Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.32 as permitted sender) client-ip=23.128.96.32;
Message-ID: <3fbc49da-99ac-4499-84df-064f3dc429d2@quicinc.com>
Date:   Mon, 20 Nov 2023 10:00:40 +0800
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v2 1/1] ARM: kprobes: Explicitly reserve r7 for local
 variables
To:     Ard Biesheuvel <ardb@kernel.org>,
        Nathan Chancellor <nathan@kernel.org>
CC:     <linux@armlinux.org.uk>, <mhiramat@kernel.org>,
        <kernel@quicinc.com>, <linux-arm-kernel@lists.infradead.org>,
        <linux-arm-msm@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
        <quic_lijuang@quicinc.com>, <stable@vger.kernel.org>
References: <20231115095830.20607-1-quic_aiquny@quicinc.com>
 <20231116172418.GA174808@dev-arch.thelio-3990X>
 <CAMj1kXE4stcFikmxBMY19CR4J0+EpaNW8J=xVyFVY3A7F7BpRQ@mail.gmail.com>
From:   "Aiqun(Maria) Yu" <quic_aiquny@quicinc.com>
In-Reply-To: <CAMj1kXE4stcFikmxBMY19CR4J0+EpaNW8J=xVyFVY3A7F7BpRQ@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 11/17/2023 1:36 AM, Ard Biesheuvel wrote:
> On Fri, 17 Nov 2023 at 03:24, Nathan Chancellor <nathan@kernel.org> wrote:
>>
>> On Wed, Nov 15, 2023 at 05:58:30PM +0800, Maria Yu wrote:
>>> Registers r7 is removed in clobber list, so compiler may choose r7 for
>>> local variables usage, while r7 will be actually updated by the inline asm
>>> code. This caused the runtime behavior wrong.
>>> While those kind of reserved registers cannot be set to clobber list
>>> because of error like "inline asm clobber list contains reserved
>>> registers".
>>> Explicitly reserve r7 by adding attribute no-omit-frame-pointer for needed
>>> function, then in T32 asm code r7 is used as a frame pointer and is not
>>> available for use as a general-purpose register.
>>> Note that "no-omit-frame-pointer" will make the code size a little bigger
>>> to store the stack frame pointer. So limited to needed functions can have
>>> the less impact than the full source file.
>>>
>>> Fixes: dd12e97f3c72 ("ARM: kprobes: treat R7 as the frame pointer register in Thumb2 builds")
>>> Signed-off-by: Maria Yu <quic_aiquny@quicinc.com>
>>> Cc: stable@vger.kernel.org
>>
>> This causes warnings with clang:
>>
>>    arch/arm/probes/kprobes/actions-thumb.c:441:47: warning: unknown attribute 'optimize' ignored [-Wunknown-attributes]
>>      441 | static unsigned long __kprobes __attribute__((optimize("no-omit-frame-pointer")))
>>          |                                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>    arch/arm/probes/kprobes/actions-thumb.c:524:38: warning: unknown attribute 'optimize' ignored [-Wunknown-attributes]
>>      524 | static void __kprobes __attribute__((optimize("no-omit-frame-pointer")))
>>          |                                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>    arch/arm/probes/kprobes/actions-thumb.c:560:38: warning: unknown attribute 'optimize' ignored [-Wunknown-attributes]
>>      560 | static void __kprobes __attribute__((optimize("no-omit-frame-pointer")))
>>          |                                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>    arch/arm/probes/kprobes/actions-thumb.c:579:38: warning: unknown attribute 'optimize' ignored [-Wunknown-attributes]
>>      579 | static void __kprobes __attribute__((optimize("no-omit-frame-pointer")))
>>          |                                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>    4 warnings generated.
>>
>> Furthermore, as far as I am aware, the optimize attribute has other issues so
>> its use is discouraged, see commits 080b6f407635 ("bpf: Don't rely on GCC
>> __attribute__((optimize)) to disable GCSE") and a7223f5bfcae ("powerpc: Avoid
>> broken GCC __attribute__((optimize))").

Thx for trying this. I was tested with arm-linux-gnueabihf-gcc and not 
notice this.
>>
> 
> Ah yes, apalogies for missing that. I did a test build with Clang but
> I did not spot the errors. So
> 
> Unreviewed-by: ....
> 
> Maria, please use the Makefile based per-file CFLAGS override that I
> suggested before. There is really no reason to make this per-function.
Ok. I will upload a new patchset.

-- 
Thx and BRs,
Aiqun(Maria) Yu