event_groups源文件分析
/**
* An event group is a collection of bits to which an application can assign a
* meaning. For example, an application may create an event group to convey
* the status of various CAN bus related events in which bit 0 might mean "A CAN
* message has been received and is ready for processing", bit 1 might mean "The
* application has queued a message that is ready for sending onto the CAN
* network", and bit 2 might mean "It is time to send a SYNC message onto the
* CAN network" etc. A task can then test the bit values to see which events
* are active, and optionally enter the Blocked state to wait for a specified
* bit or a group of specified bits to be active. To continue the CAN bus
* example, a CAN controlling task can enter the Blocked state (and therefore
* not consume any processing time) until either bit 0, bit 1 or bit 2 are
* active, at which time the bit that was actually active would inform the task
* which action it had to take (process a received message, send a message, or
* send a SYNC).
*
* The event groups implementation contains intelligence to avoid race
* conditions that would otherwise occur were an application to use a simple
* variable for the same purpose. This is particularly important with respect
* to when a bit within an event group is to be cleared, and when bits have to
* be set and then tested atomically - as is the case where event groups are
* used to create a synchronisation point between multiple tasks (a
* 'rendezvous').
*
* \defgroup EventGroup
*/
首先是头文件里对于event_groups的介绍,event_groups相当于是事件驱动器,应用程序可以使用它来标记某个状态,进而促成事件驱动。
比如,某个应用程序可以创建一个event_group来传递CAN总线刚完成了接收亦或是需要发送信息,等等状态信息。
任务系统可以通过event_group中的信息来确定是否该任务是阻塞或者是就绪态,刚才的例子里可以通过状态信息来判断当前应该是让这个CAN总线任务是堵塞还是就绪,然后通知调度系统去处理他。
既然event_group中存的是各种状态信息,那么多个任务之间的同步自然都会涉及到他,那么就不可避免的有可能会发生资源的竞争问题。event_group的实现里自然包括了智能避免竞争,避免发生这种情况。
环境
编译环境:keil
固件库:Keil.STM32F7xx_DFP.2.9.0
目标开发板:STM32F767IG
目标系统:FreeRTOS 9.0
EventGroup创建函数
如果定义了configSUPPORT_STATIC_ALLOCATION==1的情况下,事件驱动一般就使用静态创建方式。
event_group需要一小片空间来存储其自身,如果调用xEventGropuCreate()函数就会自动申请其所需的内存空间。
但是如果xEventGropuCreateStatic()来创建的话,那么就需要使用者提供所需的内存空间才行。
所以xEventGropuCreateStatic()函数就是静态创建函数,不需要动态分配的。
虽然EventGroup和系统时钟无关,但是由于FreeRTOSConfig.h中configUSE_16_BIT_TICKS对于寄存器大小的设置,会导致EventGroup的大小与其相关。
configUSE_16_BIT_TICKS 如果是 1 的话,event group 就只有8位可用
configUSE_16_BIT_TICKS 如果是 0 的话,event group 就有24位可用。
创建成功返回其指针,失败的话返回NULL
静态创建
#if( configSUPPORT_STATIC_ALLOCATION == 1 )
EventGroupHandle_t xEventGroupCreateStatic( StaticEventGroup_t *pxEventGroupBuffer )
{
EventGroup_t *pxEventBits;
/* A StaticEventGroup_t object must be provided. */
configASSERT( pxEventGroupBuffer );
/* The user has provided a statically allocated event group - use it. */
pxEventBits = ( EventGroup_t * ) pxEventGroupBuffer;
/*lint !e740 EventGroup_t and StaticEventGroup_t are guaranteed to have the same size and alignment requirement - checked by configASSERT(). */
if( pxEventBits != NULL )
{
pxEventBits->uxEventBits = 0;
vListInitialise( &( pxEventBits->xTasksWaitingForBits ) );
#if( configSUPPORT_DYNAMIC_ALLOCATION == 1 )
{
/* Both static and dynamic allocation can be used, so note that
this event group was created statically in case the event group
is later deleted. */
pxEventBits->ucStaticallyAllocated = pdTRUE;
}
#endif /* configSUPPORT_DYNAMIC_ALLOCATION */
traceEVENT_GROUP_CREATE( pxEventBits );
}
else
{
traceEVENT_GROUP_CREATE_FAILED();
}
return ( EventGroupHandle_t ) pxEventBits;
}
#endif /* configSUPPORT_STATIC_ALLOCATION */
EventGroup_t 结构体
typedef struct xEventGroupDefinition
{
EventBits_t uxEventBits;
List_t xTasksWaitingForBits; /*< List of tasks waiting for a bit to be set. */
#if( configUSE_TRACE_FACILITY == 1 )
UBaseType_t uxEventGroupNumber;
#endif
#if( ( configSUPPORT_STATIC_ALLOCATION == 1 ) && ( configSUPPORT_DYNAMIC_ALLOCATION == 1 ) )
uint8_t ucStaticallyAllocated; /*< Set to pdTRUE if the event group is statically allocated to ensure no attempt is made to free the memory. */
#endif
} EventGroup_t;
可以看到这里的 EventBits_t 是与 TickType_t 相关的
typedef TickType_t EventBits_t;
这个结构体里有状态存储的变量,对应的任务列表,如果可以静态或动态分配的情况下分配的标志位,标识其是否使用静态分配。
xSTATIC_EVENT_GROUP 结构体
typedef struct xSTATIC_EVENT_GROUP
{
TickType_t xDummy1;
StaticList_t xDummy2;
#if( configUSE_TRACE_FACILITY == 1 )
UBaseType_t uxDummy3;
#endif
#if( ( configSUPPORT_STATIC_ALLOCATION == 1 ) && ( configSUPPORT_DYNAMIC_ALLOCATION == 1 ) )
uint8_t ucDummy4;
#endif
} StaticEventGroup_t;
静态结构和上面的差不多。
再看这个函数,首先传入了其所需要的空间,检查其是否合法
清空各位,与其相关的任务列表初始化
如果同时可以静态或者动态申请的情况下,默认用静态申请,保证其不会被意外删除了
最后返回动态的地址。
静态申请的方式更像是一个初始化函数,给新建的变量初始化一次而已。
动态创建
#if( configSUPPORT_DYNAMIC_ALLOCATION == 1 )
EventGroupHandle_t xEventGroupCreate( void )
{
...
pxEventBits = ( EventGroup_t * ) pvPortMalloc( sizeof( EventGroup_t ) );
...
动态创建基本都相同,只是没有传入参数,而且这里用了动态内存申请而已。
等待条件函数
static BaseType_t prvTestWaitCondition( const EventBits_t uxCurrentEventBits, const EventBits_t uxBitsToWaitFor, const BaseType_t xWaitForAllBits )
{
BaseType_t xWaitConditionMet = pdFALSE;
if( xWaitForAllBits == pdFALSE )
{
/* Task only has to wait for one bit within uxBitsToWaitFor to be
set. Is one already set? */
if( ( uxCurrentEventBits & uxBitsToWaitFor ) != ( EventBits_t ) 0 )
{
xWaitConditionMet = pdTRUE;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
else
{
/* Task has to wait for all the bits in uxBitsToWaitFor to be set.
Are they set already? */
if( ( uxCurrentEventBits & uxBitsToWaitFor ) == uxBitsToWaitFor )
{
xWaitConditionMet = pdTRUE;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
return xWaitConditionMet;
}
这个函数用来检测uxCurrentEventBits中的位是否被置为了目标位。
如果xWaitForAllBits==pdTRUE的情况下,等待条件就是全部位都需要达到目标状态才行。
如果不是这种情况,那么就是任何一位到达目标状态就ok
完成要求的情况下返回1,没完成的情况下返回0
EventGroup等待位函数
EventBits_t xEventGroupWaitBits( EventGroupHandle_t xEventGroup, const EventBits_t uxBitsToWaitFor, const BaseType_t xClearOnExit, const BaseType_t xWaitForAllBits, TickType_t xTicksToWait )
{
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
EventBits_t uxReturn, uxControlBits = 0;
BaseType_t xWaitConditionMet, xAlreadyYielded;
BaseType_t xTimeoutOccurred = pdFALSE;
/* Check the user is not attempting to wait on the bits used by the kernel
itself, and that at least one bit is being requested. */
configASSERT( xEventGroup );
configASSERT( ( uxBitsToWaitFor & eventEVENT_BITS_CONTROL_BYTES ) == 0 );
configASSERT( uxBitsToWaitFor != 0 );
#if ( ( INCLUDE_xTaskGetSchedulerState == 1 ) || ( configUSE_TIMERS == 1 ) )
{
configASSERT( !( ( xTaskGetSchedulerState() == taskSCHEDULER_SUSPENDED ) && ( xTicksToWait != 0 ) ) );
}
#endif
vTaskSuspendAll();
{
const EventBits_t uxCurrentEventBits = pxEventBits->uxEventBits;
/* Check to see if the wait condition is already met or not. */
xWaitConditionMet = prvTestWaitCondition( uxCurrentEventBits, uxBitsToWaitFor, xWaitForAllBits );
if( xWaitConditionMet != pdFALSE )
{
/* The wait condition has already been met so there is no need to
block. */
uxReturn = uxCurrentEventBits;
xTicksToWait = ( TickType_t ) 0;
/* Clear the wait bits if requested to do so. */
if( xClearOnExit != pdFALSE )
{
pxEventBits->uxEventBits &= ~uxBitsToWaitFor;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
else if( xTicksToWait == ( TickType_t ) 0 )
{
/* The wait condition has not been met, but no block time was
specified, so just return the current value. */
uxReturn = uxCurrentEventBits;
}
else
{
/* The task is going to block to wait for its required bits to be
set. uxControlBits are used to remember the specified behaviour of
this call to xEventGroupWaitBits() - for use when the event bits
unblock the task. */
if( xClearOnExit != pdFALSE )
{
uxControlBits |= eventCLEAR_EVENTS_ON_EXIT_BIT;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
if( xWaitForAllBits != pdFALSE )
{
uxControlBits |= eventWAIT_FOR_ALL_BITS;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
/* Store the bits that the calling task is waiting for in the
task's event list item so the kernel knows when a match is
found. Then enter the blocked state. */
vTaskPlaceOnUnorderedEventList( &( pxEventBits->xTasksWaitingForBits ), ( uxBitsToWaitFor | uxControlBits ), xTicksToWait );
/* This is obsolete as it will get set after the task unblocks, but
some compilers mistakenly generate a warning about the variable
being returned without being set if it is not done. */
uxReturn = 0;
traceEVENT_GROUP_WAIT_BITS_BLOCK( xEventGroup, uxBitsToWaitFor );
}
}
xAlreadyYielded = xTaskResumeAll();
if( xTicksToWait != ( TickType_t ) 0 )
{
if( xAlreadyYielded == pdFALSE )
{
portYIELD_WITHIN_API();
}
else
{
mtCOVERAGE_TEST_MARKER();
}
/* The task blocked to wait for its required bits to be set - at this
point either the required bits were set or the block time expired. If
the required bits were set they will have been stored in the task's
event list item, and they should now be retrieved then cleared. */
uxReturn = uxTaskResetEventItemValue();
if( ( uxReturn & eventUNBLOCKED_DUE_TO_BIT_SET ) == ( EventBits_t ) 0 )
{
taskENTER_CRITICAL();
{
/* The task timed out, just return the current event bit value. */
uxReturn = pxEventBits->uxEventBits;
/* It is possible that the event bits were updated between this
task leaving the Blocked state and running again. */
if( prvTestWaitCondition( uxReturn, uxBitsToWaitFor, xWaitForAllBits ) != pdFALSE )
{
if( xClearOnExit != pdFALSE )
{
pxEventBits->uxEventBits &= ~uxBitsToWaitFor;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
taskEXIT_CRITICAL();
/* Prevent compiler warnings when trace macros are not used. */
xTimeoutOccurred = pdFALSE;
}
else
{
/* The task unblocked because the bits were set. */
}
/* The task blocked so control bits may have been set. */
uxReturn &= ~eventEVENT_BITS_CONTROL_BYTES;
}
traceEVENT_GROUP_WAIT_BITS_END( xEventGroup, uxBitsToWaitFor, xTimeoutOccurred );
return uxReturn;
}
这个函数也是不能中断调用。
此函数用来等待某些位被置位
参数检测之后,挂起调度系统
先检查一次是否达到目标状态。
- 如果达到了目标状态,就不需要阻塞了,返回当前状态,等待时间清空
- 判断是否需要置位后清除位,需要的话就清除一下。
-
如果时间到了,返回当前状态
- 没到达目标状态,时间没耗尽的情况下,把等待位任务加入到等待列表中去,进行阻塞
恢复调度系统
再次检测是否耗尽时间,没耗尽时间,进入阻塞
-
耗尽时间的情况下,还没到达目标位,PV操作 返回当前位
-
耗尽时间,PV操作,到达目标位,根据需要清空当前位
返回最终状态。
同步函数
EventBits_t xEventGroupSync( EventGroupHandle_t xEventGroup, const EventBits_t uxBitsToSet, const EventBits_t uxBitsToWaitFor, TickType_t xTicksToWait )
{
EventBits_t uxOriginalBitValue, uxReturn;
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
BaseType_t xAlreadyYielded;
BaseType_t xTimeoutOccurred = pdFALSE;
configASSERT( ( uxBitsToWaitFor & eventEVENT_BITS_CONTROL_BYTES ) == 0 );
configASSERT( uxBitsToWaitFor != 0 );
#if ( ( INCLUDE_xTaskGetSchedulerState == 1 ) || ( configUSE_TIMERS == 1 ) )
{
configASSERT( !( ( xTaskGetSchedulerState() == taskSCHEDULER_SUSPENDED ) && ( xTicksToWait != 0 ) ) );
}
#endif
vTaskSuspendAll();
{
uxOriginalBitValue = pxEventBits->uxEventBits;
( void ) xEventGroupSetBits( xEventGroup, uxBitsToSet );
if( ( ( uxOriginalBitValue | uxBitsToSet ) & uxBitsToWaitFor ) == uxBitsToWaitFor )
{
/* All the rendezvous bits are now set - no need to block. */
uxReturn = ( uxOriginalBitValue | uxBitsToSet );
/* Rendezvous always clear the bits. They will have been cleared
already unless this is the only task in the rendezvous. */
pxEventBits->uxEventBits &= ~uxBitsToWaitFor;
xTicksToWait = 0;
}
else
{
if( xTicksToWait != ( TickType_t ) 0 )
{
traceEVENT_GROUP_SYNC_BLOCK( xEventGroup, uxBitsToSet, uxBitsToWaitFor );
/* Store the bits that the calling task is waiting for in the
task's event list item so the kernel knows when a match is
found. Then enter the blocked state. */
vTaskPlaceOnUnorderedEventList( &( pxEventBits->xTasksWaitingForBits ), ( uxBitsToWaitFor | eventCLEAR_EVENTS_ON_EXIT_BIT | eventWAIT_FOR_ALL_BITS ), xTicksToWait );
/* This assignment is obsolete as uxReturn will get set after
the task unblocks, but some compilers mistakenly generate a
warning about uxReturn being returned without being set if the
assignment is omitted. */
uxReturn = 0;
}
else
{
/* The rendezvous bits were not set, but no block time was
specified - just return the current event bit value. */
uxReturn = pxEventBits->uxEventBits;
}
}
}
xAlreadyYielded = xTaskResumeAll();
if( xTicksToWait != ( TickType_t ) 0 )
{
if( xAlreadyYielded == pdFALSE )
{
portYIELD_WITHIN_API();
}
else
{
mtCOVERAGE_TEST_MARKER();
}
/* The task blocked to wait for its required bits to be set - at this
point either the required bits were set or the block time expired. If
the required bits were set they will have been stored in the task's
event list item, and they should now be retrieved then cleared. */
uxReturn = uxTaskResetEventItemValue();
if( ( uxReturn & eventUNBLOCKED_DUE_TO_BIT_SET ) == ( EventBits_t ) 0 )
{
/* The task timed out, just return the current event bit value. */
taskENTER_CRITICAL();
{
uxReturn = pxEventBits->uxEventBits;
/* Although the task got here because it timed out before the
bits it was waiting for were set, it is possible that since it
unblocked another task has set the bits. If this is the case
then it needs to clear the bits before exiting. */
if( ( uxReturn & uxBitsToWaitFor ) == uxBitsToWaitFor )
{
pxEventBits->uxEventBits &= ~uxBitsToWaitFor;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
taskEXIT_CRITICAL();
xTimeoutOccurred = pdTRUE;
}
else
{
/* The task unblocked because the bits were set. */
}
/* Control bits might be set as the task had blocked should not be
returned. */
uxReturn &= ~eventEVENT_BITS_CONTROL_BYTES;
}
traceEVENT_GROUP_SYNC_END( xEventGroup, uxBitsToSet, uxBitsToWaitFor, xTimeoutOccurred );
return uxReturn;
}
这是任务同步函数,简单说就是要所有和该event_group相关的任务在到达同步点之后进行等待,等待其他所有人都到达同步点再继续下一步。
这个函数不能在中断里调用,参数一是对应的事件组,其必须初始化以后才能传入这里,参数二是要置位状态位,参数三是目标状态位,参数四是最大等待同步时间,返回就两种情况,一个是到达目标状态了,那么返回目标值,没到达目标状态,也就是时间耗尽了,返回当前状态。
再看代码具体做了什么。
首先检查了参数,挂起调度系统。
给当前事件组写入要置为的状态,检查目标状态和当前状态是否相同
-
相同的情况下,所有位都被置位了,无需等待,清空事件组中的位。
-
不相同的情况下,如果最大等待时间不为0的情况下,进入阻塞,把该事件组加入到等待标志位的事件组任务列表中去。
- 时间为0时,返回当前状态
恢复调度系统。
再次检测是否耗尽了时间
- 如果还有时间,调用上下文切换,进行阻塞。直到时间耗尽或者被置位,再次进行检测
- 如果是时间耗尽还没被置位,进入PV操作阶段,返回当前状态。
- 如果时间耗尽但是被置位了,进入PV操作阶段,清空被置位。
返回事件组状态
清空位函数
EventBits_t xEventGroupClearBits( EventGroupHandle_t xEventGroup, const EventBits_t uxBitsToClear )
{
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
EventBits_t uxReturn;
/* Check the user is not attempting to clear the bits used by the kernel
itself. */
configASSERT( xEventGroup );
configASSERT( ( uxBitsToClear & eventEVENT_BITS_CONTROL_BYTES ) == 0 );
taskENTER_CRITICAL();
{
traceEVENT_GROUP_CLEAR_BITS( xEventGroup, uxBitsToClear );
/* The value returned is the event group value prior to the bits being
cleared. */
uxReturn = pxEventBits->uxEventBits;
/* Clear the bits. */
pxEventBits->uxEventBits &= ~uxBitsToClear;
}
taskEXIT_CRITICAL();
return uxReturn;
}
清空事件组中的位,这里先检测一波,防止把内核用的位给清除了。
然后先PV操作,保存之前的值,清除位,结束PV操作
返回清除前的值
获取位函数
EventBits_t xEventGroupSetBits( EventGroupHandle_t xEventGroup, const EventBits_t uxBitsToSet )
{
ListItem_t *pxListItem, *pxNext;
ListItem_t const *pxListEnd;
List_t *pxList;
EventBits_t uxBitsToClear = 0, uxBitsWaitedFor, uxControlBits;
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
BaseType_t xMatchFound = pdFALSE;
/* Check the user is not attempting to set the bits used by the kernel
itself. */
configASSERT( xEventGroup );
configASSERT( ( uxBitsToSet & eventEVENT_BITS_CONTROL_BYTES ) == 0 );
pxList = &( pxEventBits->xTasksWaitingForBits );
pxListEnd = listGET_END_MARKER( pxList ); /*lint !e826 !e740 The mini list structure is used as the list end to save RAM. This is checked and valid. */
vTaskSuspendAll();
{
traceEVENT_GROUP_SET_BITS( xEventGroup, uxBitsToSet );
pxListItem = listGET_HEAD_ENTRY( pxList );
/* Set the bits. */
pxEventBits->uxEventBits |= uxBitsToSet;
/* See if the new bit value should unblock any tasks. */
while( pxListItem != pxListEnd )
{
pxNext = listGET_NEXT( pxListItem );
uxBitsWaitedFor = listGET_LIST_ITEM_VALUE( pxListItem );
xMatchFound = pdFALSE;
/* Split the bits waited for from the control bits. */
uxControlBits = uxBitsWaitedFor & eventEVENT_BITS_CONTROL_BYTES;
uxBitsWaitedFor &= ~eventEVENT_BITS_CONTROL_BYTES;
if( ( uxControlBits & eventWAIT_FOR_ALL_BITS ) == ( EventBits_t ) 0 )
{
/* Just looking for single bit being set. */
if( ( uxBitsWaitedFor & pxEventBits->uxEventBits ) != ( EventBits_t ) 0 )
{
xMatchFound = pdTRUE;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
else if( ( uxBitsWaitedFor & pxEventBits->uxEventBits ) == uxBitsWaitedFor )
{
/* All bits are set. */
xMatchFound = pdTRUE;
}
else
{
/* Need all bits to be set, but not all the bits were set. */
}
if( xMatchFound != pdFALSE )
{
/* The bits match. Should the bits be cleared on exit? */
if( ( uxControlBits & eventCLEAR_EVENTS_ON_EXIT_BIT ) != ( EventBits_t ) 0 )
{
uxBitsToClear |= uxBitsWaitedFor;
}
else
{
mtCOVERAGE_TEST_MARKER();
}
/* Store the actual event flag value in the task's event list
item before removing the task from the event list. The
eventUNBLOCKED_DUE_TO_BIT_SET bit is set so the task knows
that is was unblocked due to its required bits matching, rather
than because it timed out. */
( void ) xTaskRemoveFromUnorderedEventList( pxListItem, pxEventBits->uxEventBits | eventUNBLOCKED_DUE_TO_BIT_SET );
}
/* Move onto the next list item. Note pxListItem->pxNext is not
used here as the list item may have been removed from the event list
and inserted into the ready/pending reading list. */
pxListItem = pxNext;
}
/* Clear any bits that matched when the eventCLEAR_EVENTS_ON_EXIT_BIT
bit was set in the control word. */
pxEventBits->uxEventBits &= ~uxBitsToClear;
}
( void ) xTaskResumeAll();
return pxEventBits->uxEventBits;
}
首先挂起调度,置位
然后开始遍历等待位链表,分离控制位和等待位,检查是否满足了其等待
- 满足了一位等待条件,xMatchFound = pdTRUE;
- 满足所有位等待条件,xMatchFound = pdTRUE;
- 不满足,xMatchFound = pdFalse;
然后检查满足条件的任务,是否需要置位之后清除位
将满足条件的任务从队列里移除,并且更新其被置位后清除的标识
就这样遍历一遍等待置位队列,然后恢复调度,返回。
中断可用获取位函数
EventBits_t xEventGroupGetBitsFromISR( EventGroupHandle_t xEventGroup )
{
UBaseType_t uxSavedInterruptStatus;
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
EventBits_t uxReturn;
uxSavedInterruptStatus = portSET_INTERRUPT_MASK_FROM_ISR();
{
uxReturn = pxEventBits->uxEventBits;
}
portCLEAR_INTERRUPT_MASK_FROM_ISR( uxSavedInterruptStatus );
return uxReturn;
}
简单说要想在中断里调用这个函数,首先需要把中断级别提到最高,防止有其他中断嵌套这里造成意外。
当然这里的最高是系统所能用的最高优先级,而不是硬件最高级。
然后或许需要的位,再下调中断级别
返回位
删除EventGroup
void vEventGroupDelete( EventGroupHandle_t xEventGroup )
{
EventGroup_t *pxEventBits = ( EventGroup_t * ) xEventGroup;
const List_t *pxTasksWaitingForBits = &( pxEventBits->xTasksWaitingForBits );
vTaskSuspendAll();
{
traceEVENT_GROUP_DELETE( xEventGroup );
while( listCURRENT_LIST_LENGTH( pxTasksWaitingForBits ) > ( UBaseType_t ) 0 )
{
/* Unblock the task, returning 0 as the event list is being deleted
and cannot therefore have any bits set. */
configASSERT( pxTasksWaitingForBits->xListEnd.pxNext != ( ListItem_t * ) &( pxTasksWaitingForBits->xListEnd ) );
( void ) xTaskRemoveFromUnorderedEventList( pxTasksWaitingForBits->xListEnd.pxNext, eventUNBLOCKED_DUE_TO_BIT_SET );
}
#if( ( configSUPPORT_DYNAMIC_ALLOCATION == 1 ) && ( configSUPPORT_STATIC_ALLOCATION == 0 ) )
{
/* The event group can only have been allocated dynamically - free
it again. */
vPortFree( pxEventBits );
}
#elif( ( configSUPPORT_DYNAMIC_ALLOCATION == 1 ) && ( configSUPPORT_STATIC_ALLOCATION == 1 ) )
{
/* The event group could have been allocated statically or
dynamically, so check before attempting to free the memory. */
if( pxEventBits->ucStaticallyAllocated == ( uint8_t ) pdFALSE )
{
vPortFree( pxEventBits );
}
else
{
mtCOVERAGE_TEST_MARKER();
}
}
#endif /* configSUPPORT_DYNAMIC_ALLOCATION */
}
( void ) xTaskResumeAll();
}
挂起调度
因为要删除了这个EventGroup,所以对应需要他的置位的所有任务都应该被不阻塞了。
如果是动态申请的,那么用系统的释放函数释放它。
如果动态和静态都可以用的情况下,判断一下是用什么的,然后释放它。
恢复调度
总结
到这里基本就差不多了,整个EventGroup看的其实还是有点懵的,但总体上就这样了,日后发现了什么再补充吧
还有一些小函数,以及中断用函数,就不说了基本上都是这么一个套路。
Quote
event_groups.c
event_groups.h
http://cstriker1407.info/blog/freertos-event-groups/