Any example of USB gadget for DRD device mode on Intel Gemini Lake?

Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Dmitry N. Mikushin]

Dear All,

I'm confused by the USB gadget mode on the recent Intel SoCs, such as
Gemini Lake. The /sys/class/udc is empty, and a SoC can't present
itself as a IoT gadget this way, yet definitely being designed as
such. I've noticed the concept of dual-role-device, which seems to
replace the OTG. Particularly, Harry Pan mentioned that Gemini Lake
supports DRD in port 0.

Indeed, if I do:

sudo sh -c "echo device >/sys/class/usb_role/intel_xhci_usb_sw-role-switch/role"


then one of the ports gets successfully recognized as a new USB
device, if connected with a cable to another laptop.

But I'm wondering, how exactly could I construct a logical device
class for this case, e.g. similarly to uvc gadget for OTG? Completely
lacking any example of code, currently it's a big confusion.

Thanks & Kind regards,
- Dmitry.

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Heikki Krogerus]

On Mon, Oct 19, 2020 at 01:27:35PM +0200, Dmitry N. Mikushin wrote:
> Dear All,
> 
> I'm confused by the USB gadget mode on the recent Intel SoCs, such as
> Gemini Lake. The /sys/class/udc is empty, and a SoC can't present
> itself as a IoT gadget this way, yet definitely being designed as
> such. I've noticed the concept of dual-role-device, which seems to
> replace the OTG. Particularly, Harry Pan mentioned that Gemini Lake
> supports DRD in port 0.

So do you have the dwc3 (the USB device controller) PCI device
available/visible on your system? What do you get if you run:

        lspci -nn|grep USB

The DWC3 PCI device ID on Gemini lake is 0x31aa (search
PCI_DEVICE_ID_INTEL_GLK in drivers/usb/dwc3/dwc3-pci.c).

Which board/product are you using.
Which kernel are you using?
Is the dwc3 driver enabled in your kernel?
Which gadget are you testing with? g_zero?

thanks,

-- 
heikki

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Dmitry N. Mikushin]

Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:

00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)

Board is M1K.
My kernel is 5.4.0-48-lowlatency with the following additional patch:
https://lore.kernel.org/patchwork/patch/1037542/
I have dwc3 as a kernel module.
I'm not testing any gadget, I just have no idea what to test. Could
you please suggest some sources on how I should map gadget onto my
dwc3 port?

Thanks & Kind Regards,
- Dmitry.

пн, 19 окт. 2020 г. в 15:36, Heikki Krogerus <heikki.krogerus@linux.intel.com>:
>
> On Mon, Oct 19, 2020 at 01:27:35PM +0200, Dmitry N. Mikushin wrote:
> > Dear All,
> >
> > I'm confused by the USB gadget mode on the recent Intel SoCs, such as
> > Gemini Lake. The /sys/class/udc is empty, and a SoC can't present
> > itself as a IoT gadget this way, yet definitely being designed as
> > such. I've noticed the concept of dual-role-device, which seems to
> > replace the OTG. Particularly, Harry Pan mentioned that Gemini Lake
> > supports DRD in port 0.
>
> So do you have the dwc3 (the USB device controller) PCI device
> available/visible on your system? What do you get if you run:
>
>         lspci -nn|grep USB
>
> The DWC3 PCI device ID on Gemini lake is 0x31aa (search
> PCI_DEVICE_ID_INTEL_GLK in drivers/usb/dwc3/dwc3-pci.c).
>
> Which board/product are you using.
> Which kernel are you using?
> Is the dwc3 driver enabled in your kernel?
> Which gadget are you testing with? g_zero?
>
> thanks,
>
> --
> heikki

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Sid Spry]

On Thu, Oct 22, 2020, at 4:06 PM, Dmitry N. Mikushin wrote:
> Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> 
> 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
> 
> Board is M1K.
> My kernel is 5.4.0-48-lowlatency with the following additional patch:
> https://lore.kernel.org/patchwork/patch/1037542/
> I have dwc3 as a kernel module.
> I'm not testing any gadget, I just have no idea what to test. Could
> you please suggest some sources on how I should map gadget onto my
> dwc3 port?
> 
> Thanks & Kind Regards,
> - Dmitry.
> 

After you set the DRD switch to device do you see a device in /sys/class/udc?

I will show you how to use configfs. You must enable USB gadget configfs. It
is easier to use in my opinion. The other gadget drivers simply need to be
loaded, but I am unsure how they work with newer DWC3 hardware. Most
of the code is written for android and android uses configfs.

`modprobe libcomposite` then execute the following, using that name:

#!/usr/bin/env bash
cd /sys/kernel/config/usb_gadget
if [[ -d "g0" ]]; then
        echo "" > g0/UDC
        rm -rf g0;
fi

mkdir g0
cd g0

echo "0x1d6b" > idVendor
echo "0x0104" > idProduct

mkdir strings/0x409
echo "0000000000" > strings/0x409/serialnumber
echo "Your Company" > strings/0x409/manufacturer
echo "Your Device" > strings/0x409/product

mkdir functions/ncm.usb0

mkdir configs/c.1
ln -s functions/ncm.usb0 configs/c.1

# List /sys/class/udc and take a name for this line:
echo "" > UDC

This will set up a Ethernet NCM device. There are similar instructions
from the configfs author in a powerpoint. More device types are in the
documentation.

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Dmitry N. Mikushin]

Hi Sid, alas no device in udc, what can I do?

marcusmae@m1k:~$ cat /sys/class/usb_role/intel_xhci_usb_sw-role-switch/role
device
marcusmae@m1k:~$ ls -l /sys/class/udc
total 0


чт, 22 окт. 2020 г. в 23:26, Sid Spry <sid@aeam.us>:
>
> On Thu, Oct 22, 2020, at 4:06 PM, Dmitry N. Mikushin wrote:
> > Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> >
> > 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
> >
> > Board is M1K.
> > My kernel is 5.4.0-48-lowlatency with the following additional patch:
> > https://lore.kernel.org/patchwork/patch/1037542/
> > I have dwc3 as a kernel module.
> > I'm not testing any gadget, I just have no idea what to test. Could
> > you please suggest some sources on how I should map gadget onto my
> > dwc3 port?
> >
> > Thanks & Kind Regards,
> > - Dmitry.
> >
>
> After you set the DRD switch to device do you see a device in /sys/class/udc?
>
> I will show you how to use configfs. You must enable USB gadget configfs. It
> is easier to use in my opinion. The other gadget drivers simply need to be
> loaded, but I am unsure how they work with newer DWC3 hardware. Most
> of the code is written for android and android uses configfs.
>
> `modprobe libcomposite` then execute the following, using that name:
>
> #!/usr/bin/env bash
> cd /sys/kernel/config/usb_gadget
> if [[ -d "g0" ]]; then
>         echo "" > g0/UDC
>         rm -rf g0;
> fi
>
> mkdir g0
> cd g0
>
> echo "0x1d6b" > idVendor
> echo "0x0104" > idProduct
>
> mkdir strings/0x409
> echo "0000000000" > strings/0x409/serialnumber
> echo "Your Company" > strings/0x409/manufacturer
> echo "Your Device" > strings/0x409/product
>
> mkdir functions/ncm.usb0
>
> mkdir configs/c.1
> ln -s functions/ncm.usb0 configs/c.1
>
> # List /sys/class/udc and take a name for this line:
> echo "" > UDC
>
> This will set up a Ethernet NCM device. There are similar instructions
> from the configfs author in a powerpoint. More device types are in the
> documentation.

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Sid Spry]

On Thu, Oct 22, 2020, at 4:51 PM, Dmitry N. Mikushin wrote:
> Hi Sid, alas no device in udc, what can I do?
> 
> marcusmae@m1k:~$ cat /sys/class/usb_role/intel_xhci_usb_sw-role-switch/role
> device
> marcusmae@m1k:~$ ls -l /sys/class/udc
> total 0

I don't have an exact answer, but the device in /sys/class/udc must be there
(as far as I am aware) if the kernel is properly communicating with the hardware.

On my ARM systems I've found that DRD seems to not work in general, and
I've had to set device mode operation via devicetree. On an Intel board I
assume this would be a UEFI/BIOS option.

If Intel is still providing support for the hardware you should try to contact
them. In the meantime, hopefully someone else can comment.

> 
> чт, 22 окт. 2020 г. в 23:26, Sid Spry <sid@aeam.us>:
> >
> > On Thu, Oct 22, 2020, at 4:06 PM, Dmitry N. Mikushin wrote:
> > > Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> > >
> > > 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
> > >
> > > Board is M1K.
> > > My kernel is 5.4.0-48-lowlatency with the following additional patch:
> > > https://lore.kernel.org/patchwork/patch/1037542/
> > > I have dwc3 as a kernel module.
> > > I'm not testing any gadget, I just have no idea what to test. Could
> > > you please suggest some sources on how I should map gadget onto my
> > > dwc3 port?
> > >
> > > Thanks & Kind Regards,
> > > - Dmitry.
> > >
> >
> > After you set the DRD switch to device do you see a device in /sys/class/udc?
> >
> > I will show you how to use configfs. You must enable USB gadget configfs. It
> > is easier to use in my opinion. The other gadget drivers simply need to be
> > loaded, but I am unsure how they work with newer DWC3 hardware. Most
> > of the code is written for android and android uses configfs.
> >
> > `modprobe libcomposite` then execute the following, using that name:
> >
> > #!/usr/bin/env bash
> > cd /sys/kernel/config/usb_gadget
> > if [[ -d "g0" ]]; then
> >         echo "" > g0/UDC
> >         rm -rf g0;
> > fi
> >
> > mkdir g0
> > cd g0
> >
> > echo "0x1d6b" > idVendor
> > echo "0x0104" > idProduct
> >
> > mkdir strings/0x409
> > echo "0000000000" > strings/0x409/serialnumber
> > echo "Your Company" > strings/0x409/manufacturer
> > echo "Your Device" > strings/0x409/product
> >
> > mkdir functions/ncm.usb0
> >
> > mkdir configs/c.1
> > ln -s functions/ncm.usb0 configs/c.1
> >
> > # List /sys/class/udc and take a name for this line:
> > echo "" > UDC
> >
> > This will set up a Ethernet NCM device. There are similar instructions
> > from the configfs author in a powerpoint. More device types are in the
> > documentation.
>

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Dmitry N. Mikushin]

This is exactly the purpose of my message here. Intel's approach to
UDC is different to what we've been used up to in ARMs. No USB options
are provided in BIOS; actually BIOS is almost empty :) Indeed, I
really hope to get this clarified with the help of Intel folks.

Kind regards,
- Dmitry.

пт, 23 окт. 2020 г. в 01:05, Sid Spry <sid@aeam.us>:
>
> On Thu, Oct 22, 2020, at 4:51 PM, Dmitry N. Mikushin wrote:
> > Hi Sid, alas no device in udc, what can I do?
> >
> > marcusmae@m1k:~$ cat /sys/class/usb_role/intel_xhci_usb_sw-role-switch/role
> > device
> > marcusmae@m1k:~$ ls -l /sys/class/udc
> > total 0
>
> I don't have an exact answer, but the device in /sys/class/udc must be there
> (as far as I am aware) if the kernel is properly communicating with the hardware.
>
> On my ARM systems I've found that DRD seems to not work in general, and
> I've had to set device mode operation via devicetree. On an Intel board I
> assume this would be a UEFI/BIOS option.
>
> If Intel is still providing support for the hardware you should try to contact
> them. In the meantime, hopefully someone else can comment.
>
> >
> > чт, 22 окт. 2020 г. в 23:26, Sid Spry <sid@aeam.us>:
> > >
> > > On Thu, Oct 22, 2020, at 4:06 PM, Dmitry N. Mikushin wrote:
> > > > Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> > > >
> > > > 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
> > > >
> > > > Board is M1K.
> > > > My kernel is 5.4.0-48-lowlatency with the following additional patch:
> > > > https://lore.kernel.org/patchwork/patch/1037542/
> > > > I have dwc3 as a kernel module.
> > > > I'm not testing any gadget, I just have no idea what to test. Could
> > > > you please suggest some sources on how I should map gadget onto my
> > > > dwc3 port?
> > > >
> > > > Thanks & Kind Regards,
> > > > - Dmitry.
> > > >
> > >
> > > After you set the DRD switch to device do you see a device in /sys/class/udc?
> > >
> > > I will show you how to use configfs. You must enable USB gadget configfs. It
> > > is easier to use in my opinion. The other gadget drivers simply need to be
> > > loaded, but I am unsure how they work with newer DWC3 hardware. Most
> > > of the code is written for android and android uses configfs.
> > >
> > > `modprobe libcomposite` then execute the following, using that name:
> > >
> > > #!/usr/bin/env bash
> > > cd /sys/kernel/config/usb_gadget
> > > if [[ -d "g0" ]]; then
> > >         echo "" > g0/UDC
> > >         rm -rf g0;
> > > fi
> > >
> > > mkdir g0
> > > cd g0
> > >
> > > echo "0x1d6b" > idVendor
> > > echo "0x0104" > idProduct
> > >
> > > mkdir strings/0x409
> > > echo "0000000000" > strings/0x409/serialnumber
> > > echo "Your Company" > strings/0x409/manufacturer
> > > echo "Your Device" > strings/0x409/product
> > >
> > > mkdir functions/ncm.usb0
> > >
> > > mkdir configs/c.1
> > > ln -s functions/ncm.usb0 configs/c.1
> > >
> > > # List /sys/class/udc and take a name for this line:
> > > echo "" > UDC
> > >
> > > This will set up a Ethernet NCM device. There are similar instructions
> > > from the configfs author in a powerpoint. More device types are in the
> > > documentation.
> >

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Heikki Krogerus]

On Thu, Oct 22, 2020 at 11:06:59PM +0200, Dmitry N. Mikushin wrote:
> Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> 
> 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)

That is the xHCI controller, and it is not what you need if you want
to use the connector in device mode. The xHCI and DWC3 IPs are
separate IPs on GLK. That is why there is a mux between the two. The
dwc3 USB device controller has device ID 31aa, so you want to see a
PCI device with the device ID. It's not there.

So the dwc3 PCI device is not enabled on your board, which means you
do not have USB device controller to deal with. The connector is in
host mode only. Sorry.

If you can enter the BIOS menu, then you can try to find an option
named XDCI (so that's "XDCI" not "xHCI"). It is usually somewhere
under some USB menu. If you have that, then enable it, and you should
see the dwc3 PCI device in the operating system.

Br,

-- 
heikki

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Felipe Balbi]

[-- Attachment #1: Type: text/plain, Size: 1231 bytes --]

Heikki Krogerus <heikki.krogerus@linux.intel.com> writes:

> On Thu, Oct 22, 2020 at 11:06:59PM +0200, Dmitry N. Mikushin wrote:
>> Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
>> 
>> 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
>
> That is the xHCI controller, and it is not what you need if you want
> to use the connector in device mode. The xHCI and DWC3 IPs are
> separate IPs on GLK. That is why there is a mux between the two. The
> dwc3 USB device controller has device ID 31aa, so you want to see a
> PCI device with the device ID. It's not there.
>
> So the dwc3 PCI device is not enabled on your board, which means you
> do not have USB device controller to deal with. The connector is in
> host mode only. Sorry.
>
> If you can enter the BIOS menu, then you can try to find an option
> named XDCI (so that's "XDCI" not "xHCI"). It is usually somewhere
> under some USB menu. If you have that, then enable it, and you should
> see the dwc3 PCI device in the operating system.

Also, have a look at acpidump. See if the device even exists in your
DSDT but, perhaps, disabled (look at the _STA method for OTDG or XDCI)

-- 
balbi

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 857 bytes --]

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Dmitry N. Mikushin]

This is what I get from acpidump:

marcusmae@m1k:~$ sudo acpidump | grep OTDG
marcusmae@m1k:~$ sudo acpidump | grep XDCI
    1AA0: 5F 41 44 52 0C 00 00 15 00 5B 82 0F 58 44 43 49  _ADR.....[..XDCI
    3160: 42 38 58 44 43 49 14 1A 5F 50 4C 44 00 A4 5E 5E  B8XDCI.._PLD..^^
    31B0: A0 0E 90 50 4D 45 45 60 86 58 44 43 49 0A 02 5B  ...PMEE`.XDCI..[
    3200: 44 44 4E 0D 42 72 6F 78 74 6F 6E 20 58 44 43 49  DDN.Broxton XDCI
    B170: 00 5C 2F 04 5F 53 42 5F 50 43 49 30 58 44 43 49  .\/._SB_PCI0XDCI

Does this look sufficient?

Kind regards,
- Dmitry.

пт, 23 окт. 2020 г. в 12:09, Felipe Balbi <balbi@kernel.org>:
>
> Heikki Krogerus <heikki.krogerus@linux.intel.com> writes:
>
> > On Thu, Oct 22, 2020 at 11:06:59PM +0200, Dmitry N. Mikushin wrote:
> >> Yes, AFAIK PCI_DEVICE_ID_INTEL_GLK_XHCI is actually 0x31a8, and I do have it:
> >>
> >> 00:15.0 USB controller [0c03]: Intel Corporation Device [8086:31a8] (rev 03)
> >
> > That is the xHCI controller, and it is not what you need if you want
> > to use the connector in device mode. The xHCI and DWC3 IPs are
> > separate IPs on GLK. That is why there is a mux between the two. The
> > dwc3 USB device controller has device ID 31aa, so you want to see a
> > PCI device with the device ID. It's not there.
> >
> > So the dwc3 PCI device is not enabled on your board, which means you
> > do not have USB device controller to deal with. The connector is in
> > host mode only. Sorry.
> >
> > If you can enter the BIOS menu, then you can try to find an option
> > named XDCI (so that's "XDCI" not "xHCI"). It is usually somewhere
> > under some USB menu. If you have that, then enable it, and you should
> > see the dwc3 PCI device in the operating system.
>
> Also, have a look at acpidump. See if the device even exists in your
> DSDT but, perhaps, disabled (look at the _STA method for OTDG or XDCI)
>
> --
> balbi

Re: Any example of USB gadget for DRD device mode on Intel Gemini Lake?

[Felipe Balbi]

[-- Attachment #1: Type: text/plain, Size: 768 bytes --]


Hi,

(please avoid top-posting)

"Dmitry N. Mikushin" <maemarcus@gmail.com> writes:
> This is what I get from acpidump:
>
> marcusmae@m1k:~$ sudo acpidump | grep OTDG
> marcusmae@m1k:~$ sudo acpidump | grep XDCI
>     1AA0: 5F 41 44 52 0C 00 00 15 00 5B 82 0F 58 44 43 49  _ADR.....[..XDCI
>     3160: 42 38 58 44 43 49 14 1A 5F 50 4C 44 00 A4 5E 5E  B8XDCI.._PLD..^^
>     31B0: A0 0E 90 50 4D 45 45 60 86 58 44 43 49 0A 02 5B  ...PMEE`.XDCI..[
>     3200: 44 44 4E 0D 42 72 6F 78 74 6F 6E 20 58 44 43 49  DDN.Broxton XDCI
>     B170: 00 5C 2F 04 5F 53 42 5F 50 43 49 30 58 44 43 49  .\/._SB_PCI0XDCI
>
> Does this look sufficient?

you need to decompile it to look at the sources. Have a look at the help
for acpidump and iasl ;-)

-- 
balbi

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 857 bytes --]