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;
Message-ID: <9bd50ce2-0ac8-a5dc-a584-0610891c7805@bytedance.com>
Date:   Sat, 8 Apr 2023 13:26:25 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0)
 Gecko/20100101 Thunderbird/102.9.1
Subject: Re: [PATCH 1/2] x86: make profile_pc() use arch_stack_walk()
Content-Language: en-US
To:     Josh Poimboeuf <jpoimboe@kernel.org>
Cc:     peterz@infradead.org, keescook@chromium.org,
        dave.hansen@linux.intel.com, bp@alien8.de, mingo@redhat.com,
        tglx@linutronix.de, rostedt@goodmis.org, x86@kernel.org,
        linux-kernel@vger.kernel.org,
        Chen Zhongjin <chenzhongjin@huawei.com>,
        Andi Kleen <ak@linux.intel.com>
References: <20230330081552.54178-1-zhengqi.arch@bytedance.com>
 <20230330081552.54178-2-zhengqi.arch@bytedance.com>
 <20230408045619.m5zfbispodf4zjvz@treble>
From:   Qi Zheng <zhengqi.arch@bytedance.com>
In-Reply-To: <20230408045619.m5zfbispodf4zjvz@treble>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 2023/4/8 12:56, Josh Poimboeuf wrote:
> On Thu, Mar 30, 2023 at 04:15:51PM +0800, Qi Zheng wrote:
>> The profile_pc() try to get pc by doing a trick to read
>> the contents of the stack. This may cause false positives
>> for KASAN, like the following:
>>
>>   BUG: KASAN: stack-out-of-bounds in profile_pc+0x5b/0x90
>>   Read of size 8 at addr ffff8881062a7a00 by task id/130040
> 
> I don't think this was actually a false positive.  The !FRAME_POINTER
> code in profile_pc() has been badly broken for many years.
> 
> BTW, there was a similar patch here:
> 
>    https://lore.kernel.org/lkml/20230224021858.120078-1-chenzhongjin@huawei.com/

Ah.

> 
> I thought CONFIG_PROFILING was obsolete but Andi said previously he
> wants to keep it for at least boot-time profiling.
> 
> Andi did suggest removing the lock profiling hacks, which means all the
> profile_pc() implementations can just be removed in favor of the generic
> instruction_pointer().

That's great, and I see Chen Zhongjin will send a new patch for this,
let him continue this work. :)

> 

-- 
Thanks,
Qi