Ring Buffers

A ring buffer is a circular buffer, whose contents are stored in first-in-first-out order.

For circumstances where an application needs to implement asynchronous “streaming” copying of data, Zephyr provides a struct ring_buf abstraction to manage copies of such data in and out of a shared buffer of memory.

Two content data modes are supported:

Data item mode: Multiple 32-bit word data items with metadata can be enqueued and dequeued from the ring buffer in chunks of up to 1020 bytes. Each data item also has two associated metadata values: a type identifier and a 16-bit integer value, both of which are application-specific.
Byte mode: raw bytes can be enqueued and dequeued.

While the underlying data structure is the same, it is not legal to mix these two modes on a single ring buffer instance. A ring buffer initialized with a byte count must be used only with the “bytes” API, one initialized with a word count must use the “items” calls.

Concepts
Implementation
Configuration Options
API Reference

Concepts 

Any number of ring buffers can be defined (limited only by available RAM). Each ring buffer is referenced by its memory address.

A ring buffer has the following key properties:

A data buffer of 32-bit words or bytes. The data buffer contains the data items or raw bytes that have been added to the ring buffer but not yet removed.
A data buffer size, measured in 32-bit words or bytes. This governs the maximum amount of data (including metadata values) the ring buffer can hold.

A ring buffer must be initialized before it can be used. This sets its data buffer to empty.

A struct ring_buf may be placed anywhere in user-accessible memory, and must be initialized with ring_buf_init() before use. This must be provided a region of user-controlled memory for use as the buffer itself. Note carefully that the units of the size of the buffer passed change (either bytes or words) depending on how the ring buffer will be used later. Macros for combining these steps in a single static declaration exist for convenience. RING_BUF_DECLARE will declare and statically initialize a ring buffer with a specified byte count, where RING_BUF_ITEM_DECLARE_SIZE will declare and statically initialize a buffer with a given count of 32 bit words. RING_BUF_ITEM_DECLARE_POW2 can be used to initialize an items-mode buffer with a memory region guaranteed to be a power of two, which enables various optimizations internal to the implementation. No power-of-two initialization is available for bytes-mode ring buffers.

“Bytes” data may be copied into the ring buffer using ring_buf_put(), passing a data pointer and byte count. These bytes will be copied into the buffer in order, as many as will fit in the allocated buffer. The total number of bytes copied (which may be fewer than provided) will be returned. Likewise ring_buf_get() will copy bytes out of the ring buffer in the order that they were written, into a user-provided buffer, returning the number of bytes that were transferred.

To avoid multiply-copied-data situations, a “claim” API exists for byte mode. ring_buf_put_claim() takes a byte size value from the user and returns a pointer to memory internal to the ring buffer that can be used to receive those bytes, along with a size of the contiguous internal region (which may be smaller than requested). The user can then copy data into that region at a later time without assembling all the bytes in a single region first. When complete, ring_buf_put_finish() can be used to signal the buffer that the transfer is complete, passing the number of bytes actually transferred. At this point a new transfer can be initiated. Similarly, ring_buf_get_claim() returns a pointer to internal ring buffer data from which the user can read without making a verbatim copy, and ring_buf_get_finish() signals the buffer with how many bytes have been consumed and allows for a new transfer to begin.

“Items” mode works similarly to bytes mode, except that all transfers are in units of 32 bit words and all memory is assumed to be aligned on 32 bit boundaries. The write and read operations are ring_buf_item_put() and ring_buf_item_get(), and work otherwise identically to the bytes mode APIs. There no “claim” API provided for items mode. One important difference is that unlike ring_buf_put(), ring_buf_item_put() will not do a partial transfer; it will return an error in the case where the provided data does not fit in its entirety.

The user can manage the capacity of a ring buffer without modifying it using the ring_buf_space_get() call (which returns a value of either bytes or items depending on how the ring buffer has been used), or by testing the ring_buf_is_empty() predicate.

Finally, a ring_buf_reset() call exists to immediately empty a ring buffer, discarding the tracking of any bytes or items already written to the buffer. It does not modify the memory contents of the buffer itself, however.

Data item mode 

A data item mode ring buffer instance is declared using RING_BUF_ITEM_DECLARE_POW2() or RING_BUF_ITEM_DECLARE_SIZE() and accessed using ring_buf_item_put() and ring_buf_item_get().

A ring buffer data item is an array of 32-bit words from 0 to 1020 bytes in length. When a data item is enqueued (ring_buf_item_put()) its contents are copied to the data buffer, along with its associated metadata values (which occupy one additional 32-bit word). If the ring buffer has insufficient space to hold the new data item the enqueue operation fails.

A data items is dequeued (ring_buf_item_get()) from a ring buffer by removing the oldest enqueued item. The contents of the dequeued data item, as well as its two metadata values, are copied to areas supplied by the retriever. If the ring buffer is empty, or if the data array supplied by the retriever is not large enough to hold the data item’s data, the dequeue operation fails.

Byte mode 

A byte mode ring buffer instance is declared using RING_BUF_ITEM_DECLARE_SIZE() and accessed using: ring_buf_put_claim(), ring_buf_put_finish(), ring_buf_get_claim(), ring_buf_get_finish(), ring_buf_put() and ring_buf_get().

Data can be copied into the ring buffer (see ring_buf_put()) or ring buffer memory can be used directly by the user. In the latter case, the operation is split into three stages:

allocating the buffer (ring_buf_put_claim()) when user requests the destination location where data can be written.
writing the data by the user (e.g. buffer written by DMA).
indicating the amount of data written to the provided buffer (ring_buf_put_finish()). The amount can be less than or equal to the allocated amount.

Data can be retrieved from a ring buffer through copying (see ring_buf_get()) or accessed directly by address. In the latter case, the operation is split into three stages:

retrieving source location with valid data written to a ring buffer (see ring_buf_get_claim()).
processing data
freeing processed data (see ring_buf_get_finish()). The amount freed can be less than or equal or to the retrieved amount.

Concurrency 

The ring buffer APIs do not provide any concurrency control. Depending on usage (particularly with respect to number of concurrent readers/writers) applications may need to protect the ring buffer with mutexes and/or use semaphores to notify consumers that there is data to read.

For the trivial case of one producer and one consumer, concurrency shouldn’t be needed.

Internal Operation 

If the size of the data buffer is a power of two, the ring buffer uses efficient masking operations instead of expensive modulo operations when enqueuing and dequeuing data items. This option is applicable only for data item mode.

Data streamed through a ring buffer is always written to the next byte within the buffer, wrapping around to the first element after reaching the end, thus the “ring” structure. Internally, the struct ring_buf contains its own buffer pointer and its size, and also a “head” and “tail” index representing where the next read and write

This boundary is invisible to the user using the normal put/get APIs, but becomes a barrier to the “claim” API, because obviously no contiguous region can be returned that crosses the end of the buffer. This can be surprising to application code, and produce performance artifacts when transfers need to alias closely to the size of the buffer, as the number of calls to claim/finish need to double for such transfers.

When running in items mode (only), the ring buffer contains two implementations for the modular arithmetic required to compute “next element” offsets. One is used for arbitrary sized buffers, but the other is optimized for power of two sizes and can replace the compare and subtract steps with a simple bitmask in several places, at the cost of testing the “mask” value for each call.

Implementation 

Defining a Ring Buffer 

A ring buffer is defined using a variable of type ring_buf. It must then be initialized by calling ring_buf_init().

The following code defines and initializes an empty data item mode ring buffer (which is part of a larger data structure). The ring buffer’s data buffer is capable of holding 64 words of data and metadata information.

#define MY_RING_BUF_SIZE 64

struct my_struct {
    struct ring_buf rb;
    uint32_t buffer[MY_RING_BUF_SIZE];
    ...
};
struct my_struct ms;

void init_my_struct {
    ring_buf_init(&ms.rb, sizeof(ms.buffer), ms.buffer);
    ...
}

Alternatively, a ring buffer can be defined and initialized at compile time using one of two macros at file scope. Each macro defines both the ring buffer itself and its data buffer.

The following code defines a ring buffer with a power-of-two sized data buffer, which can be accessed using efficient masking operations.

/* Buffer with 2^8 (or 256) words */
RING_BUF_ITEM_DECLARE_POW2(my_ring_buf, 8);

The following code defines an application-specific sized byte mode ring buffer enqueued and dequeued as raw bytes:

#define MY_RING_BUF_WORDS 93
RING_BUF_ITEM_DECLARE_SIZE(my_ring_buf, MY_RING_BUF_WORDS);

The following code defines a ring buffer with an arbitrary-sized data buffer, which can be accessed using less efficient modulo operations. Ring buffer is intended to be used for raw bytes.

#define MY_RING_BUF_BYTES 93
RING_BUF_DECLARE_SIZE(my_ring_buf, MY_RING_BUF_BYTES);

Enqueuing Data 

A data item is added to a ring buffer by calling ring_buf_item_put().

uint32_t data[MY_DATA_WORDS];
int ret;

ret = ring_buf_item_put(&ring_buf, TYPE_FOO, 0, data, SIZE32_OF(data));
if (ret == -EMSGSIZE) {
    /* not enough room for the data item */
    ...
}

If the data item requires only the type or application-specific integer value (i.e. it has no data array), a size of 0 and data pointer of NULL can be specified.

int ret;

ret = ring_buf_item_put(&ring_buf, TYPE_BAR, 17, NULL, 0);
if (ret == -EMSGSIZE) {
    /* not enough room for the data item */
    ...
}

Bytes are copied to a byte mode ring buffer by calling ring_buf_put().

uint8_t my_data[MY_RING_BUF_BYTES];
uint32_t ret;

ret = ring_buf_put(&ring_buf, my_data, SIZE_OF(my_data));
if (ret != SIZE_OF(my_data)) {
    /* not enough room, partial copy. */
    ...
}

Data can be added to a byte mode ring buffer by directly accessing the ring buffer’s memory. For example:

uint32_t size;
uint32_t rx_size;
uint8_t *data;
int err;

/* Allocate buffer within a ring buffer memory. */
size = ring_buf_put_claim(&ring_buf, &data, MY_RING_BUF_BYTES);

/* Work directly on a ring buffer memory. */
rx_size = uart_rx(data, size);

/* Indicate amount of valid data. rx_size can be equal or less than size. */
err = ring_buf_put_finish(&ring_buf, rx_size);
if (err != 0) {
    /* No space to put requested amount of data to ring buffer. */
    ...
}

Retrieving Data 

A data item is removed from a ring buffer by calling ring_buf_item_get().

uint32_t my_data[MY_DATA_WORDS];
uint16_t my_type;
uint8_t  my_value;
uint8_t  my_size;
int ret;

my_size = SIZE32_OF(my_data);
ret = ring_buf_item_get(&ring_buf, &my_type, &my_value, my_data, &my_size);
if (ret == -EMSGSIZE) {
    printk("Buffer is too small, need %d uint32_t\n", my_size);
} else if (ret == -EAGAIN) {
    printk("Ring buffer is empty\n");
} else {
    printk("Got item of type %u value &u of size %u dwords\n",
           my_type, my_value, my_size);
    ...
}

Data bytes are copied out from a byte mode ring buffer by calling ring_buf_get(). For example:

uint8_t my_data[MY_DATA_BYTES];
size_t  ret;

ret = ring_buf_get(&ring_buf, my_data, sizeof(my_data));
if (ret != sizeof(my_size)) {
    /* Less bytes copied. */
} else {
    /* Requested amount of bytes retrieved. */
    ...
}

Data can be retrieved from a byte mode ring buffer by direct operations on the ring buffer’s memory. For example:

uint32_t size;
uint32_t proc_size;
uint8_t *data;
int err;

/* Get buffer within a ring buffer memory. */
size = ring_buf_get_claim(&ring_buf, &data, MY_RING_BUF_BYTES);

/* Work directly on a ring buffer memory. */
proc_size = process(data, size);

/* Indicate amount of data that can be freed. proc_size can be equal or less
 * than size.
 */
err = ring_buf_get_finish(&ring_buf, proc_size);
if (err != 0) {
    /* proc_size exceeds amount of valid data in a ring buffer. */
    ...
}

Configuration Options 

Related configuration options:

CONFIG_RING_BUFFER: Enable ring buffer.

API Reference 

The following ring buffer APIs are provided by include/sys/ring_buffer.h:

group ring_buffer_apis

Defines

RING_BUF_ITEM_DECLARE_POW2(name, pow)

Define and initialize a high performance ring buffer.

This macro establishes a ring buffer whose size must be a power of 2; that is, the ring buffer contains 2^pow 32-bit words, where pow is the specified ring buffer size exponent. A high performance ring buffer doesn’t require the use of modulo arithmetic operations to maintain itself.

The ring buffer can be accessed outside the module where it is defined using:

extern struct ring_buf <name>; 

Parameters

name – Name of the ring buffer.
pow – Ring buffer size exponent.

RING_BUF_ITEM_DECLARE_SIZE(name, size32)

Define and initialize a standard ring buffer.

This macro establishes a ring buffer of an arbitrary size. A standard ring buffer uses modulo arithmetic operations to maintain itself.

The ring buffer can be accessed outside the module where it is defined using:

extern struct ring_buf <name>; 

Parameters

name – Name of the ring buffer.
size32 – Size of ring buffer (in 32-bit words).

RING_BUF_DECLARE(name, size8)

Define and initialize a ring buffer for byte data.

This macro establishes a ring buffer of an arbitrary size.

The ring buffer can be accessed outside the module where it is defined using:

extern struct ring_buf <name>; 

Parameters

name – Name of the ring buffer.
size8 – Size of ring buffer (in bytes).

Functions

static inline void ring_buf_init(struct ring_buf *buf, uint32_t size, void *data)

Initialize a ring buffer.

This routine initializes a ring buffer, prior to its first use. It is only used for ring buffers not defined using RING_BUF_DECLARE, RING_BUF_ITEM_DECLARE_POW2 or RING_BUF_ITEM_DECLARE_SIZE.

Setting size to a power of 2 establishes a high performance ring buffer that doesn’t require the use of modulo arithmetic operations to maintain itself.

Parameters

buf – Address of ring buffer.
size – Ring buffer size (in 32-bit words or bytes).
data – Ring buffer data area (uint32_t data[size] or uint8_t data[size] for bytes mode).

int ring_buf_is_empty(struct ring_buf *buf)

Determine if a ring buffer is empty.

Parameters

buf – Address of ring buffer.

Returns

1 if the ring buffer is empty, or 0 if not.

static inline void ring_buf_reset(struct ring_buf *buf)

Reset ring buffer state.

Parameters

buf – Address of ring buffer.

uint32_t ring_buf_space_get(struct ring_buf *buf)

Determine free space in a ring buffer.

Parameters

buf – Address of ring buffer.

Returns

Ring buffer free space (in 32-bit words or bytes).

static inline uint32_t ring_buf_capacity_get(struct ring_buf *buf)

Return ring buffer capacity.

Parameters

buf – Address of ring buffer.

Returns

Ring buffer capacity (in 32-bit words or bytes).

uint32_t ring_buf_size_get(struct ring_buf *buf)

Determine used space in a ring buffer.

Parameters

buf – Address of ring buffer.

Returns

Ring buffer space used (in 32-bit words or bytes).

int ring_buf_item_put(struct ring_buf *buf, uint16_t type, uint8_t value, uint32_t *data, uint8_t size32)

Write a data item to a ring buffer.

This routine writes a data item to ring buffer buf. The data item is an array of 32-bit words (from zero to 1020 bytes in length), coupled with a 16-bit type identifier and an 8-bit integer value.

Warning

Use cases involving multiple writers to the ring buffer must prevent concurrent write operations, either by preventing all writers from being preempted or by using a mutex to govern writes to the ring buffer.

Parameters

buf – Address of ring buffer.
type – Data item’s type identifier (application specific).
value – Data item’s integer value (application specific).
data – Address of data item.
size32 – Data item size (number of 32-bit words).

Return values

0 – Data item was written.
-EMSGSIZE – Ring buffer has insufficient free space.

int ring_buf_item_get(struct ring_buf *buf, uint16_t *type, uint8_t *value, uint32_t *data, uint8_t *size32)

Read a data item from a ring buffer.

This routine reads a data item from ring buffer buf. The data item is an array of 32-bit words (up to 1020 bytes in length), coupled with a 16-bit type identifier and an 8-bit integer value.

Warning

Use cases involving multiple reads of the ring buffer must prevent concurrent read operations, either by preventing all readers from being preempted or by using a mutex to govern reads to the ring buffer.

Parameters

buf – Address of ring buffer.
type – Area to store the data item’s type identifier.
value – Area to store the data item’s integer value.
data – Area to store the data item. Can be NULL to discard data.
size32 – Size of the data item storage area (number of 32-bit chunks).

Return values

0 – Data item was fetched; size32 now contains the number of 32-bit words read into data area data.
-EAGAIN – Ring buffer is empty.
-EMSGSIZE – Data area data is too small; size32 now contains the number of 32-bit words needed.

uint32_t ring_buf_put_claim(struct ring_buf *buf, uint8_t **data, uint32_t size)

Allocate buffer for writing data to a ring buffer.

With this routine, memory copying can be reduced since internal ring buffer can be used directly by the user. Once data is written to allocated area number of bytes written can be confirmed (see ring_buf_put_finish).

Warning

Use cases involving multiple writers to the ring buffer must prevent concurrent write operations, either by preventing all writers from being preempted or by using a mutex to govern writes to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item access (calls prefixed with ring_buf_item_).

Parameters

buf – [in] Address of ring buffer.
data – [out] Pointer to the address. It is set to a location within ring buffer.
size – [in] Requested allocation size (in bytes).

Returns

Size of allocated buffer which can be smaller than requested if there is not enough free space or buffer wraps.

int ring_buf_put_finish(struct ring_buf *buf, uint32_t size)

Indicate number of bytes written to allocated buffers.

Warning

Use cases involving multiple writers to the ring buffer must prevent concurrent write operations, either by preventing all writers from being preempted or by using a mutex to govern writes to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item access (calls prefixed with ring_buf_item_).

Parameters

buf – Address of ring buffer.
size – Number of valid bytes in the allocated buffers.

Return values

0 – Successful operation.
-EINVAL – Provided size exceeds free space in the ring buffer.

uint32_t ring_buf_put(struct ring_buf *buf, const uint8_t *data, uint32_t size)

Write (copy) data to a ring buffer.

This routine writes data to a ring buffer buf.

Warning

Use cases involving multiple writers to the ring buffer must prevent concurrent write operations, either by preventing all writers from being preempted or by using a mutex to govern writes to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item access (calls prefixed with ring_buf_item_).

Parameters

buf – Address of ring buffer.
data – Address of data.
size – Data size (in bytes).

Return values

Number – of bytes written.

uint32_t ring_buf_get_claim(struct ring_buf *buf, uint8_t **data, uint32_t size)

Get address of a valid data in a ring buffer.

With this routine, memory copying can be reduced since internal ring buffer can be used directly by the user. Once data is processed it can be freed using ring_buf_get_finish.

Warning

Use cases involving multiple reads of the ring buffer must prevent concurrent read operations, either by preventing all readers from being preempted or by using a mutex to govern reads to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item access (calls prefixed with ring_buf_item_).

Parameters

buf – [in] Address of ring buffer.
data – [out] Pointer to the address. It is set to a location within ring buffer.
size – [in] Requested size (in bytes).

Returns

Number of valid bytes in the provided buffer which can be smaller than requested if there is not enough free space or buffer wraps.

int ring_buf_get_finish(struct ring_buf *buf, uint32_t size)

Indicate number of bytes read from claimed buffer.

Warning

Use cases involving multiple reads of the ring buffer must prevent concurrent read operations, either by preventing all readers from being preempted or by using a mutex to govern reads to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item mode (calls prefixed with ring_buf_item_).

Parameters

buf – Address of ring buffer.
size – Number of bytes that can be freed.

Return values

0 – Successful operation.
-EINVAL – Provided size exceeds valid bytes in the ring buffer.

uint32_t ring_buf_get(struct ring_buf *buf, uint8_t *data, uint32_t size)

Read data from a ring buffer.

This routine reads data from a ring buffer buf.

Warning

Use cases involving multiple reads of the ring buffer must prevent concurrent read operations, either by preventing all readers from being preempted or by using a mutex to govern reads to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item mode (calls prefixed with ring_buf_item_).

Parameters

buf – Address of ring buffer.
data – Address of the output buffer. Can be NULL to discard data.
size – Data size (in bytes).

Return values

Number – of bytes written to the output buffer.

uint32_t ring_buf_peek(struct ring_buf *buf, uint8_t *data, uint32_t size)

Peek at data from a ring buffer.

This routine reads data from a ring buffer buf without removal.

Warning

Use cases involving multiple reads of the ring buffer must prevent concurrent read operations, either by preventing all readers from being preempted or by using a mutex to govern reads to the ring buffer.

Warning

Ring buffer instance should not mix byte access and item mode (calls prefixed with ring_buf_item_).

Warning

Multiple calls to peek will result in the same data being ‘peeked’ multiple times. To remove data, use either ring_buf_get or ring_buf_get_claim followed by ring_buf_get_finish with a non-zero size.

Parameters

buf – Address of ring buffer.
data – Address of the output buffer. Cannot be NULL.
size – Data size (in bytes).

Return values

Number – of bytes written to the output buffer.