Passing reset tests prove the scenarios that ran. Coverage proves the scenarios ran at all. Reset coverage asks, for every reset and every interesting design state, whether a reset landed there during the regression. An empty answer is a hole in the signoff, not a pass.
Question | What a hole means |
|---|---|
Did each reset assert at time zero? | The baseline release sequence for that reset never ran. |
Did each reset assert while its domain was busy? | Mid-operation recovery is untested for that domain. |
Did each reset assert while its neighbour kept running? | Partial-reset behaviour is untested. |
Did each sequenced domain release in every allowed order? | Only the easy order is proven; the risky orders are assumed. |
Did back-to-back resets occur? | Recovery-during-recovery is untested. |
Did reset coincide with a stalled handshake or a full buffer? | Reset-under-backpressure is untested. |
Did every waiver-relevant condition occur? | A waived crossing claims a condition, such as a settled input, that no test ever produced. |
Each question becomes a small coverable event: the reset name, the design state, and their coincidence. A simple approach that works without special tooling is a checklist table filled from the test plan and confirmed against logs: for each reset and each scenario, the seed and the test that covers it. Whatever the mechanism, the review reads the holes, not the hits: one uncovered row re-opens the dynamic-checks item for that reset.
The review maps every reset in the intent to at least the time-zero and mid-operation rows, and every sequenced domain to the ordering rows. Coverage collected only on the default configuration, or only with all resets moving together, is rejected: partial resets and staggered releases must appear in the evidence.
No. One scenario ran for many cycles. Coverage asks which scenarios ran, and mid-operation, partial, ordered, and back-to-back resets are all missing.
A row showing reset released while that input held its settled value, in every configuration the waiver covers. Without it the waiver argument is untested.