Re: [PATCH] m68k/irq: don't use pr_crit in an header

Re: [PATCH] m68k/irq: don't use pr_crit in an header

[Geert Uytterhoeven]

2011/12/17 Uwe Kleine-König <u.kleine-koenig@pengutronix.de>:
> Using pr_crit in an header results in funny messages. Consider
>
>        #define pr_fmt(fmt) "mydriver: " fmt
>        #include <linux/hardirq.h>
>
> which makes the message from ack_bad_irq
>
>        mydriver: unexpected IRQ trap...
>
> so better use plain printk with KERN_CRIT directly.

Yep, that's expected behavior, as defining pr_fmt() modifies all kernel messages
generated from that module.

> This fixes a build problem on m68k with aufs3 en passant because the
> latter builds with
>
>        ccflags-y += -D'pr_fmt(fmt)=AUFS_NAME"\040%s:%d:%s[%d]:\040"fmt,__func__,__LINE__,current->comm,current->pid'
>
> without providing AUFS_NAME early enough for ack_bad_irq (which is the
> problem of aufs).

Isn't this a problem with (out of tree) aufs?
Why does it put a define that relies on an (apparently sometimes still
undefined)
variable on the build command line?
Any header may contain calls to pr_*().

> Cc: Thorsten Glaser <tg@debian.org>
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> ---
>  arch/m68k/include/asm/hardirq.h |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/arch/m68k/include/asm/hardirq.h b/arch/m68k/include/asm/hardirq.h
> index db30ed2..1f652e0 100644
> --- a/arch/m68k/include/asm/hardirq.h
> +++ b/arch/m68k/include/asm/hardirq.h
> @@ -20,7 +20,7 @@
>
>  static inline void ack_bad_irq(unsigned int irq)
>  {
> -       pr_crit("unexpected IRQ trap at vector %02x\n", irq);
> +       printk(KERN_CRIT "unexpected IRQ trap at vector %02x\n", irq);

Nack. Nowadays pr_crit(...) is recommended over "printk(KERN_CRIT ...)".

Besides, there are (albeit not that many yet) other callers of pr_*() in
header files. Do you plan to revert them to printk(), too?

Please fix aufs instead. Thanks!

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] m68k/irq: don't use pr_crit in an header

[Uwe Kleine-König]

Hi Geert,

(thanks for adding Joe to Cc:, I noticed that when I wanted to add him
myself :-)

On Sun, Dec 18, 2011 at 11:32:21AM +0100, Geert Uytterhoeven wrote:
> 2011/12/17 Uwe Kleine-König <u.kleine-koenig@pengutronix.de>:
> > Using pr_crit in an header results in funny messages. Consider
> >
> >        #define pr_fmt(fmt) "mydriver: " fmt
> >        #include <linux/hardirq.h>
> >
> > which makes the message from ack_bad_irq
> >
> >        mydriver: unexpected IRQ trap...
> >
> > so better use plain printk with KERN_CRIT directly.
> 
> Yep, that's expected behavior, as defining pr_fmt() modifies all kernel messages
> generated from that module.
I'm aware it is expected, I only wondered if it is also desirable to
have messages in headers modified depending on the module the header is
included in.
 
> > This fixes a build problem on m68k with aufs3 en passant because the
> > latter builds with
> >
> >        ccflags-y += -D'pr_fmt(fmt)=AUFS_NAME"\040%s:%d:%s[%d]:\040"fmt,__func__,__LINE__,current->comm,current->pid'
> >
> > without providing AUFS_NAME early enough for ack_bad_irq (which is the
> > problem of aufs).
> 
> Isn't this a problem with (out of tree) aufs?
> Why does it put a define that relies on an (apparently sometimes still
> undefined)
> variable on the build command line?
This is definitily a bug in aufs that needs fixing independant of the
issue of using or not using pr_... in headers.

> Any header may contain calls to pr_*().
> 
> > Cc: Thorsten Glaser <tg@debian.org>
> > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> > ---
> >  arch/m68k/include/asm/hardirq.h |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> >
> > diff --git a/arch/m68k/include/asm/hardirq.h b/arch/m68k/include/asm/hardirq.h
> > index db30ed2..1f652e0 100644
> > --- a/arch/m68k/include/asm/hardirq.h
> > +++ b/arch/m68k/include/asm/hardirq.h
> > @@ -20,7 +20,7 @@
> >
> >  static inline void ack_bad_irq(unsigned int irq)
> >  {
> > -       pr_crit("unexpected IRQ trap at vector %02x\n", irq);
> > +       printk(KERN_CRIT "unexpected IRQ trap at vector %02x\n", irq);
> 
> Nack. Nowadays pr_crit(...) is recommended over "printk(KERN_CRIT ...)".
I know that, I just wonder if the proponents of this recommendation are
aware of the issue when using pr_* in headers. Joe?
 
> Besides, there are (albeit not that many yet) other callers of pr_*() in
> header files. Do you plan to revert them to printk(), too?
That depends on the outcome of this discussion.

> Please fix aufs instead. Thanks!
I already provided a patch for that, too. (Currently only on the Debian
kernel ML.)

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

Re: [PATCH] m68k/irq: don't use pr_crit in an header

[Joe Perches]

On Sun, 2011-12-18 at 11:42 +0100, Uwe Kleine-König wrote:
> On Sun, Dec 18, 2011 at 11:32:21AM +0100, Geert Uytterhoeven wrote:
> > 2011/12/17 Uwe Kleine-König <u.kleine-koenig@pengutronix.de>:
> > > Using pr_crit in an header results in funny messages. Consider
> > >        #define pr_fmt(fmt) "mydriver: " fmt
> > >        #include <linux/hardirq.h>
> > > which makes the message from ack_bad_irq
> > >        mydriver: unexpected IRQ trap...
> > > so better use plain printk with KERN_CRIT directly.

Why or when is that inappropriate?

> I only wondered if it is also desirable to
> have messages in headers modified depending on the module the header is
> included in.
[]
> > Nack. Nowadays pr_crit(...) is recommended over "printk(KERN_CRIT ...)".
> I know that, I just wonder if the proponents of this recommendation are
> aware of the issue when using pr_* in headers. Joe?

I believe it to be a feature rather than a defect.

For instance:  commit 256ee435b9a9ee9cca69602fe8046b27ca99fbee

    netdevice: Convert printk to pr_info in netif_tx_stop_queue
    
    This allows any caller to be prefaced by any specific
    pr_fmt to better identify which device driver is using
    this function inappropriately. 

cheers, Joe