esp-idf/components/esp_ringbuf/ringbuf.c

1433 lines
65 KiB
C
Raw Normal View History

/*
* SPDX-FileCopyrightText: 2015-2021 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <stdlib.h>
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
#include <string.h>
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "freertos/semphr.h"
#include "freertos/ringbuf.h"
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//32-bit alignment macros
#define rbALIGN_MASK (0x03)
#define rbALIGN_SIZE( xSize ) ( ( xSize + rbALIGN_MASK ) & ~rbALIGN_MASK )
#define rbCHECK_ALIGNED( pvPtr ) ( ( ( UBaseType_t ) ( pvPtr ) & rbALIGN_MASK ) == 0 )
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Ring buffer flags
#define rbALLOW_SPLIT_FLAG ( ( UBaseType_t ) 1 ) //The ring buffer allows items to be split
#define rbBYTE_BUFFER_FLAG ( ( UBaseType_t ) 2 ) //The ring buffer is a byte buffer
#define rbBUFFER_FULL_FLAG ( ( UBaseType_t ) 4 ) //The ring buffer is currently full (write pointer == free pointer)
#define rbBUFFER_STATIC_FLAG ( ( UBaseType_t ) 8 ) //The ring buffer is statically allocated
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Item flags
#define rbITEM_FREE_FLAG ( ( UBaseType_t ) 1 ) //Item has been retrieved and returned by application, free to overwrite
#define rbITEM_DUMMY_DATA_FLAG ( ( UBaseType_t ) 2 ) //Data from here to end of the ring buffer is dummy data. Restart reading at start of head of the buffer
#define rbITEM_SPLIT_FLAG ( ( UBaseType_t ) 4 ) //Valid for RINGBUF_TYPE_ALLOWSPLIT, indicating that rest of the data is wrapped around
#define rbITEM_WRITTEN_FLAG ( ( UBaseType_t ) 8 ) //Item has been written to by the application, thus can be read
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Static allocation related
#if ( configSUPPORT_STATIC_ALLOCATION == 1 )
#define rbGET_TX_SEM_HANDLE( pxRingbuffer ) ( (SemaphoreHandle_t) &(pxRingbuffer->xTransSemStatic) )
#define rbGET_RX_SEM_HANDLE( pxRingbuffer ) ( (SemaphoreHandle_t) &(pxRingbuffer->xRecvSemStatic) )
#else
#define rbGET_TX_SEM_HANDLE( pxRingbuffer ) ( pxRingbuffer->xTransSemHandle )
#define rbGET_RX_SEM_HANDLE( pxRingbuffer ) ( pxRingbuffer->xRecvSemHandle )
#endif
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
typedef struct {
//This size of this structure must be 32-bit aligned
size_t xItemLen;
UBaseType_t uxItemFlags;
} ItemHeader_t;
#define rbHEADER_SIZE sizeof(ItemHeader_t)
typedef struct RingbufferDefinition Ringbuffer_t;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
typedef BaseType_t (*CheckItemFitsFunction_t)(Ringbuffer_t *pxRingbuffer, size_t xItemSize);
typedef void (*CopyItemFunction_t)(Ringbuffer_t *pxRingbuffer, const uint8_t *pcItem, size_t xItemSize);
typedef BaseType_t (*CheckItemAvailFunction_t) (Ringbuffer_t *pxRingbuffer);
typedef void *(*GetItemFunction_t)(Ringbuffer_t *pxRingbuffer, BaseType_t *pxIsSplit, size_t xMaxSize, size_t *pxItemSize);
typedef void (*ReturnItemFunction_t)(Ringbuffer_t *pxRingbuffer, uint8_t *pvItem);
typedef size_t (*GetCurMaxSizeFunction_t)(Ringbuffer_t *pxRingbuffer);
typedef struct RingbufferDefinition {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xSize; //Size of the data storage
size_t xMaxItemSize; //Maximum item size
UBaseType_t uxRingbufferFlags; //Flags to indicate the type and status of ring buffer
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
CheckItemFitsFunction_t xCheckItemFits; //Function to check if item can currently fit in ring buffer
CopyItemFunction_t vCopyItem; //Function to copy item to ring buffer
GetItemFunction_t pvGetItem; //Function to get item from ring buffer
ReturnItemFunction_t vReturnItem; //Function to return item to ring buffer
GetCurMaxSizeFunction_t xGetCurMaxSize; //Function to get current free size
2019-06-14 02:10:48 -04:00
uint8_t *pucAcquire; //Acquire Pointer. Points to where the next item should be acquired.
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
uint8_t *pucWrite; //Write Pointer. Points to where the next item should be written
uint8_t *pucRead; //Read Pointer. Points to where the next item should be read from
uint8_t *pucFree; //Free Pointer. Points to the last item that has yet to be returned to the ring buffer
uint8_t *pucHead; //Pointer to the start of the ring buffer storage area
uint8_t *pucTail; //Pointer to the end of the ring buffer storage area
BaseType_t xItemsWaiting; //Number of items/bytes(for byte buffers) currently in ring buffer that have not yet been read
/*
* TransSem: Binary semaphore used to indicate to a blocked transmitting tasks
* that more free space has become available or that the block has
* timed out.
*
* RecvSem: Binary semaphore used to indicate to a blocked receiving task that
* new data/item has been written to the ring buffer.
*
* Note - When static allocation is enabled, the two semaphores are always
* statically stored in the ring buffer's control structure
* regardless of whether the ring buffer is allocated dynamically or
* statically. When static allocation is disabled, the two semaphores
* are allocated dynamically and their handles stored instead, thus
* making the ring buffer's control structure slightly smaller when
* static allocation is disabled.
*/
#if ( configSUPPORT_STATIC_ALLOCATION == 1 )
StaticSemaphore_t xTransSemStatic;
StaticSemaphore_t xRecvSemStatic;
#else
SemaphoreHandle_t xTransSemHandle;
SemaphoreHandle_t xRecvSemHandle;
#endif
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
portMUX_TYPE mux; //Spinlock required for SMP
} Ringbuffer_t;
#if ( configSUPPORT_STATIC_ALLOCATION == 1 )
#if __GNUC_PREREQ(4, 6)
_Static_assert(sizeof(StaticRingbuffer_t) == sizeof(Ringbuffer_t), "StaticRingbuffer_t != Ringbuffer_t");
#endif
#endif
/*
Remark: A counting semaphore for items_buffered_sem would be more logical, but counting semaphores in
FreeRTOS need a maximum count, and allocate more memory the larger the maximum count is. Here, we
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
would need to set the maximum to the maximum amount of times a null-byte unit first in the buffer,
which is quite high and so would waste a fair amount of memory.
*/
/* --------------------------- Static Declarations -------------------------- */
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
/*
* WARNING: All of the following static functions (except generic functions)
* ARE NOT THREAD SAFE. Therefore they should only be called within a critical
* section (using spin locks)
*/
//Initialize a ring buffer after space has been allocated for it
static void prvInitializeNewRingbuffer(size_t xBufferSize,
RingbufferType_t xBufferType,
Ringbuffer_t *pxNewRingbuffer,
uint8_t *pucRingbufferStorage);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Calculate current amount of free space (in bytes) in the ring buffer
static size_t prvGetFreeSize(Ringbuffer_t *pxRingbuffer);
//Checks if an item/data is currently available for retrieval
static BaseType_t prvCheckItemAvail(Ringbuffer_t *pxRingbuffer);
//Checks if an item will currently fit in a no-split/allow-split ring buffer
static BaseType_t prvCheckItemFitsDefault( Ringbuffer_t *pxRingbuffer, size_t xItemSize);
//Checks if an item will currently fit in a byte buffer
static BaseType_t prvCheckItemFitsByteBuffer( Ringbuffer_t *pxRingbuffer, size_t xItemSize);
/*
Copies an item to a no-split ring buffer
Entry:
- Must have already guaranteed there is sufficient space for item by calling prvCheckItemFitsDefault()
Exit:
- New item copied into ring buffer
- pucAcquire and pucWrite updated.
- Dummy item added if necessary
*/
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvCopyItemNoSplit(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize);
/*
Copies an item to a allow-split ring buffer
Entry:
- Must have already guaranteed there is sufficient space for item by calling prvCheckItemFitsDefault()
Exit:
- New item copied into ring buffer
- pucAcquire and pucWrite updated
- Item may be split
*/
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvCopyItemAllowSplit(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize);
//Copies an item to a byte buffer. Only call this function after calling prvCheckItemFitsByteBuffer()
static void prvCopyItemByteBuf(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize);
//Retrieve item from no-split/allow-split ring buffer. *pxIsSplit is set to pdTRUE if the retrieved item is split
/*
Entry:
- Must have already guaranteed that there is an item available for retrieval by calling prvCheckItemAvail()
- Guaranteed that pucREAD points to a valid item (i.e., not a dummy item)
Exit:
- Item is returned. Only first half returned if split
- pucREAD updated to point to next valid item to read, or equals to pucWrite if there are no more valid items to read
- pucREAD update must skip over dummy items
*/
static void *prvGetItemDefault(Ringbuffer_t *pxRingbuffer,
BaseType_t *pxIsSplit,
size_t xUnusedParam,
size_t *pxItemSize);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Retrieve data from byte buffer. If xMaxSize is 0, all continuous data is retrieved
static void *prvGetItemByteBuf(Ringbuffer_t *pxRingbuffer,
BaseType_t *pxUnusedParam,
size_t xMaxSize,
size_t *pxItemSize);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
/*
Return an item to a split/no-split ring buffer
Exit:
- Item is marked free rbITEM_FREE_FLAG
- pucFree is progressed as far as possible, skipping over already freed items or dummy items
*/
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvReturnItemDefault(Ringbuffer_t *pxRingbuffer, uint8_t *pucItem);
//Return data to a byte buffer
static void prvReturnItemByteBuf(Ringbuffer_t *pxRingbuffer, uint8_t *pucItem);
//Get the maximum size an item that can currently have if sent to a no-split ring buffer
static size_t prvGetCurMaxSizeNoSplit(Ringbuffer_t *pxRingbuffer);
//Get the maximum size an item that can currently have if sent to a allow-split ring buffer
static size_t prvGetCurMaxSizeAllowSplit(Ringbuffer_t *pxRingbuffer);
//Get the maximum size an item that can currently have if sent to a byte buffer
static size_t prvGetCurMaxSizeByteBuf(Ringbuffer_t *pxRingbuffer);
/**
* Generic function used to retrieve an item/data from ring buffers. If called on
* an allow-split buffer, and pvItem2 and xItemSize2 are not NULL, both parts of
* a split item will be retrieved. xMaxSize will only take effect if called on
* byte buffers.
*/
static BaseType_t prvReceiveGeneric(Ringbuffer_t *pxRingbuffer,
void **pvItem1,
void **pvItem2,
size_t *xItemSize1,
size_t *xItemSize2,
size_t xMaxSize,
TickType_t xTicksToWait);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Generic function used to retrieve an item/data from ring buffers in an ISR
static BaseType_t prvReceiveGenericFromISR(Ringbuffer_t *pxRingbuffer,
void **pvItem1,
void **pvItem2,
size_t *xItemSize1,
size_t *xItemSize2,
size_t xMaxSize);
/* --------------------------- Static Definitions --------------------------- */
static void prvInitializeNewRingbuffer(size_t xBufferSize,
RingbufferType_t xBufferType,
Ringbuffer_t *pxNewRingbuffer,
uint8_t *pucRingbufferStorage)
{
//Initialize values
pxNewRingbuffer->xSize = xBufferSize;
pxNewRingbuffer->pucHead = pucRingbufferStorage;
pxNewRingbuffer->pucTail = pucRingbufferStorage + xBufferSize;
pxNewRingbuffer->pucFree = pucRingbufferStorage;
pxNewRingbuffer->pucRead = pucRingbufferStorage;
pxNewRingbuffer->pucWrite = pucRingbufferStorage;
2019-06-14 02:10:48 -04:00
pxNewRingbuffer->pucAcquire = pucRingbufferStorage;
pxNewRingbuffer->xItemsWaiting = 0;
pxNewRingbuffer->uxRingbufferFlags = 0;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Initialize type dependent values and function pointers
if (xBufferType == RINGBUF_TYPE_NOSPLIT) {
pxNewRingbuffer->xCheckItemFits = prvCheckItemFitsDefault;
pxNewRingbuffer->vCopyItem = prvCopyItemNoSplit;
pxNewRingbuffer->pvGetItem = prvGetItemDefault;
pxNewRingbuffer->vReturnItem = prvReturnItemDefault;
/*
2019-06-14 02:10:48 -04:00
* Worst case scenario is when the read/write/acquire/free pointers are all
* pointing to the halfway point of the buffer.
*/
pxNewRingbuffer->xMaxItemSize = rbALIGN_SIZE(pxNewRingbuffer->xSize / 2) - rbHEADER_SIZE;
pxNewRingbuffer->xGetCurMaxSize = prvGetCurMaxSizeNoSplit;
} else if (xBufferType == RINGBUF_TYPE_ALLOWSPLIT) {
pxNewRingbuffer->uxRingbufferFlags |= rbALLOW_SPLIT_FLAG;
pxNewRingbuffer->xCheckItemFits = prvCheckItemFitsDefault;
pxNewRingbuffer->vCopyItem = prvCopyItemAllowSplit;
pxNewRingbuffer->pvGetItem = prvGetItemDefault;
pxNewRingbuffer->vReturnItem = prvReturnItemDefault;
//Worst case an item is split into two, incurring two headers of overhead
pxNewRingbuffer->xMaxItemSize = pxNewRingbuffer->xSize - (sizeof(ItemHeader_t) * 2);
pxNewRingbuffer->xGetCurMaxSize = prvGetCurMaxSizeAllowSplit;
} else { //Byte Buffer
pxNewRingbuffer->uxRingbufferFlags |= rbBYTE_BUFFER_FLAG;
pxNewRingbuffer->xCheckItemFits = prvCheckItemFitsByteBuffer;
pxNewRingbuffer->vCopyItem = prvCopyItemByteBuf;
pxNewRingbuffer->pvGetItem = prvGetItemByteBuf;
pxNewRingbuffer->vReturnItem = prvReturnItemByteBuf;
//Byte buffers do not incur any overhead
pxNewRingbuffer->xMaxItemSize = pxNewRingbuffer->xSize;
pxNewRingbuffer->xGetCurMaxSize = prvGetCurMaxSizeByteBuf;
}
xSemaphoreGive(rbGET_TX_SEM_HANDLE(pxNewRingbuffer));
portMUX_INITIALIZE(&pxNewRingbuffer->mux);
}
2016-10-24 09:18:02 -04:00
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static size_t prvGetFreeSize(Ringbuffer_t *pxRingbuffer)
{
size_t xReturn;
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
xReturn = 0;
2016-10-24 09:18:02 -04:00
} else {
2019-06-14 02:10:48 -04:00
BaseType_t xFreeSize = pxRingbuffer->pucFree - pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check if xFreeSize has underflowed
if (xFreeSize <= 0) {
xFreeSize += pxRingbuffer->xSize;
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xReturn = xFreeSize;
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
configASSERT(xReturn <= pxRingbuffer->xSize);
return xReturn;
}
2016-10-24 09:18:02 -04:00
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static BaseType_t prvCheckItemFitsDefault( Ringbuffer_t *pxRingbuffer, size_t xItemSize)
{
//Check arguments and buffer state
2019-06-14 02:10:48 -04:00
configASSERT(rbCHECK_ALIGNED(pxRingbuffer->pucAcquire)); //pucAcquire is always aligned in no-split/allow-split ring buffers
configASSERT(pxRingbuffer->pucAcquire >= pxRingbuffer->pucHead && pxRingbuffer->pucAcquire < pxRingbuffer->pucTail); //Check write pointer is within bounds
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xTotalItemSize = rbALIGN_SIZE(xItemSize) + rbHEADER_SIZE; //Rounded up aligned item size with header
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucFree) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Buffer is either complete empty or completely full
return (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) ? pdFALSE : pdTRUE;
}
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucFree > pxRingbuffer->pucAcquire) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Free space does not wrap around
2019-06-14 02:10:48 -04:00
return (xTotalItemSize <= pxRingbuffer->pucFree - pxRingbuffer->pucAcquire) ? pdTRUE : pdFALSE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Free space wraps around
2019-06-14 02:10:48 -04:00
if (xTotalItemSize <= pxRingbuffer->pucTail - pxRingbuffer->pucAcquire) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return pdTRUE; //Item fits without wrapping around
}
//Check if item fits by wrapping
if (pxRingbuffer->uxRingbufferFlags & rbALLOW_SPLIT_FLAG) {
//Allow split wrapping incurs an extra header
2019-06-14 02:10:48 -04:00
return (xTotalItemSize + rbHEADER_SIZE <= pxRingbuffer->xSize - (pxRingbuffer->pucAcquire - pxRingbuffer->pucFree)) ? pdTRUE : pdFALSE;
} else {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return (xTotalItemSize <= pxRingbuffer->pucFree - pxRingbuffer->pucHead) ? pdTRUE : pdFALSE;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static BaseType_t prvCheckItemFitsByteBuffer( Ringbuffer_t *pxRingbuffer, size_t xItemSize)
{
//Check arguments and buffer state
2019-06-14 02:10:48 -04:00
configASSERT(pxRingbuffer->pucAcquire >= pxRingbuffer->pucHead && pxRingbuffer->pucAcquire < pxRingbuffer->pucTail); //Check acquire pointer is within bounds
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucFree) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Buffer is either complete empty or completely full
return (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) ? pdFALSE : pdTRUE;
}
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucFree > pxRingbuffer->pucAcquire) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Free space does not wrap around
2019-06-14 02:10:48 -04:00
return (xItemSize <= pxRingbuffer->pucFree - pxRingbuffer->pucAcquire) ? pdTRUE : pdFALSE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Free space wraps around
2019-06-14 02:10:48 -04:00
return (xItemSize <= pxRingbuffer->xSize - (pxRingbuffer->pucAcquire - pxRingbuffer->pucFree)) ? pdTRUE : pdFALSE;
}
2019-06-14 02:10:48 -04:00
static uint8_t* prvAcquireItemNoSplit(Ringbuffer_t *pxRingbuffer, size_t xItemSize)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
//Check arguments and buffer state
size_t xAlignedItemSize = rbALIGN_SIZE(xItemSize); //Rounded up aligned item size
2019-06-14 02:10:48 -04:00
size_t xRemLen = pxRingbuffer->pucTail - pxRingbuffer->pucAcquire; //Length from pucAcquire until end of buffer
configASSERT(rbCHECK_ALIGNED(pxRingbuffer->pucAcquire)); //pucAcquire is always aligned in no-split ring buffers
configASSERT(pxRingbuffer->pucAcquire >= pxRingbuffer->pucHead && pxRingbuffer->pucAcquire < pxRingbuffer->pucTail); //Check write pointer is within bounds
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
configASSERT(xRemLen >= rbHEADER_SIZE); //Remaining length must be able to at least fit an item header
//If remaining length can't fit item, set as dummy data and wrap around
if (xRemLen < xAlignedItemSize + rbHEADER_SIZE) {
2019-06-14 02:10:48 -04:00
ItemHeader_t *pxDummy = (ItemHeader_t *)pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxDummy->uxItemFlags = rbITEM_DUMMY_DATA_FLAG; //Set remaining length as dummy data
pxDummy->xItemLen = 0; //Dummy data should have no length
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead; //Reset acquire pointer to wrap around
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Item should be guaranteed to fit at this point. Set item header and copy data
2019-06-14 02:10:48 -04:00
ItemHeader_t *pxHeader = (ItemHeader_t *)pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxHeader->xItemLen = xItemSize;
pxHeader->uxItemFlags = 0;
2019-06-14 02:10:48 -04:00
//hold the buffer address without touching pucWrite
uint8_t* item_address = pxRingbuffer->pucAcquire + rbHEADER_SIZE;
pxRingbuffer->pucAcquire += rbHEADER_SIZE + xAlignedItemSize; //Advance pucAcquire past header and the item to next aligned address
//After the allocation, add some padding after the buffer and correct the flags
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//If current remaining length can't fit a header, wrap around write pointer
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucTail - pxRingbuffer->pucAcquire < rbHEADER_SIZE) {
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead; //Wrap around pucAcquire
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Check if buffer is full
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucFree) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Mark the buffer as full to distinguish with an empty buffer
pxRingbuffer->uxRingbufferFlags |= rbBUFFER_FULL_FLAG;
}
2019-06-14 02:10:48 -04:00
return item_address;
}
static void prvSendItemDoneNoSplit(Ringbuffer_t *pxRingbuffer, uint8_t* pucItem)
{
//Check arguments and buffer state
configASSERT(rbCHECK_ALIGNED(pucItem));
configASSERT(pucItem >= pxRingbuffer->pucHead);
configASSERT(pucItem <= pxRingbuffer->pucTail); //Inclusive of pucTail in the case of zero length item at the very end
//Get and check header of the item
ItemHeader_t *pxCurHeader = (ItemHeader_t *)(pucItem - rbHEADER_SIZE);
configASSERT(pxCurHeader->xItemLen <= pxRingbuffer->xMaxItemSize);
configASSERT((pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG) == 0); //Dummy items should never have been written
configASSERT((pxCurHeader->uxItemFlags & rbITEM_WRITTEN_FLAG) == 0); //Indicates item has already been written before
pxCurHeader->uxItemFlags &= ~rbITEM_SPLIT_FLAG; //Clear wrap flag if set (not strictly necessary)
pxCurHeader->uxItemFlags |= rbITEM_WRITTEN_FLAG; //Mark as written
pxRingbuffer->xItemsWaiting++;
/*
* Items might not be written in the order they were acquired. Move the
* write pointer up to the next item that has not been marked as written (by
* written flag) or up till the acquire pointer. When advancing the write
* pointer, items that have already been written or items with dummy data
* should be skipped over
*/
pxCurHeader = (ItemHeader_t *)pxRingbuffer->pucWrite;
//Skip over Items that have already been written or are dummy items
while (((pxCurHeader->uxItemFlags & rbITEM_WRITTEN_FLAG) || (pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG)) && pxRingbuffer->pucWrite != pxRingbuffer->pucAcquire) {
if (pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG) {
pxCurHeader->uxItemFlags |= rbITEM_WRITTEN_FLAG; //Mark as freed (not strictly necessary but adds redundancy)
pxRingbuffer->pucWrite = pxRingbuffer->pucHead; //Wrap around due to dummy data
} else {
//Item with data that has already been written, advance write pointer past this item
size_t xAlignedItemSize = rbALIGN_SIZE(pxCurHeader->xItemLen);
pxRingbuffer->pucWrite += xAlignedItemSize + rbHEADER_SIZE;
//Redundancy check to ensure write pointer has not overshot buffer bounds
configASSERT(pxRingbuffer->pucWrite <= pxRingbuffer->pucHead + pxRingbuffer->xSize);
}
//Check if pucWrite requires wrap around
2019-06-14 02:10:48 -04:00
if ((pxRingbuffer->pucTail - pxRingbuffer->pucWrite) < rbHEADER_SIZE) {
pxRingbuffer->pucWrite = pxRingbuffer->pucHead;
}
pxCurHeader = (ItemHeader_t *)pxRingbuffer->pucWrite; //Update header to point to item
}
}
static void prvCopyItemNoSplit(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize)
{
uint8_t* item_addr = prvAcquireItemNoSplit(pxRingbuffer, xItemSize);
memcpy(item_addr, pucItem, xItemSize);
prvSendItemDoneNoSplit(pxRingbuffer, item_addr);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
2016-10-24 09:18:02 -04:00
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvCopyItemAllowSplit(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize)
2016-10-24 09:18:02 -04:00
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments and buffer state
size_t xAlignedItemSize = rbALIGN_SIZE(xItemSize); //Rounded up aligned item size
2019-06-14 02:10:48 -04:00
size_t xRemLen = pxRingbuffer->pucTail - pxRingbuffer->pucAcquire; //Length from pucAcquire until end of buffer
configASSERT(rbCHECK_ALIGNED(pxRingbuffer->pucAcquire)); //pucAcquire is always aligned in split ring buffers
configASSERT(pxRingbuffer->pucAcquire >= pxRingbuffer->pucHead && pxRingbuffer->pucAcquire < pxRingbuffer->pucTail); //Check write pointer is within bounds
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
configASSERT(xRemLen >= rbHEADER_SIZE); //Remaining length must be able to at least fit an item header
//Split item if necessary
if (xRemLen < xAlignedItemSize + rbHEADER_SIZE) {
//Write first part of the item
2019-06-14 02:10:48 -04:00
ItemHeader_t *pxFirstHeader = (ItemHeader_t *)pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxFirstHeader->uxItemFlags = 0;
pxFirstHeader->xItemLen = xRemLen - rbHEADER_SIZE; //Fill remaining length with first part
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire += rbHEADER_SIZE; //Advance pucAcquire past header
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xRemLen -= rbHEADER_SIZE;
if (xRemLen > 0) {
2019-06-14 02:10:48 -04:00
memcpy(pxRingbuffer->pucAcquire, pucItem, xRemLen);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->xItemsWaiting++;
//Update item arguments to account for data already copied
pucItem += xRemLen;
xItemSize -= xRemLen;
xAlignedItemSize -= xRemLen;
pxFirstHeader->uxItemFlags |= rbITEM_SPLIT_FLAG; //There must be more data
} else {
//Remaining length was only large enough to fit header
pxFirstHeader->uxItemFlags |= rbITEM_DUMMY_DATA_FLAG; //Item will completely be stored in 2nd part
}
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead; //Reset acquire pointer to start of buffer
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Item (whole or second part) should be guaranteed to fit at this point
2019-06-14 02:10:48 -04:00
ItemHeader_t *pxSecondHeader = (ItemHeader_t *)pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxSecondHeader->xItemLen = xItemSize;
pxSecondHeader->uxItemFlags = 0;
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire += rbHEADER_SIZE; //Advance acquire pointer past header
memcpy(pxRingbuffer->pucAcquire, pucItem, xItemSize);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->xItemsWaiting++;
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire += xAlignedItemSize; //Advance pucAcquire past item to next aligned address
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//If current remaining length can't fit a header, wrap around write pointer
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucTail - pxRingbuffer->pucAcquire < rbHEADER_SIZE) {
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead; //Wrap around pucAcquire
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Check if buffer is full
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucFree) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Mark the buffer as full to distinguish with an empty buffer
pxRingbuffer->uxRingbufferFlags |= rbBUFFER_FULL_FLAG;
}
2019-06-14 02:10:48 -04:00
//currently the Split mode is not supported, pucWrite tracks the pucAcquire
pxRingbuffer->pucWrite = pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvCopyItemByteBuf(Ringbuffer_t *pxRingbuffer, const uint8_t *pucItem, size_t xItemSize)
{
//Check arguments and buffer state
2019-06-14 02:10:48 -04:00
configASSERT(pxRingbuffer->pucAcquire >= pxRingbuffer->pucHead && pxRingbuffer->pucAcquire < pxRingbuffer->pucTail); //Check acquire pointer is within bounds
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
2019-06-14 02:10:48 -04:00
size_t xRemLen = pxRingbuffer->pucTail - pxRingbuffer->pucAcquire; //Length from pucAcquire until end of buffer
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xRemLen < xItemSize) {
//Copy as much as possible into remaining length
2019-06-14 02:10:48 -04:00
memcpy(pxRingbuffer->pucAcquire, pucItem, xRemLen);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->xItemsWaiting += xRemLen;
//Update item arguments to account for data already written
pucItem += xRemLen;
xItemSize -= xRemLen;
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead; //Reset acquire pointer to start of buffer
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Copy all or remaining portion of the item
2019-06-14 02:10:48 -04:00
memcpy(pxRingbuffer->pucAcquire, pucItem, xItemSize);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->xItemsWaiting += xItemSize;
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucAcquire += xItemSize;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
2019-06-14 02:10:48 -04:00
//Wrap around pucAcquire if it reaches the end
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucTail) {
pxRingbuffer->pucAcquire = pxRingbuffer->pucHead;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
//Check if buffer is full
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucFree) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->uxRingbufferFlags |= rbBUFFER_FULL_FLAG; //Mark the buffer as full to avoid confusion with an empty buffer
2016-10-24 09:18:02 -04:00
}
2019-06-14 02:10:48 -04:00
//Currently, acquiring memory is not supported in byte mode. pucWrite tracks the pucAcquire.
pxRingbuffer->pucWrite = pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
2016-10-24 09:18:02 -04:00
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static BaseType_t prvCheckItemAvail(Ringbuffer_t *pxRingbuffer)
{
if ((pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) && pxRingbuffer->pucRead != pxRingbuffer->pucFree) {
return pdFALSE; //Byte buffers do not allow multiple retrievals before return
}
if ((pxRingbuffer->xItemsWaiting > 0) && ((pxRingbuffer->pucRead != pxRingbuffer->pucWrite) || (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG))) {
return pdTRUE; //Items/data available for retrieval
} else {
return pdFALSE; //No items/data available for retrieval
2016-10-24 09:18:02 -04:00
}
}
static void *prvGetItemDefault(Ringbuffer_t *pxRingbuffer,
BaseType_t *pxIsSplit,
size_t xUnusedParam,
size_t *pxItemSize)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments and buffer state
ItemHeader_t *pxHeader = (ItemHeader_t *)pxRingbuffer->pucRead;
configASSERT(pxIsSplit != NULL);
configASSERT((pxRingbuffer->xItemsWaiting > 0) && ((pxRingbuffer->pucRead != pxRingbuffer->pucWrite) || (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG))); //Check there are items to be read
configASSERT(rbCHECK_ALIGNED(pxRingbuffer->pucRead)); //pucRead is always aligned in split ring buffers
configASSERT(pxRingbuffer->pucRead >= pxRingbuffer->pucHead && pxRingbuffer->pucRead < pxRingbuffer->pucTail); //Check read pointer is within bounds
configASSERT((pxHeader->xItemLen <= pxRingbuffer->xMaxItemSize) || (pxHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG));
uint8_t *pcReturn;
//Wrap around if dummy data (dummy data indicates wrap around in no-split buffers)
if (pxHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG) {
pxRingbuffer->pucRead = pxRingbuffer->pucHead;
//Check for errors with the next item
pxHeader = (ItemHeader_t *)pxRingbuffer->pucRead;
configASSERT(pxHeader->xItemLen <= pxRingbuffer->xMaxItemSize);
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pcReturn = pxRingbuffer->pucRead + rbHEADER_SIZE; //Get pointer to part of item containing data (point past the header)
if (pxHeader->xItemLen == 0) {
//Inclusive of pucTail for special case where item of zero length just fits at the end of the buffer
configASSERT(pcReturn >= pxRingbuffer->pucHead && pcReturn <= pxRingbuffer->pucTail);
} else {
//Exclusive of pucTail if length is larger than zero, pcReturn should never point to pucTail
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
configASSERT(pcReturn >= pxRingbuffer->pucHead && pcReturn < pxRingbuffer->pucTail);
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
*pxItemSize = pxHeader->xItemLen; //Get length of item
pxRingbuffer->xItemsWaiting --; //Update item count
*pxIsSplit = (pxHeader->uxItemFlags & rbITEM_SPLIT_FLAG) ? pdTRUE : pdFALSE;
pxRingbuffer->pucRead += rbHEADER_SIZE + rbALIGN_SIZE(pxHeader->xItemLen); //Update pucRead
//Check if pucRead requires wrap around
if ((pxRingbuffer->pucTail - pxRingbuffer->pucRead) < rbHEADER_SIZE) {
pxRingbuffer->pucRead = pxRingbuffer->pucHead;
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return (void *)pcReturn;
}
static void *prvGetItemByteBuf(Ringbuffer_t *pxRingbuffer,
BaseType_t *pxUnusedParam,
size_t xMaxSize,
size_t *pxItemSize)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
//Check arguments and buffer state
configASSERT((pxRingbuffer->xItemsWaiting > 0) && ((pxRingbuffer->pucRead != pxRingbuffer->pucWrite) || (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG))); //Check there are items to be read
configASSERT(pxRingbuffer->pucRead >= pxRingbuffer->pucHead && pxRingbuffer->pucRead < pxRingbuffer->pucTail); //Check read pointer is within bounds
configASSERT(pxRingbuffer->pucRead == pxRingbuffer->pucFree);
uint8_t *ret = pxRingbuffer->pucRead;
if ((pxRingbuffer->pucRead > pxRingbuffer->pucWrite) || (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG)) { //Available data wraps around
//Return contiguous piece from read pointer until buffer tail, or xMaxSize
if (xMaxSize == 0 || pxRingbuffer->pucTail - pxRingbuffer->pucRead <= xMaxSize) {
//All contiguous data from read pointer to tail
*pxItemSize = pxRingbuffer->pucTail - pxRingbuffer->pucRead;
pxRingbuffer->xItemsWaiting -= pxRingbuffer->pucTail - pxRingbuffer->pucRead;
pxRingbuffer->pucRead = pxRingbuffer->pucHead; //Wrap around read pointer
2016-10-24 09:18:02 -04:00
} else {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Return xMaxSize amount of data
*pxItemSize = xMaxSize;
pxRingbuffer->xItemsWaiting -= xMaxSize;
pxRingbuffer->pucRead += xMaxSize; //Advance read pointer past retrieved data
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
} else { //Available data is contiguous between read and write pointer
if (xMaxSize == 0 || pxRingbuffer->pucWrite - pxRingbuffer->pucRead <= xMaxSize) {
//Return all contiguous data from read to write pointer
*pxItemSize = pxRingbuffer->pucWrite - pxRingbuffer->pucRead;
pxRingbuffer->xItemsWaiting -= pxRingbuffer->pucWrite - pxRingbuffer->pucRead;
pxRingbuffer->pucRead = pxRingbuffer->pucWrite;
2016-10-24 09:18:02 -04:00
} else {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Return xMaxSize data from read pointer
*pxItemSize = xMaxSize;
pxRingbuffer->xItemsWaiting -= xMaxSize;
pxRingbuffer->pucRead += xMaxSize; //Advance read pointer past retrieved data
2016-10-24 09:18:02 -04:00
}
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return (void *)ret;
}
static void prvReturnItemDefault(Ringbuffer_t *pxRingbuffer, uint8_t *pucItem)
{
//Check arguments and buffer state
configASSERT(rbCHECK_ALIGNED(pucItem));
configASSERT(pucItem >= pxRingbuffer->pucHead);
configASSERT(pucItem <= pxRingbuffer->pucTail); //Inclusive of pucTail in the case of zero length item at the very end
//Get and check header of the item
ItemHeader_t *pxCurHeader = (ItemHeader_t *)(pucItem - rbHEADER_SIZE);
configASSERT(pxCurHeader->xItemLen <= pxRingbuffer->xMaxItemSize);
configASSERT((pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG) == 0); //Dummy items should never have been read
configASSERT((pxCurHeader->uxItemFlags & rbITEM_FREE_FLAG) == 0); //Indicates item has already been returned before
pxCurHeader->uxItemFlags &= ~rbITEM_SPLIT_FLAG; //Clear wrap flag if set (not strictly necessary)
pxCurHeader->uxItemFlags |= rbITEM_FREE_FLAG; //Mark as free
/*
* Items might not be returned in the order they were retrieved. Move the free pointer
* up to the next item that has not been marked as free (by free flag) or up
* till the read pointer. When advancing the free pointer, items that have already been
* freed or items with dummy data should be skipped over
*/
pxCurHeader = (ItemHeader_t *)pxRingbuffer->pucFree;
//Skip over Items that have already been freed or are dummy items
while (((pxCurHeader->uxItemFlags & rbITEM_FREE_FLAG) || (pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG)) && pxRingbuffer->pucFree != pxRingbuffer->pucRead) {
if (pxCurHeader->uxItemFlags & rbITEM_DUMMY_DATA_FLAG) {
pxCurHeader->uxItemFlags |= rbITEM_FREE_FLAG; //Mark as freed (not strictly necessary but adds redundancy)
pxRingbuffer->pucFree = pxRingbuffer->pucHead; //Wrap around due to dummy data
} else {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Item with data that has already been freed, advance free pointer past this item
size_t xAlignedItemSize = rbALIGN_SIZE(pxCurHeader->xItemLen);
pxRingbuffer->pucFree += xAlignedItemSize + rbHEADER_SIZE;
//Redundancy check to ensure free pointer has not overshot buffer bounds
configASSERT(pxRingbuffer->pucFree <= pxRingbuffer->pucHead + pxRingbuffer->xSize);
}
//Check if pucFree requires wrap around
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if ((pxRingbuffer->pucTail - pxRingbuffer->pucFree) < rbHEADER_SIZE) {
pxRingbuffer->pucFree = pxRingbuffer->pucHead;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxCurHeader = (ItemHeader_t *)pxRingbuffer->pucFree; //Update header to point to item
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check if the buffer full flag should be reset
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucFree != pxRingbuffer->pucAcquire) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->uxRingbufferFlags &= ~rbBUFFER_FULL_FLAG;
2019-06-14 02:10:48 -04:00
} else if (pxRingbuffer->pucFree == pxRingbuffer->pucAcquire && pxRingbuffer->pucFree == pxRingbuffer->pucRead) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Special case where a full buffer is completely freed in one go
pxRingbuffer->uxRingbufferFlags &= ~rbBUFFER_FULL_FLAG;
}
}
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static void prvReturnItemByteBuf(Ringbuffer_t *pxRingbuffer, uint8_t *pucItem)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check pointer points to address inside buffer
configASSERT((uint8_t *)pucItem >= pxRingbuffer->pucHead);
configASSERT((uint8_t *)pucItem < pxRingbuffer->pucTail);
//Free the read memory. Simply moves free pointer to read pointer as byte buffers do not allow multiple outstanding reads
pxRingbuffer->pucFree = pxRingbuffer->pucRead;
//If buffer was full before, reset full flag as free pointer has moved
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
pxRingbuffer->uxRingbufferFlags &= ~rbBUFFER_FULL_FLAG;
}
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static size_t prvGetCurMaxSizeNoSplit(Ringbuffer_t *pxRingbuffer)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xFreeSize;
//Check if buffer is full
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
return 0;
}
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire < pxRingbuffer->pucFree) {
//Free space is contiguous between pucAcquire and pucFree
xFreeSize = pxRingbuffer->pucFree - pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
} else {
//Free space wraps around (or overlapped at pucHead), select largest
//contiguous free space as no-split items require contiguous space
2019-06-14 02:10:48 -04:00
size_t xSize1 = pxRingbuffer->pucTail - pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xSize2 = pxRingbuffer->pucFree - pxRingbuffer->pucHead;
xFreeSize = (xSize1 > xSize2) ? xSize1 : xSize2;
}
//No-split ring buffer items need space for a header
xFreeSize -= rbHEADER_SIZE;
//Check for xFreeSize < 0 before checking xFreeSize > pxRingbuffer->xMaxItemSize
//to avoid incorrect comparison operation when xFreeSize is negative
if (xFreeSize < 0) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Occurs when free space is less than header size
xFreeSize = 0;
} else if (xFreeSize > pxRingbuffer->xMaxItemSize) {
//Limit free size to be within bounds
xFreeSize = pxRingbuffer->xMaxItemSize;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xFreeSize;
}
2016-10-24 09:18:02 -04:00
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static size_t prvGetCurMaxSizeAllowSplit(Ringbuffer_t *pxRingbuffer)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xFreeSize;
//Check if buffer is full
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
return 0;
}
2019-06-14 02:10:48 -04:00
if (pxRingbuffer->pucAcquire == pxRingbuffer->pucHead && pxRingbuffer->pucFree == pxRingbuffer->pucHead) {
//Check for special case where pucAcquire and pucFree are both at pucHead
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xFreeSize = pxRingbuffer->xSize - rbHEADER_SIZE;
2019-06-14 02:10:48 -04:00
} else if (pxRingbuffer->pucAcquire < pxRingbuffer->pucFree) {
//Free space is contiguous between pucAcquire and pucFree, requires single header
xFreeSize = (pxRingbuffer->pucFree - pxRingbuffer->pucAcquire) - rbHEADER_SIZE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
} else {
//Free space wraps around, requires two headers
xFreeSize = (pxRingbuffer->pucFree - pxRingbuffer->pucHead) +
2019-06-14 02:10:48 -04:00
(pxRingbuffer->pucTail - pxRingbuffer->pucAcquire) -
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
(rbHEADER_SIZE * 2);
}
//Check for xFreeSize < 0 before checking xFreeSize > pxRingbuffer->xMaxItemSize
//to avoid incorrect comparison operation when xFreeSize is negative
if (xFreeSize < 0) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xFreeSize = 0;
} else if (xFreeSize > pxRingbuffer->xMaxItemSize) {
//Limit free size to be within bounds
xFreeSize = pxRingbuffer->xMaxItemSize;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xFreeSize;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
static size_t prvGetCurMaxSizeByteBuf(Ringbuffer_t *pxRingbuffer)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xFreeSize;
//Check if buffer is full
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_FULL_FLAG) {
return 0;
}
/*
* Return whatever space is available depending on relative positions of the free
2019-06-14 02:10:48 -04:00
* pointer and Acquire pointer. There is no overhead of headers in this mode
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
*/
2019-06-14 02:10:48 -04:00
xFreeSize = pxRingbuffer->pucFree - pxRingbuffer->pucAcquire;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xFreeSize <= 0) {
xFreeSize += pxRingbuffer->xSize;
}
return xFreeSize;
}
static BaseType_t prvReceiveGeneric(Ringbuffer_t *pxRingbuffer,
void **pvItem1,
void **pvItem2,
size_t *xItemSize1,
size_t *xItemSize2,
size_t xMaxSize,
TickType_t xTicksToWait)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xReturn = pdFALSE;
BaseType_t xReturnSemaphore = pdFALSE;
TickType_t xTicksEnd = xTaskGetTickCount() + xTicksToWait;
TickType_t xTicksRemaining = xTicksToWait;
while (xTicksRemaining <= xTicksToWait) { //xTicksToWait will underflow once xTaskGetTickCount() > ticks_end
//Block until more free space becomes available or timeout
if (xSemaphoreTake(rbGET_RX_SEM_HANDLE(pxRingbuffer), xTicksRemaining) != pdTRUE) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xReturn = pdFALSE; //Timed out attempting to get semaphore
break;
}
//Semaphore obtained, check if item can be retrieved
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (prvCheckItemAvail(pxRingbuffer) == pdTRUE) {
//Item is available for retrieval
BaseType_t xIsSplit = pdFALSE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) {
//Second argument (pxIsSplit) is unused for byte buffers
*pvItem1 = pxRingbuffer->pvGetItem(pxRingbuffer, NULL, xMaxSize, xItemSize1);
} else {
//Third argument (xMaxSize) is unused for no-split/allow-split buffers
*pvItem1 = pxRingbuffer->pvGetItem(pxRingbuffer, &xIsSplit, 0, xItemSize1);
}
//Check for item split if configured to do so
if ((pxRingbuffer->uxRingbufferFlags & rbALLOW_SPLIT_FLAG) && (pvItem2 != NULL) && (xItemSize2 != NULL)) {
if (xIsSplit == pdTRUE) {
*pvItem2 = pxRingbuffer->pvGetItem(pxRingbuffer, &xIsSplit, 0, xItemSize2);
configASSERT(*pvItem2 < *pvItem1); //Check wrap around has occurred
configASSERT(xIsSplit == pdFALSE); //Second part should not have wrapped flag
} else {
*pvItem2 = NULL;
}
}
xReturn = pdTRUE;
if (pxRingbuffer->xItemsWaiting > 0) {
xReturnSemaphore = pdTRUE;
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
break;
}
//No item available for retrieval, adjust ticks and take the semaphore again
if (xTicksToWait != portMAX_DELAY) {
xTicksRemaining = xTicksEnd - xTaskGetTickCount();
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
/*
* Gap between critical section and re-acquiring of the semaphore. If
* semaphore is given now, priority inversion might occur (see docs)
*/
}
if (xReturnSemaphore == pdTRUE) {
xSemaphoreGive(rbGET_RX_SEM_HANDLE(pxRingbuffer)); //Give semaphore back so other tasks can retrieve
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xReturn;
}
static BaseType_t prvReceiveGenericFromISR(Ringbuffer_t *pxRingbuffer,
void **pvItem1,
void **pvItem2,
size_t *xItemSize1,
size_t *xItemSize2,
size_t xMaxSize)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xReturn = pdFALSE;
BaseType_t xReturnSemaphore = pdFALSE;
portENTER_CRITICAL_ISR(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if(prvCheckItemAvail(pxRingbuffer) == pdTRUE) {
BaseType_t xIsSplit = pdFALSE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) {
//Second argument (pxIsSplit) is unused for byte buffers
*pvItem1 = pxRingbuffer->pvGetItem(pxRingbuffer, NULL, xMaxSize, xItemSize1);
} else {
//Third argument (xMaxSize) is unused for no-split/allow-split buffers
*pvItem1 = pxRingbuffer->pvGetItem(pxRingbuffer, &xIsSplit, 0, xItemSize1);
}
//Check for item split if configured to do so
if ((pxRingbuffer->uxRingbufferFlags & rbALLOW_SPLIT_FLAG) && pvItem2 != NULL && xItemSize2 != NULL) {
if (xIsSplit == pdTRUE) {
*pvItem2 = pxRingbuffer->pvGetItem(pxRingbuffer, &xIsSplit, 0, xItemSize2);
configASSERT(*pvItem2 < *pvItem1); //Check wrap around has occurred
configASSERT(xIsSplit == pdFALSE); //Second part should not have wrapped flag
} else {
*pvItem2 = NULL;
}
}
xReturn = pdTRUE;
if (pxRingbuffer->xItemsWaiting > 0) {
xReturnSemaphore = pdTRUE;
}
}
portEXIT_CRITICAL_ISR(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xReturnSemaphore == pdTRUE) {
xSemaphoreGiveFromISR(rbGET_RX_SEM_HANDLE(pxRingbuffer), NULL); //Give semaphore back so other tasks can retrieve
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xReturn;
}
/* --------------------------- Public Definitions --------------------------- */
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
RingbufHandle_t xRingbufferCreate(size_t xBufferSize, RingbufferType_t xBufferType)
{
configASSERT(xBufferSize > 0);
configASSERT(xBufferType < RINGBUF_TYPE_MAX);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Allocate memory
if (xBufferType != RINGBUF_TYPE_BYTEBUF) {
xBufferSize = rbALIGN_SIZE(xBufferSize); //xBufferSize is rounded up for no-split/allow-split buffers
}
Ringbuffer_t *pxNewRingbuffer = calloc(1, sizeof(Ringbuffer_t));
uint8_t *pucRingbufferStorage = malloc(xBufferSize);
if (pxNewRingbuffer == NULL || pucRingbufferStorage == NULL) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
goto err;
}
//Initialize Semaphores
#if ( configSUPPORT_STATIC_ALLOCATION == 1)
//We don't use the handles for static semaphores, and xSemaphoreCreateBinaryStatic will never fail thus no need to check static case
xSemaphoreCreateBinaryStatic(&(pxNewRingbuffer->xTransSemStatic));
xSemaphoreCreateBinaryStatic(&(pxNewRingbuffer->xRecvSemStatic));
#else
pxNewRingbuffer->xTransSemHandle = xSemaphoreCreateBinary();
pxNewRingbuffer->xRecvSemHandle = xSemaphoreCreateBinary();
if (pxNewRingbuffer->xTransSemHandle == NULL || pxNewRingbuffer->xRecvSemHandle == NULL) {
if (pxNewRingbuffer->xTransSemHandle != NULL) {
vSemaphoreDelete(pxNewRingbuffer->xTransSemHandle);
}
if (pxNewRingbuffer->xRecvSemHandle != NULL) {
vSemaphoreDelete(pxNewRingbuffer->xRecvSemHandle);
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
goto err;
}
#endif
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
prvInitializeNewRingbuffer(xBufferSize, xBufferType, pxNewRingbuffer, pucRingbufferStorage);
return (RingbufHandle_t)pxNewRingbuffer;
err:
//An error has occurred, Free memory and return NULL
free(pxNewRingbuffer);
free(pucRingbufferStorage);
return NULL;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
RingbufHandle_t xRingbufferCreateNoSplit(size_t xItemSize, size_t xItemNum)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return xRingbufferCreate((rbALIGN_SIZE(xItemSize) + rbHEADER_SIZE) * xItemNum, RINGBUF_TYPE_NOSPLIT);
}
#if ( configSUPPORT_STATIC_ALLOCATION == 1 )
RingbufHandle_t xRingbufferCreateStatic(size_t xBufferSize,
RingbufferType_t xBufferType,
uint8_t *pucRingbufferStorage,
StaticRingbuffer_t *pxStaticRingbuffer)
{
//Check arguments
configASSERT(xBufferSize > 0);
configASSERT(xBufferType < RINGBUF_TYPE_MAX);
configASSERT(pucRingbufferStorage != NULL && pxStaticRingbuffer != NULL);
if (xBufferType != RINGBUF_TYPE_BYTEBUF) {
//No-split/allow-split buffer sizes must be 32-bit aligned
configASSERT(rbCHECK_ALIGNED(xBufferSize));
}
Ringbuffer_t *pxNewRingbuffer = (Ringbuffer_t *)pxStaticRingbuffer;
xSemaphoreCreateBinaryStatic(&(pxNewRingbuffer->xTransSemStatic));
xSemaphoreCreateBinaryStatic(&(pxNewRingbuffer->xRecvSemStatic));
prvInitializeNewRingbuffer(xBufferSize, xBufferType, pxNewRingbuffer, pucRingbufferStorage);
pxNewRingbuffer->uxRingbufferFlags |= rbBUFFER_STATIC_FLAG;
return (RingbufHandle_t)pxNewRingbuffer;
}
#endif
BaseType_t xRingbufferSendAcquire(RingbufHandle_t xRingbuffer, void **ppvItem, size_t xItemSize, TickType_t xTicksToWait)
{
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(ppvItem != NULL || xItemSize == 0);
//currently only supported in NoSplit buffers
configASSERT((pxRingbuffer->uxRingbufferFlags & (rbBYTE_BUFFER_FLAG | rbALLOW_SPLIT_FLAG)) == 0);
*ppvItem = NULL;
if (xItemSize > pxRingbuffer->xMaxItemSize) {
return pdFALSE; //Data will never ever fit in the queue.
}
if ((pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) && xItemSize == 0) {
return pdTRUE; //Sending 0 bytes to byte buffer has no effect
}
//Attempt to send an item
BaseType_t xReturn = pdFALSE;
BaseType_t xReturnSemaphore = pdFALSE;
TickType_t xTicksEnd = xTaskGetTickCount() + xTicksToWait;
TickType_t xTicksRemaining = xTicksToWait;
while (xTicksRemaining <= xTicksToWait) { //xTicksToWait will underflow once xTaskGetTickCount() > ticks_end
//Block until more free space becomes available or timeout
if (xSemaphoreTake(rbGET_TX_SEM_HANDLE(pxRingbuffer), xTicksRemaining) != pdTRUE) {
xReturn = pdFALSE;
break;
}
//Semaphore obtained, check if item can fit
portENTER_CRITICAL(&pxRingbuffer->mux);
if(pxRingbuffer->xCheckItemFits(pxRingbuffer, xItemSize) == pdTRUE) {
//Item will fit, copy item
*ppvItem = prvAcquireItemNoSplit(pxRingbuffer, xItemSize);
xReturn = pdTRUE;
//Check if the free semaphore should be returned to allow other tasks to send
if (prvGetFreeSize(pxRingbuffer) > 0) {
xReturnSemaphore = pdTRUE;
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
break;
}
//Item doesn't fit, adjust ticks and take the semaphore again
if (xTicksToWait != portMAX_DELAY) {
xTicksRemaining = xTicksEnd - xTaskGetTickCount();
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
/*
* Gap between critical section and re-acquiring of the semaphore. If
* semaphore is given now, priority inversion might occur (see docs)
*/
}
if (xReturnSemaphore == pdTRUE) {
xSemaphoreGive(rbGET_TX_SEM_HANDLE(pxRingbuffer)); //Give back semaphore so other tasks can acquire
}
return xReturn;
}
BaseType_t xRingbufferSendComplete(RingbufHandle_t xRingbuffer, void *pvItem)
{
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pvItem != NULL);
configASSERT((pxRingbuffer->uxRingbufferFlags & (rbBYTE_BUFFER_FLAG | rbALLOW_SPLIT_FLAG)) == 0);
portENTER_CRITICAL(&pxRingbuffer->mux);
prvSendItemDoneNoSplit(pxRingbuffer, pvItem);
portEXIT_CRITICAL(&pxRingbuffer->mux);
xSemaphoreGive(rbGET_RX_SEM_HANDLE(pxRingbuffer));
return pdTRUE;
}
BaseType_t xRingbufferSend(RingbufHandle_t xRingbuffer,
const void *pvItem,
size_t xItemSize,
TickType_t xTicksToWait)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pvItem != NULL || xItemSize == 0);
if (xItemSize > pxRingbuffer->xMaxItemSize) {
return pdFALSE; //Data will never ever fit in the queue.
}
if ((pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) && xItemSize == 0) {
return pdTRUE; //Sending 0 bytes to byte buffer has no effect
}
//Attempt to send an item
BaseType_t xReturn = pdFALSE;
BaseType_t xReturnSemaphore = pdFALSE;
TickType_t xTicksEnd = xTaskGetTickCount() + xTicksToWait;
TickType_t xTicksRemaining = xTicksToWait;
while (xTicksRemaining <= xTicksToWait) { //xTicksToWait will underflow once xTaskGetTickCount() > ticks_end
//Block until more free space becomes available or timeout
if (xSemaphoreTake(rbGET_TX_SEM_HANDLE(pxRingbuffer), xTicksRemaining) != pdTRUE) {
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xReturn = pdFALSE;
break;
}
//Semaphore obtained, check if item can fit
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if(pxRingbuffer->xCheckItemFits(pxRingbuffer, xItemSize) == pdTRUE) {
//Item will fit, copy item
pxRingbuffer->vCopyItem(pxRingbuffer, pvItem, xItemSize);
xReturn = pdTRUE;
//Check if the free semaphore should be returned to allow other tasks to send
if (prvGetFreeSize(pxRingbuffer) > 0) {
xReturnSemaphore = pdTRUE;
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
break;
}
//Item doesn't fit, adjust ticks and take the semaphore again
if (xTicksToWait != portMAX_DELAY) {
xTicksRemaining = xTicksEnd - xTaskGetTickCount();
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
/*
* Gap between critical section and re-acquiring of the semaphore. If
* semaphore is given now, priority inversion might occur (see docs)
*/
}
if (xReturnSemaphore == pdTRUE) {
xSemaphoreGive(rbGET_TX_SEM_HANDLE(pxRingbuffer)); //Give back semaphore so other tasks can send
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xReturn == pdTRUE) {
//Indicate item was successfully sent
xSemaphoreGive(rbGET_RX_SEM_HANDLE(pxRingbuffer));
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xReturn;
}
BaseType_t xRingbufferSendFromISR(RingbufHandle_t xRingbuffer,
const void *pvItem,
size_t xItemSize,
BaseType_t *pxHigherPriorityTaskWoken)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pvItem != NULL || xItemSize == 0);
if (xItemSize > pxRingbuffer->xMaxItemSize) {
return pdFALSE; //Data will never ever fit in the queue.
}
if ((pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG) && xItemSize == 0) {
return pdTRUE; //Sending 0 bytes to byte buffer has no effect
}
//Attempt to send an item
BaseType_t xReturn;
BaseType_t xReturnSemaphore = pdFALSE;
portENTER_CRITICAL_ISR(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (pxRingbuffer->xCheckItemFits(xRingbuffer, xItemSize) == pdTRUE) {
pxRingbuffer->vCopyItem(xRingbuffer, pvItem, xItemSize);
xReturn = pdTRUE;
//Check if the free semaphore should be returned to allow other tasks to send
if (prvGetFreeSize(pxRingbuffer) > 0) {
xReturnSemaphore = pdTRUE;
}
} else {
xReturn = pdFALSE;
}
portEXIT_CRITICAL_ISR(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xReturnSemaphore == pdTRUE) {
xSemaphoreGiveFromISR(rbGET_TX_SEM_HANDLE(pxRingbuffer), pxHigherPriorityTaskWoken); //Give back semaphore so other tasks can send
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xReturn == pdTRUE) {
//Indicate item was successfully sent
xSemaphoreGiveFromISR(rbGET_RX_SEM_HANDLE(pxRingbuffer), pxHigherPriorityTaskWoken);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
return xReturn;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void *xRingbufferReceive(RingbufHandle_t xRingbuffer, size_t *pxItemSize, TickType_t xTicksToWait)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
//Attempt to retrieve an item
void *pvTempItem;
size_t xTempSize;
if (prvReceiveGeneric(pxRingbuffer, &pvTempItem, NULL, &xTempSize, NULL, 0, xTicksToWait) == pdTRUE) {
if (pxItemSize != NULL) {
*pxItemSize = xTempSize;
}
return pvTempItem;
} else {
return NULL;
}
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void *xRingbufferReceiveFromISR(RingbufHandle_t xRingbuffer, size_t *pxItemSize)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
//Attempt to retrieve an item
void *pvTempItem;
size_t xTempSize;
if (prvReceiveGenericFromISR(pxRingbuffer, &pvTempItem, NULL, &xTempSize, NULL, 0) == pdTRUE) {
if (pxItemSize != NULL) {
*pxItemSize = xTempSize;
}
return pvTempItem;
} else {
return NULL;
}
}
BaseType_t xRingbufferReceiveSplit(RingbufHandle_t xRingbuffer,
void **ppvHeadItem,
void **ppvTailItem,
size_t *pxHeadItemSize,
size_t *pxTailItemSize,
TickType_t xTicksToWait)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pxRingbuffer->uxRingbufferFlags & rbALLOW_SPLIT_FLAG);
configASSERT(ppvHeadItem != NULL && ppvTailItem != NULL);
//Attempt to retrieve multiple items
void *pvTempHeadItem, *pvTempTailItem;
size_t xTempHeadSize, xTempTailSize;
if (prvReceiveGeneric(pxRingbuffer, &pvTempHeadItem, &pvTempTailItem, &xTempHeadSize, &xTempTailSize, 0, xTicksToWait) == pdTRUE) {
//At least one item was retrieved
*ppvHeadItem = pvTempHeadItem;
if(pxHeadItemSize != NULL){
*pxHeadItemSize = xTempHeadSize;
}
//Check to see if a second item was also retrieved
if (pvTempTailItem != NULL) {
*ppvTailItem = pvTempTailItem;
if (pxTailItemSize != NULL) {
*pxTailItemSize = xTempTailSize;
}
} else {
*ppvTailItem = NULL;
}
return pdTRUE;
} else {
//No items retrieved
*ppvHeadItem = NULL;
*ppvTailItem = NULL;
return pdFALSE;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
BaseType_t xRingbufferReceiveSplitFromISR(RingbufHandle_t xRingbuffer,
void **ppvHeadItem,
void **ppvTailItem,
size_t *pxHeadItemSize,
size_t *pxTailItemSize)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pxRingbuffer->uxRingbufferFlags & rbALLOW_SPLIT_FLAG);
configASSERT(ppvHeadItem != NULL && ppvTailItem != NULL);
//Attempt to retrieve multiple items
void *pvTempHeadItem = NULL, *pvTempTailItem = NULL;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xTempHeadSize, xTempTailSize;
if (prvReceiveGenericFromISR(pxRingbuffer, &pvTempHeadItem, &pvTempTailItem, &xTempHeadSize, &xTempTailSize, 0) == pdTRUE) {
//At least one item was received
*ppvHeadItem = pvTempHeadItem;
if (pxHeadItemSize != NULL) {
*pxHeadItemSize = xTempHeadSize;
}
//Check to see if a second item was also retrieved
if (pvTempTailItem != NULL) {
*ppvTailItem = pvTempTailItem;
if (pxTailItemSize != NULL) {
*pxTailItemSize = xTempTailSize;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
} else {
*ppvTailItem = NULL;
}
return pdTRUE;
} else {
*ppvHeadItem = NULL;
*ppvTailItem = NULL;
return pdFALSE;
}
}
void *xRingbufferReceiveUpTo(RingbufHandle_t xRingbuffer,
size_t *pxItemSize,
TickType_t xTicksToWait,
size_t xMaxSize)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG); //This function should only be called for byte buffers
if (xMaxSize == 0) {
return NULL;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Attempt to retrieve up to xMaxSize bytes
void *pvTempItem;
size_t xTempSize;
if (prvReceiveGeneric(pxRingbuffer, &pvTempItem, NULL, &xTempSize, NULL, xMaxSize, xTicksToWait) == pdTRUE) {
if (pxItemSize != NULL) {
*pxItemSize = xTempSize;
}
return pvTempItem;
} else {
return NULL;
}
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void *xRingbufferReceiveUpToFromISR(RingbufHandle_t xRingbuffer, size_t *pxItemSize, size_t xMaxSize)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Check arguments
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pxRingbuffer->uxRingbufferFlags & rbBYTE_BUFFER_FLAG); //This function should only be called for byte buffers
if (xMaxSize == 0) {
return NULL;
}
//Attempt to retrieve up to xMaxSize bytes
void *pvTempItem;
size_t xTempSize;
if (prvReceiveGenericFromISR(pxRingbuffer, &pvTempItem, NULL, &xTempSize, NULL, xMaxSize) == pdTRUE) {
if (pxItemSize != NULL) {
*pxItemSize = xTempSize;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return pvTempItem;
} else {
return NULL;
}
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void vRingbufferReturnItem(RingbufHandle_t xRingbuffer, void *pvItem)
2016-10-24 09:18:02 -04:00
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pvItem != NULL);
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->vReturnItem(pxRingbuffer, (uint8_t *)pvItem);
portEXIT_CRITICAL(&pxRingbuffer->mux);
xSemaphoreGive(rbGET_TX_SEM_HANDLE(pxRingbuffer));
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void vRingbufferReturnItemFromISR(RingbufHandle_t xRingbuffer, void *pvItem, BaseType_t *pxHigherPriorityTaskWoken)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
configASSERT(pvItem != NULL);
portENTER_CRITICAL_ISR(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->vReturnItem(pxRingbuffer, (uint8_t *)pvItem);
portEXIT_CRITICAL_ISR(&pxRingbuffer->mux);
xSemaphoreGiveFromISR(rbGET_TX_SEM_HANDLE(pxRingbuffer), pxHigherPriorityTaskWoken);
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
void vRingbufferDelete(RingbufHandle_t xRingbuffer)
{
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
vSemaphoreDelete(rbGET_TX_SEM_HANDLE(pxRingbuffer));
vSemaphoreDelete(rbGET_RX_SEM_HANDLE(pxRingbuffer));
#if ( configSUPPORT_STATIC_ALLOCATION == 1 )
if (pxRingbuffer->uxRingbufferFlags & rbBUFFER_STATIC_FLAG) {
//Ring buffer was statically allocated, no need to free
return;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
#endif
free(pxRingbuffer->pucHead);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
free(pxRingbuffer);
2016-10-24 09:18:02 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xRingbufferGetMaxItemSize(RingbufHandle_t xRingbuffer)
2016-10-24 09:18:02 -04:00
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
return pxRingbuffer->xMaxItemSize;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
size_t xRingbufferGetCurFreeSize(RingbufHandle_t xRingbuffer)
{
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
size_t xFreeSize;
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
xFreeSize = pxRingbuffer->xGetCurMaxSize(pxRingbuffer);
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return xFreeSize;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xRingbufferAddToQueueSetRead(RingbufHandle_t xRingbuffer, QueueSetHandle_t xQueueSet)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
BaseType_t xReturn;
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Cannot add semaphore to queue set if semaphore is not empty. Temporarily hold semaphore
BaseType_t xHoldSemaphore = xSemaphoreTake(rbGET_RX_SEM_HANDLE(pxRingbuffer), 0);
xReturn = xQueueAddToSet(rbGET_RX_SEM_HANDLE(pxRingbuffer), xQueueSet);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xHoldSemaphore == pdTRUE) {
//Return semaphore if temporarily held
configASSERT(xSemaphoreGive(rbGET_RX_SEM_HANDLE(pxRingbuffer)) == pdTRUE);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return xReturn;
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xRingbufferCanRead(RingbufHandle_t xRingbuffer, QueueSetMemberHandle_t xMember)
{
//Check if the selected queue set member is the ring buffer's read semaphore
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
return (rbGET_RX_SEM_HANDLE(pxRingbuffer) == xMember) ? pdTRUE : pdFALSE;
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
BaseType_t xRingbufferRemoveFromQueueSetRead(RingbufHandle_t xRingbuffer, QueueSetHandle_t xQueueSet)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
BaseType_t xReturn;
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
//Cannot remove semaphore from queue set if semaphore is not empty. Temporarily hold semaphore
BaseType_t xHoldSemaphore = xSemaphoreTake(rbGET_RX_SEM_HANDLE(pxRingbuffer), 0);
xReturn = xQueueRemoveFromSet(rbGET_RX_SEM_HANDLE(pxRingbuffer), xQueueSet);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (xHoldSemaphore == pdTRUE) {
//Return semaphore if temporarily held
configASSERT(xSemaphoreGive(rbGET_RX_SEM_HANDLE(pxRingbuffer)) == pdTRUE);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
return xReturn;
}
void vRingbufferGetInfo(RingbufHandle_t xRingbuffer,
UBaseType_t *uxFree,
UBaseType_t *uxRead,
UBaseType_t *uxWrite,
2019-06-14 02:10:48 -04:00
UBaseType_t *uxAcquire,
UBaseType_t *uxItemsWaiting)
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
{
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
portENTER_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (uxFree != NULL) {
*uxFree = (UBaseType_t)(pxRingbuffer->pucFree - pxRingbuffer->pucHead);
}
if (uxRead != NULL) {
*uxRead = (UBaseType_t)(pxRingbuffer->pucRead - pxRingbuffer->pucHead);
}
if (uxWrite != NULL) {
*uxWrite = (UBaseType_t)(pxRingbuffer->pucWrite - pxRingbuffer->pucHead);
}
2019-06-14 02:10:48 -04:00
if (uxAcquire != NULL) {
*uxAcquire = (UBaseType_t)(pxRingbuffer->pucAcquire - pxRingbuffer->pucHead);
}
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
if (uxItemsWaiting != NULL) {
*uxItemsWaiting = (UBaseType_t)(pxRingbuffer->xItemsWaiting);
}
portEXIT_CRITICAL(&pxRingbuffer->mux);
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
}
void xRingbufferPrintInfo(RingbufHandle_t xRingbuffer)
{
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
Ringbuffer_t *pxRingbuffer = (Ringbuffer_t *)xRingbuffer;
configASSERT(pxRingbuffer);
2019-06-14 02:10:48 -04:00
printf("Rb size:%d\tfree: %d\trptr: %d\tfreeptr: %d\twptr: %d, aptr: %d\n",
freeRTOS/Re-factor ring buffers This fixes multiple bugs with ring buffers and re-factors the code. The public API has not changed, however the underlying implementation have various private functions have been changed. The following behavioral changes have been made - Size of ring buffers for No-Split/Allow-Split buffers will not be rounded up to the nearest 32-bit aligned size. This was done to simplify the implementation - Item size for No-Split/Allow-Split buffers will also be rounded up to the nearest 32-bit aligned size. The following bugs have been fixed - In copyItemToRingbufAllowSplit(), when copying an item where the aligned size is smaller than the remaining length, the function does not consider the case where the true size of the item is less than 4 bytes. - The copy functions will automatically wrap around the write pointers when the remaining length of the buffer is not large enough to fit a header, but does not consider if wrapping around will cause an overlap with the read pointer. This will make a full buffer be mistaken for an empty buffer closes #1711 - xRingbufferSend() can get stuck in a infinite loop when the size of the free memory is larger than the needed_size, but too small to fit in the ring buffer due to alignment and extra overhead of wrapping around. closes #1846 - Fixed documentation with ring buffer queue set API - Adding and removing from queue set does not consider the case where the read/write semaphores actually hold a value. The following functions have been deprecated - xRingbufferIsNextItemWrapped() due to lack of thread safety - xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite() as adding the queue sets only work under receive operations. The following functions have been added - xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread safe way to receive from allow-split buffers - vRingbufferGetInfo() Documentation for ring buffers has also been added.
2018-04-18 13:20:34 -04:00
pxRingbuffer->xSize, prvGetFreeSize(pxRingbuffer),
pxRingbuffer->pucRead - pxRingbuffer->pucHead,
pxRingbuffer->pucFree - pxRingbuffer->pucHead,
2019-06-14 02:10:48 -04:00
pxRingbuffer->pucWrite - pxRingbuffer->pucHead,
pxRingbuffer->pucAcquire - pxRingbuffer->pucHead);
}