Re: [PATCH] RCU torture-testing kernel module

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

On Mon, Oct 24, 2005 at 11:10:58AM -0700, Badari Pulavarty wrote:
> On Mon, 2005-10-24 at 13:59 -0400, Kyle Moffett wrote:
> > On Oct 24, 2005, at 12:24:33, Badari Pulavarty wrote:
> > > Paul,
> > >
> > > I enabled RCU_TORTURE_TEST in 2.6.14-rc5-mm1. My machine took 10+  
> > > minutes to boot and let me login. RCU kthreads are hogging the  
> > > CPU.  Is this expected ?
> > 
> > Uhh...  It's a torture test.  What exactly do _you_ expect it will  
> > do?  I think the idea is to enable it as a module and load it when  
> > you want to start torture testing, and unload it when done.   
> > "TORTURE_TEST"s are not for production systems :-D.

Hey, and if you think that is fun, just try compiling it in
(CONFIG_RCU_TORTURE_TEST=y) and then build rcutorture.c as an
external module, then insmod'ing it!  You get two torture tests
running concurrently.  I found this out the hard way while
learning how the "tristate" directive works.

> I was expecting that - even if its compiled in, there would be
> a way to turn on/off the tests from /proc or something :)

Well, I submitted -that- patch a couple of weeks ago, and it was
roundly denounced for /proc pollution, hence the shiny new modules
implementation of it.

I must admit I was rather negative on the idea of using modules
for this sort of thing, but after actually trying it, I found that
it really works quite nicely.  The module loader even parses all
your arguments for you, so you can very easily parameterize the
tests.

						Thanx, Paul