Update hw for benchmarking and api changes

The hw platform is maintained in order to run benchmark comparisons with
the seL4 platform. As we develop the seL4 platform we change some of
the library apis. This commit adds:
- Extra interrupt numbers to the PIC
- A serial input interrupt handler for saving, resetting and printing
the tsc for measuring idle time
- Changes default console output to use COM port instead of VGA
- Changes for changed library functions
- A rumprun-bake configuration for only using ethernet modules

x86: keep state for both PICs

Improve detection of pvclock

There are two MSR variants, old-style and new-style. We want to prefer
the new-style one, and neither if the relevant flags aren't set in
CPUID.

Improve separation of pvclock/tscclock

Allows us to init pvclock and fall back to tscclock if pvclock is
unavailable. Also allows for easier implementation of e.g. Xen HVM use
of pvclock later.

pvclock_read_wall_clock reads boot time, not current time

Fix mul64_32() for i386 clobbering CF

Implement KVM pvclock support (#36)

Remove broken abstraction in x86_cpuid()

CPUID is a mess, remove the broken abstraction in x86_cpuid() which
tries to detect the maximum supported level, as it gets in the way of
using other CPUID leaves (hypervisor detection).

Reduce export of clock routines where unnecessary.

Make delta_ticks unsigned, it's expected to be unsigned

Fish for interrupts if returning from block() before timeout.

Shortcircuit block is sleeptime is <=0, not only when <0.

Enable interrupts "atomically" in x86 bmk_platform_cpu_halt().

Otherwise, we might even miss the clock interrupt, let alone other
interrupts which should prevent us from sleeping.

bmk_platform_block() -> bmk_platform_cpu_block()

It's arch-dependent anyway, no point in wrapping it.

Make the clock routines bmk_platform_cpu_clock_foo

bmk can now directly call into the relevant code instead of a bounce.

actually commit the code changes too (for previous commit)

remove bmk_ prefix from routines private to platform/hw

remove unnecessary critical section

clarify RTC_UIP comment

if bmk_isr_clock() does nothing, let's simply not call it

formatting nits

no functional change

Cosmetic fix

hw: Check for TSC support and invariant TSC

hw: Fix conditions in bmk_cpu_block and simplify common codepath

While removing the while loop in bmk_cpu_block(), the wrong condition
for running into the main codepath crept in. This was manifesting as
clock running slow on x86_64.

This commit fixes the conditions and cleans up the codepath to make the
control flow for the common case easier to follow.

hw: Return from bmk_cpu_block() on any interrupt

If we get a non-timer interrupt then this may schedule work, so just
return into the scheduler after coming out of hlt().

hw: Decouple clock.c from rump kernel HZ

Clock setup does not actually depend on HZ in any way, so don't refer to
it.

hw: Add critical sections in arch/x86/clock.c

Add three critical sections:

1) While reading RTC wall time.
2) While calibrating the TSC.
3) When updating time_base and tsc_base in bmk_cpu_clock_now().

Technically the first two are not required since interrupts should not
be running at this point, but play it safe in case someone reorders the
platform initialisation code later.

hw: Move rdtsc() to include/arch/x86/inline.h

Fix typo in 64-bit version of mul64_32

hw: Implement wall time using RTC offset at boot

Addresses issue #31.

hw: Calculate tick/nsec conversions using 128-bit multiplication

Calculate tick to nsec / nsec to tick conversions using
128-bit multiplication. This avoids overflows, and in the case of
converting nanoseconds to PIT ticks is also slightly faster as it does
not use division.

hw: Dynamically program PIT in one-shot mode when idle

Instead of running the clock interrupt with a fixed periodic tick when
idle, dynamically program the PIT in one-shot mode to fire a single
interrupt after the requested delay has elapsed.

hw: Calibrate TSC and correctly scale monotonic time

This is the first phase of addressing issue #30. We calibrate the TSC at
boot against the i8254 timer and calculate a scaleing factor for
monotonic time.

hw: Consolidate clock-related code

This commit consolidates the common clock-related functions for the x86
and amd64 hw platform into platform/hw/arch/x86/clock.c.

No functional change intended.