Re: [PATCH net] net: Drop the lock in skb_may_tx_timestamp()

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2026-02-16 16:47:03 [+0000], Vadim Fedorenko wrote:
> > > Such as a packet socket sending and hitting the __skb_tstamp_tx
> > > SCM_TSTAMP_SCHED at the top of __dev_queue_xmit.
> > 
> > If that is the case, then we could create a RCU read section here.
> 
> well, __dev_queue_xmit() grabs rcu_read_lock_bh() already, we just have
> to move it right before timestamping check:

For __skb_tstamp_tx() there is:
- hci_conn_tx_queue() -> __skb_tstamp_tx() which appears to be called
  from a worker and has a GPF_KERNEL allocation later.
- hci_num_comp_pkts_evt() (does acquire a mutex) ->
  hci_conn_tx_dequeue() -> __skb_tstamp_tx()
- prueth_rx_mgm_ts_thread_sr1() (explicit threaded IRQ) ->
  prueth_tx_ts_sr1() -> skb_tstamp_tx()

There is also skb_complete_tx_timestamp():
- ines_txtstamp_work() -> skb_complete_tx_timestamp()
- nxp_c45_do_aux_work() -> nxp_c45_process_txts() ->
  skb_complete_tx_timestamp()
- ksz_ptp_txtstamp_skb() -> skb_complete_tx_timestamp()
  and ksz_ptp_txtstamp_skb() has a wait_for_completion_timeout() so no
  RCU here either.
- hellcreek_txtstamp_work() -> skb_complete_tx_timestamp()

so maybe an explicit RCU region is easier.

Sebastian