Sasan Golchin authored 9 years ago and Federico committed 3 years ago

The kernel's API has been amended such that the periodic context
switch works along with exact awakening of the sleeping threads:

1) SleepData is not in terms of kernel's tick anymore but directly
in terms of context switch timer tick.
2) Sleeping threads are being awakened in the time they have requested
rather than just waiting for the context switch to happen.
3) Three functions are added to the kernel's API:
-tickSleepUntil(tick) which is private to the kernel and is the main
reference of other sleep functions of a thread and it's implemented as
an inline function to keep both maintainability and efficiency
-Thread::nanoSleep
-Thread::nanoSleepUntil

4) Besides sleep functions, the following functions has been updated
in order to keep the timer interrupt sync with the head of the sleeping
list.
I'm not sure about this being correct and it may have some conflicts
with the scheduler's policy:
-IRQwakeThreads: After removing the head of the sleeping list, the next
interrupt should be set to the head again! Another change in this
is that it does not expect the exact equivalence between timer's tick
and wakeup_time as when the kernel reaches that particular point the
timer's tick has already passed the wakeup_time.
-IRQaddToSleepingList: Due to the change in order of items in the
sleeping list, the interrupt should be set again.

5) Constructor and Interrupt handler of the context switch timer has
been amended in order to put the next context switch time as a record
in the sleeping list.

As the meaning of tick in sleep functions has been changed, the
test suit does not work properly with option t anymore but some critical
tests has been done on the current state of the project and everything
seems to work fine but some optimizations are needed specially in the
case that the time window between items in the list is very short.

Signed-off-by: Sasan Golchin <ahmad.golchin@mail.polimi.it>
Signed-off-by: Terraneo Federico <fede.tft@miosix.org>

Sasan Golchin authored 9 years ago and Federico committed 3 years ago

A generic interface miosix::ContextSwitchTimer has been added to the
kernel that can play the role of either periodic/aperiodic system timer.
Like other kernel interfaces, there should be provided a proper
implementation for cstimer.

Currently the only available implementation is done for cortexM4_stm32f
which is based on TIM2 timer. Although the implementation is an
aperiodic timer, it is set to act like a periodic timer with T=1ms.

Signed-off-by: Sasan Golchin <ahmad.golchin@mail.polimi.it>
Signed-off-by: Terraneo Federico <fede.tft@miosix.org>

Sasan Golchin authored 9 years ago and Federico committed 3 years ago

In this subproject, an 64 bit apriodic timer is created based on
TIM2 in which the overflow is managed automatically and the user
is able to set checkpoints (in terms of timer ticks) to fire IRQ.

An expected usecase is the use this driver as the context switch
tool for the kernel.
There might be some errors in which overflows/checkpoints are not
managed correctly that will be OK asap.

Signed-off-by: sasan_golchin <ahmad.golchin@mail.polimi.it>

Signed-off-by: sasan_golchin <ahmad.golchin@mail.polimi.it>
Signed-off-by: Terraneo Federico <fede.tft@miosix.org>