Protocol decoder API

From sigrok
Revision as of 01:57, 1 May 2010 by Uwe Hermann (talk | contribs) (Created page with 'This page describes how Protocol Decoders (PDs) work in sigrok. == Architecture == The frontend gets input from the user on which PDs to use in an acquisition session. It then …')
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

This page describes how Protocol Decoders (PDs) work in sigrok.

Architecture

The frontend gets input from the user on which PDs to use in an acquisition session. It then configures these into the session with session_pd_add(). As the first PD is added, the session sets up an additional datafeed callback to itself, which it uses as input to the first PD in the stack. The output of that both sent to the frontend, along with its original datafeed, as well as fed into the next PD in the stack.

The frontend thus gets the raw datafeed as well as a feed from every PD in the stack. Which of these different feeds is actually displayed to the user is a matter of configuration or selection by the user; it should be possible, for example, to have sigrok-cli print only the top of the PD stack's output on stdout.

  • All PDs are written in Python. Only source code will be used (i.e. no .pyc or .pyo files).
  • Every PD registers its name, descriptionm, capabilities, etc by populating a hash (dict) in its own main namespace called "register".
  • PDs will be stacked together, so the user can construct a decoding pipeline. The control of communication to/from PDs is done by the python controller.
  • The data feed into the PDs will be streamed, so they will run in real time as the data comes in from the hardware.
  • In order to keep PDs simple, they don't have to deal with the intricacies of the datafeed packets. instead, the python interface in libsigrokdecode will hide the details from the python code:
    • when receiving a DF_HEADER packet going to a PD, the controller intercepts the packet and instead generates a DF_HEADER "coming from" that PD across the session bus, with that PD's output characteristics.
    • data packets get translated into a bytestream, which the PDs access through an API: the function get_sample() is a blocking call into the controller, which only returns when a datafeed packet has arrived, and its payload queued up for the PD.
    • DF_END packets are translated into an EOF on the get_sample() call.

API

A PD must contain at least two definitions

  • a populated dict register which defines the name, capabilities etc of the PD.

This is an example:

register = {
 'id': 'spi',
 'description': 'SPI',
 'protocol in': 'raw',
 'probes': {
   ['sclk', "sck", "clk"],
   ["mosi", "simo", "sdi", "di", "si"],
   ["miso", "somi", "sdo", "do", "so"],
   ["ncs", "cs", "csb", "nss", "ste"],
 ],
 'protocol out': 'bytes'
}
  • a function called run() which is called by libsigrokdecode. It is defined like this:
def run(num_probes, samplerate=None)

This function is called by libsigrokdecode at the start of a session which has this PD in its pipeline. It should implement a loop calling get_sample(), and only return when an EOF is detected in that loop.

Protocol decoders

See Protocol decoders for a list of planned or already supported protocol decoders in sigrok.

Ideas

  • Plugin system for protocol decoding:
    • Should support SPI, I2C, RS232/UART and many many more protocols, see above.
    • Should be easy to add support for additional/custom protocols, e.g. AT93C46:
      • CS (chip select)
      • SK (clock)
      • DI (data from chip to outside)
      • DO (data from outside into chip)
      • Has start bit, opcodes, data following transmission
    • A protocol as simple as this should be doable to implement without code, only the description of the meaning of the various pins.
  • PD plugins must be GUI-independent: they must always work on any GUI present in sigrok. The plugin interface must therefore provide hooks for:
    • Providing data from core to plugin - stream or dump.
    • Sending analysis results back from the plugin, e.g. timestamps or sample IDs with structured results:
      • Type:
        • Protocol overhead e.g. start bits
        • Commands e.g. opcodes
        • Command parameters e.g. address following a READ command
        • Extracted data
        • String to display over the data ("opcode READ")
        • Data (e.g. 10) + length in bits (e.g. 2)
        • Results can overlap, as in e.g. "READ address 0x08" and "opcode 10"
  • This also opens up the possibility of producing e.g. a protocol analysis report from the main code.
  • All PD plugins are written in Python.
    • We embed a scripting language (Python) for very simple transforms and/or additional display; e.g. 4-bit interface to HD44780 LCD, take two nibbles in sequence, based on clock line, and assemble them into a whole byte. Scripted plugin could do this, then pass the data back to the UI for display.