From mboxrd@z Thu Jan 1 00:00:00 1970 From: Peppe CAVALLARO Subject: Re: [PATCH (sh-2.6) 1/4] clksource: Generic timer infrastructure Date: Fri, 4 Mar 2011 07:53:50 +0100 Message-ID: <4D708C7E.8010204@st.com> References: <1298369864-24429-1-git-send-email-peppe.cavallaro@st.com> <201103030945.07486.arnd@arndb.de> <4D6F6C96.1010303@st.com> <201103031455.25089.arnd@arndb.de> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7BIT Cc: Thomas Gleixner , Stuart MENEFY , "linux-sh@vger.kernel.org" , "netdev@vger.kernel.org" , John Stultz , "linux-kernel@vger.kernel.org" , "linux@arm.linux.org.uk" To: Arnd Bergmann Return-path: In-Reply-To: <201103031455.25089.arnd@arndb.de> Sender: linux-sh-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On 3/3/2011 2:55 PM, Arnd Bergmann wrote: > > On Thursday 03 March 2011, Peppe CAVALLARO wrote: > > This logic is already in the driver, indeed. > > What I've seen on our embedded systems is that the > > cost of RX interrupts is very hight and NAPI partially helps. > > Typically, in an IP-STB, I receive a burst of UDP pkt > > and this means that many interrupts occur (~99% of CPU > > usage on slow platforms). > > With the ext timer I was able to reduce the CPU usage in > > these kind of scenarios to ~50%. > > I don't understand. Shouldn't the interrupts be stopped as long > as the system is busy? This sounds like a bug in your NAPI > handling, or maybe you just need to use a lower NAPI_WEIGHT > so you stay in polling mode longer. > Hi Arnd, yes you are right, in fact, I do not receive an interrupt for each frame received but the average of the frames I found in the ring is very low. At any rate, let me do some tests so i'll come back with some numbers. In the meantime, I'm looking at the napi support inside the driver and experiment with lower NAPI_WEIGHT (default is 64). Peppe > > > Arnd >