netdev - Re: [Intel-wired-lan] [PATCH iwl-next v5 08/12] iavf: periodically cache PHC time

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <9d8e656f-1b04-4fc5-a5b4-c04df7d34fdc@intel.com>
Date: Mon, 22 Apr 2024 11:23:05 +0200
From: Mateusz Polchlopek <mateusz.polchlopek@...el.com>
To: Rahul Rameshbabu <rrameshbabu@...dia.com>
CC: <intel-wired-lan@...ts.osuosl.org>, <netdev@...r.kernel.org>,
	<horms@...nel.org>, <anthony.l.nguyen@...el.com>, Jacob Keller
	<jacob.e.keller@...el.com>, Wojciech Drewek <wojciech.drewek@...el.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@...el.com> wrote:
>> From: Jacob Keller <jacob.e.keller@...el.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@...el.com>
>> Signed-off-by: Jacob Keller <jacob.e.keller@...el.com>
>> Co-developed-by: Mateusz Polchlopek <mateusz.polchlopek@...el.com>
>> Signed-off-by: Mateusz Polchlopek <mateusz.polchlopek@...el.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>