Re: [RFC PATCH 2/3] net: sparx5: Add Sparx5 switchdev driver

[Steen Hegelund]

On 28.11.2020 19:45, Andrew Lunn wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>
> > +/* Add a potentially wrapping 32 bit value to a 64 bit counter */
> > +static inline void sparx5_update_counter(u64 *cnt, u32 val)
> > +{
> > +     if (val < (*cnt & U32_MAX))
> > +             *cnt += (u64)1 << 32; /* value has wrapped */
> > +
> > +     *cnt = (*cnt & ~(u64)U32_MAX) + val;
> > +}
>
> I don't follow what this is doing. Could you give some examples?

The statistics counters comes from different sources, and unfortunately
have different layouts: Some are 32 bit and some a 40 bit, so this
function is a wrapper to be able to handle them in the same way, polling
the counters often enough to be able to catch overruns.

> > +static const char *const sparx5_stats_layout[] = {
> > +     "rx_in_bytes",
> > +     "rx_symbol_err",
> > +     "rx_pause",
> > +     "rx_unsup_opcode",
>
> > +static void sparx5_update_port_stats(struct sparx5 *sparx5, int portno)
> > +{
> > +     struct sparx5_port *spx5_port = sparx5->ports[portno];
> > +     bool high_speed_dev = sparx5_is_high_speed_device(&spx5_port->conf);
>
> Reverse christmas tree. Which in this case, means you need to move the
> assignment into the body of the code.

OK.
> > +static void sparx5_get_sset_strings(struct net_device *ndev, u32 sset, u8 *data)
> > +{
> > +     struct sparx5_port *port = netdev_priv(ndev);
> > +     struct sparx5  *sparx5 = port->sparx5;
> > +     int idx;
> > +
> > +     if (sset != ETH_SS_STATS)
> > +             return;
> > +
> > +     for (idx = 0; idx < sparx5->num_stats; idx++)
> > +             memcpy(data + idx * ETH_GSTRING_LEN,
> > +                    sparx5->stats_layout[idx], ETH_GSTRING_LEN);
>
> You cannot use memcpy here, because the strings you have defined are
> not ETH_GSTRING_LEN long. We once had a driver which happened to have
> its strings at the end of a page. The memcpy would copy the string,
> but keep going passed the end of string, over the page boundary, and
> trigger a segmentation fault.

Yes, I see that.

Thanks for the comments
/Steen
>        Andrew

BR
Steen

---------------------------------------
Steen Hegelund
steen.hegelund@xxxxxxxxxxxxx