Re: [PATCH v3] omap: dmtimer: convert printk to pr_err

[Joe Perches <joe@perches.com>]

On Fri, 2011-10-07 at 10:22 +0100, Russell King - ARM Linux wrote:
> On Fri, Oct 07, 2011 at 10:50:16AM +0200, Víctor Manuel Jáquez Leal wrote:
> > Convert a printk(KERN_ERR) message in the driver to pr_err().
> ...
> > @@ -111,7 +111,7 @@ static void omap_dm_timer_wait_for_reset(struct omap_dm_timer *timer)
> >  	while (!(__raw_readl(timer->sys_stat) & 1)) {
> >  		c++;
> >  		if (c > 100000) {
> > -			printk(KERN_ERR "Timer failed to reset\n");
> > +			pr_err("Timer failed to reset\n");
> 
> What is the reason behind this change?  It looks like it's to use the
> latest and greatest function.

Hi Russell

I'm not promoting this patch, just commenting.

At some point in the next couple of years, I want
to convert all of, or as many as possible of, the
remaining printk uses to pr_<level>.

This would allow finer grained control over the
prefixing of KBUILD_MODNAME and __func__, and
could possibly make the kernel image smaller.

Today, arch/arm has ~3:1 ratio of printk to pr_<level>.
grep shows 1427 printks, 468 pr_<level>, 405 pr_debug's.

> If so, please don't make these changes - we have on many occasions been
> blamed for size of diffstat, churn, needless change, and this patch is
> exactly that.

True.

These trivial changes could wait until arch/arm settles
down a bit more.

> By all means fix printk's without KERN_ constants,

There's still more than 250 of those in arch/arm.
Even more with the uses of secondary things like:
#define PRINTK(x...) (foo && printk(x))

> possibly converting
> them to pr_*, but don't go around replacing printk(KERN_* with pr_*(
> without ensuring that there's a real benefit to the change.

Style consistency patches do need to be governed by
acceptable churn rate.