MIME-Version: 1.0
References: <20230129060452.7380-1-zhanghongchen@loongson.cn>
 <CAHk-=wjw-rrT59k6VdeLu4qUarQOzicsZPFGAO5J8TKM=oukUw@mail.gmail.com>
 <Y+EjmnRqpLuBFPX1@bombadil.infradead.org> <4ffbb0c8-c5d0-73b3-7a4e-2da9a7b03669@inria.fr>
 <Y+EupX1jX1c5BAHv@kadam>
In-Reply-To: <Y+EupX1jX1c5BAHv@kadam>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Mon, 6 Feb 2023 10:25:03 -0800
Message-ID: <CAHk-=wiibPvFEGy_Y=VDRNkBZyBxNB5oo0_4p_g3i2MduRZGig@mail.gmail.com>
Subject: Re: [PATCH v4] pipe: use __pipe_{lock,unlock} instead of spinlock
To:     Dan Carpenter <error27@gmail.com>
Cc:     Julia Lawall <julia.lawall@inria.fr>,
        Luis Chamberlain <mcgrof@kernel.org>,
        Hongchen Zhang <zhanghongchen@loongson.cn>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Andrew Morton <akpm@linux-foundation.org>,
        "Christian Brauner (Microsoft)" <brauner@kernel.org>,
        David Howells <dhowells@redhat.com>,
        Mauro Carvalho Chehab <mchehab@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        "Fabio M. De Francesco" <fmdefrancesco@gmail.com>,
        Christophe JAILLET <christophe.jaillet@wanadoo.fr>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        maobibo <maobibo@loongson.cn>,
        Matthew Wilcox <willy@infradead.org>,
        Sedat Dilek <sedat.dilek@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Feb 6, 2023 at 8:45 AM Dan Carpenter <error27@gmail.com> wrote:
>
> You need the cross function database to review these warnings.  [...]
>
> hl_device_set_debug_mode() take a mutex.  Then you do
> `smdb.py preempt hl_ctx_fini` and it prints out the call tree which
> disables preemption.
>
> cs_ioctl_unreserve_signals() <- disables preempt
> -> hl_ctx_put()
>    -> hl_ctx_do_release()
>       -> hl_ctx_fini()
>
> And so on.

Hmm. Do you have automation to do that at least for the non-driver (ie
"core kernel code") ones?

They are *hopefully* false positives, but if not they are obviously
the most interesting.

And they are presumably not quite as overwhelming as all the driver
ones, so even if they *are* false positives, maybe they would then be
the point to start looking at why the tool gives the wrong answer?

                Linus