Skip to content

mepc low bits are not masked and mtvec.MODE=3 is retained on write #448

Closed

Description

@Risid

Environment Setup

Stock XLEN=32 configuration, F enabled, C disabled, IALIGN=32. Tested at commit b447578d9f3bb116d658c614cf3407153f0536da. Driver: rtlsim built DEBUG=1; tests are bare-metal ELFs.

  • f3_warl/21_mepc_warl/21_mepc_warl.S writes mepc with values whose low bits are non-zero and reads it back.
  • f3_warl/22_mtvec_warl/22_mtvec_warl.S writes mtvec with MODE=3 and reads it back.
# rtlsim built DEBUG=1 from the commit above (XLEN=32, F on, C off).
P=f3_warl/21_mepc_warl # or f3_warl/22_mtvec_warl
spike -l --log-commits --isa=rv32imaf -m128 $P/ref.elf
./rtlsim $P/dut.elf

Both tests exited 0 on both engines.

Expected behavior

Per the RISC-V Privileged Specification, machine.adoc, Machine Exception Program Counter (mepc) register:

On implementations that support only IALIGN=32, the two low bits (mepc[1:0]) are always zero.

Observed results

mepc test (21_mepc_warl). It first writes the aligned value 0x80001234 as a control (both engines read back 0x80001234), then writes 0x80001233 and reads it back:

write Vortex readback Spike readback
0x80001234 (control, aligned) 0x80001234 0x80001234
0x80001233 0x80001233 (bits [1:0] retained) 0x80001230 (bits [1:0] masked)

DUT (logs/21_mepc_warl_dut.dbg.log, commit of csrr t4,mepc at PC 0x8000001c):

361: ...-commit commit: ... PC=0x8000001c, ex=SFU, rd=29, ... data={0x80001233, ...} <--

Spike (logs/<test>_ref.commits, same <test> as above):

core 0: 3 0x80000008 (0x34129373) x6 0x00000000 c833_mepc 0x80001234
core 0: 3 0x8000000c (0x341023f3) x7 0x80001234
core 0: 3 0x80000018 (0x34129e73) x28 0x80001234 c833_mepc 0x80001230
core 0: 3 0x8000001c (0x34102ef3) x29 0x80001230 <--

mtvec test (22_mtvec_warl). Its writes use MODE=3 (reserved; only 0 and 1 are defined). Spike legalises the BASE alignment and keeps MODE=1; Vortex retains the written value including the low MODE bits:

write Vortex readback Spike readback
0x80000100 (MODE=0) 0x80000100 0x80000100
0x80000101 (MODE=1) 0x80000101 0x80000101
0x80000103 (MODE=3) 0x80000103 0x80000101

DUT (logs/22_mtvec_warl_dut.dbg.log, commits of the three csrr ...,mtvec reads):

289: ... PC=0x8000000c ... data={0x80000100, ...}
361: ... PC=0x8000001c ... data={0x80000101, ...}
433: ... PC=0x8000002c ... data={0x80000103, ...} <--

Spike (logs/<test>_ref.commits, same <test> as above):

core 0: 3 0x80000028 (0x30529f73) x30 0x80000101 c773_mtvec 0x80000101
core 0: 3 0x8000002c (0x30502ff3) x31 0x80000101 <--

The mtvec difference is a WARL question: Spike picks MODE=1 as one legal result, and the spec only requires MODE to be a legal value.

The trap CSRs appear to be stored as plain latches, with no WARL legalisation on the write path.

Affected revision and environment

  • Tested at commit b447578d9f3bb116d658c614cf3407153f0536da
  • VERSION: VORTEX_VERSION=3.0, TOOLCHAIN_REV=v3.0.1
  • DUT simulator: rtlsim DEBUG=1
  • Reference: Spike --isa=rv32imaf -m128
  • OS: Linux 6.8.0-90-generic x86_64.

Minimal RTL reproducer

The log excerpts are from logs/21_mepc_warl_dut.dbg.log and logs/22_mtvec_warl_dut.dbg.log in the attachment.

Attachment: testcase.zip contains both .S sources, their prelinked dut.elf / ref.elf, link.ld, probe_macros.h, and the full DUT and Spike logs for both tests (logs/21_mepc_warl_*, logs/22_mtvec_warl_*).

The mepc masking is the less ambiguous of the two; the mtvec MODE legalisation may be a spec reading Vortex intends differently.

testcase.zip

Activity

  1. tinebp commented on Nov 19, 2026

    @tinebp
    Collaborator

    Fixed by #487.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions