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:   Tue, 20 Mar 2018 16:58:35 +0800
From:   "Liu, Changcheng" <changcheng.liu@intel.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     tglx@linutronix.de, hpa@zytor.com, douly.fnst@cn.fujitsu.com,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] x86/ioapic: don't use unstable TSC to detect timer IRQ
Message-ID: <20180320085835.GA56497@sofia>
References: <20180320084255.GA187704@sofia>
 <20180320084929.GP4043@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180320084929.GP4043@hirez.programming.kicks-ass.net>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 09:49 Tue 20 Mar, Peter Zijlstra wrote:
> On Tue, Mar 20, 2018 at 04:42:55PM +0800, Liu, Changcheng wrote:
> > In rare case, the TSC is every unstable or can't sync with
> > real time hardware clock. 
> 
> However did you manage that? Please provide _FAR_ more details.
[Changcheng] TSC is simulated and HPET is hardware implemented.
TSC can't sync with HPET. When running linux, the TSC grows too
fast and HPET can't trigger periodic timer interrupt in time which
is used to update jiffies.

> 
> > diff --git a/arch/x86/include/asm/tsc.h b/arch/x86/include/asm/tsc.h
> > index cf5d53c..dcfc5b9 100644
> > --- a/arch/x86/include/asm/tsc.h
> > +++ b/arch/x86/include/asm/tsc.h
> > @@ -17,6 +17,7 @@ typedef unsigned long long cycles_t;
> >  
> >  extern unsigned int cpu_khz;
> >  extern unsigned int tsc_khz;
> > +extern int tsc_unstable;
> >  
> >  extern void disable_TSC(void);
> >  
> > diff --git a/arch/x86/kernel/apic/io_apic.c b/arch/x86/kernel/apic/io_apic.c
> > index 7c55387..0809ec6 100644
> > --- a/arch/x86/kernel/apic/io_apic.c
> > +++ b/arch/x86/kernel/apic/io_apic.c
> > @@ -1643,7 +1643,7 @@ static int __init timer_irq_works(void)
> >  	local_save_flags(flags);
> >  	local_irq_enable();
> >  
> > -	if (boot_cpu_has(X86_FEATURE_TSC))
> > +	if (boot_cpu_has(X86_FEATURE_TSC) && !tsc_unstable)
> >  		delay_with_tsc();
> >  	else
> >  		delay_without_tsc();
> > diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
> > index fb43027..27b1bae 100644
> > --- a/arch/x86/kernel/tsc.c
> > +++ b/arch/x86/kernel/tsc.c
> > @@ -36,7 +36,8 @@ EXPORT_SYMBOL(tsc_khz);
> >  /*
> >   * TSC can be unstable due to cpufreq or due to unsynced TSCs
> >   */
> > -static int __read_mostly tsc_unstable;
> > +int __read_mostly tsc_unstable;
> > +EXPORT_SYMBOL(tsc_unstable);
> >  
> >  /* native_sched_clock() is called before tsc_init(), so
> >     we must start with the TSC soft disabled to prevent
> 
> No, absolutely not. Even when the TSC is normally deemed unstable, which
> typically means it is not sync'ed between cores, it is still perfectly
> suitable to be used for delay loops.