Re: [PATCH] Second set of console changes.

[James S. <jsi...@tr...>]

On Thu, 25 Jul 2002, Ewan Mac Mahon wrote:

> On Wed, 24 Jul 2002, James Simmons wrote:
> >
> >    To the people with the devfs issues. Please send me a log of what
> > exactly happened and a detail ksymoop if you can. I just tried it on my
> > system with devfs enabled and it works for me.
>
> It doesn't oops, it just doesn't register the devices so you can't open
> gettys on them. Other than that the kernel boots fine and you can log in
> over the network. Doing that you can see a couple of big difference in
> /dev:

I tracked down the problem. Originally the code initialized the VT tty
early before kmalloc. So we had this:

console_driver.flags |= TTY_DRIVER_NO_DEVFS;

Now in tty_register_driver, which was called right afterwards, we have
this bit of code.

if ( !(driver->flags & TTY_DRIVER_NO_DEVFS) ) {
                for(i = 0; i < driver->num; i++)
                    tty_register_devfs(driver, 0, driver->minor_start + i);
}

In the old code code the above was never called. Instead the code in
con_init_devfs was called.

Now in the new code we don't have TTY_DRIVER_NO_DEVFS set so the above is
called. The problem is the default flag that is passed into
tty_register_devfs. It is automatically 0 whereas before it was
DEVFS_FL_AOPEN_NOTIFY. The problem is the flag being passed.

I tried out devfs and found the problem is only root is now only allowed
to access vc/X. This is the problem. I haven't figured out a solution yet.
Any ideas anyone?