Re: [PATCH v4] net: mvneta: fix calls to page_pool_get_stats

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1606 bytes --]

[...]
> > 
> > do we need to check pp->is_stopped here? (we already check if page_pool
> > pointer is NULL in mvneta_ethtool_pp_stats).
> > Moreover in mvneta_ethtool_get_sset_count() and in mvneta_ethtool_get_strings()
> > we just check pp->bm_priv pointer. Are the stats disaligned in this case?
> 
> Hi Lorenzo,
> 
> so the buffer manager (bm) does not support the page pool.
> If this mode is used we can skip any page pool references.
> 
> The question is do we end up with a race condition when we skip the is_stopped check
> as the variable is set to true just before the page pools are
> deallocated on suspend or interface stop calls.

Do we really have a race here? e.g. both ndo_stop and ethtool_get_stats() run under rtnl.
Moreover it seems is_stopped is not set for armada-3700 devices (e.g. my
espressobin). Do we have the issue for this kind of devices?

Regards,
Lorenzo

> 
> Best
> Sven
> 
> > 
> > > +		mvneta_ethtool_pp_stats(pp, data);
> > >  }
> > >  
> > >  static int mvneta_ethtool_get_sset_count(struct net_device *dev, int sset)
> > >  {
> > > -	if (sset == ETH_SS_STATS)
> > > -		return ARRAY_SIZE(mvneta_statistics) +
> > > -		       page_pool_ethtool_stats_get_count();
> > > +	if (sset == ETH_SS_STATS) {
> > > +		int count = ARRAY_SIZE(mvneta_statistics);
> > > +		struct mvneta_port *pp = netdev_priv(dev);
> > > +
> > > +		if (!pp->bm_priv)
> > > +			count += page_pool_ethtool_stats_get_count();
> > > +
> > > +		return count;
> > > +	}
> > >  
> > >  	return -EOPNOTSUPP;
> > >  }
> > > -- 
> > > 2.42.0
> > > 
> 
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]