[PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

[PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

[Peter Martincic]

From 529fcea5d296c22b1dc6c23d55bd6417794b3cda Mon Sep 17 00:00:00 2001
From: Peter Martincic <pmartincic@microsoft.com>
Date: Mon, 16 Oct 2023 16:41:10 -0700
Subject: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

Windows hosts can omit the _SYNC flag to due a bug on resume from modern
suspend. If the guest is sufficiently behind, treat a _SAMPLE the same
as if _SYNC was received.

This is hidden behind param hv_utils.timesync_implicit.

Signed-off-by: Peter Martincic <pmartincic@microsoft.com>
---
 drivers/hv/hv_util.c | 31 ++++++++++++++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)

diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c
index 42aec2c5606a..158f5ff4b809 100644
--- a/drivers/hv/hv_util.c
+++ b/drivers/hv/hv_util.c
@@ -296,6 +296,11 @@ static struct {
        spinlock_t                      lock;
 } host_ts;

+static bool timesync_implicit;
+
+module_param(timesync_implicit, bool, 0644);
+MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when clock is behind");
+
 static inline u64 reftime_to_ns(u64 reftime)
 {
        return (reftime - WLTIMEDELTA) * 100;
@@ -344,6 +349,29 @@ static void hv_set_host_time(struct work_struct *work)
                do_settimeofday64(&ts);
 }

+/*
+ * Due to a bug on Windows hosts, the sync flag may not always be sent on resume.
+ * Force a sync if it's behind.
+ */
+static inline bool hv_implicit_sync(u64 host_time)
+{
+       struct timespec64 new_ts;
+       struct timespec64 threshold_ts;
+
+       new_ts = ns_to_timespec64(reftime_to_ns(host_time));
+       ktime_get_real_ts64(&threshold_ts);
+
+       threshold_ts.tv_sec += 5;
+
+       /*
+        * If guest behind the host by 5 or more seconds.
+        */
+       if (timespec64_compare(&new_ts, &threshold_ts) >= 0)
+               return true;
+
+       return false;
+}
+
 /*
  * Synchronize time with host after reboot, restore, etc.
  *
@@ -384,7 +412,8 @@ static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
        spin_unlock_irqrestore(&host_ts.lock, flags);

        /* Schedule work to do do_settimeofday64() */
-       if (adj_flags & ICTIMESYNCFLAG_SYNC)
+       if ((adj_flags & ICTIMESYNCFLAG_SYNC) ||
+           (timesync_implicit && hv_implicit_sync(host_ts.host_time)))
                schedule_work(&adj_time_work);
 }

--
2.34.1

Re: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

[Wei Liu]

On Fri, Oct 27, 2023 at 08:42:33PM +0000, Peter Martincic wrote:
> From 529fcea5d296c22b1dc6c23d55bd6417794b3cda Mon Sep 17 00:00:00 2001
> From: Peter Martincic <pmartincic@microsoft.com>
> Date: Mon, 16 Oct 2023 16:41:10 -0700
> Subject: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC
> 
> Windows hosts can omit the _SYNC flag to due a bug on resume from modern
> suspend. If the guest is sufficiently behind, treat a _SAMPLE the same
> as if _SYNC was received.
> 
> This is hidden behind param hv_utils.timesync_implicit.
> 
> Signed-off-by: Peter Martincic <pmartincic@microsoft.com>

Boqun, what do you think about this patch?

> ---
>  drivers/hv/hv_util.c | 31 ++++++++++++++++++++++++++++++-
>  1 file changed, 30 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c
> index 42aec2c5606a..158f5ff4b809 100644
> --- a/drivers/hv/hv_util.c
> +++ b/drivers/hv/hv_util.c
> @@ -296,6 +296,11 @@ static struct {
>         spinlock_t                      lock;
>  } host_ts;
> 
> +static bool timesync_implicit;
> +
> +module_param(timesync_implicit, bool, 0644);
> +MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when clock is behind");
> +
>  static inline u64 reftime_to_ns(u64 reftime)
>  {
>         return (reftime - WLTIMEDELTA) * 100;
> @@ -344,6 +349,29 @@ static void hv_set_host_time(struct work_struct *work)
>                 do_settimeofday64(&ts);
>  }
> 
> +/*
> + * Due to a bug on Windows hosts, the sync flag may not always be sent on resume.
> + * Force a sync if it's behind.
> + */
> +static inline bool hv_implicit_sync(u64 host_time)
> +{
> +       struct timespec64 new_ts;
> +       struct timespec64 threshold_ts;
> +
> +       new_ts = ns_to_timespec64(reftime_to_ns(host_time));
> +       ktime_get_real_ts64(&threshold_ts);
> +
> +       threshold_ts.tv_sec += 5;
> +
> +       /*
> +        * If guest behind the host by 5 or more seconds.
> +        */
> +       if (timespec64_compare(&new_ts, &threshold_ts) >= 0)
> +               return true;
> +
> +       return false;
> +}
> +
>  /*
>   * Synchronize time with host after reboot, restore, etc.
>   *
> @@ -384,7 +412,8 @@ static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
>         spin_unlock_irqrestore(&host_ts.lock, flags);
> 
>         /* Schedule work to do do_settimeofday64() */
> -       if (adj_flags & ICTIMESYNCFLAG_SYNC)
> +       if ((adj_flags & ICTIMESYNCFLAG_SYNC) ||
> +           (timesync_implicit && hv_implicit_sync(host_ts.host_time)))
>                 schedule_work(&adj_time_work);
>  }
> 
> --
> 2.34.1
> 

RE: [EXTERNAL] Re: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

[Peter Martincic]

Sorry for the formatting/recipient/procedure mistakes. I've an updated commit message based on feedback from Michael. I'll wait to hear back from you and Boqun before I post V2/updates.

Thanks again,
Peter

-----Original Message-----
From: Wei Liu <wei.liu@kernel.org> 
Sent: Sunday, November 5, 2023 5:31 PM
To: Peter Martincic <pmartincic@microsoft.com>
Cc: linux-hyperv@vger.kernel.org; Michael Kelley (LINUX) <mikelley@microsoft.com>; Boqun Feng <Boqun.Feng@microsoft.com>; Wei Liu <liuwe@microsoft.com>; Wei Liu <wei.liu@kernel.org>
Subject: [EXTERNAL] Re: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

On Fri, Oct 27, 2023 at 08:42:33PM +0000, Peter Martincic wrote:
> From 529fcea5d296c22b1dc6c23d55bd6417794b3cda Mon Sep 17 00:00:00 2001
> From: Peter Martincic <pmartincic@microsoft.com>
> Date: Mon, 16 Oct 2023 16:41:10 -0700
> Subject: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC
> 
> Windows hosts can omit the _SYNC flag to due a bug on resume from 
> modern suspend. If the guest is sufficiently behind, treat a _SAMPLE 
> the same as if _SYNC was received.
> 
> This is hidden behind param hv_utils.timesync_implicit.
> 
> Signed-off-by: Peter Martincic <pmartincic@microsoft.com>

Boqun, what do you think about this patch?

> ---
>  drivers/hv/hv_util.c | 31 ++++++++++++++++++++++++++++++-
>  1 file changed, 30 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c index 
> 42aec2c5606a..158f5ff4b809 100644
> --- a/drivers/hv/hv_util.c
> +++ b/drivers/hv/hv_util.c
> @@ -296,6 +296,11 @@ static struct {
>         spinlock_t                      lock;
>  } host_ts;
> 
> +static bool timesync_implicit;
> +
> +module_param(timesync_implicit, bool, 0644); 
> +MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when 
> +clock is behind");
> +
>  static inline u64 reftime_to_ns(u64 reftime)  {
>         return (reftime - WLTIMEDELTA) * 100; @@ -344,6 +349,29 @@ 
> static void hv_set_host_time(struct work_struct *work)
>                 do_settimeofday64(&ts);  }
> 
> +/*
> + * Due to a bug on Windows hosts, the sync flag may not always be sent on resume.
> + * Force a sync if it's behind.
> + */
> +static inline bool hv_implicit_sync(u64 host_time) {
> +       struct timespec64 new_ts;
> +       struct timespec64 threshold_ts;
> +
> +       new_ts = ns_to_timespec64(reftime_to_ns(host_time));
> +       ktime_get_real_ts64(&threshold_ts);
> +
> +       threshold_ts.tv_sec += 5;
> +
> +       /*
> +        * If guest behind the host by 5 or more seconds.
> +        */
> +       if (timespec64_compare(&new_ts, &threshold_ts) >= 0)
> +               return true;
> +
> +       return false;
> +}
> +
>  /*
>   * Synchronize time with host after reboot, restore, etc.
>   *
> @@ -384,7 +412,8 @@ static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
>         spin_unlock_irqrestore(&host_ts.lock, flags);
> 
>         /* Schedule work to do do_settimeofday64() */
> -       if (adj_flags & ICTIMESYNCFLAG_SYNC)
> +       if ((adj_flags & ICTIMESYNCFLAG_SYNC) ||
> +           (timesync_implicit && 
> + hv_implicit_sync(host_ts.host_time)))
>                 schedule_work(&adj_time_work);  }
> 
> --
> 2.34.1
> 

Re: [EXTERNAL] Re: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC

[Boqun Feng]

On Mon, Nov 06, 2023 at 06:18:48PM +0000, Peter Martincic wrote:
> Sorry for the formatting/recipient/procedure mistakes. I've an updated commit message based on feedback from Michael. I'll wait to hear back from you and Boqun before I post V2/updates.
> 
> Thanks again,
> Peter
> 
> -----Original Message-----
> From: Wei Liu <wei.liu@kernel.org> 
> Sent: Sunday, November 5, 2023 5:31 PM
> To: Peter Martincic <pmartincic@microsoft.com>
> Cc: linux-hyperv@vger.kernel.org; Michael Kelley (LINUX) <mikelley@microsoft.com>; Boqun Feng <Boqun.Feng@microsoft.com>; Wei Liu <liuwe@microsoft.com>; Wei Liu <wei.liu@kernel.org>
> Subject: [EXTERNAL] Re: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC
> 
> On Fri, Oct 27, 2023 at 08:42:33PM +0000, Peter Martincic wrote:
> > From 529fcea5d296c22b1dc6c23d55bd6417794b3cda Mon Sep 17 00:00:00 2001
> > From: Peter Martincic <pmartincic@microsoft.com>
> > Date: Mon, 16 Oct 2023 16:41:10 -0700
> > Subject: [PATCH] hv_utils: Allow implicit ICTIMESYNCFLAG_SYNC
> > 
> > Windows hosts can omit the _SYNC flag to due a bug on resume from 
> > modern suspend. If the guest is sufficiently behind, treat a _SAMPLE 
> > the same as if _SYNC was received.
> > 
> > This is hidden behind param hv_utils.timesync_implicit.
> > 
> > Signed-off-by: Peter Martincic <pmartincic@microsoft.com>
> 
> Boqun, what do you think about this patch?
> 

The patch looks good to me. Peter, feel free to send the V2 and I will
give my Acked-by.

Regards,
Boqun