Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753929AbYLCSLS (ORCPT ); Wed, 3 Dec 2008 13:11:18 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751305AbYLCSLI (ORCPT ); Wed, 3 Dec 2008 13:11:08 -0500 Received: from tomts40.bellnexxia.net ([209.226.175.97]:55836 "EHLO tomts40-srv.bellnexxia.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751268AbYLCSLH (ORCPT ); Wed, 3 Dec 2008 13:11:07 -0500 X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: Av0EAL9aNklMROB9/2dsb2JhbACBbdE+gn8 Date: Wed, 3 Dec 2008 13:11:05 -0500 From: Mathieu Desnoyers To: Neil Horman Cc: ltt-dev@lists.casi.polymtl.ca, linux-kernel@vger.kernel.org Subject: Re: [ltt-dev] [PATCH] add tracepoints to track change in napi states on network receive Message-ID: <20081203181105.GA6794@Krystal> References: <20081202195129.GA30042@hmsreliant.think-freely.org> <20081202200150.GA27282@Krystal> <20081203160134.GA6549@hmsreliant.think-freely.org> <20081203172507.GA4471@Krystal> <20081203174500.GB6549@hmsreliant.think-freely.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline In-Reply-To: <20081203174500.GB6549@hmsreliant.think-freely.org> X-Editor: vi X-Info: http://krystal.dyndns.org:8080 X-Operating-System: Linux/2.6.21.3-grsec (i686) X-Uptime: 13:10:10 up 16 days, 18:50, 2 users, load average: 2.86, 1.05, 0.70 User-Agent: Mutt/1.5.16 (2007-06-11) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5673 Lines: 195 * Neil Horman (nhorman@tuxdriver.com) wrote: > On Wed, Dec 03, 2008 at 12:25:07PM -0500, Mathieu Desnoyers wrote: > > Almost there :) > > > > There is a extra whitespace above, > > > > and also "schedule napi_struct" should probably be changed into > > napi_struct only, this removes the whitespace and duplicated information > > (I guess we know that given we are in event net_napi_schedule, this will > > have something to do with schedule...). Same applies to trace_mark_tp > > below. > > > > Thanks, > > > > Mathieu > > > Sure, here you go, new patch > > Patch to trace napi states for various net devices and their napi instances. > Traces scheduling, servicing and completion of a napi poll request. > > Regards > Neil > > Signed-off-by: Neil Horman > Actually I notice that you used the old "DEFINE_TRACE" tracepoint API. I'll update your patch to the new DECLARE/DEFINE_TRACE tracepoint API. Mathieu > > include/linux/netdevice.h | 5 +++++ > include/trace/netdevice.h | 30 ++++++++++++++++++++++++++++++ > ltt/probes/net-trace.c | 22 ++++++++++++++++++++++ > net/core/dev.c | 7 ++++++- > 4 files changed, 63 insertions(+), 1 deletion(-) > > > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h > index 488c56e..7da71eb 100644 > --- a/include/linux/netdevice.h > +++ b/include/linux/netdevice.h > @@ -43,6 +43,9 @@ > > #include > > +struct napi_struct; > +#include > + > struct vlan_group; > struct ethtool_ops; > struct netpoll_info; > @@ -386,6 +389,7 @@ static inline void napi_complete(struct napi_struct *n) > local_irq_save(flags); > __napi_complete(n); > local_irq_restore(flags); > + trace_napi_complete(n); > } > > /** > @@ -1726,6 +1730,7 @@ static inline int skb_bond_should_drop(struct sk_buff *skb) > return 0; > } > > + > #endif /* __KERNEL__ */ > > #endif /* _LINUX_DEV_H */ > diff --git a/include/trace/netdevice.h b/include/trace/netdevice.h > new file mode 100644 > index 0000000..ba48f6c > --- /dev/null > +++ b/include/trace/netdevice.h > @@ -0,0 +1,30 @@ > +#ifndef _INCLUDE_NETDEVICE_H_ > +#define _INCLUDE_NETDEVICE_H_ > + > +#include > + > +/* Tracepoints */ > + > +/* > + * Note these first 2 traces are actually in __napi_schedule and net_rx_action > + * respectively. The former is in __napi_schedule because it uses at-most-once > + * logic and placing it in the calling routine (napi_schedule) would produce > + * countless trace events that were effectively no-ops. napi_poll is > + * implemented in net_rx_action, because thats where we do our polling on > + * devices. The last trace point is in napi_complete, right where you would > + * think it would be. > + */ > +DEFINE_TRACE(napi_schedule, > + TPPROTO(struct napi_struct *n), > + TPARGS(n)); > + > +DEFINE_TRACE(napi_poll, > + TPPROTO(struct napi_struct *n), > + TPARGS(n)); > + > +DEFINE_TRACE(napi_complete, > + TPPROTO(struct napi_struct *n), > + TPARGS(n)); > + > +#endif > + > diff --git a/ltt/probes/net-trace.c b/ltt/probes/net-trace.c > index bac2b21..e887e70 100644 > --- a/ltt/probes/net-trace.c > +++ b/ltt/probes/net-trace.c > @@ -9,6 +9,7 @@ > #include > #include > #include > +#include > > void probe_net_dev_xmit(struct sk_buff *skb) > { > @@ -104,6 +105,27 @@ void probe_socket_call(int call, unsigned long a0) > "call %d a0 %lu", call, a0); > } > > +void probe_napi_schedule(struct napi_struct *n) > +{ > + trace_mark_tp(net_napi_schedule, napi_schedule, probe_napi_schedule, > + "napi_struct %p name %s", > + n, n->dev->name); > +} > + > +void probe_napi_poll(struct napi_struct *n) > +{ > + trace_mark_tp(net_napi_poll, napi_poll, probe_napi_poll, > + "napi_struct %p name %s", > + n, n->dev->name); > +} > + > +void probe_napi_complete(struct napi_struct *n) > +{ > + trace_mark_tp(net_napi_complete, napi_complete, probe_napi_complete, > + "napi_struct %p name %s", > + n, n->dev->name); > +} > + > MODULE_LICENSE("GPL"); > MODULE_AUTHOR("Mathieu Desnoyers"); > MODULE_DESCRIPTION("Net Tracepoint Probes"); > diff --git a/net/core/dev.c b/net/core/dev.c > index 836fe6e..0c55c0f 100644 > --- a/net/core/dev.c > +++ b/net/core/dev.c > @@ -128,6 +128,7 @@ > #include > #include > #include > +#include > > #include "net-sysfs.h" > > @@ -2329,6 +2330,8 @@ void __napi_schedule(struct napi_struct *n) > { > unsigned long flags; > > + trace_napi_poll(n); > + > local_irq_save(flags); > list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list); > __raise_softirq_irqoff(NET_RX_SOFTIRQ); > @@ -2380,8 +2383,10 @@ static void net_rx_action(struct softirq_action *h) > * accidently calling ->poll() when NAPI is not scheduled. > */ > work = 0; > - if (test_bit(NAPI_STATE_SCHED, &n->state)) > + if (test_bit(NAPI_STATE_SCHED, &n->state)) { > + trace_napi_poll(n); > work = n->poll(n, weight); > + } > > WARN_ON_ONCE(work > weight); > > -- > /**************************************************** > * Neil Horman > * Software Engineer, Red Hat > ****************************************************/ > -- Mathieu Desnoyers OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/