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;
Subject: Re: [RFC PATCH v7 6/9] media: tegra: Add Tegra210 Video input driver
From:   Sowjanya Komatineni <skomatineni@nvidia.com>
To:     Dmitry Osipenko <digetx@gmail.com>, <thierry.reding@gmail.com>,
        <jonathanh@nvidia.com>, <frankc@nvidia.com>, <hverkuil@xs4all.nl>,
        <sakari.ailus@iki.fi>, <helen.koike@collabora.com>
CC:     <sboyd@kernel.org>, <linux-media@vger.kernel.org>,
        <devicetree@vger.kernel.org>, <linux-clk@vger.kernel.org>,
        <linux-tegra@vger.kernel.org>, <linux-kernel@vger.kernel.org>
References: <1586919463-30542-1-git-send-email-skomatineni@nvidia.com>
 <1586919463-30542-7-git-send-email-skomatineni@nvidia.com>
 <4118112f-f865-5460-6319-d71271fd78d1@gmail.com>
 <a69a8b34-beea-3ad0-e08e-f7df8b9e7047@nvidia.com>
Message-ID: <6afa951e-d904-f3c0-053f-82a02fb18979@nvidia.com>
Date:   Wed, 15 Apr 2020 10:21:15 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.9.0
MIME-Version: 1.0
In-Reply-To: <a69a8b34-beea-3ad0-e08e-f7df8b9e7047@nvidia.com>
Content-Type: text/plain; charset="utf-8"; format=flowed
Content-Transfer-Encoding: quoted-printable
Content-Language: en-US
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 4/15/20 9:54 AM, Sowjanya Komatineni wrote:
>
> On 4/15/20 7:22 AM, Dmitry Osipenko wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> 15.04.2020 05:57, Sowjanya Komatineni =D0=BF=D0=B8=D1=88=D0=B5=D1=82:
>>> +static int tegra_csi_remove(struct platform_device *pdev)
>>> +{
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 struct tegra_csi *csi =3D platform_get_drvdat=
a(pdev);
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 int err;
>>> +
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 err =3D host1x_client_unregister(&csi->client=
);
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 if (err < 0) {
>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0 dev_err(csi->dev,
>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 "failed to unregister h=
ost1x client: %d\n", err);
>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0 return err;
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 }
>>> +
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 pm_runtime_disable(csi->dev);
>>> +=C2=A0=C2=A0=C2=A0=C2=A0 kfree(csi);
>> IIRC, the driver removal is invoked on the unbinding. Hence, I'm not
>> sure how moving away from the resource-managed API helps here. Could you
>> please explain in a more details?
>>
>> Have you tried to test this driver under KASAN? I suspect that you just
>> masked the problem, instead of fixing it.
> Using devm_kzalloc for vi/csi structures based on prior feedback=20
> request to switch to use kzalloc all over this driver.
>
> Hi Hans,
>
> video devices lifetime is till video device nodes are released. So,=20
> v4l2 device release callback does the release of tegra channel=20
> allocation which hold video device.
>
> Below are the 3 possible cases of unbind/unload,
>
> 1. during tegra-video module unload, if v4l2 device refcnt is not 0=20
> which is the case when any of video device node handle is kept opened=20
> then unloading module will not happen and module refcnt is also=20
> non-zero and unloading tegra-video module reports module in use.
>
> 2. during tegra-video driver unbind, tegra-video driver removal will=20
> do vi/csi clients exit ops which unregisters video device allocated=20
> memory during release callback of v4l2 device. vi/csi structure=20
> allocation remains same as vi/csi driver removal will not happen in=20
> this case.
>
>
> 3. during direct host1x client drivers vi/csi unbind, both=20
> host1x_clients vi/csi gets unregistered, deletes host1x logical device=20
> which executes tegra-video driver removal() -> vi/csi exit() before=20
> vi/csi memory gets freed in vi/csi driver remove().
>
> So, any active streaming will stop and video devices are unregistered=20
> during direct client driver unbind prior to freeing vi/csi memory.
>
> Also vi/csi driver remove does explicit free vi/csi as its allocated=20
> with kzalloc. So not sure how using kzalloc is different to=20
> devm_kzalloc for vi/csi structure in terms of when vi/csi memory gets=20
> freed?
>
> Except for channel allocation which holds video device and as video=20
> device life time is beyond tegra-video module unbind->vi exit(), looks=20
> like we can use devm_kzalloc for vi/csi.
>
>
> Can you please comment if you still think we need to use kzalloc=20
> rather than devm_kzalloc for vi/csi structure allocation?
>
> Thanks
>
> Sowjanya
>
One more case is when video device node is kept opened with v4l2-ctl=20
sleep (rather than streaming), where it will keep device node open for=20
specified time and if direct vi client driver unbind happens then vi=20
driver remove() will free vi memory before v4l2 device release happens.

But I don't see any crash or errors with this case.

Also if we allow direct client driver unbind, then vi structure memory=20
lifetime should also be till v4l2 device release happens.

But we can free vi in v4l2 device release callback as in case when=20
device node is not kept opened, video device release happens immediate=20
and we cant free vi that early.

Hans/Thierry, Can you please comment on this case?

Thanks

Sowjanya