USB device mode do not work

[YuukiHogo]

In branch allwinner-v3.0-android-v2 USB gadget modules modprobe without issues, but usbc do not switch controller to client mode - upon connection to USB host it doesn't detect anything.
dmesg:

[  613.650000] ------------IRQ DISCONNECT-------------
[  613.650000] 
[  613.660000] usb 1-1: USB disconnect, device number 2
[  613.760000] [sw_hcd]: sw_hcd_h_disable, epnum = 0
[  613.760000] [sw_hcd]: sw_hcd_h_disable, epnum = 0
[  613.790000] [sw_hcd]: sw_hcd_h_disable, epnum = 1
[  613.790000] [sw_hcd]: sw_hcd_h_disable, epnum = 82
[  613.870000] 
[  613.870000] 
[  613.870000] rmmod_host_driver
[  613.870000] 
[  613.870000] [sw_hcd0]: sw_usb_host0_disable start
[  613.870000] -------sw_hcd0_soft_disconnect---------
[  613.870000] [sw_hcd_host0]: Set USB Power OFF
[  613.870000] sw_hcd_stop: sw_hcd disabled
[  613.940000] [sw_hcd0]: close_usb_clock
[  613.940000] [sw_hcd0]: sw_usb_host0_disable end
[  678.370000] survey done event(b)
[  678.540000] [sw_udc]: [sw_usb_udc]: binding gadget driver 'g_serial'
[  678.550000] [sw_udc]: alloc request: ep(0xc076de00, ep0, 64), req(0xe690ef00)
[  678.560000] gserial_setup: registered 1 ttyGS* device
[  678.560000] g_serial gadget: adding config #2 'CDC ACM config'/bf18b3a8
[  678.570000] g_serial gadget: adding 'acm0'/e72a2f00 to config 'CDC ACM config'/bf18b3a8
[  678.580000] ep_matches, wrn: endpoint already claimed, ep(0xc076de40, 0xe690ee00, ep1-bulk)
[  678.600000] ep_matches, wrn: endpoint already claimed, ep(0xc076de40, 0xe690ee00, ep1-bulk)
[  678.600000] ep_matches, wrn: endpoint already claimed, ep(0xc076de80, 0xe690ee00, ep2-bulk)
[  678.620000] 3333333
[  678.620000] 3333333
[  678.620000] [sw_udc]: alloc request: ep(0xc076df40, ep5-int, 512), req(0xe68e2980)
[  678.620000] g_serial gadget: acm ttyGS0: dual speed IN/ep1-bulk OUT/ep2-bulk NOTIFY/ep5-int
[  678.640000] g_serial gadget: cfg 2/bf18b3a8 speeds: high full
[  678.640000] g_serial gadget:   interface 0 = acm0/e72a2f00
[  678.640000] g_serial gadget:   interface 1 = acm0/e72a2f00
[  678.660000] g_serial gadget: Gadget Serial v2.4
[  678.660000] g_serial gadget: g_serial ready
[  701.810000] 
[  701.810000] 
[  701.810000] insmod_device_driver
[  701.810000] 
[  701.810000] [sw_udc]: sw_usb_device_enable start
[  701.810000] [sw_udc]: usb_vbase  = 0xf1c13000
[  701.810000] [sw_udc]: sram_vbase = 0xf1c00000
[  701.830000] [sw_udc]: open_usb_clock
[  701.870000] [sw_udc]: usbd_stop_work
[  701.870000] [sw_udc]: sw_usb_device_enable end

[amery]

what defconfig?

[YuukiHogo]

Using sun4i_defconfig with gadget-related modules.

[amery]

can you try with sun4i_crane_defconfig instead? Most of the gadget integration Allwinner did is sadly dependent on Android.

[YuukiHogo]

Just tried it. Hangs at

Starting kernel ...

and often self-reboots right after this message. No success boot until now with this config.

[YuukiHogo]

I made it to boot to user-space using sun4i_crane_defconfig. Plugging it to host device makes no log entries on host, although on device side there still all same

[   59.750000] android_work: did not send uevent (0 0   (null))
[   70.860000] 
[   70.860000] 
[   70.860000] insmod_device_driver
[   70.860000] 
[   70.860000] [sw_udc]: sw_usb_device_enable start
[   70.860000] [sw_udc]: usb_vbase  = 0xf1c13000
[   70.860000] [sw_udc]: sram_vbase = 0xf1c00000
[   70.860000] [sw_udc]: open_usb_clock
[   70.920000] [sw_udc]: usbd_stop_work
[   70.920000] [sw_udc]: sw_usb_device_enable end

[YuukiHogo]

OK. I found way to make it seem to work. Dunno is it right or wrong way, but now at least host get a bunch of events and logs about USB device. And detects it correctly.

diff --git a/drivers/usb/sun4i_usb/udc/sw_udc.c b/drivers/usb/sun4i_usb/udc/sw_udc.c
index 28741a2..8a58989 100644
--- a/drivers/usb/sun4i_usb/udc/sw_udc.c
+++ b/drivers/usb/sun4i_usb/udc/sw_udc.c
@@ -3279,7 +3279,7 @@ int sw_usb_device_enable(void)
                goto err;
        }

-       if(udc->driver && is_udc_enable){
+       if(udc->driver && !is_udc_enable){
                sw_udc_enable(udc);
                cfg_udc_command(SW_UDC_P_ENABLE);
        }

[amery]

@YuukiHogo can you check you have that usb controller enabled in your script.bin?

[YuukiHogo]

evb.bin and script.bin was taken from NAND with Android where USB device mode seems to work.
evb.bin:

[usbc0]
usb_used = 1
usb_port_type = 2
usb_detect_type = 1
usb_id_gpio = port:PH04<0><1><default><default>
usb_det_vbus_gpio = port:PH05<0><0><default><default>
usb_drv_vbus_gpio = port:PB09<1><0><default><0>
usb_host_init_state = 0

[usbc1]
usb_used = 0
usb_port_type = 1
usb_detect_type = 0
usb_id_gpio =
usb_det_vbus_gpio =
usb_drv_vbus_gpio = port:PH06<1><0><default><0>
usb_host_init_state = 1

[usbc2]
usb_used = 1
usb_port_type = 1
usb_detect_type = 0
usb_id_gpio =
usb_det_vbus_gpio =
usb_drv_vbus_gpio = port:PH12<1><0><default><0>
usb_host_init_state = 0

I think usbc2 is a link to rtl8192cu WiFi module.

[YuukiHogo]

Changing usb_used to 1 in [usbc1] section doesn't makes difference.

[YuukiHogo]

Value of is_udc_enable changes only in sw_udc_set_pullup() but it never fires up. It all ends in struct usb_gadget_ops with .pullup and .vbus_session members those never fire up also. Can't get which driver must call them...

[YuukiHogo]

OK. Simply changing init value of is_udc_enable to 1 almost solves initial issue. I think reverting changes related to drivers/usb/sun4i_usb/udc/sw_udc.c made by 9384245 must solve that issue completely.

But there come one more issue. When I use any gadget module - it basically works with Unix-based OSes (Linux & Mac OS X 10.6) but fails with OS Windows with code 10 during driver installation. Sniffing USB shows incomplete initialization without any endpoints on host Windows PC. Looks like something goes wrong during GET_DESCRIPTOR phase when host PC run Windows. On Linux host PC (according sysfs) endpoints appear correctly (i.e. for g_zero gadget there are two endpoints).

Will try to go deeper...

[techn]

Hi,

Would this commit help?
amery@4dfaa32

-Jari

2012/9/29 YuukiHogo notifications@github.com

OK. Simply changing init value of is_udc_enable to 1 almost solves
initial issue. I think reverting canges related to
drivers/usb/sun4i_usbudc/sw_udc.c made by 9384245amery@93842453763f63fbab41875189aa76cf9a145e35must solve that issue completely.

But there comes one more issue. When I use any gadget - it basically works
with Unix-based OSes (Linux & Mac OS X 10.6) but fails with OS Windows.
Sniffing USB shows incomplete initialization without any endpoints on host
PC.

—
Reply to this email directly or view it on GitHubhttps://github.com/amery/linux-allwinner/issues/76#issuecomment-9003201.

[YuukiHogo]

Seems HEAD already has it applied to sun4i_usb...

[cloud-rocket]

I tried changing initial value of is_udc_enable to 1, but it makes no difference - gadget drivers mention that controller is not loaded, I tried g_file_storage or g_hid (sw_udc.c throws "ERR: usb device is not active" error when trying to insmod gadget driver)

I also tried to switch OTG port to device only (using menuconfig), but in this case kernel has compilation errors....
I use 3.0.36 (crane config), any ideas?
Meir Tseitlin

[RobVanDerHoeven]

I did some browsing of the kernel source and found this function in usbc.c

void USBC_OTG_SelectMode(__hdle hUSB, __u32 mode)
{
    __usbc_otg_t *usbc_otg = (__usbc_otg_t *)hUSB;

    if(usbc_otg == NULL){
        return ;
    }

    if(mode == USBC_OTG_HOST){

    }else{

    }
}

Maybe this is part of the problem?

[YuukiHogo]

As meirts said:

I tried changing initial value of is_udc_enable to 1, but it makes no difference - gadget drivers mention that controller is not loaded, I tried g_file_storage or g_hid (sw_udc.c throws "ERR: usb device is not active" error when trying to insmod gadget driver)

I also tried to switch OTG port to device only (using menuconfig), but in this case kernel has compilation errors....
I use 3.0.36 (crane config), any ideas?

sun4i_crane_defconfig works with only one gadget driver and defaults to android_usb build-in module. So you need to switch it to another in .config (i.e. by $ make menuconfig). There is no way to # insmod gadget driver with sun4i_crane_defconfig because USB device controller is busy with built-in gadget module.

[amery]

alaihs the gadget code is heavily bound to android (even inside android.c!) so it's on the list of "needs to be rewritten" :<

[YuukiHogo]

Only android.c (which is android_usb) is bound to android infrastructure. Dunno why there is need to rewrite android_usb. Can't see a point to use it without android...

[amery]

but iirc there is where they injected the 'allwinnerization' code which we might need to move to a sunxi.c or so. But I haven't had time to look there in detail yet so I could be totally wrong.

[amery]

anyhow, can we move this to the mailing list?

• http://linux-sunxi.org/Mailing_list
• https://groups.google.com/d/forum/linux-sunxi

[cloud-rocket]

RobVanDerHoeven, USBC_OTG_SelectMode function is not called anywhere (and it is not system function as far as I know - so it is not important)

YuukiHogo, I know that UCD supports single gadget driver on top of it, but inside menuconfig you can preselect a driver (I tried putting serial or hid drivers instead of android) or compile it with no in-kernel driver and insmod one of the modules later (the injection procedure seems to be implemented). Anyway, I tried both ways and from my observation, the UDC device is not activated, so both cases are not working as a result of it.

amery, I am not sure android.c needs to be rewritten, because from my observation it is mostly related to adb and custom mass storage implementation. We just need to find out how to enable the UDC device.

I was inspecting their code for a little and I think that sw_udc.c::sw_usb_device_enable function needs to be called and it is not.
The call stack should be:
sw_udc.c::sw_usb_device_enable
from usb_msg_center.c::insmod_device_driver
usb_msg_center.c::do_usb_role_null
usb_msg_center.c::usb_msg_center

Observing dmesg log of android, these function are called when you attach usb host to the OTG port, but they are not when running ubuntu.

Interesting part is that I didn't find any traces of the relevant code inside android.c - so I think it is not related anyhow.

Ideas?

Also, let us know where are you willing to continue this conversation.

Meir

[amery]

@meirts the allwinnerization chunk in android.c is added by 9384245

[cloud-rocket]

Sorry don't understand what is "allwinnerization chunk" and how is it related :<

Isn't android.c just a gadget driver which suppose to load on top of UDC?

[amery]

@meirts this code was written by Allwinner (who makes the chip, and who has the documentation) and for them android is the goal, also script.bin (sys_config) is their wait to configure boards. integration and feature enabling is done via code like 9384245, where they decide to enable OTG or not.

[YuukiHogo]

@meirts I think it is related to UDC initialization. Check script.bin and try to use *_defconfig. Also it is not seems gadget-related issue. Similar to mine it is related to UDC code and HCD.

[cloud-rocket]

Yep, I know that their goal is android and that android.c is written by Allwinner for this purpose only. But I still can't figure out how to enable OTG device support inside sun4i_usb (which seems like regular linux usb controller driver written by Allwinner and knows hwo to work with just a single gadget on top of it).
As far as I understand android.c contains gadget only functionality.

android.c contains calls to usb_gadget_connect (strange function I don't recognize), after usb_add_config (which is standard Gadget API function)....

Maybe this is the missing point?

[cloud-rocket]

YuukiHogo, my device is Mini X and it works with crane config only. It uses evb.bin and I checked it already - seems all ok there:

[usbc0]
usb_used = 1
usb_port_type = 2
usb_detect_type = 1
usb_id_gpio = port:PH04<0><1>
usb_det_vbus_gpio = port:PH05<0><0>
usb_drv_vbus_gpio = port:PB09<1><0><0>
usb_host_init_state = 0

[YuukiHogo]

Why it doesn't work with sun4i_defconfig? I think it must to work with both types.

[amery]

as I told you. USB gadget/slave support assumes android and this dependency is assumed all over the code.

[YuukiHogo]

Anyway I have identical issue with sun4i_defconfig and sun4i_crane_defconfig.

[amery]

hi, can you give a try using the stage/sunxi-3.0 branch?

[YuukiHogo]

Just tried. Same behavior. Device completely ignores connection to host.

[amery]

@YuukiHogo as it's a VERY intrusive change for a problem that only you seem to have, please submit it to the linux-sunxi mailing list where the sunxi-usb "team" can see it, comment and suggest a proper solution.

[IDmedia]

I'm having the same issue. Has this issue been resolved?
Host mode works, but not switching to client/OTG mode. I'm on kernel version 3.4 (using BananaPi).

Output from dmesg:

[   86.631328] sd 2:0:0:0: [sda] Attached SCSI removable disk
[  108.515479] WRN:L2728(drivers/usb/sunxi_usb/udc/sw_udc.c):ERR: usb device is not active
[  108.523868] usb open backing file: /dev/sda, 0xeeb99800
[  108.530182] g_mass_storage gadget: Mass Storage Function, version: 2009/09/11
[  108.541530] g_mass_storage gadget: Number of LUNs=1
[  108.550433]  lun0: LUN: removable file: /dev/sda
[  108.562358] ep_matches, wrn: endpoint already claimed, ep(0xc08f406c, 0xbf02528c, ep1-bulk)
[  108.568320] g_mass_storage gadget: Mass Storage Gadget, version: 2009/09/11
[  108.581626] g_mass_storage gadget: userspace failed to provide iSerialNumber
[  108.593200] g_mass_storage gadget: g_mass_storage ready
[  108.602686] WRN:L2762(drivers/usb/sunxi_usb/udc/sw_udc.c):ERR: usb device is not active

[joyhope]

I have the similar problem in kernel version 3.4.103 (using BananaPro)

root@lemaker:~# WRN:L2728(/home/bananapi/bananapi/lemaker-bsp/linux->sunxi/drivers/usb/sunxi_usb/udc/sw_udc.c):ERR: usb device is not active
ep_matches, wrn: endpoint already claimed, ep(0xc08f7fbc, 0xede54a00, ep1-bulk)
ep_matches, wrn: endpoint already claimed, ep(0xc08f7fbc, 0xede54a00, ep1-bulk)
ep_matches, wrn: endpoint already claimed, ep(0xc08f8008, 0xede54a00, ep2-bulk)
<6>g_serial gadget: Gadget Serial v2.4
<6>g_serial gadget: g_serial ready
WRN:L2762(/home/bananapi/bananapi/lemaker-bsp/linux->sunxi/drivers/usb/sunxi_usb/udc/sw_udc.c):ERR: usb device is not active

Whether the issue is solved?

[strelec]

Damn, the only reason I went with the more expensive Banana PI was the OTG capability.

Now, it doesn't even work. Huge disappointment! Is this being worked on? Why is the bug closed if it is not fixed?

[EduardMe]

I got the android composite driver working. All kinds of gadgets can be loaded into it (check under /sys/class/android_usb/android0/functions):

# 1 = host role, 2 = device role
echo 1 > /sys/bus/platform/devices/sunxi_usb_udc/otg_role
echo 2 > /sys/bus/platform/devices/sunxi_usb_udc/otg_role

echo 0 > /sys/class/android_usb/android0/enable
echo 1d6b > /sys/class/android_usb/android0/idVendor
echo 0001 > /sys/class/android_usb/android0/idProduct
echo mtp > /sys/class/android_usb/android0/functions
echo 1 > /sys/class/android_usb/android0/enable

HID for example is detected on a host in my tests. Others need to be configured. Worked on the OrangePI PC, maybe it helps for BananaPI.

[lauraruoyun]

Hi, has someone find the solution for the problem reported in 2012 : When I use any gadget module - it basically works with Unix-based OSes (Linux & Mac OS X 10.6) but fails with OS Windows with code 10 during driver installation. Sniffing USB shows incomplete initialization without any endpoints on host Windows PC. Looks like something goes wrong during GET_DESCRIPTOR phase when host PC run Windows. On Linux host PC (according sysfs) endpoints appear correctly (i.e. for g_zero gadget there are two endpoints). We're in 2016, I have exactly the same problem, I use Orange PI Lite with Kernel Linux Sunxi 3.4. Can anyone help? Thanks!