netdev - Re: [RFC net-next] net: mvneta: align xdp stats naming scheme to mlx5 driver

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200218171716.GA13376@localhost.localdomain>
Date:   Tue, 18 Feb 2020 18:17:16 +0100
From:   Lorenzo Bianconi <lorenzo@...nel.org>
To:     Jesper Dangaard Brouer <brouer@...hat.com>
Cc:     netdev@...r.kernel.org, ilias.apalodimas@...aro.org,
        davem@...emloft.net, lorenzo.bianconi@...hat.com, andrew@...n.ch,
        dsahern@...nel.org, bpf@...r.kernel.org,
        Toke Høiland-Jørgensen <toke@...hat.com>
Subject: Re: [RFC net-next] net: mvneta: align xdp stats naming scheme to
 mlx5 driver

> On Tue, 18 Feb 2020 01:14:29 +0100
> Lorenzo Bianconi <lorenzo@...nel.org> wrote:
> 
> > Introduce "rx" prefix in the name scheme for xdp counters
> > on rx path.
> > Differentiate between XDP_TX and ndo_xdp_xmit counters
> > 
> > Signed-off-by: Lorenzo Bianconi <lorenzo@...nel.org>
> > ---
> >  drivers/net/ethernet/marvell/mvneta.c | 22 +++++++++++++++++-----
> >  1 file changed, 17 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > index b7045b6a15c2..6223700dc3df 100644
> > --- a/drivers/net/ethernet/marvell/mvneta.c
> > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > @@ -344,6 +344,7 @@ enum {
> >  	ETHTOOL_XDP_REDIRECT,
> >  	ETHTOOL_XDP_PASS,
> >  	ETHTOOL_XDP_DROP,
> > +	ETHTOOL_XDP_XMIT,
> >  	ETHTOOL_XDP_TX,
> >  	ETHTOOL_MAX_STATS,
> >  };
> > @@ -399,10 +400,11 @@ static const struct mvneta_statistic mvneta_statistics[] = {
> >  	{ ETHTOOL_STAT_EEE_WAKEUP, T_SW, "eee_wakeup_errors", },
> >  	{ ETHTOOL_STAT_SKB_ALLOC_ERR, T_SW, "skb_alloc_errors", },
> >  	{ ETHTOOL_STAT_REFILL_ERR, T_SW, "refill_errors", },
> > -	{ ETHTOOL_XDP_REDIRECT, T_SW, "xdp_redirect", },
> > -	{ ETHTOOL_XDP_PASS, T_SW, "xdp_pass", },
> > -	{ ETHTOOL_XDP_DROP, T_SW, "xdp_drop", },
> > -	{ ETHTOOL_XDP_TX, T_SW, "xdp_tx", },
> > +	{ ETHTOOL_XDP_REDIRECT, T_SW, "rx_xdp_redirect", },
> > +	{ ETHTOOL_XDP_PASS, T_SW, "rx_xdp_pass", },
> > +	{ ETHTOOL_XDP_DROP, T_SW, "rx_xdp_drop", },
> > +	{ ETHTOOL_XDP_TX, T_SW, "rx_xdp_tx_xmit", },
> 
> Hmmm... "rx_xdp_tx_xmit", I expected this to be named "rx_xdp_tx" to
> count the XDP_TX actions, but I guess this means something else.

just reused mlx5 naming scheme here :)

> 
> > +	{ ETHTOOL_XDP_XMIT, T_SW, "tx_xdp_xmit", },
> 
> Okay, maybe.  I guess, this will still be valid for when we add an XDP
> egress/TX-hook point.

same here

> 
> >  };
> >  
> >  struct mvneta_stats {
> > @@ -414,6 +416,7 @@ struct mvneta_stats {
> >  	u64	xdp_redirect;
> >  	u64	xdp_pass;
> >  	u64	xdp_drop;
> > +	u64	xdp_xmit;
> >  	u64	xdp_tx;
> >  };
> >  
> > @@ -2050,7 +2053,10 @@ mvneta_xdp_submit_frame(struct mvneta_port *pp, struct mvneta_tx_queue *txq,
> >  	u64_stats_update_begin(&stats->syncp);
> >  	stats->es.ps.tx_bytes += xdpf->len;
> >  	stats->es.ps.tx_packets++;
> > -	stats->es.ps.xdp_tx++;
> > +	if (buf->type == MVNETA_TYPE_XDP_NDO)
> > +		stats->es.ps.xdp_xmit++;
> > +	else
> > +		stats->es.ps.xdp_tx++;
> 
> I don't like that you add a branch (if-statement) in this fast-path code.
> 
> Do we really need to account in the xmit frame function, if this was a
> XDP_REDIRECT or XDP_TX that started the xmit?  I mean we already have
> action counters for XDP_REDIRECT and XDP_TX (but I guess you skipped
> the XDP_TX action counter). 

ack, good point..I think we can move the code in
mvneta_xdp_xmit_back/mvneta_xdp_xmit in order to avoid the if() condition.
Moreover we can move it out the for loop in mvneta_xdp_xmit().
I will fix in a formal patch

> 
> 
> >  	u64_stats_update_end(&stats->syncp);
> >  
> >  	mvneta_txq_inc_put(txq);
> > @@ -4484,6 +4490,7 @@ mvneta_ethtool_update_pcpu_stats(struct mvneta_port *pp,
> >  		u64 xdp_redirect;
> >  		u64 xdp_pass;
> >  		u64 xdp_drop;
> > +		u64 xdp_xmit;
> >  		u64 xdp_tx;
> >  
> >  		stats = per_cpu_ptr(pp->stats, cpu);
> > @@ -4494,6 +4501,7 @@ mvneta_ethtool_update_pcpu_stats(struct mvneta_port *pp,
> >  			xdp_redirect = stats->es.ps.xdp_redirect;
> >  			xdp_pass = stats->es.ps.xdp_pass;
> >  			xdp_drop = stats->es.ps.xdp_drop;
> > +			xdp_xmit = stats->es.ps.xdp_xmit;
> >  			xdp_tx = stats->es.ps.xdp_tx;
> >  		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
> >  
> > @@ -4502,6 +4510,7 @@ mvneta_ethtool_update_pcpu_stats(struct mvneta_port *pp,
> >  		es->ps.xdp_redirect += xdp_redirect;
> >  		es->ps.xdp_pass += xdp_pass;
> >  		es->ps.xdp_drop += xdp_drop;
> > +		es->ps.xdp_xmit += xdp_xmit;
> >  		es->ps.xdp_tx += xdp_tx;
> >  	}
> >  }
> > @@ -4555,6 +4564,9 @@ static void mvneta_ethtool_update_stats(struct mvneta_port *pp)
> >  			case ETHTOOL_XDP_TX:
> >  				pp->ethtool_stats[i] = stats.ps.xdp_tx;
> >  				break;
> > +			case ETHTOOL_XDP_XMIT:
> > +				pp->ethtool_stats[i] = stats.ps.xdp_xmit;
> > +				break;
> >  			}
> >  			break;
> >  		}
> 
> It doesn't look like you have an action counter for XDP_TX, but we have
> one for XDP_REDIRECT?

I did not get you here sorry, I guess they should be accounted in two separated
counters.

Regards,
Lorenzo

> 
> -- 
> Best regards,
>   Jesper Dangaard Brouer
>   MSc.CS, Principal Kernel Engineer at Red Hat
>   LinkedIn: http://www.linkedin.com/in/brouer
> 

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)