Re: Mysterious CFQ crash and RCU

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

On Wed, May 25, 2011 at 07:44:51PM +0200, Paul Bolle wrote:
> On Wed, 2011-05-25 at 08:33 -0700, Paul E. McKenney wrote:
> > On Wed, May 25, 2011 at 12:17:16PM +0200, Paul Bolle wrote:
> > > 2) So shouldn't either the config entry be set to "tristate" or the
> > > module support removed from kernel/rcutree_trace.c?
> > 
> > No.
> > 
> > Just set CONFIG_RCU_TRACE=y and you will have RCU tracing.  Here is why:
> > 
> >[...]
> >
> > So you can set CONFIG_RCU_TRACE, and if you are running CONFIG_SMP=y,
> > you will have RCU tracing in your kernel.
> 
> The point - which I must have expressed poorly - is that there's a
> mismatch between the code in kernel/rcutree_trace.c (which allows for
> that file to be built as a module) and its kconfig setup (which requires
> it to be either builtin or not built at all, since it's boolean).
> Anyhow, your explanation makes clear that this kconfig setup is actually
> correct and that the module support in kernel/rcutree_trace.c might as
> well be removed, since that file can never be part of a module.

Ah -- this is using (perhaps abusing) the module capabilities to
get initialization and cleanup done.

Interesting point, though -- might be worth allowing this to be built
as a module.

Regardless, apologies for missing your point.

							Thanx, Paul