On 05/17/2012 04:27 PM, Peter Stuge wrote:
> Uri Lublin wrote:
>>>> Only if old backend api is UNSUPPORTED.
>>> Hm. Shouldn't the check be done unconditionally - and for all
>>> devices, ie. in every pass except perhaps HCD? Otherwise I think
>>> the bug still bites when installing the previous driver after
>>> unredirecting the device?
>> Yes, it may be a problem if an old libusb driver was already installed.
>>
>> I was concerned about the device being opened, and existing in-flight
>> messages.
> If a device becomes unavailable (which can happen without this patch
> too of course) then the backend has to start failing transfers with
> LIBUSB_ERROR_NO_DEVICE. Is this what currently happens?
I suppose you are right. I did not test such a test case.
I wanted to fix the problem for no-driver --> driver case, and change
nothing
for other cases, such that the chance of me breaking something I did not
think of
would be smaller.
In my case libusb_open fails even after the driver got installed, since
the backend
was still UNSUPPORTED.
>
>> So I preferred to "play it safe" and only change if it was UNSUPPORTED
>> before.
> I think it should be done in all cases, or the bug (_get_device_list
> does not reflect current system state) is only partially fixed. Can
> you do some tests also for the other way around? A simple way to test
> is that you in your program first ask the clerk to install the libusb
> driver, then open the device, then ask the clerk to return the old
> driver, then try to use the device. It would also be intersting to
> test libusb_open() in this path, ie. ask clerk to install driver,
> call _get_device_list(), ask clerk to return the old driver, then
> call libusb_open().
>
> I would very much appreciate seeing the results from such testing, it
> would be a good way to determine if there are some concerns for
> fixing this bug.
>
OK, I'll send results once I have them.
Currently I'm looking into changing the application such that it would
not keep
reference to devices for which the driver is not installed. This should fix
the problem I was seeing as Pete proved with his xusb patch.
Thanks,
Uri.
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
libusbx-devel mailing list
libusbx-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libusbx-devel
