mbed library sources. Supersedes mbed-src. Add PORTG support for STM32L476JG (SensorTile kit)
Fork of mbed-dev by
Diff: targets/TARGET_Maxim/TARGET_MAX32620/spi_api.c
- Revision:
- 149:156823d33999
diff -r 21d94c44109e -r 156823d33999 targets/TARGET_Maxim/TARGET_MAX32620/spi_api.c --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/targets/TARGET_Maxim/TARGET_MAX32620/spi_api.c Fri Oct 28 11:17:30 2016 +0100 @@ -0,0 +1,560 @@ +/******************************************************************************* + * Copyright (C) 2016 Maxim Integrated Products, Inc., All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included + * in all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL MAXIM INTEGRATED BE LIABLE FOR ANY CLAIM, DAMAGES + * OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * Except as contained in this notice, the name of Maxim Integrated + * Products, Inc. shall not be used except as stated in the Maxim Integrated + * Products, Inc. Branding Policy. + * + * The mere transfer of this software does not imply any licenses + * of trade secrets, proprietary technology, copyrights, patents, + * trademarks, maskwork rights, or any other form of intellectual + * property whatsoever. Maxim Integrated Products, Inc. retains all + * ownership rights. + ******************************************************************************* + */ + +#include <string.h> +#include "mbed_assert.h" +#include "cmsis.h" +#include "spi_api.h" +#include "spi_multi_api.h" +#include "pinmap.h" +#include "ioman_regs.h" +#include "clkman_regs.h" +#include "PeripheralPins.h" + +#define DEFAULT_CHAR 8 +#define DEFAULT_MODE 0 +#define DEFAULT_FREQ 1000000 + +// BYTE maximums for FIFO and page writes; FIFO depth spec'd as 16-bit words +#define SPI_MAX_BYTE_LEN (MXC_CFG_SPI_FIFO_DEPTH * 2) +#define SPI_MAX_PAGE_LEN (MXC_CFG_SPI_FIFO_DEPTH * 2) + +#if DEVICE_SPI_ASYNCH +// Instance references for async transactions +static struct spi_s *state[MXC_CFG_SPI_INSTANCES] = {NULL}; +#endif + +//****************************************************************************** +void spi_init(spi_t *obj, PinName mosi, PinName miso, PinName sclk, PinName ssel) +{ + // Make sure pins are pointing to the same SPI instance + SPIName spi_mosi = (SPIName)pinmap_peripheral(mosi, PinMap_SPI_MOSI); + SPIName spi_miso = (SPIName)pinmap_peripheral(miso, PinMap_SPI_MISO); + SPIName spi_sclk = (SPIName)pinmap_peripheral(sclk, PinMap_SPI_SCLK); + SPIName spi_ssel = (SPIName)pinmap_peripheral(ssel, PinMap_SPI_SSEL); + + SPIName spi_data = (SPIName)pinmap_merge(spi_mosi, spi_miso); + SPIName spi_cntl; + + // Give the application the option to manually control Slave Select + if ((SPIName)spi_ssel != (SPIName)NC) { + spi_cntl = (SPIName)pinmap_merge(spi_sclk, spi_ssel); + // Slave select is currently limited to slave select zero. If others are + // to be supported a function to map PinName to a value suitable for use + // in mstr_cfg.slave_sel will be required. + obj->spi.ssel = 0; + } else { + spi_cntl = spi_sclk; + obj->spi.ssel = -1; + } + + SPIName spi = (SPIName)pinmap_merge(spi_data, spi_cntl); + + MBED_ASSERT((SPIName)spi != (SPIName)NC); + + // Set the obj pointer to the proper SPI Instance + obj->spi.spi = (mxc_spi_regs_t*)spi; + + // Set the SPI index and FIFOs + obj->spi.index = MXC_SPI_GET_IDX(obj->spi.spi); + obj->spi.fifo = MXC_SPI_GET_SPI_FIFO(obj->spi.index); + + // Configure the pins + pinmap_pinout(mosi, PinMap_SPI_MOSI); + pinmap_pinout(miso, PinMap_SPI_MISO); + pinmap_pinout(sclk, PinMap_SPI_SCLK); + pinmap_pinout(ssel, PinMap_SPI_SSEL); + +#if DEVICE_SPI_ASYNCH + // Configure default page size; size is known to async interface + obj->spi.spi->mstr_cfg = (obj->spi.spi->mstr_cfg & ~MXC_F_SPI_MSTR_CFG_PAGE_SIZE) | MXC_S_SPI_MSTR_CFG_PAGE_32B; +#endif + + // Enable SPI and FIFOs + obj->spi.spi->gen_ctrl = (MXC_F_SPI_GEN_CTRL_SPI_MSTR_EN | + MXC_F_SPI_GEN_CTRL_TX_FIFO_EN | + MXC_F_SPI_GEN_CTRL_RX_FIFO_EN ); + + obj->spi.sclk = sclk; // save the sclk PinName in the object as a key for Quad SPI pin mapping lookup + spi_master_width(obj, 0); // default this for Single SPI communications +} + +//****************************************************************************** +void spi_format(spi_t *obj, int bits, int mode, int slave) +{ + // Check the validity of the inputs + MBED_ASSERT(((bits >= 1) && (bits <= 32)) && ((mode >= 0) && (mode <= 3))); + + // Only supports master mode + MBED_ASSERT(!slave); + + // Save formatting data + obj->spi.bits = bits; + + // Set the mode + MXC_SET_FIELD(&obj->spi.spi->mstr_cfg, MXC_F_SPI_MSTR_CFG_SPI_MODE, mode << MXC_F_SPI_MSTR_CFG_SPI_MODE_POS); +} + +//****************************************************************************** +void spi_frequency(spi_t *obj, int hz) +{ + // Maximum frequency is half the system frequency + MBED_ASSERT((unsigned int)hz <= (SystemCoreClock / 2)); + unsigned clocks = ((SystemCoreClock / 2) / hz); + + // Figure out the divider ratio + int clk_div = 1; + while (clk_div < 10) { + if (clocks < 0x10) { + break; + } + clk_div++; + clocks = clocks >> 1; + } + + // Turn on the SPI clock + if (obj->spi.index == 0) { + MXC_CLKMAN->sys_clk_ctrl_11_spi0 = clk_div; + } else if (obj->spi.index == 1) { + MXC_CLKMAN->sys_clk_ctrl_12_spi1 = clk_div; + } else if (obj->spi.index == 2) { + MXC_CLKMAN->sys_clk_ctrl_13_spi2 = clk_div; + } else { + MBED_ASSERT(0); + } + + // Set the number of clocks to hold sclk high and low + MXC_SET_FIELD(&obj->spi.spi->mstr_cfg, (MXC_F_SPI_MSTR_CFG_SCK_HI_CLK | MXC_F_SPI_MSTR_CFG_SCK_LO_CLK), + ((clocks << MXC_F_SPI_MSTR_CFG_SCK_HI_CLK_POS) | (clocks << MXC_F_SPI_MSTR_CFG_SCK_LO_CLK_POS))); +} + +//****************************************************************************** +void spi_master_width(spi_t *obj, SpiWidth width) +{ + // Save the width to be used in the SPI header + switch (width) { + case WidthSingle: + obj->spi.width = MXC_S_SPI_FIFO_WIDTH_SINGLE; + break; + case WidthDual: + obj->spi.width = MXC_S_SPI_FIFO_WIDTH_DUAL; + break; + case WidthQuad: + obj->spi.width = MXC_S_SPI_FIFO_WIDTH_QUAD; + // do pin mapping for SDIO[2] and SDIO[3] if Quad SPI is selected + pinmap_pinout(obj->spi.sclk, PinMap_SPI_QUAD); + break; + default: + MBED_ASSERT(0); + } +} + +//****************************************************************************** +/** Performs a master write or read transaction + * + * @param[in] obj The SPI peripheral to use for sending + * @param[in] value The value to send + * @param[in] direction Direction of the transaction, TX, RX or both + * @return Returns the value received during send + */ +static int spi_master_transaction(spi_t *obj, int value, uint32_t direction) +{ + int bits; + + // Create the header + uint16_t header = (direction | // direction based on SPI object + MXC_S_SPI_FIFO_UNIT_BITS | // unit size + ((obj->spi.bits == 32) ? 0 : obj->spi.bits << MXC_F_SPI_FIFO_SIZE_POS) | // Number of units + obj->spi.width | // I/O width + ((obj->spi.ssel == -1) ? 0 : 1 << MXC_F_SPI_FIFO_DASS_POS)); + + // Send the message header + *obj->spi.fifo->trans_16 = header; + + // Send the data + if (obj->spi.bits < 17) { + *obj->spi.fifo->trans_16 = (uint16_t)value; + } else { + *obj->spi.fifo->trans_32 = (uint32_t)value; + } + + // Get the data + bits = obj->spi.bits; + int result = 0; + int i = 0; + while (bits > 0) { + // Wait for data + while (((obj->spi.spi->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED) + >> MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED_POS) < 1); + + result |= (*obj->spi.fifo->rslts_8 << (i++*8)); + bits-=8; + } + + return result; +} + +//****************************************************************************** +int spi_master_write(spi_t *obj, int value) +{ + // set the fifo direction for full duplex, TX and RX simultaneously + return spi_master_transaction(obj, value, MXC_S_SPI_FIFO_DIR_BOTH); +} + +//****************************************************************************** +int spi_master_read(spi_t *obj) +{ + return spi_master_transaction(obj, 0xFF, MXC_S_SPI_FIFO_DIR_RX); +} + +//****************************************************************************** +// spi_busy() is part of the synchronous API, it is not used by the asynchronous API. +int spi_busy(spi_t *obj) +{ + return !(obj->spi.spi->intfl & MXC_F_SPI_INTFL_TX_READY); +} + +#if DEVICE_SPI_ASYNCH +//****************************************************************************** +static uint32_t spi_master_read_rxfifo(mxc_spi_regs_t *spim, mxc_spi_fifo_regs_t *fifo, uint8_t *data, uint32_t len) +{ + uint32_t num = 0; + uint32_t avail = ((spim->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED) >> MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED_POS); + + // Get data from the RXFIFO + while (avail && (len - num)) { + // Save data from the RXFIFO + if ((avail >= 4) && ((len - num) >= 4)) { + uint32_t temp = *fifo->rslts_32; + data[num++] = temp; + data[num++] = temp >> 8; + data[num++] = temp >> 16; + data[num++] = temp >> 24; + avail -= 4; + } else if ((avail >= 2) && ((len - num) >= 2)) { + uint16_t temp = *fifo->rslts_16; + data[num++] = temp; + data[num++] = temp >> 8; + avail -= 2; + } else { + data[num++] = *fifo->rslts_8; + avail--; + } + + // Check to see if there is more data in the FIFO + if (avail == 0) { + avail = ((spim->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED) >> MXC_F_SPI_FIFO_CTRL_RX_FIFO_USED_POS); + } + } + + return num; +} + +//****************************************************************************** +static uint32_t spi_master_transfer_handler(spi_t *obj) +{ + uint8_t read; + uint8_t write; + uint16_t header; + uint32_t pages; + uint32_t bytes; + uint32_t inten; + unsigned remain; + unsigned bytes_read; + unsigned head_rem_temp; + unsigned avail; + struct spi_s *req = &obj->spi; + mxc_spi_regs_t *spim = obj->spi.spi; + mxc_spi_fifo_regs_t *fifo = obj->spi.fifo; + + inten = 0; + + // Figure out if we're reading + read = (req->rx_data != NULL) ? 1 : 0; + + // Figure out if we're writing + write = (req->tx_data != NULL) ? 1 : 0; + + // Read byte from the FIFO if we are reading + if (read) { + + // Read all of the data in the RXFIFO, or until we don't need anymore + bytes_read = spi_master_read_rxfifo(spim, fifo, &req->rx_data[req->read_num], (req->len - req->read_num)); + + req->read_num += bytes_read; + + // Adjust head_rem if we are only reading + if (!write && (req->head_rem > 0)) { + req->head_rem -= bytes_read; + } + + // Figure out how many bytes we have left to read + if (req->head_rem > 0) { + remain = req->head_rem; + } else { + remain = req->len - req->read_num; + } + + if (remain) { + + // Set the RX interrupts + if (remain > MXC_CFG_SPI_FIFO_DEPTH) { + spim->fifo_ctrl = ((spim->fifo_ctrl & ~MXC_F_SPI_FIFO_CTRL_RX_FIFO_AF_LVL) | + ((MXC_CFG_SPI_FIFO_DEPTH - 2) << MXC_F_SPI_FIFO_CTRL_RX_FIFO_AF_LVL_POS)); + } else { + spim->fifo_ctrl = ((spim->fifo_ctrl & ~MXC_F_SPI_FIFO_CTRL_RX_FIFO_AF_LVL) | + ((remain - 1) << MXC_F_SPI_FIFO_CTRL_RX_FIFO_AF_LVL_POS)); + } + + inten |= MXC_F_SPI_INTEN_RX_FIFO_AF; + } + } + + // Figure out how many bytes we have left to send headers for + if (write) { + remain = req->len - req->write_num; + } else { + remain = req->len - req->read_num; + } + + // See if we need to send a new header + if ((req->head_rem <= 0) && remain) { + + // Set the transaction configuration in the header + header = ((write | (read << 1)) << MXC_F_SPI_FIFO_DIR_POS) | (req->width << MXC_F_SPI_FIFO_WIDTH_POS); + + if (remain >= SPI_MAX_BYTE_LEN) { + + // Send a 32 byte header + if (remain == SPI_MAX_BYTE_LEN) { + + header |= (MXC_S_SPI_FIFO_UNIT_BYTES | MXC_F_SPI_FIFO_DASS); + + // Save the number of bytes we need to write to the FIFO + bytes = SPI_MAX_BYTE_LEN; + + } else { + // Send in increments of 32 byte pages + header |= MXC_S_SPI_FIFO_UNIT_PAGES; + pages = remain / SPI_MAX_PAGE_LEN; + + if (pages >= 32) { + // 0 maps to 32 in the header + bytes = 32 * SPI_MAX_PAGE_LEN; + } else { + header |= (pages << MXC_F_SPI_FIFO_SIZE_POS); + bytes = pages * SPI_MAX_PAGE_LEN; + } + + // Check if this is the last header we will send + if ((remain - bytes) == 0) { + header |= MXC_F_SPI_FIFO_DASS; + } + } + + fifo->trans_16[0] = header; + + // Save the number of bytes we need to write to the FIFO + req->head_rem = bytes; + + } else { + // Send final header with the number of bytes remaining and de-assert the SS at the end of the transaction + header |= (MXC_S_SPI_FIFO_UNIT_BYTES | (remain << MXC_F_SPI_FIFO_SIZE_POS) | MXC_F_SPI_FIFO_DASS); + fifo->trans_16[0] = header; + req->head_rem = remain; + } + } + + // Put data into the FIFO if we are writing + remain = req->len - req->write_num; + head_rem_temp = req->head_rem; + if (write && head_rem_temp) { + + // Fill the FIFO + avail = (MXC_CFG_SPI_FIFO_DEPTH - ((spim->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_TX_FIFO_USED) >> MXC_F_SPI_FIFO_CTRL_TX_FIFO_USED_POS)); + + // Use memcpy for everything except the last byte in odd length transactions + while ((avail >= 2) && (head_rem_temp >= 2)) { + + unsigned length; + if (head_rem_temp < avail) { + length = head_rem_temp; + } else { + length = avail; + } + + // Only memcpy even numbers + length = ((length / 2) * 2); + + memcpy((void*)fifo->trans_32, &(req->tx_data[req->write_num]), length); + + head_rem_temp -= length; + req->write_num += length; + + avail = (MXC_CFG_SPI_FIFO_DEPTH - ((spim->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_TX_FIFO_USED) >> MXC_F_SPI_FIFO_CTRL_TX_FIFO_USED_POS)); + } + + // Copy the last byte and pad with 0xF0 to not get confused as header + if ((avail >= 1) && (head_rem_temp == 1)) { + + // Write the last byte + fifo->trans_16[0] = (0xF000 | req->tx_data[req->write_num]); + + avail -= 1; + req->write_num += 1; + head_rem_temp -= 1; + } + + req->head_rem = head_rem_temp; + remain = req->len - req->write_num; + + // Set the TX interrupts + if (remain) { + + // Set the TX FIFO almost empty interrupt if we have to refill + spim->fifo_ctrl = ((spim->fifo_ctrl & ~MXC_F_SPI_FIFO_CTRL_TX_FIFO_AE_LVL) | + ((MXC_CFG_SPI_FIFO_DEPTH - 2) << MXC_F_SPI_FIFO_CTRL_TX_FIFO_AE_LVL_POS)); + + inten |= MXC_F_SPI_INTEN_TX_FIFO_AE; + } + } + + // Check to see if we've finished reading and writing + if (((read && (req->read_num == req->len)) || !read) && + ((req->write_num == req->len) || !write)) { + + // Disable interrupts + spim->inten = 0; + } + + // Enable the SPIM interrupts + return inten; +} + +//****************************************************************************** +void spi_master_transfer(spi_t *obj, const void *tx, size_t tx_length, void *rx, size_t rx_length, uint8_t bit_width, uint32_t handler, uint32_t event, DMAUsage hint) +{ + MBED_ASSERT(tx_length == rx_length); + MBED_ASSERT(bit_width == obj->spi.bits); + + // Save object reference for callback + state[obj->spi.index] = &obj->spi; + + // Initialize request info + obj->spi.tx_data = tx; + obj->spi.rx_data = rx; + obj->spi.len = tx_length; + obj->spi.callback = (void(*)())handler; + obj->spi.event = event; + // Clear transfer state + obj->spi.read_num = 0; + obj->spi.write_num = 0; + obj->spi.head_rem = 0; + + NVIC_EnableIRQ(MXC_SPI_GET_IRQ(obj->spi.index)); + + obj->spi.spi->inten = spi_master_transfer_handler(obj); +} + +//****************************************************************************** +uint32_t spi_irq_handler_asynch(spi_t *obj) +{ + mxc_spi_regs_t *spim = obj->spi.spi; + uint32_t flags; + + // Clear the interrupt flags + spim->inten = 0; + flags = spim->intfl; + spim->intfl = flags; + + // Figure out if this SPIM has an active request + if (flags) { + if ((spim->inten = spi_master_transfer_handler(obj)) != 0) { + return 0; + } + } + + state[obj->spi.index] = NULL; + + return SPI_EVENT_COMPLETE; +} + +//****************************************************************************** +uint8_t spi_active(spi_t *obj) +{ + mxc_spi_regs_t *spim = obj->spi.spi; + + // Check to see if there are any ongoing transactions + if ((state[obj->spi.index] == NULL) && + !(spim->fifo_ctrl & MXC_F_SPI_FIFO_CTRL_TX_FIFO_USED)) { + return 0; + } + + return 1; +} + +//****************************************************************************** +void spi_abort_asynch(spi_t *obj) +{ + mxc_spi_regs_t *spim = obj->spi.spi; + + // Disable interrupts, clear the flags + spim->inten = 0; + spim->intfl = spim->intfl; + + // Reset the SPIM to cancel the on ongoing transaction + spim->gen_ctrl &= ~(MXC_F_SPI_GEN_CTRL_SPI_MSTR_EN); + spim->gen_ctrl |= (MXC_F_SPI_GEN_CTRL_SPI_MSTR_EN); + + state[obj->spi.index] = NULL; +} + +//****************************************************************************** +static void SPI_IRQHandler(int spim_num) +{ + if (state[spim_num] != NULL) { + if (state[spim_num]->callback != NULL) { + state[spim_num]->callback(); + return; + } + } + mxc_spi_regs_t *spim = MXC_SPI_GET_SPI(spim_num); + spim->inten = 0; +} + +//****************************************************************************** +void SPI0_IRQHandler(void) { SPI_IRQHandler(0); } +void SPI1_IRQHandler(void) { SPI_IRQHandler(1); } +void SPI2_IRQHandler(void) { SPI_IRQHandler(2); } + +#endif