Zephyr Deep Dive: Ring Buffers (2023)

Zephyr includes many built-in features like stacks for networking and BLE, Flashstorage APIs, and many kernel services. These components allow you to quicklyget up and running with a project and maintain less code! Taking advantage ofthese is a huge win for small firmware teams and was a huge motivation inbringing Zephyr to my teams.

This post covers Zephyr’s built-in ring buffer API, a componentcommonly used in producer-consumer scenarios. We will cover how ring buffersin Zephyr work, when to use them, and their strengths and weaknesses. This postwill close with an example of augmenting ring buffers with waiting capabilities.

Like Interrupt? Subscribe to get our latest posts straight to your mailbox.

Table of Contents

  • Zephyr Ring Buffers
    • Bytes API
    • Items API
  • Ring Buffer Weaknesses
  • Augmenting With Wait Capabilities
  • Wrap-Up

Zephyr Ring Buffers

Ring buffers1 in Zephyr provide a way to pass data through a shared memorybuffer. Ring buffers are safely used in single-consumer, single-producerscenarios. Their usage generally follows the diagram below:

Zephyr Deep Dive: Ring Buffers (1)

The producer writes to the ring buffer via the ring_buf_put functions, whilethe reader reads from the ring buffer with the ring_buf_get functions. Next,we will explore the two interfaces provided: the bytes-based API and theitems-based API.

Bytes API

The ring buffer bytes API is often used in driver implementations, where a pieceof hardware needs to send chunks of data up to a higher layer for furtherprocessing. We will look at the eswifi driver2 as a guide for ring bufferbytes API operation and walk through some of the driver code to understand itsuse. This driver uses AT commands to interface with a WiFi module over a UART.Let’s start with the initialization code:

struct eswifi_uart_data { // ... rest of code uint8_t iface_rb_buf[ESWIFI_RING_BUF_SIZE]; struct ring_buf rx_rb;};int eswifi_uart_init(struct eswifi_dev *eswifi){ // ... rest of code ring_buf_init(&uart->rx_rb, sizeof(uart->iface_rb_buf), uart->iface_rb_buf); // ... rest of code}

We can see from this initialization that we have two components to set up. Thestructure containing the state of the ring buffer object,eswifi_uart_data.rx_rb, and the underlying shared buffer,eswifi_uart_data.iface_rb_buf. Next, let’s examine how the driver writes tothe ring buffer:

static void eswifi_iface_uart_isr(const struct device *uart_dev, void *user_data){struct eswifi_uart_data *uart = &eswifi_uart0; /* Static instance */ int rx = 0; uint8_t *dst; uint32_t partial_size = 0; uint32_t total_size = 0; ARG_UNUSED(user_data); while (uart_irq_update(uart->dev) && uart_irq_rx_ready(uart->dev)) { if (!partial_size) { partial_size = ring_buf_put_claim(&uart->rx_rb, &dst, UINT32_MAX); } if (!partial_size) { LOG_ERR("Rx buffer doesn't have enough space"); eswifi_iface_uart_flush(uart); break; } rx = uart_fifo_read(uart->dev, dst, partial_size); if (rx <= 0) { continue; } dst += rx; total_size += rx; partial_size -= rx; } ring_buf_put_finish(&uart->rx_rb, total_size);}

When the MCU receives data over the UART from the WiFi module, the UARTperipheral runs the interrupt handler, eswifi_iface_uart_isr. Next, theinterrupt handler repeatedly checks for available data. If data is available,the driver calls ring_buf_put_claim. The put_claim operation reserves up tothe provided size in the ring buffer for writing and returns the number of bytesclaimed. A return of 0 indicates the ring buffer is full. The write operationthen becomes a simple copy into the shared buffer. After copying the data, thehandler calls ring_buf_put_finish to signal completion.

If the ring buffer is full (i.e. the put_claim returns 0), the driver dropsthe current batch of data received because we cannot block to wait for new datain this interrupt. Resolving this issue may require increasing the ring buffersize to account for pressure on the ring buffer. Problems like these can be hardto trace the specific instance when the buffer fills completely. One way todetermine the proper size is to add an assertion when the put_claim fails toreserve any data. Asserting in this manner is a technique of offensiveprogramming3, as this modification proactively looks for an issue and failspurposefully.

The main driver work is done within the context of a dedicated driver workqueue.The workqueue operates as a dedicated cooperative thread that sleeps until newwork is submitted. In our case, the work starts with requests sent to the WiFimodule and ends with responses sent back. The code below shows how the function,eswifi_uart_get_resp reads data from the ring buffer to parse responses fromthe WiFi module:

static int eswifi_uart_get_resp(struct eswifi_uart_data *uart){ uint8_t c; while (ring_buf_get(&uart->rx_rb, &c, 1) > 0) { LOG_DBG("FSM: %c, RX: 0x%02x : %c", get_fsm_char(uart->fsm), c, c); if (uart->rx_buf_size > 0) { uart->rx_buf[uart->rx_count++] = c; if (uart->rx_count == uart->rx_buf_size) { return -ENOMEM; } } // ... rest of code }// ... rest of code}

Two things stand out compared to the code that wrote data to the ring buffer.The first is the usage of ring_buf_get instead of the claim-based API. When theinterrupt handler receives UART data, the handler does not immediately know howmuch to write into the ring buffer. The claim-based function allows the handlerto use the buffer directly, minimizing copying operations. However, the driverfunction, eswifi_uart_get_resp, must copy the data for later processing. Thereis a bit of an asymmetry by design here. The interrupt handler does do anyprocessing on the data. It simply hands the data off to the workqueue and moveson to the next read. The workqueue, on the other hand, does need to keep thisdata around to complete processing the response! The utility of the ring bufferis that it supports both cases and allows for clear code.

The second thing to note is that the interrupt handler writes in variablelengths while read operations are a constant length. The ring buffer offersflexibility in the amount of data being written to or read from the buffer.

Items API

In addition to the standard bytes-based API, there is the ring buffer itemsAPI. In this version, data is written to and read from the ring buffer as anitem with three components:

  1. An application-defined type
  2. An application-defined integer value
  3. An optional array of associated data

This metadata adds some structure to the data. The type helps identify what theitem contains. The integer value represents either additional metadata or thevalue of the item. We can view ring-buffers using this API as better suited topassing data with some structure rather than a raw array of bytes (e.g.packet-oriented protocols). For instance, we could parse different items by typeto produce different messages, or route data to different destinations based onthe integer value.

Ring Buffer Weaknesses

There are a few weaknesses present in ring buffers. First, they requireconcurrency protections when their usage expands to multiple producers orconsumers. It is safe to use with a single producer and a single consumer, butexpanding beyond this simple case will require additional design, especially ifused in an interrupt context. The ring buffer does have separate state variablesfor reading and writing. These allow for the scenario where an interrupt handleris the only producer, allowing it to write without acquiring a lock, whilemultiple consumer threads must use a lock for exclusive access.

The second weakness is ring buffers lack synchronization capabilities. Producerscannot wait until space is available in the buffer. Consumers cannot wait untildata is available in the buffer. The ring buffer APIs are non-blocking, theyreturn immediately. We could use the ring_buf_is_empty function to poll for achange, but this introduces many undesirable effects. A better option would beto use a kernel synchronization primitive, such as a semaphore, to augment thering buffer.

Augmenting With Wait Capabilities

In the simplest case of a single producer and consumer, the producer can use asemaphore to signal to the consumer data is available. Let’s take a look at theeswifi driver, but this time at the function that wrapseswifi_uart_get_resp:

static int eswifi_uart_wait_prompt(struct eswifi_uart_data *uart){ unsigned int max_retries = 60 * 1000; /* 1 minute */ int err; while (--max_retries) { err = eswifi_uart_get_resp(uart); if (err) { LOG_DBG("Err: 0x%08x - %d", err, err); return err; } if (uart->fsm == ESWIFI_UART_FSM_END) { LOG_DBG("Success!"); return uart->rx_count; } /* allow other threads to be scheduled */ k_sleep(K_MSEC(1)); } LOG_DBG("Timeout"); return -ETIMEDOUT;}

We can see that this function relies on the response being sent within 60seconds of the call to wait for the response. In the worst case, this code willrequire a context-switch 60,000 times as each call to k_sleep allows a readythread to take over. Additionally, many short sleeps like this introduce morejitter than a single timeout. So not only does this cause many more contextswitches, but it is also less accurate in terms of timing. A semaphore signaledfrom eswifi_iface_uart_isr would only require a single switch after timingout. Here is an example of what this improvement could look like:

// Initialize the semaphore with an initial count of 0 and a max of 1K_SEM_DEFINE(response_sem, 0, 1);static void eswifi_iface_uart_isr(const struct device *uart_dev, void *user_data){ // ... rest of code ring_buf_put_finish(&uart->rx_rb, total_size); // Raise signal to eswifi_uart_wait_prompt k_sem_give(&response_sem);}static int eswifi_uart_wait_prompt(struct eswifi_uart_data *uart) { int err; // Wait for the UART state machine to reach end state // Timeout occurs after 10 seconds since last data received while (uart->fsm != ESWIFI_UART_FSM_END) { // Returns non-zero value on timeout if (k_sem_take(&response_sem, K_SECONDS(10))) { return -ETIMEDOUT; } err = eswifi_uart_get_resp(uart); if (err) { LOG_DBG("Err: 0x%08x - %d", err, err); return err; } } LOG_DBG("Success"); return uart->rx_count;}

The reworked code has several advantages:

  • A more responsive driver when waiting for responses (pun intended)
  • eswifi_uart_wait_prompt has simpler logic. It fails quickly if it does notreceive data
  • We reduce the number of unnecessary context switches
  • The macro, K_SEM_DEFINE4, prevents annoying race conditions if the kernelobject has not yet been initialized before the UART interrupt is enabled

The only disadvantage here is a small hit to RAM due to the new semaphore. Ithink this is a worthwhile trade-off.


Zephyr has a variety of components that we can combine to suit practicallyanything your device requires. The hard part can be knowing where and what tostart with. As we have covered in this post, ring buffers shine when used toimplement single producer single consumer designs. The ring buffer designintroduces little overhead and is efficient in reading and writing through theuse of its internal buffer and indices. The data structure is flexible inmultiple ways. It allows for a variable length of data operations. It offersmultiple data passing methods, whether through direct copy or referencing theinternal buffer. Finally, its API offers either an unstructured bytes-orientedAPI or a more structured items API geared toward messages.

This post touches on just one of the many features provided out of the box byZephyr. Future posts in this series will examine its other offerings such asdata passing with FIFOs and mailboxes, memory management with slabs and pools,and subsystems like RTIO and zbus. Please feel free to leave feedback, comments,and Zephyr topic suggestions below!

Like Interrupt? Subscribe to get our latest posts straight to your mailbox.

See anything you'd like to change? Submit a pull request or open an issue at GitHub


  1. Zephyr Ring Buffer API

  2. eswifi driver

  3. Offensive Programming

  4. Zephyr Semaphore API

Zephyr Deep Dive: Ring Buffers (2) Eric Johnson is a Firmware Solutions Engineer at Memfault. Eric previously worked on embedded software teams at Walgreens Health and Athos

Top Articles
Latest Posts
Article information

Author: Clemencia Bogisich Ret

Last Updated: 02/20/2023

Views: 6605

Rating: 5 / 5 (80 voted)

Reviews: 95% of readers found this page helpful

Author information

Name: Clemencia Bogisich Ret

Birthday: 2001-07-17

Address: Suite 794 53887 Geri Spring, West Cristentown, KY 54855

Phone: +5934435460663

Job: Central Hospitality Director

Hobby: Yoga, Electronics, Rafting, Lockpicking, Inline skating, Puzzles, scrapbook

Introduction: My name is Clemencia Bogisich Ret, I am a super, outstanding, graceful, friendly, vast, comfortable, agreeable person who loves writing and wants to share my knowledge and understanding with you.