[PATCH v12 net-next 0/3] Implement PTP support in netdevsim

[Maciek Machnikowski <maciek@machnikowski.net>]

This patchset adds support to the PTP HW timestamping emulation in the
netdevsim. It uses existing binding between netdevsim and ptp_mock
driver to generate transmit and receive timestamps.

It also adds the selftest to verify the hw timestamping functionality
running over netdevsim.

v12:
- Unclone the skb before writing hwtstamp to not corrupt the TX errqueue clone
- Clear the TX hwtstamp when peer RX timestamping is off
- Set SKBTX_IN_PROGRESS before skb_tx_timestamp() to avoid SW+HW duplicates
- Use skb_unclone() instead of skb_unclone_full()
- Return -EOPNOTSUPP from nsim_get_ts_config() when there is no PHC
- Use isolated ptp4l UDS paths in the selftest
- Explicitly kill the remote ptp4l leader in the selftest
- Tolerate the deferred pkill finding no ptp4l process in the selftest

v11:
- Avoid aliasing the RX hwtstamp write with AF_XDP completion metadata
  (skb_shared_info's hwtstamps/xsk_meta union) by orphaning the skb
  first, addressing the Critical finding from Simon Horman's review.
  Link: https://lore.kernel.org/netdev/20260819101706.GP265046@horms.kernel.org/
- Only unshare the skb when an RX timestamp is written into it
- Reject non-zero hwtstamp flags and accept the PTP V2 L2 RX filters
- Use READ_ONCE() in ndo_hwtstamp_get and stop overwriting the whole
  config struct
- Skip the selftest if either the local or the remote device lacks
  hardware timestamping support
- Forward declare struct ptp_clock_info in ptp_mock.h
- Fix the ptp_mock commit message wording

v10:
- Drop the local_irq_save()/local_irq_restore() around the PHC reads.

v9:
- Unshare the skb instead of copying it
- Move the PTP timestamping off the xmit fast path
- Sample the TX and RX PHCs with IRQs disabled to keep the reads adjacent
- Validate the hwtstamp config before touching device state
- Report a -1 PHC index from ethtool for ports without a PHC (VFs)
- Clamp the ptp4l poll timeout in the selftest to a non-negative value

v8:
- Fix the lockless tstamp_config access in netdevsim with
  READ_ONCE()/WRITE_ONCE()
- Clarify the TX/RX timestamp handling in netdevsim comments

v7:
- Moved the test script to tools/testing/selftest/drivers/net
- Optimized implementation to avoid potential edge cases reported

v6:
- Enable PTP Mock in the tools/testing/selftests/net/config

v5:
- Rebase

v4:
- Check if Rx timestamps are enabled before generating a timestamp
- Replace bash selftest script with a python one
- Optimized Tx timestamp generation

v3:
- Fixed shellcheck issues in the selftest/net/ptp.sh
- Added selftest/net/ptp.sh to the selftest/net/Makefile
- Modified ptp_mock to use spin_lock_bh

v2:
- Added selftest/net/ptp.sh
- Modified ptp_mock to use spin_lock_bh
- Populate ethtool defaults using ethtool_op_get_ts_info

Maciek Machnikowski (3):
  ptp_mock: Expose ptp_clock_info to external drivers
  netdevsim: Implement basic ptp support
  selftests: drivers/net: Implement ptp4l sync test using netdevsim

 drivers/net/netdevsim/ethtool.c              |  15 +++
 drivers/net/netdevsim/netdev.c               | 124 +++++++++++++++++++
 drivers/net/netdevsim/netdevsim.h            |   1 +
 drivers/ptp/ptp_mock.c                       |  26 ++--
 include/linux/ptp_mock.h                     |   7 ++
 tools/testing/selftests/drivers/net/Makefile |   1 +
 tools/testing/selftests/drivers/net/config   |   1 +
 tools/testing/selftests/drivers/net/ptp.py   | 112 +++++++++++++++++
 8 files changed, 279 insertions(+), 8 deletions(-)
 create mode 100755 tools/testing/selftests/drivers/net/ptp.py

-- 
2.56.0