intent/property-strap

Remedy ¶

Move the strap resistor to the rail that latches the declared level. The board boots either way, configured as something other than what was intended.

What it checks ¶

The design intent can declare a net as a boot/config strap. That net must not be biased to the opposite level from the one it is meant to latch. Declared high with a pull-down, or declared low with a pull-up, means the part will come up configured the other way from what the design intended.

It is not property-reset-polarity under another name ¶

The two read the same evidence and ask inverted questions. If a reviewer assumes they are the same check, the reviewer will misread both.

reset-polarity’s value is the level that asserts reset. Bias toward it is the defect, bias away from it is correct, so the rule can only ever catch one of the two ways a reset line goes wrong.

strap’s value is the level the pin should latch. Bias toward it is correct and bias away from it is the defect, in either direction. So this rule catches both mistakes, not one.

Read this before binding a review item to it ¶

A pass means “no contradiction found”, not “strap confirmed”. property-reset-polarity carried the same caveat until it began reporting an unbiased line as inconclusive.

Strap pins almost always carry an internal pull, and the standard datasheet instruction is to fit an external resistor only for the non-default state. So a design may declare the default level and fit no resistor on that net. That design is correct, and the case is extremely common. Firing on absent bias would report a non-defect on the majority of real straps, so the rule says nothing there.

It checks resistance only against a band you declare. The rule now asks whether a strap resistor has an appropriate value, strong enough to hold against leakage and weak enough not to fight an active driver. It asks only where the declaration states min_ohms / max_ohms. There is no built-in band and there deliberately will not be one. A CMOS input with nanoamp leakage is happy past 100k. An active driver has to override some straps, and those want a few hundred ohms. So any universal range the engine invented would flag correct boards. The person declaring the strap holds the datasheet, so the band is theirs to state.

Declare neither bound and the value half simply does not run, and the direction half still does.

It does not identify strap pins. The declaration names the nets. Strap nets are usually not named for their function. On such a design the pins have to be read off the part’s datasheet pinout and written into the declaration by hand. Nothing here can infer them.

For hardware engineers ¶

A strap is a resistor that parks a pin at a known level so the part can sample it at reset and pick a configuration such as boot source, bus width, clock ratio, or its own address on a shared bus.

Getting one backwards does not look like a wiring fault. The board powers up, the part runs, and it runs in the wrong mode. It boots from the wrong flash. Or it answers on an address that another device on the same bus already uses, and then two parts drive that bus against each other. The whole segment is then unreliable in a way that looks intermittent rather than wired wrong.

What counts as bias ¶

A resistor on the declared net counts as bias when its other end reaches a power rail (pull-up, latches high) or a ground net (pull-down, latches low). The far end may reach its rail through further passives.

A net with both, a divider, reports neither. Some parts use a divider for a tri-level strap, and the level it selects depends on the ratio of the two resistances rather than on either one, so the rule declines to guess rather than reporting a direction it does not know. A declared band over a divider reports inconclusive and names both resistors, since neither is “the” strap pull, so judging one would name an arbitrary part.

Declaring it ¶

intent:
  nets:
    # direction only
    BOOT_MODE0: {strap: high}
    # direction plus an acceptable pull band, in ohms
    PHYAD1: {strap: low, min_ohms: 1000, max_ohms: 100000}
    # one-sided: bound the weak end only
    BOOT_CFG2: {strap: high, max_ohms: 47000}

The strap level is the intended latched level, and it is required. Without it the rule has nothing to contradict. So an omitted or misspelled level is rejected at load, rather than becoming a rule that silently never fires.

min_ohms and max_ohms are optional and independent, so a one-sided band checks one side only. Both are rejected at load on a property kind that has no resistance to bound. A min_ohms above its max_ohms is rejected too. No value can satisfy that band, so it would compile to a check that can never pass. That is a declaration error rather than a design finding.

A declared band over a resistor with no value, a value that does not parse, or a value in another unit reports inconclusive and names the resistor. Firing on a number the engine could not read would report a defect on no evidence, and passing would claim a comparison nobody made, so the finding says the band was not checked. A wrong-way strap still fails, whatever its value reads as. With no strap resistor on the net, a declared band has nothing to bound and the rule says nothing, for the default-state reason above. An unfitted resistor counts as absent.

Fixing a finding ¶

An inconclusive finding asks for a readable value on the named resistor, or a look at the divider by hand. Otherwise, either the strap resistor goes to the wrong net, or the declared configuration is wrong. Check the part’s datasheet strap table before moving the resistor. If the declaration is wrong, moving the resistor creates the bug that the rule was reporting.