Re: [PATCH v2 5/5] io_uring/netcmd: add tx timestamping cmd support

[Jason Xing <kerneljasonxing@gmail.com>]

On Fri, Jun 6, 2025 at 7:55 AM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> Vadim Fedorenko wrote:
> > On 05/06/2025 01:59, Willem de Bruijn wrote:
> > > Pavel Begunkov wrote:
> > >> Add a new socket command which returns tx time stamps to the user. It
> > >> provide an alternative to the existing error queue recvmsg interface.
> > >> The command works in a polled multishot mode, which means io_uring will
> > >> poll the socket and keep posting timestamps until the request is
> > >> cancelled or fails in any other way (e.g. with no space in the CQ). It
> > >> reuses the net infra and grabs timestamps from the socket's error queue.
> > >>
> > >> The command requires IORING_SETUP_CQE32. All non-final CQEs (marked with
> > >> IORING_CQE_F_MORE) have cqe->res set to the tskey, and the upper 16 bits
> > >> of cqe->flags keep tstype (i.e. offset by IORING_CQE_BUFFER_SHIFT). The
> > >> timevalue is store in the upper part of the extended CQE. The final
> > >> completion won't have IORING_CQR_F_MORE and will have cqe->res storing
> > >> 0/error.
> > >>
> > >> Suggested-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
> > >> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> > >> ---
> > >>   include/uapi/linux/io_uring.h |  6 +++
> > >>   io_uring/cmd_net.c            | 77 +++++++++++++++++++++++++++++++++++
> > >>   2 files changed, 83 insertions(+)
> > >>
> > >> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> > >> index cfd17e382082..0bc156eb96d4 100644
> > >> --- a/include/uapi/linux/io_uring.h
> > >> +++ b/include/uapi/linux/io_uring.h
> > >> @@ -960,6 +960,11 @@ struct io_uring_recvmsg_out {
> > >>    __u32 flags;
> > >>   };
> > >>
> > >> +struct io_timespec {
> > >> +  __u64           tv_sec;
> > >> +  __u64           tv_nsec;
> > >> +};
> > >> +
> > >>   /*
> > >>    * Argument for IORING_OP_URING_CMD when file is a socket
> > >>    */
> > >> @@ -968,6 +973,7 @@ enum io_uring_socket_op {
> > >>    SOCKET_URING_OP_SIOCOUTQ,
> > >>    SOCKET_URING_OP_GETSOCKOPT,
> > >>    SOCKET_URING_OP_SETSOCKOPT,
> > >> +  SOCKET_URING_OP_TX_TIMESTAMP,
> > >>   };
> > >>
> > >>   /* Zero copy receive refill queue entry */
> > >> diff --git a/io_uring/cmd_net.c b/io_uring/cmd_net.c
> > >> index e99170c7d41a..dae59aea5847 100644
> > >> --- a/io_uring/cmd_net.c
> > >> +++ b/io_uring/cmd_net.c
> > >> @@ -1,5 +1,6 @@
> > >>   #include <asm/ioctls.h>
> > >>   #include <linux/io_uring/net.h>
> > >> +#include <linux/errqueue.h>
> > >>   #include <net/sock.h>
> > >>
> > >>   #include "uring_cmd.h"
> > >> @@ -51,6 +52,80 @@ static inline int io_uring_cmd_setsockopt(struct socket *sock,
> > >>                              optlen);
> > >>   }
> > >>
> > >> +static bool io_process_timestamp_skb(struct io_uring_cmd *cmd, struct sock *sk,
> > >> +                               struct sk_buff *skb, unsigned issue_flags)
> > >> +{
> > >> +  struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
> > >> +  struct io_uring_cqe cqe[2];
> > >> +  struct io_timespec *iots;
> > >> +  struct timespec64 ts;
> > >> +  u32 tskey;
> > >> +
> > >> +  BUILD_BUG_ON(sizeof(struct io_uring_cqe) != sizeof(struct io_timespec));
> > >> +
> > >> +  if (!skb_get_tx_timestamp(skb, sk, &ts))
> > >> +          return false;
> > >> +
> > >> +  tskey = serr->ee.ee_data;
> > >> +
> > >> +  cqe->user_data = 0;
> > >> +  cqe->res = tskey;
> > >> +  cqe->flags = IORING_CQE_F_MORE;
> > >> +  cqe->flags |= (u32)serr->ee.ee_info << IORING_CQE_BUFFER_SHIFT;
> > >> +
> > >> +  iots = (struct io_timespec *)&cqe[1];
> > >> +  iots->tv_sec = ts.tv_sec;
> > >> +  iots->tv_nsec = ts.tv_nsec;
> > >
> > > skb_get_tx_timestamp loses the information whether this is a
> > > software or a hardware timestamp. Is that loss problematic?
> > >
> > > If a process only requests one type of timestamp, it will not be.
> > >
> > > But when requesting both (SOF_TIMESTAMPING_OPT_TX_SWHW) this per cqe
> > > annotation may be necessary.
> >
> > skb_has_tx_timestamp() helper has clear priority of software timestamp,
> > if enabled for the socket. Looks like SOF_TIMESTAMPING_OPT_TX_SWHW case
> > won't produce both timestamps with the current implementation. Am I
> > missing something?
>
> The point of that option is to request both SW and HW tx
> timestamps. Before that option, the SW timestamp was suppressed if a
> HW timestamp was pending.

Agree, the regular driver does such an implementation in such an order:
XXX_xmit()
{
    if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)...
        skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;

    skb_tx_timestamp(skb); // here, it would not generate timestamp
with SKBTX_IN_PROGRESS being set in advance
}

If both sw and hw flags are set, the software one will not take any effect.

Thanks,
Jason