|
| 1 | +Linux I2C slave interface description |
| 2 | +===================================== |
| 3 | + |
| 4 | +by Wolfram Sang < [email protected]> in 2014-15 |
| 5 | + |
| 6 | +Linux can also be an I2C slave in case I2C controllers have slave support. |
| 7 | +Besides this HW requirement, one also needs a software backend providing the |
| 8 | +actual functionality. An example for this is the slave-eeprom driver, which |
| 9 | +acts as a dual memory driver. While another I2C master on the bus can access it |
| 10 | +like a regular EEPROM, the Linux I2C slave can access the content via sysfs and |
| 11 | +retrieve/provide information as needed. The software backend driver and the I2C |
| 12 | +bus driver communicate via events. Here is a small graph visualizing the data |
| 13 | +flow and the means by which data is transported. The dotted line marks only one |
| 14 | +example. The backend could also use e.g. a character device, be in-kernel |
| 15 | +only, or something completely different: |
| 16 | + |
| 17 | + |
| 18 | + e.g. sysfs I2C slave events I/O registers |
| 19 | + +-----------+ v +---------+ v +--------+ v +------------+ |
| 20 | + | Userspace +........+ Backend +-----------+ Driver +-----+ Controller | |
| 21 | + +-----------+ +---------+ +--------+ +------------+ |
| 22 | + | | |
| 23 | + ----------------------------------------------------------------+-- I2C |
| 24 | + --------------------------------------------------------------+---- Bus |
| 25 | + |
| 26 | +Note: Technically, there is also the I2C core between the backend and the |
| 27 | +driver. However, at this time of writing, the layer is transparent. |
| 28 | + |
| 29 | + |
| 30 | +User manual |
| 31 | +=========== |
| 32 | + |
| 33 | +I2C slave backends behave like standard I2C clients. So, you can instantiate |
| 34 | +them like described in the document 'instantiating-devices'. A quick example |
| 35 | +for instantiating the slave-eeprom driver from userspace: |
| 36 | + |
| 37 | + # echo 0-0064 > /sys/bus/i2c/drivers/i2c-slave-eeprom/bind |
| 38 | + |
| 39 | +Each backend should come with separate documentation to describe its specific |
| 40 | +behaviour and setup. |
| 41 | + |
| 42 | + |
| 43 | +Developer manual |
| 44 | +================ |
| 45 | + |
| 46 | +I2C slave events |
| 47 | +---------------- |
| 48 | + |
| 49 | +The bus driver sends an event to the backend using the following function: |
| 50 | + |
| 51 | + ret = i2c_slave_event(client, event, &val) |
| 52 | + |
| 53 | +'client' describes the i2c slave device. 'event' is one of the special event |
| 54 | +types described hereafter. 'val' holds an u8 value for the data byte to be |
| 55 | +read/written and is thus bidirectional. The pointer to val must always be |
| 56 | +provided even if val is not used for an event, i.e. don't use NULL here. 'ret' |
| 57 | +is the return value from the backend. Mandatory events must be provided by the |
| 58 | +bus drivers and must be checked for by backend drivers. |
| 59 | + |
| 60 | +Event types: |
| 61 | + |
| 62 | +* I2C_SLAVE_WRITE_REQUESTED (mandatory) |
| 63 | + |
| 64 | +'val': unused |
| 65 | +'ret': always 0 |
| 66 | + |
| 67 | +Another I2C master wants to write data to us. This event should be sent once |
| 68 | +our own address and the write bit was detected. The data did not arrive yet, so |
| 69 | +there is nothing to process or return. Wakeup or initialization probably needs |
| 70 | +to be done, though. |
| 71 | + |
| 72 | +* I2C_SLAVE_READ_REQUESTED (mandatory) |
| 73 | + |
| 74 | +'val': backend returns first byte to be sent |
| 75 | +'ret': always 0 |
| 76 | + |
| 77 | +Another I2C master wants to read data from us. This event should be sent once |
| 78 | +our own address and the read bit was detected. After returning, the bus driver |
| 79 | +should transmit the first byte. |
| 80 | + |
| 81 | +* I2C_SLAVE_WRITE_RECEIVED (mandatory) |
| 82 | + |
| 83 | +'val': bus driver delivers received byte |
| 84 | +'ret': 0 if the byte should be acked, some errno if the byte should be nacked |
| 85 | + |
| 86 | +Another I2C master has sent a byte to us which needs to be set in 'val'. If 'ret' |
| 87 | +is zero, the bus driver should ack this byte. If 'ret' is an errno, then the byte |
| 88 | +should be nacked. |
| 89 | + |
| 90 | +* I2C_SLAVE_READ_PROCESSED (mandatory) |
| 91 | + |
| 92 | +'val': backend returns next byte to be sent |
| 93 | +'ret': always 0 |
| 94 | + |
| 95 | +The bus driver requests the next byte to be sent to another I2C master in |
| 96 | +'val'. Important: This does not mean that the previous byte has been acked, it |
| 97 | +only means that the previous byte is shifted out to the bus! To ensure seamless |
| 98 | +transmission, most hardware requests the next byte when the previous one is |
| 99 | +still shifted out. If the master sends NACK and stops reading after the byte |
| 100 | +currently shifted out, this byte requested here is never used. It very likely |
| 101 | +needs to be sent again on the next I2C_SLAVE_READ_REQUEST, depending a bit on |
| 102 | +your backend, though. |
| 103 | + |
| 104 | +* I2C_SLAVE_STOP (mandatory) |
| 105 | + |
| 106 | +'val': unused |
| 107 | +'ret': always 0 |
| 108 | + |
| 109 | +A stop condition was received. This can happen anytime and the backend should |
| 110 | +reset its state machine for I2C transfers to be able to receive new requests. |
| 111 | + |
| 112 | + |
| 113 | +Software backends |
| 114 | +----------------- |
| 115 | + |
| 116 | +If you want to write a software backend: |
| 117 | + |
| 118 | +* use a standard i2c_driver and its matching mechanisms |
| 119 | +* write the slave_callback which handles the above slave events |
| 120 | + (best using a state machine) |
| 121 | +* register this callback via i2c_slave_register() |
| 122 | + |
| 123 | +Check the i2c-slave-eeprom driver as an example. |
| 124 | + |
| 125 | + |
| 126 | +Bus driver support |
| 127 | +------------------ |
| 128 | + |
| 129 | +If you want to add slave support to the bus driver: |
| 130 | + |
| 131 | +* implement calls to register/unregister the slave and add those to the |
| 132 | + struct i2c_algorithm. When registering, you probably need to set the i2c |
| 133 | + slave address and enable slave specific interrupts. If you use runtime pm, you |
| 134 | + should use pm_runtime_forbid() because your device usually needs to be powered |
| 135 | + on always to be able to detect its slave address. When unregistering, do the |
| 136 | + inverse of the above. |
| 137 | + |
| 138 | +* Catch the slave interrupts and send appropriate i2c_slave_events to the backend. |
| 139 | + |
| 140 | +Check the i2c-rcar driver as an example. |
| 141 | + |
| 142 | + |
| 143 | +About ACK/NACK |
| 144 | +-------------- |
| 145 | + |
| 146 | +It is good behaviour to always ACK the address phase, so the master knows if a |
| 147 | +device is basically present or if it mysteriously disappeared. Using NACK to |
| 148 | +state being busy is troublesome. SMBus demands to always ACK the address phase, |
| 149 | +while the I2C specification is more loose on that. Most I2C controllers also |
| 150 | +automatically ACK when detecting their slave addresses, so there is no option |
| 151 | +to NACK them. For those reasons, this API does not support NACK in the address |
| 152 | +phase. |
| 153 | + |
| 154 | +Currently, there is no slave event to report if the master did ACK or NACK a |
| 155 | +byte when it reads from us. We could make this an optional event if the need |
| 156 | +arises. However, cases should be extremely rare because the master is expected |
| 157 | +to send STOP after that and we have an event for that. Also, keep in mind not |
| 158 | +all I2C controllers have the possibility to report that event. |
| 159 | + |
| 160 | + |
| 161 | +About buffers |
| 162 | +------------- |
| 163 | + |
| 164 | +During development of this API, the question of using buffers instead of just |
| 165 | +bytes came up. Such an extension might be possible, usefulness is unclear at |
| 166 | +this time of writing. Some points to keep in mind when using buffers: |
| 167 | + |
| 168 | +* Buffers should be opt-in and slave drivers will always have to support |
| 169 | + byte-based transactions as the ultimate fallback because this is how the |
| 170 | + majority of HW works. |
| 171 | + |
| 172 | +* For backends simulating hardware registers, buffers are not helpful because |
| 173 | + on writes an action should be immediately triggered. For reads, the data in |
| 174 | + the buffer might get stale. |
| 175 | + |
| 176 | +* A master can send STOP at any time. For partially transferred buffers, this |
| 177 | + means additional code to handle this exception. Such code tends to be |
| 178 | + error-prone. |
| 179 | + |
0 commit comments