Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 3 Feb 2020 10:15:16 +0530
From:   Gautham R Shenoy <ego@linux.vnet.ibm.com>
To:     Nathan Lynch <nathanl@linux.ibm.com>
Cc:     "Gautham R. Shenoy" <ego@linux.vnet.ibm.com>,
        linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org,
        Michael Ellerman <mpe@ellerman.id.au>,
        Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>,
        Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>,
        "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com>,
        Tyrel Datwyler <tyreld@linux.ibm.com>
Subject: Re: [PATCH 1/3] powerpc/pseries: Account for SPURR ticks on idle CPUs
Message-ID: <20200203044516.GA13468@in.ibm.com>
Reply-To: ego@linux.vnet.ibm.com
References: <1574856072-30972-1-git-send-email-ego@linux.vnet.ibm.com>
 <1574856072-30972-2-git-send-email-ego@linux.vnet.ibm.com>
 <87o8wnu3t7.fsf@linux.ibm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <87o8wnu3t7.fsf@linux.ibm.com>
User-Agent: Mutt/1.5.23 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hello Nathan,

On Wed, Dec 04, 2019 at 04:24:52PM -0600, Nathan Lynch wrote:
> "Gautham R. Shenoy" <ego@linux.vnet.ibm.com> writes:
> > diff --git a/arch/powerpc/kernel/idle.c b/arch/powerpc/kernel/idle.c
> > index a36fd05..708ec68 100644
> > --- a/arch/powerpc/kernel/idle.c
> > +++ b/arch/powerpc/kernel/idle.c
> > @@ -33,6 +33,8 @@
> >  unsigned long cpuidle_disable = IDLE_NO_OVERRIDE;
> >  EXPORT_SYMBOL(cpuidle_disable);
> >  
> > +DEFINE_PER_CPU(u64, idle_spurr_cycles);
> > +
> 
> Does idle_spurr_cycles need any special treatment for CPU
> online/offline?

If offline uses extended cede, then we need to take a snapshot of the
idle_spurr_cycles before going offline and add the delta once we are
back online. However, since the plan is to deprecate the use of
extended cede for CPU-Offline and use only rtas-stop-self, we don't
need any special handling there.


> 
> >  static int __init powersave_off(char *arg)
> >  {
> >  	ppc_md.power_save = NULL;
> > diff --git a/drivers/cpuidle/cpuidle-pseries.c b/drivers/cpuidle/cpuidle-pseries.c
> > index 74c2479..45e2be4 100644
> > --- a/drivers/cpuidle/cpuidle-pseries.c
> > +++ b/drivers/cpuidle/cpuidle-pseries.c
> > @@ -30,11 +30,14 @@ struct cpuidle_driver pseries_idle_driver = {
> >  static struct cpuidle_state *cpuidle_state_table __read_mostly;
> >  static u64 snooze_timeout __read_mostly;
> >  static bool snooze_timeout_en __read_mostly;
> > +DECLARE_PER_CPU(u64, idle_spurr_cycles);
> 
> This belongs in a header...

Will move it to the header file.

> 
> 
> > -static inline void idle_loop_prolog(unsigned long *in_purr)
> > +static inline void idle_loop_prolog(unsigned long *in_purr,
> > +				    unsigned long *in_spurr)
> >  {
> >  	ppc64_runlatch_off();
> >  	*in_purr = mfspr(SPRN_PURR);
> > +	*in_spurr = mfspr(SPRN_SPURR);
> >  	/*
> >  	 * Indicate to the HV that we are idle. Now would be
> >  	 * a good time to find other work to dispatch.
> > @@ -42,13 +45,16 @@ static inline void idle_loop_prolog(unsigned long *in_purr)
> >  	get_lppaca()->idle = 1;
> >  }
> >  
> > -static inline void idle_loop_epilog(unsigned long in_purr)
> > +static inline void idle_loop_epilog(unsigned long in_purr,
> > +				    unsigned long in_spurr)
> >  {
> >  	u64 wait_cycles;
> > +	u64 *idle_spurr_cycles_ptr = this_cpu_ptr(&idle_spurr_cycles);
> >  
> >  	wait_cycles = be64_to_cpu(get_lppaca()->wait_state_cycles);
> >  	wait_cycles += mfspr(SPRN_PURR) - in_purr;
> >  	get_lppaca()->wait_state_cycles = cpu_to_be64(wait_cycles);
> > +	*idle_spurr_cycles_ptr += mfspr(SPRN_SPURR) - in_spurr;
> 
> ... and the sampling and increment logic probably should be further
> encapsulated in accessor functions that can be used in both the cpuidle
> driver and the default/generic idle implementation. Or is there some
> reason this is specific to the pseries cpuidle driver?

I am not sure if we use SPURR and PURR for performing accounting on
Bare-Metal systems. IIUC, the patches proposed by Kamalesh is only to
use idle_[s]purr and [s]purr on POWERVM LPARs. This is why I coded the
sampling/increment logic in the pseries cpuidle driver. But you are
right, in the absence of cpuidle, when we use the default idle
implementation, we will still need to note the value of
idle_purr/spurr.

--
Thanks and Regards
gautham.