Re: [PATCH 0/3] timers: Allocate per-cpu tvec_base's statically

[Viresh Kumar <viresh.kumar@linaro.org>]

On 30 March 2015 at 13:48, Peter Zijlstra <peterz@infradead.org> wrote:
> I did the below on top; I'll sit on these patches for a wee while in the
> hope of the blackfin maintainer explaining his reasons for wrecking
> ____cacheline_aligned.

Two things I wanted to mention:

1.) We may need to drop the patch 2/3 as that breaks build for xtensa and tile.

   kernel/time/timer.c: In function 'init_timers_cpu':
   kernel/time/timer.c:1539:1: error: section attribute cannot be
specified for local variables
   kernel/time/timer.c:1539:1: error: section attribute cannot be
specified for local variables
>> kernel/time/timer.c:1539:1: error: declaration of '__pcpu_unique___tvec_bases' with no linkage follows extern declaration
   kernel/time/timer.c:1539:1: note: previous declaration of
'__pcpu_unique___tvec_bases' was here
   kernel/time/timer.c:1539:9: error: section attribute cannot be
specified for local variables
   kernel/time/timer.c:1539:9: error: section attribute cannot be
specified for local variables
   kernel/time/timer.c:1539:9: error: weak declaration of
'__tvec_bases' must be public
   kernel/time/timer.c:1539:9: error: declaration of '__tvec_bases'
with no linkage follows extern declaration
   kernel/time/timer.c:1539:9: note: previous declaration of
'__tvec_bases' was here

Can these be fixed somehow ?


2.) We probably don't need to wait for the blackfin guys as we
haven't broken anything yet. It will be broken only once we start
using an extra bit from base pointer, which will be done by a separate
patch trying to migrate running timers.

> ---
> Subject: timer: Further simplify SMP and HOTPLUG logic
> From: Peter Zijlstra <peterz@infradead.org>
> Date: Mon Mar 30 10:09:19 CEST 2015
>
> Remove one CONFIG_HOTPLUG_CPU #ifdef in trade for introducing one
> CONFIG_SMP #ifdef.
>
> The CONFIG_SMP ifdef avoids declaring the per-cpu __tvec_bases storage
> on UP systems since they already have boot_tvec_bases.
>
> Also (re)add a runtime check on the base alignment -- for the paranoid
> amongst us :-)
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  kernel/time/timer.c |   28 +++++++++++++---------------
>  1 file changed, 13 insertions(+), 15 deletions(-)

Reviewed-by: Viresh Kumar <viresh.kumar@linaro.org>