frontend: Report the documented stall polarity in the inst64 events

[DanielKellerM]

The Snitch cluster defines these counters explicitly, in snitch_cluster_peripheral_reg.rdl:

dma_r_stall     = 12  "Incremented whenever r_ready = 1 but r_valid = 0"
dma_w_stall     = 13  "Incremented whenever w_valid = 1 but w_ready = 0"
dma_buf_w_stall = 14  "Incremented whenever w_ready = 1 but w_valid = 0"
dma_buf_r_stall = 15  "Incremented whenever r_valid = 1 but r_ready = 0"

snitch_cluster_peripheral.sv maps each metric to the identically named dma_events_t field with no remapping, so that text is a direct contract on the field.

The two lines this removes are, verbatim, the definitions of 14 and 15 sitting in the fields for 12 and 13, where they overrode the correct assignments made earlier in the same always_comb. Removing them restores 12 and 13:

field	after this PR	RDL
r_stall	r_ready && !r_valid	whenever r_ready = 1 but r_valid = 0
w_stall	!w_ready && w_valid	whenever w_valid = 1 but w_ready = 0

The asymmetry that remains is the intended one: iDMA drives w_valid but r_ready, so waiting on the fabric is valid-without-ready on AW, AR and W, and ready-without-valid on R.

Why it looks like this today

Both pairs arrived in one commit (097f6140, PR #40) that flattened axi_dma_perf_counters.sv's if-guarded set-only writes over a '0 default into unconditional assignments. Under the guards the two conditions fed four separate counters; unconditional, the later pair simply clobbers the earlier one. Accidental, not a design decision.

Testbench

tb_idma_inst64_axi_copy asserted the old behaviour, so its two goldens move with the RTL in the same commit. Verified both ways with Questa:

• with this change: TEST PASSED, rc=0
• with either side reverted: Fatal: events.EvRStall disagrees with the AXI pins, rc=2

Worth noting that no CI job runs that testbench today: it is reachable only through make idma_sim_tb_idma_inst64_axi_copy. Without running it by hand, devel would have stayed green while shipping RTL that contradicts its own assertions.

Impact

This is a silent change to reported performance counters. Metric 13 currently counts mostly bus idle, since a registered W slice parks WREADY high; metric 12 currently reads near zero because inst64 ties rsp_ready high. Existing profiles change meaning with no version marker, so it wants a CHANGELOG line alongside #201.

No programmatic reader of these counters exists in snitch_cluster outside the RDL and its generated packages, so anyone reading them reads them against the RDL, which is to say against the corrected values.

Not done here

Renaming the two lines to buf_w_stall / buf_r_stall rather than deleting them would fix 12 and 13 and light up 14 and 15, which read constant zero before and after this PR. The cost is integrators whose dma_events_t is narrower than the cluster's, including this repo's own testbench mirror in test/frontend/idma_inst64_tb_pkg.sv, which omits both fields. Left for a follow-up rather than folded in here.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.