[Linux-stm32] [PATCH net-next 4/4] net: stmmac: Always use TX coalesce timer value when rescheduling

[Jakub Kicinski]

On Mon, 16 Dec 2019 09:20:53 +0000, Jose Abreu wrote:
> > > diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> > > index f61780ae30ac..726a17d9cc35 100644
> > > --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> > > +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> > > @@ -1975,7 +1975,7 @@ static int stmmac_tx_clean(struct stmmac_priv *priv, int budget, u32 queue)
> > >  
> > >  	/* We still have pending packets, let's call for a new scheduling */
> > >  	if (tx_q->dirty_tx != tx_q->cur_tx)
> > > -		mod_timer(&tx_q->txtimer, STMMAC_COAL_TIMER(10));
> > > +		mod_timer(&tx_q->txtimer, STMMAC_COAL_TIMER(priv->tx_coal_timer));  
> > 
> > I think intent of this code is to re-check the ring soon. The same
> > value of 10 is used in stmmac_tx_timer() for quick re-check.
> > 
> > tx_coal_timer defaults to 1000, so it's quite a jump from 10 to 1000.
> > 
> > I think the commit message leaves too much unsaid.
> > 
> > Also if you want to change to the ethtool timeout value, could you move 
> > stmmac_tx_timer_arm() and reuse that helper?  
> 
> Yeah, it's a quick re-check but 10us can be too low on some speeds and 
> leads to CPU useless-looping. The intent is to let this always be 
> configurable by user.

Okay, please do mention the bump from 10us to the default of 1ms in the
commit message, though.