88ecf576f4
Added note k_busy_wait may not work when CONFIG_SYSTEM_CLOCK_SLOPPY_IDLE and CONFIG_PM are enabled. In such case timer used for time measurement may be disabled. Fixes: #53522 Signed-off-by: Artur Lipowski <artur.lipowski@hidglobal.com>
6107 lines
184 KiB
C
6107 lines
184 KiB
C
/*
|
|
* Copyright (c) 2016, Wind River Systems, Inc.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
/**
|
|
* @file
|
|
*
|
|
* @brief Public kernel APIs.
|
|
*/
|
|
|
|
#ifndef ZEPHYR_INCLUDE_KERNEL_H_
|
|
#define ZEPHYR_INCLUDE_KERNEL_H_
|
|
|
|
#if !defined(_ASMLANGUAGE)
|
|
#include <zephyr/kernel_includes.h>
|
|
#include <errno.h>
|
|
#include <limits.h>
|
|
#include <stdbool.h>
|
|
#include <zephyr/toolchain.h>
|
|
#include <zephyr/tracing/tracing_macros.h>
|
|
#include <zephyr/sys/mem_stats.h>
|
|
#include <zephyr/sys/iterable_sections.h>
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
/*
|
|
* Zephyr currently assumes the size of a couple standard types to simplify
|
|
* print string formats. Let's make sure this doesn't change without notice.
|
|
*/
|
|
BUILD_ASSERT(sizeof(int32_t) == sizeof(int));
|
|
BUILD_ASSERT(sizeof(int64_t) == sizeof(long long));
|
|
BUILD_ASSERT(sizeof(intptr_t) == sizeof(long));
|
|
|
|
/**
|
|
* @brief Kernel APIs
|
|
* @defgroup kernel_apis Kernel APIs
|
|
* @{
|
|
* @}
|
|
*/
|
|
|
|
#define K_ANY NULL
|
|
|
|
#if CONFIG_NUM_COOP_PRIORITIES + CONFIG_NUM_PREEMPT_PRIORITIES == 0
|
|
#error Zero available thread priorities defined!
|
|
#endif
|
|
|
|
#define K_PRIO_COOP(x) (-(CONFIG_NUM_COOP_PRIORITIES - (x)))
|
|
#define K_PRIO_PREEMPT(x) (x)
|
|
|
|
#define K_HIGHEST_THREAD_PRIO (-CONFIG_NUM_COOP_PRIORITIES)
|
|
#define K_LOWEST_THREAD_PRIO CONFIG_NUM_PREEMPT_PRIORITIES
|
|
#define K_IDLE_PRIO K_LOWEST_THREAD_PRIO
|
|
#define K_HIGHEST_APPLICATION_THREAD_PRIO (K_HIGHEST_THREAD_PRIO)
|
|
#define K_LOWEST_APPLICATION_THREAD_PRIO (K_LOWEST_THREAD_PRIO - 1)
|
|
|
|
#ifdef CONFIG_POLL
|
|
#define _POLL_EVENT_OBJ_INIT(obj) \
|
|
.poll_events = SYS_DLIST_STATIC_INIT(&obj.poll_events),
|
|
#define _POLL_EVENT sys_dlist_t poll_events
|
|
#else
|
|
#define _POLL_EVENT_OBJ_INIT(obj)
|
|
#define _POLL_EVENT
|
|
#endif
|
|
|
|
struct k_thread;
|
|
struct k_mutex;
|
|
struct k_sem;
|
|
struct k_msgq;
|
|
struct k_mbox;
|
|
struct k_pipe;
|
|
struct k_queue;
|
|
struct k_fifo;
|
|
struct k_lifo;
|
|
struct k_stack;
|
|
struct k_mem_slab;
|
|
struct k_timer;
|
|
struct k_poll_event;
|
|
struct k_poll_signal;
|
|
struct k_mem_domain;
|
|
struct k_mem_partition;
|
|
struct k_futex;
|
|
struct k_event;
|
|
|
|
enum execution_context_types {
|
|
K_ISR = 0,
|
|
K_COOP_THREAD,
|
|
K_PREEMPT_THREAD,
|
|
};
|
|
|
|
/* private, used by k_poll and k_work_poll */
|
|
struct k_work_poll;
|
|
typedef int (*_poller_cb_t)(struct k_poll_event *event, uint32_t state);
|
|
|
|
/**
|
|
* @addtogroup thread_apis
|
|
* @{
|
|
*/
|
|
|
|
typedef void (*k_thread_user_cb_t)(const struct k_thread *thread,
|
|
void *user_data);
|
|
|
|
/**
|
|
* @brief Iterate over all the threads in the system.
|
|
*
|
|
* This routine iterates over all the threads in the system and
|
|
* calls the user_cb function for each thread.
|
|
*
|
|
* @param user_cb Pointer to the user callback function.
|
|
* @param user_data Pointer to user data.
|
|
*
|
|
* @note @kconfig{CONFIG_THREAD_MONITOR} must be set for this function
|
|
* to be effective.
|
|
* @note This API uses @ref k_spin_lock to protect the _kernel.threads
|
|
* list which means creation of new threads and terminations of existing
|
|
* threads are blocked until this API returns.
|
|
*/
|
|
extern void k_thread_foreach(k_thread_user_cb_t user_cb, void *user_data);
|
|
|
|
/**
|
|
* @brief Iterate over all the threads in the system without locking.
|
|
*
|
|
* This routine works exactly the same like @ref k_thread_foreach
|
|
* but unlocks interrupts when user_cb is executed.
|
|
*
|
|
* @param user_cb Pointer to the user callback function.
|
|
* @param user_data Pointer to user data.
|
|
*
|
|
* @note @kconfig{CONFIG_THREAD_MONITOR} must be set for this function
|
|
* to be effective.
|
|
* @note This API uses @ref k_spin_lock only when accessing the _kernel.threads
|
|
* queue elements. It unlocks it during user callback function processing.
|
|
* If a new task is created when this @c foreach function is in progress,
|
|
* the added new task would not be included in the enumeration.
|
|
* If a task is aborted during this enumeration, there would be a race here
|
|
* and there is a possibility that this aborted task would be included in the
|
|
* enumeration.
|
|
* @note If the task is aborted and the memory occupied by its @c k_thread
|
|
* structure is reused when this @c k_thread_foreach_unlocked is in progress
|
|
* it might even lead to the system behave unstable.
|
|
* This function may never return, as it would follow some @c next task
|
|
* pointers treating given pointer as a pointer to the k_thread structure
|
|
* while it is something different right now.
|
|
* Do not reuse the memory that was occupied by k_thread structure of aborted
|
|
* task if it was aborted after this function was called in any context.
|
|
*/
|
|
extern void k_thread_foreach_unlocked(
|
|
k_thread_user_cb_t user_cb, void *user_data);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @defgroup thread_apis Thread APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
#endif /* !_ASMLANGUAGE */
|
|
|
|
|
|
/*
|
|
* Thread user options. May be needed by assembly code. Common part uses low
|
|
* bits, arch-specific use high bits.
|
|
*/
|
|
|
|
/**
|
|
* @brief system thread that must not abort
|
|
* */
|
|
#define K_ESSENTIAL (BIT(0))
|
|
|
|
#if defined(CONFIG_FPU_SHARING)
|
|
/**
|
|
* @brief FPU registers are managed by context switch
|
|
*
|
|
* @details
|
|
* This option indicates that the thread uses the CPU's floating point
|
|
* registers. This instructs the kernel to take additional steps to save
|
|
* and restore the contents of these registers when scheduling the thread.
|
|
* No effect if @kconfig{CONFIG_FPU_SHARING} is not enabled.
|
|
*/
|
|
#define K_FP_IDX 1
|
|
#define K_FP_REGS (BIT(K_FP_IDX))
|
|
#endif
|
|
|
|
/**
|
|
* @brief user mode thread
|
|
*
|
|
* This thread has dropped from supervisor mode to user mode and consequently
|
|
* has additional restrictions
|
|
*/
|
|
#define K_USER (BIT(2))
|
|
|
|
/**
|
|
* @brief Inherit Permissions
|
|
*
|
|
* @details
|
|
* Indicates that the thread being created should inherit all kernel object
|
|
* permissions from the thread that created it. No effect if
|
|
* @kconfig{CONFIG_USERSPACE} is not enabled.
|
|
*/
|
|
#define K_INHERIT_PERMS (BIT(3))
|
|
|
|
/**
|
|
* @brief Callback item state
|
|
*
|
|
* @details
|
|
* This is a single bit of state reserved for "callback manager"
|
|
* utilities (p4wq initially) who need to track operations invoked
|
|
* from within a user-provided callback they have been invoked.
|
|
* Effectively it serves as a tiny bit of zero-overhead TLS data.
|
|
*/
|
|
#define K_CALLBACK_STATE (BIT(4))
|
|
|
|
#ifdef CONFIG_ARC
|
|
/* ARC processor Bitmask definitions for threads user options */
|
|
|
|
#if defined(CONFIG_ARC_DSP_SHARING)
|
|
/**
|
|
* @brief DSP registers are managed by context switch
|
|
*
|
|
* @details
|
|
* This option indicates that the thread uses the CPU's DSP registers.
|
|
* This instructs the kernel to take additional steps to save and
|
|
* restore the contents of these registers when scheduling the thread.
|
|
* No effect if @kconfig{CONFIG_ARC_DSP_SHARING} is not enabled.
|
|
*/
|
|
#define K_DSP_IDX 6
|
|
#define K_ARC_DSP_REGS (BIT(K_DSP_IDX))
|
|
#endif
|
|
|
|
#if defined(CONFIG_ARC_AGU_SHARING)
|
|
/**
|
|
* @brief AGU registers are managed by context switch
|
|
*
|
|
* @details
|
|
* This option indicates that the thread uses the ARC processor's XY
|
|
* memory and DSP feature. Often used with @kconfig{CONFIG_ARC_AGU_SHARING}.
|
|
* No effect if @kconfig{CONFIG_ARC_AGU_SHARING} is not enabled.
|
|
*/
|
|
#define K_AGU_IDX 7
|
|
#define K_ARC_AGU_REGS (BIT(K_AGU_IDX))
|
|
#endif
|
|
#endif
|
|
|
|
#ifdef CONFIG_X86
|
|
/* x86 Bitmask definitions for threads user options */
|
|
|
|
#if defined(CONFIG_FPU_SHARING) && defined(CONFIG_X86_SSE)
|
|
/**
|
|
* @brief FP and SSE registers are managed by context switch on x86
|
|
*
|
|
* @details
|
|
* This option indicates that the thread uses the x86 CPU's floating point
|
|
* and SSE registers. This instructs the kernel to take additional steps to
|
|
* save and restore the contents of these registers when scheduling
|
|
* the thread. No effect if @kconfig{CONFIG_X86_SSE} is not enabled.
|
|
*/
|
|
#define K_SSE_REGS (BIT(7))
|
|
#endif
|
|
#endif
|
|
|
|
/* end - thread options */
|
|
|
|
#if !defined(_ASMLANGUAGE)
|
|
/**
|
|
* @brief Dynamically allocate a thread stack.
|
|
*
|
|
* Relevant stack creation flags include:
|
|
* - @ref K_USER allocate a userspace thread (requires `CONFIG_USERSPACE=y`)
|
|
*
|
|
* @param size Stack size in bytes.
|
|
* @param flags Stack creation flags, or 0.
|
|
*
|
|
* @retval the allocated thread stack on success.
|
|
* @retval NULL on failure.
|
|
*
|
|
* @see CONFIG_DYNAMIC_THREAD
|
|
*/
|
|
__syscall k_thread_stack_t *k_thread_stack_alloc(size_t size, int flags);
|
|
|
|
/**
|
|
* @brief Free a dynamically allocated thread stack.
|
|
*
|
|
* @param stack Pointer to the thread stack.
|
|
*
|
|
* @retval 0 on success.
|
|
* @retval -EBUSY if the thread stack is in use.
|
|
* @retval -EINVAL if @p stack is invalid.
|
|
* @retval -ENOSYS if dynamic thread stack allocation is disabled
|
|
*
|
|
* @see CONFIG_DYNAMIC_THREAD
|
|
*/
|
|
__syscall int k_thread_stack_free(k_thread_stack_t *stack);
|
|
|
|
/**
|
|
* @brief Create a thread.
|
|
*
|
|
* This routine initializes a thread, then schedules it for execution.
|
|
*
|
|
* The new thread may be scheduled for immediate execution or a delayed start.
|
|
* If the newly spawned thread does not have a delayed start the kernel
|
|
* scheduler may preempt the current thread to allow the new thread to
|
|
* execute.
|
|
*
|
|
* Thread options are architecture-specific, and can include K_ESSENTIAL,
|
|
* K_FP_REGS, and K_SSE_REGS. Multiple options may be specified by separating
|
|
* them using "|" (the logical OR operator).
|
|
*
|
|
* Stack objects passed to this function must be originally defined with
|
|
* either of these macros in order to be portable:
|
|
*
|
|
* - K_THREAD_STACK_DEFINE() - For stacks that may support either user or
|
|
* supervisor threads.
|
|
* - K_KERNEL_STACK_DEFINE() - For stacks that may support supervisor
|
|
* threads only. These stacks use less memory if CONFIG_USERSPACE is
|
|
* enabled.
|
|
*
|
|
* The stack_size parameter has constraints. It must either be:
|
|
*
|
|
* - The original size value passed to K_THREAD_STACK_DEFINE() or
|
|
* K_KERNEL_STACK_DEFINE()
|
|
* - The return value of K_THREAD_STACK_SIZEOF(stack) if the stack was
|
|
* defined with K_THREAD_STACK_DEFINE()
|
|
* - The return value of K_KERNEL_STACK_SIZEOF(stack) if the stack was
|
|
* defined with K_KERNEL_STACK_DEFINE().
|
|
*
|
|
* Using other values, or sizeof(stack) may produce undefined behavior.
|
|
*
|
|
* @param new_thread Pointer to uninitialized struct k_thread
|
|
* @param stack Pointer to the stack space.
|
|
* @param stack_size Stack size in bytes.
|
|
* @param entry Thread entry function.
|
|
* @param p1 1st entry point parameter.
|
|
* @param p2 2nd entry point parameter.
|
|
* @param p3 3rd entry point parameter.
|
|
* @param prio Thread priority.
|
|
* @param options Thread options.
|
|
* @param delay Scheduling delay, or K_NO_WAIT (for no delay).
|
|
*
|
|
* @return ID of new thread.
|
|
*
|
|
*/
|
|
__syscall k_tid_t k_thread_create(struct k_thread *new_thread,
|
|
k_thread_stack_t *stack,
|
|
size_t stack_size,
|
|
k_thread_entry_t entry,
|
|
void *p1, void *p2, void *p3,
|
|
int prio, uint32_t options, k_timeout_t delay);
|
|
|
|
/**
|
|
* @brief Drop a thread's privileges permanently to user mode
|
|
*
|
|
* This allows a supervisor thread to be re-used as a user thread.
|
|
* This function does not return, but control will transfer to the provided
|
|
* entry point as if this was a new user thread.
|
|
*
|
|
* The implementation ensures that the stack buffer contents are erased.
|
|
* Any thread-local storage will be reverted to a pristine state.
|
|
*
|
|
* Memory domain membership, resource pool assignment, kernel object
|
|
* permissions, priority, and thread options are preserved.
|
|
*
|
|
* A common use of this function is to re-use the main thread as a user thread
|
|
* once all supervisor mode-only tasks have been completed.
|
|
*
|
|
* @param entry Function to start executing from
|
|
* @param p1 1st entry point parameter
|
|
* @param p2 2nd entry point parameter
|
|
* @param p3 3rd entry point parameter
|
|
*/
|
|
extern FUNC_NORETURN void k_thread_user_mode_enter(k_thread_entry_t entry,
|
|
void *p1, void *p2,
|
|
void *p3);
|
|
|
|
/**
|
|
* @brief Grant a thread access to a set of kernel objects
|
|
*
|
|
* This is a convenience function. For the provided thread, grant access to
|
|
* the remaining arguments, which must be pointers to kernel objects.
|
|
*
|
|
* The thread object must be initialized (i.e. running). The objects don't
|
|
* need to be.
|
|
* Note that NULL shouldn't be passed as an argument.
|
|
*
|
|
* @param thread Thread to grant access to objects
|
|
* @param ... list of kernel object pointers
|
|
*/
|
|
#define k_thread_access_grant(thread, ...) \
|
|
FOR_EACH_FIXED_ARG(k_object_access_grant, (;), thread, __VA_ARGS__)
|
|
|
|
/**
|
|
* @brief Assign a resource memory pool to a thread
|
|
*
|
|
* By default, threads have no resource pool assigned unless their parent
|
|
* thread has a resource pool, in which case it is inherited. Multiple
|
|
* threads may be assigned to the same memory pool.
|
|
*
|
|
* Changing a thread's resource pool will not migrate allocations from the
|
|
* previous pool.
|
|
*
|
|
* @param thread Target thread to assign a memory pool for resource requests.
|
|
* @param heap Heap object to use for resources,
|
|
* or NULL if the thread should no longer have a memory pool.
|
|
*/
|
|
static inline void k_thread_heap_assign(struct k_thread *thread,
|
|
struct k_heap *heap)
|
|
{
|
|
thread->resource_pool = heap;
|
|
}
|
|
|
|
#if defined(CONFIG_INIT_STACKS) && defined(CONFIG_THREAD_STACK_INFO)
|
|
/**
|
|
* @brief Obtain stack usage information for the specified thread
|
|
*
|
|
* User threads will need to have permission on the target thread object.
|
|
*
|
|
* Some hardware may prevent inspection of a stack buffer currently in use.
|
|
* If this API is called from supervisor mode, on the currently running thread,
|
|
* on a platform which selects @kconfig{CONFIG_NO_UNUSED_STACK_INSPECTION}, an
|
|
* error will be generated.
|
|
*
|
|
* @param thread Thread to inspect stack information
|
|
* @param unused_ptr Output parameter, filled in with the unused stack space
|
|
* of the target thread in bytes.
|
|
* @return 0 on success
|
|
* @return -EBADF Bad thread object (user mode only)
|
|
* @return -EPERM No permissions on thread object (user mode only)
|
|
* #return -ENOTSUP Forbidden by hardware policy
|
|
* @return -EINVAL Thread is uninitialized or exited (user mode only)
|
|
* @return -EFAULT Bad memory address for unused_ptr (user mode only)
|
|
*/
|
|
__syscall int k_thread_stack_space_get(const struct k_thread *thread,
|
|
size_t *unused_ptr);
|
|
#endif
|
|
|
|
#if (CONFIG_HEAP_MEM_POOL_SIZE > 0)
|
|
/**
|
|
* @brief Assign the system heap as a thread's resource pool
|
|
*
|
|
* Similar to k_thread_heap_assign(), but the thread will use
|
|
* the kernel heap to draw memory.
|
|
*
|
|
* Use with caution, as a malicious thread could perform DoS attacks on the
|
|
* kernel heap.
|
|
*
|
|
* @param thread Target thread to assign the system heap for resource requests
|
|
*
|
|
*/
|
|
void k_thread_system_pool_assign(struct k_thread *thread);
|
|
#endif /* (CONFIG_HEAP_MEM_POOL_SIZE > 0) */
|
|
|
|
/**
|
|
* @brief Sleep until a thread exits
|
|
*
|
|
* The caller will be put to sleep until the target thread exits, either due
|
|
* to being aborted, self-exiting, or taking a fatal error. This API returns
|
|
* immediately if the thread isn't running.
|
|
*
|
|
* This API may only be called from ISRs with a K_NO_WAIT timeout,
|
|
* where it can be useful as a predicate to detect when a thread has
|
|
* aborted.
|
|
*
|
|
* @param thread Thread to wait to exit
|
|
* @param timeout upper bound time to wait for the thread to exit.
|
|
* @retval 0 success, target thread has exited or wasn't running
|
|
* @retval -EBUSY returned without waiting
|
|
* @retval -EAGAIN waiting period timed out
|
|
* @retval -EDEADLK target thread is joining on the caller, or target thread
|
|
* is the caller
|
|
*/
|
|
__syscall int k_thread_join(struct k_thread *thread, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Put the current thread to sleep.
|
|
*
|
|
* This routine puts the current thread to sleep for @a duration,
|
|
* specified as a k_timeout_t object.
|
|
*
|
|
* @note if @a timeout is set to K_FOREVER then the thread is suspended.
|
|
*
|
|
* @param timeout Desired duration of sleep.
|
|
*
|
|
* @return Zero if the requested time has elapsed or the number of milliseconds
|
|
* left to sleep, if thread was woken up by \ref k_wakeup call.
|
|
*/
|
|
__syscall int32_t k_sleep(k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Put the current thread to sleep.
|
|
*
|
|
* This routine puts the current thread to sleep for @a duration milliseconds.
|
|
*
|
|
* @param ms Number of milliseconds to sleep.
|
|
*
|
|
* @return Zero if the requested time has elapsed or the number of milliseconds
|
|
* left to sleep, if thread was woken up by \ref k_wakeup call.
|
|
*/
|
|
static inline int32_t k_msleep(int32_t ms)
|
|
{
|
|
return k_sleep(Z_TIMEOUT_MS(ms));
|
|
}
|
|
|
|
/**
|
|
* @brief Put the current thread to sleep with microsecond resolution.
|
|
*
|
|
* This function is unlikely to work as expected without kernel tuning.
|
|
* In particular, because the lower bound on the duration of a sleep is
|
|
* the duration of a tick, @kconfig{CONFIG_SYS_CLOCK_TICKS_PER_SEC} must be
|
|
* adjusted to achieve the resolution desired. The implications of doing
|
|
* this must be understood before attempting to use k_usleep(). Use with
|
|
* caution.
|
|
*
|
|
* @param us Number of microseconds to sleep.
|
|
*
|
|
* @return Zero if the requested time has elapsed or the number of microseconds
|
|
* left to sleep, if thread was woken up by \ref k_wakeup call.
|
|
*/
|
|
__syscall int32_t k_usleep(int32_t us);
|
|
|
|
/**
|
|
* @brief Cause the current thread to busy wait.
|
|
*
|
|
* This routine causes the current thread to execute a "do nothing" loop for
|
|
* @a usec_to_wait microseconds.
|
|
*
|
|
* @note The clock used for the microsecond-resolution delay here may
|
|
* be skewed relative to the clock used for system timeouts like
|
|
* k_sleep(). For example k_busy_wait(1000) may take slightly more or
|
|
* less time than k_sleep(K_MSEC(1)), with the offset dependent on
|
|
* clock tolerances.
|
|
*
|
|
* @note In case when @kconfig{CONFIG_SYSTEM_CLOCK_SLOPPY_IDLE} and
|
|
* @kconfig{CONFIG_PM} options are enabled, this function may not work.
|
|
* The timer/clock used for delay processing may be disabled/inactive.
|
|
*/
|
|
__syscall void k_busy_wait(uint32_t usec_to_wait);
|
|
|
|
/**
|
|
* @brief Check whether it is possible to yield in the current context.
|
|
*
|
|
* This routine checks whether the kernel is in a state where it is possible to
|
|
* yield or call blocking API's. It should be used by code that needs to yield
|
|
* to perform correctly, but can feasibly be called from contexts where that
|
|
* is not possible. For example in the PRE_KERNEL initialization step, or when
|
|
* being run from the idle thread.
|
|
*
|
|
* @return True if it is possible to yield in the current context, false otherwise.
|
|
*/
|
|
bool k_can_yield(void);
|
|
|
|
/**
|
|
* @brief Yield the current thread.
|
|
*
|
|
* This routine causes the current thread to yield execution to another
|
|
* thread of the same or higher priority. If there are no other ready threads
|
|
* of the same or higher priority, the routine returns immediately.
|
|
*/
|
|
__syscall void k_yield(void);
|
|
|
|
/**
|
|
* @brief Wake up a sleeping thread.
|
|
*
|
|
* This routine prematurely wakes up @a thread from sleeping.
|
|
*
|
|
* If @a thread is not currently sleeping, the routine has no effect.
|
|
*
|
|
* @param thread ID of thread to wake.
|
|
*/
|
|
__syscall void k_wakeup(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Query thread ID of the current thread.
|
|
*
|
|
* This unconditionally queries the kernel via a system call.
|
|
*
|
|
* @note Use k_current_get() unless absolutely sure this is necessary.
|
|
* This should only be used directly where the thread local
|
|
* variable cannot be used or may contain invalid values
|
|
* if thread local storage (TLS) is enabled. If TLS is not
|
|
* enabled, this is the same as k_current_get().
|
|
*
|
|
* @return ID of current thread.
|
|
*/
|
|
__attribute_const__
|
|
__syscall k_tid_t k_sched_current_thread_query(void);
|
|
|
|
/**
|
|
* @brief Get thread ID of the current thread.
|
|
*
|
|
* @return ID of current thread.
|
|
*
|
|
*/
|
|
__attribute_const__
|
|
static inline k_tid_t k_current_get(void)
|
|
{
|
|
#ifdef CONFIG_THREAD_LOCAL_STORAGE
|
|
/* Thread-local cache of current thread ID, set in z_thread_entry() */
|
|
extern __thread k_tid_t z_tls_current;
|
|
|
|
return z_tls_current;
|
|
#else
|
|
return k_sched_current_thread_query();
|
|
#endif
|
|
}
|
|
|
|
/**
|
|
* @brief Abort a thread.
|
|
*
|
|
* This routine permanently stops execution of @a thread. The thread is taken
|
|
* off all kernel queues it is part of (i.e. the ready queue, the timeout
|
|
* queue, or a kernel object wait queue). However, any kernel resources the
|
|
* thread might currently own (such as mutexes or memory blocks) are not
|
|
* released. It is the responsibility of the caller of this routine to ensure
|
|
* all necessary cleanup is performed.
|
|
*
|
|
* After k_thread_abort() returns, the thread is guaranteed not to be
|
|
* running or to become runnable anywhere on the system. Normally
|
|
* this is done via blocking the caller (in the same manner as
|
|
* k_thread_join()), but in interrupt context on SMP systems the
|
|
* implementation is required to spin for threads that are running on
|
|
* other CPUs. Note that as specified, this means that on SMP
|
|
* platforms it is possible for application code to create a deadlock
|
|
* condition by simultaneously aborting a cycle of threads using at
|
|
* least one termination from interrupt context. Zephyr cannot detect
|
|
* all such conditions.
|
|
*
|
|
* @param thread ID of thread to abort.
|
|
*/
|
|
__syscall void k_thread_abort(k_tid_t thread);
|
|
|
|
|
|
/**
|
|
* @brief Start an inactive thread
|
|
*
|
|
* If a thread was created with K_FOREVER in the delay parameter, it will
|
|
* not be added to the scheduling queue until this function is called
|
|
* on it.
|
|
*
|
|
* @param thread thread to start
|
|
*/
|
|
__syscall void k_thread_start(k_tid_t thread);
|
|
|
|
extern k_ticks_t z_timeout_expires(const struct _timeout *timeout);
|
|
extern k_ticks_t z_timeout_remaining(const struct _timeout *timeout);
|
|
|
|
#ifdef CONFIG_SYS_CLOCK_EXISTS
|
|
|
|
/**
|
|
* @brief Get time when a thread wakes up, in system ticks
|
|
*
|
|
* This routine computes the system uptime when a waiting thread next
|
|
* executes, in units of system ticks. If the thread is not waiting,
|
|
* it returns current system time.
|
|
*/
|
|
__syscall k_ticks_t k_thread_timeout_expires_ticks(const struct k_thread *t);
|
|
|
|
static inline k_ticks_t z_impl_k_thread_timeout_expires_ticks(
|
|
const struct k_thread *t)
|
|
{
|
|
return z_timeout_expires(&t->base.timeout);
|
|
}
|
|
|
|
/**
|
|
* @brief Get time remaining before a thread wakes up, in system ticks
|
|
*
|
|
* This routine computes the time remaining before a waiting thread
|
|
* next executes, in units of system ticks. If the thread is not
|
|
* waiting, it returns zero.
|
|
*/
|
|
__syscall k_ticks_t k_thread_timeout_remaining_ticks(const struct k_thread *t);
|
|
|
|
static inline k_ticks_t z_impl_k_thread_timeout_remaining_ticks(
|
|
const struct k_thread *t)
|
|
{
|
|
return z_timeout_remaining(&t->base.timeout);
|
|
}
|
|
|
|
#endif /* CONFIG_SYS_CLOCK_EXISTS */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct _static_thread_data {
|
|
struct k_thread *init_thread;
|
|
k_thread_stack_t *init_stack;
|
|
unsigned int init_stack_size;
|
|
k_thread_entry_t init_entry;
|
|
void *init_p1;
|
|
void *init_p2;
|
|
void *init_p3;
|
|
int init_prio;
|
|
uint32_t init_options;
|
|
const char *init_name;
|
|
#ifdef CONFIG_TIMER_READS_ITS_FREQUENCY_AT_RUNTIME
|
|
int32_t init_delay_ms;
|
|
#else
|
|
k_timeout_t init_delay;
|
|
#endif
|
|
};
|
|
|
|
#ifdef CONFIG_TIMER_READS_ITS_FREQUENCY_AT_RUNTIME
|
|
#define Z_THREAD_INIT_DELAY_INITIALIZER(ms) .init_delay_ms = (ms)
|
|
#define Z_THREAD_INIT_DELAY(thread) SYS_TIMEOUT_MS((thread)->init_delay_ms)
|
|
#else
|
|
#define Z_THREAD_INIT_DELAY_INITIALIZER(ms) .init_delay = SYS_TIMEOUT_MS(ms)
|
|
#define Z_THREAD_INIT_DELAY(thread) (thread)->init_delay
|
|
#endif
|
|
|
|
#define Z_THREAD_INITIALIZER(thread, stack, stack_size, \
|
|
entry, p1, p2, p3, \
|
|
prio, options, delay, tname) \
|
|
{ \
|
|
.init_thread = (thread), \
|
|
.init_stack = (stack), \
|
|
.init_stack_size = (stack_size), \
|
|
.init_entry = (k_thread_entry_t)entry, \
|
|
.init_p1 = (void *)p1, \
|
|
.init_p2 = (void *)p2, \
|
|
.init_p3 = (void *)p3, \
|
|
.init_prio = (prio), \
|
|
.init_options = (options), \
|
|
.init_name = STRINGIFY(tname), \
|
|
Z_THREAD_INIT_DELAY_INITIALIZER(delay) \
|
|
}
|
|
|
|
/*
|
|
* Refer to K_THREAD_DEFINE() and K_KERNEL_THREAD_DEFINE() for
|
|
* information on arguments.
|
|
*/
|
|
#define Z_THREAD_COMMON_DEFINE(name, stack_size, \
|
|
entry, p1, p2, p3, \
|
|
prio, options, delay) \
|
|
struct k_thread _k_thread_obj_##name; \
|
|
STRUCT_SECTION_ITERABLE(_static_thread_data, \
|
|
_k_thread_data_##name) = \
|
|
Z_THREAD_INITIALIZER(&_k_thread_obj_##name, \
|
|
_k_thread_stack_##name, stack_size,\
|
|
entry, p1, p2, p3, prio, options, \
|
|
delay, name); \
|
|
const k_tid_t name = (k_tid_t)&_k_thread_obj_##name
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @brief Statically define and initialize a thread.
|
|
*
|
|
* The thread may be scheduled for immediate execution or a delayed start.
|
|
*
|
|
* Thread options are architecture-specific, and can include K_ESSENTIAL,
|
|
* K_FP_REGS, and K_SSE_REGS. Multiple options may be specified by separating
|
|
* them using "|" (the logical OR operator).
|
|
*
|
|
* The ID of the thread can be accessed using:
|
|
*
|
|
* @code extern const k_tid_t <name>; @endcode
|
|
*
|
|
* @param name Name of the thread.
|
|
* @param stack_size Stack size in bytes.
|
|
* @param entry Thread entry function.
|
|
* @param p1 1st entry point parameter.
|
|
* @param p2 2nd entry point parameter.
|
|
* @param p3 3rd entry point parameter.
|
|
* @param prio Thread priority.
|
|
* @param options Thread options.
|
|
* @param delay Scheduling delay (in milliseconds), zero for no delay.
|
|
*
|
|
* @note Static threads with zero delay should not normally have
|
|
* MetaIRQ priority levels. This can preempt the system
|
|
* initialization handling (depending on the priority of the main
|
|
* thread) and cause surprising ordering side effects. It will not
|
|
* affect anything in the OS per se, but consider it bad practice.
|
|
* Use a SYS_INIT() callback if you need to run code before entrance
|
|
* to the application main().
|
|
*/
|
|
#define K_THREAD_DEFINE(name, stack_size, \
|
|
entry, p1, p2, p3, \
|
|
prio, options, delay) \
|
|
K_THREAD_STACK_DEFINE(_k_thread_stack_##name, stack_size); \
|
|
Z_THREAD_COMMON_DEFINE(name, stack_size, entry, p1, p2, p3, \
|
|
prio, options, delay)
|
|
|
|
/**
|
|
* @brief Statically define and initialize a thread intended to run only in kernel mode.
|
|
*
|
|
* The thread may be scheduled for immediate execution or a delayed start.
|
|
*
|
|
* Thread options are architecture-specific, and can include K_ESSENTIAL,
|
|
* K_FP_REGS, and K_SSE_REGS. Multiple options may be specified by separating
|
|
* them using "|" (the logical OR operator).
|
|
*
|
|
* The ID of the thread can be accessed using:
|
|
*
|
|
* @code extern const k_tid_t <name>; @endcode
|
|
*
|
|
* @note Threads defined by this can only run in kernel mode, and cannot be
|
|
* transformed into user thread via k_thread_user_mode_enter().
|
|
*
|
|
* @warning Depending on the architecture, the stack size (@p stack_size)
|
|
* may need to be multiples of CONFIG_MMU_PAGE_SIZE (if MMU)
|
|
* or in power-of-two size (if MPU).
|
|
*
|
|
* @param name Name of the thread.
|
|
* @param stack_size Stack size in bytes.
|
|
* @param entry Thread entry function.
|
|
* @param p1 1st entry point parameter.
|
|
* @param p2 2nd entry point parameter.
|
|
* @param p3 3rd entry point parameter.
|
|
* @param prio Thread priority.
|
|
* @param options Thread options.
|
|
* @param delay Scheduling delay (in milliseconds), zero for no delay.
|
|
*/
|
|
#define K_KERNEL_THREAD_DEFINE(name, stack_size, \
|
|
entry, p1, p2, p3, \
|
|
prio, options, delay) \
|
|
K_KERNEL_STACK_DEFINE(_k_thread_stack_##name, stack_size); \
|
|
Z_THREAD_COMMON_DEFINE(name, stack_size, entry, p1, p2, p3, \
|
|
prio, options, delay)
|
|
|
|
/**
|
|
* @brief Get a thread's priority.
|
|
*
|
|
* This routine gets the priority of @a thread.
|
|
*
|
|
* @param thread ID of thread whose priority is needed.
|
|
*
|
|
* @return Priority of @a thread.
|
|
*/
|
|
__syscall int k_thread_priority_get(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Set a thread's priority.
|
|
*
|
|
* This routine immediately changes the priority of @a thread.
|
|
*
|
|
* Rescheduling can occur immediately depending on the priority @a thread is
|
|
* set to:
|
|
*
|
|
* - If its priority is raised above the priority of the caller of this
|
|
* function, and the caller is preemptible, @a thread will be scheduled in.
|
|
*
|
|
* - If the caller operates on itself, it lowers its priority below that of
|
|
* other threads in the system, and the caller is preemptible, the thread of
|
|
* highest priority will be scheduled in.
|
|
*
|
|
* Priority can be assigned in the range of -CONFIG_NUM_COOP_PRIORITIES to
|
|
* CONFIG_NUM_PREEMPT_PRIORITIES-1, where -CONFIG_NUM_COOP_PRIORITIES is the
|
|
* highest priority.
|
|
*
|
|
* @param thread ID of thread whose priority is to be set.
|
|
* @param prio New priority.
|
|
*
|
|
* @warning Changing the priority of a thread currently involved in mutex
|
|
* priority inheritance may result in undefined behavior.
|
|
*/
|
|
__syscall void k_thread_priority_set(k_tid_t thread, int prio);
|
|
|
|
|
|
#ifdef CONFIG_SCHED_DEADLINE
|
|
/**
|
|
* @brief Set deadline expiration time for scheduler
|
|
*
|
|
* This sets the "deadline" expiration as a time delta from the
|
|
* current time, in the same units used by k_cycle_get_32(). The
|
|
* scheduler (when deadline scheduling is enabled) will choose the
|
|
* next expiring thread when selecting between threads at the same
|
|
* static priority. Threads at different priorities will be scheduled
|
|
* according to their static priority.
|
|
*
|
|
* @note Deadlines are stored internally using 32 bit unsigned
|
|
* integers. The number of cycles between the "first" deadline in the
|
|
* scheduler queue and the "last" deadline must be less than 2^31 (i.e
|
|
* a signed non-negative quantity). Failure to adhere to this rule
|
|
* may result in scheduled threads running in an incorrect deadline
|
|
* order.
|
|
*
|
|
* @note Despite the API naming, the scheduler makes no guarantees the
|
|
* the thread WILL be scheduled within that deadline, nor does it take
|
|
* extra metadata (like e.g. the "runtime" and "period" parameters in
|
|
* Linux sched_setattr()) that allows the kernel to validate the
|
|
* scheduling for achievability. Such features could be implemented
|
|
* above this call, which is simply input to the priority selection
|
|
* logic.
|
|
*
|
|
* @note You should enable @kconfig{CONFIG_SCHED_DEADLINE} in your project
|
|
* configuration.
|
|
*
|
|
* @param thread A thread on which to set the deadline
|
|
* @param deadline A time delta, in cycle units
|
|
*
|
|
*/
|
|
__syscall void k_thread_deadline_set(k_tid_t thread, int deadline);
|
|
#endif
|
|
|
|
#ifdef CONFIG_SCHED_CPU_MASK
|
|
/**
|
|
* @brief Sets all CPU enable masks to zero
|
|
*
|
|
* After this returns, the thread will no longer be schedulable on any
|
|
* CPUs. The thread must not be currently runnable.
|
|
*
|
|
* @note You should enable @kconfig{CONFIG_SCHED_CPU_MASK} in your project
|
|
* configuration.
|
|
*
|
|
* @param thread Thread to operate upon
|
|
* @return Zero on success, otherwise error code
|
|
*/
|
|
int k_thread_cpu_mask_clear(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Sets all CPU enable masks to one
|
|
*
|
|
* After this returns, the thread will be schedulable on any CPU. The
|
|
* thread must not be currently runnable.
|
|
*
|
|
* @note You should enable @kconfig{CONFIG_SCHED_CPU_MASK} in your project
|
|
* configuration.
|
|
*
|
|
* @param thread Thread to operate upon
|
|
* @return Zero on success, otherwise error code
|
|
*/
|
|
int k_thread_cpu_mask_enable_all(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Enable thread to run on specified CPU
|
|
*
|
|
* The thread must not be currently runnable.
|
|
*
|
|
* @note You should enable @kconfig{CONFIG_SCHED_CPU_MASK} in your project
|
|
* configuration.
|
|
*
|
|
* @param thread Thread to operate upon
|
|
* @param cpu CPU index
|
|
* @return Zero on success, otherwise error code
|
|
*/
|
|
int k_thread_cpu_mask_enable(k_tid_t thread, int cpu);
|
|
|
|
/**
|
|
* @brief Prevent thread to run on specified CPU
|
|
*
|
|
* The thread must not be currently runnable.
|
|
*
|
|
* @note You should enable @kconfig{CONFIG_SCHED_CPU_MASK} in your project
|
|
* configuration.
|
|
*
|
|
* @param thread Thread to operate upon
|
|
* @param cpu CPU index
|
|
* @return Zero on success, otherwise error code
|
|
*/
|
|
int k_thread_cpu_mask_disable(k_tid_t thread, int cpu);
|
|
|
|
/**
|
|
* @brief Pin a thread to a CPU
|
|
*
|
|
* Pin a thread to a CPU by first clearing the cpu mask and then enabling the
|
|
* thread on the selected CPU.
|
|
*
|
|
* @param thread Thread to operate upon
|
|
* @param cpu CPU index
|
|
* @return Zero on success, otherwise error code
|
|
*/
|
|
int k_thread_cpu_pin(k_tid_t thread, int cpu);
|
|
#endif
|
|
|
|
/**
|
|
* @brief Suspend a thread.
|
|
*
|
|
* This routine prevents the kernel scheduler from making @a thread
|
|
* the current thread. All other internal operations on @a thread are
|
|
* still performed; for example, kernel objects it is waiting on are
|
|
* still handed to it. Note that any existing timeouts
|
|
* (e.g. k_sleep(), or a timeout argument to k_sem_take() et. al.)
|
|
* will be canceled. On resume, the thread will begin running
|
|
* immediately and return from the blocked call.
|
|
*
|
|
* If @a thread is already suspended, the routine has no effect.
|
|
*
|
|
* @param thread ID of thread to suspend.
|
|
*/
|
|
__syscall void k_thread_suspend(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Resume a suspended thread.
|
|
*
|
|
* This routine allows the kernel scheduler to make @a thread the current
|
|
* thread, when it is next eligible for that role.
|
|
*
|
|
* If @a thread is not currently suspended, the routine has no effect.
|
|
*
|
|
* @param thread ID of thread to resume.
|
|
*/
|
|
__syscall void k_thread_resume(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Set time-slicing period and scope.
|
|
*
|
|
* This routine specifies how the scheduler will perform time slicing of
|
|
* preemptible threads.
|
|
*
|
|
* To enable time slicing, @a slice must be non-zero. The scheduler
|
|
* ensures that no thread runs for more than the specified time limit
|
|
* before other threads of that priority are given a chance to execute.
|
|
* Any thread whose priority is higher than @a prio is exempted, and may
|
|
* execute as long as desired without being preempted due to time slicing.
|
|
*
|
|
* Time slicing only limits the maximum amount of time a thread may continuously
|
|
* execute. Once the scheduler selects a thread for execution, there is no
|
|
* minimum guaranteed time the thread will execute before threads of greater or
|
|
* equal priority are scheduled.
|
|
*
|
|
* When the current thread is the only one of that priority eligible
|
|
* for execution, this routine has no effect; the thread is immediately
|
|
* rescheduled after the slice period expires.
|
|
*
|
|
* To disable timeslicing, set both @a slice and @a prio to zero.
|
|
*
|
|
* @param slice Maximum time slice length (in milliseconds).
|
|
* @param prio Highest thread priority level eligible for time slicing.
|
|
*/
|
|
extern void k_sched_time_slice_set(int32_t slice, int prio);
|
|
|
|
/**
|
|
* @brief Set thread time slice
|
|
*
|
|
* As for k_sched_time_slice_set, but (when
|
|
* CONFIG_TIMESLICE_PER_THREAD=y) sets the timeslice for a specific
|
|
* thread. When non-zero, this timeslice will take precedence over
|
|
* the global value.
|
|
*
|
|
* When such a thread's timeslice expires, the configured callback
|
|
* will be called before the thread is removed/re-added to the run
|
|
* queue. This callback will occur in interrupt context, and the
|
|
* specified thread is guaranteed to have been preempted by the
|
|
* currently-executing ISR. Such a callback is free to, for example,
|
|
* modify the thread priority or slice time for future execution,
|
|
* suspend the thread, etc...
|
|
*
|
|
* @note Unlike the older API, the time slice parameter here is
|
|
* specified in ticks, not milliseconds. Ticks have always been the
|
|
* internal unit, and not all platforms have integer conversions
|
|
* between the two.
|
|
*
|
|
* @note Threads with a non-zero slice time set will be timesliced
|
|
* always, even if they are higher priority than the maximum timeslice
|
|
* priority set via k_sched_time_slice_set().
|
|
*
|
|
* @note The callback notification for slice expiration happens, as it
|
|
* must, while the thread is still "current", and thus it happens
|
|
* before any registered timeouts at this tick. This has the somewhat
|
|
* confusing side effect that the tick time (c.f. k_uptime_get()) does
|
|
* not yet reflect the expired ticks. Applications wishing to make
|
|
* fine-grained timing decisions within this callback should use the
|
|
* cycle API, or derived facilities like k_thread_runtime_stats_get().
|
|
*
|
|
* @param th A valid, initialized thread
|
|
* @param slice_ticks Maximum timeslice, in ticks
|
|
* @param expired Callback function called on slice expiration
|
|
* @param data Parameter for the expiration handler
|
|
*/
|
|
void k_thread_time_slice_set(struct k_thread *th, int32_t slice_ticks,
|
|
k_thread_timeslice_fn_t expired, void *data);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @addtogroup isr_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Determine if code is running at interrupt level.
|
|
*
|
|
* This routine allows the caller to customize its actions, depending on
|
|
* whether it is a thread or an ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @return false if invoked by a thread.
|
|
* @return true if invoked by an ISR.
|
|
*/
|
|
extern bool k_is_in_isr(void);
|
|
|
|
/**
|
|
* @brief Determine if code is running in a preemptible thread.
|
|
*
|
|
* This routine allows the caller to customize its actions, depending on
|
|
* whether it can be preempted by another thread. The routine returns a 'true'
|
|
* value if all of the following conditions are met:
|
|
*
|
|
* - The code is running in a thread, not at ISR.
|
|
* - The thread's priority is in the preemptible range.
|
|
* - The thread has not locked the scheduler.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @return 0 if invoked by an ISR or by a cooperative thread.
|
|
* @return Non-zero if invoked by a preemptible thread.
|
|
*/
|
|
__syscall int k_is_preempt_thread(void);
|
|
|
|
/**
|
|
* @brief Test whether startup is in the before-main-task phase.
|
|
*
|
|
* This routine allows the caller to customize its actions, depending on
|
|
* whether it being invoked before the kernel is fully active.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @return true if invoked before post-kernel initialization
|
|
* @return false if invoked during/after post-kernel initialization
|
|
*/
|
|
static inline bool k_is_pre_kernel(void)
|
|
{
|
|
extern bool z_sys_post_kernel; /* in init.c */
|
|
|
|
return !z_sys_post_kernel;
|
|
}
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @addtogroup thread_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Lock the scheduler.
|
|
*
|
|
* This routine prevents the current thread from being preempted by another
|
|
* thread by instructing the scheduler to treat it as a cooperative thread.
|
|
* If the thread subsequently performs an operation that makes it unready,
|
|
* it will be context switched out in the normal manner. When the thread
|
|
* again becomes the current thread, its non-preemptible status is maintained.
|
|
*
|
|
* This routine can be called recursively.
|
|
*
|
|
* Owing to clever implementation details, scheduler locks are
|
|
* extremely fast for non-userspace threads (just one byte
|
|
* inc/decrement in the thread struct).
|
|
*
|
|
* @note This works by elevating the thread priority temporarily to a
|
|
* cooperative priority, allowing cheap synchronization vs. other
|
|
* preemptible or cooperative threads running on the current CPU. It
|
|
* does not prevent preemption or asynchrony of other types. It does
|
|
* not prevent threads from running on other CPUs when CONFIG_SMP=y.
|
|
* It does not prevent interrupts from happening, nor does it prevent
|
|
* threads with MetaIRQ priorities from preempting the current thread.
|
|
* In general this is a historical API not well-suited to modern
|
|
* applications, use with care.
|
|
*/
|
|
extern void k_sched_lock(void);
|
|
|
|
/**
|
|
* @brief Unlock the scheduler.
|
|
*
|
|
* This routine reverses the effect of a previous call to k_sched_lock().
|
|
* A thread must call the routine once for each time it called k_sched_lock()
|
|
* before the thread becomes preemptible.
|
|
*/
|
|
extern void k_sched_unlock(void);
|
|
|
|
/**
|
|
* @brief Set current thread's custom data.
|
|
*
|
|
* This routine sets the custom data for the current thread to @ value.
|
|
*
|
|
* Custom data is not used by the kernel itself, and is freely available
|
|
* for a thread to use as it sees fit. It can be used as a framework
|
|
* upon which to build thread-local storage.
|
|
*
|
|
* @param value New custom data value.
|
|
*
|
|
*/
|
|
__syscall void k_thread_custom_data_set(void *value);
|
|
|
|
/**
|
|
* @brief Get current thread's custom data.
|
|
*
|
|
* This routine returns the custom data for the current thread.
|
|
*
|
|
* @return Current custom data value.
|
|
*/
|
|
__syscall void *k_thread_custom_data_get(void);
|
|
|
|
/**
|
|
* @brief Set current thread name
|
|
*
|
|
* Set the name of the thread to be used when @kconfig{CONFIG_THREAD_MONITOR}
|
|
* is enabled for tracing and debugging.
|
|
*
|
|
* @param thread Thread to set name, or NULL to set the current thread
|
|
* @param str Name string
|
|
* @retval 0 on success
|
|
* @retval -EFAULT Memory access error with supplied string
|
|
* @retval -ENOSYS Thread name configuration option not enabled
|
|
* @retval -EINVAL Thread name too long
|
|
*/
|
|
__syscall int k_thread_name_set(k_tid_t thread, const char *str);
|
|
|
|
/**
|
|
* @brief Get thread name
|
|
*
|
|
* Get the name of a thread
|
|
*
|
|
* @param thread Thread ID
|
|
* @retval Thread name, or NULL if configuration not enabled
|
|
*/
|
|
const char *k_thread_name_get(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Copy the thread name into a supplied buffer
|
|
*
|
|
* @param thread Thread to obtain name information
|
|
* @param buf Destination buffer
|
|
* @param size Destination buffer size
|
|
* @retval -ENOSPC Destination buffer too small
|
|
* @retval -EFAULT Memory access error
|
|
* @retval -ENOSYS Thread name feature not enabled
|
|
* @retval 0 Success
|
|
*/
|
|
__syscall int k_thread_name_copy(k_tid_t thread, char *buf,
|
|
size_t size);
|
|
|
|
/**
|
|
* @brief Get thread state string
|
|
*
|
|
* This routine generates a human friendly string containing the thread's
|
|
* state, and copies as much of it as possible into @a buf.
|
|
*
|
|
* @param thread_id Thread ID
|
|
* @param buf Buffer into which to copy state strings
|
|
* @param buf_size Size of the buffer
|
|
*
|
|
* @retval Pointer to @a buf if data was copied, else a pointer to "".
|
|
*/
|
|
const char *k_thread_state_str(k_tid_t thread_id, char *buf, size_t buf_size);
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @addtogroup clock_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Generate null timeout delay.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* not to wait if the requested operation cannot be performed immediately.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_NO_WAIT Z_TIMEOUT_NO_WAIT
|
|
|
|
/**
|
|
* @brief Generate timeout delay from nanoseconds.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API to
|
|
* wait up to @a t nanoseconds to perform the requested operation.
|
|
* Note that timer precision is limited to the tick rate, not the
|
|
* requested value.
|
|
*
|
|
* @param t Duration in nanoseconds.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_NSEC(t) Z_TIMEOUT_NS(t)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from microseconds.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a t microseconds to perform the requested operation.
|
|
* Note that timer precision is limited to the tick rate, not the
|
|
* requested value.
|
|
*
|
|
* @param t Duration in microseconds.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_USEC(t) Z_TIMEOUT_US(t)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from cycles.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a t cycles to perform the requested operation.
|
|
*
|
|
* @param t Duration in cycles.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_CYC(t) Z_TIMEOUT_CYC(t)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from system ticks.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a t ticks to perform the requested operation.
|
|
*
|
|
* @param t Duration in system ticks.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_TICKS(t) Z_TIMEOUT_TICKS(t)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from milliseconds.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a ms milliseconds to perform the requested operation.
|
|
*
|
|
* @param ms Duration in milliseconds.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_MSEC(ms) Z_TIMEOUT_MS(ms)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from seconds.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a s seconds to perform the requested operation.
|
|
*
|
|
* @param s Duration in seconds.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_SECONDS(s) K_MSEC((s) * MSEC_PER_SEC)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from minutes.
|
|
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a m minutes to perform the requested operation.
|
|
*
|
|
* @param m Duration in minutes.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_MINUTES(m) K_SECONDS((m) * 60)
|
|
|
|
/**
|
|
* @brief Generate timeout delay from hours.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait up to @a h hours to perform the requested operation.
|
|
*
|
|
* @param h Duration in hours.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_HOURS(h) K_MINUTES((h) * 60)
|
|
|
|
/**
|
|
* @brief Generate infinite timeout delay.
|
|
*
|
|
* This macro generates a timeout delay that instructs a kernel API
|
|
* to wait as long as necessary to perform the requested operation.
|
|
*
|
|
* @return Timeout delay value.
|
|
*/
|
|
#define K_FOREVER Z_FOREVER
|
|
|
|
#ifdef CONFIG_TIMEOUT_64BIT
|
|
|
|
/**
|
|
* @brief Generates an absolute/uptime timeout value from system ticks
|
|
*
|
|
* This macro generates a timeout delay that represents an expiration
|
|
* at the absolute uptime value specified, in system ticks. That is, the
|
|
* timeout will expire immediately after the system uptime reaches the
|
|
* specified tick count.
|
|
*
|
|
* @param t Tick uptime value
|
|
* @return Timeout delay value
|
|
*/
|
|
#define K_TIMEOUT_ABS_TICKS(t) \
|
|
Z_TIMEOUT_TICKS(Z_TICK_ABS((k_ticks_t)MAX(t, 0)))
|
|
|
|
/**
|
|
* @brief Generates an absolute/uptime timeout value from milliseconds
|
|
*
|
|
* This macro generates a timeout delay that represents an expiration
|
|
* at the absolute uptime value specified, in milliseconds. That is,
|
|
* the timeout will expire immediately after the system uptime reaches
|
|
* the specified tick count.
|
|
*
|
|
* @param t Millisecond uptime value
|
|
* @return Timeout delay value
|
|
*/
|
|
#define K_TIMEOUT_ABS_MS(t) K_TIMEOUT_ABS_TICKS(k_ms_to_ticks_ceil64(t))
|
|
|
|
/**
|
|
* @brief Generates an absolute/uptime timeout value from microseconds
|
|
*
|
|
* This macro generates a timeout delay that represents an expiration
|
|
* at the absolute uptime value specified, in microseconds. That is,
|
|
* the timeout will expire immediately after the system uptime reaches
|
|
* the specified time. Note that timer precision is limited by the
|
|
* system tick rate and not the requested timeout value.
|
|
*
|
|
* @param t Microsecond uptime value
|
|
* @return Timeout delay value
|
|
*/
|
|
#define K_TIMEOUT_ABS_US(t) K_TIMEOUT_ABS_TICKS(k_us_to_ticks_ceil64(t))
|
|
|
|
/**
|
|
* @brief Generates an absolute/uptime timeout value from nanoseconds
|
|
*
|
|
* This macro generates a timeout delay that represents an expiration
|
|
* at the absolute uptime value specified, in nanoseconds. That is,
|
|
* the timeout will expire immediately after the system uptime reaches
|
|
* the specified time. Note that timer precision is limited by the
|
|
* system tick rate and not the requested timeout value.
|
|
*
|
|
* @param t Nanosecond uptime value
|
|
* @return Timeout delay value
|
|
*/
|
|
#define K_TIMEOUT_ABS_NS(t) K_TIMEOUT_ABS_TICKS(k_ns_to_ticks_ceil64(t))
|
|
|
|
/**
|
|
* @brief Generates an absolute/uptime timeout value from system cycles
|
|
*
|
|
* This macro generates a timeout delay that represents an expiration
|
|
* at the absolute uptime value specified, in cycles. That is, the
|
|
* timeout will expire immediately after the system uptime reaches the
|
|
* specified time. Note that timer precision is limited by the system
|
|
* tick rate and not the requested timeout value.
|
|
*
|
|
* @param t Cycle uptime value
|
|
* @return Timeout delay value
|
|
*/
|
|
#define K_TIMEOUT_ABS_CYC(t) K_TIMEOUT_ABS_TICKS(k_cyc_to_ticks_ceil64(t))
|
|
|
|
#endif
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_timer {
|
|
/*
|
|
* _timeout structure must be first here if we want to use
|
|
* dynamic timer allocation. timeout.node is used in the double-linked
|
|
* list of free timers
|
|
*/
|
|
struct _timeout timeout;
|
|
|
|
/* wait queue for the (single) thread waiting on this timer */
|
|
_wait_q_t wait_q;
|
|
|
|
/* runs in ISR context */
|
|
void (*expiry_fn)(struct k_timer *timer);
|
|
|
|
/* runs in the context of the thread that calls k_timer_stop() */
|
|
void (*stop_fn)(struct k_timer *timer);
|
|
|
|
/* timer period */
|
|
k_timeout_t period;
|
|
|
|
/* timer status */
|
|
uint32_t status;
|
|
|
|
/* user-specific data, also used to support legacy features */
|
|
void *user_data;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_timer)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_TIMER
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
#define Z_TIMER_INITIALIZER(obj, expiry, stop) \
|
|
{ \
|
|
.timeout = { \
|
|
.node = {},\
|
|
.fn = z_timer_expiration_handler, \
|
|
.dticks = 0, \
|
|
}, \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.expiry_fn = expiry, \
|
|
.stop_fn = stop, \
|
|
.status = 0, \
|
|
.user_data = 0, \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup timer_apis Timer APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @typedef k_timer_expiry_t
|
|
* @brief Timer expiry function type.
|
|
*
|
|
* A timer's expiry function is executed by the system clock interrupt handler
|
|
* each time the timer expires. The expiry function is optional, and is only
|
|
* invoked if the timer has been initialized with one.
|
|
*
|
|
* @param timer Address of timer.
|
|
*/
|
|
typedef void (*k_timer_expiry_t)(struct k_timer *timer);
|
|
|
|
/**
|
|
* @typedef k_timer_stop_t
|
|
* @brief Timer stop function type.
|
|
*
|
|
* A timer's stop function is executed if the timer is stopped prematurely.
|
|
* The function runs in the context of call that stops the timer. As
|
|
* k_timer_stop() can be invoked from an ISR, the stop function must be
|
|
* callable from interrupt context (isr-ok).
|
|
*
|
|
* The stop function is optional, and is only invoked if the timer has been
|
|
* initialized with one.
|
|
*
|
|
* @param timer Address of timer.
|
|
*/
|
|
typedef void (*k_timer_stop_t)(struct k_timer *timer);
|
|
|
|
/**
|
|
* @brief Statically define and initialize a timer.
|
|
*
|
|
* The timer can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_timer <name>; @endcode
|
|
*
|
|
* @param name Name of the timer variable.
|
|
* @param expiry_fn Function to invoke each time the timer expires.
|
|
* @param stop_fn Function to invoke if the timer is stopped while running.
|
|
*/
|
|
#define K_TIMER_DEFINE(name, expiry_fn, stop_fn) \
|
|
STRUCT_SECTION_ITERABLE(k_timer, name) = \
|
|
Z_TIMER_INITIALIZER(name, expiry_fn, stop_fn)
|
|
|
|
/**
|
|
* @brief Initialize a timer.
|
|
*
|
|
* This routine initializes a timer, prior to its first use.
|
|
*
|
|
* @param timer Address of timer.
|
|
* @param expiry_fn Function to invoke each time the timer expires.
|
|
* @param stop_fn Function to invoke if the timer is stopped while running.
|
|
*/
|
|
extern void k_timer_init(struct k_timer *timer,
|
|
k_timer_expiry_t expiry_fn,
|
|
k_timer_stop_t stop_fn);
|
|
|
|
/**
|
|
* @brief Start a timer.
|
|
*
|
|
* This routine starts a timer, and resets its status to zero. The timer
|
|
* begins counting down using the specified duration and period values.
|
|
*
|
|
* Attempting to start a timer that is already running is permitted.
|
|
* The timer's status is reset to zero and the timer begins counting down
|
|
* using the new duration and period values.
|
|
*
|
|
* @param timer Address of timer.
|
|
* @param duration Initial timer duration.
|
|
* @param period Timer period.
|
|
*/
|
|
__syscall void k_timer_start(struct k_timer *timer,
|
|
k_timeout_t duration, k_timeout_t period);
|
|
|
|
/**
|
|
* @brief Stop a timer.
|
|
*
|
|
* This routine stops a running timer prematurely. The timer's stop function,
|
|
* if one exists, is invoked by the caller.
|
|
*
|
|
* Attempting to stop a timer that is not running is permitted, but has no
|
|
* effect on the timer.
|
|
*
|
|
* @note The stop handler has to be callable from ISRs if @a k_timer_stop is to
|
|
* be called from ISRs.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param timer Address of timer.
|
|
*/
|
|
__syscall void k_timer_stop(struct k_timer *timer);
|
|
|
|
/**
|
|
* @brief Read timer status.
|
|
*
|
|
* This routine reads the timer's status, which indicates the number of times
|
|
* it has expired since its status was last read.
|
|
*
|
|
* Calling this routine resets the timer's status to zero.
|
|
*
|
|
* @param timer Address of timer.
|
|
*
|
|
* @return Timer status.
|
|
*/
|
|
__syscall uint32_t k_timer_status_get(struct k_timer *timer);
|
|
|
|
/**
|
|
* @brief Synchronize thread to timer expiration.
|
|
*
|
|
* This routine blocks the calling thread until the timer's status is non-zero
|
|
* (indicating that it has expired at least once since it was last examined)
|
|
* or the timer is stopped. If the timer status is already non-zero,
|
|
* or the timer is already stopped, the caller continues without waiting.
|
|
*
|
|
* Calling this routine resets the timer's status to zero.
|
|
*
|
|
* This routine must not be used by interrupt handlers, since they are not
|
|
* allowed to block.
|
|
*
|
|
* @param timer Address of timer.
|
|
*
|
|
* @return Timer status.
|
|
*/
|
|
__syscall uint32_t k_timer_status_sync(struct k_timer *timer);
|
|
|
|
#ifdef CONFIG_SYS_CLOCK_EXISTS
|
|
|
|
/**
|
|
* @brief Get next expiration time of a timer, in system ticks
|
|
*
|
|
* This routine returns the future system uptime reached at the next
|
|
* time of expiration of the timer, in units of system ticks. If the
|
|
* timer is not running, current system time is returned.
|
|
*
|
|
* @param timer The timer object
|
|
* @return Uptime of expiration, in ticks
|
|
*/
|
|
__syscall k_ticks_t k_timer_expires_ticks(const struct k_timer *timer);
|
|
|
|
static inline k_ticks_t z_impl_k_timer_expires_ticks(
|
|
const struct k_timer *timer)
|
|
{
|
|
return z_timeout_expires(&timer->timeout);
|
|
}
|
|
|
|
/**
|
|
* @brief Get time remaining before a timer next expires, in system ticks
|
|
*
|
|
* This routine computes the time remaining before a running timer
|
|
* next expires, in units of system ticks. If the timer is not
|
|
* running, it returns zero.
|
|
*/
|
|
__syscall k_ticks_t k_timer_remaining_ticks(const struct k_timer *timer);
|
|
|
|
static inline k_ticks_t z_impl_k_timer_remaining_ticks(
|
|
const struct k_timer *timer)
|
|
{
|
|
return z_timeout_remaining(&timer->timeout);
|
|
}
|
|
|
|
/**
|
|
* @brief Get time remaining before a timer next expires.
|
|
*
|
|
* This routine computes the (approximate) time remaining before a running
|
|
* timer next expires. If the timer is not running, it returns zero.
|
|
*
|
|
* @param timer Address of timer.
|
|
*
|
|
* @return Remaining time (in milliseconds).
|
|
*/
|
|
static inline uint32_t k_timer_remaining_get(struct k_timer *timer)
|
|
{
|
|
return k_ticks_to_ms_floor32(k_timer_remaining_ticks(timer));
|
|
}
|
|
|
|
#endif /* CONFIG_SYS_CLOCK_EXISTS */
|
|
|
|
/**
|
|
* @brief Associate user-specific data with a timer.
|
|
*
|
|
* This routine records the @a user_data with the @a timer, to be retrieved
|
|
* later.
|
|
*
|
|
* It can be used e.g. in a timer handler shared across multiple subsystems to
|
|
* retrieve data specific to the subsystem this timer is associated with.
|
|
*
|
|
* @param timer Address of timer.
|
|
* @param user_data User data to associate with the timer.
|
|
*/
|
|
__syscall void k_timer_user_data_set(struct k_timer *timer, void *user_data);
|
|
|
|
/**
|
|
* @internal
|
|
*/
|
|
static inline void z_impl_k_timer_user_data_set(struct k_timer *timer,
|
|
void *user_data)
|
|
{
|
|
timer->user_data = user_data;
|
|
}
|
|
|
|
/**
|
|
* @brief Retrieve the user-specific data from a timer.
|
|
*
|
|
* @param timer Address of timer.
|
|
*
|
|
* @return The user data.
|
|
*/
|
|
__syscall void *k_timer_user_data_get(const struct k_timer *timer);
|
|
|
|
static inline void *z_impl_k_timer_user_data_get(const struct k_timer *timer)
|
|
{
|
|
return timer->user_data;
|
|
}
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @addtogroup clock_apis
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Get system uptime, in system ticks.
|
|
*
|
|
* This routine returns the elapsed time since the system booted, in
|
|
* ticks (c.f. @kconfig{CONFIG_SYS_CLOCK_TICKS_PER_SEC}), which is the
|
|
* fundamental unit of resolution of kernel timekeeping.
|
|
*
|
|
* @return Current uptime in ticks.
|
|
*/
|
|
__syscall int64_t k_uptime_ticks(void);
|
|
|
|
/**
|
|
* @brief Get system uptime.
|
|
*
|
|
* This routine returns the elapsed time since the system booted,
|
|
* in milliseconds.
|
|
*
|
|
* @note
|
|
* While this function returns time in milliseconds, it does
|
|
* not mean it has millisecond resolution. The actual resolution depends on
|
|
* @kconfig{CONFIG_SYS_CLOCK_TICKS_PER_SEC} config option.
|
|
*
|
|
* @return Current uptime in milliseconds.
|
|
*/
|
|
static inline int64_t k_uptime_get(void)
|
|
{
|
|
return k_ticks_to_ms_floor64(k_uptime_ticks());
|
|
}
|
|
|
|
/**
|
|
* @brief Get system uptime (32-bit version).
|
|
*
|
|
* This routine returns the lower 32 bits of the system uptime in
|
|
* milliseconds.
|
|
*
|
|
* Because correct conversion requires full precision of the system
|
|
* clock there is no benefit to using this over k_uptime_get() unless
|
|
* you know the application will never run long enough for the system
|
|
* clock to approach 2^32 ticks. Calls to this function may involve
|
|
* interrupt blocking and 64-bit math.
|
|
*
|
|
* @note
|
|
* While this function returns time in milliseconds, it does
|
|
* not mean it has millisecond resolution. The actual resolution depends on
|
|
* @kconfig{CONFIG_SYS_CLOCK_TICKS_PER_SEC} config option
|
|
*
|
|
* @return The low 32 bits of the current uptime, in milliseconds.
|
|
*/
|
|
static inline uint32_t k_uptime_get_32(void)
|
|
{
|
|
return (uint32_t)k_uptime_get();
|
|
}
|
|
|
|
/**
|
|
* @brief Get elapsed time.
|
|
*
|
|
* This routine computes the elapsed time between the current system uptime
|
|
* and an earlier reference time, in milliseconds.
|
|
*
|
|
* @param reftime Pointer to a reference time, which is updated to the current
|
|
* uptime upon return.
|
|
*
|
|
* @return Elapsed time.
|
|
*/
|
|
static inline int64_t k_uptime_delta(int64_t *reftime)
|
|
{
|
|
int64_t uptime, delta;
|
|
|
|
uptime = k_uptime_get();
|
|
delta = uptime - *reftime;
|
|
*reftime = uptime;
|
|
|
|
return delta;
|
|
}
|
|
|
|
/**
|
|
* @brief Read the hardware clock.
|
|
*
|
|
* This routine returns the current time, as measured by the system's hardware
|
|
* clock.
|
|
*
|
|
* @return Current hardware clock up-counter (in cycles).
|
|
*/
|
|
static inline uint32_t k_cycle_get_32(void)
|
|
{
|
|
return arch_k_cycle_get_32();
|
|
}
|
|
|
|
/**
|
|
* @brief Read the 64-bit hardware clock.
|
|
*
|
|
* This routine returns the current time in 64-bits, as measured by the
|
|
* system's hardware clock, if available.
|
|
*
|
|
* @see CONFIG_TIMER_HAS_64BIT_CYCLE_COUNTER
|
|
*
|
|
* @return Current hardware clock up-counter (in cycles).
|
|
*/
|
|
static inline uint64_t k_cycle_get_64(void)
|
|
{
|
|
if (!IS_ENABLED(CONFIG_TIMER_HAS_64BIT_CYCLE_COUNTER)) {
|
|
__ASSERT(0, "64-bit cycle counter not enabled on this platform. "
|
|
"See CONFIG_TIMER_HAS_64BIT_CYCLE_COUNTER");
|
|
return 0;
|
|
}
|
|
|
|
return arch_k_cycle_get_64();
|
|
}
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
struct k_queue {
|
|
sys_sflist_t data_q;
|
|
struct k_spinlock lock;
|
|
_wait_q_t wait_q;
|
|
|
|
_POLL_EVENT;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_queue)
|
|
};
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
#define Z_QUEUE_INITIALIZER(obj) \
|
|
{ \
|
|
.data_q = SYS_SFLIST_STATIC_INIT(&obj.data_q), \
|
|
.lock = { }, \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
_POLL_EVENT_OBJ_INIT(obj) \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup queue_apis Queue APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a queue.
|
|
*
|
|
* This routine initializes a queue object, prior to its first use.
|
|
*
|
|
* @param queue Address of the queue.
|
|
*/
|
|
__syscall void k_queue_init(struct k_queue *queue);
|
|
|
|
/**
|
|
* @brief Cancel waiting on a queue.
|
|
*
|
|
* This routine causes first thread pending on @a queue, if any, to
|
|
* return from k_queue_get() call with NULL value (as if timeout expired).
|
|
* If the queue is being waited on by k_poll(), it will return with
|
|
* -EINTR and K_POLL_STATE_CANCELLED state (and per above, subsequent
|
|
* k_queue_get() will return NULL).
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
*/
|
|
__syscall void k_queue_cancel_wait(struct k_queue *queue);
|
|
|
|
/**
|
|
* @brief Append an element to the end of a queue.
|
|
*
|
|
* This routine appends a data item to @a queue. A queue data item must be
|
|
* aligned on a word boundary, and the first word of the item is reserved
|
|
* for the kernel's use.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*/
|
|
extern void k_queue_append(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Append an element to a queue.
|
|
*
|
|
* This routine appends a data item to @a queue. There is an implicit memory
|
|
* allocation to create an additional temporary bookkeeping data structure from
|
|
* the calling thread's resource pool, which is automatically freed when the
|
|
* item is removed. The data itself is not copied.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if there isn't sufficient RAM in the caller's resource pool
|
|
*/
|
|
__syscall int32_t k_queue_alloc_append(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Prepend an element to a queue.
|
|
*
|
|
* This routine prepends a data item to @a queue. A queue data item must be
|
|
* aligned on a word boundary, and the first word of the item is reserved
|
|
* for the kernel's use.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*/
|
|
extern void k_queue_prepend(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Prepend an element to a queue.
|
|
*
|
|
* This routine prepends a data item to @a queue. There is an implicit memory
|
|
* allocation to create an additional temporary bookkeeping data structure from
|
|
* the calling thread's resource pool, which is automatically freed when the
|
|
* item is removed. The data itself is not copied.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if there isn't sufficient RAM in the caller's resource pool
|
|
*/
|
|
__syscall int32_t k_queue_alloc_prepend(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Inserts an element to a queue.
|
|
*
|
|
* This routine inserts a data item to @a queue after previous item. A queue
|
|
* data item must be aligned on a word boundary, and the first word of
|
|
* the item is reserved for the kernel's use.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param prev Address of the previous data item.
|
|
* @param data Address of the data item.
|
|
*/
|
|
extern void k_queue_insert(struct k_queue *queue, void *prev, void *data);
|
|
|
|
/**
|
|
* @brief Atomically append a list of elements to a queue.
|
|
*
|
|
* This routine adds a list of data items to @a queue in one operation.
|
|
* The data items must be in a singly-linked list, with the first word
|
|
* in each data item pointing to the next data item; the list must be
|
|
* NULL-terminated.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param head Pointer to first node in singly-linked list.
|
|
* @param tail Pointer to last node in singly-linked list.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -EINVAL on invalid supplied data
|
|
*
|
|
*/
|
|
extern int k_queue_append_list(struct k_queue *queue, void *head, void *tail);
|
|
|
|
/**
|
|
* @brief Atomically add a list of elements to a queue.
|
|
*
|
|
* This routine adds a list of data items to @a queue in one operation.
|
|
* The data items must be in a singly-linked list implemented using a
|
|
* sys_slist_t object. Upon completion, the original list is empty.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param list Pointer to sys_slist_t object.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -EINVAL on invalid data
|
|
*/
|
|
extern int k_queue_merge_slist(struct k_queue *queue, sys_slist_t *list);
|
|
|
|
/**
|
|
* @brief Get an element from a queue.
|
|
*
|
|
* This routine removes first data item from @a queue. The first word of the
|
|
* data item is reserved for the kernel's use.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param timeout Non-negative waiting period to obtain a data item
|
|
* or one of the special values K_NO_WAIT and
|
|
* K_FOREVER.
|
|
*
|
|
* @return Address of the data item if successful; NULL if returned
|
|
* without waiting, or waiting period timed out.
|
|
*/
|
|
__syscall void *k_queue_get(struct k_queue *queue, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Remove an element from a queue.
|
|
*
|
|
* This routine removes data item from @a queue. The first word of the
|
|
* data item is reserved for the kernel's use. Removing elements from k_queue
|
|
* rely on sys_slist_find_and_remove which is not a constant time operation.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @return true if data item was removed
|
|
*/
|
|
bool k_queue_remove(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Append an element to a queue only if it's not present already.
|
|
*
|
|
* This routine appends data item to @a queue. The first word of the data
|
|
* item is reserved for the kernel's use. Appending elements to k_queue
|
|
* relies on sys_slist_is_node_in_list which is not a constant time operation.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @return true if data item was added, false if not
|
|
*/
|
|
bool k_queue_unique_append(struct k_queue *queue, void *data);
|
|
|
|
/**
|
|
* @brief Query a queue to see if it has data available.
|
|
*
|
|
* Note that the data might be already gone by the time this function returns
|
|
* if other threads are also trying to read from the queue.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue Address of the queue.
|
|
*
|
|
* @return Non-zero if the queue is empty.
|
|
* @return 0 if data is available.
|
|
*/
|
|
__syscall int k_queue_is_empty(struct k_queue *queue);
|
|
|
|
static inline int z_impl_k_queue_is_empty(struct k_queue *queue)
|
|
{
|
|
return (int)sys_sflist_is_empty(&queue->data_q);
|
|
}
|
|
|
|
/**
|
|
* @brief Peek element at the head of queue.
|
|
*
|
|
* Return element from the head of queue without removing it.
|
|
*
|
|
* @param queue Address of the queue.
|
|
*
|
|
* @return Head element, or NULL if queue is empty.
|
|
*/
|
|
__syscall void *k_queue_peek_head(struct k_queue *queue);
|
|
|
|
/**
|
|
* @brief Peek element at the tail of queue.
|
|
*
|
|
* Return element from the tail of queue without removing it.
|
|
*
|
|
* @param queue Address of the queue.
|
|
*
|
|
* @return Tail element, or NULL if queue is empty.
|
|
*/
|
|
__syscall void *k_queue_peek_tail(struct k_queue *queue);
|
|
|
|
/**
|
|
* @brief Statically define and initialize a queue.
|
|
*
|
|
* The queue can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_queue <name>; @endcode
|
|
*
|
|
* @param name Name of the queue.
|
|
*/
|
|
#define K_QUEUE_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_queue, name) = \
|
|
Z_QUEUE_INITIALIZER(name)
|
|
|
|
/** @} */
|
|
|
|
#ifdef CONFIG_USERSPACE
|
|
/**
|
|
* @brief futex structure
|
|
*
|
|
* A k_futex is a lightweight mutual exclusion primitive designed
|
|
* to minimize kernel involvement. Uncontended operation relies
|
|
* only on atomic access to shared memory. k_futex are tracked as
|
|
* kernel objects and can live in user memory so that any access
|
|
* bypasses the kernel object permission management mechanism.
|
|
*/
|
|
struct k_futex {
|
|
atomic_t val;
|
|
};
|
|
|
|
/**
|
|
* @brief futex kernel data structure
|
|
*
|
|
* z_futex_data are the helper data structure for k_futex to complete
|
|
* futex contended operation on kernel side, structure z_futex_data
|
|
* of every futex object is invisible in user mode.
|
|
*/
|
|
struct z_futex_data {
|
|
_wait_q_t wait_q;
|
|
struct k_spinlock lock;
|
|
};
|
|
|
|
#define Z_FUTEX_DATA_INITIALIZER(obj) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q) \
|
|
}
|
|
|
|
/**
|
|
* @defgroup futex_apis FUTEX APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Pend the current thread on a futex
|
|
*
|
|
* Tests that the supplied futex contains the expected value, and if so,
|
|
* goes to sleep until some other thread calls k_futex_wake() on it.
|
|
*
|
|
* @param futex Address of the futex.
|
|
* @param expected Expected value of the futex, if it is different the caller
|
|
* will not wait on it.
|
|
* @param timeout Non-negative waiting period on the futex, or
|
|
* one of the special values K_NO_WAIT or K_FOREVER.
|
|
* @retval -EACCES Caller does not have read access to futex address.
|
|
* @retval -EAGAIN If the futex value did not match the expected parameter.
|
|
* @retval -EINVAL Futex parameter address not recognized by the kernel.
|
|
* @retval -ETIMEDOUT Thread woke up due to timeout and not a futex wakeup.
|
|
* @retval 0 if the caller went to sleep and was woken up. The caller
|
|
* should check the futex's value on wakeup to determine if it needs
|
|
* to block again.
|
|
*/
|
|
__syscall int k_futex_wait(struct k_futex *futex, int expected,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Wake one/all threads pending on a futex
|
|
*
|
|
* Wake up the highest priority thread pending on the supplied futex, or
|
|
* wakeup all the threads pending on the supplied futex, and the behavior
|
|
* depends on wake_all.
|
|
*
|
|
* @param futex Futex to wake up pending threads.
|
|
* @param wake_all If true, wake up all pending threads; If false,
|
|
* wakeup the highest priority thread.
|
|
* @retval -EACCES Caller does not have access to the futex address.
|
|
* @retval -EINVAL Futex parameter address not recognized by the kernel.
|
|
* @retval Number of threads that were woken up.
|
|
*/
|
|
__syscall int k_futex_wake(struct k_futex *futex, bool wake_all);
|
|
|
|
/** @} */
|
|
#endif
|
|
|
|
/**
|
|
* @defgroup event_apis Event APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* Event Structure
|
|
* @ingroup event_apis
|
|
*/
|
|
|
|
struct k_event {
|
|
_wait_q_t wait_q;
|
|
uint32_t events;
|
|
struct k_spinlock lock;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_event)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_EVENT
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
|
|
};
|
|
|
|
#define Z_EVENT_INITIALIZER(obj) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.events = 0 \
|
|
}
|
|
|
|
/**
|
|
* @brief Initialize an event object
|
|
*
|
|
* This routine initializes an event object, prior to its first use.
|
|
*
|
|
* @param event Address of the event object.
|
|
*/
|
|
__syscall void k_event_init(struct k_event *event);
|
|
|
|
/**
|
|
* @brief Post one or more events to an event object
|
|
*
|
|
* This routine posts one or more events to an event object. All tasks waiting
|
|
* on the event object @a event whose waiting conditions become met by this
|
|
* posting immediately unpend.
|
|
*
|
|
* Posting differs from setting in that posted events are merged together with
|
|
* the current set of events tracked by the event object.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of events to post to @a event
|
|
*
|
|
* @retval Previous value of the events in @a event
|
|
*/
|
|
__syscall uint32_t k_event_post(struct k_event *event, uint32_t events);
|
|
|
|
/**
|
|
* @brief Set the events in an event object
|
|
*
|
|
* This routine sets the events stored in event object to the specified value.
|
|
* All tasks waiting on the event object @a event whose waiting conditions
|
|
* become met by this immediately unpend.
|
|
*
|
|
* Setting differs from posting in that set events replace the current set of
|
|
* events tracked by the event object.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of events to set in @a event
|
|
*
|
|
* @retval Previous value of the events in @a event
|
|
*/
|
|
__syscall uint32_t k_event_set(struct k_event *event, uint32_t events);
|
|
|
|
/**
|
|
* @brief Set or clear the events in an event object
|
|
*
|
|
* This routine sets the events stored in event object to the specified value.
|
|
* All tasks waiting on the event object @a event whose waiting conditions
|
|
* become met by this immediately unpend. Unlike @ref k_event_set, this routine
|
|
* allows specific event bits to be set and cleared as determined by the mask.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of events to set/clear in @a event
|
|
* @param events_mask Mask to be applied to @a events
|
|
*
|
|
* @retval Previous value of the events in @a events_mask
|
|
*/
|
|
__syscall uint32_t k_event_set_masked(struct k_event *event, uint32_t events,
|
|
uint32_t events_mask);
|
|
|
|
/**
|
|
* @brief Clear the events in an event object
|
|
*
|
|
* This routine clears (resets) the specified events stored in an event object.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of events to clear in @a event
|
|
*
|
|
* @retval Previous value of the events in @a event
|
|
*/
|
|
__syscall uint32_t k_event_clear(struct k_event *event, uint32_t events);
|
|
|
|
/**
|
|
* @brief Wait for any of the specified events
|
|
*
|
|
* This routine waits on event object @a event until any of the specified
|
|
* events have been delivered to the event object, or the maximum wait time
|
|
* @a timeout has expired. A thread may wait on up to 32 distinctly numbered
|
|
* events that are expressed as bits in a single 32-bit word.
|
|
*
|
|
* @note The caller must be careful when resetting if there are multiple threads
|
|
* waiting for the event object @a event.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of desired events on which to wait
|
|
* @param reset If true, clear the set of events tracked by the event object
|
|
* before waiting. If false, do not clear the events.
|
|
* @param timeout Waiting period for the desired set of events or one of the
|
|
* special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval set of matching events upon success
|
|
* @retval 0 if matching events were not received within the specified time
|
|
*/
|
|
__syscall uint32_t k_event_wait(struct k_event *event, uint32_t events,
|
|
bool reset, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Wait for all of the specified events
|
|
*
|
|
* This routine waits on event object @a event until all of the specified
|
|
* events have been delivered to the event object, or the maximum wait time
|
|
* @a timeout has expired. A thread may wait on up to 32 distinctly numbered
|
|
* events that are expressed as bits in a single 32-bit word.
|
|
*
|
|
* @note The caller must be careful when resetting if there are multiple threads
|
|
* waiting for the event object @a event.
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events Set of desired events on which to wait
|
|
* @param reset If true, clear the set of events tracked by the event object
|
|
* before waiting. If false, do not clear the events.
|
|
* @param timeout Waiting period for the desired set of events or one of the
|
|
* special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval set of matching events upon success
|
|
* @retval 0 if matching events were not received within the specified time
|
|
*/
|
|
__syscall uint32_t k_event_wait_all(struct k_event *event, uint32_t events,
|
|
bool reset, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Test the events currently tracked in the event object
|
|
*
|
|
* @param event Address of the event object
|
|
* @param events_mask Set of desired events to test
|
|
*
|
|
* @retval Current value of events in @a events_mask
|
|
*/
|
|
static inline uint32_t k_event_test(struct k_event *event, uint32_t events_mask)
|
|
{
|
|
return k_event_wait(event, events_mask, false, K_NO_WAIT);
|
|
}
|
|
|
|
/**
|
|
* @brief Statically define and initialize an event object
|
|
*
|
|
* The event can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_event <name>; @endcode
|
|
*
|
|
* @param name Name of the event object.
|
|
*/
|
|
#define K_EVENT_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_event, name) = \
|
|
Z_EVENT_INITIALIZER(name);
|
|
|
|
/** @} */
|
|
|
|
struct k_fifo {
|
|
struct k_queue _queue;
|
|
#ifdef CONFIG_OBJ_CORE_FIFO
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
#define Z_FIFO_INITIALIZER(obj) \
|
|
{ \
|
|
._queue = Z_QUEUE_INITIALIZER(obj._queue) \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup fifo_apis FIFO APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a FIFO queue.
|
|
*
|
|
* This routine initializes a FIFO queue, prior to its first use.
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
*/
|
|
#define k_fifo_init(fifo) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, init, fifo); \
|
|
k_queue_init(&(fifo)->_queue); \
|
|
K_OBJ_CORE_INIT(K_OBJ_CORE(fifo), _obj_type_fifo); \
|
|
K_OBJ_CORE_LINK(K_OBJ_CORE(fifo)); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, init, fifo); \
|
|
})
|
|
|
|
/**
|
|
* @brief Cancel waiting on a FIFO queue.
|
|
*
|
|
* This routine causes first thread pending on @a fifo, if any, to
|
|
* return from k_fifo_get() call with NULL value (as if timeout
|
|
* expired).
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
*/
|
|
#define k_fifo_cancel_wait(fifo) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, cancel_wait, fifo); \
|
|
k_queue_cancel_wait(&(fifo)->_queue); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, cancel_wait, fifo); \
|
|
})
|
|
|
|
/**
|
|
* @brief Add an element to a FIFO queue.
|
|
*
|
|
* This routine adds a data item to @a fifo. A FIFO data item must be
|
|
* aligned on a word boundary, and the first word of the item is reserved
|
|
* for the kernel's use.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO.
|
|
* @param data Address of the data item.
|
|
*/
|
|
#define k_fifo_put(fifo, data) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, put, fifo, data); \
|
|
k_queue_append(&(fifo)->_queue, data); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, put, fifo, data); \
|
|
})
|
|
|
|
/**
|
|
* @brief Add an element to a FIFO queue.
|
|
*
|
|
* This routine adds a data item to @a fifo. There is an implicit memory
|
|
* allocation to create an additional temporary bookkeeping data structure from
|
|
* the calling thread's resource pool, which is automatically freed when the
|
|
* item is removed. The data itself is not copied.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if there isn't sufficient RAM in the caller's resource pool
|
|
*/
|
|
#define k_fifo_alloc_put(fifo, data) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, alloc_put, fifo, data); \
|
|
int fap_ret = k_queue_alloc_append(&(fifo)->_queue, data); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, alloc_put, fifo, data, fap_ret); \
|
|
fap_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Atomically add a list of elements to a FIFO.
|
|
*
|
|
* This routine adds a list of data items to @a fifo in one operation.
|
|
* The data items must be in a singly-linked list, with the first word of
|
|
* each data item pointing to the next data item; the list must be
|
|
* NULL-terminated.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
* @param head Pointer to first node in singly-linked list.
|
|
* @param tail Pointer to last node in singly-linked list.
|
|
*/
|
|
#define k_fifo_put_list(fifo, head, tail) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, put_list, fifo, head, tail); \
|
|
k_queue_append_list(&(fifo)->_queue, head, tail); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, put_list, fifo, head, tail); \
|
|
})
|
|
|
|
/**
|
|
* @brief Atomically add a list of elements to a FIFO queue.
|
|
*
|
|
* This routine adds a list of data items to @a fifo in one operation.
|
|
* The data items must be in a singly-linked list implemented using a
|
|
* sys_slist_t object. Upon completion, the sys_slist_t object is invalid
|
|
* and must be re-initialized via sys_slist_init().
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
* @param list Pointer to sys_slist_t object.
|
|
*/
|
|
#define k_fifo_put_slist(fifo, list) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, put_slist, fifo, list); \
|
|
k_queue_merge_slist(&(fifo)->_queue, list); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, put_slist, fifo, list); \
|
|
})
|
|
|
|
/**
|
|
* @brief Get an element from a FIFO queue.
|
|
*
|
|
* This routine removes a data item from @a fifo in a "first in, first out"
|
|
* manner. The first word of the data item is reserved for the kernel's use.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
* @param timeout Waiting period to obtain a data item,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @return Address of the data item if successful; NULL if returned
|
|
* without waiting, or waiting period timed out.
|
|
*/
|
|
#define k_fifo_get(fifo, timeout) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, get, fifo, timeout); \
|
|
void *fg_ret = k_queue_get(&(fifo)->_queue, timeout); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, get, fifo, timeout, fg_ret); \
|
|
fg_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Query a FIFO queue to see if it has data available.
|
|
*
|
|
* Note that the data might be already gone by the time this function returns
|
|
* if other threads is also trying to read from the FIFO.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
*
|
|
* @return Non-zero if the FIFO queue is empty.
|
|
* @return 0 if data is available.
|
|
*/
|
|
#define k_fifo_is_empty(fifo) \
|
|
k_queue_is_empty(&(fifo)->_queue)
|
|
|
|
/**
|
|
* @brief Peek element at the head of a FIFO queue.
|
|
*
|
|
* Return element from the head of FIFO queue without removing it. A usecase
|
|
* for this is if elements of the FIFO object are themselves containers. Then
|
|
* on each iteration of processing, a head container will be peeked,
|
|
* and some data processed out of it, and only if the container is empty,
|
|
* it will be completely remove from the FIFO queue.
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
*
|
|
* @return Head element, or NULL if the FIFO queue is empty.
|
|
*/
|
|
#define k_fifo_peek_head(fifo) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, peek_head, fifo); \
|
|
void *fph_ret = k_queue_peek_head(&(fifo)->_queue); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, peek_head, fifo, fph_ret); \
|
|
fph_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Peek element at the tail of FIFO queue.
|
|
*
|
|
* Return element from the tail of FIFO queue (without removing it). A usecase
|
|
* for this is if elements of the FIFO queue are themselves containers. Then
|
|
* it may be useful to add more data to the last container in a FIFO queue.
|
|
*
|
|
* @param fifo Address of the FIFO queue.
|
|
*
|
|
* @return Tail element, or NULL if a FIFO queue is empty.
|
|
*/
|
|
#define k_fifo_peek_tail(fifo) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_fifo, peek_tail, fifo); \
|
|
void *fpt_ret = k_queue_peek_tail(&(fifo)->_queue); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_fifo, peek_tail, fifo, fpt_ret); \
|
|
fpt_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Statically define and initialize a FIFO queue.
|
|
*
|
|
* The FIFO queue can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_fifo <name>; @endcode
|
|
*
|
|
* @param name Name of the FIFO queue.
|
|
*/
|
|
#define K_FIFO_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_fifo, name) = \
|
|
Z_FIFO_INITIALIZER(name)
|
|
|
|
/** @} */
|
|
|
|
struct k_lifo {
|
|
struct k_queue _queue;
|
|
#ifdef CONFIG_OBJ_CORE_LIFO
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
#define Z_LIFO_INITIALIZER(obj) \
|
|
{ \
|
|
._queue = Z_QUEUE_INITIALIZER(obj._queue) \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup lifo_apis LIFO APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a LIFO queue.
|
|
*
|
|
* This routine initializes a LIFO queue object, prior to its first use.
|
|
*
|
|
* @param lifo Address of the LIFO queue.
|
|
*/
|
|
#define k_lifo_init(lifo) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_lifo, init, lifo); \
|
|
k_queue_init(&(lifo)->_queue); \
|
|
K_OBJ_CORE_INIT(K_OBJ_CORE(lifo), _obj_type_lifo); \
|
|
K_OBJ_CORE_LINK(K_OBJ_CORE(lifo)); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_lifo, init, lifo); \
|
|
})
|
|
|
|
/**
|
|
* @brief Add an element to a LIFO queue.
|
|
*
|
|
* This routine adds a data item to @a lifo. A LIFO queue data item must be
|
|
* aligned on a word boundary, and the first word of the item is
|
|
* reserved for the kernel's use.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param lifo Address of the LIFO queue.
|
|
* @param data Address of the data item.
|
|
*/
|
|
#define k_lifo_put(lifo, data) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_lifo, put, lifo, data); \
|
|
k_queue_prepend(&(lifo)->_queue, data); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_lifo, put, lifo, data); \
|
|
})
|
|
|
|
/**
|
|
* @brief Add an element to a LIFO queue.
|
|
*
|
|
* This routine adds a data item to @a lifo. There is an implicit memory
|
|
* allocation to create an additional temporary bookkeeping data structure from
|
|
* the calling thread's resource pool, which is automatically freed when the
|
|
* item is removed. The data itself is not copied.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param lifo Address of the LIFO.
|
|
* @param data Address of the data item.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if there isn't sufficient RAM in the caller's resource pool
|
|
*/
|
|
#define k_lifo_alloc_put(lifo, data) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_lifo, alloc_put, lifo, data); \
|
|
int lap_ret = k_queue_alloc_prepend(&(lifo)->_queue, data); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_lifo, alloc_put, lifo, data, lap_ret); \
|
|
lap_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Get an element from a LIFO queue.
|
|
*
|
|
* This routine removes a data item from @a LIFO in a "last in, first out"
|
|
* manner. The first word of the data item is reserved for the kernel's use.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param lifo Address of the LIFO queue.
|
|
* @param timeout Waiting period to obtain a data item,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @return Address of the data item if successful; NULL if returned
|
|
* without waiting, or waiting period timed out.
|
|
*/
|
|
#define k_lifo_get(lifo, timeout) \
|
|
({ \
|
|
SYS_PORT_TRACING_OBJ_FUNC_ENTER(k_lifo, get, lifo, timeout); \
|
|
void *lg_ret = k_queue_get(&(lifo)->_queue, timeout); \
|
|
SYS_PORT_TRACING_OBJ_FUNC_EXIT(k_lifo, get, lifo, timeout, lg_ret); \
|
|
lg_ret; \
|
|
})
|
|
|
|
/**
|
|
* @brief Statically define and initialize a LIFO queue.
|
|
*
|
|
* The LIFO queue can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_lifo <name>; @endcode
|
|
*
|
|
* @param name Name of the fifo.
|
|
*/
|
|
#define K_LIFO_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_lifo, name) = \
|
|
Z_LIFO_INITIALIZER(name)
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
#define K_STACK_FLAG_ALLOC ((uint8_t)1) /* Buffer was allocated */
|
|
|
|
typedef uintptr_t stack_data_t;
|
|
|
|
struct k_stack {
|
|
_wait_q_t wait_q;
|
|
struct k_spinlock lock;
|
|
stack_data_t *base, *next, *top;
|
|
|
|
uint8_t flags;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_stack)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_STACK
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
#define Z_STACK_INITIALIZER(obj, stack_buffer, stack_num_entries) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.base = stack_buffer, \
|
|
.next = stack_buffer, \
|
|
.top = stack_buffer + stack_num_entries, \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup stack_apis Stack APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a stack.
|
|
*
|
|
* This routine initializes a stack object, prior to its first use.
|
|
*
|
|
* @param stack Address of the stack.
|
|
* @param buffer Address of array used to hold stacked values.
|
|
* @param num_entries Maximum number of values that can be stacked.
|
|
*/
|
|
void k_stack_init(struct k_stack *stack,
|
|
stack_data_t *buffer, uint32_t num_entries);
|
|
|
|
|
|
/**
|
|
* @brief Initialize a stack.
|
|
*
|
|
* This routine initializes a stack object, prior to its first use. Internal
|
|
* buffers will be allocated from the calling thread's resource pool.
|
|
* This memory will be released if k_stack_cleanup() is called, or
|
|
* userspace is enabled and the stack object loses all references to it.
|
|
*
|
|
* @param stack Address of the stack.
|
|
* @param num_entries Maximum number of values that can be stacked.
|
|
*
|
|
* @return -ENOMEM if memory couldn't be allocated
|
|
*/
|
|
|
|
__syscall int32_t k_stack_alloc_init(struct k_stack *stack,
|
|
uint32_t num_entries);
|
|
|
|
/**
|
|
* @brief Release a stack's allocated buffer
|
|
*
|
|
* If a stack object was given a dynamically allocated buffer via
|
|
* k_stack_alloc_init(), this will free it. This function does nothing
|
|
* if the buffer wasn't dynamically allocated.
|
|
*
|
|
* @param stack Address of the stack.
|
|
* @retval 0 on success
|
|
* @retval -EAGAIN when object is still in use
|
|
*/
|
|
int k_stack_cleanup(struct k_stack *stack);
|
|
|
|
/**
|
|
* @brief Push an element onto a stack.
|
|
*
|
|
* This routine adds a stack_data_t value @a data to @a stack.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param stack Address of the stack.
|
|
* @param data Value to push onto the stack.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if stack is full
|
|
*/
|
|
__syscall int k_stack_push(struct k_stack *stack, stack_data_t data);
|
|
|
|
/**
|
|
* @brief Pop an element from a stack.
|
|
*
|
|
* This routine removes a stack_data_t value from @a stack in a "last in,
|
|
* first out" manner and stores the value in @a data.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param stack Address of the stack.
|
|
* @param data Address of area to hold the value popped from the stack.
|
|
* @param timeout Waiting period to obtain a value,
|
|
* or one of the special values K_NO_WAIT and
|
|
* K_FOREVER.
|
|
*
|
|
* @retval 0 Element popped from stack.
|
|
* @retval -EBUSY Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
__syscall int k_stack_pop(struct k_stack *stack, stack_data_t *data,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Statically define and initialize a stack
|
|
*
|
|
* The stack can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_stack <name>; @endcode
|
|
*
|
|
* @param name Name of the stack.
|
|
* @param stack_num_entries Maximum number of values that can be stacked.
|
|
*/
|
|
#define K_STACK_DEFINE(name, stack_num_entries) \
|
|
stack_data_t __noinit \
|
|
_k_stack_buf_##name[stack_num_entries]; \
|
|
STRUCT_SECTION_ITERABLE(k_stack, name) = \
|
|
Z_STACK_INITIALIZER(name, _k_stack_buf_##name, \
|
|
stack_num_entries)
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_work;
|
|
struct k_work_q;
|
|
struct k_work_queue_config;
|
|
extern struct k_work_q k_sys_work_q;
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup mutex_apis Mutex APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* Mutex Structure
|
|
* @ingroup mutex_apis
|
|
*/
|
|
struct k_mutex {
|
|
/** Mutex wait queue */
|
|
_wait_q_t wait_q;
|
|
/** Mutex owner */
|
|
struct k_thread *owner;
|
|
|
|
/** Current lock count */
|
|
uint32_t lock_count;
|
|
|
|
/** Original thread priority */
|
|
int owner_orig_prio;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_mutex)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_MUTEX
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
#define Z_MUTEX_INITIALIZER(obj) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.owner = NULL, \
|
|
.lock_count = 0, \
|
|
.owner_orig_prio = K_LOWEST_APPLICATION_THREAD_PRIO, \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @brief Statically define and initialize a mutex.
|
|
*
|
|
* The mutex can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_mutex <name>; @endcode
|
|
*
|
|
* @param name Name of the mutex.
|
|
*/
|
|
#define K_MUTEX_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_mutex, name) = \
|
|
Z_MUTEX_INITIALIZER(name)
|
|
|
|
/**
|
|
* @brief Initialize a mutex.
|
|
*
|
|
* This routine initializes a mutex object, prior to its first use.
|
|
*
|
|
* Upon completion, the mutex is available and does not have an owner.
|
|
*
|
|
* @param mutex Address of the mutex.
|
|
*
|
|
* @retval 0 Mutex object created
|
|
*
|
|
*/
|
|
__syscall int k_mutex_init(struct k_mutex *mutex);
|
|
|
|
|
|
/**
|
|
* @brief Lock a mutex.
|
|
*
|
|
* This routine locks @a mutex. If the mutex is locked by another thread,
|
|
* the calling thread waits until the mutex becomes available or until
|
|
* a timeout occurs.
|
|
*
|
|
* A thread is permitted to lock a mutex it has already locked. The operation
|
|
* completes immediately and the lock count is increased by 1.
|
|
*
|
|
* Mutexes may not be locked in ISRs.
|
|
*
|
|
* @param mutex Address of the mutex.
|
|
* @param timeout Waiting period to lock the mutex,
|
|
* or one of the special values K_NO_WAIT and
|
|
* K_FOREVER.
|
|
*
|
|
* @retval 0 Mutex locked.
|
|
* @retval -EBUSY Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
__syscall int k_mutex_lock(struct k_mutex *mutex, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Unlock a mutex.
|
|
*
|
|
* This routine unlocks @a mutex. The mutex must already be locked by the
|
|
* calling thread.
|
|
*
|
|
* The mutex cannot be claimed by another thread until it has been unlocked by
|
|
* the calling thread as many times as it was previously locked by that
|
|
* thread.
|
|
*
|
|
* Mutexes may not be unlocked in ISRs, as mutexes must only be manipulated
|
|
* in thread context due to ownership and priority inheritance semantics.
|
|
*
|
|
* @param mutex Address of the mutex.
|
|
*
|
|
* @retval 0 Mutex unlocked.
|
|
* @retval -EPERM The current thread does not own the mutex
|
|
* @retval -EINVAL The mutex is not locked
|
|
*
|
|
*/
|
|
__syscall int k_mutex_unlock(struct k_mutex *mutex);
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
|
|
struct k_condvar {
|
|
_wait_q_t wait_q;
|
|
|
|
#ifdef CONFIG_OBJ_CORE_CONDVAR
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
#define Z_CONDVAR_INITIALIZER(obj) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
}
|
|
|
|
/**
|
|
* @defgroup condvar_apis Condition Variables APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a condition variable
|
|
*
|
|
* @param condvar pointer to a @p k_condvar structure
|
|
* @retval 0 Condition variable created successfully
|
|
*/
|
|
__syscall int k_condvar_init(struct k_condvar *condvar);
|
|
|
|
/**
|
|
* @brief Signals one thread that is pending on the condition variable
|
|
*
|
|
* @param condvar pointer to a @p k_condvar structure
|
|
* @retval 0 On success
|
|
*/
|
|
__syscall int k_condvar_signal(struct k_condvar *condvar);
|
|
|
|
/**
|
|
* @brief Unblock all threads that are pending on the condition
|
|
* variable
|
|
*
|
|
* @param condvar pointer to a @p k_condvar structure
|
|
* @return An integer with number of woken threads on success
|
|
*/
|
|
__syscall int k_condvar_broadcast(struct k_condvar *condvar);
|
|
|
|
/**
|
|
* @brief Waits on the condition variable releasing the mutex lock
|
|
*
|
|
* Atomically releases the currently owned mutex, blocks the current thread
|
|
* waiting on the condition variable specified by @a condvar,
|
|
* and finally acquires the mutex again.
|
|
*
|
|
* The waiting thread unblocks only after another thread calls
|
|
* k_condvar_signal, or k_condvar_broadcast with the same condition variable.
|
|
*
|
|
* @param condvar pointer to a @p k_condvar structure
|
|
* @param mutex Address of the mutex.
|
|
* @param timeout Waiting period for the condition variable
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
* @retval 0 On success
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
__syscall int k_condvar_wait(struct k_condvar *condvar, struct k_mutex *mutex,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Statically define and initialize a condition variable.
|
|
*
|
|
* The condition variable can be accessed outside the module where it is
|
|
* defined using:
|
|
*
|
|
* @code extern struct k_condvar <name>; @endcode
|
|
*
|
|
* @param name Name of the condition variable.
|
|
*/
|
|
#define K_CONDVAR_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_condvar, name) = \
|
|
Z_CONDVAR_INITIALIZER(name)
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_sem {
|
|
_wait_q_t wait_q;
|
|
unsigned int count;
|
|
unsigned int limit;
|
|
|
|
_POLL_EVENT;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_sem)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_SEM
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
#define Z_SEM_INITIALIZER(obj, initial_count, count_limit) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.count = initial_count, \
|
|
.limit = count_limit, \
|
|
_POLL_EVENT_OBJ_INIT(obj) \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup semaphore_apis Semaphore APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Maximum limit value allowed for a semaphore.
|
|
*
|
|
* This is intended for use when a semaphore does not have
|
|
* an explicit maximum limit, and instead is just used for
|
|
* counting purposes.
|
|
*
|
|
*/
|
|
#define K_SEM_MAX_LIMIT UINT_MAX
|
|
|
|
/**
|
|
* @brief Initialize a semaphore.
|
|
*
|
|
* This routine initializes a semaphore object, prior to its first use.
|
|
*
|
|
* @param sem Address of the semaphore.
|
|
* @param initial_count Initial semaphore count.
|
|
* @param limit Maximum permitted semaphore count.
|
|
*
|
|
* @see K_SEM_MAX_LIMIT
|
|
*
|
|
* @retval 0 Semaphore created successfully
|
|
* @retval -EINVAL Invalid values
|
|
*
|
|
*/
|
|
__syscall int k_sem_init(struct k_sem *sem, unsigned int initial_count,
|
|
unsigned int limit);
|
|
|
|
/**
|
|
* @brief Take a semaphore.
|
|
*
|
|
* This routine takes @a sem.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param sem Address of the semaphore.
|
|
* @param timeout Waiting period to take the semaphore,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval 0 Semaphore taken.
|
|
* @retval -EBUSY Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out,
|
|
* or the semaphore was reset during the waiting period.
|
|
*/
|
|
__syscall int k_sem_take(struct k_sem *sem, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Give a semaphore.
|
|
*
|
|
* This routine gives @a sem, unless the semaphore is already at its maximum
|
|
* permitted count.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param sem Address of the semaphore.
|
|
*/
|
|
__syscall void k_sem_give(struct k_sem *sem);
|
|
|
|
/**
|
|
* @brief Resets a semaphore's count to zero.
|
|
*
|
|
* This routine sets the count of @a sem to zero.
|
|
* Any outstanding semaphore takes will be aborted
|
|
* with -EAGAIN.
|
|
*
|
|
* @param sem Address of the semaphore.
|
|
*/
|
|
__syscall void k_sem_reset(struct k_sem *sem);
|
|
|
|
/**
|
|
* @brief Get a semaphore's count.
|
|
*
|
|
* This routine returns the current count of @a sem.
|
|
*
|
|
* @param sem Address of the semaphore.
|
|
*
|
|
* @return Current semaphore count.
|
|
*/
|
|
__syscall unsigned int k_sem_count_get(struct k_sem *sem);
|
|
|
|
/**
|
|
* @internal
|
|
*/
|
|
static inline unsigned int z_impl_k_sem_count_get(struct k_sem *sem)
|
|
{
|
|
return sem->count;
|
|
}
|
|
|
|
/**
|
|
* @brief Statically define and initialize a semaphore.
|
|
*
|
|
* The semaphore can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_sem <name>; @endcode
|
|
*
|
|
* @param name Name of the semaphore.
|
|
* @param initial_count Initial semaphore count.
|
|
* @param count_limit Maximum permitted semaphore count.
|
|
*/
|
|
#define K_SEM_DEFINE(name, initial_count, count_limit) \
|
|
STRUCT_SECTION_ITERABLE(k_sem, name) = \
|
|
Z_SEM_INITIALIZER(name, initial_count, count_limit); \
|
|
BUILD_ASSERT(((count_limit) != 0) && \
|
|
((initial_count) <= (count_limit)) && \
|
|
((count_limit) <= K_SEM_MAX_LIMIT));
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_work_delayable;
|
|
struct k_work_sync;
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup workqueue_apis Work Queue APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/** @brief The signature for a work item handler function.
|
|
*
|
|
* The function will be invoked by the thread animating a work queue.
|
|
*
|
|
* @param work the work item that provided the handler.
|
|
*/
|
|
typedef void (*k_work_handler_t)(struct k_work *work);
|
|
|
|
/** @brief Initialize a (non-delayable) work structure.
|
|
*
|
|
* This must be invoked before submitting a work structure for the first time.
|
|
* It need not be invoked again on the same work structure. It can be
|
|
* re-invoked to change the associated handler, but this must be done when the
|
|
* work item is idle.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work the work structure to be initialized.
|
|
*
|
|
* @param handler the handler to be invoked by the work item.
|
|
*/
|
|
void k_work_init(struct k_work *work,
|
|
k_work_handler_t handler);
|
|
|
|
/** @brief Busy state flags from the work item.
|
|
*
|
|
* A zero return value indicates the work item appears to be idle.
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* is checked. Use locks where appropriate.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @return a mask of flags K_WORK_DELAYED, K_WORK_QUEUED,
|
|
* K_WORK_RUNNING, and K_WORK_CANCELING.
|
|
*/
|
|
int k_work_busy_get(const struct k_work *work);
|
|
|
|
/** @brief Test whether a work item is currently pending.
|
|
*
|
|
* Wrapper to determine whether a work item is in a non-idle dstate.
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* is checked. Use locks where appropriate.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @return true if and only if k_work_busy_get() returns a non-zero value.
|
|
*/
|
|
static inline bool k_work_is_pending(const struct k_work *work);
|
|
|
|
/** @brief Submit a work item to a queue.
|
|
*
|
|
* @param queue pointer to the work queue on which the item should run. If
|
|
* NULL the queue from the most recent submission will be used.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @retval 0 if work was already submitted to a queue
|
|
* @retval 1 if work was not submitted and has been queued to @p queue
|
|
* @retval 2 if work was running and has been queued to the queue that was
|
|
* running it
|
|
* @retval -EBUSY
|
|
* * if work submission was rejected because the work item is cancelling; or
|
|
* * @p queue is draining; or
|
|
* * @p queue is plugged.
|
|
* @retval -EINVAL if @p queue is null and the work item has never been run.
|
|
* @retval -ENODEV if @p queue has not been started.
|
|
*/
|
|
int k_work_submit_to_queue(struct k_work_q *queue,
|
|
struct k_work *work);
|
|
|
|
/** @brief Submit a work item to the system queue.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @return as with k_work_submit_to_queue().
|
|
*/
|
|
extern int k_work_submit(struct k_work *work);
|
|
|
|
/** @brief Wait for last-submitted instance to complete.
|
|
*
|
|
* Resubmissions may occur while waiting, including chained submissions (from
|
|
* within the handler).
|
|
*
|
|
* @note Be careful of caller and work queue thread relative priority. If
|
|
* this function sleeps it will not return until the work queue thread
|
|
* completes the tasks that allow this thread to resume.
|
|
*
|
|
* @note Behavior is undefined if this function is invoked on @p work from a
|
|
* work queue running @p work.
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @param sync pointer to an opaque item containing state related to the
|
|
* pending cancellation. The object must persist until the call returns, and
|
|
* be accessible from both the caller thread and the work queue thread. The
|
|
* object must not be used for any other flush or cancel operation until this
|
|
* one completes. On architectures with CONFIG_KERNEL_COHERENCE the object
|
|
* must be allocated in coherent memory.
|
|
*
|
|
* @retval true if call had to wait for completion
|
|
* @retval false if work was already idle
|
|
*/
|
|
bool k_work_flush(struct k_work *work,
|
|
struct k_work_sync *sync);
|
|
|
|
/** @brief Cancel a work item.
|
|
*
|
|
* This attempts to prevent a pending (non-delayable) work item from being
|
|
* processed by removing it from the work queue. If the item is being
|
|
* processed, the work item will continue to be processed, but resubmissions
|
|
* are rejected until cancellation completes.
|
|
*
|
|
* If this returns zero cancellation is complete, otherwise something
|
|
* (probably a work queue thread) is still referencing the item.
|
|
*
|
|
* See also k_work_cancel_sync().
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @return the k_work_busy_get() status indicating the state of the item after all
|
|
* cancellation steps performed by this call are completed.
|
|
*/
|
|
int k_work_cancel(struct k_work *work);
|
|
|
|
/** @brief Cancel a work item and wait for it to complete.
|
|
*
|
|
* Same as k_work_cancel() but does not return until cancellation is complete.
|
|
* This can be invoked by a thread after k_work_cancel() to synchronize with a
|
|
* previous cancellation.
|
|
*
|
|
* On return the work structure will be idle unless something submits it after
|
|
* the cancellation was complete.
|
|
*
|
|
* @note Be careful of caller and work queue thread relative priority. If
|
|
* this function sleeps it will not return until the work queue thread
|
|
* completes the tasks that allow this thread to resume.
|
|
*
|
|
* @note Behavior is undefined if this function is invoked on @p work from a
|
|
* work queue running @p work.
|
|
*
|
|
* @param work pointer to the work item.
|
|
*
|
|
* @param sync pointer to an opaque item containing state related to the
|
|
* pending cancellation. The object must persist until the call returns, and
|
|
* be accessible from both the caller thread and the work queue thread. The
|
|
* object must not be used for any other flush or cancel operation until this
|
|
* one completes. On architectures with CONFIG_KERNEL_COHERENCE the object
|
|
* must be allocated in coherent memory.
|
|
*
|
|
* @retval true if work was pending (call had to wait for cancellation of a
|
|
* running handler to complete, or scheduled or submitted operations were
|
|
* cancelled);
|
|
* @retval false otherwise
|
|
*/
|
|
bool k_work_cancel_sync(struct k_work *work, struct k_work_sync *sync);
|
|
|
|
/** @brief Initialize a work queue structure.
|
|
*
|
|
* This must be invoked before starting a work queue structure for the first time.
|
|
* It need not be invoked again on the same work queue structure.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue the queue structure to be initialized.
|
|
*/
|
|
void k_work_queue_init(struct k_work_q *queue);
|
|
|
|
/** @brief Initialize a work queue.
|
|
*
|
|
* This configures the work queue thread and starts it running. The function
|
|
* should not be re-invoked on a queue.
|
|
*
|
|
* @param queue pointer to the queue structure. It must be initialized
|
|
* in zeroed/bss memory or with @ref k_work_queue_init before
|
|
* use.
|
|
*
|
|
* @param stack pointer to the work thread stack area.
|
|
*
|
|
* @param stack_size size of the the work thread stack area, in bytes.
|
|
*
|
|
* @param prio initial thread priority
|
|
*
|
|
* @param cfg optional additional configuration parameters. Pass @c
|
|
* NULL if not required, to use the defaults documented in
|
|
* k_work_queue_config.
|
|
*/
|
|
void k_work_queue_start(struct k_work_q *queue,
|
|
k_thread_stack_t *stack, size_t stack_size,
|
|
int prio, const struct k_work_queue_config *cfg);
|
|
|
|
/** @brief Access the thread that animates a work queue.
|
|
*
|
|
* This is necessary to grant a work queue thread access to things the work
|
|
* items it will process are expected to use.
|
|
*
|
|
* @param queue pointer to the queue structure.
|
|
*
|
|
* @return the thread associated with the work queue.
|
|
*/
|
|
static inline k_tid_t k_work_queue_thread_get(struct k_work_q *queue);
|
|
|
|
/** @brief Wait until the work queue has drained, optionally plugging it.
|
|
*
|
|
* This blocks submission to the work queue except when coming from queue
|
|
* thread, and blocks the caller until no more work items are available in the
|
|
* queue.
|
|
*
|
|
* If @p plug is true then submission will continue to be blocked after the
|
|
* drain operation completes until k_work_queue_unplug() is invoked.
|
|
*
|
|
* Note that work items that are delayed are not yet associated with their
|
|
* work queue. They must be cancelled externally if a goal is to ensure the
|
|
* work queue remains empty. The @p plug feature can be used to prevent
|
|
* delayed items from being submitted after the drain completes.
|
|
*
|
|
* @param queue pointer to the queue structure.
|
|
*
|
|
* @param plug if true the work queue will continue to block new submissions
|
|
* after all items have drained.
|
|
*
|
|
* @retval 1 if call had to wait for the drain to complete
|
|
* @retval 0 if call did not have to wait
|
|
* @retval negative if wait was interrupted or failed
|
|
*/
|
|
int k_work_queue_drain(struct k_work_q *queue, bool plug);
|
|
|
|
/** @brief Release a work queue to accept new submissions.
|
|
*
|
|
* This releases the block on new submissions placed when k_work_queue_drain()
|
|
* is invoked with the @p plug option enabled. If this is invoked before the
|
|
* drain completes new items may be submitted as soon as the drain completes.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue pointer to the queue structure.
|
|
*
|
|
* @retval 0 if successfully unplugged
|
|
* @retval -EALREADY if the work queue was not plugged.
|
|
*/
|
|
int k_work_queue_unplug(struct k_work_q *queue);
|
|
|
|
/** @brief Initialize a delayable work structure.
|
|
*
|
|
* This must be invoked before scheduling a delayable work structure for the
|
|
* first time. It need not be invoked again on the same work structure. It
|
|
* can be re-invoked to change the associated handler, but this must be done
|
|
* when the work item is idle.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param dwork the delayable work structure to be initialized.
|
|
*
|
|
* @param handler the handler to be invoked by the work item.
|
|
*/
|
|
void k_work_init_delayable(struct k_work_delayable *dwork,
|
|
k_work_handler_t handler);
|
|
|
|
/**
|
|
* @brief Get the parent delayable work structure from a work pointer.
|
|
*
|
|
* This function is necessary when a @c k_work_handler_t function is passed to
|
|
* k_work_schedule_for_queue() and the handler needs to access data from the
|
|
* container of the containing `k_work_delayable`.
|
|
*
|
|
* @param work Address passed to the work handler
|
|
*
|
|
* @return Address of the containing @c k_work_delayable structure.
|
|
*/
|
|
static inline struct k_work_delayable *
|
|
k_work_delayable_from_work(struct k_work *work);
|
|
|
|
/** @brief Busy state flags from the delayable work item.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* can be inspected. Use locks where appropriate.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @return a mask of flags K_WORK_DELAYED, K_WORK_QUEUED, K_WORK_RUNNING, and
|
|
* K_WORK_CANCELING. A zero return value indicates the work item appears to
|
|
* be idle.
|
|
*/
|
|
int k_work_delayable_busy_get(const struct k_work_delayable *dwork);
|
|
|
|
/** @brief Test whether a delayed work item is currently pending.
|
|
*
|
|
* Wrapper to determine whether a delayed work item is in a non-idle state.
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* can be inspected. Use locks where appropriate.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @return true if and only if k_work_delayable_busy_get() returns a non-zero
|
|
* value.
|
|
*/
|
|
static inline bool k_work_delayable_is_pending(
|
|
const struct k_work_delayable *dwork);
|
|
|
|
/** @brief Get the absolute tick count at which a scheduled delayable work
|
|
* will be submitted.
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* can be inspected. Use locks where appropriate.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @return the tick count when the timer that will schedule the work item will
|
|
* expire, or the current tick count if the work is not scheduled.
|
|
*/
|
|
static inline k_ticks_t k_work_delayable_expires_get(
|
|
const struct k_work_delayable *dwork);
|
|
|
|
/** @brief Get the number of ticks until a scheduled delayable work will be
|
|
* submitted.
|
|
*
|
|
* @note This is a live snapshot of state, which may change before the result
|
|
* can be inspected. Use locks where appropriate.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @return the number of ticks until the timer that will schedule the work
|
|
* item will expire, or zero if the item is not scheduled.
|
|
*/
|
|
static inline k_ticks_t k_work_delayable_remaining_get(
|
|
const struct k_work_delayable *dwork);
|
|
|
|
/** @brief Submit an idle work item to a queue after a delay.
|
|
*
|
|
* Unlike k_work_reschedule_for_queue() this is a no-op if the work item is
|
|
* already scheduled or submitted, even if @p delay is @c K_NO_WAIT.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue the queue on which the work item should be submitted after the
|
|
* delay.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param delay the time to wait before submitting the work item. If @c
|
|
* K_NO_WAIT and the work is not pending this is equivalent to
|
|
* k_work_submit_to_queue().
|
|
*
|
|
* @retval 0 if work was already scheduled or submitted.
|
|
* @retval 1 if work has been scheduled.
|
|
* @retval -EBUSY if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
* @retval -EINVAL if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
* @retval -ENODEV if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
*/
|
|
int k_work_schedule_for_queue(struct k_work_q *queue,
|
|
struct k_work_delayable *dwork,
|
|
k_timeout_t delay);
|
|
|
|
/** @brief Submit an idle work item to the system work queue after a
|
|
* delay.
|
|
*
|
|
* This is a thin wrapper around k_work_schedule_for_queue(), with all the API
|
|
* characteristics of that function.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param delay the time to wait before submitting the work item. If @c
|
|
* K_NO_WAIT this is equivalent to k_work_submit_to_queue().
|
|
*
|
|
* @return as with k_work_schedule_for_queue().
|
|
*/
|
|
extern int k_work_schedule(struct k_work_delayable *dwork,
|
|
k_timeout_t delay);
|
|
|
|
/** @brief Reschedule a work item to a queue after a delay.
|
|
*
|
|
* Unlike k_work_schedule_for_queue() this function can change the deadline of
|
|
* a scheduled work item, and will schedule a work item that is in any state
|
|
* (e.g. is idle, submitted, or running). This function does not affect
|
|
* ("unsubmit") a work item that has been submitted to a queue.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param queue the queue on which the work item should be submitted after the
|
|
* delay.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param delay the time to wait before submitting the work item. If @c
|
|
* K_NO_WAIT this is equivalent to k_work_submit_to_queue() after canceling
|
|
* any previous scheduled submission.
|
|
*
|
|
* @note If delay is @c K_NO_WAIT ("no delay") the return values are as with
|
|
* k_work_submit_to_queue().
|
|
*
|
|
* @retval 0 if delay is @c K_NO_WAIT and work was already on a queue
|
|
* @retval 1 if
|
|
* * delay is @c K_NO_WAIT and work was not submitted but has now been queued
|
|
* to @p queue; or
|
|
* * delay not @c K_NO_WAIT and work has been scheduled
|
|
* @retval 2 if delay is @c K_NO_WAIT and work was running and has been queued
|
|
* to the queue that was running it
|
|
* @retval -EBUSY if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
* @retval -EINVAL if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
* @retval -ENODEV if @p delay is @c K_NO_WAIT and
|
|
* k_work_submit_to_queue() fails with this code.
|
|
*/
|
|
int k_work_reschedule_for_queue(struct k_work_q *queue,
|
|
struct k_work_delayable *dwork,
|
|
k_timeout_t delay);
|
|
|
|
/** @brief Reschedule a work item to the system work queue after a
|
|
* delay.
|
|
*
|
|
* This is a thin wrapper around k_work_reschedule_for_queue(), with all the
|
|
* API characteristics of that function.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param delay the time to wait before submitting the work item.
|
|
*
|
|
* @return as with k_work_reschedule_for_queue().
|
|
*/
|
|
extern int k_work_reschedule(struct k_work_delayable *dwork,
|
|
k_timeout_t delay);
|
|
|
|
/** @brief Flush delayable work.
|
|
*
|
|
* If the work is scheduled, it is immediately submitted. Then the caller
|
|
* blocks until the work completes, as with k_work_flush().
|
|
*
|
|
* @note Be careful of caller and work queue thread relative priority. If
|
|
* this function sleeps it will not return until the work queue thread
|
|
* completes the tasks that allow this thread to resume.
|
|
*
|
|
* @note Behavior is undefined if this function is invoked on @p dwork from a
|
|
* work queue running @p dwork.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param sync pointer to an opaque item containing state related to the
|
|
* pending cancellation. The object must persist until the call returns, and
|
|
* be accessible from both the caller thread and the work queue thread. The
|
|
* object must not be used for any other flush or cancel operation until this
|
|
* one completes. On architectures with CONFIG_KERNEL_COHERENCE the object
|
|
* must be allocated in coherent memory.
|
|
*
|
|
* @retval true if call had to wait for completion
|
|
* @retval false if work was already idle
|
|
*/
|
|
bool k_work_flush_delayable(struct k_work_delayable *dwork,
|
|
struct k_work_sync *sync);
|
|
|
|
/** @brief Cancel delayable work.
|
|
*
|
|
* Similar to k_work_cancel() but for delayable work. If the work is
|
|
* scheduled or submitted it is canceled. This function does not wait for the
|
|
* cancellation to complete.
|
|
*
|
|
* @note The work may still be running when this returns. Use
|
|
* k_work_flush_delayable() or k_work_cancel_delayable_sync() to ensure it is
|
|
* not running.
|
|
*
|
|
* @note Canceling delayable work does not prevent rescheduling it. It does
|
|
* prevent submitting it until the cancellation completes.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @return the k_work_delayable_busy_get() status indicating the state of the
|
|
* item after all cancellation steps performed by this call are completed.
|
|
*/
|
|
int k_work_cancel_delayable(struct k_work_delayable *dwork);
|
|
|
|
/** @brief Cancel delayable work and wait.
|
|
*
|
|
* Like k_work_cancel_delayable() but waits until the work becomes idle.
|
|
*
|
|
* @note Canceling delayable work does not prevent rescheduling it. It does
|
|
* prevent submitting it until the cancellation completes.
|
|
*
|
|
* @note Be careful of caller and work queue thread relative priority. If
|
|
* this function sleeps it will not return until the work queue thread
|
|
* completes the tasks that allow this thread to resume.
|
|
*
|
|
* @note Behavior is undefined if this function is invoked on @p dwork from a
|
|
* work queue running @p dwork.
|
|
*
|
|
* @param dwork pointer to the delayable work item.
|
|
*
|
|
* @param sync pointer to an opaque item containing state related to the
|
|
* pending cancellation. The object must persist until the call returns, and
|
|
* be accessible from both the caller thread and the work queue thread. The
|
|
* object must not be used for any other flush or cancel operation until this
|
|
* one completes. On architectures with CONFIG_KERNEL_COHERENCE the object
|
|
* must be allocated in coherent memory.
|
|
*
|
|
* @retval true if work was not idle (call had to wait for cancellation of a
|
|
* running handler to complete, or scheduled or submitted operations were
|
|
* cancelled);
|
|
* @retval false otherwise
|
|
*/
|
|
bool k_work_cancel_delayable_sync(struct k_work_delayable *dwork,
|
|
struct k_work_sync *sync);
|
|
|
|
enum {
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
/* The atomic API is used for all work and queue flags fields to
|
|
* enforce sequential consistency in SMP environments.
|
|
*/
|
|
|
|
/* Bits that represent the work item states. At least nine of the
|
|
* combinations are distinct valid stable states.
|
|
*/
|
|
K_WORK_RUNNING_BIT = 0,
|
|
K_WORK_CANCELING_BIT = 1,
|
|
K_WORK_QUEUED_BIT = 2,
|
|
K_WORK_DELAYED_BIT = 3,
|
|
|
|
K_WORK_MASK = BIT(K_WORK_DELAYED_BIT) | BIT(K_WORK_QUEUED_BIT)
|
|
| BIT(K_WORK_RUNNING_BIT) | BIT(K_WORK_CANCELING_BIT),
|
|
|
|
/* Static work flags */
|
|
K_WORK_DELAYABLE_BIT = 8,
|
|
K_WORK_DELAYABLE = BIT(K_WORK_DELAYABLE_BIT),
|
|
|
|
/* Dynamic work queue flags */
|
|
K_WORK_QUEUE_STARTED_BIT = 0,
|
|
K_WORK_QUEUE_STARTED = BIT(K_WORK_QUEUE_STARTED_BIT),
|
|
K_WORK_QUEUE_BUSY_BIT = 1,
|
|
K_WORK_QUEUE_BUSY = BIT(K_WORK_QUEUE_BUSY_BIT),
|
|
K_WORK_QUEUE_DRAIN_BIT = 2,
|
|
K_WORK_QUEUE_DRAIN = BIT(K_WORK_QUEUE_DRAIN_BIT),
|
|
K_WORK_QUEUE_PLUGGED_BIT = 3,
|
|
K_WORK_QUEUE_PLUGGED = BIT(K_WORK_QUEUE_PLUGGED_BIT),
|
|
|
|
/* Static work queue flags */
|
|
K_WORK_QUEUE_NO_YIELD_BIT = 8,
|
|
K_WORK_QUEUE_NO_YIELD = BIT(K_WORK_QUEUE_NO_YIELD_BIT),
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
/* Transient work flags */
|
|
|
|
/** @brief Flag indicating a work item that is running under a work
|
|
* queue thread.
|
|
*
|
|
* Accessed via k_work_busy_get(). May co-occur with other flags.
|
|
*/
|
|
K_WORK_RUNNING = BIT(K_WORK_RUNNING_BIT),
|
|
|
|
/** @brief Flag indicating a work item that is being canceled.
|
|
*
|
|
* Accessed via k_work_busy_get(). May co-occur with other flags.
|
|
*/
|
|
K_WORK_CANCELING = BIT(K_WORK_CANCELING_BIT),
|
|
|
|
/** @brief Flag indicating a work item that has been submitted to a
|
|
* queue but has not started running.
|
|
*
|
|
* Accessed via k_work_busy_get(). May co-occur with other flags.
|
|
*/
|
|
K_WORK_QUEUED = BIT(K_WORK_QUEUED_BIT),
|
|
|
|
/** @brief Flag indicating a delayed work item that is scheduled for
|
|
* submission to a queue.
|
|
*
|
|
* Accessed via k_work_busy_get(). May co-occur with other flags.
|
|
*/
|
|
K_WORK_DELAYED = BIT(K_WORK_DELAYED_BIT),
|
|
};
|
|
|
|
/** @brief A structure used to submit work. */
|
|
struct k_work {
|
|
/* All fields are protected by the work module spinlock. No fields
|
|
* are to be accessed except through kernel API.
|
|
*/
|
|
|
|
/* Node to link into k_work_q pending list. */
|
|
sys_snode_t node;
|
|
|
|
/* The function to be invoked by the work queue thread. */
|
|
k_work_handler_t handler;
|
|
|
|
/* The queue on which the work item was last submitted. */
|
|
struct k_work_q *queue;
|
|
|
|
/* State of the work item.
|
|
*
|
|
* The item can be DELAYED, QUEUED, and RUNNING simultaneously.
|
|
*
|
|
* It can be RUNNING and CANCELING simultaneously.
|
|
*/
|
|
uint32_t flags;
|
|
};
|
|
|
|
#define Z_WORK_INITIALIZER(work_handler) { \
|
|
.handler = work_handler, \
|
|
}
|
|
|
|
/** @brief A structure used to submit work after a delay. */
|
|
struct k_work_delayable {
|
|
/* The work item. */
|
|
struct k_work work;
|
|
|
|
/* Timeout used to submit work after a delay. */
|
|
struct _timeout timeout;
|
|
|
|
/* The queue to which the work should be submitted. */
|
|
struct k_work_q *queue;
|
|
};
|
|
|
|
#define Z_WORK_DELAYABLE_INITIALIZER(work_handler) { \
|
|
.work = { \
|
|
.handler = work_handler, \
|
|
.flags = K_WORK_DELAYABLE, \
|
|
}, \
|
|
}
|
|
|
|
/**
|
|
* @brief Initialize a statically-defined delayable work item.
|
|
*
|
|
* This macro can be used to initialize a statically-defined delayable
|
|
* work item, prior to its first use. For example,
|
|
*
|
|
* @code static K_WORK_DELAYABLE_DEFINE(<dwork>, <work_handler>); @endcode
|
|
*
|
|
* Note that if the runtime dependencies support initialization with
|
|
* k_work_init_delayable() using that will eliminate the initialized
|
|
* object in ROM that is produced by this macro and copied in at
|
|
* system startup.
|
|
*
|
|
* @param work Symbol name for delayable work item object
|
|
* @param work_handler Function to invoke each time work item is processed.
|
|
*/
|
|
#define K_WORK_DELAYABLE_DEFINE(work, work_handler) \
|
|
struct k_work_delayable work \
|
|
= Z_WORK_DELAYABLE_INITIALIZER(work_handler)
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
/* Record used to wait for work to flush.
|
|
*
|
|
* The work item is inserted into the queue that will process (or is
|
|
* processing) the item, and will be processed as soon as the item
|
|
* completes. When the flusher is processed the semaphore will be
|
|
* signaled, releasing the thread waiting for the flush.
|
|
*/
|
|
struct z_work_flusher {
|
|
struct k_work work;
|
|
struct k_sem sem;
|
|
};
|
|
|
|
/* Record used to wait for work to complete a cancellation.
|
|
*
|
|
* The work item is inserted into a global queue of pending cancels.
|
|
* When a cancelling work item goes idle any matching waiters are
|
|
* removed from pending_cancels and are woken.
|
|
*/
|
|
struct z_work_canceller {
|
|
sys_snode_t node;
|
|
struct k_work *work;
|
|
struct k_sem sem;
|
|
};
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/** @brief A structure holding internal state for a pending synchronous
|
|
* operation on a work item or queue.
|
|
*
|
|
* Instances of this type are provided by the caller for invocation of
|
|
* k_work_flush(), k_work_cancel_sync() and sibling flush and cancel APIs. A
|
|
* referenced object must persist until the call returns, and be accessible
|
|
* from both the caller thread and the work queue thread.
|
|
*
|
|
* @note If CONFIG_KERNEL_COHERENCE is enabled the object must be allocated in
|
|
* coherent memory; see arch_mem_coherent(). The stack on these architectures
|
|
* is generally not coherent. be stack-allocated. Violations are detected by
|
|
* runtime assertion.
|
|
*/
|
|
struct k_work_sync {
|
|
union {
|
|
struct z_work_flusher flusher;
|
|
struct z_work_canceller canceller;
|
|
};
|
|
};
|
|
|
|
/** @brief A structure holding optional configuration items for a work
|
|
* queue.
|
|
*
|
|
* This structure, and values it references, are not retained by
|
|
* k_work_queue_start().
|
|
*/
|
|
struct k_work_queue_config {
|
|
/** The name to be given to the work queue thread.
|
|
*
|
|
* If left null the thread will not have a name.
|
|
*/
|
|
const char *name;
|
|
|
|
/** Control whether the work queue thread should yield between
|
|
* items.
|
|
*
|
|
* Yielding between items helps guarantee the work queue
|
|
* thread does not starve other threads, including cooperative
|
|
* ones released by a work item. This is the default behavior.
|
|
*
|
|
* Set this to @c true to prevent the work queue thread from
|
|
* yielding between items. This may be appropriate when a
|
|
* sequence of items should complete without yielding
|
|
* control.
|
|
*/
|
|
bool no_yield;
|
|
};
|
|
|
|
/** @brief A structure used to hold work until it can be processed. */
|
|
struct k_work_q {
|
|
/* The thread that animates the work. */
|
|
struct k_thread thread;
|
|
|
|
/* All the following fields must be accessed only while the
|
|
* work module spinlock is held.
|
|
*/
|
|
|
|
/* List of k_work items to be worked. */
|
|
sys_slist_t pending;
|
|
|
|
/* Wait queue for idle work thread. */
|
|
_wait_q_t notifyq;
|
|
|
|
/* Wait queue for threads waiting for the queue to drain. */
|
|
_wait_q_t drainq;
|
|
|
|
/* Flags describing queue state. */
|
|
uint32_t flags;
|
|
};
|
|
|
|
/* Provide the implementation for inline functions declared above */
|
|
|
|
static inline bool k_work_is_pending(const struct k_work *work)
|
|
{
|
|
return k_work_busy_get(work) != 0;
|
|
}
|
|
|
|
static inline struct k_work_delayable *
|
|
k_work_delayable_from_work(struct k_work *work)
|
|
{
|
|
return CONTAINER_OF(work, struct k_work_delayable, work);
|
|
}
|
|
|
|
static inline bool k_work_delayable_is_pending(
|
|
const struct k_work_delayable *dwork)
|
|
{
|
|
return k_work_delayable_busy_get(dwork) != 0;
|
|
}
|
|
|
|
static inline k_ticks_t k_work_delayable_expires_get(
|
|
const struct k_work_delayable *dwork)
|
|
{
|
|
return z_timeout_expires(&dwork->timeout);
|
|
}
|
|
|
|
static inline k_ticks_t k_work_delayable_remaining_get(
|
|
const struct k_work_delayable *dwork)
|
|
{
|
|
return z_timeout_remaining(&dwork->timeout);
|
|
}
|
|
|
|
static inline k_tid_t k_work_queue_thread_get(struct k_work_q *queue)
|
|
{
|
|
return &queue->thread;
|
|
}
|
|
|
|
/** @} */
|
|
|
|
struct k_work_user;
|
|
|
|
/**
|
|
* @addtogroup workqueue_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @typedef k_work_user_handler_t
|
|
* @brief Work item handler function type for user work queues.
|
|
*
|
|
* A work item's handler function is executed by a user workqueue's thread
|
|
* when the work item is processed by the workqueue.
|
|
*
|
|
* @param work Address of the work item.
|
|
*/
|
|
typedef void (*k_work_user_handler_t)(struct k_work_user *work);
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_work_user_q {
|
|
struct k_queue queue;
|
|
struct k_thread thread;
|
|
};
|
|
|
|
enum {
|
|
K_WORK_USER_STATE_PENDING, /* Work item pending state */
|
|
};
|
|
|
|
struct k_work_user {
|
|
void *_reserved; /* Used by k_queue implementation. */
|
|
k_work_user_handler_t handler;
|
|
atomic_t flags;
|
|
};
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
#if defined(__cplusplus) && ((__cplusplus - 0) < 202002L)
|
|
#define Z_WORK_USER_INITIALIZER(work_handler) { NULL, work_handler, 0 }
|
|
#else
|
|
#define Z_WORK_USER_INITIALIZER(work_handler) \
|
|
{ \
|
|
._reserved = NULL, \
|
|
.handler = work_handler, \
|
|
.flags = 0 \
|
|
}
|
|
#endif
|
|
|
|
/**
|
|
* @brief Initialize a statically-defined user work item.
|
|
*
|
|
* This macro can be used to initialize a statically-defined user work
|
|
* item, prior to its first use. For example,
|
|
*
|
|
* @code static K_WORK_USER_DEFINE(<work>, <work_handler>); @endcode
|
|
*
|
|
* @param work Symbol name for work item object
|
|
* @param work_handler Function to invoke each time work item is processed.
|
|
*/
|
|
#define K_WORK_USER_DEFINE(work, work_handler) \
|
|
struct k_work_user work = Z_WORK_USER_INITIALIZER(work_handler)
|
|
|
|
/**
|
|
* @brief Initialize a userspace work item.
|
|
*
|
|
* This routine initializes a user workqueue work item, prior to its
|
|
* first use.
|
|
*
|
|
* @param work Address of work item.
|
|
* @param handler Function to invoke each time work item is processed.
|
|
*/
|
|
static inline void k_work_user_init(struct k_work_user *work,
|
|
k_work_user_handler_t handler)
|
|
{
|
|
*work = (struct k_work_user)Z_WORK_USER_INITIALIZER(handler);
|
|
}
|
|
|
|
/**
|
|
* @brief Check if a userspace work item is pending.
|
|
*
|
|
* This routine indicates if user work item @a work is pending in a workqueue's
|
|
* queue.
|
|
*
|
|
* @note Checking if the work is pending gives no guarantee that the
|
|
* work will still be pending when this information is used. It is up to
|
|
* the caller to make sure that this information is used in a safe manner.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work Address of work item.
|
|
*
|
|
* @return true if work item is pending, or false if it is not pending.
|
|
*/
|
|
static inline bool k_work_user_is_pending(struct k_work_user *work)
|
|
{
|
|
return atomic_test_bit(&work->flags, K_WORK_USER_STATE_PENDING);
|
|
}
|
|
|
|
/**
|
|
* @brief Submit a work item to a user mode workqueue
|
|
*
|
|
* Submits a work item to a workqueue that runs in user mode. A temporary
|
|
* memory allocation is made from the caller's resource pool which is freed
|
|
* once the worker thread consumes the k_work item. The workqueue
|
|
* thread must have memory access to the k_work item being submitted. The caller
|
|
* must have permission granted on the work_q parameter's queue object.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work_q Address of workqueue.
|
|
* @param work Address of work item.
|
|
*
|
|
* @retval -EBUSY if the work item was already in some workqueue
|
|
* @retval -ENOMEM if no memory for thread resource pool allocation
|
|
* @retval 0 Success
|
|
*/
|
|
static inline int k_work_user_submit_to_queue(struct k_work_user_q *work_q,
|
|
struct k_work_user *work)
|
|
{
|
|
int ret = -EBUSY;
|
|
|
|
if (!atomic_test_and_set_bit(&work->flags,
|
|
K_WORK_USER_STATE_PENDING)) {
|
|
ret = k_queue_alloc_append(&work_q->queue, work);
|
|
|
|
/* Couldn't insert into the queue. Clear the pending bit
|
|
* so the work item can be submitted again
|
|
*/
|
|
if (ret != 0) {
|
|
atomic_clear_bit(&work->flags,
|
|
K_WORK_USER_STATE_PENDING);
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
/**
|
|
* @brief Start a workqueue in user mode
|
|
*
|
|
* This works identically to k_work_queue_start() except it is callable from
|
|
* user mode, and the worker thread created will run in user mode. The caller
|
|
* must have permissions granted on both the work_q parameter's thread and
|
|
* queue objects, and the same restrictions on priority apply as
|
|
* k_thread_create().
|
|
*
|
|
* @param work_q Address of workqueue.
|
|
* @param stack Pointer to work queue thread's stack space, as defined by
|
|
* K_THREAD_STACK_DEFINE()
|
|
* @param stack_size Size of the work queue thread's stack (in bytes), which
|
|
* should either be the same constant passed to
|
|
* K_THREAD_STACK_DEFINE() or the value of K_THREAD_STACK_SIZEOF().
|
|
* @param prio Priority of the work queue's thread.
|
|
* @param name optional thread name. If not null a copy is made into the
|
|
* thread's name buffer.
|
|
*/
|
|
extern void k_work_user_queue_start(struct k_work_user_q *work_q,
|
|
k_thread_stack_t *stack,
|
|
size_t stack_size, int prio,
|
|
const char *name);
|
|
|
|
/**
|
|
* @brief Access the user mode thread that animates a work queue.
|
|
*
|
|
* This is necessary to grant a user mode work queue thread access to things
|
|
* the work items it will process are expected to use.
|
|
*
|
|
* @param work_q pointer to the user mode queue structure.
|
|
*
|
|
* @return the user mode thread associated with the work queue.
|
|
*/
|
|
static inline k_tid_t k_work_user_queue_thread_get(struct k_work_user_q *work_q)
|
|
{
|
|
return &work_q->thread;
|
|
}
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_work_poll {
|
|
struct k_work work;
|
|
struct k_work_q *workq;
|
|
struct z_poller poller;
|
|
struct k_poll_event *events;
|
|
int num_events;
|
|
k_work_handler_t real_handler;
|
|
struct _timeout timeout;
|
|
int poll_result;
|
|
};
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @addtogroup workqueue_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Initialize a statically-defined work item.
|
|
*
|
|
* This macro can be used to initialize a statically-defined workqueue work
|
|
* item, prior to its first use. For example,
|
|
*
|
|
* @code static K_WORK_DEFINE(<work>, <work_handler>); @endcode
|
|
*
|
|
* @param work Symbol name for work item object
|
|
* @param work_handler Function to invoke each time work item is processed.
|
|
*/
|
|
#define K_WORK_DEFINE(work, work_handler) \
|
|
struct k_work work = Z_WORK_INITIALIZER(work_handler)
|
|
|
|
/**
|
|
* @brief Initialize a triggered work item.
|
|
*
|
|
* This routine initializes a workqueue triggered work item, prior to
|
|
* its first use.
|
|
*
|
|
* @param work Address of triggered work item.
|
|
* @param handler Function to invoke each time work item is processed.
|
|
*/
|
|
extern void k_work_poll_init(struct k_work_poll *work,
|
|
k_work_handler_t handler);
|
|
|
|
/**
|
|
* @brief Submit a triggered work item.
|
|
*
|
|
* This routine schedules work item @a work to be processed by workqueue
|
|
* @a work_q when one of the given @a events is signaled. The routine
|
|
* initiates internal poller for the work item and then returns to the caller.
|
|
* Only when one of the watched events happen the work item is actually
|
|
* submitted to the workqueue and becomes pending.
|
|
*
|
|
* Submitting a previously submitted triggered work item that is still
|
|
* waiting for the event cancels the existing submission and reschedules it
|
|
* the using the new event list. Note that this behavior is inherently subject
|
|
* to race conditions with the pre-existing triggered work item and work queue,
|
|
* so care must be taken to synchronize such resubmissions externally.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @warning
|
|
* Provided array of events as well as a triggered work item must be placed
|
|
* in persistent memory (valid until work handler execution or work
|
|
* cancellation) and cannot be modified after submission.
|
|
*
|
|
* @param work_q Address of workqueue.
|
|
* @param work Address of delayed work item.
|
|
* @param events An array of events which trigger the work.
|
|
* @param num_events The number of events in the array.
|
|
* @param timeout Timeout after which the work will be scheduled
|
|
* for execution even if not triggered.
|
|
*
|
|
*
|
|
* @retval 0 Work item started watching for events.
|
|
* @retval -EINVAL Work item is being processed or has completed its work.
|
|
* @retval -EADDRINUSE Work item is pending on a different workqueue.
|
|
*/
|
|
extern int k_work_poll_submit_to_queue(struct k_work_q *work_q,
|
|
struct k_work_poll *work,
|
|
struct k_poll_event *events,
|
|
int num_events,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Submit a triggered work item to the system workqueue.
|
|
*
|
|
* This routine schedules work item @a work to be processed by system
|
|
* workqueue when one of the given @a events is signaled. The routine
|
|
* initiates internal poller for the work item and then returns to the caller.
|
|
* Only when one of the watched events happen the work item is actually
|
|
* submitted to the workqueue and becomes pending.
|
|
*
|
|
* Submitting a previously submitted triggered work item that is still
|
|
* waiting for the event cancels the existing submission and reschedules it
|
|
* the using the new event list. Note that this behavior is inherently subject
|
|
* to race conditions with the pre-existing triggered work item and work queue,
|
|
* so care must be taken to synchronize such resubmissions externally.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @warning
|
|
* Provided array of events as well as a triggered work item must not be
|
|
* modified until the item has been processed by the workqueue.
|
|
*
|
|
* @param work Address of delayed work item.
|
|
* @param events An array of events which trigger the work.
|
|
* @param num_events The number of events in the array.
|
|
* @param timeout Timeout after which the work will be scheduled
|
|
* for execution even if not triggered.
|
|
*
|
|
* @retval 0 Work item started watching for events.
|
|
* @retval -EINVAL Work item is being processed or has completed its work.
|
|
* @retval -EADDRINUSE Work item is pending on a different workqueue.
|
|
*/
|
|
extern int k_work_poll_submit(struct k_work_poll *work,
|
|
struct k_poll_event *events,
|
|
int num_events,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Cancel a triggered work item.
|
|
*
|
|
* This routine cancels the submission of triggered work item @a work.
|
|
* A triggered work item can only be canceled if no event triggered work
|
|
* submission.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param work Address of delayed work item.
|
|
*
|
|
* @retval 0 Work item canceled.
|
|
* @retval -EINVAL Work item is being processed or has completed its work.
|
|
*/
|
|
extern int k_work_poll_cancel(struct k_work_poll *work);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @defgroup msgq_apis Message Queue APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Message Queue Structure
|
|
*/
|
|
struct k_msgq {
|
|
/** Message queue wait queue */
|
|
_wait_q_t wait_q;
|
|
/** Lock */
|
|
struct k_spinlock lock;
|
|
/** Message size */
|
|
size_t msg_size;
|
|
/** Maximal number of messages */
|
|
uint32_t max_msgs;
|
|
/** Start of message buffer */
|
|
char *buffer_start;
|
|
/** End of message buffer */
|
|
char *buffer_end;
|
|
/** Read pointer */
|
|
char *read_ptr;
|
|
/** Write pointer */
|
|
char *write_ptr;
|
|
/** Number of used messages */
|
|
uint32_t used_msgs;
|
|
|
|
_POLL_EVENT;
|
|
|
|
/** Message queue */
|
|
uint8_t flags;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_msgq)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_MSGQ
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
|
|
#define Z_MSGQ_INITIALIZER(obj, q_buffer, q_msg_size, q_max_msgs) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&obj.wait_q), \
|
|
.msg_size = q_msg_size, \
|
|
.max_msgs = q_max_msgs, \
|
|
.buffer_start = q_buffer, \
|
|
.buffer_end = q_buffer + (q_max_msgs * q_msg_size), \
|
|
.read_ptr = q_buffer, \
|
|
.write_ptr = q_buffer, \
|
|
.used_msgs = 0, \
|
|
_POLL_EVENT_OBJ_INIT(obj) \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
|
|
#define K_MSGQ_FLAG_ALLOC BIT(0)
|
|
|
|
/**
|
|
* @brief Message Queue Attributes
|
|
*/
|
|
struct k_msgq_attrs {
|
|
/** Message Size */
|
|
size_t msg_size;
|
|
/** Maximal number of messages */
|
|
uint32_t max_msgs;
|
|
/** Used messages */
|
|
uint32_t used_msgs;
|
|
};
|
|
|
|
|
|
/**
|
|
* @brief Statically define and initialize a message queue.
|
|
*
|
|
* The message queue's ring buffer contains space for @a q_max_msgs messages,
|
|
* each of which is @a q_msg_size bytes long. Alignment of the message queue's
|
|
* ring buffer is not necessary, setting @a q_align to 1 is sufficient.
|
|
*
|
|
* The message queue can be accessed outside the module where it is defined
|
|
* using:
|
|
*
|
|
* @code extern struct k_msgq <name>; @endcode
|
|
*
|
|
* @param q_name Name of the message queue.
|
|
* @param q_msg_size Message size (in bytes).
|
|
* @param q_max_msgs Maximum number of messages that can be queued.
|
|
* @param q_align Alignment of the message queue's ring buffer (power of 2).
|
|
*
|
|
*/
|
|
#define K_MSGQ_DEFINE(q_name, q_msg_size, q_max_msgs, q_align) \
|
|
static char __noinit __aligned(q_align) \
|
|
_k_fifo_buf_##q_name[(q_max_msgs) * (q_msg_size)]; \
|
|
STRUCT_SECTION_ITERABLE(k_msgq, q_name) = \
|
|
Z_MSGQ_INITIALIZER(q_name, _k_fifo_buf_##q_name, \
|
|
(q_msg_size), (q_max_msgs))
|
|
|
|
/**
|
|
* @brief Initialize a message queue.
|
|
*
|
|
* This routine initializes a message queue object, prior to its first use.
|
|
*
|
|
* The message queue's ring buffer must contain space for @a max_msgs messages,
|
|
* each of which is @a msg_size bytes long. Alignment of the message queue's
|
|
* ring buffer is not necessary.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param buffer Pointer to ring buffer that holds queued messages.
|
|
* @param msg_size Message size (in bytes).
|
|
* @param max_msgs Maximum number of messages that can be queued.
|
|
*/
|
|
void k_msgq_init(struct k_msgq *msgq, char *buffer, size_t msg_size,
|
|
uint32_t max_msgs);
|
|
|
|
/**
|
|
* @brief Initialize a message queue.
|
|
*
|
|
* This routine initializes a message queue object, prior to its first use,
|
|
* allocating its internal ring buffer from the calling thread's resource
|
|
* pool.
|
|
*
|
|
* Memory allocated for the ring buffer can be released by calling
|
|
* k_msgq_cleanup(), or if userspace is enabled and the msgq object loses
|
|
* all of its references.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param msg_size Message size (in bytes).
|
|
* @param max_msgs Maximum number of messages that can be queued.
|
|
*
|
|
* @return 0 on success, -ENOMEM if there was insufficient memory in the
|
|
* thread's resource pool, or -EINVAL if the size parameters cause
|
|
* an integer overflow.
|
|
*/
|
|
__syscall int k_msgq_alloc_init(struct k_msgq *msgq, size_t msg_size,
|
|
uint32_t max_msgs);
|
|
|
|
/**
|
|
* @brief Release allocated buffer for a queue
|
|
*
|
|
* Releases memory allocated for the ring buffer.
|
|
*
|
|
* @param msgq message queue to cleanup
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -EBUSY Queue not empty
|
|
*/
|
|
int k_msgq_cleanup(struct k_msgq *msgq);
|
|
|
|
/**
|
|
* @brief Send a message to a message queue.
|
|
*
|
|
* This routine sends a message to message queue @a q.
|
|
*
|
|
* @note The message content is copied from @a data into @a msgq and the @a data
|
|
* pointer is not retained, so the message content will not be modified
|
|
* by this function.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param data Pointer to the message.
|
|
* @param timeout Non-negative waiting period to add the message,
|
|
* or one of the special values K_NO_WAIT and
|
|
* K_FOREVER.
|
|
*
|
|
* @retval 0 Message sent.
|
|
* @retval -ENOMSG Returned without waiting or queue purged.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
__syscall int k_msgq_put(struct k_msgq *msgq, const void *data, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Receive a message from a message queue.
|
|
*
|
|
* This routine receives a message from message queue @a q in a "first in,
|
|
* first out" manner.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param data Address of area to hold the received message.
|
|
* @param timeout Waiting period to receive the message,
|
|
* or one of the special values K_NO_WAIT and
|
|
* K_FOREVER.
|
|
*
|
|
* @retval 0 Message received.
|
|
* @retval -ENOMSG Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
__syscall int k_msgq_get(struct k_msgq *msgq, void *data, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Peek/read a message from a message queue.
|
|
*
|
|
* This routine reads a message from message queue @a q in a "first in,
|
|
* first out" manner and leaves the message in the queue.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param data Address of area to hold the message read from the queue.
|
|
*
|
|
* @retval 0 Message read.
|
|
* @retval -ENOMSG Returned when the queue has no message.
|
|
*/
|
|
__syscall int k_msgq_peek(struct k_msgq *msgq, void *data);
|
|
|
|
/**
|
|
* @brief Peek/read a message from a message queue at the specified index
|
|
*
|
|
* This routine reads a message from message queue at the specified index
|
|
* and leaves the message in the queue.
|
|
* k_msgq_peek_at(msgq, data, 0) is equivalent to k_msgq_peek(msgq, data)
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param data Address of area to hold the message read from the queue.
|
|
* @param idx Message queue index at which to peek
|
|
*
|
|
* @retval 0 Message read.
|
|
* @retval -ENOMSG Returned when the queue has no message at index.
|
|
*/
|
|
__syscall int k_msgq_peek_at(struct k_msgq *msgq, void *data, uint32_t idx);
|
|
|
|
/**
|
|
* @brief Purge a message queue.
|
|
*
|
|
* This routine discards all unreceived messages in a message queue's ring
|
|
* buffer. Any threads that are blocked waiting to send a message to the
|
|
* message queue are unblocked and see an -ENOMSG error code.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
*/
|
|
__syscall void k_msgq_purge(struct k_msgq *msgq);
|
|
|
|
/**
|
|
* @brief Get the amount of free space in a message queue.
|
|
*
|
|
* This routine returns the number of unused entries in a message queue's
|
|
* ring buffer.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
*
|
|
* @return Number of unused ring buffer entries.
|
|
*/
|
|
__syscall uint32_t k_msgq_num_free_get(struct k_msgq *msgq);
|
|
|
|
/**
|
|
* @brief Get basic attributes of a message queue.
|
|
*
|
|
* This routine fetches basic attributes of message queue into attr argument.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
* @param attrs pointer to message queue attribute structure.
|
|
*/
|
|
__syscall void k_msgq_get_attrs(struct k_msgq *msgq,
|
|
struct k_msgq_attrs *attrs);
|
|
|
|
|
|
static inline uint32_t z_impl_k_msgq_num_free_get(struct k_msgq *msgq)
|
|
{
|
|
return msgq->max_msgs - msgq->used_msgs;
|
|
}
|
|
|
|
/**
|
|
* @brief Get the number of messages in a message queue.
|
|
*
|
|
* This routine returns the number of messages in a message queue's ring buffer.
|
|
*
|
|
* @param msgq Address of the message queue.
|
|
*
|
|
* @return Number of messages.
|
|
*/
|
|
__syscall uint32_t k_msgq_num_used_get(struct k_msgq *msgq);
|
|
|
|
static inline uint32_t z_impl_k_msgq_num_used_get(struct k_msgq *msgq)
|
|
{
|
|
return msgq->used_msgs;
|
|
}
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @defgroup mailbox_apis Mailbox APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Mailbox Message Structure
|
|
*
|
|
*/
|
|
struct k_mbox_msg {
|
|
/** internal use only - needed for legacy API support */
|
|
uint32_t _mailbox;
|
|
/** size of message (in bytes) */
|
|
size_t size;
|
|
/** application-defined information value */
|
|
uint32_t info;
|
|
/** sender's message data buffer */
|
|
void *tx_data;
|
|
/** source thread id */
|
|
k_tid_t rx_source_thread;
|
|
/** target thread id */
|
|
k_tid_t tx_target_thread;
|
|
/** internal use only - thread waiting on send (may be a dummy) */
|
|
k_tid_t _syncing_thread;
|
|
#if (CONFIG_NUM_MBOX_ASYNC_MSGS > 0)
|
|
/** internal use only - semaphore used during asynchronous send */
|
|
struct k_sem *_async_sem;
|
|
#endif
|
|
};
|
|
/**
|
|
* @brief Mailbox Structure
|
|
*
|
|
*/
|
|
struct k_mbox {
|
|
/** Transmit messages queue */
|
|
_wait_q_t tx_msg_queue;
|
|
/** Receive message queue */
|
|
_wait_q_t rx_msg_queue;
|
|
struct k_spinlock lock;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_mbox)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_MAILBOX
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
#define Z_MBOX_INITIALIZER(obj) \
|
|
{ \
|
|
.tx_msg_queue = Z_WAIT_Q_INIT(&obj.tx_msg_queue), \
|
|
.rx_msg_queue = Z_WAIT_Q_INIT(&obj.rx_msg_queue), \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @brief Statically define and initialize a mailbox.
|
|
*
|
|
* The mailbox is to be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_mbox <name>; @endcode
|
|
*
|
|
* @param name Name of the mailbox.
|
|
*/
|
|
#define K_MBOX_DEFINE(name) \
|
|
STRUCT_SECTION_ITERABLE(k_mbox, name) = \
|
|
Z_MBOX_INITIALIZER(name) \
|
|
|
|
/**
|
|
* @brief Initialize a mailbox.
|
|
*
|
|
* This routine initializes a mailbox object, prior to its first use.
|
|
*
|
|
* @param mbox Address of the mailbox.
|
|
*/
|
|
extern void k_mbox_init(struct k_mbox *mbox);
|
|
|
|
/**
|
|
* @brief Send a mailbox message in a synchronous manner.
|
|
*
|
|
* This routine sends a message to @a mbox and waits for a receiver to both
|
|
* receive and process it. The message data may be in a buffer or non-existent
|
|
* (i.e. an empty message).
|
|
*
|
|
* @param mbox Address of the mailbox.
|
|
* @param tx_msg Address of the transmit message descriptor.
|
|
* @param timeout Waiting period for the message to be received,
|
|
* or one of the special values K_NO_WAIT
|
|
* and K_FOREVER. Once the message has been received,
|
|
* this routine waits as long as necessary for the message
|
|
* to be completely processed.
|
|
*
|
|
* @retval 0 Message sent.
|
|
* @retval -ENOMSG Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
extern int k_mbox_put(struct k_mbox *mbox, struct k_mbox_msg *tx_msg,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Send a mailbox message in an asynchronous manner.
|
|
*
|
|
* This routine sends a message to @a mbox without waiting for a receiver
|
|
* to process it. The message data may be in a buffer or non-existent
|
|
* (i.e. an empty message). Optionally, the semaphore @a sem will be given
|
|
* when the message has been both received and completely processed by
|
|
* the receiver.
|
|
*
|
|
* @param mbox Address of the mailbox.
|
|
* @param tx_msg Address of the transmit message descriptor.
|
|
* @param sem Address of a semaphore, or NULL if none is needed.
|
|
*/
|
|
extern void k_mbox_async_put(struct k_mbox *mbox, struct k_mbox_msg *tx_msg,
|
|
struct k_sem *sem);
|
|
|
|
/**
|
|
* @brief Receive a mailbox message.
|
|
*
|
|
* This routine receives a message from @a mbox, then optionally retrieves
|
|
* its data and disposes of the message.
|
|
*
|
|
* @param mbox Address of the mailbox.
|
|
* @param rx_msg Address of the receive message descriptor.
|
|
* @param buffer Address of the buffer to receive data, or NULL to defer data
|
|
* retrieval and message disposal until later.
|
|
* @param timeout Waiting period for a message to be received,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval 0 Message received.
|
|
* @retval -ENOMSG Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
*/
|
|
extern int k_mbox_get(struct k_mbox *mbox, struct k_mbox_msg *rx_msg,
|
|
void *buffer, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Retrieve mailbox message data into a buffer.
|
|
*
|
|
* This routine completes the processing of a received message by retrieving
|
|
* its data into a buffer, then disposing of the message.
|
|
*
|
|
* Alternatively, this routine can be used to dispose of a received message
|
|
* without retrieving its data.
|
|
*
|
|
* @param rx_msg Address of the receive message descriptor.
|
|
* @param buffer Address of the buffer to receive data, or NULL to discard
|
|
* the data.
|
|
*/
|
|
extern void k_mbox_data_get(struct k_mbox_msg *rx_msg, void *buffer);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @defgroup pipe_apis Pipe APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/** Pipe Structure */
|
|
struct k_pipe {
|
|
unsigned char *buffer; /**< Pipe buffer: may be NULL */
|
|
size_t size; /**< Buffer size */
|
|
size_t bytes_used; /**< # bytes used in buffer */
|
|
size_t read_index; /**< Where in buffer to read from */
|
|
size_t write_index; /**< Where in buffer to write */
|
|
struct k_spinlock lock; /**< Synchronization lock */
|
|
|
|
struct {
|
|
_wait_q_t readers; /**< Reader wait queue */
|
|
_wait_q_t writers; /**< Writer wait queue */
|
|
} wait_q; /** Wait queue */
|
|
|
|
_POLL_EVENT;
|
|
|
|
uint8_t flags; /**< Flags */
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_pipe)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_PIPE
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
#define K_PIPE_FLAG_ALLOC BIT(0) /** Buffer was allocated */
|
|
|
|
#define Z_PIPE_INITIALIZER(obj, pipe_buffer, pipe_buffer_size) \
|
|
{ \
|
|
.buffer = pipe_buffer, \
|
|
.size = pipe_buffer_size, \
|
|
.bytes_used = 0, \
|
|
.read_index = 0, \
|
|
.write_index = 0, \
|
|
.lock = {}, \
|
|
.wait_q = { \
|
|
.readers = Z_WAIT_Q_INIT(&obj.wait_q.readers), \
|
|
.writers = Z_WAIT_Q_INIT(&obj.wait_q.writers) \
|
|
}, \
|
|
_POLL_EVENT_OBJ_INIT(obj) \
|
|
.flags = 0, \
|
|
}
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @brief Statically define and initialize a pipe.
|
|
*
|
|
* The pipe can be accessed outside the module where it is defined using:
|
|
*
|
|
* @code extern struct k_pipe <name>; @endcode
|
|
*
|
|
* @param name Name of the pipe.
|
|
* @param pipe_buffer_size Size of the pipe's ring buffer (in bytes),
|
|
* or zero if no ring buffer is used.
|
|
* @param pipe_align Alignment of the pipe's ring buffer (power of 2).
|
|
*
|
|
*/
|
|
#define K_PIPE_DEFINE(name, pipe_buffer_size, pipe_align) \
|
|
static unsigned char __noinit __aligned(pipe_align) \
|
|
_k_pipe_buf_##name[pipe_buffer_size]; \
|
|
STRUCT_SECTION_ITERABLE(k_pipe, name) = \
|
|
Z_PIPE_INITIALIZER(name, _k_pipe_buf_##name, pipe_buffer_size)
|
|
|
|
/**
|
|
* @brief Initialize a pipe.
|
|
*
|
|
* This routine initializes a pipe object, prior to its first use.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
* @param buffer Address of the pipe's ring buffer, or NULL if no ring buffer
|
|
* is used.
|
|
* @param size Size of the pipe's ring buffer (in bytes), or zero if no ring
|
|
* buffer is used.
|
|
*/
|
|
void k_pipe_init(struct k_pipe *pipe, unsigned char *buffer, size_t size);
|
|
|
|
/**
|
|
* @brief Release a pipe's allocated buffer
|
|
*
|
|
* If a pipe object was given a dynamically allocated buffer via
|
|
* k_pipe_alloc_init(), this will free it. This function does nothing
|
|
* if the buffer wasn't dynamically allocated.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
* @retval 0 on success
|
|
* @retval -EAGAIN nothing to cleanup
|
|
*/
|
|
int k_pipe_cleanup(struct k_pipe *pipe);
|
|
|
|
/**
|
|
* @brief Initialize a pipe and allocate a buffer for it
|
|
*
|
|
* Storage for the buffer region will be allocated from the calling thread's
|
|
* resource pool. This memory will be released if k_pipe_cleanup() is called,
|
|
* or userspace is enabled and the pipe object loses all references to it.
|
|
*
|
|
* This function should only be called on uninitialized pipe objects.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
* @param size Size of the pipe's ring buffer (in bytes), or zero if no ring
|
|
* buffer is used.
|
|
* @retval 0 on success
|
|
* @retval -ENOMEM if memory couldn't be allocated
|
|
*/
|
|
__syscall int k_pipe_alloc_init(struct k_pipe *pipe, size_t size);
|
|
|
|
/**
|
|
* @brief Write data to a pipe.
|
|
*
|
|
* This routine writes up to @a bytes_to_write bytes of data to @a pipe.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
* @param data Address of data to write.
|
|
* @param bytes_to_write Size of data (in bytes).
|
|
* @param bytes_written Address of area to hold the number of bytes written.
|
|
* @param min_xfer Minimum number of bytes to write.
|
|
* @param timeout Waiting period to wait for the data to be written,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval 0 At least @a min_xfer bytes of data were written.
|
|
* @retval -EIO Returned without waiting; zero data bytes were written.
|
|
* @retval -EAGAIN Waiting period timed out; between zero and @a min_xfer
|
|
* minus one data bytes were written.
|
|
*/
|
|
__syscall int k_pipe_put(struct k_pipe *pipe, void *data,
|
|
size_t bytes_to_write, size_t *bytes_written,
|
|
size_t min_xfer, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Read data from a pipe.
|
|
*
|
|
* This routine reads up to @a bytes_to_read bytes of data from @a pipe.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
* @param data Address to place the data read from pipe.
|
|
* @param bytes_to_read Maximum number of data bytes to read.
|
|
* @param bytes_read Address of area to hold the number of bytes read.
|
|
* @param min_xfer Minimum number of data bytes to read.
|
|
* @param timeout Waiting period to wait for the data to be read,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval 0 At least @a min_xfer bytes of data were read.
|
|
* @retval -EINVAL invalid parameters supplied
|
|
* @retval -EIO Returned without waiting; zero data bytes were read.
|
|
* @retval -EAGAIN Waiting period timed out; between zero and @a min_xfer
|
|
* minus one data bytes were read.
|
|
*/
|
|
__syscall int k_pipe_get(struct k_pipe *pipe, void *data,
|
|
size_t bytes_to_read, size_t *bytes_read,
|
|
size_t min_xfer, k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Query the number of bytes that may be read from @a pipe.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
*
|
|
* @retval a number n such that 0 <= n <= @ref k_pipe.size; the
|
|
* result is zero for unbuffered pipes.
|
|
*/
|
|
__syscall size_t k_pipe_read_avail(struct k_pipe *pipe);
|
|
|
|
/**
|
|
* @brief Query the number of bytes that may be written to @a pipe
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
*
|
|
* @retval a number n such that 0 <= n <= @ref k_pipe.size; the
|
|
* result is zero for unbuffered pipes.
|
|
*/
|
|
__syscall size_t k_pipe_write_avail(struct k_pipe *pipe);
|
|
|
|
/**
|
|
* @brief Flush the pipe of write data
|
|
*
|
|
* This routine flushes the pipe. Flushing the pipe is equivalent to reading
|
|
* both all the data in the pipe's buffer and all the data waiting to go into
|
|
* that pipe into a large temporary buffer and discarding the buffer. Any
|
|
* writers that were previously pended become unpended.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
*/
|
|
__syscall void k_pipe_flush(struct k_pipe *pipe);
|
|
|
|
/**
|
|
* @brief Flush the pipe's internal buffer
|
|
*
|
|
* This routine flushes the pipe's internal buffer. This is equivalent to
|
|
* reading up to N bytes from the pipe (where N is the size of the pipe's
|
|
* buffer) into a temporary buffer and then discarding that buffer. If there
|
|
* were writers previously pending, then some may unpend as they try to fill
|
|
* up the pipe's emptied buffer.
|
|
*
|
|
* @param pipe Address of the pipe.
|
|
*/
|
|
__syscall void k_pipe_buffer_flush(struct k_pipe *pipe);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
struct k_mem_slab_info {
|
|
uint32_t num_blocks;
|
|
size_t block_size;
|
|
uint32_t num_used;
|
|
#ifdef CONFIG_MEM_SLAB_TRACE_MAX_UTILIZATION
|
|
uint32_t max_used;
|
|
#endif
|
|
};
|
|
|
|
struct k_mem_slab {
|
|
_wait_q_t wait_q;
|
|
struct k_spinlock lock;
|
|
char *buffer;
|
|
char *free_list;
|
|
struct k_mem_slab_info info;
|
|
|
|
SYS_PORT_TRACING_TRACKING_FIELD(k_mem_slab)
|
|
|
|
#ifdef CONFIG_OBJ_CORE_MEM_SLAB
|
|
struct k_obj_core obj_core;
|
|
#endif
|
|
};
|
|
|
|
#define Z_MEM_SLAB_INITIALIZER(_slab, _slab_buffer, _slab_block_size, \
|
|
_slab_num_blocks) \
|
|
{ \
|
|
.wait_q = Z_WAIT_Q_INIT(&(_slab).wait_q), \
|
|
.lock = {}, \
|
|
.buffer = _slab_buffer, \
|
|
.free_list = NULL, \
|
|
.info = {_slab_num_blocks, _slab_block_size, 0} \
|
|
}
|
|
|
|
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @defgroup mem_slab_apis Memory Slab APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Statically define and initialize a memory slab in a public (non-static) scope.
|
|
*
|
|
* The memory slab's buffer contains @a slab_num_blocks memory blocks
|
|
* that are @a slab_block_size bytes long. The buffer is aligned to a
|
|
* @a slab_align -byte boundary. To ensure that each memory block is similarly
|
|
* aligned to this boundary, @a slab_block_size must also be a multiple of
|
|
* @a slab_align.
|
|
*
|
|
* The memory slab can be accessed outside the module where it is defined
|
|
* using:
|
|
*
|
|
* @code extern struct k_mem_slab <name>; @endcode
|
|
*
|
|
* @note This macro cannot be used together with a static keyword.
|
|
* If such a use-case is desired, use @ref K_MEM_SLAB_DEFINE_STATIC
|
|
* instead.
|
|
*
|
|
* @param name Name of the memory slab.
|
|
* @param slab_block_size Size of each memory block (in bytes).
|
|
* @param slab_num_blocks Number memory blocks.
|
|
* @param slab_align Alignment of the memory slab's buffer (power of 2).
|
|
*/
|
|
#define K_MEM_SLAB_DEFINE(name, slab_block_size, slab_num_blocks, slab_align) \
|
|
char __noinit_named(k_mem_slab_buf_##name) \
|
|
__aligned(WB_UP(slab_align)) \
|
|
_k_mem_slab_buf_##name[(slab_num_blocks) * WB_UP(slab_block_size)]; \
|
|
STRUCT_SECTION_ITERABLE(k_mem_slab, name) = \
|
|
Z_MEM_SLAB_INITIALIZER(name, _k_mem_slab_buf_##name, \
|
|
WB_UP(slab_block_size), slab_num_blocks)
|
|
|
|
/**
|
|
* @brief Statically define and initialize a memory slab in a private (static) scope.
|
|
*
|
|
* The memory slab's buffer contains @a slab_num_blocks memory blocks
|
|
* that are @a slab_block_size bytes long. The buffer is aligned to a
|
|
* @a slab_align -byte boundary. To ensure that each memory block is similarly
|
|
* aligned to this boundary, @a slab_block_size must also be a multiple of
|
|
* @a slab_align.
|
|
*
|
|
* @param name Name of the memory slab.
|
|
* @param slab_block_size Size of each memory block (in bytes).
|
|
* @param slab_num_blocks Number memory blocks.
|
|
* @param slab_align Alignment of the memory slab's buffer (power of 2).
|
|
*/
|
|
#define K_MEM_SLAB_DEFINE_STATIC(name, slab_block_size, slab_num_blocks, slab_align) \
|
|
static char __noinit_named(k_mem_slab_buf_##name) \
|
|
__aligned(WB_UP(slab_align)) \
|
|
_k_mem_slab_buf_##name[(slab_num_blocks) * WB_UP(slab_block_size)]; \
|
|
static STRUCT_SECTION_ITERABLE(k_mem_slab, name) = \
|
|
Z_MEM_SLAB_INITIALIZER(name, _k_mem_slab_buf_##name, \
|
|
WB_UP(slab_block_size), slab_num_blocks)
|
|
|
|
/**
|
|
* @brief Initialize a memory slab.
|
|
*
|
|
* Initializes a memory slab, prior to its first use.
|
|
*
|
|
* The memory slab's buffer contains @a slab_num_blocks memory blocks
|
|
* that are @a slab_block_size bytes long. The buffer must be aligned to an
|
|
* N-byte boundary matching a word boundary, where N is a power of 2
|
|
* (i.e. 4 on 32-bit systems, 8, 16, ...).
|
|
* To ensure that each memory block is similarly aligned to this boundary,
|
|
* @a slab_block_size must also be a multiple of N.
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
* @param buffer Pointer to buffer used for the memory blocks.
|
|
* @param block_size Size of each memory block (in bytes).
|
|
* @param num_blocks Number of memory blocks.
|
|
*
|
|
* @retval 0 on success
|
|
* @retval -EINVAL invalid data supplied
|
|
*
|
|
*/
|
|
extern int k_mem_slab_init(struct k_mem_slab *slab, void *buffer,
|
|
size_t block_size, uint32_t num_blocks);
|
|
|
|
/**
|
|
* @brief Allocate memory from a memory slab.
|
|
*
|
|
* This routine allocates a memory block from a memory slab.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
* @note When CONFIG_MULTITHREADING=n any @a timeout is treated as K_NO_WAIT.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
* @param mem Pointer to block address area.
|
|
* @param timeout Non-negative waiting period to wait for operation to complete.
|
|
* Use K_NO_WAIT to return without waiting,
|
|
* or K_FOREVER to wait as long as necessary.
|
|
*
|
|
* @retval 0 Memory allocated. The block address area pointed at by @a mem
|
|
* is set to the starting address of the memory block.
|
|
* @retval -ENOMEM Returned without waiting.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
* @retval -EINVAL Invalid data supplied
|
|
*/
|
|
extern int k_mem_slab_alloc(struct k_mem_slab *slab, void **mem,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Free memory allocated from a memory slab.
|
|
*
|
|
* This routine releases a previously allocated memory block back to its
|
|
* associated memory slab.
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
* @param mem Pointer to the memory block (as returned by k_mem_slab_alloc()).
|
|
*/
|
|
extern void k_mem_slab_free(struct k_mem_slab *slab, void *mem);
|
|
|
|
/**
|
|
* @brief Get the number of used blocks in a memory slab.
|
|
*
|
|
* This routine gets the number of memory blocks that are currently
|
|
* allocated in @a slab.
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
*
|
|
* @return Number of allocated memory blocks.
|
|
*/
|
|
static inline uint32_t k_mem_slab_num_used_get(struct k_mem_slab *slab)
|
|
{
|
|
return slab->info.num_used;
|
|
}
|
|
|
|
/**
|
|
* @brief Get the number of maximum used blocks so far in a memory slab.
|
|
*
|
|
* This routine gets the maximum number of memory blocks that were
|
|
* allocated in @a slab.
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
*
|
|
* @return Maximum number of allocated memory blocks.
|
|
*/
|
|
static inline uint32_t k_mem_slab_max_used_get(struct k_mem_slab *slab)
|
|
{
|
|
#ifdef CONFIG_MEM_SLAB_TRACE_MAX_UTILIZATION
|
|
return slab->info.max_used;
|
|
#else
|
|
ARG_UNUSED(slab);
|
|
return 0;
|
|
#endif
|
|
}
|
|
|
|
/**
|
|
* @brief Get the number of unused blocks in a memory slab.
|
|
*
|
|
* This routine gets the number of memory blocks that are currently
|
|
* unallocated in @a slab.
|
|
*
|
|
* @param slab Address of the memory slab.
|
|
*
|
|
* @return Number of unallocated memory blocks.
|
|
*/
|
|
static inline uint32_t k_mem_slab_num_free_get(struct k_mem_slab *slab)
|
|
{
|
|
return slab->info.num_blocks - slab->info.num_used;
|
|
}
|
|
|
|
/**
|
|
* @brief Get the memory stats for a memory slab
|
|
*
|
|
* This routine gets the runtime memory usage stats for the slab @a slab.
|
|
*
|
|
* @param slab Address of the memory slab
|
|
* @param stats Pointer to memory into which to copy memory usage statistics
|
|
*
|
|
* @retval 0 Success
|
|
* @retval -EINVAL Any parameter points to NULL
|
|
*/
|
|
|
|
int k_mem_slab_runtime_stats_get(struct k_mem_slab *slab, struct sys_memory_stats *stats);
|
|
|
|
/**
|
|
* @brief Reset the maximum memory usage for a slab
|
|
*
|
|
* This routine resets the maximum memory usage for the slab @a slab to its
|
|
* current usage.
|
|
*
|
|
* @param slab Address of the memory slab
|
|
*
|
|
* @retval 0 Success
|
|
* @retval -EINVAL Memory slab is NULL
|
|
*/
|
|
int k_mem_slab_runtime_stats_reset_max(struct k_mem_slab *slab);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @addtogroup heap_apis
|
|
* @{
|
|
*/
|
|
|
|
/* kernel synchronized heap struct */
|
|
|
|
struct k_heap {
|
|
struct sys_heap heap;
|
|
_wait_q_t wait_q;
|
|
struct k_spinlock lock;
|
|
};
|
|
|
|
/**
|
|
* @brief Initialize a k_heap
|
|
*
|
|
* This constructs a synchronized k_heap object over a memory region
|
|
* specified by the user. Note that while any alignment and size can
|
|
* be passed as valid parameters, internal alignment restrictions
|
|
* inside the inner sys_heap mean that not all bytes may be usable as
|
|
* allocated memory.
|
|
*
|
|
* @param h Heap struct to initialize
|
|
* @param mem Pointer to memory.
|
|
* @param bytes Size of memory region, in bytes
|
|
*/
|
|
void k_heap_init(struct k_heap *h, void *mem, size_t bytes);
|
|
|
|
/** @brief Allocate aligned memory from a k_heap
|
|
*
|
|
* Behaves in all ways like k_heap_alloc(), except that the returned
|
|
* memory (if available) will have a starting address in memory which
|
|
* is a multiple of the specified power-of-two alignment value in
|
|
* bytes. The resulting memory can be returned to the heap using
|
|
* k_heap_free().
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
* @note When CONFIG_MULTITHREADING=n any @a timeout is treated as K_NO_WAIT.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param h Heap from which to allocate
|
|
* @param align Alignment in bytes, must be a power of two
|
|
* @param bytes Number of bytes requested
|
|
* @param timeout How long to wait, or K_NO_WAIT
|
|
* @return Pointer to memory the caller can now use
|
|
*/
|
|
void *k_heap_aligned_alloc(struct k_heap *h, size_t align, size_t bytes,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Allocate memory from a k_heap
|
|
*
|
|
* Allocates and returns a memory buffer from the memory region owned
|
|
* by the heap. If no memory is available immediately, the call will
|
|
* block for the specified timeout (constructed via the standard
|
|
* timeout API, or K_NO_WAIT or K_FOREVER) waiting for memory to be
|
|
* freed. If the allocation cannot be performed by the expiration of
|
|
* the timeout, NULL will be returned.
|
|
* Allocated memory is aligned on a multiple of pointer sizes.
|
|
*
|
|
* @note @a timeout must be set to K_NO_WAIT if called from ISR.
|
|
* @note When CONFIG_MULTITHREADING=n any @a timeout is treated as K_NO_WAIT.
|
|
*
|
|
* @funcprops \isr_ok
|
|
*
|
|
* @param h Heap from which to allocate
|
|
* @param bytes Desired size of block to allocate
|
|
* @param timeout How long to wait, or K_NO_WAIT
|
|
* @return A pointer to valid heap memory, or NULL
|
|
*/
|
|
void *k_heap_alloc(struct k_heap *h, size_t bytes,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Free memory allocated by k_heap_alloc()
|
|
*
|
|
* Returns the specified memory block, which must have been returned
|
|
* from k_heap_alloc(), to the heap for use by other callers. Passing
|
|
* a NULL block is legal, and has no effect.
|
|
*
|
|
* @param h Heap to which to return the memory
|
|
* @param mem A valid memory block, or NULL
|
|
*/
|
|
void k_heap_free(struct k_heap *h, void *mem);
|
|
|
|
/* Hand-calculated minimum heap sizes needed to return a successful
|
|
* 1-byte allocation. See details in lib/os/heap.[ch]
|
|
*/
|
|
#define Z_HEAP_MIN_SIZE (sizeof(void *) > 4 ? 56 : 44)
|
|
|
|
/**
|
|
* @brief Define a static k_heap in the specified linker section
|
|
*
|
|
* This macro defines and initializes a static memory region and
|
|
* k_heap of the requested size in the specified linker section.
|
|
* After kernel start, &name can be used as if k_heap_init() had
|
|
* been called.
|
|
*
|
|
* Note that this macro enforces a minimum size on the memory region
|
|
* to accommodate metadata requirements. Very small heaps will be
|
|
* padded to fit.
|
|
*
|
|
* @param name Symbol name for the struct k_heap object
|
|
* @param bytes Size of memory region, in bytes
|
|
* @param in_section __attribute__((section(name))
|
|
*/
|
|
#define Z_HEAP_DEFINE_IN_SECT(name, bytes, in_section) \
|
|
char in_section \
|
|
__aligned(8) /* CHUNK_UNIT */ \
|
|
kheap_##name[MAX(bytes, Z_HEAP_MIN_SIZE)]; \
|
|
STRUCT_SECTION_ITERABLE(k_heap, name) = { \
|
|
.heap = { \
|
|
.init_mem = kheap_##name, \
|
|
.init_bytes = MAX(bytes, Z_HEAP_MIN_SIZE), \
|
|
}, \
|
|
}
|
|
|
|
/**
|
|
* @brief Define a static k_heap
|
|
*
|
|
* This macro defines and initializes a static memory region and
|
|
* k_heap of the requested size. After kernel start, &name can be
|
|
* used as if k_heap_init() had been called.
|
|
*
|
|
* Note that this macro enforces a minimum size on the memory region
|
|
* to accommodate metadata requirements. Very small heaps will be
|
|
* padded to fit.
|
|
*
|
|
* @param name Symbol name for the struct k_heap object
|
|
* @param bytes Size of memory region, in bytes
|
|
*/
|
|
#define K_HEAP_DEFINE(name, bytes) \
|
|
Z_HEAP_DEFINE_IN_SECT(name, bytes, \
|
|
__noinit_named(kheap_buf_##name))
|
|
|
|
/**
|
|
* @brief Define a static k_heap in uncached memory
|
|
*
|
|
* This macro defines and initializes a static memory region and
|
|
* k_heap of the requested size in uncached memory. After kernel
|
|
* start, &name can be used as if k_heap_init() had been called.
|
|
*
|
|
* Note that this macro enforces a minimum size on the memory region
|
|
* to accommodate metadata requirements. Very small heaps will be
|
|
* padded to fit.
|
|
*
|
|
* @param name Symbol name for the struct k_heap object
|
|
* @param bytes Size of memory region, in bytes
|
|
*/
|
|
#define K_HEAP_DEFINE_NOCACHE(name, bytes) \
|
|
Z_HEAP_DEFINE_IN_SECT(name, bytes, __nocache)
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @defgroup heap_apis Heap APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Allocate memory from the heap with a specified alignment.
|
|
*
|
|
* This routine provides semantics similar to aligned_alloc(); memory is
|
|
* allocated from the heap with a specified alignment. However, one minor
|
|
* difference is that k_aligned_alloc() accepts any non-zero @p size,
|
|
* whereas aligned_alloc() only accepts a @p size that is an integral
|
|
* multiple of @p align.
|
|
*
|
|
* Above, aligned_alloc() refers to:
|
|
* C11 standard (ISO/IEC 9899:2011): 7.22.3.1
|
|
* The aligned_alloc function (p: 347-348)
|
|
*
|
|
* @param align Alignment of memory requested (in bytes).
|
|
* @param size Amount of memory requested (in bytes).
|
|
*
|
|
* @return Address of the allocated memory if successful; otherwise NULL.
|
|
*/
|
|
extern void *k_aligned_alloc(size_t align, size_t size);
|
|
|
|
/**
|
|
* @brief Allocate memory from the heap.
|
|
*
|
|
* This routine provides traditional malloc() semantics. Memory is
|
|
* allocated from the heap memory pool.
|
|
* Allocated memory is aligned on a multiple of pointer sizes.
|
|
*
|
|
* @param size Amount of memory requested (in bytes).
|
|
*
|
|
* @return Address of the allocated memory if successful; otherwise NULL.
|
|
*/
|
|
extern void *k_malloc(size_t size);
|
|
|
|
/**
|
|
* @brief Free memory allocated from heap.
|
|
*
|
|
* This routine provides traditional free() semantics. The memory being
|
|
* returned must have been allocated from the heap memory pool.
|
|
*
|
|
* If @a ptr is NULL, no operation is performed.
|
|
*
|
|
* @param ptr Pointer to previously allocated memory.
|
|
*/
|
|
extern void k_free(void *ptr);
|
|
|
|
/**
|
|
* @brief Allocate memory from heap, array style
|
|
*
|
|
* This routine provides traditional calloc() semantics. Memory is
|
|
* allocated from the heap memory pool and zeroed.
|
|
*
|
|
* @param nmemb Number of elements in the requested array
|
|
* @param size Size of each array element (in bytes).
|
|
*
|
|
* @return Address of the allocated memory if successful; otherwise NULL.
|
|
*/
|
|
extern void *k_calloc(size_t nmemb, size_t size);
|
|
|
|
/** @} */
|
|
|
|
/* polling API - PRIVATE */
|
|
|
|
#ifdef CONFIG_POLL
|
|
#define _INIT_OBJ_POLL_EVENT(obj) do { (obj)->poll_event = NULL; } while (false)
|
|
#else
|
|
#define _INIT_OBJ_POLL_EVENT(obj) do { } while (false)
|
|
#endif
|
|
|
|
/* private - types bit positions */
|
|
enum _poll_types_bits {
|
|
/* can be used to ignore an event */
|
|
_POLL_TYPE_IGNORE,
|
|
|
|
/* to be signaled by k_poll_signal_raise() */
|
|
_POLL_TYPE_SIGNAL,
|
|
|
|
/* semaphore availability */
|
|
_POLL_TYPE_SEM_AVAILABLE,
|
|
|
|
/* queue/FIFO/LIFO data availability */
|
|
_POLL_TYPE_DATA_AVAILABLE,
|
|
|
|
/* msgq data availability */
|
|
_POLL_TYPE_MSGQ_DATA_AVAILABLE,
|
|
|
|
/* pipe data availability */
|
|
_POLL_TYPE_PIPE_DATA_AVAILABLE,
|
|
|
|
_POLL_NUM_TYPES
|
|
};
|
|
|
|
#define Z_POLL_TYPE_BIT(type) (1U << ((type) - 1U))
|
|
|
|
/* private - states bit positions */
|
|
enum _poll_states_bits {
|
|
/* default state when creating event */
|
|
_POLL_STATE_NOT_READY,
|
|
|
|
/* signaled by k_poll_signal_raise() */
|
|
_POLL_STATE_SIGNALED,
|
|
|
|
/* semaphore is available */
|
|
_POLL_STATE_SEM_AVAILABLE,
|
|
|
|
/* data is available to read on queue/FIFO/LIFO */
|
|
_POLL_STATE_DATA_AVAILABLE,
|
|
|
|
/* queue/FIFO/LIFO wait was cancelled */
|
|
_POLL_STATE_CANCELLED,
|
|
|
|
/* data is available to read on a message queue */
|
|
_POLL_STATE_MSGQ_DATA_AVAILABLE,
|
|
|
|
/* data is available to read from a pipe */
|
|
_POLL_STATE_PIPE_DATA_AVAILABLE,
|
|
|
|
_POLL_NUM_STATES
|
|
};
|
|
|
|
#define Z_POLL_STATE_BIT(state) (1U << ((state) - 1U))
|
|
|
|
#define _POLL_EVENT_NUM_UNUSED_BITS \
|
|
(32 - (0 \
|
|
+ 8 /* tag */ \
|
|
+ _POLL_NUM_TYPES \
|
|
+ _POLL_NUM_STATES \
|
|
+ 1 /* modes */ \
|
|
))
|
|
|
|
/* end of polling API - PRIVATE */
|
|
|
|
|
|
/**
|
|
* @defgroup poll_apis Async polling APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
|
|
/* Public polling API */
|
|
|
|
/* public - values for k_poll_event.type bitfield */
|
|
#define K_POLL_TYPE_IGNORE 0
|
|
#define K_POLL_TYPE_SIGNAL Z_POLL_TYPE_BIT(_POLL_TYPE_SIGNAL)
|
|
#define K_POLL_TYPE_SEM_AVAILABLE Z_POLL_TYPE_BIT(_POLL_TYPE_SEM_AVAILABLE)
|
|
#define K_POLL_TYPE_DATA_AVAILABLE Z_POLL_TYPE_BIT(_POLL_TYPE_DATA_AVAILABLE)
|
|
#define K_POLL_TYPE_FIFO_DATA_AVAILABLE K_POLL_TYPE_DATA_AVAILABLE
|
|
#define K_POLL_TYPE_MSGQ_DATA_AVAILABLE Z_POLL_TYPE_BIT(_POLL_TYPE_MSGQ_DATA_AVAILABLE)
|
|
#define K_POLL_TYPE_PIPE_DATA_AVAILABLE Z_POLL_TYPE_BIT(_POLL_TYPE_PIPE_DATA_AVAILABLE)
|
|
|
|
/* public - polling modes */
|
|
enum k_poll_modes {
|
|
/* polling thread does not take ownership of objects when available */
|
|
K_POLL_MODE_NOTIFY_ONLY = 0,
|
|
|
|
K_POLL_NUM_MODES
|
|
};
|
|
|
|
/* public - values for k_poll_event.state bitfield */
|
|
#define K_POLL_STATE_NOT_READY 0
|
|
#define K_POLL_STATE_SIGNALED Z_POLL_STATE_BIT(_POLL_STATE_SIGNALED)
|
|
#define K_POLL_STATE_SEM_AVAILABLE Z_POLL_STATE_BIT(_POLL_STATE_SEM_AVAILABLE)
|
|
#define K_POLL_STATE_DATA_AVAILABLE Z_POLL_STATE_BIT(_POLL_STATE_DATA_AVAILABLE)
|
|
#define K_POLL_STATE_FIFO_DATA_AVAILABLE K_POLL_STATE_DATA_AVAILABLE
|
|
#define K_POLL_STATE_MSGQ_DATA_AVAILABLE Z_POLL_STATE_BIT(_POLL_STATE_MSGQ_DATA_AVAILABLE)
|
|
#define K_POLL_STATE_PIPE_DATA_AVAILABLE Z_POLL_STATE_BIT(_POLL_STATE_PIPE_DATA_AVAILABLE)
|
|
#define K_POLL_STATE_CANCELLED Z_POLL_STATE_BIT(_POLL_STATE_CANCELLED)
|
|
|
|
/* public - poll signal object */
|
|
struct k_poll_signal {
|
|
/** PRIVATE - DO NOT TOUCH */
|
|
sys_dlist_t poll_events;
|
|
|
|
/**
|
|
* 1 if the event has been signaled, 0 otherwise. Stays set to 1 until
|
|
* user resets it to 0.
|
|
*/
|
|
unsigned int signaled;
|
|
|
|
/** custom result value passed to k_poll_signal_raise() if needed */
|
|
int result;
|
|
};
|
|
|
|
#define K_POLL_SIGNAL_INITIALIZER(obj) \
|
|
{ \
|
|
.poll_events = SYS_DLIST_STATIC_INIT(&obj.poll_events), \
|
|
.signaled = 0, \
|
|
.result = 0, \
|
|
}
|
|
/**
|
|
* @brief Poll Event
|
|
*
|
|
*/
|
|
struct k_poll_event {
|
|
/** PRIVATE - DO NOT TOUCH */
|
|
sys_dnode_t _node;
|
|
|
|
/** PRIVATE - DO NOT TOUCH */
|
|
struct z_poller *poller;
|
|
|
|
/** optional user-specified tag, opaque, untouched by the API */
|
|
uint32_t tag:8;
|
|
|
|
/** bitfield of event types (bitwise-ORed K_POLL_TYPE_xxx values) */
|
|
uint32_t type:_POLL_NUM_TYPES;
|
|
|
|
/** bitfield of event states (bitwise-ORed K_POLL_STATE_xxx values) */
|
|
uint32_t state:_POLL_NUM_STATES;
|
|
|
|
/** mode of operation, from enum k_poll_modes */
|
|
uint32_t mode:1;
|
|
|
|
/** unused bits in 32-bit word */
|
|
uint32_t unused:_POLL_EVENT_NUM_UNUSED_BITS;
|
|
|
|
/** per-type data */
|
|
union {
|
|
void *obj;
|
|
struct k_poll_signal *signal;
|
|
struct k_sem *sem;
|
|
struct k_fifo *fifo;
|
|
struct k_queue *queue;
|
|
struct k_msgq *msgq;
|
|
#ifdef CONFIG_PIPES
|
|
struct k_pipe *pipe;
|
|
#endif
|
|
};
|
|
};
|
|
|
|
#define K_POLL_EVENT_INITIALIZER(_event_type, _event_mode, _event_obj) \
|
|
{ \
|
|
.poller = NULL, \
|
|
.type = _event_type, \
|
|
.state = K_POLL_STATE_NOT_READY, \
|
|
.mode = _event_mode, \
|
|
.unused = 0, \
|
|
{ \
|
|
.obj = _event_obj, \
|
|
}, \
|
|
}
|
|
|
|
#define K_POLL_EVENT_STATIC_INITIALIZER(_event_type, _event_mode, _event_obj, \
|
|
event_tag) \
|
|
{ \
|
|
.tag = event_tag, \
|
|
.type = _event_type, \
|
|
.state = K_POLL_STATE_NOT_READY, \
|
|
.mode = _event_mode, \
|
|
.unused = 0, \
|
|
{ \
|
|
.obj = _event_obj, \
|
|
}, \
|
|
}
|
|
|
|
/**
|
|
* @brief Initialize one struct k_poll_event instance
|
|
*
|
|
* After this routine is called on a poll event, the event it ready to be
|
|
* placed in an event array to be passed to k_poll().
|
|
*
|
|
* @param event The event to initialize.
|
|
* @param type A bitfield of the types of event, from the K_POLL_TYPE_xxx
|
|
* values. Only values that apply to the same object being polled
|
|
* can be used together. Choosing K_POLL_TYPE_IGNORE disables the
|
|
* event.
|
|
* @param mode Future. Use K_POLL_MODE_NOTIFY_ONLY.
|
|
* @param obj Kernel object or poll signal.
|
|
*/
|
|
|
|
extern void k_poll_event_init(struct k_poll_event *event, uint32_t type,
|
|
int mode, void *obj);
|
|
|
|
/**
|
|
* @brief Wait for one or many of multiple poll events to occur
|
|
*
|
|
* This routine allows a thread to wait concurrently for one or many of
|
|
* multiple poll events to have occurred. Such events can be a kernel object
|
|
* being available, like a semaphore, or a poll signal event.
|
|
*
|
|
* When an event notifies that a kernel object is available, the kernel object
|
|
* is not "given" to the thread calling k_poll(): it merely signals the fact
|
|
* that the object was available when the k_poll() call was in effect. Also,
|
|
* all threads trying to acquire an object the regular way, i.e. by pending on
|
|
* the object, have precedence over the thread polling on the object. This
|
|
* means that the polling thread will never get the poll event on an object
|
|
* until the object becomes available and its pend queue is empty. For this
|
|
* reason, the k_poll() call is more effective when the objects being polled
|
|
* only have one thread, the polling thread, trying to acquire them.
|
|
*
|
|
* When k_poll() returns 0, the caller should loop on all the events that were
|
|
* passed to k_poll() and check the state field for the values that were
|
|
* expected and take the associated actions.
|
|
*
|
|
* Before being reused for another call to k_poll(), the user has to reset the
|
|
* state field to K_POLL_STATE_NOT_READY.
|
|
*
|
|
* When called from user mode, a temporary memory allocation is required from
|
|
* the caller's resource pool.
|
|
*
|
|
* @param events An array of events to be polled for.
|
|
* @param num_events The number of events in the array.
|
|
* @param timeout Waiting period for an event to be ready,
|
|
* or one of the special values K_NO_WAIT and K_FOREVER.
|
|
*
|
|
* @retval 0 One or more events are ready.
|
|
* @retval -EAGAIN Waiting period timed out.
|
|
* @retval -EINTR Polling has been interrupted, e.g. with
|
|
* k_queue_cancel_wait(). All output events are still set and valid,
|
|
* cancelled event(s) will be set to K_POLL_STATE_CANCELLED. In other
|
|
* words, -EINTR status means that at least one of output events is
|
|
* K_POLL_STATE_CANCELLED.
|
|
* @retval -ENOMEM Thread resource pool insufficient memory (user mode only)
|
|
* @retval -EINVAL Bad parameters (user mode only)
|
|
*/
|
|
|
|
__syscall int k_poll(struct k_poll_event *events, int num_events,
|
|
k_timeout_t timeout);
|
|
|
|
/**
|
|
* @brief Initialize a poll signal object.
|
|
*
|
|
* Ready a poll signal object to be signaled via k_poll_signal_raise().
|
|
*
|
|
* @param sig A poll signal.
|
|
*/
|
|
|
|
__syscall void k_poll_signal_init(struct k_poll_signal *sig);
|
|
|
|
/*
|
|
* @brief Reset a poll signal object's state to unsignaled.
|
|
*
|
|
* @param sig A poll signal object
|
|
*/
|
|
__syscall void k_poll_signal_reset(struct k_poll_signal *sig);
|
|
|
|
/**
|
|
* @brief Fetch the signaled state and result value of a poll signal
|
|
*
|
|
* @param sig A poll signal object
|
|
* @param signaled An integer buffer which will be written nonzero if the
|
|
* object was signaled
|
|
* @param result An integer destination buffer which will be written with the
|
|
* result value if the object was signaled, or an undefined
|
|
* value if it was not.
|
|
*/
|
|
__syscall void k_poll_signal_check(struct k_poll_signal *sig,
|
|
unsigned int *signaled, int *result);
|
|
|
|
/**
|
|
* @brief Signal a poll signal object.
|
|
*
|
|
* This routine makes ready a poll signal, which is basically a poll event of
|
|
* type K_POLL_TYPE_SIGNAL. If a thread was polling on that event, it will be
|
|
* made ready to run. A @a result value can be specified.
|
|
*
|
|
* The poll signal contains a 'signaled' field that, when set by
|
|
* k_poll_signal_raise(), stays set until the user sets it back to 0 with
|
|
* k_poll_signal_reset(). It thus has to be reset by the user before being
|
|
* passed again to k_poll() or k_poll() will consider it being signaled, and
|
|
* will return immediately.
|
|
*
|
|
* @note The result is stored and the 'signaled' field is set even if
|
|
* this function returns an error indicating that an expiring poll was
|
|
* not notified. The next k_poll() will detect the missed raise.
|
|
*
|
|
* @param sig A poll signal.
|
|
* @param result The value to store in the result field of the signal.
|
|
*
|
|
* @retval 0 The signal was delivered successfully.
|
|
* @retval -EAGAIN The polling thread's timeout is in the process of expiring.
|
|
*/
|
|
|
|
__syscall int k_poll_signal_raise(struct k_poll_signal *sig, int result);
|
|
|
|
/** @} */
|
|
|
|
/**
|
|
* @defgroup cpu_idle_apis CPU Idling APIs
|
|
* @ingroup kernel_apis
|
|
* @{
|
|
*/
|
|
/**
|
|
* @brief Make the CPU idle.
|
|
*
|
|
* This function makes the CPU idle until an event wakes it up.
|
|
*
|
|
* In a regular system, the idle thread should be the only thread responsible
|
|
* for making the CPU idle and triggering any type of power management.
|
|
* However, in some more constrained systems, such as a single-threaded system,
|
|
* the only thread would be responsible for this if needed.
|
|
*
|
|
* @note In some architectures, before returning, the function unmasks interrupts
|
|
* unconditionally.
|
|
*/
|
|
static inline void k_cpu_idle(void)
|
|
{
|
|
arch_cpu_idle();
|
|
}
|
|
|
|
/**
|
|
* @brief Make the CPU idle in an atomic fashion.
|
|
*
|
|
* Similar to k_cpu_idle(), but must be called with interrupts locked.
|
|
*
|
|
* Enabling interrupts and entering a low-power mode will be atomic,
|
|
* i.e. there will be no period of time where interrupts are enabled before
|
|
* the processor enters a low-power mode.
|
|
*
|
|
* After waking up from the low-power mode, the interrupt lockout state will
|
|
* be restored as if by irq_unlock(key).
|
|
*
|
|
* @param key Interrupt locking key obtained from irq_lock().
|
|
*/
|
|
static inline void k_cpu_atomic_idle(unsigned int key)
|
|
{
|
|
arch_cpu_atomic_idle(key);
|
|
}
|
|
|
|
/**
|
|
* @}
|
|
*/
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
* @internal
|
|
*/
|
|
#ifdef ARCH_EXCEPT
|
|
/* This architecture has direct support for triggering a CPU exception */
|
|
#define z_except_reason(reason) ARCH_EXCEPT(reason)
|
|
#else
|
|
|
|
#if !defined(CONFIG_ASSERT_NO_FILE_INFO)
|
|
#define __EXCEPT_LOC() __ASSERT_PRINT("@ %s:%d\n", __FILE__, __LINE__)
|
|
#else
|
|
#define __EXCEPT_LOC()
|
|
#endif
|
|
|
|
/* NOTE: This is the implementation for arches that do not implement
|
|
* ARCH_EXCEPT() to generate a real CPU exception.
|
|
*
|
|
* We won't have a real exception frame to determine the PC value when
|
|
* the oops occurred, so print file and line number before we jump into
|
|
* the fatal error handler.
|
|
*/
|
|
#define z_except_reason(reason) do { \
|
|
__EXCEPT_LOC(); \
|
|
z_fatal_error(reason, NULL); \
|
|
} while (false)
|
|
|
|
#endif /* _ARCH__EXCEPT */
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
/**
|
|
* @brief Fatally terminate a thread
|
|
*
|
|
* This should be called when a thread has encountered an unrecoverable
|
|
* runtime condition and needs to terminate. What this ultimately
|
|
* means is determined by the _fatal_error_handler() implementation, which
|
|
* will be called will reason code K_ERR_KERNEL_OOPS.
|
|
*
|
|
* If this is called from ISR context, the default system fatal error handler
|
|
* will treat it as an unrecoverable system error, just like k_panic().
|
|
*/
|
|
#define k_oops() z_except_reason(K_ERR_KERNEL_OOPS)
|
|
|
|
/**
|
|
* @brief Fatally terminate the system
|
|
*
|
|
* This should be called when the Zephyr kernel has encountered an
|
|
* unrecoverable runtime condition and needs to terminate. What this ultimately
|
|
* means is determined by the _fatal_error_handler() implementation, which
|
|
* will be called will reason code K_ERR_KERNEL_PANIC.
|
|
*/
|
|
#define k_panic() z_except_reason(K_ERR_KERNEL_PANIC)
|
|
|
|
/**
|
|
* @cond INTERNAL_HIDDEN
|
|
*/
|
|
|
|
/*
|
|
* private APIs that are utilized by one or more public APIs
|
|
*/
|
|
|
|
/**
|
|
* @internal
|
|
*/
|
|
#ifdef CONFIG_MULTITHREADING
|
|
/**
|
|
* @internal
|
|
*/
|
|
extern void z_init_static_threads(void);
|
|
#else
|
|
/**
|
|
* @internal
|
|
*/
|
|
#define z_init_static_threads() do { } while (false)
|
|
#endif
|
|
|
|
/**
|
|
* @internal
|
|
*/
|
|
extern void z_timer_expiration_handler(struct _timeout *t);
|
|
/**
|
|
* INTERNAL_HIDDEN @endcond
|
|
*/
|
|
|
|
#ifdef CONFIG_PRINTK
|
|
/**
|
|
* @brief Emit a character buffer to the console device
|
|
*
|
|
* @param c String of characters to print
|
|
* @param n The length of the string
|
|
*
|
|
*/
|
|
__syscall void k_str_out(char *c, size_t n);
|
|
#endif
|
|
|
|
/**
|
|
* @brief Disable preservation of floating point context information.
|
|
*
|
|
* This routine informs the kernel that the specified thread
|
|
* will no longer be using the floating point registers.
|
|
*
|
|
* @warning
|
|
* Some architectures apply restrictions on how the disabling of floating
|
|
* point preservation may be requested, see arch_float_disable.
|
|
*
|
|
* @warning
|
|
* This routine should only be used to disable floating point support for
|
|
* a thread that currently has such support enabled.
|
|
*
|
|
* @param thread ID of thread.
|
|
*
|
|
* @retval 0 On success.
|
|
* @retval -ENOTSUP If the floating point disabling is not implemented.
|
|
* -EINVAL If the floating point disabling could not be performed.
|
|
*/
|
|
__syscall int k_float_disable(struct k_thread *thread);
|
|
|
|
/**
|
|
* @brief Enable preservation of floating point context information.
|
|
*
|
|
* This routine informs the kernel that the specified thread
|
|
* will use the floating point registers.
|
|
|
|
* Invoking this routine initializes the thread's floating point context info
|
|
* to that of an FPU that has been reset. The next time the thread is scheduled
|
|
* by z_swap() it will either inherit an FPU that is guaranteed to be in a
|
|
* "sane" state (if the most recent user of the FPU was cooperatively swapped
|
|
* out) or the thread's own floating point context will be loaded (if the most
|
|
* recent user of the FPU was preempted, or if this thread is the first user
|
|
* of the FPU). Thereafter, the kernel will protect the thread's FP context
|
|
* so that it is not altered during a preemptive context switch.
|
|
*
|
|
* The @a options parameter indicates which floating point register sets will
|
|
* be used by the specified thread.
|
|
*
|
|
* For x86 options:
|
|
*
|
|
* - K_FP_REGS indicates x87 FPU and MMX registers only
|
|
* - K_SSE_REGS indicates SSE registers (and also x87 FPU and MMX registers)
|
|
*
|
|
* @warning
|
|
* Some architectures apply restrictions on how the enabling of floating
|
|
* point preservation may be requested, see arch_float_enable.
|
|
*
|
|
* @warning
|
|
* This routine should only be used to enable floating point support for
|
|
* a thread that currently has such support enabled.
|
|
*
|
|
* @param thread ID of thread.
|
|
* @param options architecture dependent options
|
|
*
|
|
* @retval 0 On success.
|
|
* @retval -ENOTSUP If the floating point enabling is not implemented.
|
|
* -EINVAL If the floating point enabling could not be performed.
|
|
*/
|
|
__syscall int k_float_enable(struct k_thread *thread, unsigned int options);
|
|
|
|
/**
|
|
* @brief Get the runtime statistics of a thread
|
|
*
|
|
* @param thread ID of thread.
|
|
* @param stats Pointer to struct to copy statistics into.
|
|
* @return -EINVAL if null pointers, otherwise 0
|
|
*/
|
|
int k_thread_runtime_stats_get(k_tid_t thread,
|
|
k_thread_runtime_stats_t *stats);
|
|
|
|
/**
|
|
* @brief Get the runtime statistics of all threads
|
|
*
|
|
* @param stats Pointer to struct to copy statistics into.
|
|
* @return -EINVAL if null pointers, otherwise 0
|
|
*/
|
|
int k_thread_runtime_stats_all_get(k_thread_runtime_stats_t *stats);
|
|
|
|
/**
|
|
* @brief Enable gathering of runtime statistics for specified thread
|
|
*
|
|
* This routine enables the gathering of runtime statistics for the specified
|
|
* thread.
|
|
*
|
|
* @param thread ID of thread
|
|
* @return -EINVAL if invalid thread ID, otherwise 0
|
|
*/
|
|
extern int k_thread_runtime_stats_enable(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Disable gathering of runtime statistics for specified thread
|
|
*
|
|
* This routine disables the gathering of runtime statistics for the specified
|
|
* thread.
|
|
*
|
|
* @param thread ID of thread
|
|
* @return -EINVAL if invalid thread ID, otherwise 0
|
|
*/
|
|
extern int k_thread_runtime_stats_disable(k_tid_t thread);
|
|
|
|
/**
|
|
* @brief Enable gathering of system runtime statistics
|
|
*
|
|
* This routine enables the gathering of system runtime statistics. Note that
|
|
* it does not affect the gathering of similar statistics for individual
|
|
* threads.
|
|
*/
|
|
extern void k_sys_runtime_stats_enable(void);
|
|
|
|
/**
|
|
* @brief Disable gathering of system runtime statistics
|
|
*
|
|
* This routine disables the gathering of system runtime statistics. Note that
|
|
* it does not affect the gathering of similar statistics for individual
|
|
* threads.
|
|
*/
|
|
extern void k_sys_runtime_stats_disable(void);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|
|
|
|
#include <zephyr/tracing/tracing.h>
|
|
#include <syscalls/kernel.h>
|
|
|
|
#endif /* !_ASMLANGUAGE */
|
|
|
|
#endif /* ZEPHYR_INCLUDE_KERNEL_H_ */
|