Re: your mail

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sun, Feb 12, 2012 at 01:40:47PM +0100, Jiri Slaby wrote:
> > Is tty_kref_put() safe in interrupt?  Here it seems to be OK, but in other
> > callers...  More or less at random: drivers/tty/serial/lantiq.c has it
> > called from lqasc_rx_int().  It seems to be possible to have it end up
> > calling ->ops->shutdown() and in this case that'd be lqasc_shutdown().
> > Which does a bunch of free_irq(), including the ->rx_irq, i.e. the one
> > we have it called from.  Alan?
> 
> I'm not Alan, but will reply anyway. Yes, it is safe (unless the driver
> does something tricky). In the driver you mention, this is uart_ops,
> called from tty_port_operations' ->shutdown. And that's a different from
> tty_operations' ->shutdown.
> 
> Yes, there are:
> * tty->ops
> * tty_port->ops
> * uart_port->ops
> 
> uart_port->ops->shutdown is supposed to tear down interrupts like in
> lantiq.c. It is called from tty_port->ops->shutdown. And that one is
> allowed to be called only from user context (tty->ops->close and
> tty->ops->hangup).

Yecchhh...  If I'm reading (and grepping) it right, there are only two
non-default instance of tty_operations ->shutdown() - pty and vt ones.
Lovely...  And while we are at it, vt instance is definitely not safe
from interrupts - calls console_lock().  Not that it was relevant in
this case...

It's probably too late in this case, but I would've called that method
->sync_cleanup().  Assuming I'm not misreading its intent and history...