Add PHC support with error bound value for ENA driver

Add PHC support with error bound value for ENA driver

[Bernstein, Amit]

Hi Thomas, Richard and all,

In one of the upcoming releases, we're planning to add PHC support to the ENA driver (/drivers/net/ethernet/amazon/ena/)
To provide the best experience for service customers, the ENA driver will expose an error_bound parameter (expressed in nanoseconds), 
which will represent the maximal clock error on each given PHC timestamp.
The error bound is calculated by the device,  taking into account the accuracy and delays of all the device's PHC components.

Based on our search, there is no similar functionality in other drivers, meaning there is no user interface to expose it.
We're currently exploring the best method of exposing this capability to the user.
Our debate is between:
- Extending the gettimex64 API in ptp_clock_info
- Introducing a new devlink entry
- Updating the time-related ethtool option (-T flag).

As our device sends each PHC timestamp with an error_bound value together, gettimex64 is the reasonable option for us and our recommended solution.
We would like to ask for your recommendation.

We have already consulted with Jakub, who recommended that we consult with all of you.

For more information, you may visit: https://aws.amazon.com/blogs/compute/its-about-time-microsecond-accurate-clocks-on-amazon-ec2-instances/

Thanks,
Amit

Re: Add PHC support with error bound value for ENA driver

[Richard Cochran]

On Thu, Feb 01, 2024 at 07:24:51AM +0000, Bernstein, Amit wrote:

> In one of the upcoming releases, we're planning to add PHC support
> to the ENA driver (/drivers/net/ethernet/amazon/ena/) To provide the
> best experience for service customers, the ENA driver will expose an
> error_bound parameter (expressed in nanoseconds), which will
> represent the maximal clock error on each given PHC timestamp.

> As our device sends each PHC timestamp with an error_bound value
> together, gettimex64 is the reasonable option for us and our
> recommended solution.  We would like to ask for your recommendation.

Sounds like gettimex64 won't help you.

You will need to extend the socket time stamping API.  That is how
time stamps are delivered to user space.

HTH,
Richard

RE: Add PHC support with error bound value for ENA driver

[Bernstein, Amit]

> On Thu, Feb 01, 2024 at 07:24:51AM +0000, Bernstein, Amit wrote:
> 
> > In one of the upcoming releases, we're planning to add PHC support to
> > the ENA driver (/drivers/net/ethernet/amazon/ena/) To provide the best
> > experience for service customers, the ENA driver will expose an
> > error_bound parameter (expressed in nanoseconds), which will represent
> > the maximal clock error on each given PHC timestamp.
> 
> > As our device sends each PHC timestamp with an error_bound value
> > together, gettimex64 is the reasonable option for us and our
> > recommended solution.  We would like to ask for your recommendation.
> 
> Sounds like gettimex64 won't help you.
> 
> You will need to extend the socket time stamping API.  That is how time
> stamps are delivered to user space.
> 
> HTH,
> Richard

Richard, thank you for your response.
However, I forgot to mention that the ENA driver currently doesn't support any socket timestamping.
It only supports the PTP hardware clock infrastructure, which is exported using the clock driver.
User space programs can control the clock driver via the PTP_SYS_OFFSET ioctl.
Therefore, we can only use the basic clock operations of ptp_clock_info, such as gettime/gettime64/gettimex64.

Thanks,
Amit