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;
Subject: Re: [PATCH 3/5] ARM: stacktrace: Allow stack trace saving for
 non-current tasks
To:     Linus Walleij <linus.walleij@linaro.org>
CC:     <linux@armlinux.org.uk>, <rmk+kernel@armlinux.org.uk>,
        <ardb@kernel.org>, <will@kernel.org>, <mark.rutland@arm.com>,
        <broonie@kernel.org>, <peterz@infradead.org>, <mingo@redhat.com>,
        <acme@kernel.org>, <alexander.shishkin@linux.intel.com>,
        <jolsa@kernel.org>, <namhyung@kernel.org>, <arnd@arndb.de>,
        <rostedt@goodmis.org>, <nick.hawkins@hpe.com>, <john@phrozen.org>,
        <mhiramat@kernel.org>, <ast@kernel.org>, <linyujun809@huawei.com>,
        <ndesaulniers@google.com>, <linux-arm-kernel@lists.infradead.org>,
        <linux-kernel@vger.kernel.org>, <linux-perf-users@vger.kernel.org>,
        Li Huafei <lihuafei1@huawei.com>
References: <20220712021527.109921-1-lihuafei1@huawei.com>
 <20220712021527.109921-4-lihuafei1@huawei.com>
 <CACRpkdYvOjfmf=Z3pGfD-UPxfTc9PXtOyw2x+ptYiSy=gmGnpQ@mail.gmail.com>
From:   Li Huafei <lihuafei1@huawei.com>
Message-ID: <1288c73b-cf29-707d-47cb-4e2737300a29@huawei.com>
Date:   Tue, 26 Jul 2022 17:12:39 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101
 Thunderbird/78.9.0
MIME-Version: 1.0
In-Reply-To: <CACRpkdYvOjfmf=Z3pGfD-UPxfTc9PXtOyw2x+ptYiSy=gmGnpQ@mail.gmail.com>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 2022/7/18 17:07, Linus Walleij wrote:
> On Tue, Jul 12, 2022 at 4:18 AM Li Huafei <lihuafei1@huawei.com> wrote:
> 
>> The current ARM implementation of save_stack_trace_tsk() does not allow
>> saving stack trace for non-current tasks, which may limit the scenarios
>> in which stack_trace_save_tsk() can be used. Like other architectures,
>> or like ARM's unwind_backtrace(), we can leave it up to the caller to
>> ensure that the task that needs to be unwound is not running.
>>
>> Signed-off-by: Li Huafei <lihuafei1@huawei.com>
> 
> That sounds good, but:
> 
>>         if (tsk != current) {
>> -#ifdef CONFIG_SMP
>> -               /*
>> -                * What guarantees do we have here that 'tsk' is not
>> -                * running on another CPU?  For now, ignore it as we
>> -                * can't guarantee we won't explode.
>> -                */
>> -               return;
>> -#else
>> +               /* task blocked in __switch_to */
> 
> The commit text is not consistent with the comment you are removing.
> 
> The commit is talking about "non-current" tasks which is one thing,
> but the code is avoiding any tasks under SMP because they may be
> running on another CPU. So you need to update the commit
> message to say something like "non-current or running on another CPU".
> 
> If this condition will be checked at call sites in following patches,
> then mention
> that in the commit as well, so we know the end result is that we do
> not break it,

The generic code stack_trace_save_tsk() does not have this check, and by
'caller' I mean the caller of stack_trace_save_tsk(), expecting the
'caller' to ensure that the task is not running. So in effect this check
has been dropped and there is no more guarantee. Sorry for not
clarifying the change here.

But can we assume that the user should know that the stacktrace is
unreliable for a task that is running on another CPU? If not, I should
remove this patch and keep the check.

Thanks,
Huafei
> 
> I think Russell want to check this commit as well,
> 
> Yours,
> Linus Walleij
> .
>