Re: CAN message timestamping

[Austin Schuh <austin@peloton-tech.com>]

On Mon, Mar 28, 2016 at 6:51 PM, Austin Schuh <austin@peloton-tech.com> wrote:
> Hi Oliver,
>
> I spent some time trying recvmsg, but it still only gives me
> timestamps with the real-time clock.  I do like the interface much
> better.  Thanks!
>
> I was able to get
>   setsockopt(socket_, SOL_SOCKET, SO_TIMESTAMPNS,
>       &enabled, sizeof(enabled))
> and
>   const int stamping_val = SOF_TIMESTAMPING_SOFTWARE |
>       SOF_TIMESTAMPING_SYS_HARDWARE |
>       SOF_TIMESTAMPING_RAW_HARDWARE;
>   setsockopt(socket_, SOL_SOCKET, SO_TIMESTAMPING, &stamping_val,
>       sizeof(stamping_val))
> to successfully timestamp values with recvmsg.
>
> I0328 18:25:09.927222 12739 can_data_source.cc:144] 0: 1459214709 927191637
> I0328 18:25:09.927227 12739 can_data_source.cc:146] 1: 0 0
> I0328 18:25:09.927233 12739 can_data_source.cc:148] 2: 0 0
> I0328 18:25:09.927240 12739 can_data_source.cc:169] Capture time is
> 1459214709927191637ns
>
> The time-stamp is definitely using the real-time clock.  The hardware
> timestamps are not filled in.  Maybe I could modify the sja1000-peak
> driver to abuse one of the hardware timestamps to return the timestamp
> I am looking for?  I'd really rather not have to run a custom
> kernel...
>
> Austin
>
> On Wed, Mar 23, 2016 at 12:00 AM Oliver Hartkopp <socketcan@hartkopp.net> wrote:
>>
>> Hi Austin,
>>
>> On 23.03.2016 06:12, Austin Schuh wrote:
>> > We recently tried doing CAN timestamping with the ioctl(s, SIOCGSTAMP,
>> > &tv) call.
>>
>> Did you ever try to use the recvmsg() call with enabled SO_TIMESTAMPING socket
>> option?
>>
>> See:
>>
>> https://www.kernel.org/doc/Documentation/networking/timestamping.txt
>> https://www.kernel.org/doc/Documentation/networking/timestamping/
>>
>> The candump tool also uses recvmsg() - but only with SO_TIMESTAMP so far which
>> does not provide the other RX timestamps (link HW TS).
>>
>> Regards,
>> Oliver
>>

Turns out timestamping in the driver is pretty easy.  The following
seems to be working for me.  (comments welcome!)  I don't think this
is something that should be up streamed, but I'm including it here in
case there is other interest.  I'm reading both clocks in the ISR to
reduce the amount of time difference between when they are both read.

$ git diff
diff --git a/drivers/net/can/sja1000/sja1000.c
b/drivers/net/can/sja1000/sja1000.c
index 76ef900..55d6583 100644
--- a/drivers/net/can/sja1000/sja1000.c
+++ b/drivers/net/can/sja1000/sja1000.c
@@ -370,6 +370,10 @@ static void sja1000_rx(struct net_device *dev)
        /* release receive buffer */
        sja1000_write_cmdreg(priv, CMD_RRB);

+       struct skb_shared_hwtstamps *shhwtstamps =
+           skb_hwtstamps(skb);
+       shhwtstamps->syststamp = ktime_get();
+       skb->tstamp = ktime_get_real();
        netif_rx(skb);

        stats->rx_packets++;

Austin