1================= 2SPI userspace API 3================= 4 5SPI devices have a limited userspace API, supporting basic half-duplex 6read() and write() access to SPI slave devices. Using ioctl() requests, 7full duplex transfers and device I/O configuration are also available. 8 9:: 10 11 #include <fcntl.h> 12 #include <unistd.h> 13 #include <sys/ioctl.h> 14 #include <linux/types.h> 15 #include <linux/spi/spidev.h> 16 17Some reasons you might want to use this programming interface include: 18 19 * Prototyping in an environment that's not crash-prone; stray pointers 20 in userspace won't normally bring down any Linux system. 21 22 * Developing simple protocols used to talk to microcontrollers acting 23 as SPI slaves, which you may need to change quite often. 24 25Of course there are drivers that can never be written in userspace, because 26they need to access kernel interfaces (such as IRQ handlers or other layers 27of the driver stack) that are not accessible to userspace. 28 29 30DEVICE CREATION, DRIVER BINDING 31=============================== 32 33The spidev driver contains lists of SPI devices that are supported for 34the different hardware topology representations. 35 36The following are the SPI device tables supported by the spidev driver: 37 38 - struct spi_device_id spidev_spi_ids[]: list of devices that can be 39 bound when these are defined using a struct spi_board_info with a 40 .modalias field matching one of the entries in the table. 41 42 - struct of_device_id spidev_dt_ids[]: list of devices that can be 43 bound when these are defined using a Device Tree node that has a 44 compatible string matching one of the entries in the table. 45 46 - struct acpi_device_id spidev_acpi_ids[]: list of devices that can 47 be bound when these are defined using a ACPI device object with a 48 _HID matching one of the entries in the table. 49 50You are encouraged to add an entry for your SPI device name to relevant 51tables, if these don't already have an entry for the device. To do that, 52post a patch for spidev to the linux-spi@vger.kernel.org mailing list. 53 54It used to be supported to define an SPI device using the "spidev" name. 55For example, as .modalias = "spidev" or compatible = "spidev". But this 56is no longer supported by the Linux kernel and instead a real SPI device 57name as listed in one of the tables must be used. 58 59Not having a real SPI device name will lead to an error being printed and 60the spidev driver failing to probe. 61 62Sysfs also supports userspace driven binding/unbinding of drivers to 63devices that do not bind automatically using one of the tables above. 64To make the spidev driver bind to such a device, use the following:: 65 66 echo spidev > /sys/bus/spi/devices/spiB.C/driver_override 67 echo spiB.C > /sys/bus/spi/drivers/spidev/bind 68 69When the spidev driver is bound to a SPI device, the sysfs node for the 70device will include a child device node with a "dev" attribute that will 71be understood by udev or mdev (udev replacement from BusyBox; it's less 72featureful, but often enough). 73 74For a SPI device with chipselect C on bus B, you should see: 75 76 /dev/spidevB.C ... 77 character special device, major number 153 with 78 a dynamically chosen minor device number. This is the node 79 that userspace programs will open, created by "udev" or "mdev". 80 81 /sys/devices/.../spiB.C ... 82 as usual, the SPI device node will 83 be a child of its SPI master controller. 84 85 /sys/class/spidev/spidevB.C ... 86 created when the "spidev" driver 87 binds to that device. (Directory or symlink, based on whether 88 or not you enabled the "deprecated sysfs files" Kconfig option.) 89 90Do not try to manage the /dev character device special file nodes by hand. 91That's error prone, and you'd need to pay careful attention to system 92security issues; udev/mdev should already be configured securely. 93 94If you unbind the "spidev" driver from that device, those two "spidev" nodes 95(in sysfs and in /dev) should automatically be removed (respectively by the 96kernel and by udev/mdev). You can unbind by removing the "spidev" driver 97module, which will affect all devices using this driver. You can also unbind 98by having kernel code remove the SPI device, probably by removing the driver 99for its SPI controller (so its spi_master vanishes). 100 101Since this is a standard Linux device driver -- even though it just happens 102to expose a low level API to userspace -- it can be associated with any number 103of devices at a time. Just provide one spi_board_info record for each such 104SPI device, and you'll get a /dev device node for each device. 105 106 107BASIC CHARACTER DEVICE API 108========================== 109Normal open() and close() operations on /dev/spidevB.D files work as you 110would expect. 111 112Standard read() and write() operations are obviously only half-duplex, and 113the chipselect is deactivated between those operations. Full-duplex access, 114and composite operation without chipselect de-activation, is available using 115the SPI_IOC_MESSAGE(N) request. 116 117Several ioctl() requests let your driver read or override the device's current 118settings for data transfer parameters: 119 120 SPI_IOC_RD_MODE, SPI_IOC_WR_MODE ... 121 pass a pointer to a byte which will 122 return (RD) or assign (WR) the SPI transfer mode. Use the constants 123 SPI_MODE_0..SPI_MODE_3; or if you prefer you can combine SPI_CPOL 124 (clock polarity, idle high iff this is set) or SPI_CPHA (clock phase, 125 sample on trailing edge iff this is set) flags. 126 Note that this request is limited to SPI mode flags that fit in a 127 single byte. 128 129 SPI_IOC_RD_MODE32, SPI_IOC_WR_MODE32 ... 130 pass a pointer to a uin32_t 131 which will return (RD) or assign (WR) the full SPI transfer mode, 132 not limited to the bits that fit in one byte. 133 134 SPI_IOC_RD_LSB_FIRST, SPI_IOC_WR_LSB_FIRST ... 135 pass a pointer to a byte 136 which will return (RD) or assign (WR) the bit justification used to 137 transfer SPI words. Zero indicates MSB-first; other values indicate 138 the less common LSB-first encoding. In both cases the specified value 139 is right-justified in each word, so that unused (TX) or undefined (RX) 140 bits are in the MSBs. 141 142 SPI_IOC_RD_BITS_PER_WORD, SPI_IOC_WR_BITS_PER_WORD ... 143 pass a pointer to 144 a byte which will return (RD) or assign (WR) the number of bits in 145 each SPI transfer word. The value zero signifies eight bits. 146 147 SPI_IOC_RD_MAX_SPEED_HZ, SPI_IOC_WR_MAX_SPEED_HZ ... 148 pass a pointer to a 149 u32 which will return (RD) or assign (WR) the maximum SPI transfer 150 speed, in Hz. The controller can't necessarily assign that specific 151 clock speed. 152 153NOTES: 154 155 - At this time there is no async I/O support; everything is purely 156 synchronous. 157 158 - There's currently no way to report the actual bit rate used to 159 shift data to/from a given device. 160 161 - From userspace, you can't currently change the chip select polarity; 162 that could corrupt transfers to other devices sharing the SPI bus. 163 Each SPI device is deselected when it's not in active use, allowing 164 other drivers to talk to other devices. 165 166 - There's a limit on the number of bytes each I/O request can transfer 167 to the SPI device. It defaults to one page, but that can be changed 168 using a module parameter. 169 170 - Because SPI has no low-level transfer acknowledgement, you usually 171 won't see any I/O errors when talking to a non-existent device. 172 173 174FULL DUPLEX CHARACTER DEVICE API 175================================ 176 177See the spidev_fdx.c sample program for one example showing the use of the 178full duplex programming interface. (Although it doesn't perform a full duplex 179transfer.) The model is the same as that used in the kernel spi_sync() 180request; the individual transfers offer the same capabilities as are 181available to kernel drivers (except that it's not asynchronous). 182 183The example shows one half-duplex RPC-style request and response message. 184These requests commonly require that the chip not be deselected between 185the request and response. Several such requests could be chained into 186a single kernel request, even allowing the chip to be deselected after 187each response. (Other protocol options include changing the word size 188and bitrate for each transfer segment.) 189 190To make a full duplex request, provide both rx_buf and tx_buf for the 191same transfer. It's even OK if those are the same buffer. 192