C18.1 Forbidden Combinations

MediumFixVerilator coverage build · serverReview the topic

Fix a lane-by-command cross that counts link-training traffic as if the link were up.

Hardware context: A 4-lane link with read-only and config-only lanes and a training phase. Count lane-by-command products only on a live link, without impossible products.

This is a Fix problem: the covergroup in the editor has one realistic defect. Find it and fix it.

A 4-lane link carries nop, read, write and config commands. Config uses lanes 0 and 1 only, and lane 3 is read-only: a write there must never happen. Before the link comes up, training sends every command on every lane.

The plan counts lanes and commands at all times, but the lane-by-command cross only for traffic on a live link. The report shows cross products covered by tests that never brought the link up.

Signals in the Timeline

SignalWidthRole
samples8input of each check
Full specification
ItemCounts
ln, cmevery sample, training included
ln_x_cmsamples with link_up = 1 only; config on lanes 2-3 ignored; a write on lane 3 illegal

Each case samples a fresh instance and reads get_inst_coverage() times 100, rounded. The cases never send a write on lane 3 with the link up, which would stop the run on the coverage build. The full check samples more lists than the examples show.

Examples

Example 1

Training Only
Input
transfers (lane, cmd, link_up): (0, rd, 0) (1, wr, 0)
Output
coverage_percent_x100 = 3333

ln and cm each see two values (2 of 4); the cross sees nothing: (50 + 50 + 0) / 3 = 33.33%.

Example 2

The Same with the Link Up
Input
transfers (lane, cmd, link_up): (0, rd, 1) (1, wr, 1)
Output
coverage_percent_x100 = 3846

Now the cross counts both products: 2 of 13.

Example 3

Training, Then Live Traffic
Input
transfers (lane, cmd, link_up): (2, nop, 0) (2, nop, 1) (0, cfg, 1)
Output
coverage_percent_x100 = 3846

The training nop on lane 2 counts in the coverpoints; the live nop and config each add a cross product.

Constraints and Contract

  • Keep the covergroup, coverpoint, cross and bin names and the exclusions.
  • The coverpoints must keep counting during training.

Hints

Hint 1

Which samples should the cross skip?

Hint 2

A guard on a coverpoint also hides the sample from that coverpoint.

Hint 3

A cross can have its own iff.

What the grader checks: Your covergroup is sampled with 18 lists of transfers. Its instance coverage is compared with the value IEEE 1800-2023 19.6 and 19.11 give. Three lists are shown.

18 checks on Verilator coverage build · server. Run the examples first, then check the full regression. Full checks use the server judge. Downloads contain public samples only. A passing result establishes the tested cases, not a proof for all possible inputs.

See all 2 steps and the final capstone

Report a problem with this page

Your code SystemVerilogSaved locally
Ln 1, Col 1Tab indents · Ctrl/⌘ + Enter runs samples
Not run yet

Training Only

Input
transfers (lane, cmd, link_up): (0, rd, 0) (1, wr, 0)
Expected output
coverage_percent_x100 = 3333

ln and cm each see two values (2 of 4); the cross sees nothing: (50 + 50 + 0) / 3 = 33.33%.

Run samples executes all 3 public checks. Check all tests runs 18 checks on the server.

Verilator coverage build · server