Re: FUSB200 xhci issue

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

Am 27.07.2013 23:59, schrieb Christian Lamparter:
> On Wednesday, July 24, 2013 12:37:55 PM Oleksij Rempel wrote:
>> Am 23.07.2013 20:26, schrieb Christian Lamparter:
>>> On Tuesday, July 23, 2013 06:59:46 AM Oleksij Rempel wrote:
>>>> Am 22.07.2013 23:23, schrieb Christian Lamparter:
>>>>> On Monday, July 22, 2013 10:47:41 PM Oleksij Rempel wrote:
>>>>>> Am 22.07.2013 21:54, schrieb Christian Lamparter:
>>>>>>> Hello!
>>>>>>>
>>>>>>> On Monday, July 22, 2013 05:21:54 PM Oleksij Rempel wrote:
>>>>>>>> i'm one of ath9k_htc devs. Currently i'm working on usb_suspend issue of
>>>>>>>> this adapters. Looks like ar9271 and ar7010 have FUSB200, and i
>>>>>>>> accidentally discovered that 9170 have it too. Are there any issue with
>>>>>>>> usb-suspend + xhci controllers by you? Did you some how specially
>>>>>>>> handled it?
>>>>>>>
>>>>>>> No, I haven't heard any complains about xhci + suspend. In fact,
>>>>>>> it's working fine with the NEC xhci I have. I also have a AR9271
>>>>>>> and AR7010, so if you want I could try if they survive a suspend
>>>>>>> +resume cycle when attached.
>>>>>>>
>>>>>>> But, I do have a bug-report from someone else who has/had? problems
>>>>>>> with carl9170 and xhci. If you want, you can get the details from:
>>>>>>> "carl9170 A-MPDU transmit problem":
>>>>>>> <http://comments.gmane.org/gmane.linux.kernel.wireless.general/104597>
>>>>>>>
>>>>>>> The likely cause is related to Intel's xhci silicon (Ivy Bridge is
>>>>>>> affected, but I don't know about Haswell):
>>>>>>> <http://permalink.gmane.org/gmane.linux.kernel.wireless.general/104602>
>>>>>>
>>>>>> Same situation is here - i have problem on Ivy Bridge.
>>>>> (Note: I don't have any Ivy Bridge system. Just Sandy Bridge
>>>>> and AMD's new A6-1450 Temash and both xhci work. So I can't
>>>>> do any proper comparisons like you.)
>>>>>
>>>>>> Steps to reproduce:
>>>>>> - plug adapter. Module and firmware will be loaded
>>>>>> - make sure usb autosupend is enabled. By default it is not! Use
>>>>>> powertop or directly sysfs to enable autosuspend for this device
>>>>>> - rmmod .... and wait some seconds until adapter is suspended and then
>>>>>> modprobe ath9k_htc
>>>>
>>>>>> first packet which is bigger as 64Byte will kill EP4 FIFO. Size register
>>>>>> will report wrong size.. bigger as FIFO can handle. And only first ~40
>>>>>> readet bytes will be actually OK.. all the rest of packet will be trashed.
>>>>>
>>>>> This is what happens with a WN721 (ar9271):
>>>>>
>>>>> [17619.597905] usbcore: deregistering interface driver ath9k_htc
>>>>> [17619.679549] usb 2-2: ath9k_htc: USB layer deinitialized
>>>>> [17619.679602] ath9k_htc: Driver unloaded
>>>>> <suspend>
>>>>>
>>>>> <resume>
>>>>> [17667.543024] usb 2-2: reset high-speed USB device number 3 using xhci_hcd <----
>>>>> [17667.572168] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa77600
>>>>> [17667.572174] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa77640
>>>>> [17667.572177] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa77680
>>>>> [17667.572180] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa776c0
>>>>> [17667.572183] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa77700
>>>>> [17667.572185] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88031aa77740
>>>>> [17667.573826] usb 2-2: ath9k_htc: Firmware htc_9271.fw requested
>>>>> [17667.573873] usbcore: registered new interface driver ath9k_htc
>>>>> [17668.038200] usb 2-2: ath9k_htc: Transferred FW: htc_9271.fw, size: 51272
>>>>> [17668.273249] ath9k_htc 2-2:1.0: ath9k_htc: HTC initialized with 33 credits
>>>>>
>>>>> The driver loads, but there's no "wlanX" interface, no phyX interface
>>>>> and the driver can't be unloaded due to "Error: Module ath9k_htc is in use".
>>>>
>>>> So, it is exactly this bug.
>>>> After firmware was loaded ath9k trying to set some registers. Since
>>>> command buffer is trashed it will write some nonsense to registers and
>>>> some times in wrong to wrong addresses. I have some patches to do crc
>>>> check of commands, to easy detect corrupted ones.
>>>>
>>>> You reproduced it on NEC xhci? Is it possible to reproduce it in
>>>> carl9170?
>>>
>>> Ok: That's dmesg of your scenario:
>>>
>>> [  809.995966] usbcore: deregistering interface driver carl9170
>>> <suspend>
>>>
>>> <resume>
>>> [  826.365596] usb 2-2: reset high-speed USB device number 2 using xhci_hcd
>>> [  826.431154] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88045af2b900
>>> [  826.431159] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88045af2b940
>>> [  826.431162] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88045af2b980
>>> [  826.431164] xhci_hcd 0000:19:00.0: xHCI xhci_drop_endpoint called with disabled ep ffff88045af2b9c0
>>> [  826.432257] usbcore: registered new interface driver carl9170
>>> [  826.433717] usb 2-2: driver   API: 1.9.8 2013-03-23 [1-1]
>>> ...
>>> [  826.816110] usb 2-2: Atheros AR9170 is registered as 'phy3'
>>>
>>> carl9170 is able to recover and the stick is working after autosuspend cycle.
>>
>> Thank you for your tests. USB configuration and handlers look similar on
>> this two firmwares. May be problem is not in FUSB200 and it is clock
>> related issue so carl9170 do not need reset. - I don't know :(
>>
>> Can you please test this branch:
>> https://github.com/olerem/open-ath9k-htc-firmware/commits/next
>> There is a workaround to reset adapter, right after this bug is
>> detected. It is really dirty hack, but currently i do not know how to
>> fix this bug on xhci or on ath9k-htc side.
>>
> Is it still Saturday?
>
> Anyway, I tried the -next branch.
>
> commit dbbb809d592dde0b3c9ecb97b3b387ff8e40e799
> Author: Oleksij Rempel <linux@rempel-privat.de>
> Date:   Wed Jul 24 10:26:18 2013 +0200
>
>      k2_fw_usb_api: workaround for EP4 bug.
>
> but still, the device won't show up after autosuspend.

Hm... firmware probably didn't rebooted before suspend. Did interface 
was up, before autosuspend? If no, you need latest wireles-testing - 
there are patches to handle this issue. Or just make "ifconfig wlan1 up" 
  before  rmmod.
If it was up... then i would need last log messages from firmware.. If 
there is some way to attach to uart pins on it.

-- 
Regards,
Oleksij