Cover link state (up/down, speed, width), error counters, PHY margin, packet/frame evidence and software configuration. The useful order depends on available observation points; preserve the same traffic interval and transaction context instead of treating the list as a verdict hierarchy.
Narrow physical margin is evidence to investigate, not proof that RTL, clocks or software configuration are correct. Confirm counter definitions, traffic and test duration; compare channel, receiver settings and interface timing with a known-good case. Treat each controlled experiment as evidence for or against a hypothesis.
Join debug evidence across Ethernet boundaries
Illustrative Ethernet instance of cross-boundary debugging. Counters and logs must cover the same replay and time interval before they can be correlated.
Observation point | Useful evidence | Question it can help answer |
|---|---|---|
Software / driver | Operation trace and sequence IDs | Was the expected payload queued or consumed? |
MAC / frames | Packet and FCS counters | Which observed frames passed or failed the frame check? |
PCS / coding | Block lock and mode-specific coding counters | Was synchronisation/coding healthy at this boundary? |
Physical path | Applicable lane, BER and margin measurements | Does controlled physical evidence support the channel hypothesis? |
What to notice
A clean software trace is not evidence that every intervening representation was checked.
A correlation is a lead; use a reduced experiment before declaring which component caused the failure.
Software/ownership and the observed data-delivery boundaries still matter, and the scope of MAC/PCS/physical counters must be checked. One clean counter does not locate the corruption.
A throughput drop: which measurement would narrow the cause?
Observation | Next correlated evidence | Do not conclude yet |
|---|---|---|
USB bulk IN returns NAK repeatedly | Endpoint readiness, firmware queue and host schedule | A NAK alone does not prove physical corruption. |
Ethernet MAC reports bad FCS | PCS/PHY errors plus the MAC-side interface and counter definition | FCS alone does not locate the corrupting stage. |
CSI-2 frame count falls | Packet/header/payload errors, FIFO overflow and memory service | A missing frame alone does not prove insufficient lane rate. |
Lower-rate experiment improves all observed counters | Same payload validity, revised traffic and unchanged observation window | Improvement does not isolate SI from reduced queueing pressure. |
No. That experiment changed electrical stress and offered traffic. Compare matched traffic/service conditions or vary one factor at a time to distinguish physical errors from buffering, scheduling and configuration effects.