Implement HPET driver

[marysaka]

This PR adds support for the "High Precision Event Timer" device.

This permits us to get better timing resolution (PIC resolution was 10ms, we now have a resolution of 100ns)

As we currently don't have IO-APIC support, the HPET is setup in "legacy mapping" and use the IRQ0 and IRQ8 for timer 0 and timer 1.
This result in the RTC device not being able to send interrupt.
For this matter, the timer 1 is setup to send an interrupt every seconds to fake this until we get IO-APIC support.

[marysaka]

Ready for reviews

[todo]

Test the hardware behaviour on partial write of the HPET accumulator

Because we are running on i386, this cause issue on QEMU. In fact, QEMU clear the accumulator flag on every partial write. The question here is: Is that normal or a bug in QEMU?

---

SunriseOS/kernel/src/devices/hpet.rs

Lines 232 to 242 in 8e5ee83

	// TODO: Test the hardware behaviour on partial write of the HPET accumulator
	// BODY: Because we are running on i386, this cause issue on QEMU.
	// BODY: In fact, QEMU clear the accumulator flag on every partial write.
	// BODY: The question here is: Is that normal or a bug in QEMU?
	let mut config = unsafe { (*self.inner).config.read() };
	config.set_accumulator_config(true);
	unsafe { (*self.inner).config.write(config) };
	unsafe {
	(*self.inner)
	.comparator_value_low
	.write((value & 0xFFFF_FFFF) as u32)

---

This comment was generated by todo based on a TODO comment in 8e5ee83 in #256. cc @Thog.

[todo]

enable main timer

SunriseOS/kernel/src/devices/hpet.rs

Lines 431 to 441 in 8e5ee83

	// TODO: enable main timer
	let main_timer_opt = hpet_instance.get_timer(0);
	if main_timer_opt.is_none() {
	paging::kernel_memory::get_kernel_memory().unmap(virtual_address, PAGE_SIZE);
	return false;
	}
	// Set the tick rate in femtoseconds
	// Kernel needs an update frequency of 100 nanoseconds.
	let tick_rate = 100;

---

This comment was generated by todo based on a TODO comment in 8e5ee83 in #256. cc @Thog.

[todo]

remove this when IO-APIC will be implemented

SunriseOS/kernel/src/devices/hpet.rs

Lines 464 to 474 in 8e5ee83

	// TODO: remove this when IO-APIC will be implemented
	// Cannot fail, by spec there is at least 3 comparators defined.
	let rtc_timer = hpet_instance.get_timer(1).unwrap();
	if !rtc_timer.support_periodic_interrupt() {
	warn!("RTC irq cannot be faked, clock will be broken!");
	} else {
	let rtc_frequency = hpet_instance.get_frequency();
	rtc_timer.set_edge_trigger();
	rtc_timer.set_periodic_mode();
	rtc_timer.enable_interrupt();
	rtc_timer.set_accumulator_value(rtc_frequency);

---

This comment was generated by todo based on a TODO comment in 8e5ee83 in #256. cc @Thog.

[todo]

switch HPET to normal mode when IO-APIC will be implemented.

SunriseOS/kernel/src/devices/hpet.rs

Lines 478 to 488 in 8e5ee83

	// TODO: switch HPET to normal mode when IO-APIC will be implemented.
	hpet_instance.enable_legacy_mapping();
	// Clear the interrupt state
	hpet_instance.enable();
	timer::set_kernel_timer_info(0, hpet_instance.get_frequency(), tick_rate);
	HPET_INSTANCE = Some(hpet_instance);
	true
	}

---

This comment was generated by todo based on a TODO comment in 8e5ee83 in #256. cc @Thog.

[marysaka]

Rebased

[Orycterope]

Math is hard

[Orycterope]

I don't recall what was the precise reason of this PR ? Did we need a higher precision timer somewhere ? Is this a necessary step to some goal ?

[marysaka]

@Orycterope This is a preparation to get the LAPIC and IO-APIC implemented (and then multicore). The final goal of this is to have multiple timer on multiple core.

[todo]

Switch to a lower update frequency in HPET

We will maybe prefer to have a better resolution for kernel time. For that to be possible, we need to take care of the sleep_thread logic in userland first (sleep_thread(0) shouldn't be used).

---

SunriseOS/kernel/src/devices/hpet.rs

Lines 449 to 459 in dce8599

	// TODO: Switch to a lower update frequency in HPET
	// BODY: We will maybe prefer to have a better resolution for kernel time.
	// BODY: For that to be possible, we need to take care of the sleep_thread logic in userland first (sleep_thread(0) shouldn't be used).
	let irq_period_ns = 1 * 1_000_000;
	let irq_period_fs = irq_period_ns * 1_000_000;
	info!("HPET frequency: {} Hz", hpet_instance.get_frequency());
	info!("HPET IRQ period: {} fs", irq_period_fs);
	let kernel_frequency = irq_period_fs / u64::from(hpet_instance.get_period());

---

This comment was generated by todo based on a TODO comment in dce8599 in #256. cc @Thog.