CHERIoT ASR-faulting CSR read counted as retired on WritebackStage=0

未关闭
  • 哪里不对？
#2,569 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 2 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
46/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
领域
backend

调研方向

Start with the reported paths in ibex_wb_stage.sv, ibex_id_stage.sv, and ibex_controller.sv, then compare how CHERIoT faults affect instr_perf_count_id_o and perf_instr_ret_wb_o when WritebackStage=0. Reproduce with the attached csr_asr_minstret.S testcase and compare the mscratch retired count across both configurations. Done means the faulting CSR read does not retire in wb0, matching the expected count of 43.

由索引模型根据 Issue 内容生成。

描述

Type:Bug

Observed Behavior

In CHERIoT mode with WritebackStage=0, csrr mscratch raises a Permit_Access_System_Registers exception and is included in the retired-instruction count. The other test reads cycle. Only the CSR read differs.

build CSR read retired trap? record (mcause, mtval, rd)
tag_zeros (WritebackStage=1) cycle 37 no 0x0, 0x0, 0x21
tag_zeros (WritebackStage=1) mscratch 43 yes 0x1c, 0x418, 0x0
wb0 (WritebackStage=0) cycle 37 no 0x0, 0x0, 0x24
wb0 (WritebackStage=0) mscratch 44 yes 0x1c, 0x418, 0x0

On the cycle rows the third word is the counter value the test stores.

Expected Behavior

Per the RISC-V Privileged Specification, "Hardware Performance Monitor" (machine.adoc, Zicntr):

The minstret CSR counts the number of instructions the hart has retired.

An instruction that raises an exception does not retire. The wb0 mscratch row should retire 43, the same as tag_zeros.

Steps to reproduce the issue

  • cheriot_enable_i tied on in examples/simple_system/rtl/ibex_simple_system.sv. RV32E=0, SecureIbex=0.
  • tag_zeros: WritebackStage=1. wb0: WritebackStage=0.
  • riscv64-unknown-elf-gcc 13.2.0, -march=rv32imc_zicsr -mabi=ilp32. csr_asr_minstret.S and link.ld are attached.
GCC=riscv64-unknown-elf-gcc
SIM_TAG=/path/to/build_tag_zeros/sim-verilator/Vibex_simple_system
SIM_WB0=/path/to/build_wb0/sim-verilator/Vibex_simple_system

for c in 1 2; do
 $GCC -march=rv32imc_zicsr -mabi=ilp32 -nostdlib -nostartfiles \
 -DCASE=$c -T link.ld \
 -o csr_asr_case$c.elf csr_asr_minstret.S
done

for build in tag_zeros wb0; do
 for c in 1 2; do
 mkdir -p runs/${build}_case$c
 ( cd runs/${build}_case$c && \
 timeout 120 "$SIM_${build^^}" \
 --meminit=ram,csr_asr_case$c.elf -c 20000 > sim.out 2>&1 )
 grep -aE 'Instructions Retired|Cycles' \
 runs/${build}_case$c/ibex_simple_system_pcount.csv
 python3 rec.py runs/${build}_case$c/trace_core_00000000.log
 done
done

tag_zeros case1: Cycles,44 Instructions Retired,37
tag_zeros case2: Cycles,54 Instructions Retired,43
wb0 case1: Cycles,48 Instructions Retired,37
wb0 case2: Cycles,59 Instructions Retired,44

Attached: csr_asr_minstret.S, link.ld, csr_asr_case1.elf, csr_asr_case2.elf, rec.py, and logs/tag_zeros_case1/, logs/tag_zeros_case2/, logs/wb0_case1/, logs/wb0_case2/ (each has pcount.csv and trace.log).

My Environment

EDA tool and version: Verilator (Vibex_simple_system elaborations).

Operating system: Linux 6.8.0-90-generic x86_64.

Version of the Ibex source code: d513671dae09e701ceb6cd732d2da5fb52c13856. The only source change is the cheriot_enable_i tie-off above.

With WritebackStage=0, perf_instr_ret_wb_o is instr_perf_count_id_i & en_wb_i & ~dummy_instr_wb_o & ~(lsu_resp_valid_i & lsu_resp_err_i) (ibex_wb_stage.sv:269). instr_perf_count_id_o (ibex_id_stage.sv:1209) does not include cheriot_asr_err_d (ibex_controller.sv:269). cheriot_wb_err_i is unused (ibex_wb_stage.sv:294).

#2552 (fab0d86) masks CHERIoT faults raised in writeback, and says traps raised in ID are never counted. This fault is csrr mscratch in ID.

testcase.zip

主要语言
SystemVerilog
星标
2.2k
派生
836
平均合并
5 天 8 小时
30 天内合并 PR
22

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

lowRISC/ibex 的其他 Issue

查看 lowRISC/ibex 的全部 Issue

相似的 Issue

更多 Backend & API Design Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。