On Fri, Dec 27, 2019 at 06:30:11PM +0000, Sudip Mukherjee wrote:
> On Mon, Dec 23, 2019 at 07:06:51AM -0500, Greg KH wrote:
> > On Fri, Dec 20, 2019 at 10:08:03AM -0800, Linus Torvalds wrote:
> > > On Thu, Dec 19, 2019 at 11:07 PM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > The last tty core fix should resolve a long-standing bug with a race
> > > > at port creation time that some people would see, and Sudip finally
> > > > tracked down.
> > >
> > > Hmm, looks good. But it makes me wonder if we should now try to remove
> > > the second call to tty_port_link_device()?
> > >
> > > Now we have a number of helpers that do that tty_port_link_device()
> > > call for the driver (eg tty_port_register_device_attr_serdev(),
> > > tty_port_register_device_attr(), and the just added
> > > uart_add_one_port()).
> > >
> > > But we also have drivers doing it by hand, and presumably we now have
> > > drivers that do it through multiple paths? I guess it's harmless, but
> > > it feels a bit odd. No?
> >
> > It does. I'll try to look at this after the holidays unless Sudip beats
> > me to it.
>
> The second call to tty_port_link_device() is in
> tty_port_register_device_attr_serdev() and tty_port_register_device_attr()
> is being called from many other places apart from uart_add_one_port().
> The attached patch should be safe. I will test and send it properly unless
> someone objects to it.
No, this is horrid. There's certainly room for some clean up here but we
shouldn't make things worse. ;)
Why not look into registering the console only after the port has been
set up and registered and revert fb2b90014d78 ("tty: link tty and port
before configuring it as console") completely instead?
Johan
