Use sel4runtime for TLS management.

sel4runtime will initilise the TLS for the first thread in the process
(this should not be overridden as it was in entry.c).

Instead of using the nasty trick of de-initialising the TLS with a null
pointer to set variables in remote TLS, we use the tools from
sel4runtime that aren't so destructive to configure the state for libbmk
and to also ensure that the IPC buffer pointer is set on each thread.

The existing tooling for creating and allocating TLS can still be used
as sel4runtime makes no assumptions or changes to the size or layout of
TLS.

THe platform-independant interface for configuring the TLS from
sel4runtime is also used rather than specific implementations for
different architectures.

clock routines don't belong in sched.h, move to platform.h

Retire sched_nanosleep(), blockprepare_timeout() is sufficient

Hide raw bmk_sched() from public and provide bmk_sched_block()

In addition to being a responsible abstraction, fixes
scheduler assert meltdown reported as part of issue #20

Remove the "thread" argument from sched_block()

Also rename sched_block() to sched_blockprepare(). The only perverted
routines that need to pass the thread argument to cancel threads
from being able to be scheduled derives from libpthread, and that
has completely different semantics from block, so give it its own
suspend()/unsuspend() interfaces (currently unimplemented).

bmk_sched_current() -> bmk_current

Use TLS for "current" instead of stack-based retrieval.

separate tls switch from context switch

Use TLS for the _lwp implementation.

Allocate extra space in TLS for the stack protector canary.

It's difficult to catch all the places where -fno-stack-protector
is required, so at least ensure that the stack protector doesn't
dereference random garbage.

Give "sleep indefinitely" a name: BMK_SCHED_BLOCK_INFTIME.

Rename bmk_clock_*() to bmk_platform_clock_*()

They are expected to be implemented by the underlying platform.

Common name for thread switch routine: bmk_platform_cpu_sched_switch().

Not sure if it can be unified, since it needs to handle the TLS context
switch, which is apparently slightly different in a virtualized env and
on bare metal, so keep the implementations separate for now.

allow bmk_sched_init() to pass an arg to the main function

Match baremetal "curlwp" with xen

i.e. store thread pointer at the end of the stack. This is done
merely to minimize diffs for now, and will be fixed with tls.

Make bmk_sched_init() jump from bootstrap thread to main thread.

Convert baremetal.

Provide bmk_sched_threadname() in both schedulers

unify timed sleeps in the scheduler

Add a scheduler debug routine for dumping the thread queue.

Fix sched_yield() to actually yield.

add bmk_clock_epochoffset()

Returns bmk_time_t diff between monotonic clock and wall clock
at the time of the call.

use common thread-switching assembly between platforms

declare bmk_clock_monotonic()

returns monotonically increasing nanoseconds

Add copyright/license.

Better late than never ...

Make bmk_time_t a bmk-core type

It's still unsure how many types we should provide in bmk-core, but
this one is useful since it avoids having to keep typing "long long"
all over the place.

Move the baremetal scheduler interface definition to bmk-core.

The implementations for baremetal and mini-os still remain private
(for now).

Add a platform-independent tcb definition to bmk-core

It's MI for now. Even if we have to go MD at some point, this
initial MI approach avoids having identical copies.