verdi-bus-debug
/verdi-bus-debug
Diagnozuje zawieszone magistrale i transakcje w Verdi, sprawdzając sygnały, uzgadnianie i reguły protokołu.
- Autor
- fayememe ↗
- Repozytorium
- verdimcp ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 29.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 29.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Diagnozuje zawieszone magistrale i transakcje w Verdi, sprawdzając sygnały, uzgadnianie i reguły protokołu.
Oryginalny opis (EN)
Use when a bus/handshake is stuck or a transaction never arrives. Drives the protocol-agnostic primitives (signal map, handshake beats, change stats, window scan, declarative checks) and applies protocol rules from this skill's grounding text — the MCP server contains no protocol semantics.
Jak działa
- Scope
- Decision flow
- Tool notes
- Grounding text (protocol knowledge lives here, not in the server)
- Must-not
- Reporting
Podgląd SKILL.md
---
name: verdi-bus-debug
description: Use when a bus/handshake is stuck or a transaction never arrives. Drives the protocol-agnostic primitives (signal map, handshake beats, change stats, window scan, declarative checks) and applies protocol rules from this skill's grounding text — the MCP server contains no protocol semantics.
---
# Bus / handshake debug
## Scope
**In:** "the write never landed", "valid is high but ready is not", "the response came back on the wrong ID",
"the bus never arrived". A handshake that is *blocked* or *misordered*.
**Out:** X/Z on the bus itself (→ `verdi-xz-debug`), performance/outstanding analysis (not implemented),
and any protocol decoder (there is none on purpose; see the must-not list).
## Decision flow
1. **Identify the interface and protocol** from the system docs or the user's description. Write down
which channel(s) matter before touching a tool.
2. **Build a signal map** — `verdi_signal_map_put {name, signals:{role: path}}`, roles are free-form
(`clock`, `valid`, `ready`, `data`, `id`, `last`, `strobe`). Reuse existing ones with
`verdi_signal_map_list` / `verdi_signal_map_get`. Details: `verdi-signal-map`.
3. **Measure before reading raw data**
- `verdi_change_stats` on the channel signals → volume, decide GUI vs NPI. Read `fsdb.is_completed`.
- `verdi_handshake_beats {channel, clock, valid, ready, data?, id?, last?, start, end}` → every cycle
where `valid && ready`, with the payload. This is the transaction list you actually reason about.
4. **Find the first violation** with `verdi_window_scan`, e.g. `valid && !ready`, optionally
`min_duration`. Report `truncated`, the `step`, and that it is sampled.
5. **Apply the protocol rules** below, one rule at a time. Each rule either holds, fails at a stated time,
or is not checkable with the available signals — say which.
6. **Freeze a deterministic check** once the rule proves useful:
`verdi_check_define {name, predicate, params}` → `verdi_check_run {check_id}`.
7. **Produce evidence** with `verdi-evidence` (annotated capture + handshake beats + check result).
## Tool notes
- `handshake_beats` needs a `clock` role for per-cycle beats; without one it samples at value-change times
and you must not call them "cycles".
- `engine.used` tells you whether the transition list can have skipped glitches (`gui` can, `npi` cannot).
- `change_stats` never returns the raw change stream. Do not try to reconstruct it from other calls.
- Never return a million transitions to the agent; aggregate first, then narrow.
- A blocked handshake proves a **stall**, not a **cause**. The cause is one hop upstream: go to
`verdi-design-trace` or capture the cone with `verdi-cone-evidence`.
## Grounding text (protocol knowledge lives here, not in the server)
### AXI4 / AXI4-Lite
- `AWVALID`/`AWREADY`, `WVALID`/`WREADY`, `BVALID`/`BREADY`, `ARVALID`/`ARREADY`, `RVALID`/`RREADY` are
independent valid/ready pairs.
- Once `VALID` asserts it must stay asserted until the matching `READY` (no withdrawal).
- A beat transfers only when `VALID && READY` in the same cycle.
- `WLAST` marks the final write beat; `RLAST` the final read beat.
- `B` must not precede the last `W` beat; response ordering follows the protocol.
- AXI4-Lite has no bursts: no `AWLEN/AWSIZE/AWBURST`, no `WLAST/RLAST`.
- `AWID/BID/ARID/RID` pair responses to requests — check ID consistency.
### AXI4-Stream
- `TVALID`/`TREADY`; `TLAST` marks a packet boundary.
- `TVALID` must not be withdrawn before `TREADY`.
- `TKEEP`/`TSTRB` define valid bytes; check them against the `TDATA` width.
### APB
- `PSEL && PENABLE` marks the access phase; `PREADY` completes it.
- `PWRITE`, `PADDR`, `PWDATA`, `PRDATA` are sampled during the access phase.
- `PSLVERR` reports an error response — check it is captured.
- APB has no burst/ID; do not apply AXI rules.
### AHB / AHB-Lite
- `HTRANS` (`NONSEQ`/`SEQ`/`IDLE`/`BUSY`), `HREADY`, `HRESP` define address/data phases.
- A transfer completes when `HREADY` is high; wait states insert `HREADY=0`.
- `HRESP` error responses take two cycles — check pipelining.
- `HBURST`/`HSIZE` define burst behaviour; AHB-Lite has a single master.
### Wishbone
- `STB`/`ACK` (optionally `CYC`) form the handshake; `WE`, `ADR`, `DAT` carry the operation.
- `ACK` must not assert without `STB`; check `CYC` framing.
- Classic vs pipelined differ in when `STB` may drop — state which variant you assume.
> These rules are prompt knowledge, not MCP code. If a rule is not in the system docs, label the
> conclusion `likely` / `uncertain` and cite where it came from.
## Must-not
- Do not add protocol strings to the MCP server, and do not ask for a protocol decoder tool.
- Do not apply one protocol's rules to another because the signal names look similar.
- Do not claim a protocol violation from sampled data alone.
## Reporting
```
channel: wr (clock clk, valid wr_valid, ready wr_ready, data wr_data)
beats: 12 in 0..240 ns (NPI, fsdb.completed=true, not truncated)
first bad: 115.43 ns wr_valid=1, wr_ready=0 (VERIFIED sampled, step 0.48 ns)
stall len: >= 124 ns (to end of window) (VERIFIED from transitions: wr_ready never rose)
upstream: fifo_full held high since 84 ns (VERIFIED via transitions_get)
rule: AXI4 once-VALID-asserted — not violated; this is backpressure, not a protocol error (INFERRED)
check: wr_stall (7dcf91b9) frozen for re-run
```
Wersje
W bazie nie ma innych skilli o nazwie verdi-bus-debug.
Migawka bazy z 13.11.2026 · wersja paczki be8054860387 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100verdi-signal-map
Buduje i sprawdza profile ról protokołu wobec ścieżek sygnałów, np. valid, ready, data, id i last. - 100verdi
Kieruje sprawy z Synopsys Verdi, nWave i FSDB do właściwych narzędzi i przypomina zasady bezpiecznej pracy. - 100protocol-version-handshake
Negocjuje jawnie wersję protokołu i przypina tę, którą serwer faktycznie obsługuje, gdy klient może działać na innej wersji. - 90esp32-open-drain-bus-planning
Planuje magistrale open-drain na ESP32, w tym dobór rezystorów podciągających i długość linii, żeby sygnały były czyste. - 81bus-debugging
Diagnozuje błędy magistral SPI, UART, CAN i I2C na podstawie zapisów transakcji i ograniczeń elektrycznych. - 90esp32-can-bus-off-recovery-policy
Projektuje politykę odzyskiwania po bus-off w CAN dla ESP32, z ograniczeniami układu, analizą awarii i weryfikacją.