ExceptionsZalrsc test generator with 92% coverage

[Teemooooooooo]

ExceptionsZalrsc python test generator at 92% coverage

• Note that illegal sc.w does not get coverage as SAIL stores content in by bytes instead of giving exceptions

• Sail issue: Misaligned sc.w does not trigger store-address-misaligned exception sail-riscv#1574

• Note that QEMU also has a bug where sc instructions at illegal address do not trigger access fault exception

• QEMU issue: https://gitlab.com/qemu-project/qemu/-/issues/3323
  _generate_store_address_misaligned_tests is not hitting the following coverpoints due to SAIL bug

• Cross cp_store_address_misaligned_illegal_w

• Cross cp_store_address_misaligned_illegal_d

[davidharrishmc]

Many little things. Going a good direction!

[davidharrishmc]

Sure. Put a add_test case just before each.

…

On Thu, Mar 5, 2026 at 8:52 PM Jinghe Yu ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In generators/testgen/src/testgen/priv/extensions/ExceptionsZalrsc.py <#1019 (comment)> : > + """Generate instruction address misaligned and access fault exception tests.""" + covergroup, coverpoint = "ExceptionsZalrsc_cg", "cp_store_misaligned_priority" + addr_reg, data_reg, rd_reg = test_data.int_regs.get_registers(3, exclude_regs=[0]) + + lines = [comment_banner(coverpoint)] + + lines.extend( + [ + f" LA(x{addr_reg}, RVMODEL_ACCESS_FAULT_ADDRESS)", + f" addi x{addr_reg}, x{addr_reg}, 1", + f" LI(x{data_reg}, 0xDECAFCAB)", # Match original value + f" LI(x{rd_reg}, 0xDECAFCAB)", + test_data.add_testcase("sc.w_off1_priority", coverpoint, covergroup), + f" lr.w x{rd_reg}, (x{addr_reg})", + " nop", + write_sigupd(rd_reg, test_data), Should I just sigupd for both .w and lr.w? — Reply to this email directly, view it on GitHub <#1019 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AR4AA37F76WW74M6BZYBKED4PJKPNAVCNFSM6AAAAACWDU3DD2VHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTSMBRGI3DSMZRHE> . You are receiving this because you commented.Message ID: ***@***.***>

[davidharrishmc]

Good to go after minor spacing fixes and merging into the latest version.

[davidharrishmc]

Could you add a comment in the coverage file by the coverpoints that aren't hit indicating why and pointing to the Sail bug, so we don't forget this and have to debug again later?

[jordancarlin]

One last comment and then this looks good to go.

[jordancarlin]

There are limits on how many instruction can be between lr.w and sc.w. To avoid conflating this test with that limit, let's move the LI(x{rd_reg}, 0xBAD) before the lr.w. So initialize all of the registers, then do the lr.w and then do the sc.w (with a nop in between). Let's also move the write_sigupd for the temp_reg after the sc.w since it should not change in between helps avoid running extra instructions between them. This applies to all sc.w/sc.d tests.

[Teemooooooooo]

One of the reason that I put the LI(x{rd_reg}, 0xBAD) before the sc.w is because one of the coverpoint is rd_gt_one_prev. If I move it, it would just be nop before the sc.w, should I still move it?

[davidharrishmc]

Good point. We’d best keep that LI before the sc. We may have to drop the write_sigupd after the lr to prevent too much code between the two.

…

On Mar 10, 2026, at 8:21 AM, Jinghe Yu ***@***.***> wrote: @Teemooooooooo commented on this pull request. In generators/testgen/src/testgen/priv/extensions/ExceptionsZalrsc.py <#1019 (comment)>: > + f" LI(x{data_reg}, 0xDECAFCAB)", + f" LI(x{temp_reg}, 0xBAD)", + test_data.add_testcase(f"lr.w_off{offset}", coverpoint, covergroup), + f" lr.w x{temp_reg}, (x{addr_reg})", # establish reservation + " nop", + write_sigupd(temp_reg, test_data), + f" LI(x{rd_reg}, 0xBAD)", # previous rd greater than 1 + test_data.add_testcase(f"sc.w_off{offset}", coverpoint, covergroup), + f" sc.w x{rd_reg}, x{data_reg}, (x{addr_reg})", One of the reason that I put the LI(x{rd_reg}, 0xBAD) before the sc.w is because one of the coverpoint is rd_gt_one_prev. If I move it, it would just be nop before the sc.w, should I still move it? — Reply to this email directly, view it on GitHub <#1019 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AR4AA37TLIKQU43Q54TLDLD4QAXHZAVCNFSM6AAAAACWDU3DD2VHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTSMRTGI4TGMZRHA>. You are receiving this because you commented.