[PATCH v2 net-next 0/7] net: dsa: netc: add PTP support for NETC switch

From: Wei Fang <[email protected]>

This series adds PTP hardware timestamping support to the NETC switch
DSA driver. The NETC switch has no time registers of its own; it shares
the PTP time base of the NETC Timer, which is a separate PCIe function
driven by the ptp_netc timer driver. The series therefore first prepares
the timer driver to expose the current PTP time to other drivers, then
builds RX, two-step TX and one-step TX timestamping on top of it in the
switch driver and its DSA tagger.

The series is organized in two parts.

Preparation of the NETC Timer driver (patches 1-3):

  1) Convert the open-coded 64-bit register accesses to the
     ioread64_lo_hi()/iowrite64_lo_hi() helpers, wrapped in
     netc_timer_rd64()/netc_timer_wr64(), to reduce boilerplate and make
     the access width explicit. No functional change.

  2) Drop the pcie_flr() call in probe. Per the reference manual, function
     level reset does not apply to the Timer as a supporting function, so
     the call has no effect.

  3) Export netc_timer_get_current_time() so the switch driver, a separate
     PCIe function, can read the shared PTP time. The helper is declared
     in <linux/fsl/netc_global.h> under CONFIG_PTP_NETC_V4_TIMER with a
     stub returning 0 when the timer is disabled, so callers need no hard
     dependency on the timer driver, and returns 0 when the timer has not
     probed or is gone so the caller can handle it gracefully.

NETC switch PTP support (patches 4-7):

  4) Track the host flood rule by entry ID instead of by ipft_entry_data
     pointer, freeing the descriptor as soon as the hardware entry is
     committed. This removes a long-lived heap allocation and lets
     netc_free_host_flood_rules() go away, in preparation for the
     timestamping rules added next.

  5) Enable the ingress port filtering lookup (IPFT) by default. A frame
     that matches no entry is simply passed on, so leaving the lookup
     always enabled simplifies the logic and avoids tracking whether a
     port already has an IPFT entry, which the RX timestamping rules rely
     on.

  6) Add two-step TX timestamping and RX timestamping. RX installs IPFT
     rules that redirect PTP frames (L2, L4 over IPv4/IPv6, event and
     general) to the CPU port; the hardware prepends a To_Host tag with
     the 64-bit ingress timestamp, which the tagger hands to
     netc_port_rxtstamp(). Two-step TX clones the skb, allocates a 4-bit
     request ID carried in a To_Port subtype 2 tag, and completes the
     clone when the hardware echoes the ID and transmit timestamp back in
     a To_Host response frame.

  7) Add one-step TX timestamping for PTP Sync frames. The MAC updates the
     correction field in flight using the per-port PM_SINGLE_STEP register,
     which is a single register that must be programmed per frame, so
     one-step Sync transmission is serialized per port with a per-port PTP
     spinlock guarding an in-flight slot and a deferral queue.

---
v2:
1. Change IS_ENABLED to IS_REACHABLE in netc_global.h
2. Move spin_lock_init() before pci_set_drvdata() in
   netc_timer_pci_probe()
3. Add device_lock() in netc_timer_get_current_time() to avoid
   use-after-free
4. Remove SOF_TIMESTAMPING_*_SOFTWARE flags
5. Use READ_ONCE/WRITE_ONCE to access np->ptp_tx_type
6. Clear NETC_SKB_CB(skb)->tstamp in netc_rcv()
7. Set np->ptp_tx_type at the end of netc_port_hwtstamp_set()
8. netc_port_txtstamp_twostep() return type changed from int from void
9. Add netc_port_purge_txtstamp_queue, which is called in
   netc_port_hwtstamp_set() when HWTSTAMP_TX_OFF is set
10. Remove clone from struct netc_skb_cb
11. Get ts_req_id from skb->cb (since clone is removed from netc_skb_cb)
    in netc_fill_tp_tag_subtype2()
12. Change the return type of netc_rx_tstamp_process() to int and add
    pskb_may_pull() to check whether the Ethertype header is in the
    linear buffer of the skb
13. netc_get_phc_index() return -1 instead of -ENODEV when priv->tmr_dev
    is NULL
14. Refactor the entire patch 7, introduce struct netc_onestep and
    related interfaces, such as netc_onestep_get/put/release and so on
v1 link: https://lore.kernel.org/imx/[email protected]/
---

Wei Fang (7):
  ptp: netc: use ioread64_lo_hi/iowrite64_lo_hi for 64-bit register
    access
  ptp: netc: remove unnecessary pcie_flr() call in probe
  ptp: netc: export netc_timer_get_current_time() for cross-driver use
  net: dsa: netc: use entry ID instead of pointer to track host flood
    rule
  net: dsa: netc: enable ingress port filtering lookup by default
  net: dsa: netc: add PTP two-step timestamping support
  net: dsa: netc: add PTP one-step timestamping support

 drivers/net/dsa/netc/Kconfig          |   1 +
 drivers/net/dsa/netc/Makefile         |   3 +-
 drivers/net/dsa/netc/netc_main.c      | 183 ++++--
 drivers/net/dsa/netc/netc_platform.c  |   1 +
 drivers/net/dsa/netc/netc_ptp.c       | 866 ++++++++++++++++++++++++++
 drivers/net/dsa/netc/netc_switch.h    |  91 ++-
 drivers/net/dsa/netc/netc_switch_hw.h |   5 +
 drivers/ptp/ptp_netc.c                | 110 ++--
 include/linux/dsa/tag_netc.h          |  43 ++
 include/linux/fsl/netc_global.h       |  10 +
 net/dsa/tag_netc.c                    | 214 ++++++-
 11 files changed, 1428 insertions(+), 99 deletions(-)
 create mode 100644 drivers/net/dsa/netc/netc_ptp.c

-- 
2.34.1