While the majority of hardware interfacing should be done in kernel-space (using device driver kernel modules), some applications may be simplified by doing I/O from userspace. When using SPI from userspace, the spidev driver is used for generic SPI devices.
While the kernel does have some documentation1 on using spidev, some of the best information available can be gleaned from the associated headers and kernel source, which is a pretty common theme with Linux drivers and interfaces (unfortunately). When you are looking for the various available options for things like the SPI mode and device configuration, you should check in the kernel source tree in include/uapi/linux/spi/spidev.h or in the installed header at /usr/include/linux/spi/spidev.h (or a similar location for cross-compile toolchains and such). Additionally, the kernel source includes an example program using the SPI userspace API, which is a good reference.
Here is an example of opening a SPI device and configuring it for 10MHz in SPI mode 2 with 8 bits per word:
#include <errno.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <unistd.h>
#include <linux/spi/spidev.h>
#include <linux/types.h>
int openspidev(int bus, int cs) {
char devpath[35];
if (bus < 0 || cs < 0) {
errno = EINVAL;
return -1;
}
snprintf(devpath, 35, "/dev/spidev%d.%d", bus, cs);
// check if the device exists
if (access(devpath, F_OK) != 0) {
return -1;
}
return open(devpath, O_SYNC | O_RDWR | O_CLOEXEC);
}
int main(int argc, char** argv) {
int bus;
int cs;
int fd;
uint8_t tmp8;
uint32_t tmp32;
int ret;
if (argc != 3) {
fprintf(stderr, "usage: %s bus cs\n", argv[0]);
return 1;
}
bus = atoi(argv[1]);
cs = atoi(argv[2]);
fd = openspidev(bus, cs);
if (fd < 0) {
perror("openspidev");
return 1;
}
/* configure the SPI device */
// bits per word = 8
tmp8 = 8;
ret = ioctl(fd, SPI_IOC_WR_BITS_PER_WORD, &tmp8);
if (ret < 0) {
perror("ioctl: SPI_IOC_WR_BITS_PER_WORD");
return 1;
}
// max speed = 1MHz
tmp32 = 1 * 1000 * 1000;
ret = ioctl(fd, SPI_IOC_WR_MAX_SPEED_HZ, &tmp32);
if (ret < 0) {
perror("ioctl: SPI_IOC_WR_MAX_SPEED_HZ");
return 1;
}
// set mode to mode 2
// you can use the additional mode flags which don't fit in a single byte
// here, since we will use WR_MODE32, while the normal WR_MODE only accepts
// a uin8_t
tmp32 = SPI_MODE_2; // alternatively SPI_CPOL & ~SPI_CPHA, etc.
ret = ioctl(fd, SPI_IOC_WR_MODE32, &tmp32);
if (ret < 0) {
perror("ioctl: SPI_IOC_WR_MODE32");
return 1;
}
/* use the device as necessary here */
close(fd);
return 0;
}Spidev is capable of both half-duplex and full-duplex SPI transfers. This can be done by simply not specifying a TX or RX buffer in a SPI transaction (though of course at least one must be present). Also, both pointers may point to the same buffer, allowing full-duplex reading and writing with only one buffer. This example will write, then read, then perform a full-duplex transmission. Error handling is omitted for brevity, so make sure to check for errors. SPI errors are pretty rare, though, since there's no acknowledgment or handshaking with SPI devices.
/* pick up where we left off above, so 'fd' is our SPI device file descriptor */
// setup the SPI transfer structure to be used in the examples
struct spi_ioc_transfer m;
memset(&m, 0, sizeof(m));
/* half-duplex: write 5 bytes */
uint8_t txbuf[5] = { 0xAA, 0xBB, 0xCC, 0xDD, 0xEE };
m.tx_buf = (__u64)txbuf; // yes, they pass pointers as __u64 :(
m.rx_buf = 0;
m.len = 5;
ioctl(fd, SPI_IOC_MESSAGE(1), &m);
/* half-duplex: read 5 bytes */
uint8_t rxbuf[5] = {0};
m.tx_buf = 0;
m.rx_buf = (__u64)rxbuf;
m.len = 5;
ioctl(fd, SPI_IOC_MESSAGE(1), &m);
/* full-duplex: write 2 bytes, read 3 */
uint8_t txrxbuf[5] = { 0xAA, 0xBB, 0, 0, 0 };
m.tx_buf = (__u64)txrxbuf;
m.rx_buf = (__u64)txrxbuf;
m.len = 5;
ioctl(fd, SPI_IOC_MESSAGE(1), &m);While the above examples should suffice for most uses, there are some handy
features available that were not mentioned. The spi_ioc_transfer structure
contains a lot of useful flags and configuration options which are documented in
spidev.h. These allow overriding certain configuration options on a per-transfer
basis, such as bits per word, speed, and others. See spidev.h for the full suite
of options.
Additionally, the SPI_IOC_MESSAGE macro takes an argument which is the number
of messages to be transmitted. Above, we only used 1, as there was only one
message. However, you can pass entire arrays of messages to be transmitted, and
you can even specify whether or not the CS line should change after each message
using a flag in the spi_ioc_transfer structure. Here's an example sending
3 messages, receiving data on the last one, and only toggling the CS line after
the second and third transmissions. If all goes well, a logic analyzer should
show you something like this:
MOSI: _/[ 1 ]\_/[ 2 ]\___/[ 3 ]\_
MISO: _XXXXXXXXXXXXXXX___/[ 3 ]\_
CS: \_______________/‾\_______/
Here's the code:
struct spi_ioc_transfer msgs[3];
memset(msgs, 0, 3 * sizeof(*msgs));
for (int i = 0; i < 3; i++) {
msgs[i].len = 5;
msgs[i].tx_buf = /* buffer */;
}
// we don't have to set this for msgs[2], as it's the last transfer, so the
// deselect is implicit (which is why you don't have to set this to 1 if you're
// only sending one message)
msgs[1].cs_change = 1;
// receive data on the third packet
msgs[2].rx_buf = /* buffer */;
// finally, send our data
ioctl(fd, SPI_IOC_MESSAGE(3), msgs);