On Tuesday, February 25, 2025 11:19:04 AM EST Petr Mladek wrote:
> Hi Adam,
>
> please add printk maintainers into Cc as already suggested by Andy
> at https://lore.kernel.org/r/CAHp75VeBaetiQBykfLk_weBHdzZF1nWp=k8BJu+OKNp6iYRRTg@mail.gmail.com
>
OK, I have added the other printk maintainers in this thread, and will update
my git send email script.
> The motivation is that the console registration code is in
> kernel/printk/printk.c. It is historically pretty tricky.
> Some ordering is defined rather by chance than by design.
> And we should be careful when adding new rules and hacks.
>
> On Mon 2025-02-24 07:39:14, adamsimonelli@gmail.com wrote:
> > From: Adam Simonelli <adamsimonelli@gmail.com>
> >
> > The new config option, CONFIG_NULL_TTY_CONSOLE will allow ttynull to be
> > initialized by console_initcall() and selected as a possible console
> > device.
> >
> > Signed-off-by: Adam Simonelli <adamsimonelli@gmail.com>
> > ---
> > drivers/tty/Kconfig | 15 ++++++++++++++-
> > drivers/tty/ttynull.c | 20 +++++++++++++++++++-
> > 2 files changed, 33 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/tty/Kconfig b/drivers/tty/Kconfig
> > index 63a494d36a1f..b4afae8b0e74 100644
> > --- a/drivers/tty/Kconfig
> > +++ b/drivers/tty/Kconfig
> > @@ -383,7 +383,20 @@ config NULL_TTY
> > available or desired.
> >
> > In order to use this driver, you should redirect the console to this
> > - TTY, or boot the kernel with console=ttynull.
> > + TTY, boot the kernel with console=ttynull, or enable
> > + CONFIG_NULL_TTY_CONSOLE.
> > +
> > + If unsure, say N.
> > +
> > +config NULL_TTY_CONSOLE
>
> It makes sense to enable this behavior by a CONFIG_ setting
> but the name is misleading.
>
> > +
> > + bool "Support for console on ttynull"
> > + depends on NULL_TTY=y && !VT_CONSOLE
> > + help
> > + Say Y here if you want the NULL TTY to be used as a /dev/console
> > + device.
> > +
> > + This is similar to CONFIG_VT_CONSOLE, but without the dependency on
> > + CONFIG_VT. It uses the ttynull driver as the system console.
>
> It is true that CONFIG_VT_CONSOLE causes that the virtual terminal
> will get associated with /dev/console. But it works only "by chance".
>
> It works because "register_console(&vt_console_driver)" in
> con_init() is the first register_console() call. And it also
> works only by chance because of the linking order.
>
> Anyway, there are more similar CONFIG_ options, for example,
> CONFIG_LP_CONSOLE, or CONFIG_VIRTIO_CONSOLE. And they are not
> default when CONFIG_VT_CONSOLE is enabled. They are registered only
> when the related console= option is defined on the command line.
>
> I want to say that CONFIG_<BLA>_CONSOLE does not mean
> that the BLA console will be registered by default.
> And we should us a better descriptive name, for example,
>
> NULL_TTY_DEFAULT_CONSOLE
> NULL_TTY_DEV_CONSOLE
>
OK. I can change the option name.
> > If unsure, say N.
> >
> > diff --git a/drivers/tty/ttynull.c b/drivers/tty/ttynull.c
> > index 6b2f7208b564..ec3dd3fd41c0 100644
> > --- a/drivers/tty/ttynull.c
> > +++ b/drivers/tty/ttynull.c
> > @@ -57,6 +57,13 @@ static struct tty_driver *ttynull_device(struct console *c, int *index)
> > static struct console ttynull_console = {
> > .name = "ttynull",
> > .device = ttynull_device,
> > +
> > + /*
> > + * Match the index and flags from other boot consoles when CONFIG_NULL_TTY_CONSOLE is
> > + * enabled, otherwise, use the default values for the index and flags.
> > + */
> > + .index = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? -1 : 0,
>
> This should not be needed. "con->index" is always initialized to "0"
> for the default console, see:
>
OK, I had this in an #ifdef before, it was the cleanest way to set it to -1
that I could think of, other than the ifdef... If I still need this, I will try
to think of something else to set it to -1 when the option is enabled
> static void try_enable_default_console(struct console *newcon)
> {
> if (newcon->index < 0)
> newcon->index = 0;
> [...]
> }
>
> > + .flags = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? CON_PRINTBUFFER : 0,
>
> This does not make much sense to me.
>
> CON_PRINTBUFFER prevents duplicated output when the same device has
> already been registered as a boot console. But ttynull does not have
> a boot console variant. Also it is a "null" device. It never prints
> anything. The output could never be duplicated by definition.
>
OK, I was duplicating what I saw in other consoles. I can try to remove it
> > };
> >
> > static int __init ttynull_init(void)
> > @@ -90,11 +97,22 @@ static int __init ttynull_init(void)
> > }
> >
> > ttynull_driver = driver;
> > - register_console(&ttynull_console);
> > + if (!console_is_registered(&ttynull_console))
> > + register_console(&ttynull_console);
> >
> > return 0;
> > }
> >
> > +#ifdef CONFIG_NULL_TTY_CONSOLE
> > +static int __init ttynull_register(void)
> > +{
> > + if (!console_is_registered(&ttynull_console))
> > + register_console(&ttynull_console);
> > + return 0;
> > +}
> > +console_initcall(ttynull_register);
> > +#endif
>
> This looks strange. I guess that you needed to move this into
> console_initcall() because it is called earlier together
> with the other console_initcall() calls for serial ports.
> Otherwise, the hack with the linking order (2nd patch) did
> not work.
>
> But you needed to keep it in ttynull_init() so that ttynull
> did not get registered prematurely when CONFIG_NULL_TTY_CONSOLE
> was not enabled.
>
> Sigh, it looks like a dirty hack which works rather by chance
> than by design.
>
>
> Thinking loudly:
>
> The register_console() code is a historic mess. I dream about
> having time to clean it up. Anyway, there are basically two
> modes:
>
> 1. try_enable_default_console(newcon) is called only when
> there is no @preferred_console and there is no registered
> console with tty binding (valid con->device).
>
> The first register_console() caller wins. The order is defined
> by the __con_initcall section. Which is defined by the linking
> order.
>
> IMHO, it is quite fragile and non-intuitive.
>
>
> 2. try_enable_preferred_console() is called when some
> console is preferred via console_cmdline[]. The entries
> are added by __add_preferred_console() calls.
>
> This approach was created to handle console= command line
> parameters. But it was later used to define default consoles
> also via SPCR and device tree, see add_preferred_console()
> callers.
>
> It is also a bit tricky because the last added entry
> is preferred. Plus the .user_specified entries are
> preferred over the entries added via SPCR or device tree.
>
> Anyway, I think that the preference and ordering defined
> by console_cmdline[] array is a more intuitive approach.
>
>
> My proposal is to call:
>
> #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> add_preferred_console("ttynull", 0, NULL);
> #endif
>
> somewhere in the kernel code. The question is where.
> I wonder if the following would work:
>
> #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> static int __init ttynull_default_console(void)
> {
> add_preferred_console("ttynull", 0, NULL);
> return 0;
> }
> console_initcall(ttynull_register);
> #endif
>
> Best Regards,
> Petr
>
OK, actually in earlier revisions locally, I did actually have
diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index dddb15f48d59..c1554a789de8 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -3712,6 +3712,11 @@ void __init console_init(void)
initcall_t call;
initcall_entry_t *ce;
+#ifdef CONFIG_NULL_TTY_CONSOLE
+ if (!strstr(boot_command_line, "console="))
+ add_preferred_console("ttynull", 0, NULL);
+#endif
+
/* Setup the default TTY line discipline. */
n_tty_init();
Which worked as far as I could tell, at least on x86. Not sure if that was the
right place, and yeah, I was trying to better copy how CONFIG_VT_CONSOLE worked
because I thought that was more correct.On Wed 2025-02-26 08:39:23, Adam Simonelli wrote:
> On Tuesday, February 25, 2025 11:19:04 AM EST Petr Mladek wrote:
> > On Mon 2025-02-24 07:39:14, adamsimonelli@gmail.com wrote:
> > > From: Adam Simonelli <adamsimonelli@gmail.com>
> > >
> > > The new config option, CONFIG_NULL_TTY_CONSOLE will allow ttynull to be
> > > initialized by console_initcall() and selected as a possible console
> > > device.
> > >
> > > diff --git a/drivers/tty/ttynull.c b/drivers/tty/ttynull.c
> > > index 6b2f7208b564..ec3dd3fd41c0 100644
> > > --- a/drivers/tty/ttynull.c
> > > +++ b/drivers/tty/ttynull.c
> > > @@ -57,6 +57,13 @@ static struct tty_driver *ttynull_device(struct console *c, int *index)
> > > static struct console ttynull_console = {
> > > .name = "ttynull",
> > > .device = ttynull_device,
> > > +
> > > + /*
> > > + * Match the index and flags from other boot consoles when CONFIG_NULL_TTY_CONSOLE is
> > > + * enabled, otherwise, use the default values for the index and flags.
> > > + */
> > > + .index = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? -1 : 0,
> >
> > This should not be needed. "con->index" is always initialized to "0"
> > for the default console, see:
> >
> OK, I had this in an #ifdef before, it was the cleanest way to set it to -1
> that I could think of, other than the ifdef... If I still need this, I will try
> to think of something else to set it to -1 when the option is enabled
Ah, I was not clear enough. It should be perfectly fine to always
statically initialize the value to -1. We should not need any
#ifdef or IS_ENABLED.
I mean to do:
static struct console ttynull_console = {
.name = "ttynull",
.device = ttynull_device,
.index = -1,
};
We might even do this in a separate patch. IMHO, it should have
been done this way since the beginning.
> > static void try_enable_default_console(struct console *newcon)
> > {
> > if (newcon->index < 0)
> > newcon->index = 0;
> > [...]
> > }
> >
> > > + .flags = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? CON_PRINTBUFFER : 0,
> >
> > This does not make much sense to me.
> >
> > CON_PRINTBUFFER prevents duplicated output when the same device has
> > already been registered as a boot console. But ttynull does not have
> > a boot console variant. Also it is a "null" device. It never prints
> > anything. The output could never be duplicated by definition.
> >
> OK, I was duplicating what I saw in other consoles. I can try to remove it
Again, I was not clear enough. My primary concern was that it did not make
much sense to use the IS_ENABLED() check and initialize the value
different way.
Anyway, I would omit the flag. It is a NULL device. It does not matter
whether it prints existing (old) messages during registration or not.
> > > };
> > >
> > My proposal is to call:
> >
> > #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> > static int __init ttynull_default_console(void)
> > {
> > add_preferred_console("ttynull", 0, NULL);
> > return 0;
> > }
> > console_initcall(ttynull_register);
> > #endif
> >
> OK, actually in earlier revisions locally, I did actually have
>
>
>
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index dddb15f48d59..c1554a789de8 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -3712,6 +3712,11 @@ void __init console_init(void)
> initcall_t call;
> initcall_entry_t *ce;
>
> +#ifdef CONFIG_NULL_TTY_CONSOLE
> + if (!strstr(boot_command_line, "console="))
> + add_preferred_console("ttynull", 0, NULL);
Good point! We should call add_preferred_console() only when
the is no console= command line parameter. Otherwise, it could
not get overridden by the command line.
We could check "console_set_on_cmdline", similar to
xenfb_make_preferred_console().
> +#endif
> +
> /* Setup the default TTY line discipline. */
> n_tty_init();
>
> Which worked as far as I could tell, at least on x86. Not sure if that was the
> right place,
I would prefer to keep it in drivers/tty/ttynull.c when possible.
The following might do the trick:
#ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
static int __init ttynull_default_console(void)
{
if (!console_set_on_cmdline)
add_preferred_console("ttynull", 0, NULL);
return 0;
}
console_initcall(ttynull_register);
#endif
Best Regards,
Petr
On Monday, March 3, 2025 8:20:11 AM EST Petr Mladek wrote:
> On Wed 2025-02-26 08:39:23, Adam Simonelli wrote:
> > On Tuesday, February 25, 2025 11:19:04 AM EST Petr Mladek wrote:
> > > On Mon 2025-02-24 07:39:14, adamsimonelli@gmail.com wrote:
> > > > From: Adam Simonelli <adamsimonelli@gmail.com>
> > > >
> > > > The new config option, CONFIG_NULL_TTY_CONSOLE will allow ttynull to be
> > > > initialized by console_initcall() and selected as a possible console
> > > > device.
> > > >
> > > > diff --git a/drivers/tty/ttynull.c b/drivers/tty/ttynull.c
> > > > index 6b2f7208b564..ec3dd3fd41c0 100644
> > > > --- a/drivers/tty/ttynull.c
> > > > +++ b/drivers/tty/ttynull.c
> > > > @@ -57,6 +57,13 @@ static struct tty_driver *ttynull_device(struct console *c, int *index)
> > > > static struct console ttynull_console = {
> > > > .name = "ttynull",
> > > > .device = ttynull_device,
> > > > +
> > > > + /*
> > > > + * Match the index and flags from other boot consoles when CONFIG_NULL_TTY_CONSOLE is
> > > > + * enabled, otherwise, use the default values for the index and flags.
> > > > + */
> > > > + .index = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? -1 : 0,
> > >
> > > This should not be needed. "con->index" is always initialized to "0"
> > > for the default console, see:
> > >
> > OK, I had this in an #ifdef before, it was the cleanest way to set it to -1
> > that I could think of, other than the ifdef... If I still need this, I will try
> > to think of something else to set it to -1 when the option is enabled
>
> Ah, I was not clear enough. It should be perfectly fine to always
> statically initialize the value to -1. We should not need any
> #ifdef or IS_ENABLED.
>
> I mean to do:
>
> static struct console ttynull_console = {
> .name = "ttynull",
> .device = ttynull_device,
> .index = -1,
> };
>
> We might even do this in a separate patch. IMHO, it should have
> been done this way since the beginning.
>
OK, will do. This makes sense.
> > > static void try_enable_default_console(struct console *newcon)
> > > {
> > > if (newcon->index < 0)
> > > newcon->index = 0;
> > > [...]
> > > }
> > >
> > > > + .flags = IS_ENABLED(CONFIG_NULL_TTY_CONSOLE) ? CON_PRINTBUFFER : 0,
> > >
> > > This does not make much sense to me.
> > >
> > > CON_PRINTBUFFER prevents duplicated output when the same device has
> > > already been registered as a boot console. But ttynull does not have
> > > a boot console variant. Also it is a "null" device. It never prints
> > > anything. The output could never be duplicated by definition.
> > >
> > OK, I was duplicating what I saw in other consoles. I can try to remove it
>
> Again, I was not clear enough. My primary concern was that it did not make
> much sense to use the IS_ENABLED() check and initialize the value
> different way.
>
> Anyway, I would omit the flag. It is a NULL device. It does not matter
> whether it prints existing (old) messages during registration or not.
>
Understood, this makes sense.
> > > > };
> > > >
> > > My proposal is to call:
> > >
> > > #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> > > static int __init ttynull_default_console(void)
> > > {
> > > add_preferred_console("ttynull", 0, NULL);
> > > return 0;
> > > }
> > > console_initcall(ttynull_register);
> > > #endif
> > >
> > OK, actually in earlier revisions locally, I did actually have
> >
> >
> >
> > diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> > index dddb15f48d59..c1554a789de8 100644
> > --- a/kernel/printk/printk.c
> > +++ b/kernel/printk/printk.c
> > @@ -3712,6 +3712,11 @@ void __init console_init(void)
> > initcall_t call;
> > initcall_entry_t *ce;
> >
> > +#ifdef CONFIG_NULL_TTY_CONSOLE
> > + if (!strstr(boot_command_line, "console="))
> > + add_preferred_console("ttynull", 0, NULL);
>
> Good point! We should call add_preferred_console() only when
> the is no console= command line parameter. Otherwise, it could
> not get overridden by the command line.
>
> We could check "console_set_on_cmdline", similar to
> xenfb_make_preferred_console().
>
> > +#endif
> > +
> > /* Setup the default TTY line discipline. */
> > n_tty_init();
> >
> > Which worked as far as I could tell, at least on x86. Not sure if that was the
> > right place,
>
> I would prefer to keep it in drivers/tty/ttynull.c when possible.
> The following might do the trick:
>
> #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> static int __init ttynull_default_console(void)
> {
> if (!console_set_on_cmdline)
> add_preferred_console("ttynull", 0, NULL);
>
> return 0;
> }
> console_initcall(ttynull_register);
> #endif
>
> Best Regards,
> Petr
>
Thanks for that, that works.
On Wed, Feb 26, 2025 at 3:39 PM Adam Simonelli <adamsimonelli@gmail.com> wrote:
> On Tuesday, February 25, 2025 11:19:04 AM EST Petr Mladek wrote:
...
> > My proposal is to call:
> >
> > #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> > add_preferred_console("ttynull", 0, NULL);
> > #endif
> >
> > somewhere in the kernel code. The question is where.
> > I wonder if the following would work:
> >
>
> > #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
> > static int __init ttynull_default_console(void)
> > {
> > add_preferred_console("ttynull", 0, NULL);
> > return 0;
> > }
> > console_initcall(ttynull_register);
> > #endif
> >
> OK, actually in earlier revisions locally, I did actually have
>
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index dddb15f48d59..c1554a789de8 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -3712,6 +3712,11 @@ void __init console_init(void)
> initcall_t call;
> initcall_entry_t *ce;
>
> +#ifdef CONFIG_NULL_TTY_CONSOLE
> + if (!strstr(boot_command_line, "console="))
Just a side note: strstr() is fragile as theoretically "console=" can
be part of an argument unrelated to the console, like
foo="bar,baz,console=10,key=value". Although I haven't checked if this
is allowed by cmdline parser (lib/cmdline.c).
> + add_preferred_console("ttynull", 0, NULL);
> +#endif
> +
> /* Setup the default TTY line discipline. */
> n_tty_init();
>
>
>
> Which worked as far as I could tell, at least on x86. Not sure if that was the
> right place, and yeah, I was trying to better copy how CONFIG_VT_CONSOLE worked
> because I thought that was more correct.
--
With Best Regards,
Andy Shevchenko
© 2016 - 2025 Red Hat, Inc.