search

raspberrypi / linux

Kernel source tree for Raspberry Pi-provided kernel builds. Issues unrelated to the linux kernel should be posted on the community forum at https://forums.raspberrypi.com/
Other
11k stars 4.95k forks source link

dwc2 device mode causes enumeration errors if the udc is not bound to a gadget #2684

Open ali1234 opened 5 years ago

ali1234 commented 5 years ago

To reproduce, enable the dwc2 overlay but don't configure any modules to use in device mode. Boot the Pi and plug it into a Linux PC and you get this:

[168362.857069] usb 2-2: new high-speed USB device number 6 using xhci_hcd
[168368.153068] usb 2-2: device descriptor read/64, error -110
[168383.769030] usb 2-2: device descriptor read/64, error -110
[168384.005014] usb 2-2: new high-speed USB device number 7 using xhci_hcd
[168389.145021] usb 2-2: device descriptor read/64, error -110
[168404.760992] usb 2-2: device descriptor read/64, error -110
[168404.869037] usb usb2-port2: attempt power cycle
[168405.520966] usb 2-2: new high-speed USB device number 8 using xhci_hcd
[168410.776991] xhci_hcd 0000:00:14.0: Timeout while waiting for setup device command
[168416.152996] xhci_hcd 0000:00:14.0: Timeout while waiting for setup device command
[168416.360932] usb 2-2: device not accepting address 8, error -62
[168416.488943] usb 2-2: new high-speed USB device number 9 using xhci_hcd
[168421.528973] xhci_hcd 0000:00:14.0: Timeout while waiting for setup device command
[168426.904948] xhci_hcd 0000:00:14.0: Timeout while waiting for setup device command
[168427.112924] usb 2-2: device not accepting address 9, error -62
[168427.113041] usb usb2-port2: unable to enumerate USB device

...and if you are very unlucky, all the USB ports on your PC will get disabled and you will have to power cycle it to make them work again.

Using gadgetfs it is possible to bind a gadget to the UDC from userspace, at run time. It takes a while to reach this state, during which time the above errors will be seen on the host.

If a driver is bound to the UDC and then unbound, then Pi correctly signals a disconnect to the host like this:

[169017.304085] usb 2-2: USB disconnect, device number 12

Once in this state the host will not attempt to enumerate it and it will not cause errors.

If you then bind the UDC again, it will correctly enumerate on the host.

So it seems to me that on first boot the dwc2 hardware is in an incorrect state where it signals to the host that a device is connected, but then it does not respond to enumeration because there is nothing bound to it. It should instead be put into the disconnected state until such time as it gets bound to a driver.

I have two theories as to why this could be the case: the firmware has a USB device mode boot method and may be leaving the dwc2 controller in the wrong state, or the Linux kernel driver may simply be incorrectly initializing the hardware state (or not initializing it at all).

lategoodbye commented 5 years ago

Do you use the Raspbian Kernel 4.14? Are you able to reproduce this issue with 4.18?

ali1234 commented 5 years ago

Tested with Raspbian kernel (4.14.50+) rpi-update kernel (4.14.69+), and 'BRANCH=next rpi-update' (4.14.17+). They all behave the same.

I am building a 4.18 kernel to test now.

ali1234 commented 5 years ago

I have now tested 4.18.7+ and it is the same.

lategoodbye commented 5 years ago

Please report this issue to the following: Minas Harutyunyan hminas@synopsys.com Felipe Balbi balbi@kernel.org Greg Kroah-Hartman gregkh@linuxfoundation.org linux-usb@vger.kernel.org

ali1234 commented 5 years ago

I didn't get around to reporting this upstream yet - I will need to test upstream kernel before doing that.

However someone reported the same issue on IRC today when using the gpioexpander firmware.

It occured to me that this may be a firmware bug, if the firmware is leaving the USB hardware in an odd state. Although technically Linux should handle that, it may not be quick enough if it needs to unpack a large initrd or something.

JamesH65 commented 5 years ago

Please try the latest release kernel and report back results.

This issue will be closed within 30 days unless further interactions are posted. If you wish this issue to remain open, please add a comment. A closed issue may be reopened if requested.

ali1234 commented 5 years ago

Please name a commit that you want testing, and include the reason why you think it fixes the issue.

JamesH65 commented 5 years ago

The very latest release kernel (from apt), just to ensure the issue is either still there, nor not.

ali1234 commented 5 years ago

It still happens on Linux buster 4.19.57+ #1244 Thu Jul 4 18:42:50 BST 2019 armv6l

JamesH65 commented 5 years ago

OK, thanks. will leave open, although I doubt we will have much time to look at this one.