Re: [PATCH] usb: echi-hcd: Add register access check in shutdown

[Srinivas Kandagatla <srinivas.kandagatla@linaro.org>]

On 18/05/16 17:15, Alan Stern wrote:
> On Wed, 18 May 2016, Srinivas Kandagatla wrote:
>
>> On 18/05/16 15:56, Alan Stern wrote:
>>> On Wed, 18 May 2016, Srinivas Kandagatla wrote:
>>>
>>>> This patch adds a check in ehci_shutdown(), to make sure
>>>> that the register access is available before accessing registers.
>>>>
>>>> The use case is simple, for boards like DB410c where the usb host
>>>> or device functionality is decided based on the micro-usb cable
>>>> presence. If the board boots up with micro-usb connected and the
>>>> host driver is probed, but the ehci_setup() has not been done yet,
>>>> then a system shutdown would trigger below NULL pointer exception
>>>> without this patch.
>>>
>>> How can that happen?  While the host driver is probed, the probing
>>> thread holds the device lock.  But the system shutdown routine acquires
>>> the device lock before invoking the ->shutdown callback.  Therefore the
>>> two things cannot happen concurrently.
>>
>> No, I did not mean them happening concurrently, I mean that the host
>> driver is up, however ehci_setup() is not done yet.
>
> I don't understand.  ehci_setup() is called as part of the probe
> procedure.  How can the host driver be up if ehci_setup() is not done
> yet?
>
Yes, this is true in ehci-msm driver, The driver does not add usb host 
by default in probe when phy is otg capable.

The usb host is added dynamically by the msm_otg driver depending on the 
the micro USB cable plug/un-plug events via extcon.

> Are you saying that when the system is plugged into the "B" end of an
> OTG cable, ehci_setup() doesn't get called at all?
>
Yes, for echi-msm driver, not sure about other host controller drivers.

> And would the same thing happen if the system started out as the host
> but then used HNP to change into the peripheral?
I don't think so, As the ehci->regs get populated once we enter the 
ehci_setup(), so ehci_halt() will never get chance to dereference null 
in this case.

Fault occurs only if the driver did not enter into host mode and system 
reboot/shutdown is requested.

--srini


>
> Alan Stern
>