Date:   Fri, 10 Feb 2023 13:30:53 -0800
From:   Josh Poimboeuf <jpoimboe@kernel.org>
To:     Chen Zhongjin <chenzhongjin@huawei.com>
Cc:     linux-kernel@vger.kernel.org, peterz@infradead.org,
        tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
        dave.hansen@linux.intel.com, x86@kernel.org, hpa@zytor.com
Subject: Re: [PATCH] x86/unwind/orc: Fix unwind ip when kprobes hits push/pop
Message-ID: <20230210213053.fscxeamu7h56mjaj@treble>
References: <20221201085311.249883-1-chenzhongjin@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20221201085311.249883-1-chenzhongjin@huawei.com>
Precedence: bulk

On Thu, Dec 01, 2022 at 04:53:11PM +0800, Chen Zhongjin wrote:
> When unwind stack at asm_exc_int3, the orc type is UNWIND_HINT_TYPE_REGS
> and the unwinder will use pt_regs->ip to find next orc, which point to
> the probed insn + INT3_INSN_SIZE.
> 
> If the probed insn is push/pop, it will point to the next insn which has
> different orc state. Before the probed insn has been really excuted in
> single step, using next insn ip to find orc will get a wrong unwinding
> result.
> 
> So, when there is kprobe running and the previous op code is int3,
> state->signal should be false so that ip - 1 will be used to find next
> orc, until the probed push/pop has been single steped and kprobe set
> kprobe_status as KPROBE_HIT_SSDONE.
> 
> Fixes: ee9f8fce9964 ("x86/unwind: Add the ORC unwinder")
> Signed-off-by: Chen Zhongjin <chenzhongjin@huawei.com>

Hi Chen,

Thanks for reporting this, and sorry for the delay.  I would rather fix
this in a more general way, by specifying 'signal' (or lack thereof) in
the ORC data.  I will work up a patch.

-- 
Josh