RE: [PATCH v3 net-next 12/13] net: enetc: add preliminary support for i.MX95 ENETC PF

[Wei Fang]

> > -----Original Message-----
> > From: Wei Fang <wei.fang@xxxxxxx>
> > Sent: Thursday, October 17, 2024 10:47 AM
> [...]
> > Subject: [PATCH v3 net-next 12/13] net: enetc: add preliminary support for
> > i.MX95 ENETC PF
> >
> [...]
> > diff --git a/drivers/net/ethernet/freescale/enetc/enetc.c
> > b/drivers/net/ethernet/freescale/enetc/enetc.c
> > index bccbeb1f355c..927beccffa6b 100644
> > --- a/drivers/net/ethernet/freescale/enetc/enetc.c
> > +++ b/drivers/net/ethernet/freescale/enetc/enetc.c
> > @@ -2,6 +2,7 @@
> >  /* Copyright 2017-2019 NXP */
> >
> >  #include "enetc.h"
> > +#include <linux/clk.h>
> >  #include <linux/bpf_trace.h>
> >  #include <linux/tcp.h>
> >  #include <linux/udp.h>
> > @@ -21,7 +22,7 @@ void enetc_port_mac_wr(struct enetc_si *si, u32 reg,
> > u32 val)
> >  {
> >  	enetc_port_wr(&si->hw, reg, val);
> >  	if (si->hw_features & ENETC_SI_F_QBU)
> > -		enetc_port_wr(&si->hw, reg + ENETC_PMAC_OFFSET, val);
> > +		enetc_port_wr(&si->hw, reg + si->pmac_offset, val);
> >  }
> >  EXPORT_SYMBOL_GPL(enetc_port_mac_wr);
> >
> > @@ -700,8 +701,10 @@ static void enetc_rx_dim_work(struct work_struct
> > *w)
> >  		net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
> >  	struct enetc_int_vector	*v =
> >  		container_of(dim, struct enetc_int_vector, rx_dim);
> > +	struct enetc_ndev_priv *priv = netdev_priv(v->rx_ring.ndev);
> > +	u64 clk_freq = priv->si->clk_freq;
> 
> Not happy to access 'priv' struct here and redirect again to the 'si' struct just
> to get some init time parameter value like 'clk_freq'. enetc_rx_dim_work()
> should
> be fast and have a small footprint. Messing up caches by accessing these 2 extra
> structures periodically doesn't help. Pls move 'clk_freq' to 'priv' to get rid of one
> indirection at least (I don't have a better idea now).

Not sure if this helps much, but I could move the clk_freq to priv.

> 
> >
> > -	v->rx_ictt = enetc_usecs_to_cycles(moder.usec);
> > +	v->rx_ictt = enetc_usecs_to_cycles(moder.usec, clk_freq);
> >  	dim->state = DIM_START_MEASURE;
> >  }
> >
> > @@ -1721,14 +1724,25 @@ void enetc_get_si_caps(struct enetc_si *si)
> >  	struct enetc_hw *hw = &si->hw;
> >  	u32 val;
> >
> > +	if (is_enetc_rev1(si))
> > +		si->clk_freq = ENETC_CLK;
> > +	else
> > +		si->clk_freq = ENETC_CLK_333M;
> > +
> 
> [...]
> 
> > @@ -2079,10 +2096,11 @@ void enetc_init_si_rings_params(struct
> > enetc_ndev_priv *priv)
> >  	 * TODO: Make # of TX rings run-time configurable
> >  	 */
> >  	priv->num_rx_rings = min_t(int, cpus, si->num_rx_rings);
> > +	priv->num_rx_rings = min_t(int, cpus, si->num_rx_rings);
> 
> Duplicated statement.

Oh, I was not aware of this, thanks

> [...]