problem w/ kernel init (rootfs vs initial console / init process)

problem w/ kernel init (rootfs vs initial console / init process)

[mtx]

Hi folks,


I've got a very strange problem w/ v4.13 (on imx6):

it tries to open the initial console, before even attempt to mount the
root fs, which obviously is doomed to fail. (directly passing root=
via cmdline - no initrd etc)

What puzzles me most: prepare_namespace() is only called from
kernel_init_freeable() - after it tried to open console (only when no
ramdisk_execute_command found).


Anyone have an idea what could be wrong here ?


--mtx

unable to open initial console (ENODEV) (WAS: problem w/ kernel init (rootfs vs initial console / init process))

[mtx]

On 10.09.2017 11:09, mtx wrote:

> it tries to open the initial console, before even attempt to mount the > root fs, which obviously is doomed to fail. (directly passing root=> 
via cmdline - no initrd etc)
Solved it. For the record: the uart2 node was still disabled, so the
kernel did allocate a console for the port, but no tty_driver assigned.

tty_lookup_driver() (tty_io.c) is called on opening tty devices (such as
/dev/console). it asks console_device() (in printk.c), which scans the
consoles list for one with a tty_driver.

As my DT left the console's uart port disabled, the corresponding device
wasn't initialized, thus leaving the console w/o assigned tty_driver.

BTW: maybe we should think about some logic which spits out a warning,
or maybe implement an own special tty_driver for /dev/console, which
goes to the active console(s), just like printk() does.

--mtx