linux-kernel - Re: [RFC PATCH] trace: prevent preemption in perf_ftrace_function

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <d82068a0-8bb0-b824-49f4-c7f55525ac95@linux.alibaba.com>
Date:   Fri, 24 Sep 2021 11:36:53 +0800
From:   王贇 <yun.wang@...ux.alibaba.com>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Ingo Molnar <mingo@...hat.com>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH] trace: prevent preemption in
 perf_ftrace_function_call()



On 2021/9/24 上午11:26, Steven Rostedt wrote:
> On Fri, 24 Sep 2021 10:08:10 +0800
> 王贇 <yun.wang@...ux.alibaba.com> wrote:
> 
>> I found the rcu tree implementation of rcu_is_watching() will check
>> this_cpu_ptr(&rcu_data.dynticks), and after that enable the preemption.
>>
>> If preemption happened after that and before we disable here, there are
>> still possibility that the CPU changed and make the dynticks checking
>> invalid, isn't it?
> 
> If it can be scheduled, then RCU is definitely watching ;-)
> 
> The rcu_is_watching() is a safe guard for places that are in between
> context switches. Not task context switches, but transitioning between
> kernel and user space, or going into or out of idle, or transitioning
> in and out of an interrupt.  There are small critical sections that RCU
> is not watching, and we are actually working on making those locations
> disable instrumentation (like tracing), where rcu_is_watching() will no
> longer be needed.

Thanks for the explain :-)

Context available for scheduling should not in these situations, will
move down the 'disable' in v2.

Regards,
Michael Wang

> 
> -- Steve
>