ADB requires new userland setup w/ linux-linaro-android 3.0-2011.07

[John Stultz]

ADB was reworked in the linux-3.0 kernel, removing the need for device specific platform drivers (similar to the change I did in 2.6.39). However, for adb to work over USB, some userland commands have to be issued at boot:

Workaround:
echo 0 > /sys/class/android_usb/android0/enable
echo adb,acm > /sys/class/android_usb/android0/functions
echo 2 > /sys/class/android_usb/android0/f_acm/instances
echo 1 > /sys/class/android_usb/android0/enable

(The above were found in the commit log here: https://review.source.android.com/#/c/23890/ )

After those commands are issued, I was able to get ADB functioning with the android-3.0 kernel.

[Zach Pfeffer (pfefferz)]

When I run the last command I get an oops.

[Zach Pfeffer (pfefferz)]

The above crash was on: https://android-build.linaro.org/builds/~linaro-android/panda/#build=190

[John Stultz (jstultz)]

Yuck. So the crash only happens on panda when no USB cable is plugged into the USB gadget port.
If the cable is plugged in when the enable sysfs call is made, it seems to work ok.

Still digging.

[John Stultz (jstultz)]

Ok. Got a potential fix posted to lkml for review:
https://lkml.org/lkml/2011/7/20/357

[Alexander Sack (asac)]

can we commit this fix as a workaround to tree for 11.07 and verify before it gets confirmed on lkml?

[John Stultz (jstultz)]

Ok, nagged the maintainer and he said he's a bit too busy to look at it right now, but its in his queue for post 3.1-rc1.

Thus, the fix has been commited to the linaro+android tree for now and I've pushed it out so that its available.

Zach: You ok with adding the adb setup commands now in the init scripts?

[Ken Werner (kwerner)]

I've added a "service" by appending the follwoing two lines to the init.rc:

<snip>
service adbfix /system/etc/adbfix.sh
    oneshot
</snip>

and put the commands mentioned above in a shell script at /system/etc/adbfix.sh - works fine for me.

[John Stultz (jstultz)]

So with the crash resolved, there shouldn't be any more kernel side functionality required, so I'm assigning this to Zach for the integration of the userland modifications (similar to what Ken did).

[Alexander Sack (asac)]

this bug has a solution it seems. From the info I see here it feels like a low hanging fruit.

Potential board specific issues can be followed up on in individual bugs afterwards. The sooner we land this fix in platform, the sooner we'll know.

[vishal (vishalbhoj)]

with the commit http://review.android.git.linaro.org/#change,135 , adb should have worked.But I see the following behaviour.

On Panda build, adb works on panda provided cable is plugged into the device after its booted up(android driver gets configured completely after cable is plugged in), appears to be stable this way works on multiple times plugging/unplugging of cable.
On Beagle-xM: adb is not stable,keeps disconnecting when you run commands like "logcat OR ls -l" on the adb shell.
On LEB-panda,adb works but kernel crashes when you unplug the cable.

[Zach Pfeffer (pfefferz)]

Vishal,

Would you file 3 target specific bugs with each behavior and assign them to both the TI LT (Andy) and the Android Team (You). Make sure you mark them public.

[vishal (vishalbhoj)]

adb is not working on the devices listed below.

Beagle-xM https://bugs.launchpad.net/linux-linaro/+bug/845642
Beagle https://bugs.launchpad.net/linux-linaro/+bug/845642
Staging-Origen https://bugs.launchpad.net/linaro-android/+bug/859896
Staging-Snowball https://bugs.launchpad.net/linaro-android/+bug/847652
Staging-iMX53 https://bugs.launchpad.net/linaro-android/+bug/859995