The ring_buffer_discard_commit makes better usage of the ring_buffer
when an event has been discarded. It tries to remove it completely if
possible.

This patch converts the trace event filtering to use
ring_buffer_discard_commit instead of the ring_buffer_event_discard.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Frederic Weisbecker suggested that the trace_special event shouldn't be
filterable; this patch adds a TRACE_EVENT_FORMAT_NOFILTER event macro
that allows an event format to be exported without having a filter
attached, and removes filtering from the trace_special event.
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This patch adds run-time field descriptions to all the event formats
exported using TRACE_EVENT_FORMAT.  It also hooks up all the tracers
that use them (i.e. the tracers in the 'ftrace subsystem') so they can
also have their output filtered by the event-filtering mechanism.

When I was testing this, there were a couple of things that fooled me
into thinking the filters weren't working, when actually they were -
I'll mention them here so others don't make the same mistakes (and file
bug reports. ;-)

One is that some of the tracers trace multiple events e.g. the
sched_switch tracer uses the context_switch and wakeup events, and if
you don't set filters on all of the traced events, the unfiltered output
from the events without filters on them can make it look like the
filtering as a whole isn't working properly, when actually it is doing
what it was asked to do - it just wasn't asked to do the right thing.

The other is that for the really high-volume tracers e.g. the function
tracer, the volume of filtered events can be so high that it pushes the
unfiltered events out of the ring buffer before they can be read so e.g.
cat'ing the trace file repeatedly shows either no output, or once in
awhile some output but that isn't there the next time you read the
trace, which isn't what you normally expect when reading the trace file.
If you read from the trace_pipe file though, you can catch them before
they disappear.

Changes from v1:

As suggested by Frederic Weisbecker:

- get rid of externs in functions
- added unlikely() to filter_check_discard()
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: refactor code for future changes

Current kmemtrace.h is used both as header file of kmemtrace and kmem's
tracepoints definition.

Tracepoints' definition file may be used by other code, and should only have
definition of tracepoint.

We can separate include/trace/kmemtrace.h into 2 files:

  include/linux/kmemtrace.h: header file for kmemtrace
  include/trace/kmem.h:      definition of kmem tracepoints
Signed-off-by: NZhao Lei <zhaolei@cn.fujitsu.com>
Acked-by: NEduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Acked-by: NPekka Enberg <penberg@cs.helsinki.fi>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <49DEE68A.5040902@cn.fujitsu.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: cleanup

Most of the tracing files creation follow the same pattern:

ret = debugfs_create_file(...)
if (!ret)
	pr_warning("Couldn't create ... entry\n")

Unify it!
Reported-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1238109938-11840-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Impact: fix time output bug in 32bits system

ns2usecs() returns 'long', it's incorrect.

(In i386)
...
          <idle>-0     [000]   521.442100: _spin_lock <-tick_do_update_jiffies64
          <idle>-0     [000]   521.442101: do_timer <-tick_do_update_jiffies64
          <idle>-0     [000]   521.442102: update_wall_time <-do_timer
          <idle>-0     [000]   521.442102: update_xtime_cache <-update_wall_time
....
(It always print the time less than 2200 seconds besides ...)
Because 'long' is 32bits in i386. ( (1<<31) useconds is about 2200 seconds)

...
          <idle>-0     [001] 4154502640.134759: rcu_bh_qsctr_inc <-__do_softirq
          <idle>-0     [001] 4154502640.134760: _local_bh_enable <-__do_softirq
          <idle>-0     [001] 4154502640.134761: idle_cpu <-irq_exit
...
(very large value)
Because 'long' is a signed type and it is 32bits in i386.

Changes in v2:
return 'unsigned long long' instead of 'cycle_t'
Signed-off-by: NLai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <49D05D10.4030009@cn.fujitsu.com>
Reported-by: NLi Zefan <lizf@cn.fujitsu.com>
Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

kmemtrace now uses tracepoints instead of markers. We no longer need to
use format specifiers to pass arguments.
Signed-off-by: NEduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
[ folded: Use the new TP_PROTO and TP_ARGS to fix the build.     ]
[ folded: fix build when CONFIG_KMEMTRACE is disabled.           ]
[ folded: define tracepoints when CONFIG_TRACEPOINTS is enabled. ]
Signed-off-by: NPekka Enberg <penberg@cs.helsinki.fi>
LKML-Reference: <ae61c0f37156db8ec8dc0d5778018edde60a92e3.1237813499.git.eduard.munteanu@linux360.ro>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

graph time is the time that a function is executing another function.
Thus if function A calls B, if graph-time is set, then the time for
A includes B. This is the default behavior. But if graph-time is off,
then the time spent executing B is subtracted from A.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

If the function graph trace is enabled, the function profiler will
use it to take the timing of the functions.

 cat /debug/tracing/trace_stat/functions

  Function                               Hit    Time
  --------                               ---    ----
  mwait_idle                             127    183028.4 us
  schedule                                26    151997.7 us
  __schedule                              31    151975.1 us
  sys_wait4                                2    74080.53 us
  do_wait                                  2    74077.80 us
  sys_newlstat                           138    39929.16 us
  do_path_lookup                         179    39845.79 us
  vfs_lstat_fd                           138    39761.97 us
  user_path_at                           153    39469.58 us
  path_walk                              179    39435.76 us
  __link_path_walk                       189    39143.73 us
[...]

Note the times are skewed due to the function graph tracer not taking
into account schedules.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: give user a choice to show times spent while sleeping

The user may want to see the time a function spent sleeping.
This patch adds the trace option "sleep-time" to allow that.
The "sleep-time" option is default on.

 echo sleep-time > /debug/tracing/trace_options

produces:

 ------------------------------------------
 2)  avahi-d-3428  =>    <idle>-0
 ------------------------------------------

 2)               |      finish_task_switch() {
 2)   0.621 us    |        _spin_unlock_irq();
 2)   2.202 us    |      }
 2) ! 1002.197 us |    }
 2) ! 1003.521 us |  }

where as,

 echo nosleep-time > /debug/tracing/trace_options

produces:

 0)    <idle>-0    =>  yum-upd-3416
 ------------------------------------------

 0)               |              finish_task_switch() {
 0)   0.643 us    |                _spin_unlock_irq();
 0)   2.342 us    |              }
 0) + 41.302 us   |            }
 0) + 42.453 us   |          }
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: cleanup

Instead of just using the trace_seq buffer to print the filters, use
trace_seq_printf() as it was intended to be used.
Reported-by: NSteven Rostedt <rostedt@goodmis.org>
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237878871.8339.59.camel@charm-linux>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: fix hard-lockup with sched switch events

Some ftrace events, such as sched wakeup, can be traced
while the runqueue lock is hold. Since they are using
trace_current_buffer_unlock_commit(), they call wake_up()
which can try to grab the runqueue lock too, resulting in
a deadlock.

Now for all event, we call a new helper:
trace_nowake_buffer_unlock_commit() which do pretty the same than
trace_current_buffer_unlock_commit() except than it doesn't call
trace_wake_up().
Reported-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237759847-21025-4-git-send-email-fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This patch adds per-subsystem filtering to the event tracing subsystem.

It adds a 'filter' debugfs file to each subsystem directory.  This file
can be written to to set filters; reading from it will display the
current set of filters set for that subsystem.

Basically what it does is propagate the filter down to each event
contained in the subsystem.  If a particular event doesn't have a field
with the name specified in the filter, it simply doesn't get set for
that event.  You can verify whether or not the filter was set for a
particular event by looking at the filter file for that event.

As with per-event filters, compound expressions are supported, echoing
'0' to the subsystem's filter file clears all filters in the subsystem,
etc.
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710677.7703.49.camel@charm-linux>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This patch adds per-event filtering to the event tracing subsystem.

It adds a 'filter' debugfs file to each event directory.  This file can
be written to to set filters; reading from it will display the current
set of filters set for that event.

Basically, any field listed in the 'format' file for an event can be
filtered on (including strings, but not yet other array types) using
either matching ('==') or non-matching ('!=') 'predicates'.  A
'predicate' can be either a single expression:

 # echo pid != 0 > filter

 # cat filter
 pid != 0

or a compound expression of up to 8 sub-expressions combined using '&&'
or '||':

 # echo comm == Xorg > filter
 # echo "&& sig != 29" > filter

 # cat filter
 comm == Xorg
 && sig != 29

Only events having field values matching an expression will be available
in the trace output; non-matching events are discarded.

Note that a compound expression is built up by echoing each
sub-expression separately - it's not the most efficient way to do
things, but it keeps the parser simple and assumes that compound
expressions will be relatively uncommon.  In any case, a subsequent
patch introducing a way to set filters for entire subsystems should
mitigate any need to do this for lots of events.

Setting a filter without an '&&' or '||' clears the previous filter
completely and sets the filter to the new expression:

 # cat filter
 comm == Xorg
 && sig != 29

 # echo comm != Xorg

 # cat filter
 comm != Xorg

To clear a filter, echo 0 to the filter file:

 # echo 0 > filter
 # cat filter
 none

The limit of 8 predicates for a compound expression is arbitrary - for
efficiency, it's implemented as an array of pointers to predicates, and
8 seemed more than enough for any filter...
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710665.7703.48.camel@charm-linux>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

This patch makes the field descriptions defined for event tracing
available at run-time, for the event-filtering mechanism introduced
in a subsequent patch.

The common event fields are prepended with 'common_' in the format
display, allowing them to be distinguished from the other fields
that might internally have same name and can therefore be
unambiguously used in filters.
Signed-off-by: NTom Zanussi <tzanussi@gmail.com>
Acked-by: NFrederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1237710639.7703.46.camel@charm-linux>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: new tracing infrastructure feature

Provide infrastructure to generate software perf counter events
from tracepoints.
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20090319194233.557364871@chello.nl>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

The function depth in trace_printk was to facilitate the function
graph output. Now that the function graph calculates the depth within
the trace output, we no longer need to record the depth when the
trace_printk is called.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: feature to allow better serialized clock

This patch adds an option called "global-clock" that will allow
the tracer to switch to a slower but more accurate (across CPUs)
clock.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: fix to one cause of incorrect comm outputs in trace

The spinlock only protected the creation of a comm <=> pid pair.
But it was possible that a reader could look up a pid, and get the
wrong comm because it had no locking.

This also required changing trace_find_cmdline to copy the comm cache
and not just send back a pointer to it.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: new feature

This adds the generic support for syscalls tracing. This is
currently exploited through a devoted tracer but other tracing
engines can use it. (They just have to play with
{start,stop}_ftrace_syscalls() and use the display callbacks
unless they want to override them.)

The syscalls prototypes definitions are abused here to steal
some metadata informations:

- syscall name, param types, param names, number of params

The syscall addr is not directly saved during this definition
because we don't know if its prototype is available in the
namespace. But we don't really need it. The arch has just to
build a function able to resolve the syscall number to its
metadata struct.

The current tracer prints the syscall names, parameters names
and values (and their types optionally). Currently the value is
a raw hex but higher level values diplaying is on my TODO list.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1236955332-10133-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Provide basic callbacks to do syscall tracing.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Acked-by: NSteven Rostedt <rostedt@goodmis.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <1236401580-5758-2-git-send-email-fweisbec@gmail.com>
[ simplified it to a trace_printk() for now. ]
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: documentation

The use of the double __builtin_contant_p checks in the event_trace_printk
can be confusing to developers and reviewers. This patch adds a comment
to explain why it is there.
Requested-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
LKML-Reference: <20090313122235.43EB.A69D9226@jp.fujitsu.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: speed up on event tracing

The event_trace_printk is currently a wrapper function that calls
trace_vprintk. Because it uses a variable for the fmt it misses out
on the optimization of using the binary printk.

This patch makes event_trace_printk into a macro wrapper to use the
fmt as the same as the trace_printks.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: fix callsites with dynamic format strings

Since its new binary implementation, trace_printk() internally uses static
containers for the format strings on each callsites. But the value is
assigned once at build time, which means that it can't take dynamic
formats.

So this patch unearthes the raw trace_printk implementation for the callers
that will need trace_printk to be able to carry these dynamic format
strings. The trace_printk() macro will use the appropriate implementation
for each callsite. Most of the time however, the binary implementation will
still be used.

The other impact of this patch is that mmiotrace_printk() will use the old
implementation because it calls the low level trace_vprintk and we can't
guess here whether the format passed in it is dynamic or not.

Some parts of this patch have been written by Steven Rostedt (most notably
the part that chooses the appropriate implementation for each callsites).
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

To save memory, the tracer ring buffers are set to a minimum.
The activating of a trace expands the ring buffer size. This patch
adds this expanding, when an event is activated.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: clean up and enhancement

The TRACE_EVENT_FORMAT macro looks quite ugly and is limited in its
ability to save data as well as to print the record out. Working with
Ingo Molnar, we came up with a new format that is much more pleasing to
the eye of C developers. This new macro is more C style than the old
macro, and is more obvious to what it does.

Here's the example. The only updated macro in this patch is the
sched_switch trace point.

The old method looked like this:

 TRACE_EVENT_FORMAT(sched_switch,
        TP_PROTO(struct rq *rq, struct task_struct *prev,
                struct task_struct *next),
        TP_ARGS(rq, prev, next),
        TP_FMT("task %s:%d ==> %s:%d",
              prev->comm, prev->pid, next->comm, next->pid),
        TRACE_STRUCT(
                TRACE_FIELD(pid_t, prev_pid, prev->pid)
                TRACE_FIELD(int, prev_prio, prev->prio)
                TRACE_FIELD_SPECIAL(char next_comm[TASK_COMM_LEN],
                                    next_comm,
                                    TP_CMD(memcpy(TRACE_ENTRY->next_comm,
                                                 next->comm,
                                                 TASK_COMM_LEN)))
                TRACE_FIELD(pid_t, next_pid, next->pid)
                TRACE_FIELD(int, next_prio, next->prio)
        ),
        TP_RAW_FMT("prev %d:%d ==> next %s:%d:%d")
        );

The above method is hard to read and requires two format fields.

The new method:

 /*
  * Tracepoint for task switches, performed by the scheduler:
  *
  * (NOTE: the 'rq' argument is not used by generic trace events,
  *        but used by the latency tracer plugin. )
  */
 TRACE_EVENT(sched_switch,

	TP_PROTO(struct rq *rq, struct task_struct *prev,
		 struct task_struct *next),

	TP_ARGS(rq, prev, next),

	TP_STRUCT__entry(
		__array(	char,	prev_comm,	TASK_COMM_LEN	)
		__field(	pid_t,	prev_pid			)
		__field(	int,	prev_prio			)
		__array(	char,	next_comm,	TASK_COMM_LEN	)
		__field(	pid_t,	next_pid			)
		__field(	int,	next_prio			)
	),

	TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
		__entry->next_comm, __entry->next_pid, __entry->next_prio),

	TP_fast_assign(
		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
		__entry->prev_pid	= prev->pid;
		__entry->prev_prio	= prev->prio;
		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
		__entry->next_pid	= next->pid;
		__entry->next_prio	= next->prio;
	)
 );

This macro is called TRACE_EVENT, it is broken up into 5 parts:

 TP_PROTO:        the proto type of the trace point
 TP_ARGS:         the arguments of the trace point
 TP_STRUCT_entry: the structure layout of the entry in the ring buffer
 TP_printk:       the printk format
 TP_fast_assign:  the method used to write the entry into the ring buffer

The structure is the definition of how the event will be saved in the
ring buffer. The printk is used by the internal tracing in case of
an oops, and the kernel needs to print out the format of the record
to the console. This the TP_printk gives a means to show the records
in a human readable format. It is also used to print out the data
from the trace file.

The TP_fast_assign is executed directly. It is basically like a C function,
where the __entry is the handle to the record.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: cleanup

Remove a few leftovers and clean up the code a bit.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1236356510-8381-5-git-send-email-fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: faster and lighter tracing

Now that we have trace_bprintk() which is faster and consume lesser
memory than trace_printk() and has the same purpose, we can now drop
the old implementation in favour of the binary one from trace_bprintk(),
which means we move all the implementation of trace_bprintk() to
trace_printk(), so the Api doesn't change except that we must now use
trace_seq_bprintk() to print the TRACE_PRINT entries.

Some changes result of this:

- Previously, trace_bprintk depended of a single tracer and couldn't
  work without. This tracer has been dropped and the whole implementation
  of trace_printk() (like the module formats management) is now integrated
  in the tracing core (comes with CONFIG_TRACING), though we keep the file
  trace_printk (previously trace_bprintk.c) where we can find the module
  management. Thus we don't overflow trace.c

- changes some parts to use trace_seq_bprintk() to print TRACE_PRINT entries.

- change a bit trace_printk/trace_vprintk macros to support non-builtin formats
  constants, and fix 'const' qualifiers warnings. But this is all transparent for
  developers.

- etc...

V2:

- Rebase against last changes
- Fix mispell on the changelog

V3:

- Rebase against last changes (moving trace_printk() to kernel.h)
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Acked-by: NSteven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1236356510-8381-5-git-send-email-fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: save on memory for tracing

Current tracers are typically using a struct(like struct ftrace_entry,
struct ctx_switch_entry, struct special_entr etc...)to record a binary
event. These structs can only record a their own kind of events.
A new kind of tracer need a new struct and a lot of code too handle it.

So we need a generic binary record for events. This infrastructure
is for this purpose.

[fweisbec@gmail.com: rebase against latest -tip, make it safe while sched
tracing as reported by Steven Rostedt]
Signed-off-by: NLai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Acked-by: NSteven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1236356510-8381-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: cleanup

Use a more generic name - this also allows the prototype to move
to kernel.h and be generally available to kernel developers who
want to do some quick tracing.
Signed-off-by: NIngo Molnar <mingo@elte.hu>

With the removal of the latency_trace file, we lost the ability
to see some of the finer details in a trace. Like the state of
interrupts enabled, the preempt count, need resched, and if we
are in an interrupt handler, softirq handler or not.

This patch simply creates an option to bring back the old format.
This also removes the warning about an unused variable that held
the latency_trace file operations.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: new feature

This patch creates a directory of files that correspond to the
per CPU ring buffers. These are binary files and are made to
be used with splice. This is the fastest way to extract data from
the ftrace ring buffers.

Thanks to Jiaying Zhang for pushing me to get this code fixed,
 and to Eduard - Gabriel Munteanu for his splice code that helped
 me debug my code.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

This patch adds the "format" file to the trace point event directory.
This is based off of work by Tom Zanussi, in which a file is exported
to be tread from user land such that a user space app may read the
binary record stored in the ring buffer.

 # cat /debug/tracing/events/sched/sched_switch/format
        field:pid_t prev_pid;   offset:12;      size:4;
        field:int prev_prio;    offset:16;      size:4;
        field special:char next_comm[TASK_COMM_LEN];    offset:20;      size:16;
        field:pid_t next_pid;   offset:36;      size:4;
        field:int next_prio;    offset:40;      size:4;

Idea-from: Tom Zanussi <tzanussi@gmail.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: clean up

The trace_seq functions may be used separately outside of the ftrace
iterator. The trace_seq_reset is needed for these operations.

This patch also renames trace_seq_reset to the more appropriate
trace_seq_init.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

This patch adds the interface to enable the C style trace points.
In the directory /debugfs/tracing/events/subsystem/event
We now have three files:

 enable : values 0 or 1 to enable or disable the trace event.

 available_types: values 'raw' and 'printf' which indicate the tracing
       types available for the trace point. If a developer does not
       use the TRACE_EVENT_FORMAT macro and just uses the TRACE_FORMAT
       macro, then only 'printf' will be available. This file is
       read only.

 type: values 'raw' or 'printf'. This indicates which type of tracing
       is active for that trace point. 'printf' is the default and
       if 'raw' is not available, this file is read only.

 # echo raw > /debug/tracing/events/sched/sched_wakeup/type
 # echo 1 > /debug/tracing/events/sched/sched_wakeup/enable

 Will enable the C style tracing for the sched_wakeup trace point.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Impact: lower overhead tracing

The current event tracer can automatically pick up trace points
that are registered with the TRACE_FORMAT macro. But it required
a printf format string and parsing. Although, this adds the ability
to get guaranteed information like task names and such, it took
a hit in overhead processing. This processing can add about 500-1000
nanoseconds overhead, but in some cases that too is considered
too much and we want to shave off as much from this overhead as
possible.

Tom Zanussi recently posted tracing patches to lkml that are based
on a nice idea about capturing the data via C structs using
STRUCT_ENTER, STRUCT_EXIT type of macros.

I liked that method very much, but did not like the implementation
that required a developer to add data/code in several disjoint
locations.

This patch extends the event_tracer macros to do a similar "raw C"
approach that Tom Zanussi did. But instead of having the developers
needing to tweak a bunch of code all over the place, they can do it
all in one macro - preferably placed near the code that it is
tracing. That makes it much more likely that tracepoints will be
maintained on an ongoing basis by the code they modify.

The new macro TRACE_EVENT_FORMAT is created for this approach. (Note,
a developer may still utilize the more low level DECLARE_TRACE macros
if they don't care about getting their traces automatically in the event
tracer.)

They can also use the existing TRACE_FORMAT if they don't need to code
the tracepoint in C, but just want to use the convenience of printf.

So if the developer wants to "hardwire" a tracepoint in the fastest
possible way, and wants to acquire their data via a user space utility
in a raw binary format, or wants to see it in the trace output but not
sacrifice any performance, then they can implement the faster but
more complex TRACE_EVENT_FORMAT macro.

Here's what usage looks like:

  TRACE_EVENT_FORMAT(name,
	TPPROTO(proto),
	TPARGS(args),
	TPFMT(fmt, fmt_args),
	TRACE_STUCT(
		TRACE_FIELD(type1, item1, assign1)
		TRACE_FIELD(type2, item2, assign2)
			[...]
	),
	TPRAWFMT(raw_fmt)
	);

Note name, proto, args, and fmt, are all identical to what TRACE_FORMAT
uses.

 name: is the unique identifier of the trace point
 proto: The proto type that the trace point uses
 args: the args in the proto type
 fmt: printf format to use with the event printf tracer
 fmt_args: the printf argments to match fmt

 TRACE_STRUCT starts the ability to create a structure.
 Each item in the structure is defined with a TRACE_FIELD

  TRACE_FIELD(type, item, assign)

 type: the C type of item.
 item: the name of the item in the stucture
 assign: what to assign the item in the trace point callback

 raw_fmt is a way to pretty print the struct. It must match
  the order of the items are added in TRACE_STUCT

 An example of this would be:

 TRACE_EVENT_FORMAT(sched_wakeup,
	TPPROTO(struct rq *rq, struct task_struct *p, int success),
	TPARGS(rq, p, success),
	TPFMT("task %s:%d %s",
	      p->comm, p->pid, success?"succeeded":"failed"),
	TRACE_STRUCT(
		TRACE_FIELD(pid_t, pid, p->pid)
		TRACE_FIELD(int, success, success)
	),
	TPRAWFMT("task %d success=%d")
	);

 This creates us a unique struct of:

 struct {
	pid_t		pid;
	int		success;
 };

 And the way the call back would assign these values would be:

	entry->pid = p->pid;
	entry->success = success;

The nice part about this is that the creation of the assignent is done
via macro magic in the event tracer.  Once the TRACE_EVENT_FORMAT is
created, the developer will then have a faster method to record
into the ring buffer. They do not need to worry about the tracer itself.

The developer would only need to touch the files in include/trace/*.h

Again, I would like to give special thanks to Tom Zanussi for this
nice idea.

Idea-from: Tom Zanussi <tzanussi@gmail.com>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Right now all tracers must manage their own trace buffers. This was
to enforce tracers to be independent in case we finally decide to
allow each tracer to have their own trace buffer.

But now we are adding event tracing that writes to the current tracer's
buffer. This adds an interface to allow events to write to the current
tracer buffer without having to manage its own. Since event tracing
has no "tracer", and is just a way to hook into any other tracer.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Now that several per-cpu files can be read or spliced at the
same, we want the read/splice callbacks for tracing files to be
reentrants.

Until now, a single global mutex (trace_types_lock) serialized
the access to tracing_read_pipe(), tracing_splice_read_pipe(),
and the seq helpers.

Ie: it means that if a user tries to read trace_pipe0 and
trace_pipe1 at the same time, the access to the function
tracing_read_pipe() is contended and one reader must wait for
the other to finish its read call.

The trace_type_lock mutex is mostly here to serialize the access
to the global current tracer (current_trace), which can be
changed concurrently. Although the iter struct keeps a private
pointer to this tracer, its callbacks can be changed by another
function.

The method used here is to not keep anymore private reference to
the tracer inside the iterator but to make a copy of it inside
the iterator. Then it checks on subsequents read calls if the
tracer has changed. This is not costly because the current
tracer is not expected to be changed often, so we use a branch
prediction for that.

Moreover, we add a private mutex to the iterator (there is one
iterator per file descriptor) to serialize the accesses in case
of multiple consumers per file descriptor (which would be a
silly idea from the user). Note that this is not to protect the
ring buffer, since the ring buffer already serializes the
readers accesses. This is to prevent from traces weirdness in
case of concurrent consumers. But these mutexes can be dropped
anyway, that would not result in any crash. Just tell me what
you think about it.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: split up tracing output per cpu

Currently, on the tracing debugfs directory, three files are
available to the user to let him extracting the trace output:

- trace is an iterator through the ring-buffer. It's a reader
  but not a consumer It doesn't block when no more traces are
  available.

- trace pretty similar to the former, except that it adds more
  informations such as prempt count, irq flag, ...

- trace_pipe is a reader and a consumer, it will also block
  waiting for traces if necessary (heh, yes it's a pipe).

The traces coming from different cpus are curretly mixed up
inside these files. Sometimes it messes up the informations,
sometimes it's useful, depending on what does the tracer
capture.

The tracing_cpumask file is useful to filter the output and
select only the traces captured a custom defined set of cpus.
But still it is not enough powerful to extract at the same time
one trace buffer per cpu.

So this patch creates a new directory: /debug/tracing/per_cpu/.

Inside this directory, you will now find one trace_pipe file and
one trace file per cpu.

Which means if you have two cpus, you will have:

 trace0
 trace1
 trace_pipe0
 trace_pipe1

And of course, reading these files will have the same effect
than with the usual tracing files, except that you will only see
the traces from the given cpu.

The original all-in-one cpu trace file are still available on
their original place.

Until now, only one consumer was allowed on trace_pipe to avoid
racy consuming on the ring-buffer. Now the approach changed a
bit, you can have only one consumer per cpu.

Which means you are allowed to read concurrently trace_pipe0 and
trace_pipe1 But you can't have two readers on trace_pipe0 or
trace_pipe1.

Following the same logic, if there is one reader on the common
trace_pipe, you can not have at the same time another reader on
trace_pipe0 or in trace_pipe1. Because in trace_pipe is already
a consumer in all cpu buffers in essence.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: api and pipe waiting change

Currently, the waiting used in tracing_read_pipe() is done through a
100 msecs schedule_timeout() loop which periodically check if there
are traces on the buffer.

This can cause small latencies for programs which are reading the incoming
events.

This patch makes the reader waiting for the trace_wait waitqueue except
for few tracers such as the sched and functions tracers which might be
already hold the runqueue lock while waking up the reader.

This is performed through a new callback wait_pipe() on struct tracer.
If none is implemented on a specific tracer, the default waiting for
trace_wait queue is attached.
Signed-off-by: NFrederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>