Re: [PATCH tip/core/rcu 19/55] rcutorture: Abstract TOROUT_STRING() and friends

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Feb 17, 2014 at 05:23:18PM -0800, Josh Triplett wrote:
> On Mon, Feb 17, 2014 at 04:58:02PM -0800, Paul E. McKenney wrote:
> > On Mon, Feb 17, 2014 at 04:26:06PM -0800, Josh Triplett wrote:
> > > On Mon, Feb 17, 2014 at 02:12:23PM -0800, Paul E. McKenney wrote:
> > > > From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> > > > 
> > > > These diagnostic macros are not confined to torturing RCU, so this commit
> > > > makes them available to other torture tests.
> > > > 
> > > > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > 
> > > This removes the do {} while (0) around the first of the macros, without
> > > any explanation in the commit message.
> > 
> > checkpatch.pl yelled at me about it.  ;-)
> 
> That's a fine reason, but the change needs documenting in the commit
> message.
> 
> > > Also, to what extent could these be made redundant with pr_fmt?
> > > if(verbose) seems like it really ought to become a priority level or
> > > dynamic debugging (and it's really awful to further propagate macros
> > > that reference an out-of-macro variable).  Ideally all of these could go
> > > away in favor of pr_fmt, and then the individual calls to them would
> > > become pr_alert, pr_debug, pr_error, or similar.
> > 
> > I don't immediately see how to get the "!!!" to be there or not with
> > pr_fmt(), since the same pr_fmt() would apply to all the macros.
> 
> Replace VERBOSE_TOROUT_ERRSTRING with something like pr_emerg, or leave
> it as pr_alert and drop the priority of the other variations to
> something less; then instead of "!!!" you have the existing message
> priority mechanism.

Is this message-priority mechanism apparent over a serial console?
As in can a script reading the serial console output tell what the
message priority is?

> Also, if it helps, note that you can add parameters in pr_fmt, as long
> as they're the initial parameters and you add corresponding format
> specifiers before the passed format string.  (Many modules do this to
> add __func__, for instance.)

Agreed, and I do see some uses of this sort.

> > I am also not all that excited about including "verbose" or similar
> > into the argument list of all calls to these guys.
> 
> verbose could go away entirely in favor of dynamic debug; you could then
> turn debug on or off for the relevant torture test module instead of
> setting verbose.

But I need a pretty fully functional userspace to control the dynamic
debug from what I can see.  I really would like to avoid relying on
having much of any userspace during rcutorture and locktorture testing.
The reason is that avoiding a userspace greatly reduces my memory and
disk requirements, allowing me to run more instances of rcutorture and
locktorture in parallel on smaller systems.

							Thanx, Paul