| Message ID | 20221202201528.26634-1-rrameshbabu@nvidia.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | ptp: Introduce .getfine callback to ptp_clock_info | expand |
On Fri, Dec 02, 2022 at 12:15:27PM -0800, Rahul Rameshbabu wrote: > The current state of the ptp driver provides the ability to query the frequency > of the ptp clock device by caching the frequency values used in previous > adjustments done through the ptp driver. This works great when the ptp driver is > the only means for changing the clock frequency. However, some devices support > ways to adjust the frequency outside the ptp driver stack. The kernel provides no other way to adjust the frequency. So NAK on this series. Thanks, Richard
On 12/2/2022 12:45 PM, Richard Cochran wrote: > On Fri, Dec 02, 2022 at 12:15:27PM -0800, Rahul Rameshbabu wrote: >> The current state of the ptp driver provides the ability to query the frequency >> of the ptp clock device by caching the frequency values used in previous >> adjustments done through the ptp driver. This works great when the ptp driver is >> the only means for changing the clock frequency. However, some devices support >> ways to adjust the frequency outside the ptp driver stack. > > The kernel provides no other way to adjust the frequency. > > So NAK on this series. > > Thanks, > Richard > Agreed. Just because your driver does something silly such as modifying the frequency out of band of PTP stack doesn't mean we need to support it. As for running multiple ptp4l instances on the same PTP clock, we already have the ability to report the frequency out to user space. The ptp4l application checks this at init but then begins assuming that it is the only software controlling the clock. That seems like a reasonable assumption to me. If you want to write your own user space software which handles this, you already have the mechanism to do so. Thanks, Jake
The current state of the ptp driver provides the ability to query the frequency of the ptp clock device by caching the frequency values used in previous adjustments done through the ptp driver. This works great when the ptp driver is the only means for changing the clock frequency. However, some devices support ways to adjust the frequency outside the ptp driver stack. When this occurs, the ptp cached value is inaccurate. Also, if the device for some reason does not start with a frequency of zero, the initial frequency queried before the first frequency adjustment would be inaccurate. The changes proposed enable a callback that various drivers can implement to provide the frequency, in units of scaled ppm, of the ptp clock. When the callback is implemented, it will be called to provide the frequency value instead of the cached dialed_frequency value. There are cases in linuxptp usage where the caching of the frequency has proven to be inadequate. One example was the case where concurrent ptp4l processes adjust frequency, but the cached frequency values in the ptp4l processes are stale. Permitting implementers the ability to provide the frequency would resolve issues involving stale frequency values due to changes occurring outside the ptp driver stack. https://sourceforge.net/p/linuxptp/mailman/message/37720193/ Signed-off-by: Rahul Rameshbabu <rrameshbabu@nvidia.com> Rahul Rameshbabu (2): ptp: Add .getfine function to support reporting frequency offset from hardware net/mlx5: Implement ptp_clock_info .getfine function .../ethernet/mellanox/mlx5/core/lib/clock.c | 35 +++++++++++++++++++ drivers/ptp/ptp_clock.c | 18 +++++++++- include/linux/ptp_clock_kernel.h | 6 ++++ 3 files changed, 58 insertions(+), 1 deletion(-)