Re: [PATCH 18/18] rcu: Use pr_fmt to prefix "rcu: " to logging output

From: Joe Perches
Date: Mon May 14 2018 - 17:42:09 EST


On Mon, 2018-05-14 at 13:29 -0700, Paul E. McKenney wrote:
> On Thu, May 10, 2018 at 08:45:44AM -0700, Joe Perches wrote:
> > Use a consistent logging prefix for all rcu related output.
> >
> > Signed-off-by: Joe Perches <joe@xxxxxxxxxxx>
>
> I took parts of this (thank you!) but have concerns about other parts.
[]
> > diff --git a/kernel/rcu/rcuperf.c b/kernel/rcu/rcuperf.c
> > index 076a50fb22ad..ebdd77b45470 100644
> > --- a/kernel/rcu/rcuperf.c
> > +++ b/kernel/rcu/rcuperf.c
> > @@ -19,6 +19,9 @@
> > *
> > * Authors: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > */
> > +
> > +#define pr_fmt(fmt) "rcu: " fmt
>
> This is going to get us messages of the form "rcu: rcu-perf:", not?
> (And other odd combinations, depending on the flavor of RCU under test.)
> If so, this does not seem to be an improvement.

That depends on the existing embedded content of the format.
This will prefix just "rcu: " to pr_<level> output.

> > diff --git a/kernel/rcu/rcutorture.c b/kernel/rcu/rcutorture.c
[]
> > @@ -908,7 +909,7 @@ rcu_torture_writer(void *arg)
> > VERBOSE_TOROUT_STRING("rcu_torture_writer task started");
> > if (!can_expedite)
> > pr_alert("%s" TORTURE_FLAG
> > - " GP expediting controlled from boot/sysfs for %s.\n",
> > + " GP expediting controlled from boot/sysfs for %s\n",
> > torture_type, cur_ops->name);

As there is _no_ pr_fmt #defined in this file,
output will/could be prefixed with KBUILD_MODNAME ": "
(in this case "rcutorture: ") if/when the generic
pr_fmt conversion patch is applied.

> > } else if (gp_sync && !cur_ops->sync) {
> > - pr_alert("%s: gp_sync without primitives.\n", __func__);
> > + pr_alert("%s: gp_sync without primitives\n", __func__);
>
> I used a CDC Cyber 73 in the 1970s.

Fancy. I used a CDC 6400 and an IBM 1620, but
those were pretty old when I started.

> It had tiny memory by today's
> standards, but even it had periods in its error messages. We can easily
> afford them today, especially given that rcutorture is not included in
> small-memory Linux configurations.

OK, but I like consistency too.

~90 percent of linux logging does not use terminating periods.
For instance, on my laptop:

$ uptime -p
up 1 week, 1 day, 13 hours, 37 minute
$ dmesg | wc -l
4240
$ dmesg | grep -P "\w\.$"| wc -l
381

> > @@ -1711,11 +1712,11 @@ static void rcu_test_debug_objects(void)
> >
> > /* Wait for them all to get done so we can safely return. */
> > rcu_barrier();
> > - pr_alert("%s: WARN: Duplicate call_rcu() test complete.\n", KBUILD_MODNAME);
> > + pr_alert("WARN: Duplicate call_rcu() test complete\n");
>
> I would like to keep these, as they mark the region of console output where
> splats are expected.

The prefixes are still there...

> > if (cur_ops->fqs == NULL && fqs_duration != 0) {
> > - pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
> > + pr_alert("->fqs NULL and non-zero fqs_duration, fqs disabled\n");
>
> This I would like to keep. Easier to find.

One thing that you could use to validate the
output string format is after compilation:

$ strings kernel/rcu/rcutorture.o | grep -P "^[0-6]\w+:"

With your change, you will see duplicated prefixes.

cheers, Joe
>