Re: [PATCH 00/13] tty.h cleanups

[Johan Hovold <johan@kernel.org>]

On Thu, Apr 15, 2021 at 10:21:54AM +0200, Greg Kroah-Hartman wrote:
> On Fri, Apr 09, 2021 at 09:32:45AM +0200, Johan Hovold wrote:
> > On Thu, Apr 08, 2021 at 08:01:08PM +0200, Greg Kroah-Hartman wrote:
> > > On Thu, Apr 08, 2021 at 04:25:22PM +0200, Johan Hovold wrote:
> > > > On Thu, Apr 08, 2021 at 02:51:21PM +0200, Greg Kroah-Hartman wrote:
> > > > > Turns out there is a lot of tty-internal stuff in include/linux/tty.h
> > > > > that do not belong there.  Create a internal-to-the-tty-layer .h file
> > > > > for these types of things and move function prototypes to it instead of
> > > > > being in the system-wide header file.
> > > > > 
> > > > > Along the way clean up the use of some old tty-only debugging macros and
> > > > > use the in-kernel dev_*() calls instead.
> > > > 
> > > > I'm afraid that's not a good idea since not all ttys have a
> > > > corresponding class device. Notable exception include pseudo terminals
> > > > and serdev.
> > > > 
> > > > While dev_printk() can handle a NULL device argument without crashing,
> > > > we'll actually lose log information by removing the tty printk helpers.
> > > 
> > > I think the same info will be printed here as before, just some NULL
> > > information at the beginning, right?  And the benifits overall (for real
> > > tty devices), should outweigh the few devices that do not have this
> > > information.
> > 
> > No, you'll only be losing information (tty driver and tty name). Here's
> > a pty example, where the first line in each pair use dev_info() and the
> > second tty_info():
> > 
> > [   10.235331] (NULL device *): tty_get_device
> > [   10.235441] ptm ptm0: tty_get_device
> > 
> > [   10.235586] (NULL device *): tty_get_device
> > [   10.235674] pts pts0: tty_get_device
> > 
> > and similar for serdev, which is becoming more and more common.
> 
> Ok, good point, I'll go apply only the first 2 patches in this series
> (moving the macros out of tty.h and removing the unused one) and then
> will redo this set of patches again.

Perhaps no harm in leaving the tty_info() on in there for consistency.
We have users of the _ratelimited() flavour of it (even if there's no
dependency).

> I think a better tty_msg() macro is warrented so that we can provide
> dev_*() output if we have a device, otherwise fall back to the old
> style to preserve functionality.

Possibly, but the dev_printk() for the tty class devices wouldn't
provide any more info than what's already there (i.e. driver name + tty
name).

(And associating ttys with other devices and drivers (e.g. a serdev
client and its driver) might not be what we want since you lose the
connection to the underlying tty driver.)

Johan