The current event implementation relies on a dedicated wait object,
which introduces several issues:
1. Increases memory usage
2. Complicates the event logic
3. Makes the API design less clean, as it requires a separate
nxevent_tickwait_wait() function
This patch removes the event’s dependency on the wait object
and eliminates the nxevent_tickwait_wait() API accordingly.
BREAKING CHANGE: this commit removed the nxevent_tickwait_wait function
so the related docs should be updated accordingly
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
If the thread is blocked waiting on a event, then the
thread must be unblocked from the evnet to handle
the task cancellation.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
The current event implementation uses semaphores for wait and post
operations. Since semaphores are relatively heavy-weight and intended
for resource-based synchronization, this is suboptimal.
So this patch replaced the semaphore-based mechanism with direct
scheduler operations to improve performance and reduce memory footprint.
This patch also introduce a new task state TSTATE_WAIT_EVENT to indicate
the task is waiting for a event.
BREAKING CHANGE: This commit introduced a new task state TSTATE_WAIT_EVENT
so apps/nshlib/, procfs/ and tools/pynuttx/nxgdb/ are needed to be updated accordingly.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Restore the use of critical sections to provide mutual exclusion
between event wait and post operations. This allows replacing the
heavier semaphore-based mechanism with direct scheduler operations
for synchronization.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Replace the spinlock/sched_lock pair in the sched wqueue module with
spin_lock_irqsave_nopreempt() to improve code clarity and consistency.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Replace the spinlock/sched_lock pair in the sched clock module with
spin_lock_irqsave_nopreempt() to improve code clarity and consistency.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Refactor nxsched_timeout() to use nxsched_wakeup() in order to
eliminate code duplication and improve maintainability.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Add a new function nxsched_wakeup() to the scheduler,
which allows waking up a sleeping task before its timeout.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Currently, there is duplicated code in nxsig_wait_irq() and nxsig_timeout().
This patch updates the input parameters of nxsig_wait_irq() so that it can
be reused in nxsig_timeout(), reducing code duplication and improving
maintainability.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
nxsig_wait_irq() is always invoked within an existing critical section,
and the task state is already verified, similar to nxsem_wait_irq().
Therefore, the additional critical section protection and task state
check inside the function are redundant and have been removed.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
Replace direct access to the g_waitingforsignal list with the
NuttX-defined macro list_waitingforsignal() to improve code
consistency and maintainability.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Introduce a new function nxsched_tick_expiration() to replace
nxsched_alarm_tick_expiration(). The new function provides a
common implementation that can be shared by both the alarm
and timer architectures.
This change reduces code duplication and provides a unified
function that can be directly reused.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
Improve nxclock_gettime(), to get the system time excluding
the time that the system is suspended, when the clockid is
CLOCK_MONOTONIC
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Replace direct usage of g_system_ticks with clock_increase_sched_ticks()
and clock_get_sched_ticks()
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
In the current NuttX kernel implementation, there is no function
to represent pure scheduler ticks.
The existing clock_systime_ticks() cannot serve this purpose,
because it keeps increasing even when the scheduler is not running.
This happens since the returned value reflects real hardware time:
as long as the timer or alarm hardware is active,
clock_systime_ticks() will continuously increase.
This patch introduces two new functions that provide access to
scheduler ticks — a counter that only increases when the scheduler
itself is running.
Signed-off-by: Chengdong Wang wangchengdong@lixiang.com
Nuttx currently has 2 types of sleep interfaces:
1. Signal-scheduled sleep: nxsig_sleep() / nxsig_usleep() / nxsig_nanosleep()
Weaknesses:
a. Signal-dependent: The signal-scheduled sleep method is bound to the signal framework, while some driver sleep operations do not depend on signals.
b. Timespec conversion: Signal-scheduled sleep involves timespec conversion, which has a significant impact on performance.
2. Busy sleep: up_mdelay() / up_udelay()
Weaknesses:
a. Does not actively trigger scheduling, occupy the CPU loading.
3. New interfaces: Scheduled sleep: nxsched_sleep() / nxsched_usleep() / nxsched_msleep() / nxsched_ticksleep()
Strengths:
a. Does not depend on the signal framework.
b. Tick-based, without additional computational overhead.
Currently, the Nuttx driver framework extensively uses nxsig_* interfaces. However, the driver does not need to rely on signals or timespec conversion.
Therefore, a new set of APIs is added to reduce dependencies on other modules.
(This PR also aims to make signals optional, further reducing the code size of Nuttx.)
Signed-off-by: chao an <anchao.archer@bytedance.com>
Nuttx currently has 2 types of sleep interfaces:
1. Signal-scheduled sleep: nxsig_sleep() / nxsig_usleep() / nxsig_nanosleep()
Weaknesses:
a. Signal-dependent: The signal-scheduled sleep method is bound to the signal framework, while some driver sleep operations do not depend on signals.
b. Timespec conversion: Signal-scheduled sleep involves timespec conversion, which has a significant impact on performance.
2. Busy sleep: up_mdelay() / up_udelay()
Weaknesses:
a. Does not actively trigger scheduling, occupy the CPU loading.
3. New interfaces: Scheduled sleep: nxsched_sleep() / nxsched_usleep() / nxsched_msleep() / nxsched_ticksleep()
Strengths:
a. Does not depend on the signal framework.
b. Tick-based, without additional computational overhead.
Currently, the Nuttx driver framework extensively uses nxsig_* interfaces. However, the driver does not need to rely on signals or timespec conversion.
Therefore, a new set of APIs is added to reduce dependencies on other modules.
(This PR also aims to make signals optional, further reducing the code size of Nuttx.)
Signed-off-by: chao an <anchao.archer@bytedance.com>
`up_check_intstack` exists only when CONFIG_ARCH_INTERRUPTSTACK
is greater than 0, so the irq should first determine whether
CONFIG_ARCH_INTERRUPTSTACK is greater than 0 to determine
whether up_check_intstack can be called.
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
1. Remove STACKCHECK_SOFTWARE config,
2. Do sp value checking when STACKCHECK_MARGIN == 0,
3. Do margin-based stack check when STACKCHECK_MARGIN > 0,
4. Disable stack check when STACKCHECK_MARGIN == -1
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
clock_systime_timespec() always returns 0, so there is no need to
check the return value in the caller code, let us remove the return
value directly.
Signed-off-by: chao an <anchao.archer@bytedance.com>
The wdog implementation redefined `list_node`, which could cause
conflicts or unexpected behavior when both `wdog/wdog.h` and
`list.h` are included in the same source file. This patch removes
the redundant definition to avoid such issues.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
Previously sigpending() returned all pending signals in the group, which is not
POSIX compliant. It should return signals pending only for the caller, so
a signal send with pthread_kill() intended for another thread should not be
returned (it's not pending for a caller).
This fixes the pthread_create.c/test9 test case from
PSE52 Open Group Threads Test Suite.
Signed-off-by: p-szafonimateusz <p-szafonimateusz@xiaomi.com>
Merge suspend and resume into one switch to optimize performance
Co-authored-by: yinshengkai <yinshengkai@xiaomi.com>
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
move both suspned and resume logic to nxsched_switch_context
Co-authored-by: yinshengkai <yinshengkai@xiaomi.com>
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
After irq_dispatch finished, the interrupt stack will be checked to
determine whether overflow occurs.
The relevant configuration reuses the configuration of stack overflow
detection during context switching.
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
Many different architectures re-implemented the exact same code for
`up_*delay` because it was originally written as architecture dependent
code. Busy-waiting can be done regardless of architecture, so this
commit moves that duplicated implementation to a common file with weak
definitions so that individual architectures (see tc32) are still able
to override the definition if needed/desired.
Default implementation is not included if ARCH_TIMER is enabled, since
it is more accurate and provides its own weak definitions to override.
Signed-off-by: Matteo Golin <matteo.golin@gmail.com>
Many times, context switching does not occur when the thread stack
memory is almost exhausted, so we need to mofify up_check_tcbstack to
accurately detect it.
Co-authored-by: Chengdong Wang <wangchengdong@lixiang.com>
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
Added two detection methods:
1. Determine by detecting the number of bytes specified at the bottom of the stack.
2. Check if the `sp` register is out of bounds.
Co-authored-by: Chengdong Wang <wangchengdong@lixiang.com>
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
Replace all nxsched_suspend/resume_*** with nxsched_switch_context
Delete nxsched_resume_scheduler in up_exit and call it uniformly in task_exit
Co-authored-by: yinshengkai <yinshengkai@xiaomi.com>
Signed-off-by: guoshengyuan1 <guoshengyuan1@xiaomi.com>
In certain cases, when the event-wait thread is awakened,
it needs to know the specific mask bit that triggered it,
so that it can branch to different paths accordingly.
This patch introduces the nxevent_getmask function to address this requirement.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
Change:
Replace list_delete_init() with list_delete() in nxevent_post.
Reason:
PR 16933 removed list_delete() in nxevent_wait() and allow other task
to preempt event-wait thread after list_add_tail() option, the even-post thread
can preempt it at this time and do list_delete_init().
The problem is that list_delete_init() will make list_in_list() return true, so
DEBUGASSERT(!list_in_list(&(wait->node))); will fail in nxevent_wait() after the event-wait thread
returns.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
Curernt implementation default semaphore max allowed value to SEM_VALUE_MAX.
In some cases, user may want to change this, so provide a function to do this: sem_setmaxvalue
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
There are cases that users do not want the event wait object to be allocated automatically
in the stack as a temporary varialbe in nxevent_tickwait, since multiple threads will
access the varialbe, they want to allocate the object as a global variable for safety
control or easy debug. To solve this problem, this patch add a new function nxevent_tickwait_wait
implementation to give user the chance to pass the global wait object to it.
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
The nxsig_dispatch should just deliver the signal to either a
thread by pid (tid) or to the process (group) by pid.
Simplify the code so that the intent is more obvious.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
The signal dispatch is called from interrupt handlers as well, so
this_task() is wrong. The thread to which the signal is supposed to
be delivered is known (stcb), use that.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
In a flat build, a separate init thread should not be mandatory,
users can create a user task in the board early initialization or late initialization hook.
For example, set sigaction after create signalfd,
the sa_sigaction was restored but sa_user not,
causing signalfd_action() get the wild private data.
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
This commit adds simple implementation of guardsize for pthreads.
At this moment this option simply increases the size of allocated pthread stack.
At default pthread guard size is set to 0.
Signed-off-by: p-szafonimateusz <p-szafonimateusz@xiaomi.com>
When the task has TCB_FLAG_CPU_LOCKED it is locked to the CPU regardless
of the affinity. There is no need to switch the affinity back and forth.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
Since g_assignedtasks only holds the running task for each CPU, it can
be just a vector. Idle tasks are already preserved in statically allocated
structures "g_idletcb", and can be used from there.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
This fixes several places, where the sched erroneously checks if the scheduling
is locked on current cpu/task, when it should check if the scheduling is locked
on the target cpu/task.
The original code randomly caused a task to be added to the pending list, and never
taken out from there, leading to system halt.
For SMP, there is no need for the pending list. Each CPU has got it's own
running list (assigned tasks list), and pending tasks can just be kept in
the unassigned (readytorun) list.
In addition, the smp scheduling is changed in a way that every CPU just picks
up the tasks from the ready-to-run list themselves. Which task to pick is
not tried to be dictated by another CPU.
This also allows using up_send_smp_sched for asynchronously
- re-prioritizing a running task
- triggering round robin scheduling switch
Iow, no separate smp call mechanism is needed for those and the code can be simplified.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
Change every occurence of up_switch_context to use this_task() as the first parameter.
"nxsched_add_readytorun" returns "true" if context switch is required. "nxsched_add_readytorun"
typically could only switch the assigned/running task to the one which is passed in as parameter.
But this is not strictly guaranteed in SMP; if other CPUs tweak with affinities or priorities,
it may be possible that the running task after the call is changed, but is some other
task from the readytorun list (and it should be, if there is higher priority one available or the
affinity of the added task prevents it to be scheduled in, but the previous head of the readytorun
list should run.
this_task() is always the correct one to switch to, since it always points to the tcb which was
just switched in by the nxsched_add_readytorun.
This is also a precursor to re-writing the SMP queue logic to remove pending lists for SMP.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
When exiting schedlock, that task should first take the critical section and
only after that decrease the lockcount to 0. Otherwise an interrupt might
cause a re-schedule before the task enters the critical section, which makes
the following code meaningless.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
The current setlogmask call used in coredump_dump_syslog specifies a raw log level
instead of a bitmask, and this causes wrong evaluations later on when that value
is checked against a mask. Therefore the LOG_UPTO macro is added for conversion.
Signed-off-by: Niccolò Maggioni <nicco.maggioni+nuttx@gmail.com>
