blob: 70c8aef702725dc8598683d2b3aebdc32dff658b [file] [log] [blame]
/*
* Copyright (c) 2017 Google LLC.
* Copyright (c) 2018 qianfan Zhao.
*
* SPDX-License-Identifier: Apache-2.0
*/
#define DT_DRV_COMPAT atmel_sam_spi
#define LOG_LEVEL CONFIG_SPI_LOG_LEVEL
#include <logging/log.h>
LOG_MODULE_REGISTER(spi_sam);
#include "spi_context.h"
#include <errno.h>
#include <device.h>
#include <drivers/spi.h>
#include <soc.h>
#define SAM_SPI_CHIP_SELECT_COUNT 4
/* Device constant configuration parameters */
struct spi_sam_config {
Spi *regs;
uint32_t periph_id;
uint32_t num_pins;
struct soc_gpio_pin pins[];
};
/* Device run time data */
struct spi_sam_data {
struct spi_context ctx;
};
static int spi_slave_to_mr_pcs(int slave)
{
int pcs[SAM_SPI_CHIP_SELECT_COUNT] = {0x0, 0x1, 0x3, 0x7};
/* SPI worked in fixed perieral mode(SPI_MR.PS = 0) and disabled chip
* select decode(SPI_MR.PCSDEC = 0), based on Atmel | SMART ARM-based
* Flash MCU DATASHEET 40.8.2 SPI Mode Register:
* PCS = xxx0 NPCS[3:0] = 1110
* PCS = xx01 NPCS[3:0] = 1101
* PCS = x011 NPCS[3:0] = 1011
* PCS = 0111 NPCS[3:0] = 0111
*/
return pcs[slave];
}
static int spi_sam_configure(const struct device *dev,
const struct spi_config *config)
{
const struct spi_sam_config *cfg = dev->config;
struct spi_sam_data *data = dev->data;
Spi *regs = cfg->regs;
uint32_t spi_mr = 0U, spi_csr = 0U;
int div;
if (spi_context_configured(&data->ctx, config)) {
return 0;
}
if (config->operation & SPI_HALF_DUPLEX) {
LOG_ERR("Half-duplex not supported");
return -ENOTSUP;
}
if (SPI_OP_MODE_GET(config->operation) != SPI_OP_MODE_MASTER) {
/* Slave mode is not implemented. */
return -ENOTSUP;
}
if (config->slave > (SAM_SPI_CHIP_SELECT_COUNT - 1)) {
LOG_ERR("Slave %d is greater than %d",
config->slave, SAM_SPI_CHIP_SELECT_COUNT - 1);
return -EINVAL;
}
/* Set master mode, disable mode fault detection, set fixed peripheral
* select mode.
*/
spi_mr |= (SPI_MR_MSTR | SPI_MR_MODFDIS);
spi_mr |= SPI_MR_PCS(spi_slave_to_mr_pcs(config->slave));
if ((config->operation & SPI_MODE_CPOL) != 0U) {
spi_csr |= SPI_CSR_CPOL;
}
if ((config->operation & SPI_MODE_CPHA) == 0U) {
spi_csr |= SPI_CSR_NCPHA;
}
if (SPI_WORD_SIZE_GET(config->operation) != 8) {
return -ENOTSUP;
} else {
spi_csr |= SPI_CSR_BITS(SPI_CSR_BITS_8_BIT);
}
/* Use the requested or next highest possible frequency */
div = SOC_ATMEL_SAM_MCK_FREQ_HZ / config->frequency;
div = CLAMP(div, 1, UINT8_MAX);
spi_csr |= SPI_CSR_SCBR(div);
regs->SPI_CR = SPI_CR_SPIDIS; /* Disable SPI */
regs->SPI_MR = spi_mr;
regs->SPI_CSR[config->slave] = spi_csr;
regs->SPI_CR = SPI_CR_SPIEN; /* Enable SPI */
data->ctx.config = config;
return 0;
}
static bool spi_sam_transfer_ongoing(struct spi_sam_data *data)
{
return spi_context_tx_on(&data->ctx) || spi_context_rx_on(&data->ctx);
}
static void spi_sam_shift_master(Spi *regs, struct spi_sam_data *data)
{
uint8_t tx;
uint8_t rx;
if (spi_context_tx_buf_on(&data->ctx)) {
tx = *(uint8_t *)(data->ctx.tx_buf);
} else {
tx = 0U;
}
while ((regs->SPI_SR & SPI_SR_TDRE) == 0) {
}
regs->SPI_TDR = SPI_TDR_TD(tx);
spi_context_update_tx(&data->ctx, 1, 1);
while ((regs->SPI_SR & SPI_SR_RDRF) == 0) {
}
rx = (uint8_t)regs->SPI_RDR;
if (spi_context_rx_buf_on(&data->ctx)) {
*data->ctx.rx_buf = rx;
}
spi_context_update_rx(&data->ctx, 1, 1);
}
/* Finish any ongoing writes and drop any remaining read data */
static void spi_sam_finish(Spi *regs)
{
while ((regs->SPI_SR & SPI_SR_TXEMPTY) == 0) {
}
while (regs->SPI_SR & SPI_SR_RDRF) {
(void)regs->SPI_RDR;
}
}
/* Fast path that transmits a buf */
static void spi_sam_fast_tx(Spi *regs, const struct spi_buf *tx_buf)
{
const uint8_t *p = tx_buf->buf;
const uint8_t *pend = (uint8_t *)tx_buf->buf + tx_buf->len;
uint8_t ch;
while (p != pend) {
ch = *p++;
while ((regs->SPI_SR & SPI_SR_TDRE) == 0) {
}
regs->SPI_TDR = SPI_TDR_TD(ch);
}
spi_sam_finish(regs);
}
/* Fast path that reads into a buf */
static void spi_sam_fast_rx(Spi *regs, const struct spi_buf *rx_buf)
{
uint8_t *rx = rx_buf->buf;
int len = rx_buf->len;
if (len <= 0) {
return;
}
/* See the comment in spi_sam_fast_txrx re: interleaving. */
/* Write the first byte */
regs->SPI_TDR = SPI_TDR_TD(0);
len--;
while (len) {
while ((regs->SPI_SR & SPI_SR_TDRE) == 0) {
}
/* Load byte N+1 into the transmit register */
regs->SPI_TDR = SPI_TDR_TD(0);
len--;
/* Read byte N+0 from the receive register */
while ((regs->SPI_SR & SPI_SR_RDRF) == 0) {
}
*rx++ = (uint8_t)regs->SPI_RDR;
}
/* Read the final incoming byte */
while ((regs->SPI_SR & SPI_SR_RDRF) == 0) {
}
*rx = (uint8_t)regs->SPI_RDR;
spi_sam_finish(regs);
}
/* Fast path that writes and reads bufs of the same length */
static void spi_sam_fast_txrx(Spi *regs,
const struct spi_buf *tx_buf,
const struct spi_buf *rx_buf)
{
const uint8_t *tx = tx_buf->buf;
const uint8_t *txend = (uint8_t *)tx_buf->buf + tx_buf->len;
uint8_t *rx = rx_buf->buf;
size_t len = rx_buf->len;
if (len == 0) {
return;
}
/*
* The code below interleaves the transmit writes with the
* receive reads to keep the bus fully utilised. The code is
* equivalent to:
*
* Transmit byte 0
* Loop:
* - Transmit byte n+1
* - Receive byte n
* Receive the final byte
*/
/* Write the first byte */
regs->SPI_TDR = SPI_TDR_TD(*tx++);
while (tx != txend) {
while ((regs->SPI_SR & SPI_SR_TDRE) == 0) {
}
/* Load byte N+1 into the transmit register. TX is
* single buffered and we have at most one byte in
* flight so skip the DRE check.
*/
regs->SPI_TDR = SPI_TDR_TD(*tx++);
/* Read byte N+0 from the receive register */
while ((regs->SPI_SR & SPI_SR_RDRF) == 0) {
}
*rx++ = (uint8_t)regs->SPI_RDR;
}
/* Read the final incoming byte */
while ((regs->SPI_SR & SPI_SR_RDRF) == 0) {
}
*rx = (uint8_t)regs->SPI_RDR;
spi_sam_finish(regs);
}
/* Fast path where every overlapping tx and rx buffer is the same length */
static void spi_sam_fast_transceive(const struct device *dev,
const struct spi_config *config,
const struct spi_buf_set *tx_bufs,
const struct spi_buf_set *rx_bufs)
{
const struct spi_sam_config *cfg = dev->config;
size_t tx_count = 0;
size_t rx_count = 0;
Spi *regs = cfg->regs;
const struct spi_buf *tx = NULL;
const struct spi_buf *rx = NULL;
if (tx_bufs) {
tx = tx_bufs->buffers;
tx_count = tx_bufs->count;
}
if (rx_bufs) {
rx = rx_bufs->buffers;
rx_count = rx_bufs->count;
}
while (tx_count != 0 && rx_count != 0) {
if (tx->buf == NULL) {
spi_sam_fast_rx(regs, rx);
} else if (rx->buf == NULL) {
spi_sam_fast_tx(regs, tx);
} else {
spi_sam_fast_txrx(regs, tx, rx);
}
tx++;
tx_count--;
rx++;
rx_count--;
}
for (; tx_count != 0; tx_count--) {
spi_sam_fast_tx(regs, tx++);
}
for (; rx_count != 0; rx_count--) {
spi_sam_fast_rx(regs, rx++);
}
}
/* Returns true if the request is suitable for the fast
* path. Specifically, the bufs are a sequence of:
*
* - Zero or more RX and TX buf pairs where each is the same length.
* - Zero or more trailing RX only bufs
* - Zero or more trailing TX only bufs
*/
static bool spi_sam_is_regular(const struct spi_buf_set *tx_bufs,
const struct spi_buf_set *rx_bufs)
{
const struct spi_buf *tx = NULL;
const struct spi_buf *rx = NULL;
size_t tx_count = 0;
size_t rx_count = 0;
if (tx_bufs) {
tx = tx_bufs->buffers;
tx_count = tx_bufs->count;
}
if (rx_bufs) {
rx = rx_bufs->buffers;
rx_count = rx_bufs->count;
}
if (!tx || !rx) {
return true;
}
while (tx_count != 0 && rx_count != 0) {
if (tx->len != rx->len) {
return false;
}
tx++;
tx_count--;
rx++;
rx_count--;
}
return true;
}
static int spi_sam_transceive(const struct device *dev,
const struct spi_config *config,
const struct spi_buf_set *tx_bufs,
const struct spi_buf_set *rx_bufs)
{
const struct spi_sam_config *cfg = dev->config;
struct spi_sam_data *data = dev->data;
Spi *regs = cfg->regs;
int err;
spi_context_lock(&data->ctx, false, NULL, config);
err = spi_sam_configure(dev, config);
if (err != 0) {
goto done;
}
spi_context_cs_control(&data->ctx, true);
/* This driver special cases the common send only, receive
* only, and transmit then receive operations. This special
* casing is 4x faster than the spi_context() routines
* and allows the transmit and receive to be interleaved.
*/
if (spi_sam_is_regular(tx_bufs, rx_bufs)) {
spi_sam_fast_transceive(dev, config, tx_bufs, rx_bufs);
} else {
spi_context_buffers_setup(&data->ctx, tx_bufs, rx_bufs, 1);
do {
spi_sam_shift_master(regs, data);
} while (spi_sam_transfer_ongoing(data));
}
spi_context_cs_control(&data->ctx, false);
done:
spi_context_release(&data->ctx, err);
return err;
}
static int spi_sam_transceive_sync(const struct device *dev,
const struct spi_config *config,
const struct spi_buf_set *tx_bufs,
const struct spi_buf_set *rx_bufs)
{
return spi_sam_transceive(dev, config, tx_bufs, rx_bufs);
}
#ifdef CONFIG_SPI_ASYNC
static int spi_sam_transceive_async(const struct device *dev,
const struct spi_config *config,
const struct spi_buf_set *tx_bufs,
const struct spi_buf_set *rx_bufs,
struct k_poll_signal *async)
{
/* TODO: implement asyc transceive */
return -ENOTSUP;
}
#endif /* CONFIG_SPI_ASYNC */
static int spi_sam_release(const struct device *dev,
const struct spi_config *config)
{
struct spi_sam_data *data = dev->data;
spi_context_unlock_unconditionally(&data->ctx);
return 0;
}
static int spi_sam_init(const struct device *dev)
{
int err;
const struct spi_sam_config *cfg = dev->config;
struct spi_sam_data *data = dev->data;
soc_pmc_peripheral_enable(cfg->periph_id);
soc_gpio_list_configure(cfg->pins, cfg->num_pins);
err = spi_context_cs_configure_all(&data->ctx);
if (err < 0) {
return err;
}
spi_context_unlock_unconditionally(&data->ctx);
/* The device will be configured and enabled when transceive
* is called.
*/
return 0;
}
static const struct spi_driver_api spi_sam_driver_api = {
.transceive = spi_sam_transceive_sync,
#ifdef CONFIG_SPI_ASYNC
.transceive_async = spi_sam_transceive_async,
#endif
.release = spi_sam_release,
};
#define SPI_SAM_DEFINE_CONFIG(n) \
static const struct spi_sam_config spi_sam_config_##n = { \
.regs = (Spi *)DT_INST_REG_ADDR(n), \
.periph_id = DT_INST_PROP(n, peripheral_id), \
.num_pins = ATMEL_SAM_DT_INST_NUM_PINS(n), \
.pins = ATMEL_SAM_DT_INST_PINS(n), \
}
#define SPI_SAM_DEVICE_INIT(n) \
SPI_SAM_DEFINE_CONFIG(n); \
static struct spi_sam_data spi_sam_dev_data_##n = { \
SPI_CONTEXT_INIT_LOCK(spi_sam_dev_data_##n, ctx), \
SPI_CONTEXT_INIT_SYNC(spi_sam_dev_data_##n, ctx), \
SPI_CONTEXT_CS_GPIOS_INITIALIZE(DT_DRV_INST(n), ctx) \
}; \
DEVICE_DT_INST_DEFINE(n, &spi_sam_init, NULL, \
&spi_sam_dev_data_##n, \
&spi_sam_config_##n, POST_KERNEL, \
CONFIG_SPI_INIT_PRIORITY, &spi_sam_driver_api);
DT_INST_FOREACH_STATUS_OKAY(SPI_SAM_DEVICE_INIT)