Re: [PATCH] mm: Fix printk format strings

["Vegard Nossum" <vegard.nossum@gmail.com>]

On 10/11/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> On Thu, 11 Oct 2007 09:04:57 +0200 "Vegard Nossum" <vegard.nossum@gmail.com> wrote:
>
> > > > -     printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
> > > > -             " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
> > > > +     printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n",
> > > >               global_page_state(NR_ACTIVE),
> > > >               global_page_state(NR_INACTIVE),
> > > >               global_page_state(NR_FILE_DIRTY),
> > > >               global_page_state(NR_WRITEBACK),
> > > > -             global_page_state(NR_UNSTABLE_NFS),
> > > > +             global_page_state(NR_UNSTABLE_NFS));
> > > > +     printk(" free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
> > > >               global_page_state(NR_FREE_PAGES),
> > > >               global_page_state(NR_SLAB_RECLAIMABLE) +
> > > >                       global_page_state(NR_SLAB_UNRECLAIMABLE),
> > >
> > > I don't understand the reason for this change.
> >
> > I'm sorry :). It helps to make one line per call only, since this
> > allows changing the printk internals for the better by reducing some
> > complexity. I believe this is a good thing. I have a patch that
> > changes printk, but it assumes that each format string only contains a
> > single line. Is this a very bad assumption to make? Or maybe I should
> > have sent that change first and made a reference to it?
> >
> > But don't you also agree, on the grounds of principle, that a single
> > line per call is better style?
>
> Well we do multiple-lines-per-printk in rather a lot of places and it has
> two advantages:
>
> - less text size (I expect)
>
> - the printk is "atomic" in that the logically-connected output lines
>   won't get tangled up with an intervening printk from another CPU, or from
>   an interrupt handler on this CPU.
>
> Those are pretty modest advantages and I guess we could live without them
> if we got a significant gain from doing so.  But it'd take quite some
> effort hunting down all the callsites, and preventing new ones from
> occuring.

I have written a sparse program to detect all callers that don't use
single-line literal strings for the printk format (posted to sparse
mailing list yesterday). So in fact, what I was doing was just that,
hunting them down and cleaning it up.

It so happens that my other patch also introduce a buffered printk
that supports atomic line continuations and multiple uninterrupted
lines, though this requires of course that the callers use these
functions. Other benefits include writing directly to the ring-buffer,
saving a 1K temporary buffer. The printk patch(es) can be viewed at
http://folk.uio.no/vegardno/xprintf/

Vegard