speedtch usbatm.c,1.26,1.27

[matthieu castet]

matthieu castet wrote:
> Roman Kagan wrote:
> 
>> On Wed, May 11, 2005 at 08:21:14PM +0200, matthieu castet wrote:
>>
>>> matthieu castet wrote:
>>>
>>>> isn't there something wrong about iso pipe in current cvs :
>>>>
>>>> usbatm_usb_probe calls bind, then init rx_channel.endpoint and 
>>>> tx_channel.endpoint to bulk pipe and finaly test if there are iso pipe.
>>>>
>>>> How the sub-driver is supposed to choose iso pipe ?
>>>>
>>>
>>> any info about this ?
>>>
>>> It could be good to fix it before inclusion in kernel ?
>>
>>
>>
>> I can only humbly repeat my suggestion to move channel's enpoints
>> initialization back to before .bind(), to allow the subdriver to
>> override it, but apparently Duncan doesn't like this approach...
>>
> If he don't like it, why don't use something like that :
> it could be statiquely initialized, changed in usb_probe and in
> usbatm_bind .
> 

There is a problem with the actual implementation : for the eagleub 
modem we must have a precise number of packets by iso frame : 16.
If I try with 4 or 17 for example, I see bogus pdu_length  and packet 
failed crc check errors.


Matthieu