DPI UVM Config Bridge: External Knobs, Deterministic Startup

Core Contract

Keep this startup boundary deterministic, observable, and fail-fast.

Treat startup as reusable infrastructure: ownership, ordering, and liveness checks must be explicit.

[UVM][ENV][DPI] startup path

external config -> DPI call
-> sanitize/default
-> config_db publish
-> run_test

UVM classes consume normalized values only

[UVM][ENV] boundary model

boundary A: ownership and lifecycle
boundary B: publication and path scope
boundary C: first-activity liveness proof

fix the first broken boundary first

• Document ownership and ordering assumptions in code, not only slides.

• Make startup diagnostics actionable and low-noise.

• Prefer deterministic startup defaults over convenience shortcuts.

Reference Implementation

import "DPI-C" context function int dpi_get_timeout_ns();
import "DPI-C" context function int dpi_get_enable_trace();

initial begin
  int timeout_ns;
  bit enable_trace;

  timeout_ns = dpi_get_timeout_ns();
  if (timeout_ns <= 0) timeout_ns = 2_000_000;

  enable_trace = (dpi_get_enable_trace() != 0);

  uvm_config_db#(int)::set(null, "uvm_test_top", "timeout_ns", timeout_ns);
  uvm_config_db#(bit)::set(null, "uvm_test_top.env", "enable_trace", enable_trace);
  run_test();
end

[ENV] implementation audit

- critical startup assumptions validated?
- missing-state paths fail immediately?
- diagnostics include component path and context?
- minimal smoke seed can exercise this boundary?

[UVM][ENV] startup observability

track timestamps/counters for:
  reset release
  first sequence launch
  first monitor publish
  first checker compare

these markers reduce triage latency dramatically

• Keep startup code simple enough to audit in code review.

• Use fatal failures for non-recoverable startup contract violations.

• Expose startup metrics in report_phase for CI artifacts.

Debug Workflow

Triage order

1. Reproduce with one seed and startup tracing enabled.

2. Verify startup boundary counters and first-activity markers.

3. Classify failure by boundary before deep waveform analysis.

4. Fix root boundary and rerun identical seed to confirm.

simv +UVM_TESTNAME=smoke_test        +UVM_VERBOSITY=UVM_LOW        +UVM_CONFIG_DB_TRACE        +UVM_OBJECTION_TRACE

[UVM][ENV] quick diagnosis

fails before build:
  run_test/factory/config publish issue

build passes, no traffic:
  reset/sequence launch gating issue

monitor active, checker idle:
  connect/subscriber pipeline issue

Key takeaways

• Use DPI as ingestion boundary, not class-logic dependency.

• Sanitize and log all imported startup values.

• Deterministic defaults are required for reproducible regressions.

• A deterministic startup workflow improves regression trust and team velocity.

Common pitfalls

• Allowing nondeterministic external state into CI runs.

• Silent fallback behavior with missing diagnostics.

• Passing raw DPI values directly into deep components.

• Skipping startup smoke and finding setup regressions only in long runs.

Applied Patterns

Use this reusable grid before landing startup-related changes.

[UVM][ENV] startup pattern grid

contract:
  ownership + ordering + liveness documented

implementation:
  fail-fast checks and deterministic defaults

observability:
  milestone counters and concise logs

verification:
  smoke seed + boundary-first triage + closure checklist

function void startup_health_report();
  `uvm_info("STARTUP_SUMMARY",
    $sformatf("rst=%0t seq=%0t mon=%0t cmp=%0t",
      first_reset_release_time,
      first_sequence_start_time,
      first_monitor_publish_time,
      first_scoreboard_compare_time),
    UVM_LOW)
endfunction

• Treat startup updates as infrastructure API changes.

• Keep code and checklist language aligned.

• Require stable startup summaries in regression logs.