Re: ath9k_htc: Target is unresponsive

[Oleksij Rempel <linux@rempel-privat.de>]

Am 15.11.2013 17:07, schrieb Luis R. Rodriguez:
> On Fri, Nov 15, 2013 at 5:04 PM, Oleksij Rempel <linux@rempel-privat.de> wrote:
>> Am 15.11.2013 16:33, schrieb Luis R. Rodriguez:
>>> On Fri, Nov 15, 2013 at 3:36 PM, Oleksij Rempel <linux@rempel-privat.de> wrote:
>>>> Am 15.11.2013 15:29, schrieb Luis R. Rodriguez:
>>>>> On Fri, Nov 15, 2013 at 3:18 PM, Oleksij Rempel <linux@rempel-privat.de> wrote:
>>>>>> I do not know which backport include patch for usb interrupt transfer.
>>>>>
>>>>> Which patch are you referring to? The backports project releases match
>>>>> the Linux kernel releases, so the backports-3.12 release is based on
>>>>> linux v3.12, we can check what kernel the patch you are describing got
>>>>> merged into Linux by doing:
>>>>>
>>>>> git describe --contains gitsum
>>>>
>>>> i mean:
>>>>
>>>> commit 2f5e3ecfc155449987d64028ff6b73f29cd1ef8b
>>>> Author: Oleksij Rempel <linux@rempel-privat.de>
>>>> Date:   Tue Aug 13 09:29:34 2013 +0200
>>>>
>>>>     ath9k_htc: do not use bulk on EP3 and EP4
>>>
>>> OK that is upstream commit ID 2b721118b7
>>>
>>> mcgrof@frijol ~/linux-stable (git::master)$ git describe --contains
>>> 2b721118b7821107757eb1d37af4b60e877b27e7
>>> v3.12-rc1~132^2~84^2^2~88
>>>
>>> This means that patch got merged into v3.12-rc1 which means the
>>> backports release based on v3.12 will have it.
>>>
>>>> If it is already included, then try to revert.
>>>
>>> What do you mean? Are you saying to try to revert that patch in case
>>> v3.12 backports release causes issues with the card?
>>
>> yes
>>
>>> It seems the user
>>> of this card had issues with the v3.10 release of backports which
>>> would not have this patch merged.
>>
>> Ok. Thank you.
> 
> So to be clear -- you would recommend testing the new v3.12 release
> that *has* the patch then?

Yes. Before this patch ath9k_htc was overwriting usb descriptor from
driver. The problem with it - USB subsystem do not have this option, so
not all usb host controller drivers will use updated information.
In case driver indeed use Bulk instead of Int, it can still cause other
problems. For example not all controller can work with 64 byte bulk frames.

This behaviour was added back in 2010 and should be workaround for speed
problem on EP4, which is intensively used by REG_READ/WRITE.

Right now there are two options:
- use Bulk, but fail on some controllers
- use Int but be damn slow on other controllers.

After i tested ath9k-htc on different usb HC, i will not wonder that
there can be other problems too. Here is some info about known usb issues:
https://github.com/qca/open-ath9k-htc-firmware/wiki/usb-related-issues

-- 
Regards,
Oleksij