Tracepoint: Make skb tracepoint use TRACE_EVENT macro
PATCH 1/1 Tracepoint: Make skb tracepoint use TRACE_EVENT macro
TRACE_EVENT is more unify and generic for define a tracepoint. It also add ftrace support for this tracepoint.
--- include/trace/skb.h | 4 +-- include/trace/skb_event_types.h | 38 +++++++++++++++++++++++++++++++++++++ include/trace/trace_event_types.h | 1 + include/trace/trace_events.h | 1 + 4 files changed, 41 insertions(+), 3 deletions(-) create mode 100644 include/trace/skb_event_types.h
include/trace/skb.h +1 -3
@@ -4,8 +4,6 @@#include <linux/skbuff.h> #include <linux/tracepoint.h> -DECLARE_TRACE (kfree_skb, - TP_PROTO(struct sk_buff *skb, void *location), -TP_ARGS (skb, location)); +#include <trace/skb_event_types.h> #endif
include/trace/skb_event_types.h new file +38 -0
@@ -0,0 +1,38 @@ + +/* use <trace/skb.h> instead */ +#ifndef TRACE_EVENT +# error Do not include this file directly. +# error Unless you know what you are doing. +#endif + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM skb + +/* +* Tracepoint for free an sk_buff: +*/ +TRACE_EVENT (kfree_skb, + + TP_PROTO(struct sk_buff *skb, void *location), + +TP_ARGS (skb, location), + +TP_STRUCT__entry ( + __field( void *, skbaddr ) + __field( unsigned short, protocol ) + __field( void *, location ) + ), + +TP_fast_assign ( + __entry->skbaddr = skb; + if (skb) { + __entry->protocol = ntohs(skb->protocol); + } + __entry->location = location; + ), + +TP_printk ("skbaddr=%p protocol=%u location=%p" , + __entry->skbaddr, __entry->protocol, __entry->location) +); + +#undef TRACE_SYSTEM
include/trace/trace_event_types.h +1 -0
@@ -3,3 +3,4 @@#include <trace/sched_event_types.h> #include <trace/irq_event_types.h> #include <trace/lockdep_event_types.h> +#include <trace/skb_event_types.h>
include/trace/trace_events.h +1 -0
@@ -3,3 +3,4 @@#include <trace/sched.h> #include <trace/irq.h> #include <trace/lockdep.h> +#include <trace/skb.h>
Signature · 1 line
1.5.5.3
Quoted 11 lines · Zhaolei
TRACE_EVENTis more unify and generic for define a tracepoint. It also add ftrace support for this tracepoint.Signed-off-by Zhao Lei <[email protected]>--- include/trace/skb.h | 4 +-- include/trace/skb_event_types.h | 38 +++++++++++++++++++++++++++++++++++++ include/trace/trace_event_types.h | 1 + include/trace/trace_events.h | 1 + 4 files changed, 41 insertions(+), 3 deletions(-) create mode 100644 include/trace/skb_event_types.h
I've Cc:-ed more networking folks, the acks of them are needed.
David, Neil: the point of this patch is to make the kfree_skb() tracepoint show up in the general event tracing framework that got introduced upstream in this merge window.
One advantage is that this event will/can show up under the function graph tracer and other generic tracers as well:
0) | handle_IRQ_event() {
0) | /* irq_handler_entry: irq=48 handler=eth0 */
0) | e1000_intr() {
0) 0.926 us | __napi_schedule();
0) 3.888 us | }
0) | /* irq_handler_exit: irq=48 return=handled */
0) 0.655 us | runqueue_is_locked();
0) | __wake_up() {
0) 0.831 us | _spin_lock_irqsave();This tracepoint [when enabled] would show up as:
skbaddr=%p protocol=%u location=%p
Another advantage is that its output can also be high-bandwidth per-cpu zero-copy traced via a no-vsprintf direct tracing path and via splice(), using the /debug/tracing/per_cpu/cpu*/trace_pipe_raw mechanism. (this is what the TP_fast_assign() portion of the TRACE_EVENT() macro achieves. )
Yet another advantage is that such tracepoints also show up under /debug/tracing/events/. For example the existing IRQ-handler-exit tracepoint shows up as:
aldebaran:/debug/tracing/events/irq/irq_handler_exit> ls -l total 0
-rw-r--r-- 1 root root 0 2009-04-08 08:56 enable -rw-r--r-- 1 root root 0 2009-04-08 08:56 filter -r--r--r-- 1 root root 0 2009-04-08 08:56 format -r--r--r-- 1 root root 0 2009-04-08 08:56 id
It can be toggled on/off, its format is exposed in a user-space-tool parse-able way:
aldebaran:/debug/tracing/events/irq/irq_handler_exit> cat format
name: irq_handler_exit
ID: 27
format:
field:unsigned char common_type; offset:0; size:1; field:unsigned char common_flags; offset:1; size:1; field:unsigned char common_preempt_count; offset:2; size:1; field:int common_pid; offset:4; size:4; field:int common_tgid; offset:8; size:4; field:int irq; offset:12; size:4; field:int ret; offset:16; size:4;
print fmt: "irq=%d return=%s", __entry->irq, __entry->ret ? "handled" : "unhandled"
The 'filter' file can be used to dynamically add optional filter expressions, which are evaluated by the kernel at the tracepoint to filter data. For example:
echo 'irq == 1' > filter
Restricts this IRQ tracepoint to keyboard interrupts only. Doing:
echo '|| irq == 0' >> filter
cat filter
irq == 1
|| irq == 0
extends the filter expression to also include timer interrupts. Etc.
All the fields enumerated in the 'format' descriptor can be used in the filter language. (there's also per subsystem filter expression to simplify the handling of a group of tracepoints, and other details.)
There's no fastpath overhead difference to old-style tracepoints in the disabled case, except some additional off-site section size increase. External tools can still hook in as well.
Ingo
Quoted 95 lines · Ingo Molnar
Quoted 11 lines · Zhaolei
TRACE_EVENTis more unify and generic for define a tracepoint. It also add ftrace support for this tracepoint.Signed-off-by Zhao Lei <[email protected]>--- include/trace/skb.h | 4 +-- include/trace/skb_event_types.h | 38 +++++++++++++++++++++++++++++++++++++ include/trace/trace_event_types.h | 1 + include/trace/trace_events.h | 1 + 4 files changed, 41 insertions(+), 3 deletions(-) create mode 100644 include/trace/skb_event_types.hI've Cc:-ed more networking folks, the acks of them are needed.
David, Neil: the point of this patch is to make the
kfree_skb()tracepoint show up in the general event tracing framework that got introduced upstream in this merge window.One advantage is that this event will/can show up under the function graph tracer and other generic tracers as well:
0) | handle_IRQ_event() { 0) | /* irq_handler_entry: irq=48 handler=eth0 */ 0) | e1000_intr() { 0) 0.926 us | __napi_schedule(); 0) 3.888 us | } 0) | /* irq_handler_exit: irq=48 return=handled */ 0) 0.655 us | runqueue_is_locked(); 0) | __wake_up() { 0) 0.831 us | _spin_lock_irqsave();This tracepoint [when enabled] would show up as:
skbaddr=%p protocol=%u location=%p
Another advantage is that its output can also be high-bandwidth per-cpu zero-copy traced via a no-vsprintf direct tracing path and via
splice(), using the /debug/tracing/per_cpu/cpu*/trace_pipe_raw mechanism. (this is what theTP_fast_assign()portion of theTRACE_EVENT()macro achieves. )Yet another advantage is that such tracepoints also show up under /debug/tracing/events/. For example the existing IRQ-handler-exit tracepoint shows up as:
aldebaran:/debug/tracing/events/irq/irq_handler_exit> ls -l total 0
-rw-r--r-- 1 root root 0 2009-04-08 08:56 enable -rw-r--r-- 1 root root 0 2009-04-08 08:56 filter -r--r--r-- 1 root root 0 2009-04-08 08:56 format -r--r--r-- 1 root root 0 2009-04-08 08:56 idIt can be toggled on/off, its format is exposed in a user-space-tool parse-able way:
aldebaran:/debug/tracing/events/irq/irq_handler_exit> cat format
name:
irq_handler_exit
ID: 27
format:field:unsigned char common_type; offset:0; size:1; field:unsigned char common_flags; offset:1; size:1; field:unsigned char common_preempt_count; offset:2; size:1; field:int common_pid; offset:4; size:4; field:int common_tgid; offset:8; size:4; field:int irq; offset:12; size:4; field:int ret; offset:16; size:4;print fmt: "irq=%d return=%s",
__entry->irq,__entry->ret? "handled" : "unhandled"The 'filter' file can be used to dynamically add optional filter expressions, which are evaluated by the kernel at the tracepoint to filter data. For example:
echo 'irq == 1' > filterRestricts this IRQ tracepoint to keyboard interrupts only. Doing:
echo '|| irq == 0' >> filter cat filter irq == 1 || irq == 0extends the filter expression to also include timer interrupts. Etc.
All the fields enumerated in the 'format' descriptor can be used in the filter language. (there's also per subsystem filter expression to simplify the handling of a group of tracepoints, and other details.)
Quoted 6 lines
There's no fastpath overhead difference to old-style tracepoints in the disabled case, except some additional off-site section size increase. External tools can still hook in as well.
Ingo
I think this seems like a reasonable idea, as long as the sites that register a hook for the tracepoint don't need to be updated (and it looks like they dont). I do have one nit though: include/trace/skb.h has been emptied and now includes include/trace/skb_event_types.h instead. For the sake of neatness I'd say we should either not make that change, and just dump the contents of skb_event_types.h into trace/skb.h, or alternatively, remove trace/skb.h entirely, and have files that need trace/skb.h include skb_event_types.h instead. Do that and it has my ACK.
Thanks
Neil
Steven Rostedt
maintainerQuoted 9 lines · Neil Horman
I think this seems like a reasonable idea, as long as the sites that register a hook for the tracepoint don't need to be updated (and it looks like they dont). I do have one nit though:
include/trace/skb.hhas been emptied and now includesinclude/trace/skb_event_types.hinstead. For the sake of neatness I'd say we should either not make that change, and just dump the contents ofskb_event_types.hintotrace/skb.h, or alternatively, removetrace/skb.hentirely, and have files that needtrace/skb.hincludeskb_event_types.hinstead. Do that and it has my ACK.
Actually that trace/skb.h change is needed. It is not emptied, it still has:
#include <linux/skbuff.h>
#include <linux/tracepoint.h>
before the include of include/trace/skb_event_types.h
For ftrace to do its magic with the TRACE_EVENT macro, the file that contains the TRACE_EVENT can not include tracepoint.h. ftrace runs the skb_event_types.h through a series of redefines of the TRACE_EVENT to automate the creation of the directory and files in debugfs/tracing/events/*.
We've already changed most of the files in include/trace/ to this format.
I hope we can still have your Ack without making your requested update.
Thanks,
-- Steve
Neil Horman
Acked-byQuoted 26 lines · Steven Rostedt
Quoted 9 lines · Neil Horman
I think this seems like a reasonable idea, as long as the sites that register a hook for the tracepoint don't need to be updated (and it looks like they dont). I do have one nit though:
include/trace/skb.hhas been emptied and now includesinclude/trace/skb_event_types.hinstead. For the sake of neatness I'd say we should either not make that change, and just dump the contents ofskb_event_types.hintotrace/skb.h, or alternatively, removetrace/skb.hentirely, and have files that needtrace/skb.hincludeskb_event_types.hinstead. Do that and it has my ACK.Actually that
trace/skb.hchange is needed. It is not emptied, it still has:#include <
linux/skbuff.h>
#include <linux/tracepoint.h>before the include of
include/trace/skb_event_types.hFor ftrace to do its magic with the
TRACE_EVENTmacro, the file that contains theTRACE_EVENTcan not includetracepoint.h. ftrace runs theskb_event_types.hthrough a series of redefines of theTRACE_EVENTto automate the creation of the directory and files in debugfs/tracing/events/*.
Quoted 9 lines
We've already changed most of the files in
include/trace/to this format.I hope we can still have your Ack without making your requested update.
Thanks,
-- Steve
That strikes me as overly complex and fragile. One would think that ftrace could include a first pass to scrub out that include prior to processing. Having a utility reliant on the format of kernel header files, especially the placement of includes is begging for somthing to break.
I'll ack it since the change I'm asking for is just cosmetic, and since any reversions in this area are going to break ftrace in userspace rather than oops the kernel, but I would really like to see ftrace hardened in such a way that we don't need to be extra careful of include formatting when adding additional tracepoints.
[tip:tracing/core] tracing, net, skb tracepoint: make skb tracepoint use the TRACE_EVENT() macro
In reply to ZhaoleiCommit-ID: 5cb3d1d9d34ac04bcaa2034139345b2a5fea54c1 Gitweb: http://git.kernel.org/tip/5cb3d1d9d34ac04bcaa2034139345b2a5fea54c1 Author: Zhaolei <[email protected]> AuthorDate: Thu, 9 Apr 2009 14:08:18 +0800 Committer: Ingo Molnar <[email protected]> CommitDate: Fri, 10 Apr 2009 12:57:55 +0200
tracing, net, skb tracepoint: make skb tracepoint use the TRACE_EVENT() macro
TRACE_EVENT is a more generic way to define a tracepoint. Doing so adds these new capabilities to this tracepoint:
- zero-copy and per-cpu splice() tracing
- binary tracing without printf overhead
- structured logging records exposed under /debug/tracing/events
- trace events embedded in function tracer output and other plugins
- user-defined, per tracepoint filter expressions
LKML-Reference: <[email protected]>
--- include/trace/skb.h | 4 +-- include/trace/skb_event_types.h | 38 +++++++++++++++++++++++++++++++++++++ include/trace/trace_event_types.h | 1 + include/trace/trace_events.h | 1 + 4 files changed, 41 insertions(+), 3 deletions(-)
include/trace/skb.h +1 -3
@@ -4,8 +4,6 @@#include <linux/skbuff.h> #include <linux/tracepoint.h> -DECLARE_TRACE (kfree_skb, - TP_PROTO(struct sk_buff *skb, void *location), -TP_ARGS (skb, location)); +#include <trace/skb_event_types.h> #endif
include/trace/skb_event_types.h new file +38 -0
@@ -0,0 +1,38 @@ + +/* use <trace/skb.h> instead */ +#ifndef TRACE_EVENT +# error Do not include this file directly. +# error Unless you know what you are doing. +#endif + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM skb + +/* +* Tracepoint for free an sk_buff: +*/ +TRACE_EVENT (kfree_skb, + + TP_PROTO(struct sk_buff *skb, void *location), + +TP_ARGS (skb, location), + +TP_STRUCT__entry ( + __field( void *, skbaddr ) + __field( unsigned short, protocol ) + __field( void *, location ) + ), + +TP_fast_assign ( + __entry->skbaddr = skb; + if (skb) { + __entry->protocol = ntohs(skb->protocol); + } + __entry->location = location; + ), + +TP_printk ("skbaddr=%p protocol=%u location=%p" , + __entry->skbaddr, __entry->protocol, __entry->location) +); + +#undef TRACE_SYSTEM
include/trace/trace_event_types.h +1 -0
@@ -3,3 +3,4 @@#include <trace/sched_event_types.h> #include <trace/irq_event_types.h> #include <trace/lockdep_event_types.h> +#include <trace/skb_event_types.h>
include/trace/trace_events.h +1 -0
@@ -3,3 +3,4 @@#include <trace/sched.h> #include <trace/irq.h> #include <trace/lockdep.h> +#include <trace/skb.h>