Xen devel - Jun 2013 - [PATCH 5/5] x86/xen: sync the CMOS RTC as well as the Xen wallclock

From: David Vrabel <david.vrabel@citrix.com>

Adjustments to Xen''s persistent clock via update_persistent_clock()
don''t actually persist, as the Xen wallclock is a software only clock
and modifications to it do not modify the underlying CMOS RTC.

The x86_platform.set_wallclock hook can be used to keep the hardware
RTC synchronized (as on bare metal).  Because the Xen wallclock is now
kept synchronized by pvclock_gtod notifier, xen_set_wallclock() need
not do this and dom0 can simply use the native implementation.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
---
 arch/x86/xen/time.c |   32 ++++++++++++++++----------------
 1 files changed, 16 insertions(+), 16 deletions(-)

diff --git a/arch/x86/xen/time.c b/arch/x86/xen/time.c
index 3364850..20e395a 100644
--- a/arch/x86/xen/time.c
+++ b/arch/x86/xen/time.c
@@ -199,37 +199,35 @@ static void xen_get_wallclock(struct timespec *now)
 
 static int xen_set_wallclock(const struct timespec *now)
 {
-	struct xen_platform_op op;
-
-	/* do nothing for domU */
-	if (!xen_initial_domain())
-		return -1;
-
-	op.cmd = XENPF_settime;
-	op.u.settime.secs = now->tv_sec;
-	op.u.settime.nsecs = now->tv_nsec;
-	op.u.settime.system_time = xen_clocksource_read();
-
-	return HYPERVISOR_dom0_op(&op);
+	return -1;
 }
 
 static int xen_pvclock_gtod_notify(struct notifier_block *nb, unsigned long
was_set,
 				   void *priv)
 {
+	static struct timespec next;
 	struct timespec now;
 	struct xen_platform_op op;
 
-	if (!was_set)
-		return NOTIFY_OK;
-
 	now = __current_kernel_time();
 
+	if (!was_set && timespec_compare(&now, &next) < 0)
+		return NOTIFY_OK;
+
 	op.cmd = XENPF_settime;
 	op.u.settime.secs = now.tv_sec;
 	op.u.settime.nsecs = now.tv_nsec;
 	op.u.settime.system_time = xen_clocksource_read();
 
 	(void)HYPERVISOR_dom0_op(&op);
+
+	/*
+	 * Don''t update the wallclock for another 11 minutes. This is
+	 * the same period as the sync_cmos_clock() work.
+	 */
+	next = now;
+	next.tv_sec += 11*60;
+
 	return NOTIFY_OK;
 }
 
@@ -513,7 +511,9 @@ void __init xen_init_time_ops(void)
 
 	x86_platform.calibrate_tsc = xen_tsc_khz;
 	x86_platform.get_wallclock = xen_get_wallclock;
-	x86_platform.set_wallclock = xen_set_wallclock;
+	/* Dom0 uses the native method to set the hardware RTC. */
+	if (!xen_initial_domain())
+		x86_platform.set_wallclock = xen_set_wallclock;
 }
 
 #ifdef CONFIG_XEN_PVHVM
-- 
1.7.2.5

On Thu, 27 Jun 2013, David Vrabel wrote:
> From: David Vrabel <david.vrabel@citrix.com>
> 
> Adjustments to Xen''s persistent clock via
update_persistent_clock()
> don''t actually persist, as the Xen wallclock is a software only
clock
> and modifications to it do not modify the underlying CMOS RTC.
> 
> The x86_platform.set_wallclock hook can be used to keep the hardware
> RTC synchronized (as on bare metal).  Because the Xen wallclock is now
> kept synchronized by pvclock_gtod notifier, xen_set_wallclock() need
> not do this and dom0 can simply use the native implementation.
I can understand that part, but ...
 
>  static int xen_pvclock_gtod_notify(struct notifier_block *nb, unsigned
long was_set,
>  				   void *priv)
>  {
> +	static struct timespec next;
>  	struct timespec now;
>  	struct xen_platform_op op;
>  
> -	if (!was_set)
> -		return NOTIFY_OK;
> -
>  	now = __current_kernel_time();
>  
> +	if (!was_set && timespec_compare(&now, &next) < 0)
> +		return NOTIFY_OK;
> +
>  	op.cmd = XENPF_settime;
>  	op.u.settime.secs = now.tv_sec;
>  	op.u.settime.nsecs = now.tv_nsec;
>  	op.u.settime.system_time = xen_clocksource_read();
>  
>  	(void)HYPERVISOR_dom0_op(&op);
> +
> +	/*
> +	 * Don''t update the wallclock for another 11 minutes. This is
> +	 * the same period as the sync_cmos_clock() work.
> +	 */
> +	next = now;
> +	next.tv_sec += 11*60;
> +
How is this related to the changelog? /me is confused .....

On 28/06/13 16:38, Thomas Gleixner wrote:
> On Thu, 27 Jun 2013, David Vrabel wrote:
> 
>> From: David Vrabel <david.vrabel@citrix.com>
>>
>> Adjustments to Xen''s persistent clock via
update_persistent_clock()
>> don''t actually persist, as the Xen wallclock is a software
only clock
>> and modifications to it do not modify the underlying CMOS RTC.
>>
>> The x86_platform.set_wallclock hook can be used to keep the hardware
>> RTC synchronized (as on bare metal).  Because the Xen wallclock is now
>> kept synchronized by pvclock_gtod notifier, xen_set_wallclock() need
>> not do this and dom0 can simply use the native implementation.
> 
> I can understand that part, but ...
>  
>>  static int xen_pvclock_gtod_notify(struct notifier_block *nb, unsigned
long was_set,
>>  				   void *priv)
>>  {
>> +	static struct timespec next;
>>  	struct timespec now;
>>  	struct xen_platform_op op;
>>  
>> -	if (!was_set)
>> -		return NOTIFY_OK;
>> -
>>  	now = __current_kernel_time();
>>  
>> +	if (!was_set && timespec_compare(&now, &next) < 0)
>> +		return NOTIFY_OK;
>> +
>>  	op.cmd = XENPF_settime;
>>  	op.u.settime.secs = now.tv_sec;
>>  	op.u.settime.nsecs = now.tv_nsec;
>>  	op.u.settime.system_time = xen_clocksource_read();
>>  
>>  	(void)HYPERVISOR_dom0_op(&op);
>> +
>> +	/*
>> +	 * Don''t update the wallclock for another 11 minutes. This
is
>> +	 * the same period as the sync_cmos_clock() work.
>> +	 */
>> +	next = now;
>> +	next.tv_sec += 11*60;
>> +
> 
> How is this related to the changelog? /me is confused .....
Before:

Xen wallclock set when time is stepped.
Xen wallclock set every 11 minutes (by sync_cmos_clock()).
Hardware RTC never set.

After:

Xen wallclock set when time is stepped.
Xen wallclock set every 11 minutes (in pvclock gtod notifier).
Hardware RTC set every 11 minutes (by sync_cmos_clock()).

I''ll update the changelog to be more descriptive:

  Adjustments to Xen''s persistent clock via update_persistent_clock()
  don''t actually persist, as the Xen wallclock is a software only clock
  and modifications to it do not modify the underlying CMOS RTC.

  The x86_platform.set_wallclock hook can be used to keep the hardware
  RTC synchronized (as on bare metal).  If (in dom0) we make the Xen
  wallclock periodically synchronized by the pvclock_gtod notifier, the
  set_wallclock hook need not update the Xen wallclock and the native
  implementation can be used.

Is that better?

David

On Fri, 28 Jun 2013, David Vrabel wrote:
> 
> Before:
> 
> Xen wallclock set when time is stepped.
> Xen wallclock set every 11 minutes (by sync_cmos_clock()).
> Hardware RTC never set.
> 
> After:
> 
> Xen wallclock set when time is stepped.
> Xen wallclock set every 11 minutes (in pvclock gtod notifier).
Ah, you are emulating the sync_cmos_clock() behaviour for the xen
wallclock via the periodic pvclock_gtod notifier call.
> Hardware RTC set every 11 minutes (by sync_cmos_clock()).
> 
> I''ll update the changelog to be more descriptive:
> 
>   Adjustments to Xen''s persistent clock via
update_persistent_clock()
>   don''t actually persist, as the Xen wallclock is a software only
clock
>   and modifications to it do not modify the underlying CMOS RTC.
> 
>   The x86_platform.set_wallclock hook can be used to keep the hardware
>   RTC synchronized (as on bare metal).  If (in dom0) we make the Xen
>   wallclock periodically synchronized by the pvclock_gtod notifier, the
>   set_wallclock hook need not update the Xen wallclock and the native
>   implementation can be used.
Yep. I''ll pick that up.

Thanks,

	tglx

On 28/06/13 17:09, Thomas Gleixner wrote:
> On Fri, 28 Jun 2013, David Vrabel wrote:
>>
>> Before:
>>
>> Xen wallclock set when time is stepped.
>> Xen wallclock set every 11 minutes (by sync_cmos_clock()).
>> Hardware RTC never set.
>>
>> After:
>>
>> Xen wallclock set when time is stepped.
>> Xen wallclock set every 11 minutes (in pvclock gtod notifier).
> 
> Ah, you are emulating the sync_cmos_clock() behaviour for the xen
> wallclock via the periodic pvclock_gtod notifier call.
> 
>> Hardware RTC set every 11 minutes (by sync_cmos_clock()).
>>
>> I''ll update the changelog to be more descriptive:
>>
>>   Adjustments to Xen''s persistent clock via
update_persistent_clock()
>>   don''t actually persist, as the Xen wallclock is a software
only clock
>>   and modifications to it do not modify the underlying CMOS RTC.
>>
>>   The x86_platform.set_wallclock hook can be used to keep the hardware
>>   RTC synchronized (as on bare metal).  If (in dom0) we make the Xen
>>   wallclock periodically synchronized by the pvclock_gtod notifier, the
>>   set_wallclock hook need not update the Xen wallclock and the native
>>   implementation can be used.
> 
> Yep. I''ll pick that up.
If it helps, I''ve made this change as well as splitting the xen part of
the hrtimers patch out into a separate commit.

The following changes since commit 52efa9eb9ea36b457b23b1b5d3dd9ce64d110715:

  x86: increase precision of x86_platform.get/set_wallclock() (2013-06-26
18:07:16 +0100)

are available in the git repository at:
  git://xenbits.xen.org/people/dvrabel/linux.git wallclock-v8

David Vrabel (6):
      hrtimers: support resuming with two or more CPUs online (but stopped)
      time: pass flags instead of multiple bools to timekeeping_update()
      time: indicate that the clock was set in the pvclock gtod notifier chain
      xen: remove unnecessary call to clock_was_set() during resume
      x86/xen: sync the wallclock when the system time is set
      x86/xen: sync the CMOS RTC as well as the Xen wallclock

 arch/x86/xen/time.c          |   42 +++++++++++++++++++++++++++++++++++-------
 drivers/xen/manage.c         |    3 ---
 include/linux/pvclock_gtod.h |    7 +++++++
 kernel/hrtimer.c             |   15 ++++++++++++---
 kernel/time/timekeeping.c    |   39 ++++++++++++++++++++++++---------------
 5 files changed, 78 insertions(+), 28 deletions(-)

David

On Fri, 28 Jun 2013, David Vrabel wrote:
> On 28/06/13 17:09, Thomas Gleixner wrote:
> > Yep. I''ll pick that up.
> 
> If it helps, I''ve made this change as well as splitting the xen
part of
> the hrtimers patch out into a separate commit.
Fun. I already have that same split in my pending queue :)

Thanks,

	tglx