x86, vmi: TSC going backwards check in vmi clocksource

x86, vmi: TSC going backwards check in vmi clocksource

[Alok Kataria]

From: Alok N Kataria <akataria@vmware.com>

Similar to the check for TSC going backwards in the TSC clocksource, we also
need this check for VMI clocksource.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Cc: stable@kernel.org
---

 arch/x86/kernel/vmiclock_32.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)


diff --git a/arch/x86/kernel/vmiclock_32.c b/arch/x86/kernel/vmiclock_32.c
index c4c1f9e..bde106c 100644
--- a/arch/x86/kernel/vmiclock_32.c
+++ b/arch/x86/kernel/vmiclock_32.c
@@ -283,10 +283,13 @@ void __devinit vmi_time_ap_init(void)
 #endif
 
 /** vmi clocksource */
+static struct clocksource clocksource_vmi;
 
 static cycle_t read_real_cycles(void)
 {
-	return vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
+	cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
+	return ret >= clocksource_vmi.cycle_last ?
+		ret : clocksource_vmi.cycle_last;
 }
 
 static struct clocksource clocksource_vmi = {

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Ingo Molnar]

* Alok Kataria <akataria@vmware.com> wrote:

> From: Alok N Kataria <akataria@vmware.com>
> 
> Similar to the check for TSC going backwards in the TSC 
> clocksource, we also need this check for VMI clocksource.

ok. Precisely what kind of bug/symptoms does this fix?

	Ingo

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Alok Kataria]

On Thu, 2009-02-19 at 03:18 -0800, Ingo Molnar wrote:
> * Alok Kataria <akataria@vmware.com> wrote:
> 
> > From: Alok N Kataria <akataria@vmware.com>
> > 
> > Similar to the check for TSC going backwards in the TSC 
> > clocksource, we also need this check for VMI clocksource.
> 
> ok. Precisely what kind of bug/symptoms does this fix?

If the read CYCLES_REAL counter value is less than the cycle_last value,
it may result in time jumps.
I also saw some messages about  process being stuck for more than
120sec, which was traced back to this.

Thanks,
Alok
> 
> 	Ingo

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Ingo Molnar]

* Alok Kataria <akataria@vmware.com> wrote:

> On Thu, 2009-02-19 at 03:18 -0800, Ingo Molnar wrote:
> > * Alok Kataria <akataria@vmware.com> wrote:
> > 
> > > From: Alok N Kataria <akataria@vmware.com>
> > > 
> > > Similar to the check for TSC going backwards in the TSC 
> > > clocksource, we also need this check for VMI clocksource.
> > 
> > ok. Precisely what kind of bug/symptoms does this fix?
> 
> If the read CYCLES_REAL counter value is less than the cycle_last value,
> it may result in time jumps.
> I also saw some messages about  process being stuck for more than
> 120sec, which was traced back to this.

Applied to tip:x86/urgent, thanks Alok!

	Ingo

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Ingo Molnar]

* Alok Kataria <akataria@vmware.com> wrote:

>  static cycle_t read_real_cycles(void)
>  {
> -	return vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> +	cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> +	return ret >= clocksource_vmi.cycle_last ?
> +		ret : clocksource_vmi.cycle_last;

btw., could you please clean this up to use max()?

	Ingo

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Alok Kataria]

On Fri, 2009-02-20 at 09:58 -0800, Ingo Molnar wrote:
> * Alok Kataria <akataria@vmware.com> wrote:
> 
> >  static cycle_t read_real_cycles(void)
> >  {
> > -	return vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> > +	cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> > +	return ret >= clocksource_vmi.cycle_last ?
> > +		ret : clocksource_vmi.cycle_last;
> 
> btw., could you please clean this up to use max()?

Please find it below, on top of current tip/master.

--

cleanup vmi_read_cycles to use max()

Signed-off-by: Alok N Kataria <akataria@vmware.com>

---

 arch/x86/kernel/vmiclock_32.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)


diff --git a/arch/x86/kernel/vmiclock_32.c b/arch/x86/kernel/vmiclock_32.c
index 9cd28c0..b77ad57 100644
--- a/arch/x86/kernel/vmiclock_32.c
+++ b/arch/x86/kernel/vmiclock_32.c
@@ -288,8 +288,7 @@ static struct clocksource clocksource_vmi;
 static cycle_t read_real_cycles(void)
 {
 	cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
-	return ret >= clocksource_vmi.cycle_last ?
-		ret : clocksource_vmi.cycle_last;
+	return max(ret, clocksource_vmi.cycle_last);
 }
 
 static struct clocksource clocksource_vmi = {

Re: x86, vmi: TSC going backwards check in vmi clocksource

[Ingo Molnar]

* Alok Kataria <akataria@vmware.com> wrote:

> On Fri, 2009-02-20 at 09:58 -0800, Ingo Molnar wrote:
> > * Alok Kataria <akataria@vmware.com> wrote:
> > 
> > >  static cycle_t read_real_cycles(void)
> > >  {
> > > -	return vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> > > +	cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
> > > +	return ret >= clocksource_vmi.cycle_last ?
> > > +		ret : clocksource_vmi.cycle_last;
> > 
> > btw., could you please clean this up to use max()?
> 
> Please find it below, on top of current tip/master.

Applied to tip:x86/cleanups, thanks!

	Ingo