Kernel: Allow parallel LAPIC timer initialization with HPET

HPET supports reading LAPIC counter without locks, so it can be done in
parallel. This makes booting much faster. Previously initializing every
timer took 100 ms, so 16 CPUs took total of 1.6 seconds. This allows
doing it all in 100 ms.

kernel/include/kernel/Timer/HPET.h

@@ -19,6 +19,7 @@ namespace Kernel
 		virtual uint64_t ns_since_boot() const override;
 		virtual timespec time_since_boot() const override;
 
+		virtual bool pre_scheduler_sleep_needs_lock() const override { return false; }
 		virtual void pre_scheduler_sleep_ns(uint64_t) override;
 
 		virtual void handle_irq() override;

kernel/include/kernel/Timer/PIT.h

@@ -16,6 +16,7 @@ namespace Kernel
 		virtual uint64_t ns_since_boot() const override;
 		virtual timespec time_since_boot() const override;
 
+		virtual bool pre_scheduler_sleep_needs_lock() const override { return true; }
 		virtual void pre_scheduler_sleep_ns(uint64_t) override;
 
 		virtual void handle_irq() override;

kernel/include/kernel/Timer/Timer.h

@@ -17,6 +17,7 @@ namespace Kernel
 		virtual uint64_t ns_since_boot() const = 0;
 		virtual timespec time_since_boot() const = 0;
 
+		virtual bool pre_scheduler_sleep_needs_lock() const = 0;
 		virtual void pre_scheduler_sleep_ns(uint64_t) = 0;
 
 		void dont_invoke_scheduler() { m_should_invoke_scheduler = false; }
@@ -39,6 +40,7 @@ namespace Kernel
 		virtual uint64_t ns_since_boot() const override;
 		virtual timespec time_since_boot() const override;
 
+		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 { return sleep_ns(ms * 1'000'000); }

kernel/kernel/APIC.cpp

@@ -253,11 +253,11 @@ namespace Kernel
 		const auto send_ipi =
 			[&](uint8_t processor, uint32_t data, uint64_t ud)
 			{
+				while ((read_from_local_apic(LAPIC_ICR_LO_REG) & ICR_LO_delivery_status_send_pending) == ICR_LO_delivery_status_send_pending)
+					__builtin_ia32_pause();
 				write_to_local_apic(LAPIC_ICR_HI_REG, (read_from_local_apic(LAPIC_ICR_HI_REG) & 0x00FFFFFF) | (processor << 24));
 				write_to_local_apic(LAPIC_ICR_LO_REG, data);
 				udelay(ud);
-				while ((read_from_local_apic(LAPIC_ICR_LO_REG) & ICR_LO_delivery_status_send_pending) == ICR_LO_delivery_status_send_pending)
-					__builtin_ia32_pause();
 			};
 
 		dprintln("System has {} processors", m_processors.size());
@@ -402,21 +402,26 @@ namespace Kernel
 
 	void APIC::initialize_timer()
 	{
-		{
+		ASSERT(Kernel::Processor::get_interrupt_state() == InterruptState::Disabled);
-			constexpr uint64_t measuring_duration_ms = 100;
 
-			SpinLockGuard _(s_timer_init_lock);
+		constexpr uint64_t measuring_duration_ms = 100;
 
-			write_to_local_apic(LAPIC_TIMER_LVT,         TimerModeOneShot | TimerMask);
+		const bool needs_lock = SystemTimer::get().pre_scheduler_sleep_needs_lock();
-			write_to_local_apic(LAPIC_TIMER_DIVIDE_REG,  DivideBy2);
+		if (needs_lock)
-			write_to_local_apic(LAPIC_TIMER_INITIAL_REG, 0xFFFFFFFF);
+			s_timer_init_lock.lock();
-			SystemTimer::get().pre_scheduler_sleep_ns(measuring_duration_ms * 1'000'000);
 
-			const uint32_t counter = read_from_local_apic(LAPIC_TIMER_CURRENT_REG);
+		write_to_local_apic(LAPIC_TIMER_LVT,         TimerModeOneShot | TimerMask);
-			m_lapic_timer_frequency_hz = static_cast<uint64_t>(0xFFFFFFFF - counter) * 2 * (1000 / measuring_duration_ms);
+		write_to_local_apic(LAPIC_TIMER_DIVIDE_REG,  DivideBy2);
+		write_to_local_apic(LAPIC_TIMER_INITIAL_REG, 0xFFFFFFFF);
+		SystemTimer::get().pre_scheduler_sleep_ns(measuring_duration_ms * 1'000'000);
 
-			dprintln("CPU {}: lapic timer frequency: {} Hz", Kernel::Processor::current_id(), m_lapic_timer_frequency_hz);
+		const uint32_t counter = read_from_local_apic(LAPIC_TIMER_CURRENT_REG);
-		}
+		m_lapic_timer_frequency_hz = static_cast<uint64_t>(0xFFFFFFFF - counter) * 2 * (1000 / measuring_duration_ms);
+
+		if (needs_lock)
+			s_timer_init_lock.unlock(InterruptState::Disabled);
+
+		dprintln("CPU {}: lapic timer frequency: {} Hz", Kernel::Processor::current_id(), m_lapic_timer_frequency_hz);
 
 		write_to_local_apic(LAPIC_TIMER_LVT,         TimerModePeriodic | (IRQ_VECTOR_BASE + IRQ_TIMER));
 		write_to_local_apic(LAPIC_TIMER_INITIAL_REG, m_lapic_timer_frequency_hz / 2 / 100);

kernel/kernel/Timer/Timer.cpp

@@ -69,6 +69,11 @@ namespace Kernel
 		return m_timer->time_since_boot();
 	}
 
+	bool SystemTimer::pre_scheduler_sleep_needs_lock() const
+	{
+		return m_timer->pre_scheduler_sleep_needs_lock();
+	}
+
 	void SystemTimer::pre_scheduler_sleep_ns(uint64_t ns)
 	{
 		return m_timer->pre_scheduler_sleep_ns(ns);