[PATCH v4 04/11] x86/intel_pstate: avoid calling cpufreq_add_cpu() twice

[PATCH v4 04/11] x86/intel_pstate: avoid calling cpufreq_add_cpu() twice

[Wei Wang]

cpufreq_add_cpu() is already called in the hypercall code path
(the bottom of set_px_pminfo() and inside cpufreq_cpu_init()).
So, we remove the redundant calling here.

v4 changes:
None.

Signed-off-by: Wei Wang <wei.w.wang@intel.com>
---
 xen/drivers/cpufreq/cpufreq.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/xen/drivers/cpufreq/cpufreq.c b/xen/drivers/cpufreq/cpufreq.c
index ab66884..91b6c25 100644
--- a/xen/drivers/cpufreq/cpufreq.c
+++ b/xen/drivers/cpufreq/cpufreq.c
@@ -632,8 +632,6 @@ static struct notifier_block cpu_nfb = {
 
 static int __init cpufreq_presmp_init(void)
 {
-    void *cpu = (void *)(long)smp_processor_id();
-    cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
     register_cpu_notifier(&cpu_nfb);
     return 0;
 }
-- 
1.9.1

Re: [PATCH v4 04/11] x86/intel_pstate: avoid calling cpufreq_add_cpu() twice

[Jan Beulich]

>>> On 25.06.15 at 13:15, <wei.w.wang@intel.com> wrote:
> cpufreq_add_cpu() is already called in the hypercall code path
> (the bottom of set_px_pminfo() and inside cpufreq_cpu_init()).
> So, we remove the redundant calling here.

While I can see that currently the call is kind of pointless (as it can't
do anything useful before Dom0 communicated the data obtained
from ACPI), it's still logically correct to call the callback on the BP
prior to registering a hook for AP bringup. Otherwise you could (and
perhaps should) as well defer the CPU notifier registration.

Otoh now that you're trying to introduce a driver independent of
ACPI (and hence initialized at boot time) I wonder why you don't
make use of what is here instead of deleting it.

Jan

> --- a/xen/drivers/cpufreq/cpufreq.c
> +++ b/xen/drivers/cpufreq/cpufreq.c
> @@ -632,8 +632,6 @@ static struct notifier_block cpu_nfb = {
>  
>  static int __init cpufreq_presmp_init(void)
>  {
> -    void *cpu = (void *)(long)smp_processor_id();
> -    cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
>      register_cpu_notifier(&cpu_nfb);
>      return 0;
>  }

Re: [PATCH v4 04/11] x86/intel_pstate: avoid calling cpufreq_add_cpu() twice

[Wang, Wei W]

On 24/07/2015 21:36,  Jan Beulich wrote:
> >>> On 25.06.15 at 13:15, <wei.w.wang@intel.com> wrote:
> > cpufreq_add_cpu() is already called in the hypercall code path (the
> > bottom of set_px_pminfo() and inside cpufreq_cpu_init()).
> > So, we remove the redundant calling here.
> 
> While I can see that currently the call is kind of pointless (as it can't do
> anything useful before Dom0 communicated the data obtained from ACPI),
> it's still logically correct to call the callback on the BP prior to registering a hook
> for AP bringup. Otherwise you could (and perhaps should) as well defer the
> CPU notifier registration.
> 
> Otoh now that you're trying to introduce a driver independent of ACPI (and
> hence initialized at boot time) I wonder why you don't make use of what is
> here instead of deleting it.
> 

Ok, I will roll back to leave cpufreq_presmp_init() there.

Best,
Wei
> 
> > --- a/xen/drivers/cpufreq/cpufreq.c
> > +++ b/xen/drivers/cpufreq/cpufreq.c
> > @@ -632,8 +632,6 @@ static struct notifier_block cpu_nfb = {
> >
> >  static int __init cpufreq_presmp_init(void)  {
> > -    void *cpu = (void *)(long)smp_processor_id();
> > -    cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
> >      register_cpu_notifier(&cpu_nfb);
> >      return 0;
> >  }
>