linux-kernel - Re: [PATCH v3] usb: core: verify devicetree nodes for USB devices

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <yw1x4l5yvfg1.fsf@mansr.com>
Date:   Mon, 13 May 2019 11:03:58 +0100
From:   Måns Rullgård <mans@...sr.com>
To:     Marek Szyprowski <m.szyprowski@...sung.com>
Cc:     Peter Chen <peter.chen@....com>,
        "linux-usb\@vger.kernel.org" <linux-usb@...r.kernel.org>,
        "linux-samsung-soc\@vger.kernel.org" 
        <linux-samsung-soc@...r.kernel.org>,
        "linux-kernel\@vger.kernel.org" <linux-kernel@...r.kernel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Bartlomiej Zolnierkiewicz <b.zolnierkie@...sung.com>,
        Markus Reichl <m.reichl@...etechno.de>,
        Krzysztof Kozlowski <krzk@...nel.org>
Subject: Re: [PATCH v3] usb: core: verify devicetree nodes for USB devices

Marek Szyprowski <m.szyprowski@...sung.com> writes:

> Hi Peter,
>
> On 2019-05-10 05:10, Peter Chen wrote:
>>
>>> Marek Szyprowski <m.szyprowski@...sung.com> writes:
>>>> Commit 69bec7259853 ("USB: core: let USB device know device node")
>>>> added support for attaching devicetree node for USB devices. The
>>>> mentioned commit however identifies the given USB device node only by the 'reg'
>>>> property in the host controller children nodes. The USB device node
>>>> however also has to have a 'compatible' property as described in
>>>> Documentation/devicetree/bindings/usb/usb-device.txt. Lack for the
>>>> 'compatible' property check might result in assigning a devicetree
>>>> node, which is not intended to be the proper node for the given USB device.
>>>>
>>>> This is important especially when USB host controller has child-nodes
>>>> for other purposes. For example, Exynos EHCI and OHCI drivers already
>>>> define child-nodes for each physical root hub port and assigns
>>>> respective PHY controller and parameters for them. Those binding
>>>> predates support for USB devicetree nodes.
>>>>
>>>> Checking for the proper compatibility string allows to mitigate the
>>>> conflict between USB device devicetree nodes and the bindings for USB
>>>> controllers with child nodes. It also fixes the side-effect of the
>>>> other commits, like 01fdf179f4b0 ("usb: core: skip interfaces disabled
>>>> in devicetree"), which incorrectly disables some devices on Exynos
>>>> based boards.
>> Hi Marek,
>>
>> The purpose of your patch is do not set of_node for device under USB
>> controller, right?
>
> Right.
>
>> I do not understand how 01fdf179f4b0 affect your boards, some nodes
>> under the USB controller with status is not "okay", but still want to
>> be enumerated?
>
> Please look at the ehci node in arch/arm/boot/dts/exynos4.dtsi and then 
> at the changes to that node in arch/arm/boot/dts/exynos4412-odroidx.dts. 
> Exynos EHCI controller has 3 subnodes, which matches to the physical 
> ports of it and allows the driver to enable given PHY ports depending on 
> which physical port is used on the particular board. All ports cannot 
> not be enabled by default, because PHY controller has limited resources 
> and shares them between USB host and USB device ports.

It seems like what's happening is that the Exynos port/phy nodes are
mistaken for standard USB device nodes attached to the root hub.  The
problem is that hub port numbering starts at 1 while the Exynos nodes
start from 0.  This causes attached devices to be associated with the
wrong DT node.

Ignoring backwards compatibility, I can see a few ways of fixing this:

- Add another child node, along side the port@N nodes, of the host
  controller to represent the root hub.  Nodes for attached devices
  would then be descendants of this new node.

- Change the Exynos HCD binding to use a more standard "phys" property
  and get rid of the child nodes for this purpose.

- Move the port@N nodes below a new dedicated child node of the HCD.

The first is probably the easiest to implement since it doesn't require
any nasty hacks to avoid breaking existing device trees.

-- 
Måns Rullgård