Re: [net-next 3/3] net: mscc: ocelot: support PTP Sync one-step timestamping

[Vladimir Oltean <olteanv@gmail.com>]

On Tue, Apr 20, 2021 at 07:33:39AM +0000, Y.b. Lu wrote:
> > > +	/* For two-step timestamp, retrieve ptp_cmd in DSA_SKB_CB_PRIV
> > > +	 * and timestamp ID in clone->cb[0].
> > > +	 * For one-step timestamp, retrieve ptp_cmd in DSA_SKB_CB_PRIV.
> > > +	 */
> > > +	u8 *ptp_cmd = DSA_SKB_CB_PRIV(skb);
> >
> > This is fine in the sense that it works, but please consider creating something
> > similar to sja1105:
> >
> > struct ocelot_skb_cb {
> > 	u8 ptp_cmd; /* For both one-step and two-step timestamping */
> > 	u8 ts_id; /* Only for two-step timestamping */ };
> >
> > #define OCELOT_SKB_CB(skb) \
> > 	((struct ocelot_skb_cb *)DSA_SKB_CB_PRIV(skb))
> >
> > And then access as OCELOT_SKB_CB(skb)->ptp_cmd,
> > OCELOT_SKB_CB(clone)->ts_id.
> >
> > and put a comment to explain that this is done in order to have common code
> > between Felix DSA and Ocelot switchdev. Basically Ocelot will not use the first
> > 8 bytes of skb->cb, but there's enough space for this to not make any
> > difference. The original skb will hold only ptp_cmd, the clone will only hold
> > ts_id, but it helps to have the same structure in place.
> >
> > If you create this ocelot_skb_cb structure, I expect the comment above to be
> > fairly redundant, you can consider removing it.
> >
>
> You're right to define the structure.
> Considering patch #1, move skb cloning to drivers, and populate DSA_SKB_CB(skb)->clone if needs to do so (per suggestion).
> Can we totally drop dsa_skb_cb in dsa core? The only usage of it is holding a skb clone pointer, for only felix and sja1105.
> Actually we can move such pointer in <device>_skb_cb, instead of reserving the space of skb for any drivers.
>
> Do you think so?

The trouble with skb->cb is that it isn't zero-initialized. But somebody
needs to initialize the clone pointer to NULL, otherwise you don't know
if this is a valid pointer or not. Because dsa_skb_tx_timestamp() is
called before p->xmit(), the driver has no way to initialize the clone
pointer by itself. So this was done directly from dsa_slave_xmit(), and
not from any driver-specific hook. So this is why there is a
DSA_SKB_CB(skb)->clone and not SJA1105_SKB_CB(skb)->clone. The
alternative would be to memset(skb->cb, 0, 48) which is a bit
sub-optimal because it initializes more than it needs. Alternatively, it
might be possible to introduce a new property in struct dsa_device_ops
which holds sizeof(struct sja1105_skb_cb), and the generic code will
only zero-initialize this number of bytes.
I don't know, if you can get it to work in a way that does not incur a
noticeable performance penalty, I'm okay with whatever you come up with.