Linux dev-4.7 tree: Request for testing

[Joel Stanley]

(+cc jeremy, I'm after some advice on how we should discover which tty
is the vuart. See below)

On Wed, Aug 10, 2016 at 10:46 AM, Patrick Williams <patrick at stwcx.xyz> wrote:
> On Wed, Aug 10, 2016 at 12:43:14AM +0930, Joel Stanley wrote:
>> Hello OpenBMCers,
>>
>> I've pushed a new Linux tree to https://github.com/openbmc/linux as
>> dev-4.7. I intend this to become the new tree used for openbmc's
>> master branch.
>
> Joel,
>
> I tried to pull this into the greater OpenBMC build at:
>     https://gerrit.openbmc-project.xyz/#/c/159/
>
> Good news: it compiles fine and gets to userspace with network!  Bad
> news: lots of other things are broken.

I booted it on my Palmetto today. To my surprise it attempted to turn
the host on; the fans started spinning. However I did not see anything
come out of the vuart.

>
> My first observations:
>
> * The host tty has changed from S5 to S4?
>     Aug 09 21:52:37 barreleye systemd[1]: Found device /dev/ttyS4.

No intentional change. I'm not sure what the userspace is trying to do
here. ttyS4 is the Linux console device.

>     Aug 09 21:52:41 barreleye obmc-console-server[588]: obmc-console-server:
>         Can't access attribute sirq on device ttyS5: Permission denied
>     Aug 09 21:52:41 barreleye obmc-console-server[588]: obmc-console-server:
>         Can't access attribute lpc_address on device ttyS5: Permission denied
>     Aug 09 21:52:41 barreleye obmc-console-server[588]: obmc-console-server:
>         Can't read tty termios: Input/output error

We need to use a mechanism to tell userspace which device is the
vuart. udev? systemd?

$ ls /sys/devices/platform/ahb/ahb:apb/1e787000.vuart/tty
ttyS0

We could also set an alias in the device tree?

>
> * There might be some 'startup_hacks' to remove in skeleton:

>     Aug 09 21:53:27 barreleye system_manager.py[606]:
>         /usr/sbin/startup_hacks.sh: line 9:
>         /sys/devices/platform/ahb/ahb:apb/1e787000.vuart/enabled: Permission
>         denied

This would have stopped working in January, when we changed the
behaviour of the vuart to enable on bind:

 https://lists.ozlabs.org/pipermail/openbmc/2016-February/001883.html

The startup_hacks script should be fixed to not touch 'enabled'.

>
> * i2c and LEDs seems to be broken in general.
>     Aug 09 21:53:33 barreleye system_manager.py[606]: sh:
>         /sys/class/i2c-adapter/i2c-3/3-0050/online: No such file or directory
>     Aug 09 21:53:33 barreleye system_manager.py[606]: sh:
>         /sys/class/i2c-adapter/i2c-3/3-0050/online: No such file or directory
>     Aug 09 21:53:37 barreleye system_manager.py[606]: No LEDs present in the
>         system
>
>     Also, no /sys/class/i2c-adapter at all and no dmesg about i2c.
>     Maybe missing in kernel config somehow?  /sys/class/leds and
>     /sys/class/hwmon are both empty.

Yes, the kernel config was out of date. It needed to be updated for
the new tree.

>
> * pinctrl problems when we try to set up GPIOs
>     aspeed-g4-pinctrl 1e6e2000.syscon:pinctrl at 1e6e2000: request() failed for
>     pin 138
>     aspeed-g4-pinctrl 1e6e2000.syscon:pinctrl at 1e6e2000: pin-138
>     (1e780000.gpio:458) status -22

GPIOR was not present in pinmux. I have added it.

>     aspeed-g4-pinctrl 1e6e2000.syscon:pinctrl at 1e6e2000: request() failed for
>     pin 34
>     aspeed-g4-pinctrl 1e6e2000.syscon:pinctrl at 1e6e2000: pin-34
>     (1e780000.gpio:354) status -22

GPIOE was not present in pinmux either. I've added it.

Cheers,

Joel