alex/esp-idf
1
0
Fork 0
mirror of https://github.com/espressif/esp-idf.git synced 2025-12-07 09:02:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add bytebuffer support to ringbuf.c

Browse Source
This commit is contained in:
Jeroen Domburg
2016-10-24 21:18:02 +08:00
parent 74aff2b9d2
commit d9005e739d
2 changed files with 303 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

344
components/freertos/ringbuf.c
View File

@@ -18,6 +18,7 @@
#include "freertos/queue.h"
#include "freertos/xtensa_api.h"
#include "freertos/ringbuf.h"
#include "esp_attr.h"
#include <stdint.h>
#include <string.h>
#include <stdlib.h>
@@ -25,6 +26,7 @@
typedef enum {
flag_allowsplit = 1,
flag_bytebuf = 2,
} rbflag_t;
typedef enum {
@@ -33,8 +35,10 @@ typedef enum {
} itemflag_t;
typedef struct ringbuf_t ringbuf_t;
//The ringbuffer structure
typedef struct {
struct ringbuf_t {
SemaphoreHandle_t free_space_sem; //Binary semaphore, wakes up writing threads when there's more free space
SemaphoreHandle_t items_buffered_sem; //Binary semaphore, indicates there are new packets in the circular buffer. See remark.
size_t size; //Size of the data storage
@@ -44,7 +48,12 @@ typedef struct {
uint8_t *data; //Data storage
portMUX_TYPE mux; //Spinlock for actual data/ptr/struct modification
rbflag_t flags;
} ringbuf_t;
size_t maxItemSize;
//The following keep function pointers to hold different implementations for ringbuffer management.
BaseType_t (*copyItemToRingbufImpl)(ringbuf_t *rb, uint8_t *buffer, size_t buffer_size);
uint8_t *(*getItemFromRingbufImpl)(ringbuf_t *rb, size_t *length, int wanted_length);
void (*returnItemToRingbufImpl)(ringbuf_t *rb, void *item);
};
@@ -73,14 +82,16 @@ static int ringbufferFreeMem(ringbuf_t *rb)
return free_size-1;
}
//Copies a single item to the ring buffer. Assumes there is space in the ringbuffer and
//Copies a single item to the ring buffer; refuses to split items. Assumes there is space in the ringbuffer and
//the ringbuffer is locked. Increases write_ptr to the next item. Returns pdTRUE on
//success, pdFALSE if it can't make the item fit and the calling routine needs to retry
//later or fail.
//This function by itself is not threadsafe, always call from within a muxed section.
static BaseType_t copyItemToRingbuf(ringbuf_t *rb, uint8_t *buffer, size_t buffer_size)
static BaseType_t copyItemToRingbufNoSplit(ringbuf_t *rb, uint8_t *buffer, size_t buffer_size)
{
size_t rbuffer_size=(buffer_size+3)&~3; //Payload length, rounded to next 32-bit value
size_t rbuffer_size;
rbuffer_size=(buffer_size+3)&~3; //Payload length, rounded to next 32-bit value
configASSERT(((int)rb->write_ptr&3)==0); //write_ptr needs to be 32-bit aligned
configASSERT(rb->write_ptr-(rb->data+rb->size) >= sizeof(buf_entry_hdr_t)); //need to have at least the size
//of a header to the end of the ringbuff
@@ -88,65 +99,28 @@ static BaseType_t copyItemToRingbuf(ringbuf_t *rb, uint8_t *buffer, size_t buffe
//See if we have enough contiguous space to write the buffer.
if (rem_len < rbuffer_size + sizeof(buf_entry_hdr_t)) {
//The buffer can't be contiguously written to the ringbuffer, but needs special handling. Do
//that depending on how the ringbuffer is configured.
//The code here is also expected to check if the buffer, mangled in whatever way is implemented,
//will still fit, and return pdFALSE if that is not the case.
if (rb->flags & flag_allowsplit) {
//Buffer plus header is not going to fit in the room from wr_pos to the end of the
//ringbuffer... we need to split the write in two.
//First, see if this will fit at all.
if (ringbufferFreeMem(rb) < (sizeof(buf_entry_hdr_t)*2)+rbuffer_size) {
//Will not fit.
return pdFALSE;
}
//Because the code at the end of the function makes sure we always have
//room for a header, this should never assert.
configASSERT(rem_len>=sizeof(buf_entry_hdr_t));
//Okay, it should fit. Write everything.
//First, place bit of buffer that does fit. Write header first...
buf_entry_hdr_t *hdr=(buf_entry_hdr_t *)rb->write_ptr;
hdr->flags=0;
hdr->len=rem_len-sizeof(buf_entry_hdr_t);
rb->write_ptr+=sizeof(buf_entry_hdr_t);
rem_len-=sizeof(buf_entry_hdr_t);
if (rem_len!=0) {
//..then write the data bit that fits.
memcpy(rb->write_ptr, buffer, rem_len);
//Update vars so the code later on will write the rest of the data.
buffer+=rem_len;
rbuffer_size-=rem_len;
buffer_size-=rem_len;
} else {
//Huh, only the header fit. Mark as dummy so the receive function doesn't receive
//an useless zero-byte packet.
hdr->flags|=iflag_dummydata;
}
rb->write_ptr=rb->data;
} else {
//Buffer plus header is not going to fit in the room from wr_pos to the end of the
//ringbuffer... but we're not allowed to split the buffer. We need to fill the
//rest of the ringbuffer with a dummy item so we can place the data at the _start_ of
//the ringbuffer..
//First, find out if we actually have enough space at the start of the ringbuffer to
//make this work (Again, we need 4 bytes extra because otherwise read_ptr==free_ptr)
if (rb->free_ptr-rb->data < rbuffer_size+sizeof(buf_entry_hdr_t)+4) {
//Will not fit.
return pdFALSE;
}
//If the read buffer hasn't wrapped around yet, there's no way this will work either.
if (rb->free_ptr > rb->write_ptr) {
//No luck.
return pdFALSE;
}
//Okay, it will fit. Mark the rest of the ringbuffer space with a dummy packet.
buf_entry_hdr_t *hdr=(buf_entry_hdr_t *)rb->write_ptr;
hdr->flags=iflag_dummydata;
//Reset the write pointer to the start of the ringbuffer so the code later on can
//happily write the data.
rb->write_ptr=rb->data;
//Buffer plus header is not going to fit in the room from wr_pos to the end of the
//ringbuffer... but we're not allowed to split the buffer. We need to fill the
//rest of the ringbuffer with a dummy item so we can place the data at the _start_ of
//the ringbuffer..
//First, find out if we actually have enough space at the start of the ringbuffer to
//make this work (Again, we need 4 bytes extra because otherwise read_ptr==free_ptr)
if (rb->free_ptr-rb->data < rbuffer_size+sizeof(buf_entry_hdr_t)+4) {
//Will not fit.
return pdFALSE;
}
//If the read buffer hasn't wrapped around yet, there's no way this will work either.
if (rb->free_ptr > rb->write_ptr) {
//No luck.
return pdFALSE;
}
//Okay, it will fit. Mark the rest of the ringbuffer space with a dummy packet.
buf_entry_hdr_t *hdr=(buf_entry_hdr_t *)rb->write_ptr;
hdr->flags=iflag_dummydata;
//Reset the write pointer to the start of the ringbuffer so the code later on can
//happily write the data.
rb->write_ptr=rb->data;
} else {
//No special handling needed. Checking if it's gonna fit probably still is a good idea.
if (ringbufferFreeMem(rb) < sizeof(buf_entry_hdr_t)+rbuffer_size) {
@@ -174,9 +148,117 @@ static BaseType_t copyItemToRingbuf(ringbuf_t *rb, uint8_t *buffer, size_t buffe
return pdTRUE;
}
//Copies a single item to the ring buffer; allows split items. Assumes there is space in the ringbuffer and
//the ringbuffer is locked. Increases write_ptr to the next item. Returns pdTRUE on
//success, pdFALSE if it can't make the item fit and the calling routine needs to retry
//later or fail.
//This function by itself is not threadsafe, always call from within a muxed section.
static BaseType_t copyItemToRingbufAllowSplit(ringbuf_t *rb, uint8_t *buffer, size_t buffer_size)
{
size_t rbuffer_size;
rbuffer_size=(buffer_size+3)&~3; //Payload length, rounded to next 32-bit value
configASSERT(((int)rb->write_ptr&3)==0); //write_ptr needs to be 32-bit aligned
configASSERT(rb->write_ptr-(rb->data+rb->size) >= sizeof(buf_entry_hdr_t)); //need to have at least the size
//of a header to the end of the ringbuff
size_t rem_len=(rb->data + rb->size) - rb->write_ptr; //length remaining until end of ringbuffer
//See if we have enough contiguous space to write the buffer.
if (rem_len < rbuffer_size + sizeof(buf_entry_hdr_t)) {
//The buffer can't be contiguously written to the ringbuffer, but needs special handling. Do
//that depending on how the ringbuffer is configured.
//The code here is also expected to check if the buffer, mangled in whatever way is implemented,
//will still fit, and return pdFALSE if that is not the case.
//Buffer plus header is not going to fit in the room from wr_pos to the end of the
//ringbuffer... we need to split the write in two.
//First, see if this will fit at all.
if (ringbufferFreeMem(rb) < (sizeof(buf_entry_hdr_t)*2)+rbuffer_size) {
//Will not fit.
return pdFALSE;
}
//Because the code at the end of the function makes sure we always have
//room for a header, this should never assert.
configASSERT(rem_len>=sizeof(buf_entry_hdr_t));
//Okay, it should fit. Write everything.
//First, place bit of buffer that does fit. Write header first...
buf_entry_hdr_t *hdr=(buf_entry_hdr_t *)rb->write_ptr;
hdr->flags=0;
hdr->len=rem_len-sizeof(buf_entry_hdr_t);
rb->write_ptr+=sizeof(buf_entry_hdr_t);
rem_len-=sizeof(buf_entry_hdr_t);
if (rem_len!=0) {
//..then write the data bit that fits.
memcpy(rb->write_ptr, buffer, rem_len);
//Update vars so the code later on will write the rest of the data.
buffer+=rem_len;
rbuffer_size-=rem_len;
buffer_size-=rem_len;
} else {
//Huh, only the header fit. Mark as dummy so the receive function doesn't receive
//an useless zero-byte packet.
hdr->flags|=iflag_dummydata;
}
rb->write_ptr=rb->data;
} else {
//No special handling needed. Checking if it's gonna fit probably still is a good idea.
if (ringbufferFreeMem(rb) < sizeof(buf_entry_hdr_t)+rbuffer_size) {
//Buffer is not going to fit, period.
return pdFALSE;
}
}
//If we are here, the buffer is guaranteed to fit in the space starting at the write pointer.
buf_entry_hdr_t *hdr=(buf_entry_hdr_t *)rb->write_ptr;
hdr->len=buffer_size;
hdr->flags=0;
rb->write_ptr+=sizeof(buf_entry_hdr_t);
memcpy(rb->write_ptr, buffer, buffer_size);
rb->write_ptr+=rbuffer_size;
//The buffer will wrap around if we don't have room for a header anymore.
if ((rb->data+rb->size)-rb->write_ptr < sizeof(buf_entry_hdr_t)) {
//'Forward' the write buffer until we are at the start of the ringbuffer.
//The read pointer will always be at the start of a full header, which cannot
//exist at the point of the current write pointer, so there's no chance of overtaking
//that.
rb->write_ptr=rb->data;
}
return pdTRUE;
}
//Copies a bunch of daya to the ring bytebuffer. Assumes there is space in the ringbuffer and
//the ringbuffer is locked. Increases write_ptr to the next item. Returns pdTRUE on
//success, pdFALSE if it can't make the item fit and the calling routine needs to retry
//later or fail.
//This function by itself is not threadsafe, always call from within a muxed section.
static BaseType_t copyItemToRingbufByteBuf(ringbuf_t *rb, uint8_t *buffer, size_t buffer_size)
{
size_t rem_len=(rb->data + rb->size) - rb->write_ptr; //length remaining until end of ringbuffer
//See if we have enough contiguous space to write the buffer.
if (rem_len < buffer_size) {
//...Nope. Write the data bit that fits.
memcpy(rb->write_ptr, buffer, rem_len);
//Update vars so the code later on will write the rest of the data.
buffer+=rem_len;
buffer_size-=rem_len;
rb->write_ptr=rb->data;
}
//If we are here, the buffer is guaranteed to fit in the space starting at the write pointer.
memcpy(rb->write_ptr, buffer, buffer_size);
rb->write_ptr+=buffer_size;
//The buffer will wrap around if we're at the end.
if ((rb->data+rb->size)==rb->write_ptr) {
rb->write_ptr=rb->data;
}
return pdTRUE;
}
//Retrieves a pointer to the data of the next item, or NULL if this is not possible.
//This function by itself is not threadsafe, always call from within a muxed section.
static uint8_t *getItemFromRingbuf(ringbuf_t *rb, size_t *length)
//Because we always return one item, this function ignores the wanted_length variable.
static uint8_t *getItemFromRingbufDefault(ringbuf_t *rb, size_t *length, int wanted_length)
{
uint8_t *ret;
configASSERT(((int)rb->read_ptr&3)==0);
@@ -210,10 +292,48 @@ static uint8_t *getItemFromRingbuf(ringbuf_t *rb, size_t *length)
return ret;
}
//Retrieves a pointer to the data in the buffer, or NULL if this is not possible.
//This function by itself is not threadsafe, always call from within a muxed section.
//This function honours the wanted_length and will never return more data than this.
static uint8_t *getItemFromRingbufByteBuf(ringbuf_t *rb, size_t *length, int wanted_length)
{
uint8_t *ret;
if (rb->read_ptr != rb->free_ptr) {
//This type of ringbuff does not support multiple outstanding buffers.
return NULL;
}
if (rb->read_ptr == rb->write_ptr) {
//No data available.
return NULL;
}
ret=rb->read_ptr;
if (rb->read_ptr > rb->write_ptr) {
//Available data wraps around. Give data until the end of the buffer.
*length=rb->size-(rb->read_ptr - rb->data);
if (wanted_length != 0 && *length > wanted_length) {
*length=wanted_length;
rb->read_ptr+=wanted_length;
} else {
rb->read_ptr=rb->data;
}
} else {
//Return data up to write pointer.
*length=rb->write_ptr -rb->read_ptr;
if (wanted_length != 0 && *length > wanted_length) {
*length=wanted_length;
rb->read_ptr+=wanted_length;
} else {
rb->read_ptr=rb->write_ptr;
}
}
return ret;
}
//Returns an item to the ringbuffer. Will mark the item as free, and will see if the free pointer
//can be increase.
//This function by itself is not threadsafe, always call from within a muxed section.
static void returnItemToRingbuf(ringbuf_t *rb, void *item) {
static void returnItemToRingbufDefault(ringbuf_t *rb, void *item) {
uint8_t *data=(uint8_t*)item;
configASSERT(((int)rb->free_ptr&3)==0);
configASSERT(data >= rb->data);
@@ -249,6 +369,17 @@ static void returnItemToRingbuf(ringbuf_t *rb, void *item) {
}
//Returns an item to the ringbuffer. Will mark the item as free, and will see if the free pointer
//can be increase.
//This function by itself is not threadsafe, always call from within a muxed section.
static void returnItemToRingbufBytebuf(ringbuf_t *rb, void *item) {
uint8_t *data=(uint8_t*)item;
configASSERT(data >= rb->data);
configASSERT(data < rb->data+rb->size);
//Free the read memory.
rb->free_ptr=rb->read_ptr;
}
void xRingbufferPrintInfo(RingbufHandle_t ringbuf)
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
@@ -259,7 +390,7 @@ void xRingbufferPrintInfo(RingbufHandle_t ringbuf)
RingbufHandle_t xRingbufferCreate(size_t buf_length, BaseType_t allow_split_items)
RingbufHandle_t xRingbufferCreate(size_t buf_length, ringbuf_type_t type)
{
ringbuf_t *rb = malloc(sizeof(ringbuf_t));
if (rb==NULL) goto err;
@@ -273,9 +404,35 @@ RingbufHandle_t xRingbufferCreate(size_t buf_length, BaseType_t allow_split_item
rb->free_space_sem = xSemaphoreCreateBinary();
rb->items_buffered_sem = xSemaphoreCreateBinary();
rb->flags=0;
if (allow_split_items) rb->flags|=flag_allowsplit;
if (type==RINGBUF_TYPE_ALLOWSPLIT) {
rb->flags|=flag_allowsplit;
rb->copyItemToRingbufImpl=copyItemToRingbufAllowSplit;
rb->getItemFromRingbufImpl=getItemFromRingbufDefault;
rb->returnItemToRingbufImpl=returnItemToRingbufDefault;
//Calculate max item size. Worst case, we need to split an item into two, which means two headers of overhead.
rb->maxItemSize=rb->size-(sizeof(buf_entry_hdr_t)*2)-4;
} else if (type==RINGBUF_TYPE_BYTEBUF) {
rb->flags|=flag_bytebuf;
rb->copyItemToRingbufImpl=copyItemToRingbufByteBuf;
rb->getItemFromRingbufImpl=getItemFromRingbufByteBuf;
rb->returnItemToRingbufImpl=returnItemToRingbufBytebuf;
//Calculate max item size. We have no headers and can split anywhere -> size is total size minus one.
rb->maxItemSize=rb->size-1;
} else if (type==RINGBUF_TYPE_NOSPLIT) {
rb->copyItemToRingbufImpl=copyItemToRingbufNoSplit;
rb->getItemFromRingbufImpl=getItemFromRingbufDefault;
rb->returnItemToRingbufImpl=returnItemToRingbufDefault;
//Calculate max item size. Worst case, we have the write ptr in such a position that we are lacking four bytes of free
//memory to put an item into the rest of the memory. If this happens, we have to dummy-fill
//(item_data-4) bytes of buffer, then we only have (size-(item_data-4) bytes left to fill
//with the real item. (item size being header+data)
rb->maxItemSize=(rb->size/2)-sizeof(buf_entry_hdr_t)-4;
} else {
configASSERT(0);
}
if (rb->free_space_sem == NULL || rb->items_buffered_sem == NULL) goto err;
vPortCPUInitializeMutex(&rb->mux);
return (RingbufHandle_t)rb;
err:
@@ -303,18 +460,7 @@ size_t xRingbufferGetMaxItemSize(RingbufHandle_t ringbuf)
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
configASSERT(rb);
//In both cases, we return 4 bytes less than what we actually can have. If the ringbuffer is
//indeed entirely filled, read_ptr==free_ptr, which throws off the free space calculation.
if (rb->flags & flag_allowsplit) {
//Worst case, we need to split an item into two, which means two headers of overhead.
return rb->size-(sizeof(buf_entry_hdr_t)*2)-4;
} else {
//Worst case, we have the write ptr in such a position that we are lacking four bytes of free
//memory to put an item into the rest of the memory. If this happens, we have to dummy-fill
//(item_data-4) bytes of buffer, then we only have (size-(item_data-4) bytes left to fill
//with the real item. (item size being header+data)
return (rb->size/2)-sizeof(buf_entry_hdr_t)-4;
}
return rb->maxItemSize;
}
BaseType_t xRingbufferSend(RingbufHandle_t ringbuf, void *data, size_t dataSize, TickType_t ticks_to_wait)
@@ -352,7 +498,7 @@ BaseType_t xRingbufferSend(RingbufHandle_t ringbuf, void *data, size_t dataSize,
portENTER_CRITICAL(&rb->mux);
//Another thread may have been able to sneak its write first. Check again now we locked the ringbuff, and retry
//everything if this is the case. Otherwise, we can write and are done.
done=copyItemToRingbuf(rb, data, dataSize);
done=rb->copyItemToRingbufImpl(rb, data, dataSize);
portEXIT_CRITICAL(&rb->mux);
}
xSemaphoreGive(rb->items_buffered_sem);
@@ -371,7 +517,7 @@ BaseType_t xRingbufferSendFromISR(RingbufHandle_t ringbuf, void *data, size_t da
//Does not fit in the remaining space in the ringbuffer.
write_succeeded=pdFALSE;
} else {
write_succeeded = copyItemToRingbuf(rb, data, dataSize);
write_succeeded = rb->copyItemToRingbufImpl(rb, data, dataSize);
}
portEXIT_CRITICAL_ISR(&rb->mux);
if (write_succeeded) {
@@ -381,7 +527,7 @@ BaseType_t xRingbufferSendFromISR(RingbufHandle_t ringbuf, void *data, size_t da
}
void *xRingbufferReceive(RingbufHandle_t ringbuf, size_t *item_size, TickType_t ticks_to_wait)
static void *xRingbufferReceiveGeneric(RingbufHandle_t ringbuf, size_t *item_size, TickType_t ticks_to_wait, size_t wanted_size)
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
uint8_t *itemData;
@@ -398,7 +544,7 @@ void *xRingbufferReceive(RingbufHandle_t ringbuf, size_t *item_size, TickType_t
}
//Okay, we seem to have data in the buffer. Grab the mux and copy it out if it's still there.
portENTER_CRITICAL(&rb->mux);
itemData=getItemFromRingbuf(rb, item_size);
itemData=rb->getItemFromRingbufImpl(rb, item_size, wanted_size);
portEXIT_CRITICAL(&rb->mux);
if (itemData) {
//We managed to get an item.
@@ -408,6 +554,11 @@ void *xRingbufferReceive(RingbufHandle_t ringbuf, size_t *item_size, TickType_t
return (void*)itemData;
}
void *xRingbufferReceive(RingbufHandle_t ringbuf, size_t *item_size, TickType_t ticks_to_wait)
{
return xRingbufferReceiveGeneric(ringbuf, item_size, ticks_to_wait, 0);
}
void *xRingbufferReceiveFromISR(RingbufHandle_t ringbuf, size_t *item_size)
{
@@ -415,7 +566,28 @@ void *xRingbufferReceiveFromISR(RingbufHandle_t ringbuf, size_t *item_size)
uint8_t *itemData;
configASSERT(rb);
portENTER_CRITICAL_ISR(&rb->mux);
itemData=getItemFromRingbuf(rb, item_size);
itemData=rb->getItemFromRingbufImpl(rb, item_size, 0);
portEXIT_CRITICAL_ISR(&rb->mux);
return (void*)itemData;
}
void *xRingbufferReceiveUpTo(RingbufHandle_t ringbuf, size_t *item_size, TickType_t ticks_to_wait, size_t wanted_size) {
ringbuf_t *rb=(ringbuf_t *)ringbuf;
if (wanted_size == 0) return NULL;
configASSERT(rb);
configASSERT(rb->flags & flag_bytebuf);
return xRingbufferReceiveGeneric(ringbuf, item_size, ticks_to_wait, wanted_size);
}
void *xRingbufferReceiveUpToFromISR(RingbufHandle_t ringbuf, size_t *item_size, size_t wanted_size)
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
uint8_t *itemData;
if (wanted_size == 0) return NULL;
configASSERT(rb);
configASSERT(rb->flags & flag_bytebuf);
portENTER_CRITICAL_ISR(&rb->mux);
itemData=rb->getItemFromRingbufImpl(rb, item_size, 0);
portEXIT_CRITICAL_ISR(&rb->mux);
return (void*)itemData;
}
@@ -425,7 +597,7 @@ void vRingbufferReturnItem(RingbufHandle_t ringbuf, void *item)
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
portENTER_CRITICAL_ISR(&rb->mux);
returnItemToRingbuf(rb, item);
rb->returnItemToRingbufImpl(rb, item);
portEXIT_CRITICAL_ISR(&rb->mux);
xSemaphoreGive(rb->free_space_sem);
}
@@ -435,7 +607,7 @@ void vRingbufferReturnItemFromISR(RingbufHandle_t ringbuf, void *item, BaseType_
{
ringbuf_t *rb=(ringbuf_t *)ringbuf;
portENTER_CRITICAL_ISR(&rb->mux);
returnItemToRingbuf(rb, item);
rb->returnItemToRingbufImpl(rb, item);
portEXIT_CRITICAL_ISR(&rb->mux);
xSemaphoreGiveFromISR(rb->free_space_sem, higher_prio_task_awoken);
}