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;
MIME-Version: 1.0
References: <20220830065413.638-1-thunder.leizhen@huawei.com>
 <CAGETcx8hfzAthF_gCMrmnVOgwgFJEPE=bwRuVeXaiB7ywUkWkg@mail.gmail.com>
 <Yw3cs6wb4PvnDN/7@shell.armlinux.org.uk> <b120e94b-4b43-6d8a-a136-41417cb250cc@huawei.com>
 <Yw3hfGV4DYyJ0Gno@shell.armlinux.org.uk> <2452e7b0-d4fd-2e27-a336-bf87b85a8d50@huawei.com>
 <Yw3oR6WhM0O5LQX5@shell.armlinux.org.uk>
In-Reply-To: <Yw3oR6WhM0O5LQX5@shell.armlinux.org.uk>
From:   Saravana Kannan <saravanak@google.com>
Date:   Tue, 30 Aug 2022 10:34:31 -0700
Message-ID: <CAGETcx_GB-r+H96hWHfoKSZjK9hodqe-+FWh6rve0E09pz0nXA@mail.gmail.com>
Subject: Re: [PATCH v2] ARM: Add sanity check for dev->periphid in amba_probe()
To:     "Russell King (Oracle)" <linux@armlinux.org.uk>
Cc:     "Leizhen (ThunderTown)" <thunder.leizhen@huawei.com>,
        linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
        patches@armlinux.org.uk, Kefeng Wang <wangkefeng.wang@huawei.com>,
        Linus Walleij <linus.walleij@linaro.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Aug 30, 2022 at 3:36 AM Russell King (Oracle)
<linux@armlinux.org.uk> wrote:
>
> On Tue, Aug 30, 2022 at 06:31:14PM +0800, Leizhen (ThunderTown) wrote:
> > On 2022/8/30 18:07, Russell King (Oracle) wrote:
> > > For an amba driver, drv->bus will always be pointing at amba_bustype.
> > > That always has a "match" operation. Therefore, the default of '1'
> > > above will *never* be used for an AMBA driver.
> > >
> > > If drv->bus does not point at amba_bustype, then amba_probe() will
> > > not be called for "drv".
> > >
> > > Therefore, amba_match() must always be called before amba_probe().
> >
> > Oh, I was careless. I think it's drv->match. But the processing flow
> > will continue to go to "dev->bus->probe".
> >
> > __driver_attach():
> >         ret = driver_match_device(drv, dev);
> >         if (ret == 0) {
> >                 /* no match */
> >                 return 0;
> >         } else if (ret == -EPROBE_DEFER) {               <------no return in this branch
> >                 dev_dbg(dev, "Device match requests probe deferral\n");
> >                 dev->can_match = true;
> >                 driver_deferred_probe_add(dev);
> >         } else if (ret < 0) {
> >                 dev_dbg(dev, "Bus failed to match device: %d\n", ret);
> >                 return ret;
> >         } /* ret > 0 means positive match */
> >
> >         ... ...
> >         driver_probe_device(drv, dev);
> >              ......
> >                   dev->bus->probe
>
> And that makes no sense, is an already known issue, and there is a patch
> to fix it:
>
> https://lore.kernel.org/all/20220817184026.3468620-1-isaacmanjarres@google.com/

Russell,

Thanks for discussing this further and pointing out the other fix. I
assumed Leizhen was talking about an instance of
device_driver_attach() which allows probe to be called without match.
Thankfully that function is used only by some specific
frameworks/buses types. I did a cursory check and I don't see any
intersection with we amba.

-Saravana
P.S: I hate that function exists, as it just throws a wrench in the
whole driver core design and adds corner cases to a lot of generic
driver core design.