diff mbox series

[iwl-next,v5,08/12] iavf: periodically cache PHC time

Message ID 20240418052500.50678-9-mateusz.polchlopek@intel.com
State Changes Requested
Headers show
Series Add support for Rx timestamping for both ice and iavf drivers. | expand

Commit Message

Mateusz Polchlopek April 18, 2024, 5:24 a.m. UTC
From: Jacob Keller <jacob.e.keller@intel.com>

The Rx timestamps reported by hardware may only have 32 bits of storage
for nanosecond time. These timestamps cannot be directly reported to the
Linux stack, as it expects 64bits of time.

To handle this, the timestamps must be extended using an algorithm that
calculates the corrected 64bit timestamp by comparison between the PHC
time and the timestamp. This algorithm requires the PHC time to be
captured within ~2 seconds of when the timestamp was captured.

Instead of trying to read the PHC time in the Rx hotpath, the algorithm
relies on a cached value that is periodically updated.

Keep this cached time up to date by using the PTP .do_aux_work kthread
function.

The iavf_ptp_do_aux_work will reschedule itself about twice a second,
and will check whether or not the cached PTP time needs to be updated.
If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
from the PF. The jitter and latency involved with this command aren't
important, because the cached time just needs to be kept up to date
within about ~2 seconds.

Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
---
 drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52 ++++++++++++++++++++++
 drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
 2 files changed, 53 insertions(+)

Comments

Rahul Rameshbabu April 18, 2024, 7:51 p.m. UTC | #1
On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek <mateusz.polchlopek@intel.com> wrote:
> From: Jacob Keller <jacob.e.keller@intel.com>
>
> The Rx timestamps reported by hardware may only have 32 bits of storage
> for nanosecond time. These timestamps cannot be directly reported to the
> Linux stack, as it expects 64bits of time.
>
> To handle this, the timestamps must be extended using an algorithm that
> calculates the corrected 64bit timestamp by comparison between the PHC
> time and the timestamp. This algorithm requires the PHC time to be
> captured within ~2 seconds of when the timestamp was captured.
>
> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
> relies on a cached value that is periodically updated.
>
> Keep this cached time up to date by using the PTP .do_aux_work kthread
> function.

Seems reasonable.

>
> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
> and will check whether or not the cached PTP time needs to be updated.
> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
> from the PF. The jitter and latency involved with this command aren't
> important, because the cached time just needs to be kept up to date
> within about ~2 seconds.
>
> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> ---
>  drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52 ++++++++++++++++++++++
>  drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
>  2 files changed, 53 insertions(+)
>
> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
<snip>
> +/**
> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
> + * @ptp: PTP clock info structure
> + *
> + * Handler to take care of periodic work required for PTP operation. This
> + * includes the following tasks:
> + *
> + *   1) updating cached_phc_time
> + *
> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
> + *      perform timestamp extension, by carefully comparing the timestamp
> + *      32bit nanosecond timestamps and determining the corrected 64bit
> + *      timestamp value to report to userspace. This algorithm only works if
> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
> + *      event. This task periodically reads the PHC time and stores it, to
> + *      ensure that timestamp extension operates correctly.
> + *
> + * Returns: time in jiffies until the periodic task should be re-scheduled.
> + */
> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
> +{
> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
> +
> +	iavf_ptp_cache_phc_time(adapter);
> +
> +	/* Check work about twice a second */
> +	return msecs_to_jiffies(500);

HZ / 2 might be more intuitive?

> +}
> +
<snip>
Mateusz Polchlopek April 22, 2024, 9:23 a.m. UTC | #2
On 4/18/2024 9:51 PM, Rahul Rameshbabu wrote:
> On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek <mateusz.polchlopek@intel.com> wrote:
>> From: Jacob Keller <jacob.e.keller@intel.com>
>>
>> The Rx timestamps reported by hardware may only have 32 bits of storage
>> for nanosecond time. These timestamps cannot be directly reported to the
>> Linux stack, as it expects 64bits of time.
>>
>> To handle this, the timestamps must be extended using an algorithm that
>> calculates the corrected 64bit timestamp by comparison between the PHC
>> time and the timestamp. This algorithm requires the PHC time to be
>> captured within ~2 seconds of when the timestamp was captured.
>>
>> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
>> relies on a cached value that is periodically updated.
>>
>> Keep this cached time up to date by using the PTP .do_aux_work kthread
>> function.
> 
> Seems reasonable.
> 
>>
>> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
>> and will check whether or not the cached PTP time needs to be updated.
>> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
>> from the PF. The jitter and latency involved with this command aren't
>> important, because the cached time just needs to be kept up to date
>> within about ~2 seconds.
>>
>> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
>> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
>> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>> ---
>>   drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52 ++++++++++++++++++++++
>>   drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
>>   2 files changed, 53 insertions(+)
>>
>> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
> <snip>
>> +/**
>> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
>> + * @ptp: PTP clock info structure
>> + *
>> + * Handler to take care of periodic work required for PTP operation. This
>> + * includes the following tasks:
>> + *
>> + *   1) updating cached_phc_time
>> + *
>> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
>> + *      perform timestamp extension, by carefully comparing the timestamp
>> + *      32bit nanosecond timestamps and determining the corrected 64bit
>> + *      timestamp value to report to userspace. This algorithm only works if
>> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
>> + *      event. This task periodically reads the PHC time and stores it, to
>> + *      ensure that timestamp extension operates correctly.
>> + *
>> + * Returns: time in jiffies until the periodic task should be re-scheduled.
>> + */
>> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
>> +{
>> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
>> +
>> +	iavf_ptp_cache_phc_time(adapter);
>> +
>> +	/* Check work about twice a second */
>> +	return msecs_to_jiffies(500);
> 
> HZ / 2 might be more intuitive?
> 

Makes sense, will take a look on that and if applicable then will send
in the next version. Thanks

>> +}
>> +
> <snip>
Jacob Keller April 24, 2024, 10:03 p.m. UTC | #3
> -----Original Message-----
> From: Polchlopek, Mateusz <mateusz.polchlopek@intel.com>
> Sent: Monday, April 22, 2024 2:23 AM
> To: Rahul Rameshbabu <rrameshbabu@nvidia.com>
> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org; horms@kernel.org;
> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Keller, Jacob E
> <jacob.e.keller@intel.com>; Drewek, Wojciech <wojciech.drewek@intel.com>
> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
> PHC time
> 
> 
> 
> On 4/18/2024 9:51 PM, Rahul Rameshbabu wrote:
> > On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek
> <mateusz.polchlopek@intel.com> wrote:
> >> From: Jacob Keller <jacob.e.keller@intel.com>
> >>
> >> The Rx timestamps reported by hardware may only have 32 bits of storage
> >> for nanosecond time. These timestamps cannot be directly reported to the
> >> Linux stack, as it expects 64bits of time.
> >>
> >> To handle this, the timestamps must be extended using an algorithm that
> >> calculates the corrected 64bit timestamp by comparison between the PHC
> >> time and the timestamp. This algorithm requires the PHC time to be
> >> captured within ~2 seconds of when the timestamp was captured.
> >>
> >> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
> >> relies on a cached value that is periodically updated.
> >>
> >> Keep this cached time up to date by using the PTP .do_aux_work kthread
> >> function.
> >
> > Seems reasonable.
> >
> >>
> >> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
> >> and will check whether or not the cached PTP time needs to be updated.
> >> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
> >> from the PF. The jitter and latency involved with this command aren't
> >> important, because the cached time just needs to be kept up to date
> >> within about ~2 seconds.
> >>
> >> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
> >> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
> >> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> >> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> >> ---
> >>   drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52 ++++++++++++++++++++++
> >>   drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
> >>   2 files changed, 53 insertions(+)
> >>
> >> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c
> b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
> > <snip>
> >> +/**
> >> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
> >> + * @ptp: PTP clock info structure
> >> + *
> >> + * Handler to take care of periodic work required for PTP operation. This
> >> + * includes the following tasks:
> >> + *
> >> + *   1) updating cached_phc_time
> >> + *
> >> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
> >> + *      perform timestamp extension, by carefully comparing the timestamp
> >> + *      32bit nanosecond timestamps and determining the corrected 64bit
> >> + *      timestamp value to report to userspace. This algorithm only works if
> >> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
> >> + *      event. This task periodically reads the PHC time and stores it, to
> >> + *      ensure that timestamp extension operates correctly.
> >> + *
> >> + * Returns: time in jiffies until the periodic task should be re-scheduled.
> >> + */
> >> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
> >> +{
> >> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
> >> +
> >> +	iavf_ptp_cache_phc_time(adapter);
> >> +
> >> +	/* Check work about twice a second */
> >> +	return msecs_to_jiffies(500);
> >
> > HZ / 2 might be more intuitive?
> >

I've always found it more intuitive to think in terms of msecs myself, but HZ / 2 is ok if other folks agree.

Thanks,
Jake
Przemek Kitszel April 25, 2024, 10:52 a.m. UTC | #4
On 4/25/24 00:03, Keller, Jacob E wrote:
> 
> 
>> -----Original Message-----
>> From: Polchlopek, Mateusz <mateusz.polchlopek@intel.com>
>> Sent: Monday, April 22, 2024 2:23 AM
>> To: Rahul Rameshbabu <rrameshbabu@nvidia.com>
>> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org; horms@kernel.org;
>> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Keller, Jacob E
>> <jacob.e.keller@intel.com>; Drewek, Wojciech <wojciech.drewek@intel.com>
>> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
>> PHC time
>>
>>
>>
>> On 4/18/2024 9:51 PM, Rahul Rameshbabu wrote:
>>> On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek
>> <mateusz.polchlopek@intel.com> wrote:
>>>> From: Jacob Keller <jacob.e.keller@intel.com>
>>>>
>>>> The Rx timestamps reported by hardware may only have 32 bits of storage
>>>> for nanosecond time. These timestamps cannot be directly reported to the
>>>> Linux stack, as it expects 64bits of time.
>>>>
>>>> To handle this, the timestamps must be extended using an algorithm that
>>>> calculates the corrected 64bit timestamp by comparison between the PHC
>>>> time and the timestamp. This algorithm requires the PHC time to be
>>>> captured within ~2 seconds of when the timestamp was captured.
>>>>
>>>> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
>>>> relies on a cached value that is periodically updated.
>>>>
>>>> Keep this cached time up to date by using the PTP .do_aux_work kthread
>>>> function.
>>>
>>> Seems reasonable.
>>>
>>>>
>>>> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
>>>> and will check whether or not the cached PTP time needs to be updated.
>>>> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
>>>> from the PF. The jitter and latency involved with this command aren't
>>>> important, because the cached time just needs to be kept up to date
>>>> within about ~2 seconds.
>>>>
>>>> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
>>>> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
>>>> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>>>> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>>>> ---
>>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52 ++++++++++++++++++++++
>>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
>>>>    2 files changed, 53 insertions(+)
>>>>
>>>> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c
>> b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
>>> <snip>
>>>> +/**
>>>> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
>>>> + * @ptp: PTP clock info structure
>>>> + *
>>>> + * Handler to take care of periodic work required for PTP operation. This
>>>> + * includes the following tasks:
>>>> + *
>>>> + *   1) updating cached_phc_time
>>>> + *
>>>> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
>>>> + *      perform timestamp extension, by carefully comparing the timestamp
>>>> + *      32bit nanosecond timestamps and determining the corrected 64bit
>>>> + *      timestamp value to report to userspace. This algorithm only works if
>>>> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
>>>> + *      event. This task periodically reads the PHC time and stores it, to
>>>> + *      ensure that timestamp extension operates correctly.
>>>> + *
>>>> + * Returns: time in jiffies until the periodic task should be re-scheduled.
>>>> + */
>>>> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
>>>> +{
>>>> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
>>>> +
>>>> +	iavf_ptp_cache_phc_time(adapter);
>>>> +
>>>> +	/* Check work about twice a second */
>>>> +	return msecs_to_jiffies(500);
>>>
>>> HZ / 2 might be more intuitive?
>>>
> 
> I've always found it more intuitive to think in terms of msecs myself, but HZ / 2 is ok if other folks agree.

HZ/2 or HZ/3 as a timer period could be understood without thinking, but
the same stands for 400ms. Problems starts when one thinks about it ;)

For me HZ, which could be both literally and colloquially understood as
"per second" should not mean 1000ms (just evaluate to).
2Hz is a frequency with half second period, but 2*HZ evaluates to 2000ms
which is 4 times more :/


> 
> Thanks,
> Jake
>
Jacob Keller April 25, 2024, 4:28 p.m. UTC | #5
> -----Original Message-----
> From: Kitszel, Przemyslaw <przemyslaw.kitszel@intel.com>
> Sent: Thursday, April 25, 2024 3:52 AM
> To: Keller, Jacob E <jacob.e.keller@intel.com>; Polchlopek, Mateusz
> <mateusz.polchlopek@intel.com>; Rahul Rameshbabu
> <rrameshbabu@nvidia.com>
> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org; horms@kernel.org;
> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Drewek, Wojciech
> <wojciech.drewek@intel.com>
> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
> PHC time
> 
> On 4/25/24 00:03, Keller, Jacob E wrote:
> >
> >
> >> -----Original Message-----
> >> From: Polchlopek, Mateusz <mateusz.polchlopek@intel.com>
> >> Sent: Monday, April 22, 2024 2:23 AM
> >> To: Rahul Rameshbabu <rrameshbabu@nvidia.com>
> >> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org;
> horms@kernel.org;
> >> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Keller, Jacob E
> >> <jacob.e.keller@intel.com>; Drewek, Wojciech <wojciech.drewek@intel.com>
> >> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
> >> PHC time
> >>
> >>
> >>
> >> On 4/18/2024 9:51 PM, Rahul Rameshbabu wrote:
> >>> On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek
> >> <mateusz.polchlopek@intel.com> wrote:
> >>>> From: Jacob Keller <jacob.e.keller@intel.com>
> >>>>
> >>>> The Rx timestamps reported by hardware may only have 32 bits of storage
> >>>> for nanosecond time. These timestamps cannot be directly reported to the
> >>>> Linux stack, as it expects 64bits of time.
> >>>>
> >>>> To handle this, the timestamps must be extended using an algorithm that
> >>>> calculates the corrected 64bit timestamp by comparison between the PHC
> >>>> time and the timestamp. This algorithm requires the PHC time to be
> >>>> captured within ~2 seconds of when the timestamp was captured.
> >>>>
> >>>> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
> >>>> relies on a cached value that is periodically updated.
> >>>>
> >>>> Keep this cached time up to date by using the PTP .do_aux_work kthread
> >>>> function.
> >>>
> >>> Seems reasonable.
> >>>
> >>>>
> >>>> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
> >>>> and will check whether or not the cached PTP time needs to be updated.
> >>>> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
> >>>> from the PF. The jitter and latency involved with this command aren't
> >>>> important, because the cached time just needs to be kept up to date
> >>>> within about ~2 seconds.
> >>>>
> >>>> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
> >>>> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
> >>>> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> >>>> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
> >>>> ---
> >>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52
> ++++++++++++++++++++++
> >>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
> >>>>    2 files changed, 53 insertions(+)
> >>>>
> >>>> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c
> >> b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
> >>> <snip>
> >>>> +/**
> >>>> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
> >>>> + * @ptp: PTP clock info structure
> >>>> + *
> >>>> + * Handler to take care of periodic work required for PTP operation. This
> >>>> + * includes the following tasks:
> >>>> + *
> >>>> + *   1) updating cached_phc_time
> >>>> + *
> >>>> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
> >>>> + *      perform timestamp extension, by carefully comparing the timestamp
> >>>> + *      32bit nanosecond timestamps and determining the corrected 64bit
> >>>> + *      timestamp value to report to userspace. This algorithm only works if
> >>>> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
> >>>> + *      event. This task periodically reads the PHC time and stores it, to
> >>>> + *      ensure that timestamp extension operates correctly.
> >>>> + *
> >>>> + * Returns: time in jiffies until the periodic task should be re-scheduled.
> >>>> + */
> >>>> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
> >>>> +{
> >>>> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
> >>>> +
> >>>> +	iavf_ptp_cache_phc_time(adapter);
> >>>> +
> >>>> +	/* Check work about twice a second */
> >>>> +	return msecs_to_jiffies(500);
> >>>
> >>> HZ / 2 might be more intuitive?
> >>>
> >
> > I've always found it more intuitive to think in terms of msecs myself, but HZ / 2 is
> ok if other folks agree.
> 
> HZ/2 or HZ/3 as a timer period could be understood without thinking, but
> the same stands for 400ms. Problems starts when one thinks about it ;)
> 
> For me HZ, which could be both literally and colloquially understood as
> "per second" should not mean 1000ms (just evaluate to).
> 2Hz is a frequency with half second period, but 2*HZ evaluates to 2000ms
> which is 4 times more :/
> 

That’s part of why I switched ice over from using HZ generally to using jiffies_to_msec in a lot of cases. It really depends on what you personally find intuitive. Those used to seeing and reading HZ may find it easier.

Thanks,
Jake

> 
> >
> > Thanks,
> > Jake
> >
Rahul Rameshbabu April 25, 2024, 5:10 p.m. UTC | #6
On Thu, 25 Apr, 2024 16:28:22 +0000 "Keller, Jacob E" <jacob.e.keller@intel.com> wrote:
>> -----Original Message-----
>> From: Kitszel, Przemyslaw <przemyslaw.kitszel@intel.com>
>> Sent: Thursday, April 25, 2024 3:52 AM
>> To: Keller, Jacob E <jacob.e.keller@intel.com>; Polchlopek, Mateusz
>> <mateusz.polchlopek@intel.com>; Rahul Rameshbabu
>> <rrameshbabu@nvidia.com>
>> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org; horms@kernel.org;
>> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Drewek, Wojciech
>> <wojciech.drewek@intel.com>
>> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
>> PHC time
>> 
>> On 4/25/24 00:03, Keller, Jacob E wrote:
>> >
>> >
>> >> -----Original Message-----
>> >> From: Polchlopek, Mateusz <mateusz.polchlopek@intel.com>
>> >> Sent: Monday, April 22, 2024 2:23 AM
>> >> To: Rahul Rameshbabu <rrameshbabu@nvidia.com>
>> >> Cc: intel-wired-lan@lists.osuosl.org; netdev@vger.kernel.org;
>> horms@kernel.org;
>> >> Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Keller, Jacob E
>> >> <jacob.e.keller@intel.com>; Drewek, Wojciech <wojciech.drewek@intel.com>
>> >> Subject: Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache
>> >> PHC time
>> >>
>> >>
>> >>
>> >> On 4/18/2024 9:51 PM, Rahul Rameshbabu wrote:
>> >>> On Thu, 18 Apr, 2024 01:24:56 -0400 Mateusz Polchlopek
>> >> <mateusz.polchlopek@intel.com> wrote:
>> >>>> From: Jacob Keller <jacob.e.keller@intel.com>
>> >>>>
>> >>>> The Rx timestamps reported by hardware may only have 32 bits of storage
>> >>>> for nanosecond time. These timestamps cannot be directly reported to the
>> >>>> Linux stack, as it expects 64bits of time.
>> >>>>
>> >>>> To handle this, the timestamps must be extended using an algorithm that
>> >>>> calculates the corrected 64bit timestamp by comparison between the PHC
>> >>>> time and the timestamp. This algorithm requires the PHC time to be
>> >>>> captured within ~2 seconds of when the timestamp was captured.
>> >>>>
>> >>>> Instead of trying to read the PHC time in the Rx hotpath, the algorithm
>> >>>> relies on a cached value that is periodically updated.
>> >>>>
>> >>>> Keep this cached time up to date by using the PTP .do_aux_work kthread
>> >>>> function.
>> >>>
>> >>> Seems reasonable.
>> >>>
>> >>>>
>> >>>> The iavf_ptp_do_aux_work will reschedule itself about twice a second,
>> >>>> and will check whether or not the cached PTP time needs to be updated.
>> >>>> If so, it issues a VIRTCHNL_OP_1588_PTP_GET_TIME to request the time
>> >>>> from the PF. The jitter and latency involved with this command aren't
>> >>>> important, because the cached time just needs to be kept up to date
>> >>>> within about ~2 seconds.
>> >>>>
>> >>>> Reviewed-by: Wojciech Drewek <wojciech.drewek@intel.com>
>> >>>> Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
>> >>>> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>> >>>> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@intel.com>
>> >>>> ---
>> >>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.c | 52
>> ++++++++++++++++++++++
>> >>>>    drivers/net/ethernet/intel/iavf/iavf_ptp.h |  1 +
>> >>>>    2 files changed, 53 insertions(+)
>> >>>>
>> >>>> diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c
>> >> b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
>> >>> <snip>
>> >>>> +/**
>> >>>> + * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
>> >>>> + * @ptp: PTP clock info structure
>> >>>> + *
>> >>>> + * Handler to take care of periodic work required for PTP operation. This
>> >>>> + * includes the following tasks:
>> >>>> + *
>> >>>> + *   1) updating cached_phc_time
>> >>>> + *
>> >>>> + *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
>> >>>> + *      perform timestamp extension, by carefully comparing the timestamp
>> >>>> + *      32bit nanosecond timestamps and determining the corrected 64bit
>> >>>> + *      timestamp value to report to userspace. This algorithm only works if
>> >>>> + *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
>> >>>> + *      event. This task periodically reads the PHC time and stores it, to
>> >>>> + *      ensure that timestamp extension operates correctly.
>> >>>> + *
>> >>>> + * Returns: time in jiffies until the periodic task should be re-scheduled.
>> >>>> + */
>> >>>> +long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
>> >>>> +{
>> >>>> +	struct iavf_adapter *adapter = clock_to_adapter(ptp);
>> >>>> +
>> >>>> +	iavf_ptp_cache_phc_time(adapter);
>> >>>> +
>> >>>> +	/* Check work about twice a second */
>> >>>> +	return msecs_to_jiffies(500);
>> >>>
>> >>> HZ / 2 might be more intuitive?
>> >>>
>> >
>> > I've always found it more intuitive to think in terms of msecs myself, but HZ / 2 is
>> ok if other folks agree.
>> 
>> HZ/2 or HZ/3 as a timer period could be understood without thinking, but
>> the same stands for 400ms. Problems starts when one thinks about it ;)
>> 
>> For me HZ, which could be both literally and colloquially understood as
>> "per second" should not mean 1000ms (just evaluate to).
>> 2Hz is a frequency with half second period, but 2*HZ evaluates to 2000ms
>> which is 4 times more :/
>> 
>
> That’s part of why I switched ice over from using HZ generally to using
> jiffies_to_msec in a lot of cases. It really depends on what you personally find
> intuitive. Those used to seeing and reading HZ may find it easier.
>

Makes sense to stick with the same if ice is using jiffies_to_msec in
general. I, recently, was re-reading the Linux Device Drivers book,
which has a section that elaborates on HZ a bit.

--
Thanks,

Rahul Rameshbabu
diff mbox series

Patch

diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.c b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
index 0d30979616a0..8f6b1a8df5df 100644
--- a/drivers/net/ethernet/intel/iavf/iavf_ptp.c
+++ b/drivers/net/ethernet/intel/iavf/iavf_ptp.c
@@ -169,6 +169,55 @@  static int iavf_ptp_gettimex64(struct ptp_clock_info *ptp,
 	return iavf_read_phc_indirect(adapter, ts, sts);
 }
 
+/**
+ * iavf_ptp_cache_phc_time - Cache PHC time for performing timestamp extension
+ * @adapter: private adapter structure
+ *
+ * Periodically cache the PHC time in order to allow for timestamp extension.
+ * This is required because the Tx and Rx timestamps only contain 32bits of
+ * nanoseconds. Timestamp extension allows calculating the corrected 64bit
+ * timestamp. This algorithm relies on the cached time being within ~1 second
+ * of the timestamp.
+ */
+static void iavf_ptp_cache_phc_time(struct iavf_adapter *adapter)
+{
+	if (time_is_before_jiffies(adapter->ptp.cached_phc_updated + HZ)) {
+		/* The response from virtchnl will store the time into
+		 * cached_phc_time
+		 */
+		iavf_send_phc_read(adapter);
+	}
+}
+
+/**
+ * iavf_ptp_do_aux_work - Perform periodic work required for PTP support
+ * @ptp: PTP clock info structure
+ *
+ * Handler to take care of periodic work required for PTP operation. This
+ * includes the following tasks:
+ *
+ *   1) updating cached_phc_time
+ *
+ *      cached_phc_time is used by the Tx and Rx timestamp flows in order to
+ *      perform timestamp extension, by carefully comparing the timestamp
+ *      32bit nanosecond timestamps and determining the corrected 64bit
+ *      timestamp value to report to userspace. This algorithm only works if
+ *      the cached_phc_time is within ~1 second of the Tx or Rx timestamp
+ *      event. This task periodically reads the PHC time and stores it, to
+ *      ensure that timestamp extension operates correctly.
+ *
+ * Returns: time in jiffies until the periodic task should be re-scheduled.
+ */
+long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp)
+{
+	struct iavf_adapter *adapter = clock_to_adapter(ptp);
+
+	iavf_ptp_cache_phc_time(adapter);
+
+	/* Check work about twice a second */
+	return msecs_to_jiffies(500);
+}
+
 /**
  * iavf_ptp_register_clock - Register a new PTP for userspace
  * @adapter: private adapter structure
@@ -189,6 +238,7 @@  static int iavf_ptp_register_clock(struct iavf_adapter *adapter)
 		 dev_name(dev));
 	ptp_info->owner = THIS_MODULE;
 	ptp_info->gettimex64 = iavf_ptp_gettimex64;
+	ptp_info->do_aux_work = iavf_ptp_do_aux_work;
 
 	adapter->ptp.clock = ptp_clock_register(ptp_info, dev);
 	if (IS_ERR(adapter->ptp.clock))
@@ -228,6 +278,8 @@  void iavf_ptp_init(struct iavf_adapter *adapter)
 		return;
 	}
 
+	ptp_schedule_worker(adapter->ptp.clock, 0);
+
 	adapter->ptp.initialized = true;
 }
 
diff --git a/drivers/net/ethernet/intel/iavf/iavf_ptp.h b/drivers/net/ethernet/intel/iavf/iavf_ptp.h
index 4f84416743e1..7a25647980f3 100644
--- a/drivers/net/ethernet/intel/iavf/iavf_ptp.h
+++ b/drivers/net/ethernet/intel/iavf/iavf_ptp.h
@@ -34,5 +34,6 @@  void iavf_ptp_release(struct iavf_adapter *adapter);
 void iavf_ptp_process_caps(struct iavf_adapter *adapter);
 bool iavf_ptp_cap_supported(struct iavf_adapter *adapter, u32 cap);
 void iavf_virtchnl_send_ptp_cmd(struct iavf_adapter *adapter);
+long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp);
 
 #endif /* _IAVF_PTP_H_ */