LittleFS on large QSPI NOR

[KRich]

I am having issues using a 2Gb NOR flash part for LittleFS storage on an nRF52840 based project (using the nRF Connect SDK). I am not sure yet if the issue is in the LittleFS module itself, the Zephyr fs (filesystem) module, or the QSPI/NOR driver. The symptom that I am seeing is that when I configure the size of the partition to larger than 16MB, after some amount of time (typically 15-20 minutes), the device begins reporting write errors (I'm writing to the device once per millisecond in my test app, in my final app, the writes will be in the range of every 40ms to every 4ms.

I have opened a ticket in the Zephyr project (https://github.com/zephyrproject-rtos/zephyr/issues/40453) which includes a piece of sample code and description for re-creating the problem, but I thought it might worth posting something here as well to see if anyone else has run into any similar issues.

Thanks!

[KRich]

After setting this field, I see that erasing the partition is much faster (is that expected?) but I am now not able to format the device (error -28 LFS_ERR_NOSPC) which seems strange given that I just erased the entire flash partition (unless, of course the partition didn't actually get erased). I have stepped through the erase operation which appears to work (returns NRF_SUCCESS), but, again, based on timing seems like it might not actually be performing the operation.

[KRich]

I believe I have found the root cause of my issue. Per https://devzone.nordicsemi.com/f/nordic-q-a/81114/zephyr-qspi-enter-32-bit-address-mode the QSPI NOR driver currently included in Zephyr does not enable 32-bit (4 byte) addressing mode. After updating the driver and DTS entry to enable 32-bit addressing mode, the issue appears to be resolved.

This leads to 2 additional questions:

1. In Zephyr/West Is there a way to "override" the driver from the zephyr tree with my modified driver (somewhere in my own project)
2. What is the most effective way to request that this functionality be added to the driver? I have attached my updated/modified changes to the driver as a starting point.

Fullscreen 2364.nrf_qspi_nor.c Download

/*
 * Copyright (c) 2019, Nordic Semiconductor ASA
 *
 * SPDX-License-Identifier: Apache-2.0
 */

#define DT_DRV_COMPAT nordic_qspi_nor

#include <errno.h>
#include <drivers/flash.h>
#include <init.h>
#include <string.h>
#include <logging/log.h>

#include "spi_nor.h"
#include "jesd216.h"
#include "flash_priv.h"
#include <nrfx_qspi.h>
#include <hal/nrf_clock.h>

struct qspi_nor_config {
       /* JEDEC id from devicetree */
       uint8_t id[SPI_NOR_MAX_ID_LEN];

       /* Size from devicetree, in bytes */
       uint32_t size;
};

/* Main config structure */
static nrfx_qspi_config_t QSPIconfig;

/* Status register bits */
#define QSPI_SECTOR_SIZE SPI_NOR_SECTOR_SIZE
#define QSPI_BLOCK_SIZE SPI_NOR_BLOCK_SIZE

/* instance 0 flash size in bytes */
#define INST_0_BYTES (DT_INST_PROP(0, size) / 8)

#define INST_0_SCK_FREQUENCY DT_INST_PROP(0, sck_frequency)
BUILD_ASSERT(INST_0_SCK_FREQUENCY >= (NRF_QSPI_BASE_CLOCK_FREQ / 16),
	     "Unsupported SCK frequency.");

/* for accessing devicetree properties of the bus node */
#define QSPI_NODE DT_BUS(DT_DRV_INST(0))
#define QSPI_PROP_AT(prop, idx) DT_PROP_BY_IDX(QSPI_NODE, prop, idx)
#define QSPI_PROP_LEN(prop) DT_PROP_LEN(QSPI_NODE, prop)

#define INST_0_QER _CONCAT(JESD216_DW15_QER_, \
			   DT_ENUM_TOKEN(DT_DRV_INST(0), \
					 quad_enable_requirements))

BUILD_ASSERT(((INST_0_QER == JESD216_DW15_QER_NONE)
	      || (INST_0_QER == JESD216_DW15_QER_S1B6)),
	     "Driver only supports NONE or S1B6 for quad-enable-requirements");

#define INST_0_4BA DT_INST_PROP(0, enter_4byte_addr)

BUILD_ASSERT(((INST_0_4BA & 0x03) != 0),
	     "Driver only supports command (0xB7) for entering 4 byte addressing mode");

#if NRF52_ERRATA_122_PRESENT
#include <hal/nrf_gpio.h>
static int anomaly_122_init(const struct device *dev);
static void anomaly_122_uninit(const struct device *dev);

#define ANOMALY_122_INIT(dev)   anomaly_122_init(dev)
#define ANOMALY_122_UNINIT(dev) anomaly_122_uninit(dev)
#else
#define ANOMALY_122_INIT(dev) 0
#define ANOMALY_122_UNINIT(dev)
#endif

#define WORD_SIZE 4

LOG_MODULE_REGISTER(qspi_nor, CONFIG_FLASH_LOG_LEVEL);

static const struct flash_parameters qspi_flash_parameters = {
	.write_block_size = 4,
	.erase_value = 0xff,
};

/**
 * @brief QSPI buffer structure
 * Structure used both for TX and RX purposes.
 *
 * @param buf is a valid pointer to a data buffer.
 * Can not be NULL.
 * @param len is the length of the data to be handled.
 * If no data to transmit/receive - pass 0.
 */
struct qspi_buf {
	uint8_t *buf;
	size_t len;
};

/**
 * @brief QSPI command structure
 * Structure used for custom command usage.
 *
 * @param op_code is a command value (i.e 0x9F - get Jedec ID)
 * @param tx_buf structure used for TX purposes. Can be NULL if not used.
 * @param rx_buf structure used for RX purposes. Can be NULL if not used.
 */
struct qspi_cmd {
	uint8_t op_code;
	const struct qspi_buf *tx_buf;
	const struct qspi_buf *rx_buf;
};

/**
 * @brief Structure for defining the QSPI NOR access
 */
struct qspi_nor_data {
#ifdef CONFIG_MULTITHREADING
	/* The semaphore to control exclusive access on write/erase. */
	struct k_sem trans;
	/* The semaphore to control exclusive access to the device. */
	struct k_sem sem;
	/* The semaphore to indicate that transfer has completed. */
	struct k_sem sync;
#if NRF52_ERRATA_122_PRESENT
	/* The semaphore to control driver init/uninit. */
	struct k_sem count;
#endif
#else /* CONFIG_MULTITHREADING */
	/* A flag that signals completed transfer when threads are
	 * not enabled.
	 */
	volatile bool ready;
#endif /* CONFIG_MULTITHREADING */
};

static int qspi_nor_write_protection_set(const struct device *dev,
					 bool write_protect);

static inline int qspi_get_mode(bool cpol, bool cpha)
{
	register int ret = -EINVAL;

	if ((!cpol) && (!cpha)) {
		ret = 0;
	} else if (cpol && cpha) {
		ret = 1;
	}
	__ASSERT(ret != -EINVAL, "Invalid QSPI mode");
	return ret;
}

static inline bool qspi_write_is_quad(nrf_qspi_writeoc_t lines)
{
	switch (lines) {
	case NRF_QSPI_WRITEOC_PP4IO:
	case NRF_QSPI_WRITEOC_PP4O:
		return true;
	default:
		return false;
	}
}

static inline bool qspi_read_is_quad(nrf_qspi_readoc_t lines)
{
	switch (lines) {
	case NRF_QSPI_READOC_READ4IO:
	case NRF_QSPI_READOC_READ4O:
		return true;
	default:
		return false;
	}
}

static inline int qspi_get_lines_write(uint8_t lines)
{
	register int ret = -EINVAL;

	switch (lines) {
	case 3:
		ret =  NRF_QSPI_WRITEOC_PP4IO;
		break;
	case 2:
		ret = NRF_QSPI_WRITEOC_PP4O;
		break;
	case 1:
		ret = NRF_QSPI_WRITEOC_PP2O;
		break;
	case 0:
		ret = NRF_QSPI_WRITEOC_PP;
		break;
	default:
		break;
	}
	__ASSERT(ret != -EINVAL, "Invalid QSPI write line");
	return ret;
}

static inline int qspi_get_lines_read(uint8_t lines)
{
	register int ret = -EINVAL;

	switch (lines) {
	case 4:
		ret = NRF_QSPI_READOC_READ4IO;
		break;
	case 3:
		ret = NRF_QSPI_READOC_READ4O;
		break;
	case 2:
		ret = NRF_QSPI_READOC_READ2IO;
		break;
	case 1:
		ret = NRF_QSPI_READOC_READ2O;
		break;
	case 0:
		ret = NRF_QSPI_READOC_FASTREAD;
		break;
	default:
		break;
	}
	__ASSERT(ret != -EINVAL, "Invalid QSPI read line");
	return ret;
}

static inline nrf_qspi_addrmode_t qspi_get_address_size(bool addr_size)
{
	return addr_size ? NRF_QSPI_ADDRMODE_32BIT : NRF_QSPI_ADDRMODE_24BIT;
}

/**
 * @brief Test whether offset is aligned.
 */
#define QSPI_IS_SECTOR_ALIGNED(_ofs) (((_ofs) & (QSPI_SECTOR_SIZE - 1U)) == 0)
#define QSPI_IS_BLOCK_ALIGNED(_ofs) (((_ofs) & (QSPI_BLOCK_SIZE - 1U)) == 0)

/**
 * @brief Main configuration structure
 */
static struct qspi_nor_data qspi_nor_memory_data = {
#ifdef CONFIG_MULTITHREADING
	.trans = Z_SEM_INITIALIZER(qspi_nor_memory_data.trans, 1, 1),
	.sem = Z_SEM_INITIALIZER(qspi_nor_memory_data.sem, 1, 1),
	.sync = Z_SEM_INITIALIZER(qspi_nor_memory_data.sync, 0, 1),
#if NRF52_ERRATA_122_PRESENT
	.count = Z_SEM_INITIALIZER(qspi_nor_memory_data.count, 0, K_SEM_MAX_LIMIT),
#endif
#endif /* CONFIG_MULTITHREADING */
};

/**
 * @brief Converts NRFX return codes to the zephyr ones
 */
static inline int qspi_get_zephyr_ret_code(nrfx_err_t res)
{
	switch (res) {
	case NRFX_SUCCESS:
		return 0;
	case NRFX_ERROR_INVALID_PARAM:
	case NRFX_ERROR_INVALID_ADDR:
		return -EINVAL;
	case NRFX_ERROR_INVALID_STATE:
		return -ECANCELED;
	case NRFX_ERROR_BUSY:
	case NRFX_ERROR_TIMEOUT:
	default:
		return -EBUSY;
	}
}

static inline struct qspi_nor_data *get_dev_data(const struct device *dev)
{
	return dev->data;
}

static inline void qspi_lock(const struct device *dev)
{
#ifdef CONFIG_MULTITHREADING
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	k_sem_take(&dev_data->sem, K_FOREVER);
#else /* CONFIG_MULTITHREADING */
	ARG_UNUSED(dev);
#endif /* CONFIG_MULTITHREADING */
}

static inline void qspi_unlock(const struct device *dev)
{
#ifdef CONFIG_MULTITHREADING
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	k_sem_give(&dev_data->sem);
#else /* CONFIG_MULTITHREADING */
	ARG_UNUSED(dev);
#endif /* CONFIG_MULTITHREADING */
}

static inline void qspi_trans_lock(const struct device *dev)
{
#ifdef CONFIG_MULTITHREADING
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	k_sem_take(&dev_data->trans, K_FOREVER);
#else /* CONFIG_MULTITHREADING */
	ARG_UNUSED(dev);
#endif /* CONFIG_MULTITHREADING */
}

static inline void qspi_trans_unlock(const struct device *dev)
{
#ifdef CONFIG_MULTITHREADING
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	k_sem_give(&dev_data->trans);
#else /* CONFIG_MULTITHREADING */
	ARG_UNUSED(dev);
#endif /* CONFIG_MULTITHREADING */
}

static inline void qspi_wait_for_completion(const struct device *dev,
					    nrfx_err_t res)
{
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	if (res == NRFX_SUCCESS) {
#ifdef CONFIG_MULTITHREADING
		k_sem_take(&dev_data->sync, K_FOREVER);
#else /* CONFIG_MULTITHREADING */
		unsigned int key = irq_lock();

		while (!dev_data->ready) {
			k_cpu_atomic_idle(key);
			key = irq_lock();
		}
		dev_data->ready = false;
		irq_unlock(key);
#endif /* CONFIG_MULTITHREADING */
	}
}

static inline void qspi_complete(struct qspi_nor_data *dev_data)
{
#ifdef CONFIG_MULTITHREADING
	k_sem_give(&dev_data->sync);
#else /* CONFIG_MULTITHREADING */
	dev_data->ready = true;
#endif /* CONFIG_MULTITHREADING */
}

/**
 * @brief QSPI handler
 *
 * @param event Driver event type
 * @param p_context Pointer to context. Use in interrupt handler.
 * @retval None
 */
static void qspi_handler(nrfx_qspi_evt_t event, void *p_context)
{
	struct qspi_nor_data *dev_data = p_context;

	if (event == NRFX_QSPI_EVENT_DONE) {
		qspi_complete(dev_data);
	}
}

#if NRF52_ERRATA_122_PRESENT
static bool qspi_initialized;

static int anomaly_122_init(const struct device *dev)
{
	struct qspi_nor_data *dev_data = get_dev_data(dev);
	nrfx_err_t res;
	int ret = 0;

	if (!nrf52_errata_122()) {
		return 0;
	}

	qspi_lock(dev);

	/* In multithreading, driver can call anomaly_122_init more than once
	 * before calling anomaly_122_uninit. Keepping count, so QSPI is
	 * uninitialized only at the last call (count == 0).
	 */
#ifdef CONFIG_MULTITHREADING
	k_sem_give(&dev_data->count);
#endif

	if (!qspi_initialized) {
		res = nrfx_qspi_init(&QSPIconfig, qspi_handler, dev_data);
		ret = qspi_get_zephyr_ret_code(res);
		qspi_initialized = (ret == 0);
	}

	qspi_unlock(dev);

	return ret;
}

static void anomaly_122_uninit(const struct device *dev)
{
	bool last = true;

	if (!nrf52_errata_122()) {
		return;
	}

	qspi_lock(dev);

#ifdef CONFIG_MULTITHREADING
	struct qspi_nor_data *dev_data = get_dev_data(dev);

	/* The last thread to finish using the driver uninit the QSPI */
	(void) k_sem_take(&dev_data->count, K_NO_WAIT);
	last = (k_sem_count_get(&dev_data->count) == 0);
#endif

	if (last) {
		while (nrfx_qspi_mem_busy_check() != NRFX_SUCCESS) {
			if (IS_ENABLED(CONFIG_MULTITHREADING)) {
				k_msleep(50);
			} else {
				k_busy_wait(50000);
			}
		}

		nrf_gpio_cfg_output(QSPI_PROP_AT(csn_pins, 0));
		nrf_gpio_pin_set(QSPI_PROP_AT(csn_pins, 0));

		nrfx_qspi_uninit();
		qspi_initialized = false;
	}

	qspi_unlock(dev);
}
#endif /* NRF52_ERRATA_122_PRESENT */


/* QSPI send custom command.
 *
 * If this is used for both send and receive the buffer sizes must be
 * equal and cover the whole transaction.
 */
static int qspi_send_cmd(const struct device *dev, const struct qspi_cmd *cmd,
			 bool wren)
{
	/* Check input parameters */
	if (!cmd) {
		return -EINVAL;
	}

	const void *tx_buf = NULL;
	size_t tx_len = 0;
	void *rx_buf = NULL;
	size_t rx_len = 0;
	size_t xfer_len = sizeof(cmd->op_code);

	if (cmd->tx_buf) {
		tx_buf = cmd->tx_buf->buf;
		tx_len = cmd->tx_buf->len;
	}

	if (cmd->rx_buf) {
		rx_buf = cmd->rx_buf->buf;
		rx_len = cmd->rx_buf->len;
	}

	if ((rx_len != 0) && (tx_len != 0)) {
		if (rx_len != tx_len) {
			return -EINVAL;
		}

		xfer_len += tx_len;
	} else {
		/* At least one of these is zero. */
		xfer_len += tx_len + rx_len;
	}

	if (xfer_len > NRF_QSPI_CINSTR_LEN_9B) {
		LOG_WRN("cinstr %02x transfer too long: %zu",
			cmd->op_code, xfer_len);
		return -EINVAL;
	}

	nrf_qspi_cinstr_conf_t cinstr_cfg = {
		.opcode = cmd->op_code,
		.length = xfer_len,
		.io2_level = true,
		.io3_level = true,
		.wipwait = false,
		.wren = wren,
	};

	qspi_lock(dev);

	int res = nrfx_qspi_cinstr_xfer(&cinstr_cfg, tx_buf, rx_buf);

	qspi_unlock(dev);
	return qspi_get_zephyr_ret_code(res);
}

/* RDSR wrapper.  Negative value is error. */
static int qspi_rdsr(const struct device *dev)
{
	uint8_t sr = -1;
	const struct qspi_buf sr_buf = {
		.buf = &sr,
		.len = sizeof(sr),
	};
	struct qspi_cmd cmd = {
		.op_code = SPI_NOR_CMD_RDSR,
		.rx_buf = &sr_buf,
	};
	int ret = qspi_send_cmd(dev, &cmd, false);

	return (ret < 0) ? ret : sr;
}

/* Wait until RDSR confirms write is not in progress. */
static int qspi_wait_while_writing(const struct device *dev)
{
	int ret;

	do {
		ret = qspi_rdsr(dev);
	} while ((ret >= 0)
		 && ((ret & SPI_NOR_WIP_BIT) != 0U));

	return (ret < 0) ? ret : 0;
}

/* QSPI erase */
static int qspi_erase(const struct device *dev, uint32_t addr, uint32_t size)
{
	/* address must be sector-aligned */
	if ((addr % QSPI_SECTOR_SIZE) != 0) {
		return -EINVAL;
	}

	/* size must be a non-zero multiple of sectors */
	if ((size == 0) || (size % QSPI_SECTOR_SIZE) != 0) {
		return -EINVAL;
	}

	int rv = 0;
	const struct qspi_nor_config *params = dev->config;

	rv = ANOMALY_122_INIT(dev);
	if (rv != 0) {
		goto out;
	}
	qspi_trans_lock(dev);
	rv = qspi_nor_write_protection_set(dev, false);
	qspi_lock(dev);
	while ((rv == 0) && (size > 0)) {
		nrfx_err_t res = !NRFX_SUCCESS;
		uint32_t adj = 0;

		if (size == params->size) {
			/* chip erase */
			res = nrfx_qspi_chip_erase();
			adj = size;
		} else if ((size >= QSPI_BLOCK_SIZE) &&
			   QSPI_IS_BLOCK_ALIGNED(addr)) {
			/* 64 kB block erase */
			res = nrfx_qspi_erase(NRF_QSPI_ERASE_LEN_64KB, addr);
			adj = QSPI_BLOCK_SIZE;
		} else if ((size >= QSPI_SECTOR_SIZE) &&
			   QSPI_IS_SECTOR_ALIGNED(addr)) {
			/* 4kB sector erase */
			res = nrfx_qspi_erase(NRF_QSPI_ERASE_LEN_4KB, addr);
			adj = QSPI_SECTOR_SIZE;
		} else {
			/* minimal erase size is at least a sector size */
			LOG_ERR("unsupported at 0x%lx size %zu", (long)addr, size);
			rv = -EINVAL;
		}

		qspi_wait_for_completion(dev, res);
		if (res == NRFX_SUCCESS) {
			addr += adj;
			size -= adj;
		} else {
			LOG_ERR("erase error at 0x%lx size %zu", (long)addr, size);
			rv = qspi_get_zephyr_ret_code(res);
		}
	}
	qspi_unlock(dev);

	int rv2 = qspi_nor_write_protection_set(dev, true);

	qspi_trans_unlock(dev);

	if (!rv) {
		rv = rv2;
	}

out:
	ANOMALY_122_UNINIT(dev);
	return rv;
}

/**
 * @brief Fills init struct
 *
 * @param config Pointer to the config struct provided by user
 * @param initstruct Pointer to the configuration struct
 * @retval None
 */
static inline void qspi_fill_init_struct(nrfx_qspi_config_t *initstruct)
{
	/* Configure XIP offset */
	initstruct->xip_offset = 0;

	/* Configure pins */
	initstruct->pins.sck_pin = DT_PROP(QSPI_NODE, sck_pin);
	initstruct->pins.csn_pin = QSPI_PROP_AT(csn_pins, 0);
	initstruct->pins.io0_pin = QSPI_PROP_AT(io_pins, 0);
	initstruct->pins.io1_pin = QSPI_PROP_AT(io_pins, 1);
#if QSPI_PROP_LEN(io_pins) > 2
	initstruct->pins.io2_pin = QSPI_PROP_AT(io_pins, 2);
	initstruct->pins.io3_pin = QSPI_PROP_AT(io_pins, 3);
#else
	initstruct->pins.io2_pin = NRF_QSPI_PIN_NOT_CONNECTED;
	initstruct->pins.io3_pin = NRF_QSPI_PIN_NOT_CONNECTED;
#endif

	/* Configure Protocol interface */
#if DT_INST_NODE_HAS_PROP(0, readoc)
	initstruct->prot_if.readoc =
		(nrf_qspi_readoc_t)qspi_get_lines_read(DT_ENUM_IDX(DT_DRV_INST(0), readoc));
#else
	initstruct->prot_if.readoc = NRF_QSPI_READOC_FASTREAD;
#endif

#if DT_INST_NODE_HAS_PROP(0, writeoc)
	initstruct->prot_if.writeoc =
		(nrf_qspi_writeoc_t)qspi_get_lines_write(DT_ENUM_IDX(DT_DRV_INST(0), writeoc));
#else
	initstruct->prot_if.writeoc = NRF_QSPI_WRITEOC_PP;
#endif
	initstruct->prot_if.addrmode =
		qspi_get_address_size(DT_INST_PROP(0, address_size_32));

	initstruct->prot_if.dpmconfig = false;

	/* Configure physical interface */
	initstruct->phy_if.sck_freq =
		(INST_0_SCK_FREQUENCY > NRF_QSPI_BASE_CLOCK_FREQ)
		? NRF_QSPI_FREQ_DIV1
		: (NRF_QSPI_BASE_CLOCK_FREQ / INST_0_SCK_FREQUENCY) - 1;
	initstruct->phy_if.sck_delay = DT_INST_PROP(0, sck_delay);
	initstruct->phy_if.spi_mode = qspi_get_mode(DT_INST_PROP(0, cpol),
						    DT_INST_PROP(0, cpha));

	initstruct->phy_if.dpmen = false;
}

/* Configures QSPI memory for the transfer */
static int qspi_nrfx_configure(const struct device *dev)
{
	if (!dev) {
		return -ENXIO;
	}

	struct qspi_nor_data *dev_data = dev->data;

	qspi_fill_init_struct(&QSPIconfig);

	nrfx_err_t res = nrfx_qspi_init(&QSPIconfig, qspi_handler, dev_data);
	int ret = qspi_get_zephyr_ret_code(res);

	if ((ret == 0)
	    && (INST_0_QER != JESD216_DW15_QER_NONE)) {
		/* Set QE to match transfer mode.  If not using quad
		 * it's OK to leave QE set, but doing so prevents use
		 * of WP#/RESET#/HOLD# which might be useful.
		 *
		 * Note build assert above ensures QER is S1B6.  Other
		 * options require more logic.
		 */

		ret = qspi_rdsr(dev);
		if (ret < 0) {
			LOG_ERR("RDSR failed: %d", ret);
			return ret;
		}

		uint8_t sr = (uint8_t)ret;
		bool qe_value =
			(qspi_write_is_quad(QSPIconfig.prot_if.writeoc))
			|| (qspi_read_is_quad(QSPIconfig.prot_if.readoc));
		const uint8_t qe_mask = BIT(6); /* only S1B6 */
		bool qe_state = ((sr & qe_mask) != 0U);

		LOG_DBG("RDSR %02x QE %d need %d: %s", sr, qe_state, qe_value,
			(qe_state != qe_value) ? "updating" : "no-change");

		ret = 0;
		if (qe_state != qe_value) {
			const struct qspi_buf sr_buf = {
				.buf = &sr,
				.len = sizeof(sr),
			};
			struct qspi_cmd cmd = {
				.op_code = SPI_NOR_CMD_WRSR,
				.tx_buf = &sr_buf,
			};

			sr ^= qe_mask;
			ret = qspi_send_cmd(dev, &cmd, true);

			/* Writing SR can take some time, and further
			 * commands sent while it's happening can be
			 * corrupted.  Wait.
			 */
			if (ret == 0) {
				ret = qspi_wait_while_writing(dev);
			}
		}

		if (ret < 0) {
			LOG_ERR("QE %s failed: %d", qe_value ? "set" : "clear",
				ret);
		}

	}

	if ((ret == 0)
	    && ((INST_0_4BA != 0) && (INST_0_4BA != 0xff))) {
    if (INST_0_4BA & 0x02)
			qspi_nor_write_protection_set(dev, false);

		struct qspi_cmd cmd = {
			.op_code = SPI_NOR_CMD_4BA,
			.rx_buf = NULL,
			.tx_buf = NULL,
		};
		ret = qspi_send_cmd(dev, &cmd, false);
	}

	return ret;
}

static int qspi_read_jedec_id(const struct device *dev,
				     uint8_t *id)
{
	const struct qspi_buf rx_buf = {
		.buf = id,
		.len = 3
	};
	const struct qspi_cmd cmd = {
		.op_code = SPI_NOR_CMD_RDID,
		.rx_buf = &rx_buf,
	};

	int ret = ANOMALY_122_INIT(dev);

	if (ret == 0) {
		ret = qspi_send_cmd(dev, &cmd, false);
	}
	ANOMALY_122_UNINIT(dev);

	return ret;
}

#if defined(CONFIG_FLASH_JESD216_API)

static int qspi_sfdp_read(const struct device *dev, off_t offset,
			  void *data, size_t len)
{
	__ASSERT(data != NULL, "null destination");

	uint8_t addr_buf[] = {
		offset >> 16,
		offset >> 8,
		offset,
		0,		/* wait state */
	};
	nrf_qspi_cinstr_conf_t cinstr_cfg = {
		.opcode = JESD216_CMD_READ_SFDP,
		.length = NRF_QSPI_CINSTR_LEN_1B,
		.io2_level = true,
		.io3_level = true,
	};

	int res = ANOMALY_122_INIT(dev);

	if (res != NRFX_SUCCESS) {
		LOG_DBG("ANOMALY_122_INIT: %x", res);
		goto out;
	}

	qspi_lock(dev);
	res = nrfx_qspi_lfm_start(&cinstr_cfg);
	if (res != NRFX_SUCCESS) {
		LOG_DBG("lfm_start: %x", res);
		goto out;
	}

	res = nrfx_qspi_lfm_xfer(addr_buf, NULL, sizeof(addr_buf), false);
	if (res != NRFX_SUCCESS) {
		LOG_DBG("lfm_xfer addr: %x", res);
		goto out;
	}

	res = nrfx_qspi_lfm_xfer(NULL, data, len, true);
	if (res != NRFX_SUCCESS) {
		LOG_DBG("lfm_xfer read: %x", res);
		goto out;
	}

out:
	qspi_unlock(dev);
	ANOMALY_122_UNINIT(dev);
	return qspi_get_zephyr_ret_code(res);
}

#endif /* CONFIG_FLASH_JESD216_API */

/**
 * @brief Retrieve the Flash JEDEC ID and compare it with the one expected
 *
 * @param dev The device structure
 * @param flash_id The flash info structure which contains the
 *		  expected JEDEC ID
 * @return 0 on success, negative errno code otherwise
 */
static inline int qspi_nor_read_id(const struct device *dev,
				   const struct qspi_nor_config *const flash_id)
{
	uint8_t id[SPI_NOR_MAX_ID_LEN];
	int ret = qspi_read_jedec_id(dev, id);

	if (ret != 0) {
		return -EIO;
	}

	if (memcmp(flash_id->id, id, SPI_NOR_MAX_ID_LEN) != 0) {
		LOG_ERR("JEDEC id [%02x %02x %02x] expect [%02x %02x %02x]",
			id[0], id[1], id[2],
			flash_id->id[0], flash_id->id[1], flash_id->id[2]);
		return -ENODEV;
	}

	return 0;
}

static inline nrfx_err_t read_non_aligned(const struct device *dev,
					  off_t addr,
					  void *dest, size_t size)
{
	uint8_t __aligned(WORD_SIZE) buf[WORD_SIZE * 2];
	uint8_t *dptr = dest;

	off_t flash_prefix = (WORD_SIZE - (addr % WORD_SIZE)) % WORD_SIZE;

	if (flash_prefix > size) {
		flash_prefix = size;
	}

	off_t dest_prefix = (WORD_SIZE - (off_t)dptr % WORD_SIZE) % WORD_SIZE;

	if (dest_prefix > size) {
		dest_prefix = size;
	}

	off_t flash_suffix = (size - flash_prefix) % WORD_SIZE;
	off_t flash_middle = size - flash_prefix - flash_suffix;
	off_t dest_middle = size - dest_prefix -
			    (size - dest_prefix) % WORD_SIZE;

	if (flash_middle > dest_middle) {
		flash_middle = dest_middle;
		flash_suffix = size - flash_prefix - flash_middle;
	}

	nrfx_err_t res = NRFX_SUCCESS;

	/* read from aligned flash to aligned memory */
	if (flash_middle != 0) {
		res = nrfx_qspi_read(dptr + dest_prefix, flash_middle,
				     addr + flash_prefix);
		qspi_wait_for_completion(dev, res);
		if (res != NRFX_SUCCESS) {
			return res;
		}

		/* perform shift in RAM */
		if (flash_prefix != dest_prefix) {
			memmove(dptr + flash_prefix, dptr + dest_prefix, flash_middle);
		}
	}

	/* read prefix */
	if (flash_prefix != 0) {
		res = nrfx_qspi_read(buf, WORD_SIZE, addr -
				     (WORD_SIZE - flash_prefix));
		qspi_wait_for_completion(dev, res);
		if (res != NRFX_SUCCESS) {
			return res;
		}
		memcpy(dptr, buf + WORD_SIZE - flash_prefix, flash_prefix);
	}

	/* read suffix */
	if (flash_suffix != 0) {
		res = nrfx_qspi_read(buf, WORD_SIZE * 2,
				     addr + flash_prefix + flash_middle);
		qspi_wait_for_completion(dev, res);
		if (res != NRFX_SUCCESS) {
			return res;
		}
		memcpy(dptr + flash_prefix + flash_middle, buf, flash_suffix);
	}

	return res;
}

static int qspi_nor_read(const struct device *dev, off_t addr, void *dest,
			 size_t size)
{
	if (!dest) {
		return -EINVAL;
	}

	/* read size must be non-zero */
	if (!size) {
		return 0;
	}

	const struct qspi_nor_config *params = dev->config;

	/* affected region should be within device */
	if (addr < 0 ||
	    (addr + size) > params->size) {
		LOG_ERR("read error: address or size "
			"exceeds expected values."
			"Addr: 0x%lx size %zu", (long)addr, size);
		return -EINVAL;
	}

	int rc = ANOMALY_122_INIT(dev);

	if (rc != 0) {
		goto out;
	}

	qspi_lock(dev);

	nrfx_err_t res = read_non_aligned(dev, addr, dest, size);

	qspi_unlock(dev);

	rc = qspi_get_zephyr_ret_code(res);

out:
	ANOMALY_122_UNINIT(dev);
	return rc;
}

/* addr aligned, sptr not null, slen less than 4 */
static inline nrfx_err_t write_sub_word(const struct device *dev, off_t addr,
					const void *sptr, size_t slen)
{
	uint8_t __aligned(4) buf[4];
	nrfx_err_t res;

	/* read out the whole word so that unchanged data can be
	 * written back
	 */
	res = nrfx_qspi_read(buf, sizeof(buf), addr);
	qspi_wait_for_completion(dev, res);

	if (res == NRFX_SUCCESS) {
		memcpy(buf, sptr, slen);
		res = nrfx_qspi_write(buf, sizeof(buf), addr);
		qspi_wait_for_completion(dev, res);
	}

	return res;
}

BUILD_ASSERT((CONFIG_NORDIC_QSPI_NOR_STACK_WRITE_BUFFER_SIZE % 4) == 0,
	     "NOR stack buffer must be multiple of 4 bytes");

#define NVMC_WRITE_OK (CONFIG_NORDIC_QSPI_NOR_STACK_WRITE_BUFFER_SIZE > 0)

/* If enabled write using a stack-allocated aligned SRAM buffer as
 * required for DMA transfers by QSPI peripheral.
 *
 * If not enabled return the error the peripheral would have produced.
 */
static inline nrfx_err_t write_from_nvmc(const struct device *dev, off_t addr,
					 const void *sptr, size_t slen)
{
#if NVMC_WRITE_OK
	uint8_t __aligned(4) buf[CONFIG_NORDIC_QSPI_NOR_STACK_WRITE_BUFFER_SIZE];
	const uint8_t *sp = sptr;
	nrfx_err_t res = NRFX_SUCCESS;

	while ((slen > 0) && (res == NRFX_SUCCESS)) {
		size_t len = MIN(slen, sizeof(buf));

		memcpy(buf, sp, len);
		res = nrfx_qspi_write(buf, sizeof(buf),
				      addr);
		qspi_wait_for_completion(dev, res);

		if (res == NRFX_SUCCESS) {
			slen -= len;
			sp += len;
			addr += len;
		}
	}
#else /* NVMC_WRITE_OK */
	nrfx_err_t res = NRFX_ERROR_INVALID_ADDR;
#endif /* NVMC_WRITE_OK */
	return res;
}

static int qspi_nor_write(const struct device *dev, off_t addr,
			  const void *src,
			  size_t size)
{
	if (!src) {
		return -EINVAL;
	}

	/* write size must be non-zero, less than 4, or a multiple of 4 */
	if ((size == 0)
	    || ((size > 4) && ((size % 4U) != 0))) {
		return -EINVAL;
	}
	/* address must be 4-byte aligned */
	if ((addr % 4U) != 0) {
		return -EINVAL;
	}

	const struct qspi_nor_config *params = dev->config;

	/* affected region should be within device */
	if (addr < 0 ||
	    (addr + size) > params->size) {
		LOG_ERR("write error: address or size "
			"exceeds expected values."
			"Addr: 0x%lx size %zu", (long)addr, size);
		return -EINVAL;
	}

	nrfx_err_t res = NRFX_SUCCESS;

	int rc = ANOMALY_122_INIT(dev);

	if (rc != 0) {
		goto out;
	}

	qspi_trans_lock(dev);
	res = qspi_nor_write_protection_set(dev, false);
	qspi_lock(dev);
	if (!res) {
		if (size < 4U) {
			res = write_sub_word(dev, addr, src, size);
		} else if (!nrfx_is_in_ram(src)) {
			res = write_from_nvmc(dev, addr, src, size);
		} else {
			res = nrfx_qspi_write(src, size, addr);
			qspi_wait_for_completion(dev, res);
		}
	}
	qspi_unlock(dev);

	int res2 = qspi_nor_write_protection_set(dev, true);

	qspi_trans_unlock(dev);
	if (!res) {
		res = res2;
	}

	rc = qspi_get_zephyr_ret_code(res);
out:
	ANOMALY_122_UNINIT(dev);
	return rc;
}

static int qspi_nor_erase(const struct device *dev, off_t addr, size_t size)
{
	const struct qspi_nor_config *params = dev->config;

	/* affected region should be within device */
	if (addr < 0 ||
	    (addr + size) > params->size) {
		LOG_ERR("erase error: address or size "
			"exceeds expected values."
			"Addr: 0x%lx size %zu", (long)addr, size);
		return -EINVAL;
	}

	int ret = qspi_erase(dev, addr, size);

	return ret;
}

static int qspi_nor_write_protection_set(const struct device *dev,
					 bool write_protect)
{
	int ret = 0;
	struct qspi_cmd cmd = {
		.op_code = ((write_protect) ? SPI_NOR_CMD_WRDI : SPI_NOR_CMD_WREN),
	};

	if (qspi_send_cmd(dev, &cmd, false) != 0) {
		ret = -EIO;
	}

	return ret;
}

/**
 * @brief Configure the flash
 *
 * @param dev The flash device structure
 * @param info The flash info structure
 * @return 0 on success, negative errno code otherwise
 */
static int qspi_nor_configure(const struct device *dev)
{
	const struct qspi_nor_config *params = dev->config;

	int ret = qspi_nrfx_configure(dev);

	if (ret != 0) {
		return ret;
	}

	ANOMALY_122_UNINIT(dev);

	/* now the spi bus is configured, we can verify the flash id */
	if (qspi_nor_read_id(dev, params) != 0) {
		return -ENODEV;
	}

	return 0;
}

/**
 * @brief Initialize and configure the flash
 *
 * @param name The flash name
 * @return 0 on success, negative errno code otherwise
 */
static int qspi_nor_init(const struct device *dev)
{
#if defined(CONFIG_SOC_SERIES_NRF53X)
	/* Make sure the PCLK192M clock, from which the SCK frequency is
	 * derived, is not prescaled (the default setting after reset is
	 * "divide by 4").
	 */
	nrf_clock_hfclk192m_div_set(NRF_CLOCK, NRF_CLOCK_HFCLK_DIV_1);
#endif

	IRQ_CONNECT(DT_IRQN(QSPI_NODE), DT_IRQ(QSPI_NODE, priority),
		    nrfx_isr, nrfx_qspi_irq_handler, 0);
	return qspi_nor_configure(dev);
}

#if defined(CONFIG_FLASH_PAGE_LAYOUT)

/* instance 0 page count */
#define LAYOUT_PAGES_COUNT (INST_0_BYTES / \
			    CONFIG_NORDIC_QSPI_NOR_FLASH_LAYOUT_PAGE_SIZE)

BUILD_ASSERT((CONFIG_NORDIC_QSPI_NOR_FLASH_LAYOUT_PAGE_SIZE *
	      LAYOUT_PAGES_COUNT)
	     == INST_0_BYTES,
	     "QSPI_NOR_FLASH_LAYOUT_PAGE_SIZE incompatible with flash size");

static const struct flash_pages_layout dev_layout = {
	.pages_count = LAYOUT_PAGES_COUNT,
	.pages_size = CONFIG_NORDIC_QSPI_NOR_FLASH_LAYOUT_PAGE_SIZE,
};
#undef LAYOUT_PAGES_COUNT

static const struct flash_parameters *
qspi_flash_get_parameters(const struct device *dev)
{
	ARG_UNUSED(dev);

	return &qspi_flash_parameters;
}

static void qspi_nor_pages_layout(const struct device *dev,
				  const struct flash_pages_layout **layout,
				  size_t *layout_size)
{
	*layout = &dev_layout;
	*layout_size = 1;
}
#endif /* CONFIG_FLASH_PAGE_LAYOUT */

static const struct flash_driver_api qspi_nor_api = {
	.read = qspi_nor_read,
	.write = qspi_nor_write,
	.erase = qspi_nor_erase,
	.get_parameters = qspi_flash_get_parameters,
#if defined(CONFIG_FLASH_PAGE_LAYOUT)
	.page_layout = qspi_nor_pages_layout,
#endif
#if defined(CONFIG_FLASH_JESD216_API)
	.sfdp_read = qspi_sfdp_read,
	.read_jedec_id = qspi_read_jedec_id,
#endif /* CONFIG_FLASH_JESD216_API */
};


static const struct qspi_nor_config flash_id = {
	.id = DT_INST_PROP(0, jedec_id),
	.size = INST_0_BYTES,
};

DEVICE_DT_INST_DEFINE(0, &qspi_nor_init, NULL,
		&qspi_nor_memory_data, &flash_id,
		POST_KERNEL, CONFIG_NORDIC_QSPI_NOR_INIT_PRIORITY,
		&qspi_nor_api);

[Kenneth]

Great! I have updated the internal jira with your findings.

Regarding 1) I am not aware of any way.
Regarding 2) Hopefully it will be followed up now that I have updated the case internally with your findings.

Kenneth

[akw]

I am still facing this issue. Regardless of partition size (tried 128 MB,100 MB, 64 MB), the code starts giving -14 (addressing) errors at about 80% of the partition size.

[agf]

did you try to set `address-size-32` DTS property for the node which describes your qspi memory - It might be that only this is missed for getting nrf-qspi working with 32-bits addressing.