Add a function probe that will cause a stack trace to be traced in
the ring buffer when the given function(s) are called.

format is:

 <function>:stacktrace[:<count>]

 echo 'schedule:stacktrace' > /debug/tracing/set_ftrace_filter
 cat /debug/tracing/trace_pipe
     kworker/2:0-4329  [002] ...2  2933.558007: <stack trace>
 => kthread
 => ret_from_fork
          <idle>-0     [000] .N.2  2933.558019: <stack trace>
 => rest_init
 => start_kernel
 => x86_64_start_reservations
 => x86_64_start_kernel
     kworker/2:0-4329  [002] ...2  2933.558109: <stack trace>
 => kthread
 => ret_from_fork
[...]

This can be set to only trace a specific amount of times:

 echo 'schedule:stacktrace:3' > /debug/tracing/set_ftrace_filter
 cat /debug/tracing/trace_pipe
           <...>-58    [003] ...2   841.801694: <stack trace>
 => kthread
 => ret_from_fork
          <idle>-0     [001] .N.2   841.801697: <stack trace>
 => start_secondary
           <...>-2059  [001] ...2   841.801736: <stack trace>
 => wait_for_common
 => wait_for_completion
 => flush_work
 => tty_flush_to_ldisc
 => input_available_p
 => n_tty_poll
 => tty_poll
 => do_select
 => core_sys_select
 => sys_select
 => system_call_fastpath

To remove these:

 echo '!schedule:stacktrace' > /debug/tracing/set_ftrace_filter
 echo '!schedule:stacktrace:0' > /debug/tracing/set_ftrace_filter
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Altough the trace_dump_stack() already skips three functions in
the call to stack trace, which gets the stack trace to start
at the caller of the function, the caller may want to skip some
more too (as it may have helper functions).

Add a skip argument to the trace_dump_stack() that lets the caller
skip back tracing functions that it doesn't care about.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add triggers to function tracer that lets an event get enabled or
disabled when a function is called:

format is:

 <function>:enable_event:<system>:<event>[:<count>]
 <function>:disable_event:<system>:<event>[:<count>]

 echo 'schedule:enable_event:sched:sched_switch' > /debug/tracing/set_ftrace_filter

Every time schedule is called, it will enable the sched_switch event.

 echo 'schedule:disable_event:sched:sched_switch:2' > /debug/tracing/set_ftrace_filter

The first two times schedule is called while the sched_switch
event is enabled, it will disable it. It will not count for a time
that the event is already disabled (or enabled for enable_event).

[ fixed return without mutex_unlock() - thanks to Dan Carpenter and smatch ]

Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

In order to let triggers enable or disable events, we need a 'soft'
method for doing so. For example, if a function probe is added that
lets a user enable or disable events when a function is called, that
change must be done without taking locks or a mutex, and definitely
it can't sleep. But the full enabling of a tracepoint is expensive.

By adding a 'SOFT_DISABLE' flag, and converting the flags to be updated
without the protection of a mutex (using set/clear_bit()), this soft
disable flag can be used to allow critical sections to enable or disable
events from being traced (after the event has been placed into "SOFT_MODE").

Some caveats though: The comm recorder (to map pids with a comm) can not
be soft disabled (yet). If you disable an event with with a "soft"
disable and wait a while before reading the trace, the comm cache may be
replaced and you'll get a bunch of <...> for comms in the trace.

Reading the "enable" file for an event that is disabled will now give
you "0*" where the '*' denotes that the tracepoint is still active but
the event itself is "disabled".

[ fixed _BIT used in & operation : thanks to Dan Carpenter and smatch ]

Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The entries to the probe hash must be freed after a synchronize_sched()
after the entry has been removed from the hash.

As the entries are registered with ops that may have their own callbacks,
and these callbacks may sleep, we can not use call_rcu_sched() because
the rcu callbacks registered with that are called from a softirq context.

Instead of using call_rcu_sched(), manually save the entries on a free_list
and at the end of the loop that removes the entries, do a synchronize_sched()
and then go through the free_list, freeing the entries.

Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

When a function probe is created, each function that the probe is
attached to, a "callback" method is called. On release of the probe,
each function entry calls the "free" method.

First, "callback" is a confusing name and does not really match what
it does. Callback sounds like it will be called when the probe
triggers. But that's not the case. This is really an "init" function,
so lets rename it as such.

Secondly, both "init" and "free" do not pass enough information back
to the handlers. Pass back the ops, ip and data for each time the
method is called. We have the information, might as well use it.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Most of the flags for the struct ftrace_event_file were moved over
to the flags of the struct ftrace_event_call, but the comments were
never updated.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

 echo 'schedule:snapshot:1' > /debug/tracing/set_ftrace_filter

This will cause the scheduler to trigger a snapshot the next time
it's called (you can use any function that's not called by NMI).

Even though it triggers only once, you still need to remove it with:

 echo '!schedule:snapshot:0' > /debug/tracing/set_ftrace_filter

The :1 can be left off for the first command:

 echo 'schedule:snapshot' > /debug/tracing/set_ftrace_filter

But this will cause all calls to schedule to trigger a snapshot.
This must be removed without the ':0'

 echo '!schedule:snapshot' > /debug/tracing/set_ftrace_filter

As adding a "count" is a different operation (internally).
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add alloc_snapshot() and free_snapshot() to allocate and free the
snapshot buffer respectively, and use these to remove duplicate
code.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Currently the function probe enables all functions and runs a "hash"
against every function call to see if it should call a probe. This
is extremely wasteful.

Note, a probe is something like:

  echo schedule:traceoff > /debug/tracing/set_ftrace_filter

When schedule is called, the probe will disable tracing. But currently,
it has a call back for *all* functions, and checks to see if the
called function is the probe that is needed.

The probe function has been created before ftrace was rewritten to
allow for more than one "op" to be registered by the function tracer.
When probes were created, it couldn't limit the functions without also
limiting normal function calls. But now we can, it's about time
to update the probe code.

Todo, have separate ops for different entries. That is, assign
a ftrace_ops per probe, instead of one op for all probes. But
as there's not many probes assigned, this may not be that urgent.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The function tracing probes that trigger traceon or traceoff can be
set to unlimited, or given a count of # of times to execute.

By separating these two types of probes, we can then use the dynamic
ftrace function filtering directly, and remove the brute force
"check if this function called is my probe" routines in ftrace.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The only thing ftrace_trace_onoff_unreg() does is to do a strcmp()
against the cmd parameter to determine what op to unregister. But
this compare is also done after the location that this function is
called (and returns). By moving the check for '!' to unregister after
the strcmp(), the callback function itself can just do the unregister
and we can get rid of the helper function.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Remove some duplicate code and replace it with a helper function.
This makes the code a it cleaner.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add EXPORT_SYMBOL_GPL() to let the tracing_snapshot() functions be
called from modules.

Also add a test to see if the snapshot was called from NMI context
and just warn in the tracing buffer if so, and return.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

There's a few places that ftrace uses trace_printk() for internal
use, but this requires context (normal, softirq, irq, NMI) buffers
to keep things lockless. But the trace_puts() does not, as it can
write the string directly into the ring buffer. Make a internal helper
for trace_puts() and have the internal functions use that.

This way the extra context buffers are not used.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Although trace_printk() is extremely fast, especially when it uses
trace_bprintk() (writes args straight to buffer instead of inserting
into string), it still has the overhead of calling one of the printf
sprintf() functions, that need to scan the fmt string to determine
what, if any args it has.

This is a waste of precious CPU cycles if the printk format has no
args but a single constant string. It is better to use trace_puts()
which does not have the overhead of the fmt scanning.

But wouldn't it be nice if the developer didn't have to think about
such things, and the compile would just do it for them?

  trace_printk("this string has no args\n");
  [...]
  trace_printk("this sting does %p %d\n", foo, bar);

As tracing is critical to have the least amount of overhead,
especially when dealing with race conditions, and you want to
eliminate any "Heisenbugs", you want the trace_printk() to use the
fastest possible means of tracing.

Currently the macro magic determines if it will use trace_bprintk()
or if the fmt is a dynamic string (a variable), it will fall
back to the slow trace_printk() method that does a full snprintf()
before copying it into the buffer, where as trace_bprintk() only
copys the pointer to the fmt and the args into the buffer.

Well, now there's a way to spend some more Hogwarts cash and come
up with new fancy macro magic.

  #define trace_printk(fmt, ...)			\
  do {							\
	char _______STR[] = __stringify((__VA_ARGS__));	\
	if (sizeof(_______STR) > 3)			\
		do_trace_printk(fmt, ##__VA_ARGS__);	\
	else						\
		trace_puts(fmt);			\
  } while (0)

The above needs a bit of explaining (both here and in the comments).

By stringifying the __VA_ARGS__, we can, at compile time, determine
the number of args that are being passed to trace_printk(). The extra
parenthesis are required, otherwise the compiler complains about
too many parameters for __stringify if there is more than one arg.

When there are no args, the __stringify((__VA_ARGS__)) converts into
"()\0", a string of 3 characters. Anything else, will be a string
containing more than 3 characters. Now we assign that string to a
dynamic char array, and then take the sizeof() of that array.
If it is greater than 3 characters, we know trace_printk() has args
and we need to do the full "do_trace_printk()" on them, otherwise
it was only passed a single arg and we can optimize to use trace_puts().

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NSteven "The King of Nasty Macros!" Rostedt <rostedt@goodmis.org>

The trace_printk() is extremely fast and is very handy as it can be
used in any context (including NMIs!). But it still requires scanning
the fmt string for parsing the args. Even the trace_bprintk() requires
a scan to know what args will be saved, although it doesn't copy the
format string itself.

Several times trace_printk() has no args, and wastes cpu cycles scanning
the fmt string.

Adding trace_puts() allows the developer to use an even faster
tracing method that only saves the pointer to the string in the
ring buffer without doing any format parsing at all. This will
help remove even more of the "Heisenbug" effect, when debugging.

Also fixed up the F_printk()s for the ftrace internal bprint and print events.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The changce to add the trace_buffer struct to have the trace array
have both the main buffer and max buffer broke the branch tracer
because the change did not update that code. As the branch tracer
adds a significant amount of overhead, and must be selected via
a selection (not a allyesconfig) it was missed in testing.
Reported-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

If debugging the kernel, and the developer wants to use
tracing_snapshot() in places where tracing_snapshot_alloc() may
be difficult (or more likely, the developer is lazy and doesn't
want to bother with tracing_snapshot_alloc() at all), then adding

  alloc_snapshot

to the kernel command line parameter will tell ftrace to allocate
the snapshot buffer (if configured) when it allocates the main
tracing buffer.

I also noticed that ring_buffer_expanded and tracing_selftest_disabled
had inconsistent use of boolean "true" and "false" with "0" and "1".
I cleaned that up too.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Move the tracing startup selftest code into its own function and
when not enabled, always have that function succeed.

This makes the register_tracer() function much more readable.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The ring buffer updates when done while the ring buffer is active,
needs to be completed on the CPU that is used for the ring buffer
per_cpu buffer. To accomplish this, schedule_work_on() is used to
schedule work on the given CPU.

Now there's no reason to use schedule_work_on() if the process
doing the update happens to be on the CPU that it is processing.
It has already filled the requirement. Instead, just do the work
and continue.

This is needed for tracing_snapshot_alloc() where it may be called
really early in boot, where the work queues have not been set up yet.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The new snapshot feature is quite handy. It's a way for the user
to take advantage of the spare buffer that, until then, only
the latency tracers used to "snapshot" the buffer when it hit
a max latency. Now users can trigger a "snapshot" manually when
some condition is hit in a program. But a snapshot currently can
not be triggered by a condition inside the kernel.

With the addition of tracing_snapshot() and tracing_snapshot_alloc(),
snapshots can now be taking when a condition is hit, and the
developer wants to snapshot the case without stopping the trace.

Note, any snapshot will overwrite the old one, so take care
in how this is done.

These new functions are to be used like tracing_on(), tracing_off()
and trace_printk() are. That is, they should never be called
in the mainline Linux kernel. They are solely for the purpose
of debugging.

The tracing_snapshot() will not allocate a buffer, but it is
safe to be called from any context (except NMIs). But if a
snapshot buffer isn't allocated when it is called, it will write
to the live buffer, complaining about the lack of a snapshot
buffer, and then stop tracing (giving you the "permanent snapshot").

tracing_snapshot_alloc() will allocate the snapshot buffer if
it was not already allocated and then take the snapshot. This routine
*may sleep*, and must be called from context that can sleep.
The allocation is done with GFP_KERNEL and not atomic.

If you need a snapshot in an atomic context, say in early boot,
then it is best to call the tracing_snapshot_alloc() before then,
where it will allocate the buffer, and then you can use the
tracing_snapshot() anywhere you want and still get snapshots.

Cc: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add a ref count to the trace_array structure and prevent removal
of instances that have open descriptors.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add the per_cpu directory to the created tracing instances:

  cd /sys/kernel/debug/tracing/instances
  mkdir foo
  ls foo/per_cpu/cpu0
buffer_size_kb	snapshot_raw  trace	  trace_pipe_raw
snapshot	stats	      trace_pipe
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add the "snapshot" file to the the multi-buffer instances.

  cd /sys/kernel/debug/tracing/instances
  mkdir foo
  ls foo
buffer_size_kb  buffer_total_size_kb  events  free_buffer  set_event
snapshot  trace  trace_clock  trace_marker  trace_options  trace_pipe
tracing_on
  cat foo/snapshot
 # tracer: nop
 #
 #
 # * Snapshot is freed *
 #
 # Snapshot commands:
 # echo 0 > snapshot : Clears and frees snapshot buffer
 # echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.
 #                      Takes a snapshot of the main buffer.
 # echo 2 > snapshot : Clears snapshot buffer (but does not allocate)
 #                      (Doesn't have to be '2' works with any number that
 #                       is not a '0' or '1')
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

There's a bit of duplicate code in creating the trace buffers for
the normal trace buffer and the max trace buffer among the instances
and the main global_trace. This code can be consolidated and cleaned
up a bit making the code cleaner and more readable as well as less
duplication.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The snapshot buffer belongs to the trace array not the tracer that is
running. The trace array should be the data structure that keeps track
of whether or not the snapshot buffer is allocated, not the tracer
desciptor. Having the trace array keep track of it makes modifications
so much easier.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add a 'snapshot_raw' per_cpu file that allows tools to read the raw
binary data of the snapshot buffer.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

When the preempt or irq latency tracers are enabled, they require
the ring buffer to be able to swap the per cpu sub buffers between
two main buffers. This adds a slight overhead to tracing as the
trace recording needs to perform some checks to synchronize
between recording and swaps that might be happening on other CPUs.

The config RING_BUFFER_ALLOW_SWAP is set when a user of the ring
buffer needs the "swap cpu" feature, otherwise the extra checks
are not implemented and removed from the tracing overhead.

The snapshot feature will swap per CPU if the RING_BUFFER_ALLOW_SWAP
config is set. But that only gets set by things like OPROFILE
and the irqs and preempt latency tracers.

This config is added to let the user decide to include this feature
with the snapshot agnostic from whether or not another user of
the ring buffer sets this config.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Add the snapshot file into the per_cpu tracing directories to allow
them to be read for an individual cpu. This also allows to clear
an individual cpu from the snapshot buffer.

If the kernel allows it (CONFIG_RING_BUFFER_ALLOW_SWAP is set), then
echoing in '1' into one of the per_cpu snapshot files will do an
individual cpu buffer swap instead of the entire file.

Cc: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Currently, the way the latency tracers and snapshot feature works
is to have a separate trace_array called "max_tr" that holds the
snapshot buffer. For latency tracers, this snapshot buffer is used
to swap the running buffer with this buffer to save the current max
latency.

The only items needed for the max_tr is really just a copy of the buffer
itself, the per_cpu data pointers, the time_start timestamp that states
when the max latency was triggered, and the cpu that the max latency
was triggered on. All other fields in trace_array are unused by the
max_tr, making the max_tr mostly bloat.

This change removes the max_tr completely, and adds a new structure
called trace_buffer, that holds the buffer pointer, the per_cpu data
pointers, the time_start timestamp, and the cpu where the latency occurred.

The trace_array, now has two trace_buffers, one for the normal trace and
one for the max trace or snapshot. By doing this, not only do we remove
the bloat from the max_trace but the instances of traces can now use
their own snapshot feature and not have just the top level global_trace have
the snapshot feature and latency tracers for itself.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The snapshot utility is extremely useful, and does not add any more
overhead in memory when another latency tracer is enabled. They use
the snapshot underneath. There's no reason to hide the snapshot file
when a latency tracer has been enabled in the kernel.

If any of the latency tracers (irq, preempt or wakeup) is enabled
then also select the snapshot facility.

Note, snapshot can be enabled without the latency tracers enabled.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Currently we do not know what buffer a module event was enabled in.
On unload, it is safest to clear all buffer instances, not just the
top level buffer.

Todo: Clear only the buffer that the event was used in. The
infrastructure is there to do this, but it makes the code a bit
more complex. Lets get the current code vetted before we add that.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

Currently, when a module with events is unloaded, the trace buffer is
cleared. This is just a safety net in case the module might have some
strange callback when its event is outputted. But there's no reason
to reset the buffer if the module didn't have any of its events traced.

Add a flag to the event "call" structure called WAS_ENABLED and gets set
when the event is ever enabled, and this flag never gets cleared. When a
module gets unloaded, if any of its events have this flag set, then the
trace buffer will get cleared.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

All the trace event flags have comments but the IGNORE_ENABLE flag
which is set for ftrace internal events that should not be enabled
via the debugfs "enable" file. That is, if the top level enable file
is set, it will enable all events. It use to just check the ftrace
event call descriptor "reg" field and skip those whithout it, but now
some ftrace internal events have a reg field but still need to be
skipped. The flag was created to ignore those events.

Now document it.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The move of blocked readers to the ring buffer left out the
init of the wait queue that is used. Tests missed this due to running
stress tests against the buffers, which didn't allow for any
readers to end up waiting. Running a simple read and wait triggered
a bug.
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

As we've added __init annotation to field-defining functions, we should
add __refdata annotation to event_call variables, which reference those
functions.

Link: http://lkml.kernel.org/r/51343C1F.2050502@huawei.comReported-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NLi Zefan <lizefan@huawei.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

The new multi-buffers added a descriptor that kept track of module
events, and the directories they use, with struct ftace_module_file_ops.
This is used to add a ref count to keep modules from unloading while
their files are being accessed.

As the descriptor is only needed when CONFIG_MODULES is enabled, it
is only declared when the config is enabled. But that struct is
dereferenced in a few areas outside the #ifdef CONFIG_MODULES.

By adding some helper routines and moving code around a little,
events can be compiled again without modules.
Reported-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

With the conversion of the data array to per cpu, sparse now complains
about the use of per_cpu_ptr() on the variable. But The variable is
allocated with alloc_percpu() and is fine to use. But since the structure
that contains the data variable does not annotate it as such, sparse
gives out a lot of false warnings.
Reported-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>

These two functions are called during kernel boot only.

Link: http://lkml.kernel.org/r/51258796.7020704@huawei.comSigned-off-by: NLi Zefan <lizefan@huawei.com>
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>