Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 4 May 2021 11:49:08 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Dmitry Vyukov <dvyukov@google.com>
Cc:     Andi Kleen <ak@linux.intel.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Andrey Ryabinin <ryabinin.a.a@gmail.com>,
        Alexander Potapenko <glider@google.com>,
        Andrey Konovalov <andreyknvl@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        kasan-dev <kasan-dev@googlegroups.com>
Subject: Re: [PATCH] stackdepot: Use a raw spinlock in stack depot
Message-ID: <YJEYlAo2HU8KfyxI@hirez.programming.kicks-ass.net>
References: <20210504024358.894950-1-ak@linux.intel.com>
 <CACT4Y+a5g5JeLJFPJEUxPFbMLXGkYEAJkK3MBctnn7UA-iTkXA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CACT4Y+a5g5JeLJFPJEUxPFbMLXGkYEAJkK3MBctnn7UA-iTkXA@mail.gmail.com>
Precedence: bulk

On Tue, May 04, 2021 at 09:23:34AM +0200, Dmitry Vyukov wrote:
> So why is this a false positive that we just need to silence?

No, it's a correctness issue for PREEMPT_RT.

> I see LOCKDEP is saying we are doing something wrong, and your
> description just describes how we are doing something wrong :)
> If this is a special false positive case, it would be good to have a
> comment on DEFINE_RAW_SPINLOCK explaining why we are using it.

Documentation/locking/locktypes.rst has the low-down IIRC

> I wonder why we never saw this on syzbot. Is it an RT kernel or some
> other special config?

IIRC the kernel isn't really PROVE_RAW_LOCK_NESTING=y clean yet, so
mostly these checks aren't on by default. printk() used to be a common
offender, but I've not checked the very latest printk status to see if
that got fixed meanwhile.

> A similar issue was discussed recently for RT kernel:
> https://groups.google.com/g/kasan-dev/c/MyHh8ov-ciU/m/nahiuqFLAQAJ
> And I think it may be fixable in the same way -- make stackdepot not
> allocate in contexts where it's not OK to allocate.

That would be preferable I think.