Re: [PATCH] rcu: Clean up rcu_init_nohz() by removing unnecessary statements

From: Paul E. McKenney
Date: Thu Mar 01 2018 - 20:10:29 EST


On Fri, Mar 02, 2018 at 09:01:19AM +0900, Byungchul Park wrote:
> On 3/1/2018 3:41 AM, Paul E. McKenney wrote:
> >On Wed, Feb 28, 2018 at 06:04:55PM +0900, Byungchul Park wrote:
> >>Since the commit 44c65ff2e3b0(rcu: Eliminate NOCBs CPU-state Kconfig
> >>options) made nocb-cpus identified only through the rcu_nocbs= boot
> >>parameter, we don't have to care NOCBs CPU-state Kconfig options
> >>anymore, which means now we can just rely on rcu_nocb_mask to
> >>decide whether going ahead in rcu_init_nohz().
> >>
> >>Remove the deprecated code.
> >>
> >>Signed-off-by: Byungchul Park <byungchul.park@xxxxxxx>
> >
> >Good catch! However, you missed a (relatively harmless) bug in my commit
> >44c65ff2e3b0, namely that if neither the nohz_full= nor the rcu_nocbs=
> >kernel boot parameters specify any CPUs, we don't need to allocate
> >rcu_nocb_mask. (That is, when both of those parameters are omitted.)
> >
> >Now, if the rcu_nocbs= kernel boot parameter was specified, we know that
> >rcu_nocb_mask was already allocated in rcu_nocb_setup(). So in
> >rcu_init_nohz() we only need to do the allocation if NO_HZ_FULL needs
> >some NOCBs CPUs, that is, when tick_nohz_full_running and when there
> >is at least one CPU specified in tick_nohz_full_mask.
>
> Why didn't I catch it in advance? :)

I guess that sentiment goes well with my "why didn't I avoid that bug
in the first place?" ;-)

> >So the change that is actually needed is to reverse the initialization
> >of need_rcu_nocb_mask, that is, to initialize it to false rather than
> >to true. I annotated your patch with my reasoning and have a patch
> >below with your Reported-by. Please take a look and let me know what
> >you think.
>
> No doubt. I agree with you.
>
> Acked-by: Byungchul Park <byungchul.park@xxxxxxx>

Applied, thank you!

Thanx, Paul

> >If I am not too confused, the only effect of this bug was to needlessly
> >allocate rcu_nocb_mask and to initialize it to all zeros bits, but please
> >let me know if I missed something.
>
> I think so as you.
>
> --
> Thanks,
> Byungchul
>