Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 20 Apr 2021 11:34:20 -0700
From:   Jakub Kicinski <kuba@kernel.org>
To:     Hayes Wang <hayeswang@realtek.com>
Cc:     "davem@davemloft.net" <davem@davemloft.net>,
        "netdev@vger.kernel.org" <netdev@vger.kernel.org>,
        nic_swsd <nic_swsd@realtek.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "linux-usb@vger.kernel.org" <linux-usb@vger.kernel.org>
Subject: Re: [PATCH net-next 4/6] r8152: support new chips
Message-ID: <20210420113420.79d7c65a@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
In-Reply-To: <0de9842749db4718b8f45a0f2fff7967@realtek.com>
References: <1394712342-15778-350-Taiwan-albertk@realtek.com>
        <1394712342-15778-354-Taiwan-albertk@realtek.com>
        <20210416145017.1946f013@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
        <0de9842749db4718b8f45a0f2fff7967@realtek.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Tue, 20 Apr 2021 07:00:39 +0000 Hayes Wang wrote:
> > > @@ -6878,7 +8942,11 @@ static int rtl8152_probe(struct usb_interface *intf,  
> > >  	set_ethernet_addr(tp);
> > >
> > >  	usb_set_intfdata(intf, tp);
> > > -	netif_napi_add(netdev, &tp->napi, r8152_poll, RTL8152_NAPI_WEIGHT);
> > > +
> > > +	if (tp->support_2500full)
> > > +		netif_napi_add(netdev, &tp->napi, r8152_poll, 256);  
> > 
> > why 256? We have 100G+ drivers all using 64 what's special here?
> >   
> > > +	else
> > > +		netif_napi_add(netdev, &tp->napi, r8152_poll, 64);  
> 
> We test 2.5G Ethernet on some embedded platform.
> And we find 64 is not large enough, and the performance
> couldn't reach 2.5 G bits/s.

Did you manage to identify what the cause is?

NAPI will keep calling your driver if the budget was exhausted, the
only difference between 64 and 256 should be the setup cost of the
driver's internal loop. And perhaps more frequent GRO flush - what's
the CONFIG_HZ set to?