Re: [ltt-dev] [PATCH] add tracepoints to track change in napi states on network receive

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Neil Horman (nhorman@tuxdriver.com) wrote:
> On Tue, Dec 02, 2008 at 03:01:50PM -0500, Mathieu Desnoyers wrote:
> > * Neil Horman (nhorman@tuxdriver.com) wrote:
> > > Hey there-
> > > 	I thought it would be handy to track the napi receive state of different
> > > network devices using ltt.  This patch adds trace points to indicate when a napi
> > > instance has been scheduled, when its serviced and when it completes.
> > > 
> > 
> > Sounds great, see comments below,
> <snip>
> 
> Hey,
> 	New patch, with previous comments taken into account.
> 
> Patch to track the napi receive stat of different network devices using ltt.
> Adds trace points to indicate when a napi instance has been scheduled, when its
> serviced and when it completes
> 
> Regards
> Neil
> 
> Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
> 
> 
>  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 <net/net_namespace.h>
>  
> +struct napi_struct;
> +#include <trace/netdevice.h>
> +
>  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 <linux/netdevice.h>
> +
> +/* 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..bd4e621 100644
> --- a/ltt/probes/net-trace.c
> +++ b/ltt/probes/net-trace.c
> @@ -9,6 +9,7 @@
>  #include <trace/ipv4.h>
>  #include <trace/ipv6.h>
>  #include <trace/socket.h>
> +#include <trace/netdevice.h>
>  
>  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,
> +		"schedule napi_struct %p name  %s",

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

> +		n, n->dev->name);
> +}
> +
> +void probe_napi_poll(struct napi_struct *n)
> +{
> +	trace_mark_tp(net_napi_poll, napi_poll, probe_napi_poll,
> +		"service 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,
> +		"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 <linux/jhash.h>
>  #include <linux/random.h>
>  #include <trace/net.h>
> +#include <trace/netdevice.h>
>  
>  #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 <nhorman@tuxdriver.com>
>  * Software Engineer, Red Hat
>  ****************************************************/
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68