Re: [PATCH v3 4/4] kdb: Switch kdb_msg_write() to use safer polling I/O

[Sumit Garg <sumit.garg@linaro.org>]

On Thu, 28 May 2020 at 20:27, Petr Mladek <pmladek@suse.com> wrote:
>
> On Thu 2020-05-28 12:26:20, Daniel Thompson wrote:
> > On Thu, May 28, 2020 at 11:48:48AM +0530, Sumit Garg wrote:
> > > On Wed, 27 May 2020 at 19:01, Daniel Thompson
> > > <daniel.thompson@linaro.org> wrote:
> > > >
> > > > On Wed, May 27, 2020 at 11:55:59AM +0530, Sumit Garg wrote:
> > > > > In kgdb NMI context, calling console handlers isn't safe due to locks
> > > > > used in those handlers which could lead to a deadlock. Although, using
> > > > > oops_in_progress increases the chance to bypass locks in most console
> > > > > handlers but it might not be sufficient enough in case a console uses
> > > > > more locks (VT/TTY is good example).
> > > > >
> > > > > Currently when a driver provides both polling I/O and a console then kdb
> > > > > will output using the console. We can increase robustness by using the
> > > > > currently active polling I/O driver (which should be lockless) instead
> > > > > of the corresponding console. For several common cases (e.g. an
> > > > > embedded system with a single serial port that is used both for console
> > > > > output and debugger I/O) this will result in no console handler being
> > > > > used.
> > > >
> > > > > diff --git a/include/linux/kgdb.h b/include/linux/kgdb.h
> > > > > index b072aeb..05d165d 100644
> > > > > --- a/include/linux/kgdb.h
> > > > > +++ b/include/linux/kgdb.h
> > > > > @@ -275,6 +275,7 @@ struct kgdb_arch {
> > > > >   * for the I/O driver.
> > > > >   * @is_console: 1 if the end device is a console 0 if the I/O device is
> > > > >   * not a console
> > > > > + * @tty_drv: Pointer to polling tty driver.
> > > > >   */
> > > > >  struct kgdb_io {
> > > > >       const char              *name;
> > > > > @@ -285,6 +286,7 @@ struct kgdb_io {
> > > > >       void                    (*pre_exception) (void);
> > > > >       void                    (*post_exception) (void);
> > > > >       int                     is_console;
> > > > > +     struct tty_driver       *tty_drv;
> > > >
> > > > Should this be a struct tty_driver or a struct console?
> > > >
> > > > In other words if the lifetime the console structure is the same as the
> > > > tty_driver then isn't it better to capture the console instead
> > > > (easier to compare and works with non-tty devices such as the
> > > > USB debug mode).
> > > >
> > >
> > > IIUC, you mean to say we can easily replace "is_console" with "struct
> > > console". This sounds feasible and should be a straightforward
> > > comparison in order to prefer "dbg_io_ops" over console handlers. So I
> > > will switch to use "struct console" instead.
> >
> > My comment contains an if ("if the lifetime of the console structure is
> > the same") so you need to check that it is true before sharing a patch to
> > make the change.
>
> Honestly, I am not completely familiar with the console an tty drivers
> code.
>
> Anyway, struct console is typically statically defined by the console
> driver code. It is not must to have but I am not aware of any
> driver where it would be dynamically defined.
>

Yes this is mine understanding as well.

> On the other hand, struct tty_driver is dynamically allocated
> when the driver gets initialized.
>
> So I would say that it is pretty safe to store struct console.

Okay.

> Well, you need to call con->device() to see if the tty_driver
> is actually initialized.

Agree and con->device() is already invoked here [1]. So we only need
to store struct console if con->device() invocation returns success.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/tty/serial/kgdboc.c#n174

-Sumit

>
> Best Regards,
> Petr