Re: [Libusb-devel] [PATCH 1/4] linux: Stop the kernel from re-attaching in kernel drivers after a reset

[Xiaofan C. <xia...@gm...>]

On Thu, Feb 10, 2011 at 3:41 PM, Hans de Goede <hde...@re...> wrote:
>> I think this does not affect Windows. For one thing, the
>> current WinUSB and HID backends do not support
>>   libusb_reset_device. And even for libusb0.sys which
>> might support that function, there is no problem, the driver
>> will be the same after reset.
>>
>> I am not so sure if this patch is really needed, even for
>> Linux. Typically after device reset, the handle may not be
>> valid and you need to redo the initial steps like to detach
>> the kernel driver again.
>
> That is not true, the handle only becomes invalid if the devices
> "morphs" into another device such as from cdrom -> modem with
> certain adsl modem. If the device does not morph the handle
> stays perfectly valid. Currently one does indeed need to re-detach
> the in kernel driver and reclaim. This is very bad for the following
> reason (example assumes a single interface device):
>
> -app does reset to get device into clean state
> -kernel driver binds talks to device ->
>  device state no longer clean
> -app unbinds kernel driver
> -app claims
> -app still does not have a clean device state
>
> With my solution this becomes:
> -app does reset to get device into clean state
> -app has device in clean state and can continue
>
> Note that this is not some hypothetical case, I've observed the
> old unwanted scenario happening, and I've also tested my code
> and it works (leads to the 2nd scenario).
>

What if the use case is indeed that device "morphs" into another
device -- like firmware updating?


-- 
Xiaofan