Re: [PATCH -v2 2/3] ACPI, APEI, Add APEI generic error status print support

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 30 Nov 2010 10:00:47 -0800 "Luck, Tony" <tony.luck@intel.com> wrote:

> > +	for (i = 0; i < strs_size; i++) {
> > +		if (!(bits & (1U << i)))
> > +			continue;
> > +		str = strs[i];
> > +		if (len && len + strlen(str) + 2 > 80) {
> > +			printk("\n");
> > +			len = 0;
> > +		}
> > +		if (!len)
> > +			len = pr_pfx(pfx, "%s", str);
> > +		else
> > +			len += printk(", %s", str);
> > +	}
> > +	if (len)
> > +		printk("\n");
> 
> Does printk() offer any guarantees about getting all the characters
> from a single printk() call out to the console without interleaving
> with messages from printk() calls on other cpus?

Yes, it uses logbuf_lock to atomically append all the output from a
printk into log_buf[].  Then it calls the console drivers against
log_buf[].  It's hard to see how a console driver could then screw that
up.

>  If it does, then
> it would be a good idea to sprintf() the parts of this message to
> a buffer and then use one printk() call.

yup.

>  I think I read that netconsole
> ends up with one packet on the wire for each call to printk().
> 
> Trying to parse output jumbled together from multiple cpus
> doesn't sound like fun.
> 
> -Tony