Is it possible to do I2C multi-master with propeller?

Comments

• jmg Posts: 15,193

  2013-07-26 13:01 edited 2013-07-26 13:01

  ?? Do you mean multiple i2c masters in a Prop, or a single master that talks to multi slaves, or a master that can clock stretch and do wired or style CAN-like bus arbitration, so that simultaneous transmits are resolved (requires significant bus quiet times)
  With a small number of bus sharing masters, one trick that could help, is a width-modulation of start, so the Props can identify when a non-Prop (and even which other prop..) is in control of the start condition. Earliest arbitration is going to be easiest.
• KeithE Posts: 957

  2013-07-26 13:03 edited 2013-07-26 13:03

  Jazzed - do you know of anybody who has multimaster working reliably on any platform? I've heard plenty of horror stories. I know of a company which mandated multimaster for a certain cellphone interface, then backed down do to reliability problems. It would be elegant to have for sending/receiving bidirectional serial streams, and not having to do any polling. So if you know of a golden implementation where someone has really done stress testing to demonstrate the reliability it would be interesting.
  
  In case it's too tricky, can someone use a side channel for bus arbitration for your application? Or would you just choose another interface at that point?
• jazzed Posts: 11,803

  2013-07-26 13:10 edited 2013-07-26 13:10

  Propeller performing as one of N masters on the I2C bus according to the I2C specification. I2C Bus Spec pdf Page 7 "Multi-master"
  
  KiethE, the subject comes up every now and then
• JonnyMac Posts: 9,408

  2013-07-26 14:52 edited 2013-07-26 14:52

  I think Guether Daubach did multi-master I2C on the SX -- if that's the case then it should port to the Propeller without too much fuss. No?
• jmg Posts: 15,193

  2013-07-26 15:21 edited 2013-07-26 15:21

  I'm sure it is do-able, the only question would be 'At what Speed' ?
  
  Because of the Async nature of the signaling, and you want to avoid lock-outs, you would code this much like a quadrature counter.
  
  ie
  Sample for edges on both SDA and SCL and run a state engine / jump table from the 8 possible change-seen answers. (plus dummy branches on the 8 possible no-change values).
  
  Start would always be visible, and would trump all other states.
  At a guess 100KHz should be attainable, 400KHz probably less likely.
  
  One nice detail with the Prop, is you could load up 8 instances of this, and have them chatter away as a stress test, and when that works, run two Props for 16 chatter points.
• Rforbes Posts: 281

  2013-07-26 15:40 edited 2013-07-26 15:40

  jazzed- I've just starting learning the details of I2c and this post is very interesting to me. What do you see as the most difficult hurdle here? I see there is an Arbitration scheme that needs to be followed for multiple masters- is that where you see difficulties?