == UINT64_MAX);
		m_cpu_time_start_ns = SystemTimer::get().ns_since_boot();
	}

	void Thread::set_cpu_time_stop()
	{
		SpinLockGuard _(m_cpu_time_lock);
		ASSERT(m_cpu_time_start_ns != UINT64_MAX);
		m_cpu_time_ns += SystemTimer::get().ns_since_boot() - m_cpu_time_start_ns;
		m_cpu_time_start_ns = UINT64_MAX;
	}

	void Thread::update_processor_index_address()
	{
		if (!is_userspace() || !has_process())
			return;
		Processor::gdt().set_cpu_index(Processor::current_index());
	}

	BAN::ErrorOr<Thread*> Thread::clone(Process* new_process, uintptr_t sp, uintptr_t ip)
	{
		ASSERT(m_is_userspace);
		ASSERT(m_state == State::Executing);

		Thread* thread = new Thread(s_next_tid++, new_process);
		if (thread == nullptr)
			return BAN::Error::from_errno(ENOMEM);
		BAN::ScopeGuard thread_deleter([thread] { delete thread; });

		thread->m_is_userspace = true;

		thread->m_kernel_stack = TRY(VirtualRange::create_to_vaddr_range(
			new_process->page_table(),
			{ userspace_stack_base, USERSPACE_END },
			kernel_stack_size,
			PageTable::Flags::ReadWrite | PageTable::Flags::Present,
			true
		));

		// NOTE: copy [sp, stack_end] so fork return works
		PageTable::with_fast_page(thread->m_kernel_stack->paddr_of(thread->kernel_stack_top() - PAGE_SIZE), [&] {
			const size_t ncopy = kernel_stack_top() - sp;
			ASSERT(ncopy <= PAGE_SIZE);
			memcpy(
				PageTable::fast_page_as_ptr(PAGE_SIZE - ncopy),
				reinterpret_cast<void*>(sp),
				ncopy
			);
		});

		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;

		thread->m_state = State::NotStarted;

		if (Processor::get_current_sse_thread() == this)
			save_sse();
		memcpy(thread->m_sse_storage, m_sse_storage, sizeof(m_sse_storage));

		thread->m_yield_registers = {};
		thread->m_yield_registers.ip = ip;
		thread->m_yield_registers.sp = sp;
		thread->m_yield_registers.ret = 0;

		thread_deleter.disable();

		return thread;
	}

	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;

		m_signal_pending_mask = 0;
		m_signal_block_mask = ~0ull;

		PageTable::with_fast_page(kernel_stack().paddr_of(kernel_stack_top() - PAGE_SIZE), [&] {
			uintptr_t sp = PageTable::fast_page() + PAGE_SIZE;
			write_to_stack(sp, this);
			write_to_stack(sp, &Thread::on_exit_trampoline);
			write_to_stack(sp, m_process);
			write_to_stack(sp, entry);
		});

		m_yield_registers = {};
		m_yield_registers.ip = reinterpret_cast<vaddr_t>(start_kernel_thread);
		m_yield_registers.sp = kernel_stack_top() - 4 * sizeof(uintptr_t);
	}

	bool Thread::is_interrupted_by_signal(bool skip_stop_and_cont) const
	{
		if (!is_userspace() || m_state != State::Executing)
			return false;
		auto& interrupt_stack = *reinterpret_cast<InterruptStack*>(kernel_stack_top() - sizeof(InterruptStack));
		if (!GDT::is_user_segment(interrupt_stack.cs))
			return false;

		uint64_t full_pending_mask = m_signal_pending_mask | process().signal_pending_mask();
		uint64_t signals = full_pending_mask & ~m_signal_block_mask;
		for (uint8_t i = 0; i < _SIGMAX; i++)
		{
			if (!(signals & ((uint64_t)1 << i)))
				continue;
			if (skip_stop_and_cont && (is_stopping_signal(i) || is_continuing_signal(i)))
				continue;

			vaddr_t signal_handler;
			{
				SpinLockGuard _(m_process->m_signal_lock);
				const auto& handler = m_process->m_signal_handlers[i];
				signal_handler = (handler.sa_flags & SA_SIGINFO)
					? reinterpret_cast<vaddr_t>(handler.sa_sigaction)
					: reinterpret_cast<vaddr_t>(handler.sa_handler);
			}
			if (signal_handler == reinterpret_cast<vaddr_t>(SIG_IGN))
				continue;
			if (signal_handler == reinterpret_cast<vaddr_t>(SIG_DFL) && is_default_ignored_signal(i))
				continue;
			return true;
		}
		return false;
	}

	bool Thread::will_exit_because_of_signal() const
	{
		const uint64_t full_pending_mask = m_signal_pending_mask | process().signal_pending_mask();
		const uint64_t signals = full_pending_mask & ~m_signal_block_mask;
		for (size_t sig = _SIGMIN; sig <= _SIGMAX; sig++)
		{
			if (!(signals & (static_cast<uint64_t>(1) << sig)))
				continue;
			if (!is_terminating_signal(sig) && !is_abnormal_terminating_signal(sig))
				continue;
			SpinLockGuard _(m_process->m_signal_lock);
			if (m_process->m_signal_handlers[sig].sa_handler == SIG_DFL)
				return true;
		}
		return false;
	}

	bool Thread::handle_signal_if_interrupted()
	{
		int signal;
		siginfo_t signal_info;
		signal_handle_info_t handle_info;

		{
			SpinLockGuard _1(m_signal_lock);
			SpinLockGuard _2(m_process->m_signal_lock);

			const uint64_t process_signal_pending_mask = m_process->m_signal_pending_mask;
			const uint64_t full_pending_mask = (m_signal_pending_mask | process_signal_pending_mask) & ~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
				signal_info = m_signal_infos[signal];
			signal_info.si_signo = signal;

			handle_info = remove_signal_and_get_info(signal);
		}

		handle_signal_impl(signal, signal_info, handle_info);

		return handle_info.has_sa_restart;
	}

	bool Thread::handle_signal(int signal, const siginfo_t& signal_info)
	{
		ASSERT(&Thread::current() == this);
		ASSERT(is_userspace());

		ASSERT(signal >= _SIGMIN);
		ASSERT(signal <= _SIGMAX);

		signal_handle_info_t handle_info;

		// If this signal is blocked or ignored, terminate the process
		bool terminate_process = false;

		{
			SpinLockGuard _1(m_signal_lock);
			SpinLockGuard _2(m_process->m_signal_lock);

			if (m_signal_block_mask & (1ull << signal))
				terminate_process = true;

			handle_info = remove_signal_and_get_info(signal);

			if (handle_info.handler == reinterpret_cast<vaddr_t>(SIG_IGN))
				terminate_process = true;
		}

		if (terminate_process)
		{
			process().exit(128 + signal, signal | 0x80);
			ASSERT_NOT_REACHED();
		}

		handle_signal_impl(signal, signal_info, handle_info);

		return handle_info.has_sa_restart;
	}

	Thread::signal_handle_info_t Thread::remove_signal_and_get_info(int signal)
	{
		ASSERT(m_signal_lock.current_processor_has_lock());
		ASSERT(m_process->m_signal_lock.current_processor_has_lock());

		ASSERT(signal >= _SIGMIN);
		ASSERT(signal <= _SIGMAX);

		const uint64_t restore_sigmask = m_signal_block_mask;

		auto& handler = m_process->m_signal_handlers[signal];

		const vaddr_t signal_handler = (handler.sa_flags & SA_SIGINFO)
			? reinterpret_cast<vaddr_t>(handler.sa_sigaction)
			: reinterpret_cast<vaddr_t>(handler.sa_handler);
		const bool has_sa_restart = !!(handler.sa_flags & SA_RESTART);

		vaddr_t signal_stack_top = 0;
		if (m_signal_alt_stack.ss_flags != SS_DISABLE && (handler.sa_flags & SA_ONSTACK) && !currently_on_alternate_stack())
			signal_stack_top = reinterpret_cast<vaddr_t>(m_signal_alt_stack.ss_sp) + m_signal_alt_stack.ss_size;

		m_signal_block_mask |= handler.sa_mask;
		if (!(handler.sa_flags & SA_NODEFER))
			m_signal_block_mask |= 1ull << signal;

		if (handler.sa_flags & SA_RESETHAND)
			handler = { .sa_handler = SIG_DFL, .sa_mask = 0, .sa_flags = 0 };

		m_signal_pending_mask &= ~(1ull << signal);
		m_process->m_signal_pending_mask &= ~(1ull << signal);

		if (m_signal_suspend_mask.has_value())
		{
			m_signal_block_mask = m_signal_suspend_mask.value();
			m_signal_suspend_mask.clear();
		}

		return {
			.handler = signal_handler,
			.stack_top = signal_stack_top,
			.restore_sigmask = restore_sigmask,
			.has_sa_restart = has_sa_restart,
		};
	}

	void Thread::handle_signal_impl(int signal, const siginfo_t& signal_info, const signal_handle_info_t& handle_info)
	{
		ASSERT(this == &Thread::current());
		ASSERT(is_userspace());
		ASSERT(Processor::get_interrupt_state() == InterruptState::Enabled);

		auto& interrupt_stack = *reinterpret_cast<InterruptStack*>(kernel_stack_top() - sizeof(InterruptStack));
		ASSERT(GDT::is_user_segment(interrupt_stack.cs));

		if (handle_info.handler == reinterpret_cast<vaddr_t>(SIG_IGN))
			;
		else if (handle_info.handler != reinterpret_cast<vaddr_t>(SIG_DFL))
		{
			// call userspace signal handlers
#if ARCH(x86_64)
			interrupt_stack.sp -= 128; // skip possible red-zone
#endif

			const auto write_to_stack =
				[&]<typename T>(uintptr_t& sp, const T& value)
				{
					static_assert(sizeof(T) >= sizeof(uintptr_t));
					sp -= sizeof(T);
					if (write_to_user(reinterpret_cast<void*>(sp), &value, sizeof(T)).is_error())
						m_process->exit(128 + SIGSEGV, SIGSEGV | 0x80);
				};

			write_to_stack(interrupt_stack.sp, interrupt_stack.ip);
			const vaddr_t old_stack = interrupt_stack.sp;
			if (handle_info.stack_top)
				interrupt_stack.sp = handle_info.stack_top;
			write_to_stack(interrupt_stack.sp, old_stack);
			write_to_stack(interrupt_stack.sp, interrupt_stack.flags);
			write_to_stack(interrupt_stack.sp, handle_info.restore_sigmask);

			ASSERT(signal_info.si_signo == signal);
			write_to_stack(interrupt_stack.sp, signal_info);

			write_to_stack(interrupt_stack.sp, static_cast<uintptr_t>(signal));
			write_to_stack(interrupt_stack.sp, handle_info.handler);
			interrupt_stack.ip = (uintptr_t)signal_trampoline;
		}
		else
		{
			if (is_abnormal_terminating_signal(signal))
			{
				process().exit(128 + signal, signal | 0x80);
				ASSERT_NOT_REACHED();
			}
			else if (is_terminating_signal(signal))
			{
				process().exit(128 + signal, signal);
				ASSERT_NOT_REACHED();
			}
			else if (is_stopping_signal(signal))
			{
				process().set_stopped(true, signal);
			}
			else if (is_continuing_signal(signal))
			{
				process().set_stopped(false, signal);
			}
			else if (is_default_ignored_signal(signal))
			{

			}
			else
			{
				panic("Executing unhandled signal {}", signal);
			}
		}
	}

	void Thread::add_signal(int signal, const siginfo_t& info)
	{
		SpinLockGuard _(m_signal_lock);

		if (m_process)
		{
			vaddr_t signal_handler;
			{
				SpinLockGuard _(m_process->m_signal_lock);
				const auto& handler = m_process->m_signal_handlers[signal];
				signal_handler = (handler.sa_flags & SA_SIGINFO)
					? reinterpret_cast<vaddr_t>(handler.sa_sigaction)
					: reinterpret_cast<vaddr_t>(handler.sa_handler);
			}
			if (signal_handler == reinterpret_cast<vaddr_t>(SIG_IGN))
				return;
			if (signal_handler == reinterpret_cast<vaddr_t>(SIG_DFL) && is_default_ignored_signal(signal))
				return;
		}

		const uint64_t mask = 1ull << signal;
		m_signal_pending_mask |= mask;
		m_signal_infos[signal] = info;

		if (this != &Thread::current())
			Processor::scheduler().unblock_thread(this);
	}

	void Thread::set_suspend_signal_mask(uint64_t sigmask)
	{
		SpinLockGuard _(m_signal_lock);
		ASSERT(!m_signal_suspend_mask.has_value());
		m_signal_suspend_mask = m_signal_block_mask;
		m_signal_block_mask = sigmask;
	}

	bool Thread::currently_on_alternate_stack() const
	{
		ASSERT(m_signal_lock.current_processor_has_lock());

		if (m_signal_alt_stack.ss_flags == SS_ONSTACK)
			return false;

		const vaddr_t stack_bottom = reinterpret_cast<vaddr_t>(m_signal_alt_stack.ss_sp);
		const vaddr_t stack_top = stack_bottom + m_signal_alt_stack.ss_size;
		const vaddr_t sp = m_yield_registers.sp;
		return stack_bottom <= sp && sp <= stack_top;
	}

	BAN::ErrorOr<void> Thread::sigaltstack(const stack_t* ss, stack_t* oss)
	{
		SpinLockGuard _(m_signal_lock);

		const bool on_alt_stack = currently_on_alternate_stack();

		if (oss)
		{
			*oss = m_signal_alt_stack;
			if (on_alt_stack)
				oss->ss_flags = SS_ONSTACK;
		}

		if (ss)
		{
			if (on_alt_stack)
				return BAN::Error::from_errno(EPERM);
			if (ss->ss_flags && ss->ss_flags != SS_DISABLE)
				return BAN::Error::from_errno(EINVAL);
			if (ss->ss_size < MINSIGSTKSZ)
				return BAN::Error::from_errno(ENOMEM);
			m_signal_alt_stack = *ss;
		}

		return {};
	}

	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_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 {};
	}

	BAN::ErrorOr<void> Thread::block_or_eintr_indefinite(ThreadBlocker& thread_blocker, BaseMutex* mutex)
	{
		if (is_interrupted_by_signal(true))
			return BAN::Error::from_errno(EINTR);
		thread_blocker.block_indefinite(mutex);
		if (is_interrupted_by_signal(true))
			return BAN::Error::from_errno(EINTR);
		return {};
	}

	BAN::ErrorOr<void> Thread::block_or_eintr_or_timeout_ns(ThreadBlocker& thread_blocker, uint64_t timeout_ns, bool etimedout, BaseMutex* mutex)
	{
		const uint64_t wake_time_ns = SystemTimer::get().ns_since_boot() + timeout_ns;
		return block_or_eintr_or_waketime_ns(thread_blocker, wake_time_ns, etimedout, mutex);
	}

	BAN::ErrorOr<void> Thread::block_or_eintr_or_waketime_ns(ThreadBlocker& thread_blocker, uint64_t wake_time_ns, bool etimedout, BaseMutex* mutex)
	{
		if (is_interrupted_by_signal(true))
			return BAN::Error::from_errno(EINTR);
		thread_blocker.block_with_wake_time_ns(wake_time_ns, mutex);
		if (is_interrupted_by_signal(true))
			return BAN::Error::from_errno(EINTR);
		if (etimedout && SystemTimer::get().ms_since_boot() >= wake_time_ns)
			return BAN::Error::from_errno(ETIMEDOUT);
		return {};
	}

	void Thread::on_exit_trampoline(Thread* thread)
	{
		thread->on_exit();
	}

	void Thread::on_exit()
	{
		ASSERT(this == &Thread::current());
		if (!m_delete_process && has_process())
		{
			if (process().on_thread_exit(*this))
			{
				Processor::set_interrupt_state(InterruptState::Disabled);
				setup_process_cleanup();
				Processor::yield();
				ASSERT_NOT_REACHED();
			}
		}

		Processor::set_interrupt_state(InterruptState::Disabled);
		m_state = State::Terminated;
		Processor::yield();
		ASSERT_NOT_REACHED();
	}

	void Thread::save_sse()
	{
#if ARCH(x86_64)
		__builtin_ia32_fxsave64(m_sse_storage);
#elif ARCH(i686)
		// no idea why the builtin don't work
		asm volatile("fxsave %0" :: "m"(m_sse_storage));
#else
#error
#endif
	}

	void Thread::load_sse()
	{
#if ARCH(x86_64)
		__builtin_ia32_fxrstor64(m_sse_storage);
#elif ARCH(i686)
		// no idea why the builtin don't work
		asm volatile("fxrstor %0" :: "m"(m_sse_storage));
#else
#error
#endif
	}

}