Re: How to set USB_PORT_QUIRK_OLD_SCHEME on an usb-port ?

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 9/17/20 10:09 PM, Alan Stern wrote:
> On Thu, Sep 17, 2020 at 07:27:03PM +0200, Hans de Goede wrote:
>> Hi,
>>
>> On 9/10/20 5:41 PM, Alan Stern wrote:
>>> On Thu, Sep 10, 2020 at 03:58:49PM +0200, Hans de Goede wrote:
>>>> Hi,
>>>>
>>>> On 9/6/20 4:22 AM, Alan Stern wrote:
>>>>> On Sat, Sep 05, 2020 at 01:37:38PM +0200, Hans de Goede wrote:
>>>>>> Hi All,
>>>>>>
>>>>>> I have been debugging an issue with a 2-in-1 which
>>>>>> consists of a tablet + a kbd-dock, where the device
>>>>>> turns into a clamshell when docked into the kbd-dock.
>>>>>>
>>>>>> The kbd dock is connected via pogo-pins. This works
>>>>>> fine when docked at boot. But there is an enumeration
>>>>>> issue when hot-docked (and the keyboard looses power
>>>>>> when the lid is closedm so this also triggers after
>>>>>> a suspend/resume):
>>>>>>
>>>>>> [ 3498.924190] usb 1-3: new full-speed USB device number 5 using xhci_hcd
>>>>>> [ 3499.041725] usb 1-3: device descriptor read/64, error -71
>>>>>> [ 3515.215890] usb 1-3: device descriptor read/64, error -110
>>>>>> [ 3515.440369] usb 1-3: new full-speed USB device number 6 using xhci_hcd
>>>>>> [ 3515.603544] usb 1-3: New USB device found, idVendor=06cb, idProduct=73f5, bcdDevice= 0.02
>>>>>> [ 3515.603574] usb 1-3: New USB device strings: Mfr=1, Product=2, SerialNumber=0
>>>>>> [ 3515.603596] usb 1-3: Product: ITE Device(8910)
>>>>>> [ 3515.603614] usb 1-3: Manufacturer: ITE Tech. Inc.
>>>>>>
>>>>>> Note there is about 6 seconds before the keyboard becomes
>>>>>> usable, which is quite long when trying to unlock the
>>>>>> laptop after opening the lid.
>>>>>>
>>>>>> If I set the USB_PORT_QUIRK_OLD_SCHEME on the port used by the kbd-dock:
>>>>>>
>>>>>> echo 1 >  /sys/devices/pci0000\:00/0000\:00\:14.0/usb1/1-0\:1.0/usb1-port3/quirks
>>>>>>
>>>>>> Then this changes to:
>>>>>>
>>>>>> [ 4467.875008] usb 1-3: new full-speed USB device number 7 using xhci_hcd
>>>>>> [ 4467.878483] usb 1-3: Device not responding to setup address.
>>>>>> [ 4468.082476] usb 1-3: Device not responding to setup address.
>>>>>> [ 4468.289990] usb 1-3: device not accepting address 7, error -71
>>>>>> [ 4468.614928] usb 1-3: new full-speed USB device number 8 using xhci_hcd
>>>>>> [ 4468.662392] usb 1-3: New USB device found, idVendor=06cb, idProduct=73f5, bcdDevice= 0.02
>>>>>> [ 4468.662423] usb 1-3: New USB device strings: Mfr=1, Product=2, SerialNumber=0
>>>>>> [ 4468.662444] usb 1-3: Product: ITE Device(8910)
>>>>>> [ 4468.662461] usb 1-3: Manufacturer: ITE Tech. Inc.
>>>>>>
>>>>>> Which is a lot better wrt making the keyboard available for
>>>>>> use in a timely manner.
>>>>>>
>>>>>> So now I'm looking into a way to automatically do this. I would
>>>>>> prefer to keep the handling of this out of the kernel, so I looked into
>>>>>> udev, but it seems that the usb_port_device_type device-s registered by
>>>>>> usb_hub_create_port_device() are not visible to udev?
>>>>>>
>>>>>> At least I'm not seeing them, in the output of "udevadm info -e"
>>>>>
>>>>> My impression is that fixing this would be the simplest approach.
>>>>
>>>> I agree that first trying to fix it is a good idea.
>>
>> <snip>
>>
>>>>> Have you tried decreasing the initial_descriptor_timeout module
>>>>> parameter for usbcore?  That would probably help, but it's kind of a
>>>>> sledgehammer.
>>>>
>>>> So I tried this:
>>>>
>>>> [root@localhost ~]# cat /sys/module/usbcore/parameters/initial_descriptor_timeout
>>>> 1000
>>>>
>>>> But it does not really seem to help:
>>>>
>>>> [ 1171.435346] usb 1-3: USB disconnect, device number 2
>>>> [ 1180.430958] usb 1-3: new full-speed USB device number 3 using xhci_hcd
>>>> [ 1180.551543] usb 1-3: device descriptor read/64, error -71
>>>> [ 1184.045548] usb 1-3: device descriptor read/64, error -110
>>>> [ 1184.270924] usb 1-3: new full-speed USB device number 4 using xhci_hcd
>>>> [ 1184.438336] usb 1-3: New USB device found, idVendor=06cb, idProduct=73f5, bcdDevice= 0.02
>>>> [ 1184.438371] usb 1-3: New USB device strings: Mfr=1, Product=2, SerialNumber=0
>>>> [ 1184.438399] usb 1-3: Product: ITE Device(8910)
>>>> [ 1184.438425] usb 1-3: Manufacturer: ITE Tech. Inc.
>>>
>>> That part of hub.c is a real mess.  Coincidentally, I just posted a
>>> patch to try and fix it up a bit:
>>>
>>> 	https://marc.info/?l=linux-usb&m=159959185227447&w=2
>>
>> Ok, I've applied the patch and tried again (with initial_descriptor_timeout still set to 1000),
>> so now the log from the reconnect looks look this:
>>
>> [ 1157.248439] usb 1-3: new full-speed USB device number 3 using xhci_hcd
>> [ 1157.254234] usb 1-3: device descriptor read/64, error -71
>> [ 1157.371442] usb 1-3: new full-speed USB device number 4 using xhci_hcd
>> [ 1158.377895] usb 1-3: device descriptor read/64, error -110
>> [ 1158.379646] usb usb1-port3: attempt power cycle
>> [ 1159.014480] usb 1-3: new full-speed USB device number 5 using xhci_hcd
>> [ 1159.176112] usb 1-3: New USB device found, idVendor=06cb, idProduct=73f5, bcdDevice= 0.02
>> [ 1159.176138] usb 1-3: New USB device strings: Mfr=1, Product=2, SerialNumber=0
>> [ 1159.176156] usb 1-3: Product: ITE Device(8910)
>> [ 1159.176172] usb 1-3: Manufacturer: ITE Tech. Inc.
>>
>> Still a bit slower then the old probe method, but much better then the
>> new probe method with the default initial_descriptor_timeout.
> 
> Yeah, okay, it's good to see that the patch helps.  But I'm doubtful
> that the change it makes will become part of the standard (i.e., not
> for embedded systems) kernel.
> 
> I still think the udev approach will be best.  That will require adding
> various *_uevent_* calls in usb_hub_create_port_device, and adding a
> .uevent member to usb_port_device_type.

So I tried this and it does not work, the problem is that
dev_uevent_filter() from drivers/base/core.c
filters out uevents for anything which is not either a device
on a bus or a class device:

static int dev_uevent_filter(struct kset *kset, struct kobject *kobj)
{
         struct kobj_type *ktype = get_ktype(kobj);

         if (ktype == &device_ktype) {
                 struct device *dev = kobj_to_dev(kobj);
                 if (dev->bus)
                         return 1;
                 if (dev->class)
                         return 1;
         }
         return 0;
}

(as mentioned this is a low priority thing for me, so hence
  the long time between replies)

Regards,

Hans