Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20220315103813.84407-1-shreeya.patel@collabora.com>
 <CAHp75VdoN+iSu1GLnxWW9BtL-p9uF1sfAw3ZxkFWNpoo44+bZg@mail.gmail.com> <b22cea07-aa2e-f5b5-d5aa-071e1084f00b@collabora.com>
In-Reply-To: <b22cea07-aa2e-f5b5-d5aa-071e1084f00b@collabora.com>
From:   Andy Shevchenko <andy.shevchenko@gmail.com>
Date:   Tue, 15 Mar 2022 15:45:39 +0200
Message-ID: <CAHp75VfsenJ64tTMdAG=CRFkwWjxf5hKpWWSa4hRxkF4ReQdHg@mail.gmail.com>
Subject: Re: [PATCH v2] gpio: Restrict usage of gc irq members before initialization
To:     Shreeya Patel <shreeya.patel@collabora.com>
Cc:     Linus Walleij <linus.walleij@linaro.org>,
        Bartosz Golaszewski <brgl@bgdev.pl>, krisman@collabora.com,
        "open list:GPIO SUBSYSTEM" <linux-gpio@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Collabora Kernel ML <kernel@collabora.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Mar 15, 2022 at 2:32 PM Shreeya Patel
<shreeya.patel@collabora.com> wrote:
> On 15/03/22 4:30 pm, Andy Shevchenko wrote:
> > On Tue, Mar 15, 2022 at 12:38 PM Shreeya Patel
> > <shreeya.patel@collabora.com> wrote:
> >
> > Thanks for the update, my comments below.
> >
> >> gc irq members are exposed before they could be completely
> > gc --> GPIO chip
> >
> >> initialized and this leads to race conditions.
> > Any example here. like ~3-4 lines of the Oops in question?
> >
> >> One such issue was observed for the gc->irq.domain variable which
> >> was accessed through the I2C interface in gpiochip_to_irq() before
> >> it could be initialized by gpiochip_add_irqchip(). This resulted in
> >> Kernel NULL pointer dereference.
> >>
> >> To avoid such scenarios, restrict usage of gc irq members before
> > gc --> GPIO chip
> >
> >> they are completely initialized.
> > ...
> >
> >> +       /*
> >> +        * Using barrier() here to prevent compiler from reordering
> >> +        * gc->irq.gc_irq_initialized before initialization of above
> >> +        * gc irq members.
> >> +        */
> >> +       barrier();
> >> +
> >> +       gc->irq.gc_irq_initialized = true;
> > There are too many duplications. Why not simply call it 'initialized'?
> >
> >> -       if (gc->to_irq) {
> >> +       if (gc->to_irq && gc->irq.gc_irq_initialized) {
> > Why can't this check be added into gpiochip_to_irq() ?
> >
> >      if (!gc->irq.initialized)
> >          return -ENXIO;
> >
> > ...
>
>
> Because we don't want to return -ENXIO in case of race condition.
>
> It should return -EPROBE_DEFER similar to how we are doing when gc->to_irq
> is NULL.

> So in this case when both gc->to_irq = NULL and gc->irq.initialized = FALSE,
> we will be returning -EPROBE_DEFER.

This is not true. The return code relies on an IRQ chip which may be
assigned (not NULL).

> This will make sure that devices
> like touchscreen
> do not become fatal due to returning -ENXIO.

So, then you need to move it to to_irq() and return there deferred
probe with a good comment in the code.

> >> +       bool gc_irq_initialized;
> > Can you move it closer to .init_hw so it will be weakly grouped by
> > logic similarities?
> > Also see above.
>
> Thanks for your comments, I'll make the necessary changes and send a v3.

-- 
With Best Regards,
Andy Shevchenko