These functions are required for std::atomic in C++11 on Cortex-M0.
The functions are only used if GCC does not know how to generate any
lock-free atomic hardware operations.
As discussed in #2725, this commit renames a number of stacksize constants to
better convey their intended usage. In addition, constants for thread priority
are given a `THREAD_` prefix. Changes are:
* KERNEL_CONF_STACKSIZE_PRINTF renamed to THREAD_EXTRA_STACKSIZE_PRINTF
* KERNEL_CONF_STACKSIZE_DEFAULT renamed to THREAD_STACKSIZE_DEFAULT
* KERNEL_CONF_STACKSIZE_IDLE renamed to THREAD_STACKSIZE_IDLE
* KERNEL_CONF_STACKSIZE_MAIN renamed to THREAD_STACKSIZE_MAIN
* Move thread stacksizes from kernel.h to thread.h, since the prefix changed
* PRIORITY_MIN renamed to THREAD_PRIORITY_MIN
* PRIORITY_IDLE renamed to THREAD_PRIORITY_IDLE
* PRIORITY_MAIN renamed to THREAD_PRIORITY_MAIN
* Move thread priorities from kernel.h to thread.h since the prefix has changed
* MINIMUM_STACK_SIZE renamed to THREAD_STACKSIZE_MINIMUM for consistency
These headers do not provide full stl functionality,
but a small subset:
* thread and this_thread
* condition_variable (some timed functions are missing)
* mutex, lock_guard and unique_lock
- Move generic implementation of atomic_set_return to core/atomic.c
- Generic implementation of atomic compare and swap in core/atomic.c
- atomic_cas is used to implement atomic counters in core/include/atomic.h
- atomic_int_t is an atomic integer type
- ATOMIC_INIT can be used as an initializer for atomic_int_t
- ATOMIC_VALUE gets a reference to the value of an atomic integer
Per default, doxygen will take only the part until the first dot into
account for the brief description. In this case this leads to an
ambiguous overview over the IPC send functions.
msg_send_int() sets `m->sender_pid = target_pid`. This was used to flag a
message as having been sent by an ISR.
This PR introduces a static inline function `msg_sent_by_int()` and a
specific define for this purpose.
`thread.c` initializes a thread with an empty message queue. `cib_put()`
will `return -1` for an empty CIB, so there is no need to test if
`thread->msg_array != NULL`.
Fixes#1942.
There were two instances were it was not checked the target thread has a
message queue before queuing the message.
This PR centralizes the check into `queue_msg()`.
If a thread sends blocking, but the target thread is not currently in
receive mode, the sender gets queued. If it has a higher priority it
should run again as soon as the target goes into receiving mode.
Fixes#1708.
Currently involuntary preemption causes the current thread not only to
yield for a higher prioritized thread, but all other threads of its own
priority class, too.
This PR adds the function `thread_yield_higher()`, which will yield the
current thread in favor of higher prioritized functions, but not for
threads of its own priority class.
Boards now need to implement `thread_yield_higher()` instead of
`thread_yield()`, but `COREIF_NG` boards are not affected in any way.
`thread_yield()` retains its old meaning: yield for every thread that
has the same or a higher priority.
This PR does not touch the occurrences of `thread_yield()` in the periph
drivers, because the author of this PR did not look into the logic of
the various driver implementations.
PR #1000 overlooked to rename `runqueues` into `sched_runqueues` in
`sched.h`. This shows that the variable is not used outside of
`sched.c`.
As the list should not be accessed outside of the scheduler, so it
can be `static`.
The current macros in hwtimer.h expect HWTIMER_SPEED to be < 1000000L, otherwise integer arithmetic will round the result down to 0.
Add a case to prevent that.
Boards should define HWTIMER_SPIN_BARRIER that is used to decide
whether it makes sense to set a timer and yield or call hwtimer_spin
instead.
Used by `core/hwtimer.c` and `sys/vtimer/vtimer.c`.
A default value is provided and a warning is printed when it is used.
sched_switch() is called by some library functions when a call unblocks
another thread. Then it needs to be tested if the current thread should
be preempted for the newly runnable thread.
A non-volutarily yield should only happen if the unblocked thread has a
_higher_ priority than the current thread. The current implementation,
which tests if the other thread has the same or a higher priority, does
not fit the documentation.
Right now the core component `clist` is a generic cyclic doubly-linked list.
In the core it is used in `tcb_t::rq_entry`.
Further it is used `net_if.c`.
This commit removes the member `clist_node_t::data` which stored the
pointer to the `tcb_t` instance of which the clist is already a member.
The needless member added `sizeof (int)` bytes to every instance of
`tcb_t`.
In `net_if.c` the clist was used in a type-punned way, so that the
change won't affect it.
Closes#1399.
> Using a different value for SCHED_PRIO_LEVELS for 16 and 32 bit
platforms hurts portability, one thing that we heavily advertise about
RIOT. if you want to write a portable application, then you have to
assume the lower value.
This PR defaults `SCHED_PRIO_LEVELS` to 16 for every board.
In many places we needlessly use `sched_active_thread->pid` whilst we
already have `sched_active_pid` with the same value, and one less
indirection.
`thread_getpid()` is made `static inline` so that there is no penalty in
using this function over accessing `sched_active_pid` directly.
The variable `node` shadows the parameter `node`. The access of
`node->first` would not compile, because there is no member `first` in
`priority_queue_t`.
`tcp_t::stack_size` is only examined by the shell command `ps` and
`DEBUG_PRINT`. For the latter one only if `DEVELHELP` was enabled.
This PR guards the member `tcp_t::stack_size` in `#ifdef DEVELHELP`.
Only if DEVELHELP was activated its value get printed by `ps`.
Closes#1287.
Instead of using differing integer types use kernel_pid_t for process
identifier. This type is introduced in a new header file to avoid
circular dependencies.
This PR converts tabs to white spaces.
The statement I used for the conversion:
'''find . -name "*.[ch]" -exec zsh -c 'expand -t 4 "$0" > /tmp/e && mv /tmp/e "$0"' {} \;'''
Afterwards, I had a quick overview of the converted files to prevent odd indentation.
pid and tcb_t were compared instead of pid and pid
SCHEDSTATISTICS:
- reduce hwtimer_now calls
- dont use thread_last_pid anymore
- increase readability
- remove stray spaces
- remove TODOs:
- MODULE_HWTIMER is not a module anymore
- checking for NULL is necessary, at least without API changes:
`sched_task_exit` sets `sched_active_thread` to `NULL`, then exits,
afterwards `cpu_switch_context_exit` calls `sched_run`
Building MSP boards gives an error, because `VERSION` is somewhere
defined in their toolchain as an integer.
This PR renames `VERSION` into `RIOT_VERSION`, because that's what it
is.
For many modules the `Makefile` contains a line like
```
MODULE:=$(shell basename $(CURDIR))
```
This conclusively shows that we do not have to set the module name
manually.
This PR removes the need to set the module name manually, if it is the
same as the basename. E.g. for `…/sys/vtimer/Makefile` the variable
make `MODULE` will still be `vtimer`, because it is the basename of the
Makefile.
The scheduling gets activated by `kernel_init()` calling
`cpu_switch_context_exit()`. Before this `sched_run()` won't be called.
When it gets called, at least the main thread and the idle thread are
spawned. The idle thread won't die / get killed. So there always is at
least one thread in `runqueue_bitcache`.
Closes#19.
For MSP430 boards oneway-malloc is already used *if* `malloc.h` was
included. The problem is that `malloc.h` is not a standard header, even
though it is common. `stdlib.h` in the right place to look for
`malloc()` and friends.
This change removes this discrepancy. `malloc()` is just named like
that, without the leading underscore. The symbols now are weak, which
means that they won't override library functions if MSP's standard
library will provide these functions at some point. (Unlikely, since
using `malloc()` on tiny systems is less then optimal ...)
Closes#1061 and #863.
Before only the hardware timer's own interrupt was being disabled.
This led to a race condition in the following scenario:
```
Thread1:
hwtimer_remove()
hwtimer_arch_disable_interrupt();
// INTERRUPT -> Thread2 (which has a higher priority than Thread1) gets scheduled
Thread2:
...
hwtimer_remove()
hwtimer_arch_disable_interrupt(); // hwtimer interrupt is already disabled
...
hwtimer_arch_enable_interrupt();
...
// yield | terminate -> Thread1 gets scheduled again
Thread1:
... // these instructions are being run with the hwtimer interrupt enabled
hwtimer_arch_enable_interrupt(); // hwtimer interrupt is already enabled
```
Fixes#924
- Included a collection of cpu-dependent headers in core/include/arch
- Extracted all interfaces that need to be implemented for a cpu
- Created a mapping between those interfaces and the old ones
- added flag for disabling arch interface
- added missing state to lpm_arch interface
- added arch interface for reboot
- fixed newline issues that were pointed out
- documentation fixes to cpu-core interface
and added `UNREACHABLE();` to hint the compiler unreachable lines
added right signature for first parameter of `thread_stack_init()`
added `UNREACHABLE();` macro to `cpu/lpc1768/atom.c` and `cpu/msp430-common/cpu.c`
Closes#726.
The thread state `STATUS_TIMER_WAITING` is not used anymore.
This PR removes the value.
The number in `STATUS_ON_RUNQUEUE` is replaced by a reference.
Change from `void reboot(void)` to `int reboot(int mode)`.
Move reboot definition to core, rename architecture implementations
from reboot to reboot_arch.
Declare reboot mode(s) in kernel.h, reboot_arch in kernel_internal.h
Currently only one reboot mode is handled, its use is enforced.
Rationale:
A reboot function is already defined in <unistd.h> on BSD systems.
(See: http://www.openbsd.org/cgi-bin/man.cgi?query=reboot&sektion=2)
This patch not only allows native to build sensibly on these systems
but also streamlines RIOTs compatability with existing software.
`thread_wakeup` did not check if target pid is invalid.
`thread_wakeup` used `dINT` and `eINT` directly.
It should use `disableIRQ` and `restoreIRQ` instead, because there might
be other (good) reasons why one might want to call `thread_wakeup` with
interrupts disabled.
`thread_wakeup` yielded even if the other thread had a lower priority
than the current thread.
