[davidharrishmc]

Three self-checking directed tests (fetchperiph, icachebusbusy, cbofaulthit) close the last rv64gc code-coverage rows. The coverage tests also now check their own results: 17 that added no coverage are dropped, and the 18 older ones that passed by running to completion, plus gshare (kept as a branch-predictor example), now compare their results, traps and CSR values with expected values. The exceptions are six tests written against specific bugs (amoAccessFault, floatmisc, misalignedInstrSpill, misalignedStoreSpill, nonleafpbmtfault, tlbMisaligned), which lockstep still checks. It also includes #1989, which follows the exclusions to main's later RTL (the ITLBMissOrUpdateRawF rename that stopped --ccov at the exclusion step, the ITLB TLBMiss row, and an unreachable TWTimeoutM row) and adds wfiSnoTW and tlbOverlap for the rows #1941 and #1927 left uncovered; with it, rv64gc code coverage is 100%.

🤖 Generated with Claude Code

[rosethompson]

This PR removes many coverage regression tests. Some of these may be useful in other context besides coverage. For example, gshare.S is useful for demonstrating how gshare and other BP function. Should we create a separate directory for tests we wish to preserve?

[davidharrishmc]

Review notes: the tests are sound and all check themselves, but a few things need attention before merging.

• Unanswered review request. On 10-13 Rose asked to keep tests/coverage/gshare.S as a gshare example, and her review asks for a directory of tests to keep. There's no reply yet and gshare.S is still deleted. Her approval (10-05, 0ab3b74) predates the force-push that dropped 18 tests. btbthrash.S and ebuIcacheDTLBMiss.S gave lockstep stress value and could move with gshare.S.
• The 100% claim can't be checked on main right now. coverage-exclusions-rv64gc.do (main :488,490) looks up ~ITLBMissF & ~CacheableF & ~SelIROM and ~ITLBMissF & CacheableF & ~SelIROM, which ifu: don't fetch while an ITLB A-update walk is pending #1934 renamed to ~ITLBMissOrUpdateRawF & … (ifu.sv:289-290), so regression-wally --ccov exits at the exclusion step. Fix the patterns here or on main.
• ITLB TLBMiss row. The -fecexprrow $line 3 exclusion on ~CAMHit & TLBAccess (main :439-440) predates mmu: page fault on non-canonical VAs without walking the page table #1928, which made the line ~CAMHit & TLBAccess & Translate & ~UpperBitsUnequal (tlbcontrol.sv:129). The unreachable row is probably 5 now, so a reachable row is excluded. A Questa run would confirm.
• The minimal-cover prune ran on RTL from before 10-09. Since then main added the tlbcamline Overlap logic (tlb: a TLB write invalidates the entries it overlaps #1927), the non-canonical path (mmu: page fault on non-canonical VAs without walking the page table #1928) and ITLB update gating (ifu: don't fetch while an ITLB A-update walk is pending #1934), so dropping tlbASID/GLB/GP/MP/TP and ifuCamlineWrite needs a ccov run on the merged result.
• The body still says it is stacked on Close the remaining rv64gc coverage holes: exclusions with reasons, tests for the September 12-17 RTL, and uniform misaligned-atomic access faults #1900, which has merged.

🤖 Generated with Claude Code

[davidharrishmc]

@rosethompson I kept gshare.S in tests/coverage instead of making a separate directory. Tests there are built by make and run in every nightly and --ccov regression, and now they check themselves too. A directory of tests that nothing runs would drift out of step with the RTL and the harness. Tell me which other dropped tests you want back and I'll restore them the same way. None of the other 17 was written against a specific bug. They remain in git history at 644ebb4^.

Also updated:

• Merged main. The Files view now shows only this PR's changes, and the description no longer says it is stacked on Close the remaining rv64gc coverage holes: exclusions with reasons, tests for the September 12-17 RTL, and uniform misaligned-atomic access faults #1900.
• regression-wally --ccov on main currently stops at the exclusion step, because coverage-exclusions-rv64gc.do still looks up ~ITLBMissF & ... in ifu.sv, which ifu: don't fetch while an ITLB A-update walk is pending #1934 renamed. That is being fixed in a separate PR. The 100% coverage result needs rechecking after it lands.

🤖 Generated with Claude Code