i2c-redundant-pull-up

Remedy ¶

Remove one of the pull-ups, or depopulate it. Which one is a board decision: keep the pull-up sized for the bus capacitance and clock rate this design actually runs at, and take out the one that came along with a module.

What it means ¶

An I2C net reaches one rail through more than one resistor. The bus is pulled up twice.

The check counts DISTINCT resistors that take the net to a rail, within three series crossings. Two routes through one resistor are one pull-up; two resistors are two, however they are drawn.

Why engineers want it ¶

Resistors in parallel are one smaller resistor. Two 2.2k pull-ups are an effective 1.1k, so the bus sinks roughly twice the current it was sized for. An open-drain device can only pull the line low by sinking that current. Every device has a maximum current it can sink while it still holds its output below VOL. So a bus pulled too hard works with some parts and not others, or its low level sits high enough that a receiver reads it as neither state.

It arrives the same way almost every time. A module carries its own termination so it works standalone, the board it plugs into already pulls the same bus, and nothing on either schematic is wrong on its own.

Impact ¶

The bus is over-pulled by whatever factor the parallel combination gives. It usually still works, on the bench, with the parts that happen to be fitted, so it is a warning worth reading rather than a fault worth stopping for.

Fires versus fine ¶

  fires                                 fine

  +3V3      +3V3                        +3V3
   |         |                           |
  R1 2k2    R2 2k2                      R1 2k2
   |         |                           |
   +----+----+                           |
        |                                |
       SDA                              SDA

Scope note ¶

• Only nets matching the I2C naming convention at a token boundary (SDA / SCL), the same predicate i2c-pull-up uses. A bus named some other way is out of scope rather than passing.
• A net with NO pull-up is not this rule’s subject. i2c-pull-up reports it as an error, and reporting the same absence in three places would treble one defect.
• Pull-ups to DIFFERENT rails are i2c-pull-up-split-rail. The two rules fire on disjoint conditions, so a bus is named once, by the rule whose remedy applies to it.
• Ground is never crossed and a resistor to ground is not counted, because that is a pull-down. Counting it would report a bus held by one pull-up and one pull-down as over-pulled.
• Severity is warning, not error, because a jumper-selectable or DNP termination is legal by design and the netlist cannot see the fit status.

Query structure ¶

Enumerate I2C nets; for each, walk out through resistors up to three crossings, collecting every distinct resistor whose far side is a rail; fire when there is more than one and they all end on the same rail.

A reachability question cannot express this. net.reaches reports a rail as reached however many resistors reach it, so a doubled bus and a correct one are the same answer.

For software readers ¶

Picture a resource with two owners, where each believes it is the only one. Counting the handles finds the second owner, and asking whether the resource is held does not. A defer close() in both the caller and the callee is harmless in one sense, because the thing does get closed. It is wrong in another, because the accounting is off. A check that asks only whether it was closed cannot see it.