Re: [ANNOUNCE] 3.14.39-rt38

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 28 May 2015 11:05:58 -0400
Paul Gortmaker <paul.gortmaker@windriver.com> wrote:

> On 15-05-20 10:55 AM, Steven Rostedt wrote:
> > 
> > Dear RT Folks,
> > 
> > I'm pleased to announce the 3.14.39-rt38 stable release.
> 
> [...]
> 
> > 
> > Changes from v3.14.39-rt37:
> > 
> > ---
> > 
> > Marcelo Tosatti (2):
> >       KVM: lapic: mark LAPIC timer handler as irqsafe
> >       KVM: use simple waitqueue for vcpu->wq
> > 
> > Mike Galbraith (2):
> >       hotplug: Use set_cpus_allowed_ptr() in sync_unplug_thread()
> >       rt, nohz_full: fix nohz_full for PREEMPT_RT_FULL
> > 
> > Sebastian Andrzej Siewior (1):
> >       kernel/irq_work: fix no_hz deadlock
> 
> Seems we can't use this commit above as-is on 3.14 (and earlier?)
> since we'll get this:
> 
> kernel/time/tick-sched.c: In function ‘can_stop_full_tick’:
> kernel/time/tick-sched.c:183:2: error: implicit declaration of function ‘arch_irq_work_has_interrupt’ [-Werror=implicit-function-declaration]
>   if (!arch_irq_work_has_interrupt()) {
>   ^
> cc1: some warnings being treated as errors
> make[2]: *** [kernel/time/tick-sched.o] Error 1
> 
> Problem arises because that fcn doesn't come in until 3.18+ with
> the merge v3.18-rc1~114 (in the 2nd commit):
> 
> git log --oneline --reverse v3.18-rc1~114^2 ^v3.18-rc1~114^
>  a80e49e2cc31 nohz: Move nohz full init call to tick init
>  c5c38ef3d703 irq_work: Introduce arch_irq_work_has_interrupt()
>  76a33061b932 irq_work: Force raised irq work to run on irq work interrupt
>  3010279f0fc3 x86: Tell irq work about self IPI support
>  09f6edd42421 arm: Tell irq work about self IPI support
>  3631073659d0 arm64: Tell irq work about self IPI support
>  4327b15f64b2 nohz: Consolidate nohz full init code
>  9b01f5bf3999 nohz: nohz full depends on irq work self IPI support
> 
> Not sure whether we want to backport more stuff or change the
> original backport to not use arch_irq_work_has_interrupt...
> 
> Config is basically a defconfig but with RT_FULL and NOHZ_FULL =y.

Grumble, I need to add NOHZ_FULL to my test configs. OK, I'll have to
look into this. Thanks for the report.

-- Steve

>