Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   =?utf-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn@kernel.org>
To:     Yunhui Cui <cuiyunhui@bytedance.com>, conor.dooley@microchip.com,
        paul.walmsley@sifive.com, palmer@dabbelt.com,
        aou@eecs.berkeley.edu, peterz@infradead.org, mpe@ellerman.id.au,
        jpoimboe@kernel.org, mark.rutland@arm.com, svens@linux.ibm.com,
        cuiyunhui@bytedance.com, guoren@kernel.org, jszhang@kernel.org,
        ebiederm@xmission.com, bjorn@rivosinc.com, heiko@sntech.de,
        xianting.tian@linux.alibaba.com, mnissler@rivosinc.com,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] riscv: Dump user opcode bytes on fatal faults
In-Reply-To: <20230329082950.726-1-cuiyunhui@bytedance.com>
References: <20230329082950.726-1-cuiyunhui@bytedance.com>
Date:   Wed, 16 Aug 2023 17:10:51 +0200
Message-ID: <87v8dfqays.fsf@all.your.base.are.belong.to.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Hi Yunhui,

Waking up the dead! ;-)

Yunhui Cui <cuiyunhui@bytedance.com> writes:

> We encountered such a problem that when the system starts to execute
> init, init exits unexpectedly with error message: "unhandled signal 4
> code 0x1 ...".
>
> We are more curious about which instruction execution caused the
> exception. After dumping it through show_opcodes(), we found that it
> was caused by a floating-point instruction.
>
> In this way, we found the problem: in the system bringup , it is
> precisely that we have not enabled the floating point function(CONFIG_FPU
> is set, but not enalbe COMPAT_HWCAP_ISA_F/D in the dts or acpi).
>
> Like commit ba54d856a9d8 ("x86/fault: Dump user opcode bytes on fatal
> faults"), when an exception occurs, it is necessary to dump the
> instruction that caused the exception.

X86's show_opcodes() is used both for kernel oops:es, and userland
unhandled signals. On RISC-V there's dump_kernel_instr() added in commit
eb165bfa8eaf ("riscv: Add instruction dump to RISC-V splats").

Wdyt about reworking that function, so that it works for userland epc as
well? I think it's useful to have the surrounding instruction context,
and not just on instruction.


Bj=C3=B6rn