Re: [PATCH net-next v1 1/1] net: dsa: microchip: add stats64 support for ksz8 series of switches

[Oleksij Rempel]

On Tue, Dec 06, 2022 at 07:08:01PM +0200, Vladimir Oltean wrote:
> On Mon, Dec 05, 2022 at 06:29:04AM +0100, Oleksij Rempel wrote:
> > +void ksz88xx_r_mib_stats64(struct ksz_device *dev, int port)
> > +{
> > +	struct ethtool_pause_stats *pstats;
> > +	struct rtnl_link_stats64 *stats;
> > +	struct ksz88xx_stats_raw *raw;
> > +	struct ksz_port_mib *mib;
> > +
> > +	mib = &dev->ports[port].mib;
> > +	stats = &mib->stats64;
> > +	pstats = &mib->pause_stats;
> > +	raw = (struct ksz88xx_stats_raw *)mib->counters;
> > +
> > +	spin_lock(&mib->stats64_lock);
> > +
> > +	stats->rx_packets = raw->rx_bcast + raw->rx_mcast + raw->rx_ucast +
> > +		raw->rx_pause;
> > +	stats->tx_packets = raw->tx_bcast + raw->tx_mcast + raw->tx_ucast +
> > +		raw->tx_pause;
> > +
> > +	/* HW counters are counting bytes + FCS which is not acceptable
> > +	 * for rtnl_link_stats64 interface
> > +	 */
> > +	stats->rx_bytes = raw->rx + raw->rx_hi - stats->rx_packets * ETH_FCS_LEN;
> > +	stats->tx_bytes = raw->tx + raw->tx_hi - stats->tx_packets * ETH_FCS_LEN;
> 
> What are rx_hi, tx_hi compared to rx, tx?

rx, tx are packets with normal priority and rx_hi, tx_hi are packets
with high prio.

> > +
> > +	stats->rx_length_errors = raw->rx_undersize + raw->rx_fragments +
> > +		raw->rx_oversize;
> > +
> > +	stats->rx_crc_errors = raw->rx_crc_err;
> > +	stats->rx_frame_errors = raw->rx_align_err;
> > +	stats->rx_dropped = raw->rx_discards;
> > +	stats->rx_errors = stats->rx_length_errors + stats->rx_crc_errors +
> > +		stats->rx_frame_errors  + stats->rx_dropped;
> > +
> > +	stats->tx_window_errors = raw->tx_late_col;
> > +	stats->tx_fifo_errors = raw->tx_discards;
> > +	stats->tx_aborted_errors = raw->tx_exc_col;
> > +	stats->tx_errors = stats->tx_window_errors + stats->tx_fifo_errors +
> > +		stats->tx_aborted_errors;
> > +
> > +	stats->multicast = raw->rx_mcast;
> > +	stats->collisions = raw->tx_total_col;
> > +
> > +	pstats->tx_pause_frames = raw->tx_pause;
> > +	pstats->rx_pause_frames = raw->rx_pause;
> 
> FWIW, ksz_get_pause_stats() can sleep, just ksz_get_stats64() can't. So
> the pause stats don't need to be periodically read (unless you want to
> do that to prevent 32-bit overflows).

KSZ driver is using worker to read stats periodically. Since all needed
locks are already taken, I copy pause stats as well.

Otherwise it will need some different locking, which will make things
look different but do not reduce CPU load. 

Regards,
Oleksij
-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |