[PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Steven Rostedt]

From: Steven Rostedt (VMware) <rostedt@goodmis.org>

There's been discussion on the fb list about the addition of
WARN_CONSOLE_UNLOCKED() inside the fb code. The complaint is that when
the fb module is loaded with lockless_register_fb the console lock is
not taken for debugging reasons. With the addition of
WARN_CONSOLE_UNLOCK() within the fb code, this causes the console to
fill up with warnings when trying to debug the fb driver.

There's also a #if 1 that enables the warning which was added before
git history, and we look down on constant #if's in the kernel nowadays
anyway.

Remove the #if 1 and add a ignore_console_lock_warning boolean that can
be set by drivers to ignore the warning in order to do debugging.

Link: http://lkml.kernel.org/r/717e6337-e7a6-7a92-1c1b-8929a25696b5@suse.de
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
diff --git a/include/linux/console.h b/include/linux/console.h
index dfd6b0e97855..9d4a6b985154 100644
--- a/include/linux/console.h
+++ b/include/linux/console.h
@@ -200,11 +200,9 @@ void vcs_make_sysfs(int index);
 void vcs_remove_sysfs(int index);
 
 /* Some debug stub to catch some of the obvious races in the VT code */
-#if 1
-#define WARN_CONSOLE_UNLOCKED()	WARN_ON(!is_console_locked() && !oops_in_progress)
-#else
-#define WARN_CONSOLE_UNLOCKED()
-#endif
+extern bool ignore_console_lock_warning;
+#define WARN_CONSOLE_UNLOCKED()						\
+	WARN_ON(!ignore_console_lock_warning && !is_console_locked() && !oops_in_progress)
 
 /* VESA Blanking Levels */
 #define VESA_NO_BLANKING        0
diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index 247808333ba4..d9056efc3739 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -66,6 +66,9 @@ int console_printk[4] = {
 	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
 };
 
+bool ignore_console_lock_warning __read_mostly;
+EXPORT_SYMBOL(ignore_console_lock_warning);
+
 /*
  * Low level drivers may need that to know if they can schedule in
  * their unblank() callback or not. So let's export it.

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Hans de Goede]

Hi,

On 11-07-18 21:17, Steven Rostedt wrote:
> 
> From: Steven Rostedt (VMware) <rostedt@goodmis.org>
> 
> There's been discussion on the fb list about the addition of
> WARN_CONSOLE_UNLOCKED() inside the fb code. The complaint is that when
> the fb module is loaded with lockless_register_fb the console lock is
> not taken for debugging reasons. With the addition of
> WARN_CONSOLE_UNLOCK() within the fb code, this causes the console to
> fill up with warnings when trying to debug the fb driver.
> 
> There's also a #if 1 that enables the warning which was added before
> git history, and we look down on constant #if's in the kernel nowadays
> anyway.
> 
> Remove the #if 1 and add a ignore_console_lock_warning boolean that can
> be set by drivers to ignore the warning in order to do debugging.
> 
> Link: http://lkml.kernel.org/r/717e6337-e7a6-7a92-1c1b-8929a25696b5@suse.de
> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

Based on the fbdev discussion this seems a good idea to me
and codewise this looks good to:

Reviewed-by: Hans de Goede <hdegoede@redhat.com>

Regards,

Hans


> ---
> diff --git a/include/linux/console.h b/include/linux/console.h
> index dfd6b0e97855..9d4a6b985154 100644
> --- a/include/linux/console.h
> +++ b/include/linux/console.h
> @@ -200,11 +200,9 @@ void vcs_make_sysfs(int index);
>   void vcs_remove_sysfs(int index);
>   
>   /* Some debug stub to catch some of the obvious races in the VT code */
> -#if 1
> -#define WARN_CONSOLE_UNLOCKED()	WARN_ON(!is_console_locked() && !oops_in_progress)
> -#else
> -#define WARN_CONSOLE_UNLOCKED()
> -#endif
> +extern bool ignore_console_lock_warning;
> +#define WARN_CONSOLE_UNLOCKED()						\
> +	WARN_ON(!ignore_console_lock_warning && !is_console_locked() && !oops_in_progress)
>   
>   /* VESA Blanking Levels */
>   #define VESA_NO_BLANKING        0
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index 247808333ba4..d9056efc3739 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -66,6 +66,9 @@ int console_printk[4] = {
>   	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
>   };
>   
> +bool ignore_console_lock_warning __read_mostly;
> +EXPORT_SYMBOL(ignore_console_lock_warning);
> +
>   /*
>    * Low level drivers may need that to know if they can schedule in
>    * their unblank() callback or not. So let's export it.
> 

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Sergey Senozhatsky]

On (07/11/18 15:17), Steven Rostedt wrote:
> +bool ignore_console_lock_warning __read_mostly;
> +EXPORT_SYMBOL(ignore_console_lock_warning);

OK. So, to recap,
We made is_console_locked() EXPORT_SYMBOL recently [it's still in linux-next],
so people could use WARN_CONSOLE_UNLOCKED in more places; this made other
people unhappy, so now we add another EXPORT_SYMBOL to the picture, which will
disable those newly added WARN_CONSOLE_UNLOCKED and make other people happy
again.

This makes me wonder - do we want to add more WARN_CONSOLE_UNLOCKED in the
first place? :)

Other than that, the patch looks OK to me I guess. I'm not super happy with
more printk EXPORT_SYMBOL-s, frankly speaking, I'm not entirely in love with
the "add a bool flag to suppress warn print outs which we added in the
previous patch" direction, but if you guys want/need it...
Probably I'm just dramatizing it, as usual ;)
The removal of "#if 1" is definitely nice.

Let's hear from Petr.

	-ss

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Steven Rostedt]

On Thu, 12 Jul 2018 10:58:48 +0900
Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> wrote:

> On (07/11/18 15:17), Steven Rostedt wrote:
> > +bool ignore_console_lock_warning __read_mostly;
> > +EXPORT_SYMBOL(ignore_console_lock_warning);  
> 
> OK. So, to recap,
> We made is_console_locked() EXPORT_SYMBOL recently [it's still in linux-next],
> so people could use WARN_CONSOLE_UNLOCKED in more places; this made other
> people unhappy, so now we add another EXPORT_SYMBOL to the picture, which will
> disable those newly added WARN_CONSOLE_UNLOCKED and make other people happy
> again.

Note, it only made people that added a module parameter that disables
grabbing the console lock in the first place for debugging purposes
only. The added WARN_CONSOLE_UNLOCKED() are good. Nobody complaining
that they exist. They are complaining that it breaks one of their
debugging work flows, and need to disable it when they are doing so.

> 
> This makes me wonder - do we want to add more WARN_CONSOLE_UNLOCKED in the
> first place? :)

I would say yes.

> 
> Other than that, the patch looks OK to me I guess. I'm not super happy with
> more printk EXPORT_SYMBOL-s, frankly speaking, I'm not entirely in love with
> the "add a bool flag to suppress warn print outs which we added in the
> previous patch" direction, but if you guys want/need it...

I could add a comment explaining why it exists. Something like:

/*
 * Set this is you need to quiet WARN_CONSOLE_UNLOCKED() for debugging
 * purposes.
 */

> Probably I'm just dramatizing it, as usual ;)
> The removal of "#if 1" is definitely nice.

Yeah, that had to go regardless.

> 
> Let's hear from Petr.

Sure.

-- Steve

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Sergey Senozhatsky]

On (07/11/18 23:09), Steven Rostedt wrote:
> > On (07/11/18 15:17), Steven Rostedt wrote:
> > > +bool ignore_console_lock_warning __read_mostly;
> > > +EXPORT_SYMBOL(ignore_console_lock_warning);  
> > 
> > OK. So, to recap,
> > We made is_console_locked() EXPORT_SYMBOL recently [it's still in linux-next],
> > so people could use WARN_CONSOLE_UNLOCKED in more places; this made other
> > people unhappy, so now we add another EXPORT_SYMBOL to the picture, which will
> > disable those newly added WARN_CONSOLE_UNLOCKED and make other people happy
> > again.
> 
> Note, it only made people that added a module parameter that disables
> grabbing the console lock in the first place for debugging purposes
> only. The added WARN_CONSOLE_UNLOCKED() are good. Nobody complaining
> that they exist. They are complaining that it breaks one of their
> debugging work flows, and need to disable it when they are doing so.

True, but at the same time nobody complained that we didn't have those
extra WARN_CONSOLE_UNLOCKED()-s. Just saying. Over all I agree that there
is probably some value in those extra WARN-s.

> I could add a comment explaining why it exists. Something like:
> 
> /*
>  * Set this is you need to quiet WARN_CONSOLE_UNLOCKED() for debugging
>  * purposes.
>  */

s/is you/if you/

Suggested-by: One-of-those-commiters@YouKnowWhoYouAre.org ;)

	-ss

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Petr Mladek]

On Thu 2018-07-12 13:25:52, Sergey Senozhatsky wrote:
> On (07/11/18 23:09), Steven Rostedt wrote:
> > > On (07/11/18 15:17), Steven Rostedt wrote:
> > > > +bool ignore_console_lock_warning __read_mostly;
> > > > +EXPORT_SYMBOL(ignore_console_lock_warning);  
> > > 
> > > OK. So, to recap,
> > > We made is_console_locked() EXPORT_SYMBOL recently [it's still in linux-next],
> > > so people could use WARN_CONSOLE_UNLOCKED in more places; this made other
> > > people unhappy, so now we add another EXPORT_SYMBOL to the picture, which will
> > > disable those newly added WARN_CONSOLE_UNLOCKED and make other people happy
> > > again.
> > 
> > Note, it only made people that added a module parameter that disables
> > grabbing the console lock in the first place for debugging purposes
> > only. The added WARN_CONSOLE_UNLOCKED() are good. Nobody complaining
> > that they exist. They are complaining that it breaks one of their
> > debugging work flows, and need to disable it when they are doing so.
> 
> True, but at the same time nobody complained that we didn't have those
> extra WARN_CONSOLE_UNLOCKED()-s. Just saying. Over all I agree that there
> is probably some value in those extra WARN-s.

I agree that there is some value in the extra WARN's. Also it seems
that it is useful to disable them for debugging purposes. I do not
see a better solution. Therefore I am fine with this one.


> > I could add a comment explaining why it exists. Something like:
> > 
> > /*
> >  * Set this is you need to quiet WARN_CONSOLE_UNLOCKED() for debugging
> >  * purposes.
> >  */
> 
> s/is you/if you/

Yes, this extra comment would be appreciated.

If you send v2, please, break also the WARN definition to 80-chars's
per line ;-)


> Suggested-by: One-of-those-commiters@YouKnowWhoYouAre.org ;)

AFAIK, the change was actually suggested by Steven. The inspiration
is described by the line:

Link:
http://lkml.kernel.org/r/717e6337-e7a6-7a92-1c1b-8929a25696b5@suse.de


Finally, IMHO, it would make sense to push this change together with
the fbcon/vt side.

Best Regards,
Petr

Re: [PATCH] console: Replace #if 1 with a bool to ignore WARN_CONSOLE_UNLOCKED()

[Sergey Senozhatsky]

On (07/12/18 09:53), Petr Mladek wrote:
> > 
> > True, but at the same time nobody complained that we didn't have those
> > extra WARN_CONSOLE_UNLOCKED()-s. Just saying. Over all I agree that there
> > is probably some value in those extra WARN-s.
> 
> I agree that there is some value in the extra WARN's. Also it seems
> that it is useful to disable them for debugging purposes. I do not
> see a better solution. Therefore I am fine with this one.

Good.

> Yes, this extra comment would be appreciated.

Yep.

> If you send v2, please, break also the WARN definition to 80-chars's
> per line ;-)

Yeah, I noticed that too.

Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>

	-ss