Re: [Xen-devel] reg dom0 console

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Sat, 2011-10-08 at 03:22 +0100, Prasad B wrote:
> On Wed, Oct 5, 2011 at 11:54 PM, Ian Campbell
> <Ian.Campbell@xxxxxxxxxx> wrote:
>         > After a few months of hiatus, we have resumed working on
>         mips port of
>         > Xen. We are at the point of booting dom0 userland (init
>         process and
>         > its children).
>         
>         
>         That sounds awesome! Are you close to having any initial
>         patches for
>         submission?
> 
> We are working with 3.4.0 which is really old. So, we would like to
> run a few heavyweight applications on Linux, forward port our MIPS
> changes to 4.1.0 and then submit the patches along with our PV Linux.
> We are using 2.6.34 Linux kernel and have not been working with pvops
> interface. So, there might have to be some collective effort to move
> our MIPS Linux changes into the pvops framework.
> 
> 
>  
>         > The init process gets spawned and in turn forks off
>         processes to run
>         > commands in /etc/rc.d/init.d/rcS. However, the getty process
>         does not
>         > seem to run subsequently and hence the login prompt is not
>         displayed.
>         > Dom 0 does not use the uart and uses the hvc0 driver.
>         >
>         > I searched the web for hints and there were suggestions to
>         add
>         > directives such as
>         >
>         > 1:2345:respawn:/sbin/getty 38400 hvc0
>         >
>         > to /etc/inittab file. But that does not seem to help.
>         >
>         > Could you please give any pointers to get getty talk to the
>         console
>         > driver correctly ?
>         
>         
>         Do you have something which is creating the actual /dev/hvc0
>         node? You
>         can probably boot with init=/bin/sh in order to get a shell to
>         poke
>         around and check this with, run the getty by hand etc.
> 
> replacing /sbin/init with /bin/sh displays the following output:
> 
> "BusyBox v1.14.0 (2009-05-01 15:38:41 PDT) built-in shell (ash)
> Enter 'help' for a list of built-in commands.
> 
> /bin/sh: can't access tty; job control turned of"
> 
> and input/commands could not be fed to the shell.
> 
> 
>         I had a vague recollection that the busybox inittab syntax was
>         a little
>         different to standard sysvinit but google threw up various
>         things which
>         suggest that what you've got is OK. Are you maybe in runlevel
>         1? You
>         could try changing the 2345->12345 to be sure. You probably
>         want this
>         getty in single user mode anyway. Alternatively lots of
>         example busybox
>         configurations I see on google seems to omit the runlevels,
>         which I
>         guess means "all"?
> 
> created an hvc device in /dev and played with different runlevels. But
> getty process is either dormant or dead and the cpu_idle() function
> kicks in.

I'm afraid I've not got any real ideas. Do you think that event channels
are working correctly? In particular you will need to be able to receive
VIRQ_CONSOLE.

Hopefully the stuff in drivers/tty/hvc/hvc_xen.c:xen_hvc_init is taking
the correct xen_initial_domain paths and not the domU ones. I think this
must be correct, or you wouldn't get any output at all, but adding a
printk or two to confirm might be handy.

I'm not 100% sure but it looks like passing xen_irq == -1 to hvc_alloc
might force things into a polled mode? That might be a useful debugging
step.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel