GitHub/exynos8895/android_kernel_samsung_universal8895.git
16 years agoring_buffer: map to cpu not page
Steven Rostedt [Thu, 2 Oct 2008 23:18:09 +0000 (19:18 -0400)]
ring_buffer: map to cpu not page

My original patch had a compile bug when NUMA was configured. I
referenced cpu when it should have been cpu_buffer->cpu.

Ingo quickly fixed this bug by replacing cpu with 'i' because that
was the loop counter. Unfortunately, the 'i' was the counter of
pages, not CPUs. This caused a crash when the number of pages allocated
for the buffers exceeded the number of pages, which would usually
be the case.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: ktime.h not included in ftrace.h
Steven Noonan [Thu, 2 Oct 2008 19:00:07 +0000 (12:00 -0700)]
ftrace: ktime.h not included in ftrace.h

Including <linux/ktime.h> eliminates the following error:

include/linux/ftrace.h:220: error: expected specifier-qualifier-list
before 'ktime_t'

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/fastboot: build fix
Ingo Molnar [Thu, 2 Oct 2008 15:45:47 +0000 (17:45 +0200)]
tracing/fastboot: build fix

fix:

 In file included from kernel/sysctl.c:52:
 include/linux/ftrace.h:217: error: 'KSYM_NAME_LEN' undeclared here (not in a function)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/fastboot: get the initcall name before it disappears
Frederic Weisbecker [Thu, 2 Oct 2008 11:26:05 +0000 (13:26 +0200)]
tracing/fastboot: get the initcall name before it disappears

After some initcall traces, some initcall names may be inconsistent.
That's because these functions will disappear from the .init section
and also their name from the symbols table.

So we have to copy the name of the function in a buffer large enough
during the trace appending. It is not costly for the ring_buffer because
the number of initcall entries is commonly not really large.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/fastboot: change the printing of boot tracer according to bootgraph.pl
Frederic Weisbecker [Thu, 2 Oct 2008 10:59:20 +0000 (12:59 +0200)]
tracing/fastboot: change the printing of boot tracer according to bootgraph.pl

Change the boot tracer printing to make it parsable for
the scripts/bootgraph.pl script.

We have now to output two lines for each initcall, according to the
printk in do_one_initcall() in init/main.c
We need now the call's time and the return's time.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring-buffer: fix build error
Ingo Molnar [Thu, 2 Oct 2008 09:04:14 +0000 (11:04 +0200)]
ring-buffer: fix build error

fix:

 kernel/trace/ring_buffer.c: In function ‘rb_allocate_pages’:
 kernel/trace/ring_buffer.c:235: error: ‘cpu’ undeclared (first use in this function)
 kernel/trace/ring_buffer.c:235: error: (Each undeclared identifier is reported only once
 kernel/trace/ring_buffer.c:235: error: for each function it appears in.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: preempt disable over interrupt disable
Steven Rostedt [Wed, 1 Oct 2008 17:14:09 +0000 (13:14 -0400)]
ftrace: preempt disable over interrupt disable

With the new ring buffer infrastructure in ftrace, I'm trying to make
ftrace a little more light weight.

This patch converts a lot of the local_irq_save/restore into
preempt_disable/enable.  The original preempt count in a lot of cases
has to be sent in as a parameter so that it can be recorded correctly.
Some places were recording it incorrectly before anyway.

This is also laying the ground work to make ftrace a little bit
more reentrant, and remove all locking. The function tracers must
still protect from reentrancy.

Note: All the function tracers must be careful when using preempt_disable.
  It must do the following:

  resched = need_resched();
  preempt_disable_notrace();
  [...]
  if (resched)
preempt_enable_no_resched_notrace();
  else
preempt_enable_notrace();

The reason is that if this function traces schedule() itself, the
preempt_enable_notrace() will cause a schedule, which will lead
us into a recursive failure.

If we needed to reschedule before calling preempt_disable, we
should have already scheduled. Since we did not, this is most
likely that we should not and are probably inside a schedule
function.

If resched was not set, we still need to catch the need resched
flag being set when preemption was off and the if case at the
end will catch that for us.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring_buffer: allocate buffer page pointer
Steven Rostedt [Wed, 1 Oct 2008 15:14:54 +0000 (11:14 -0400)]
ring_buffer: allocate buffer page pointer

The current method of overlaying the page frame as the buffer page pointer
can be very dangerous and limits our ability to do other things with
a page from the buffer, like send it off to disk.

This patch allocates the buffer_page instead of overlaying the page's
page frame. The use of the buffer_page has hardly changed due to this.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: type cast filter+verifier
Steven Rostedt [Wed, 1 Oct 2008 14:52:51 +0000 (10:52 -0400)]
ftrace: type cast filter+verifier

The mmiotrace map had a bug that would typecast the entry from
the trace to the wrong type. That is a known danger of C typecasts,
there's absolutely zero checking done on them.

Help that problem a bit by using a GCC extension to implement a
type filter that restricts the types that a trace record can be
cast into, and by adding a dynamic check (in debug mode) to verify
the type of the entry.

This patch adds a macro to assign all entries of ftrace using the type
of the variable and checking the entry id. The typecasts are now done
in the macro for only those types that it knows about, which should
be all the types that are allowed to be read from the tracer.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: adapt mmiotrace to the new type of print_line, fix
Frederic Weisbecker [Tue, 30 Sep 2008 16:13:45 +0000 (18:13 +0200)]
tracing/ftrace: adapt mmiotrace to the new type of print_line, fix

Correct the value's type of trace_empty function

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring_buffer: implement new locking
Steven Rostedt [Wed, 1 Oct 2008 04:29:53 +0000 (00:29 -0400)]
ring_buffer: implement new locking

The old "lock always" scheme had issues with lockdep, and was not very
efficient anyways.

This patch does a new design to be partially lockless on writes.
Writes will add new entries to the per cpu pages by simply disabling
interrupts. When a write needs to go to another page than it will
grab the lock.

A new "read page" has been added so that the reader can pull out a page
from the ring buffer to read without worrying about the writer writing over
it. This allows us to not take the lock for all reads. The lock is
now only taken when a read needs to go to a new page.

This is far from lockless, and interrupts still need to be disabled,
but it is a step towards a more lockless solution, and it also
solves a lot of the issues that were noticed by the first conversion
of ftrace to the ring buffers.

Note: the ring_buffer_{un}lock API has been removed.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring_buffer: remove raw from local_irq_save
Steven Rostedt [Wed, 1 Oct 2008 04:29:52 +0000 (00:29 -0400)]
ring_buffer: remove raw from local_irq_save

The raw_local_irq_save causes issues with lockdep. We don't need it
so replace them with local_irq_save.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: adapt the boot tracer to the new print_line type
Frederic Weisbecker [Mon, 29 Sep 2008 18:31:58 +0000 (20:31 +0200)]
tracing/ftrace: adapt the boot tracer to the new print_line type

This patch adapts the boot tracer to the new type of the
print_line callback.

It still relays entries it doesn't support to default output
functions.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Pekka Paalanen <pq@iki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: adapt mmiotrace to the new type of print_line
Frederic Weisbecker [Mon, 29 Sep 2008 18:27:42 +0000 (20:27 +0200)]
tracing/ftrace: adapt mmiotrace to the new type of print_line

Adapt mmiotrace to the new print_line type.
By default, it ignores (and consumes) types it doesn't support.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Pekka Paalanen <pq@iki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: fix pipe breaking
Pekka Paalanen [Mon, 29 Sep 2008 18:23:48 +0000 (20:23 +0200)]
tracing/ftrace: fix pipe breaking

This patch fixes a bug which break the pipe when the seq is empty.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: change the type of the print_line callback
Frederic Weisbecker [Mon, 29 Sep 2008 18:18:34 +0000 (20:18 +0200)]
tracing/ftrace: change the type of the print_line callback

We need a kind of disambiguation when a print_line callback
returns 0.

_There is not enough space to print all the entry.
 Please flush the seq and retry.
_I can't handle this type of entry

This patch changes the type of this callback for better information.

Also some changes have been made in this V2.

_ Only relay to default functions after the print_line callback fails.
_ This patch doesn't fix the issue with the broken pipe (see patch 2/4 for that)

Some things are still in discussion:

_ Find better names for the enum print_line_t values
_ Change the type of print_trace_line into boolean.

Patches to change that can be sent later.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Pekka Paalanen <pq@iki.fi>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: take advantage of variable length entries
Steven Rostedt [Tue, 30 Sep 2008 03:02:42 +0000 (23:02 -0400)]
ftrace: take advantage of variable length entries

Now that the underlining ring buffer for ftrace now hold variable length
entries, we can take advantage of this by only storing the size of the
actual event into the buffer. This happens to increase the number of
entries in the buffer dramatically.

We can also get rid of the "trace_cont" operation, but I'm keeping that
until we have no more users. Some of the ftrace tracers can now change
their code to adapt to this new feature.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: make work with new ring buffer
Steven Rostedt [Tue, 30 Sep 2008 03:02:41 +0000 (23:02 -0400)]
ftrace: make work with new ring buffer

This patch ports ftrace over to the new ring buffer.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring_buffer: reset buffer page when freeing
Steven Rostedt [Tue, 30 Sep 2008 03:02:40 +0000 (23:02 -0400)]
ring_buffer: reset buffer page when freeing

Mathieu Desnoyers pointed out that the freeing of the page frame needs
to be reset otherwise we might trigger BUG_ON in the page free code.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoring_buffer: add paranoid check for buffer page
Steven Rostedt [Tue, 30 Sep 2008 03:02:39 +0000 (23:02 -0400)]
ring_buffer: add paranoid check for buffer page

If for some strange reason the buffer_page gets bigger, or the page struct
gets smaller, I want to know this ASAP.  The best way is to not let the
kernel compile.

This patch adds code to test the size of the struct buffer_page against the
page struct and will cause compile issues if the buffer_page ever gets bigger
than the page struct.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing: unified trace buffer
Steven Rostedt [Tue, 30 Sep 2008 03:02:38 +0000 (23:02 -0400)]
tracing: unified trace buffer

This is a unified tracing buffer that implements a ring buffer that
hopefully everyone will eventually be able to use.

The events recorded into the buffer have the following structure:

  struct ring_buffer_event {
u32 type:2, len:3, time_delta:27;
u32 array[];
  };

The minimum size of an event is 8 bytes. All events are 4 byte
aligned inside the buffer.

There are 4 types (all internal use for the ring buffer, only
the data type is exported to the interface users).

 RINGBUF_TYPE_PADDING: this type is used to note extra space at the end
of a buffer page.

 RINGBUF_TYPE_TIME_EXTENT: This type is used when the time between events
is greater than the 27 bit delta can hold. We add another
32 bits, and record that in its own event (8 byte size).

 RINGBUF_TYPE_TIME_STAMP: (Not implemented yet). This will hold data to
help keep the buffer timestamps in sync.

RINGBUF_TYPE_DATA: The event actually holds user data.

The "len" field is only three bits. Since the data must be
4 byte aligned, this field is shifted left by 2, giving a
max length of 28 bytes. If the data load is greater than 28
bytes, the first array field holds the full length of the
data load and the len field is set to zero.

Example, data size of 7 bytes:

type = RINGBUF_TYPE_DATA
len = 2
time_delta: <time-stamp> - <prev_event-time-stamp>
array[0..1]: <7 bytes of data> <1 byte empty>

This event is saved in 12 bytes of the buffer.

An event with 82 bytes of data:

type = RINGBUF_TYPE_DATA
len = 0
time_delta: <time-stamp> - <prev_event-time-stamp>
array[0]: 84 (Note the alignment)
array[1..14]: <82 bytes of data> <2 bytes empty>

The above event is saved in 92 bytes (if my math is correct).
82 bytes of data, 2 bytes empty, 4 byte header, 4 byte length.

Do not reference the above event struct directly. Use the following
functions to gain access to the event table, since the
ring_buffer_event structure may change in the future.

ring_buffer_event_length(event): get the length of the event.
This is the size of the memory used to record this
event, and not the size of the data pay load.

ring_buffer_time_delta(event): get the time delta of the event
This returns the delta time stamp since the last event.
Note: Even though this is in the header, there should
be no reason to access this directly, accept
for debugging.

ring_buffer_event_data(event): get the data from the event
This is the function to use to get the actual data
from the event. Note, it is only a pointer to the
data inside the buffer. This data must be copied to
another location otherwise you risk it being written
over in the buffer.

ring_buffer_lock: A way to lock the entire buffer.
ring_buffer_unlock: unlock the buffer.

ring_buffer_alloc: create a new ring buffer. Can choose between
overwrite or consumer/producer mode. Overwrite will
overwrite old data, where as consumer producer will
throw away new data if the consumer catches up with the
producer.  The consumer/producer is the default.

ring_buffer_free: free the ring buffer.

ring_buffer_resize: resize the buffer. Changes the size of each cpu
buffer. Note, it is up to the caller to provide that
the buffer is not being used while this is happening.
This requirement may go away but do not count on it.

ring_buffer_lock_reserve: locks the ring buffer and allocates an
entry on the buffer to write to.
ring_buffer_unlock_commit: unlocks the ring buffer and commits it to
the buffer.

ring_buffer_write: writes some data into the ring buffer.

ring_buffer_peek: Look at a next item in the cpu buffer.
ring_buffer_consume: get the next item in the cpu buffer and
consume it. That is, this function increments the head
pointer.

ring_buffer_read_start: Start an iterator of a cpu buffer.
For now, this disables the cpu buffer, until you issue
a finish. This is just because we do not want the iterator
to be overwritten. This restriction may change in the future.
But note, this is used for static reading of a buffer which
is usually done "after" a trace. Live readings would want
to use the ring_buffer_consume above, which will not
disable the ring buffer.

ring_buffer_read_finish: Finishes the read iterator and reenables
the ring buffer.

ring_buffer_iter_peek: Look at the next item in the cpu iterator.
ring_buffer_read: Read the iterator and increment it.
ring_buffer_iter_reset: Reset the iterator to point to the beginning
of the cpu buffer.
ring_buffer_iter_empty: Returns true if the iterator is at the end
of the cpu buffer.

ring_buffer_size: returns the size in bytes of each cpu buffer.
Note, the real size is this times the number of CPUs.

ring_buffer_reset_cpu: Sets the cpu buffer to empty
ring_buffer_reset: sets all cpu buffers to empty

ring_buffer_swap_cpu: swaps a cpu buffer from one buffer with a
cpu buffer of another buffer. This is handy when you
want to take a snap shot of a running trace on just one
cpu. Having a backup buffer, to swap with facilitates this.
Ftrace max latencies use this.

ring_buffer_empty: Returns true if the ring buffer is empty.
ring_buffer_empty_cpu: Returns true if the cpu buffer is empty.

ring_buffer_record_disable: disable all cpu buffers (read only)
ring_buffer_record_disable_cpu: disable a single cpu buffer (read only)
ring_buffer_record_enable: enable all cpu buffers.
ring_buffer_record_enabl_cpu: enable a single cpu buffer.

ring_buffer_entries: The number of entries in a ring buffer.
ring_buffer_overruns: The number of entries removed due to writing wrap.

ring_buffer_time_stamp: Get the time stamp used by the ring buffer
ring_buffer_normalize_time_stamp: normalize the ring buffer time stamp
into nanosecs.

I still need to implement the GTOD feature. But we need support from
the cpu frequency infrastructure.  But this can be done at a later
time without affecting the ring buffer interface.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: give time for wakeup test to run
Steven Rostedt [Tue, 30 Sep 2008 03:02:37 +0000 (23:02 -0400)]
ftrace: give time for wakeup test to run

It is possible that the testing thread in the ftrace wakeup test does not
run before we stop the trace. This will cause the trace to fail since nothing
will be in the buffers.

This patch adds a small wait in the wakeup test to allow for the woken task
to run and be traced.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: don't consume unhandled entries by boot tracer
Frédéric Weisbecker [Thu, 25 Sep 2008 12:25:30 +0000 (13:25 +0100)]
tracing/ftrace: don't consume unhandled entries by boot tracer

When the boot tracer can't handle an entry output, it returns 1.
It should return 0 to relay on other output functions.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace/fastboot: disable tracers self-tests when boot tracer is selected
Frédéric Weisbecker [Wed, 24 Sep 2008 09:36:09 +0000 (10:36 +0100)]
ftrace/fastboot: disable tracers self-tests when boot tracer is selected

The tracing engine resets the ring buffer and the tracers touch it
too during self-tests. These self-tests happen during tracers registering
and work against boot tracing which is logging initcalls.

We have to disable tracing self-tests if the boot-tracer is selected.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: launch boot tracing after pre-smp initcalls
Frédéric Weisbecker [Tue, 23 Sep 2008 10:38:18 +0000 (11:38 +0100)]
tracing/ftrace: launch boot tracing after pre-smp initcalls

Launch the boot tracing inside the initcall_debug area. Old printk
have not been removed to keep the old way of initcall tracing for
backward compatibility.

[ mingo@elte.hu: resolved conflicts ]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: give an entry on the config for boot tracer
Frédéric Weisbecker [Tue, 23 Sep 2008 10:36:20 +0000 (11:36 +0100)]
tracing/ftrace: give an entry on the config for boot tracer

Bring the entry to choose the boot tracer on the kernel config.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: make tracing suitable to run the boot tracer
Frédéric Weisbecker [Tue, 23 Sep 2008 10:34:32 +0000 (11:34 +0100)]
tracing/ftrace: make tracing suitable to run the boot tracer

The tracing engine have now to be init in early_initcall to set the
boot tracer. Only the debugfs settings will be initialized at
fs_initcall time.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: add the boot tracer
Frédéric Weisbecker [Tue, 23 Sep 2008 10:32:08 +0000 (11:32 +0100)]
tracing/ftrace: add the boot tracer

Add the boot/initcall tracer.

It's primary purpose is to be able to trace the initcalls.

It is intended to be used with scripts/bootgraph.pl after some small
improvements.

Note that it is not active after its init. To avoid tracing (and so
crashing) before the whole tracing engine init, you have to explicitly
call start_boot_trace() after do_pre_smp_initcalls() to enable it.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/fastboot: add a script to visualize the kernel boot process / time
Arjan van de Ven [Sat, 13 Sep 2008 16:36:06 +0000 (09:36 -0700)]
tracing/fastboot: add a script to visualize the kernel boot process / time

When optimizing the kernel boot time, it's very valuable to visualize
what is going on at which time. In addition, with the fastboot asynchronous
initcall level, it's very valuable to see which initcall gets run where
and when.

This patch adds a script to turn a dmesg into a SVG graph (that can be
shown with tools such as InkScape, Gimp or Firefox) and a small change
to the initcall code to print the PID of the thread calling the initcall
(so that the script can work out the parallelism).

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
16 years agomarkers: bit-field is not thread-safe nor smp-safe
Lai Jiangshan [Fri, 10 Oct 2008 06:43:57 +0000 (14:43 +0800)]
markers: bit-field is not thread-safe nor smp-safe

bit-field is not thread-safe nor smp-safe.

struct marker_entry.rcu_pending is not protected by any lock
in rcu-callback free_old_closure().
so we must turn it into a safe type.

detail:

I suppose rcu_pending and ptype are store in struct marker_entry.tmp1

free_old_closure() side:           change ptype side:

                                |  load struct marker_entry.tmp1
--------------------------------|--------------------------------
                                |  change ptype bit in tmp1
load struct marker_entry.tmp1   |
change rcu_pending bit in tmp1  |
store tmp1                      |
--------------------------------|--------------------------------
                                |  store tmp1

now this result equals that free_old_closure() do not change rcu_pending
bit, bug! This bug will cause redundant rcu_barrier_sched() called.
not too harmful.

----- corresponding:

free_old_closure() side:           change ptype side:

load struct marker_entry.tmp1   |
--------------------------------|--------------------------------
                                |  load struct marker_entry.tmp1
change rcu_pending bit in tmp1  |
                                |  change ptype bit in tmp1
                                |  store tmp1
--------------------------------|--------------------------------
store tmp1                      |

now this result equals that change ptype side do not change ptype
bit, bug! this bug cause marker_probe_cb() access to invalid memory.
oops!

see also: http://en.wikipedia.org/wiki/Bit_field

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: fix unchecked format
Lai Jiangshan [Wed, 8 Oct 2008 02:23:36 +0000 (10:23 +0800)]
markers: fix unchecked format

when the second, third... probe is registered, its format is
not checked, this patch fixes it.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: turn marker_synchronize_unregister() into an inline
Mathieu Desnoyers [Fri, 3 Oct 2008 15:52:54 +0000 (11:52 -0400)]
markers: turn marker_synchronize_unregister() into an inline

Turn marker synchronize unregister into a static inline. There is no
reason to keep it as a macro over a static inline.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: re-enable fast batch registration
Mathieu Desnoyers [Wed, 1 Oct 2008 16:03:25 +0000 (12:03 -0400)]
markers: re-enable fast batch registration

Lai Jiangshan discovered a reentrancy issue with markers and fixed it by
adding synchronize_sched() calls at each registration/unregistraiton.

It works, but it removes the ability to do batch
registration/unregistration and can cause registration of ~100 markers
to take about 30 seconds on a loaded machine (synchronize_sched() is
much slower on such workloads).

This patch implements a version of the fix which won't slow down marker batch
registration/unregistration. It also go back to the original non-synchronized
reg/unreg.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosputrace: use marker_synchronize_unregister()
Mathieu Desnoyers [Mon, 29 Sep 2008 15:11:47 +0000 (11:11 -0400)]
sputrace: use marker_synchronize_unregister()

We need a marker_synchronize_unregister() before the end of exit() to make sure
every probe callers have exited the non preemptible section and thus are not
executing the probe code anymore.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Acked-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: documentation fix for teardown
Mathieu Desnoyers [Mon, 29 Sep 2008 15:10:34 +0000 (11:10 -0400)]
markers: documentation fix for teardown

Document the need for a marker_synchronize_unregister() before the end of
exit() to make sure every probe callers have exited the non preemptible
section and thus are not executing the probe code anymore.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: probe example, fix teardown
Mathieu Desnoyers [Mon, 29 Sep 2008 15:09:15 +0000 (11:09 -0400)]
markers: probe example, fix teardown

Need a marker_synchronize_unregister() before the end of exit() to make sure
every probe callers have exited the non preemptible section and thus are not
executing the probe code anymore.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: fix unregister bug and reenter bug, cleanup
Mathieu Desnoyers [Mon, 29 Sep 2008 15:08:03 +0000 (11:08 -0400)]
markers: fix unregister bug and reenter bug, cleanup

Use the new rcu_read_lock_sched/unlock_sched() in marker code around the call
site instead of preempt_disable/enable(). It helps reviewing the code more
easily.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: marker_synchronize_unregister()
Mathieu Desnoyers [Mon, 29 Sep 2008 15:05:13 +0000 (11:05 -0400)]
markers: marker_synchronize_unregister()

Create marker_synchronize_unregister() which must be called before the end of
exit() to make sure every probe callers have exited the non preemptible section
and thus are not executing the probe code anymore.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracepoints: fix reentrancy
Mathieu Desnoyers [Tue, 30 Sep 2008 05:51:12 +0000 (01:51 -0400)]
tracepoints: fix reentrancy

The tracepoints had the same problem markers did have wrt reentrancy. Apply a
similar fix using a rcu_barrier after each tracepoint mutex lock.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracepoints: use rcu sched
Mathieu Desnoyers [Tue, 30 Sep 2008 05:49:39 +0000 (01:49 -0400)]
tracepoints: use rcu sched

Make tracepoints use rcu sched. (cleanup)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agomarkers: fix unregister bug and reenter bug
Lai Jiangshan [Mon, 29 Sep 2008 08:00:05 +0000 (16:00 +0800)]
markers: fix unregister bug and reenter bug

unregister bug:

codes using makers are typically calling marker_probe_unregister()
and then destroying the data that marker_probe_func needs(or
unloading this module). This is bug when the corresponding
marker_probe_func is still running(on other cpus),
it is using the destroying/ed data.

we should call synchronize_sched() after marker_update_probes().

reenter bug:

marker_probe_register(), marker_probe_unregister() and
marker_probe_unregister_private_data() are not reentrant safe
functions. these 3 functions release markers_mutex and then
require it again and do "entry->oldptr = old; ...", but entry->oldptr
maybe is using now for these 3 functions may reenter when markers_mutex
is released.

we use synchronize_sched() instead of call_rcu_sched() to fix
this bug. actually we can do:
"
if (entry->rcu_pending)
rcu_barrier_sched();
"
after require markers_mutex again. but synchronize_sched()
is better and simpler. For these 3 functions are not critical path.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86/ftrace: use uaccess in atomic context
Frédéric Weisbecker [Wed, 24 Sep 2008 15:31:56 +0000 (16:31 +0100)]
x86/ftrace: use uaccess in atomic context

With latest -tip I get this bug:

[   49.439988] in_atomic():0, irqs_disabled():1
[   49.440118] INFO: lockdep is turned off.
[   49.440118] Pid: 2814, comm: modprobe Tainted: G        W 2.6.27-rc7 #4
[   49.440118]  [<c01215e1>] __might_sleep+0xe1/0x120
[   49.440118]  [<c01148ea>] ftrace_modify_code+0x2a/0xd0
[   49.440118]  [<c01148a2>] ? ftrace_test_p6nop+0x0/0xa
[   49.440118]  [<c016e80e>] __ftrace_update_code+0xfe/0x2f0
[   49.440118]  [<c01148a2>] ? ftrace_test_p6nop+0x0/0xa
[   49.440118]  [<c016f190>] ftrace_convert_nops+0x50/0x80
[   49.440118]  [<c016f1d6>] ftrace_init_module+0x16/0x20
[   49.440118]  [<c015498b>] load_module+0x185b/0x1d30
[   49.440118]  [<c01767a0>] ? find_get_page+0x0/0xf0
[   49.440118]  [<c02463c0>] ? sprintf+0x0/0x30
[   49.440118]  [<c034e012>] ? mutex_lock_interruptible_nested+0x1f2/0x350
[   49.440118]  [<c0154eb3>] sys_init_module+0x53/0x1b0
[   49.440118]  [<c0352340>] ? do_page_fault+0x0/0x740
[   49.440118]  [<c0104012>] syscall_call+0x7/0xb
[   49.440118]  =======================

It is because ftrace_modify_code() calls copy_to_user and
copy_from_user.
These functions have been inserted after guessing that there
couldn't be any race condition but copy_[to/from]_user might
sleep and __ftrace_update_code is called with local_irq_saved.

These function have been inserted since this commit:
d5e92e8978fd2574e415dc2792c5eb592978243d:
"ftrace: x86 use copy from user function"

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86: suppress trivial sparse signedness warnings
Harvey Harrison [Tue, 23 Sep 2008 22:05:46 +0000 (15:05 -0700)]
x86: suppress trivial sparse signedness warnings

Could just as easily change the three casts to cast to the correct
type...this patch changes the type of ftrace_nop instead.

Supresses sparse warnings:

 arch/x86/kernel/ftrace.c:157:14: warning: incorrect type in assignment (different signedness)
 arch/x86/kernel/ftrace.c:157:14:    expected long *static [toplevel] ftrace_nop
 arch/x86/kernel/ftrace.c:157:14:    got unsigned long *<noident>
 arch/x86/kernel/ftrace.c:161:14: warning: incorrect type in assignment (different signedness)
 arch/x86/kernel/ftrace.c:161:14:    expected long *static [toplevel] ftrace_nop
 arch/x86/kernel/ftrace.c:161:14:    got unsigned long *<noident>
 arch/x86/kernel/ftrace.c:165:14: warning: incorrect type in assignment (different signedness)
 arch/x86/kernel/ftrace.c:165:14:    expected long *static [toplevel] ftrace_nop
 arch/x86/kernel/ftrace.c:165:14:    got unsigned long *<noident>

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: warn on failure to disable mcount callers
Steven Rostedt [Mon, 22 Sep 2008 21:55:47 +0000 (14:55 -0700)]
ftrace: warn on failure to disable mcount callers

With the recent updates to ftrace, there should not be any failures when
modifying the code. If there is, then we need to warn about it.

This patch has a cleaned up version of the code that I used to discover
that the weak symbols were causing failures.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: replace none tracer by nop tracer
Frédéric Weisbecker [Sun, 21 Sep 2008 18:16:30 +0000 (20:16 +0200)]
tracing/ftrace: replace none tracer by nop tracer

Replace "none" tracer by the recently created "nop" tracer.
Both are pretty similar except that nop accepts TRACE_PRINT
or TRACE_SPECIAL entries.

And as a consequence, changing the size of the ring buffer now
requires that tracing has already been disabled.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: tracing engine depends on Nop Tracer
Frédéric Weisbecker [Sun, 21 Sep 2008 18:12:14 +0000 (20:12 +0200)]
tracing/ftrace: tracing engine depends on Nop Tracer

Now that the nop tracer is used as the default tracer by
replacing the "none" tracer, tracing engine depends on it.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotracing/ftrace: make nop tracer reset previous entries
Frédéric Weisbecker [Sun, 21 Sep 2008 18:10:14 +0000 (20:10 +0200)]
tracing/ftrace: make nop tracer reset previous entries

If nop tracer is selected, some old entries from the previous tracer
could still be enqueued. Tracing have to be reset.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agotrace: remove pointless ifdefs
Steven Noonan [Sat, 20 Sep 2008 08:00:38 +0000 (01:00 -0700)]
trace: remove pointless ifdefs

The functions are already 'extern' anyway, so there's no problem
with linkage. Removing these ifdefs also helps find any potential
compiler errors.

Suggested by Andrew Morton.

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: mcount_addr defined but not used
Steven Noonan [Sat, 20 Sep 2008 08:00:37 +0000 (01:00 -0700)]
ftrace: mcount_addr defined but not used

When CONFIG_DYNAMIC_FTRACE isn't used, neither is mcount_addr. This
patch eliminates that warning.

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: add nop tracer
Steven Noonan [Fri, 19 Sep 2008 10:06:43 +0000 (03:06 -0700)]
ftrace: add nop tracer

A no-op tracer which can serve two purposes:

 1. A template for development of a new tracer.
 2. A convenient way to see ftrace_printk() calls without
    an irrelevant trace making the output messy.

[ mingo@elte.hu: resolved conflicts ]
Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: inject markers via trace_marker file
Pekka Paalanen [Tue, 16 Sep 2008 19:06:42 +0000 (22:06 +0300)]
ftrace: inject markers via trace_marker file

Allow a user to inject a marker (TRACE_PRINT entry) into the trace ring
buffer. The related file operations are derived from code by Frédéric
Weisbecker <fweisbec@gmail.com>.

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agommiotrace: remove left-over marker cruft
Pekka Paalanen [Tue, 16 Sep 2008 19:03:56 +0000 (22:03 +0300)]
mmiotrace: remove left-over marker cruft

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agommiotrace: handle TRACE_PRINT entries
Pekka Paalanen [Tue, 16 Sep 2008 19:02:27 +0000 (22:02 +0300)]
mmiotrace: handle TRACE_PRINT entries

Also make trace_seq_print_cont() non-static, and add a newline if the
seq buffer can't hold all data.

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86 mmiotrace: implement mmiotrace_printk()
Pekka Paalanen [Tue, 16 Sep 2008 19:00:34 +0000 (22:00 +0300)]
x86 mmiotrace: implement mmiotrace_printk()

Offer mmiotrace users a function to inject markers from inside the kernel.
This depends on the trace_vprintk() patch.

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: add trace_vprintk()
Pekka Paalanen [Tue, 16 Sep 2008 18:58:24 +0000 (21:58 +0300)]
ftrace: add trace_vprintk()

trace_vprintk() for easier implementation of tracer specific *_printk
functions. Add check check for no_tracer, and implement
__ftrace_printk() as a wrapper.

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: move mmiotrace functions out of trace.c
Pekka Paalanen [Tue, 16 Sep 2008 18:56:41 +0000 (21:56 +0300)]
ftrace: move mmiotrace functions out of trace.c

Moves the mmiotrace specific functions from trace.c to
trace_mmiotrace.c. Functions trace_wake_up(), tracing_get_trace_entry(),
and tracing_generic_entry_update() are therefore made available outside
trace.c.

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86 mmiotrace: fix a rare memory leak
Pekka Paalanen [Tue, 16 Sep 2008 18:54:16 +0000 (21:54 +0300)]
x86 mmiotrace: fix a rare memory leak

Signed-off-by: Pekka Paalanen <pq@iki.fi>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: fix unlocking of hash
Steven Rostedt [Sat, 6 Sep 2008 05:06:04 +0000 (01:06 -0400)]
ftrace: fix unlocking of hash

This must be brown paper bag week for Steven Rostedt!

While working on ftrace for PPC, I discovered that the hash locking done
when CONFIG_FTRACE_MCOUNT_RECORD is not set, is totally incorrect.

With a cut and paste error, I had the hash lock macro to lock for both
hash_lock _and_ hash_unlock!

This bug did not affect x86 since this bug was introduced when
CONFIG_FTRACE_MCOUNT_RECORD was added to x86.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: use ftrace_release for all dynamic ftrace functions
Steven Rostedt [Sat, 6 Sep 2008 05:06:03 +0000 (01:06 -0400)]
ftrace: use ftrace_release for all dynamic ftrace functions

ftrace_release is necessary for all uses of dynamic ftrace and not just
the archs that have CONFIG_FTRACE_MCOUNT_RECORD defined.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: make it depend on DEBUG_KERNEL
Ingo Molnar [Thu, 4 Sep 2008 12:04:51 +0000 (14:04 +0200)]
ftrace: make it depend on DEBUG_KERNEL

make most of the tracers depend on DEBUG_KERNEL - that's their intended
purpose. (most distributions have DEBUG_KERNEL enabled anyway so this is
not a practical limitation - but it simplifies the tracing menu in the
normal case)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: sched_switch: show the wakee's cpu
Peter Zijlstra [Thu, 4 Sep 2008 08:24:16 +0000 (10:24 +0200)]
ftrace: sched_switch: show the wakee's cpu

While profiling the smp behaviour of the scheduler it was needed to know to
which cpu a task got woken.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: make ftrace_printk usable with the other tracers
Peter Zijlstra [Thu, 4 Sep 2008 08:24:14 +0000 (10:24 +0200)]
ftrace: make ftrace_printk usable with the other tracers

Currently ftrace_printk only works with the ftrace tracer, switch it to an
iter_ctrl setting so we can make us of them with other tracers too.

[rostedt@redhat.com: tweak to the disable condition]

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: print continue index fix
Steven Rostedt [Wed, 3 Sep 2008 21:42:51 +0000 (17:42 -0400)]
ftrace: print continue index fix

An item in the trace buffer that is bigger than one entry may be split
up using the TRACE_CONT entry. This makes it a virtual single entry.
The current code increments the iterator index even while traversing
TRACE_CONT entries, making it look like the iterator is further than
it actually is.

This patch adds code to not increment the iterator index while skipping
over TRACE_CONT entries.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: binary and not logical for continue test
Steven Rostedt [Wed, 3 Sep 2008 21:42:50 +0000 (17:42 -0400)]
ftrace: binary and not logical for continue test

Peter Zijlstra provided me with a nice brown paper bag while letting me know
that I was doing a logical AND and not a binary one, making a condition
true more often than it should be.

Luckily, a false true is handled by the calling function and no harm is
done. But this needs to be fixed regardless.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: make output nicely spaced for up to 999 cpus
Michael Ellerman [Wed, 20 Aug 2008 23:36:11 +0000 (16:36 -0700)]
ftrace: make output nicely spaced for up to 999 cpus

Currently some of the ftrace output goes skewiff if you have more
than 9 cpus, and some if you have more than 99.

Twiddle with the headers and format strings to make up to 999 cpus
display without causing spacing problems.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agostack tracer: depends on DEBUG_KERNEL
Ingo Molnar [Thu, 4 Sep 2008 13:04:37 +0000 (15:04 +0200)]
stack tracer: depends on DEBUG_KERNEL

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: stack trace add indexes
Steven Rostedt [Fri, 29 Aug 2008 20:51:43 +0000 (16:51 -0400)]
ftrace: stack trace add indexes

This patch adds indexes into the stack that the functions in the
stack dump were found at. As an added bonus, I also added a diff
to show which function is the most notorious consumer of the stack.

The output now looks like this:

# cat /debug/tracing/stack_trace
        Depth   Size      Location    (48 entries)
        -----   ----      --------
  0)     2476     212   blk_recount_segments+0x39/0x59
  1)     2264      12   bio_phys_segments+0x16/0x1d
  2)     2252      20   blk_rq_bio_prep+0x23/0xaf
  3)     2232      12   init_request_from_bio+0x74/0x77
  4)     2220      56   __make_request+0x294/0x331
  5)     2164     136   generic_make_request+0x34f/0x37d
  6)     2028      56   submit_bio+0xe7/0xef
  7)     1972      28   submit_bh+0xd1/0xf0
  8)     1944     112   block_read_full_page+0x299/0x2a9
  9)     1832       8   blkdev_readpage+0x14/0x16
 10)     1824      28   read_cache_page_async+0x7e/0x109
 11)     1796      16   read_cache_page+0x11/0x49
 12)     1780      32   read_dev_sector+0x3c/0x72
 13)     1748      48   read_lba+0x4d/0xaa
 14)     1700     168   efi_partition+0x85/0x61b
 15)     1532      72   rescan_partitions+0x10e/0x266
 16)     1460      40   do_open+0x1c7/0x24e
 17)     1420     292   __blkdev_get+0x79/0x84
 18)     1128      12   blkdev_get+0x12/0x14
 19)     1116      20   register_disk+0xd1/0x11e
 20)     1096      28   add_disk+0x34/0x90
 21)     1068      52   sd_probe+0x2b1/0x366
 22)     1016      20   driver_probe_device+0xa5/0x120
 23)      996       8   __device_attach+0xd/0xf
 24)      988      32   bus_for_each_drv+0x3e/0x68
 25)      956      24   device_attach+0x56/0x6c
 26)      932      16   bus_attach_device+0x26/0x4d
 27)      916      64   device_add+0x380/0x4b4
 28)      852      28   scsi_sysfs_add_sdev+0xa1/0x1c9
 29)      824     160   scsi_probe_and_add_lun+0x919/0xa2a
 30)      664      36   __scsi_add_device+0x88/0xae
 31)      628      44   ata_scsi_scan_host+0x9e/0x21c
 32)      584      28   ata_host_register+0x1cb/0x1db
 33)      556      24   ata_host_activate+0x98/0xb5
 34)      532     192   ahci_init_one+0x9bd/0x9e9
 35)      340      20   pci_device_probe+0x3e/0x5e
 36)      320      20   driver_probe_device+0xa5/0x120
 37)      300      20   __driver_attach+0x3f/0x5e
 38)      280      36   bus_for_each_dev+0x40/0x62
 39)      244      12   driver_attach+0x19/0x1b
 40)      232      28   bus_add_driver+0x9c/0x1af
 41)      204      28   driver_register+0x76/0xd2
 42)      176      20   __pci_register_driver+0x44/0x71
 43)      156       8   ahci_init+0x14/0x16
 44)      148     100   _stext+0x42/0x122
 45)       48      20   kernel_init+0x175/0x1dc
 46)       28      28   kernel_thread_helper+0x7/0x10

The first column is simply an index starting from the inner most function
and counting down to the outer most.

The next column is the location that the function was found on the stack.

The next column is the size of the stack for that function.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: remove warning of old objcopy and local functions
Steven Rostedt [Wed, 27 Aug 2008 17:02:01 +0000 (13:02 -0400)]
ftrace: remove warning of old objcopy and local functions

The warning messages about old objcopy and local functions spam the
user quite drastically.  Remove the warning until we can find a nicer
way of tell the user to upgrade their objcopy.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: remove direct reference to mcount in trace code
Steven Rostedt [Thu, 28 Aug 2008 03:24:15 +0000 (23:24 -0400)]
ftrace: remove direct reference to mcount in trace code

The mcount record method of ftrace scans objdump for references to mcount.
Using mcount as the reference to test if the calls to mcount being replaced
are indeed calls to mcount, this use of mcount was also caught as a
location to change. Using a variable that points to the mcount address
moves this reference into the data section that is not scanned, and
we do not use a false location to try and modify.

The warn on code was what was used to detect this bug.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: add stack tracer
Steven Rostedt [Thu, 28 Aug 2008 03:31:01 +0000 (23:31 -0400)]
ftrace: add stack tracer

This is another tracer using the ftrace infrastructure, that examines
at each function call the size of the stack. If the stack use is greater
than the previous max it is recorded.

You can always see (and set) the max stack size seen. By setting it
to zero will start the recording again. The backtrace is also available.

For example:

# cat /debug/tracing/stack_max_size
1856

# cat /debug/tracing/stack_trace
[<c027764d>] stack_trace_call+0x8f/0x101
[<c021b966>] ftrace_call+0x5/0x8
[<c02553cc>] clocksource_get_next+0x12/0x48
[<c02542a5>] update_wall_time+0x538/0x6d1
[<c0245913>] do_timer+0x23/0xb0
[<c0257657>] tick_do_update_jiffies64+0xd9/0xf1
[<c02576b9>] tick_sched_timer+0x4a/0xad
[<c0250fe6>] __run_hrtimer+0x3e/0x75
[<c02518ed>] hrtimer_interrupt+0xf1/0x154
[<c022c870>] smp_apic_timer_interrupt+0x71/0x84
[<c021b7e9>] apic_timer_interrupt+0x2d/0x34
[<c0238597>] finish_task_switch+0x29/0xa0
[<c05abd13>] schedule+0x765/0x7be
[<c05abfca>] schedule_timeout+0x1b/0x90
[<c05ab4d4>] wait_for_common+0xab/0x101
[<c05ab5ac>] wait_for_completion+0x12/0x14
[<c033cfc3>] blk_execute_rq+0x84/0x99
[<c0402470>] scsi_execute+0xc2/0x105
[<c040250a>] scsi_execute_req+0x57/0x7f
[<c043afe0>] sr_test_unit_ready+0x3e/0x97
[<c043bbd6>] sr_media_change+0x43/0x205
[<c046b59f>] media_changed+0x48/0x77
[<c046b5ff>] cdrom_media_changed+0x31/0x37
[<c043b091>] sr_block_media_changed+0x16/0x18
[<c02b9e69>] check_disk_change+0x1b/0x63
[<c046f4c3>] cdrom_open+0x7a1/0x806
[<c043b148>] sr_block_open+0x78/0x8d
[<c02ba4c0>] do_open+0x90/0x257
[<c02ba869>] blkdev_open+0x2d/0x56
[<c0296a1f>] __dentry_open+0x14d/0x23c
[<c0296b32>] nameidata_to_filp+0x24/0x38
[<c02a1c68>] do_filp_open+0x347/0x626
[<c02967ef>] do_sys_open+0x47/0xbc
[<c02968b0>] sys_open+0x23/0x2b
[<c021aadd>] sysenter_do_call+0x12/0x26

I've tested this on both x86_64 and i386.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agokbuild: ftrace: don't assume that scripts/recordmcount.pl is executable
Andrew Morton [Wed, 27 Aug 2008 07:08:30 +0000 (09:08 +0200)]
kbuild: ftrace: don't assume that scripts/recordmcount.pl is executable

CHK     include/linux/version.h
  CHK     include/linux/utsrelease.h
  CC      scripts/mod/empty.o
/bin/sh: /usr/src/25/scripts/recordmcount.pl: Permission denied

We shouldn't assume that files have their `x' bits set.  There are various
ways in which file permissions get lost, including use of patch(1).

It might not be correct to assume that perl lives in $PATH?

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: objcopy version test for local symbols
Steven Rostedt [Mon, 25 Aug 2008 18:52:11 +0000 (14:52 -0400)]
ftrace: objcopy version test for local symbols

The --globalize-symbols option came out in objcopy version 2.17.
If the kernel is being compiled on a system with a lower version of
objcopy, then we can not use the globalize / localize trick to
link to symbols pointing to local functions.

This patch tests the version of objcopy and will only use the trick
if the version is greater than or equal to 2.17. Otherwise, if an
object has only local functions within a section, it will give a
nice warning and recommend the user to upgrade their objcopy.

Leaving the symbols unrecorded is not that big of a deal, since the
mcount record method changes the actual mcount code to be a simple
"ret" without recording registers or anything.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: clean up macro usage
Ingo Molnar [Mon, 25 Aug 2008 06:12:04 +0000 (08:12 +0200)]
ftrace: clean up macro usage

enclose the argument in parenthesis. (especially since we cast it,
which is a high prio operation)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: fix build failure
Stephen Rothwell [Mon, 25 Aug 2008 03:08:44 +0000 (13:08 +1000)]
ftrace: fix build failure

After disabling FTRACE_MCOUNT_RECORD via a patch, a dormant build
failure surfaced:

 kernel/trace/ftrace.c: In function 'ftrace_record_ip':
 kernel/trace/ftrace.c:416: error: incompatible type for argument 1 of '_spin_lock_irqsave'
 kernel/trace/ftrace.c:433: error: incompatible type for argument 1 of '_spin_lock_irqsave'

Introduced by commit 6dad8e07f4c10b17b038e84d29f3ca41c2e55cd0 ("ftrace:
add necessary locking for ftrace records").

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: x86 use copy to and from user functions
Steven Rostedt [Wed, 20 Aug 2008 16:55:07 +0000 (12:55 -0400)]
ftrace: x86 use copy to and from user functions

The modification of code is performed either by kstop_machine, before
SMP starts, or on module code before the module is executed. There is
no reason to do the modifications from assembly. The copy to and from
user functions are sufficient and produces cleaner and easier to read
code.

Thanks to Benjamin Herrenschmidt for suggesting the idea.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: handle weak symbol functions
Steven Rostedt [Wed, 20 Aug 2008 14:07:35 +0000 (10:07 -0400)]
ftrace: handle weak symbol functions

During tests and checks, I've discovered that there were failures to
convert mcount callers into nops. Looking deeper into these failures,
code that was attempted to be changed was not an mcount caller.
The current code only updates if the code being changed is what it expects,
but I still investigate any time there is a failure.

What was happening is that a weak symbol was being used as a reference
for other mcount callers. That weak symbol was also referenced elsewhere
so the offsets were using the strong symbol and not the function symbol
that it was referenced from.

This patch changes the setting up of the mcount_loc section to search
for a global function that is not weak. It will pick a local over a weak
but if only a weak is found in a section, a warning is printed and the
mcount location is not recorded (just to be safe).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: update recordmount.pl arch changes
Steven Rostedt [Fri, 15 Aug 2008 15:40:24 +0000 (11:40 -0400)]
ftrace: update recordmount.pl arch changes

I'm trying to keep all the arch changes in recordmcount.pl in one place.
I moved your code into that area, by adding the flags to the commands
that were passed in.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: fix build problem with CONFIG_FTRACE
Jeremy Fitzhardinge [Mon, 18 Aug 2008 22:58:12 +0000 (15:58 -0700)]
ftrace: fix build problem with CONFIG_FTRACE

I'm seeing when I use separate src/build dirs:

make[3]: *** [arch/x86/kernel/time_32.o] Error 1
/bin/sh: scripts/recordmcount.pl: No such file or directory
make[3]: *** [arch/x86/kernel/irq_32.o] Error 1
/bin/sh: scripts/recordmcount.pl: No such file or directory
make[3]: *** [arch/x86/kernel/ldt.o] Error 1
/bin/sh: scripts/recordmcount.pl: No such file or directory
make[3]: *** [arch/x86/kernel/i8259.o] Error 1
/bin/sh: scripts/recordmcount.pl: No such file or directory

This fixes it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: fix incorrect comment style of __ftrace_enabled_save()
Huang Ying [Mon, 18 Aug 2008 08:24:56 +0000 (16:24 +0800)]
ftrace: fix incorrect comment style of __ftrace_enabled_save()

This patch fixes incorrect comment style of __ftrace_enabled_save().

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: add necessary locking for ftrace records
Steven Rostedt [Sat, 16 Aug 2008 01:40:05 +0000 (21:40 -0400)]
ftrace: add necessary locking for ftrace records

The new design of pre-recorded mcounts and updating the code outside of
kstop_machine has changed the way the records themselves are protected.

This patch uses the ftrace_lock to protect the records. Note, the lock
still does not need to be taken within calls that are only called via
kstop_machine, since the that code can not run while the spin lock is held.

Also removed the hash_lock needed for the daemon when MCOUNT_RECORD is
configured. Also did a slight cleanup of an unused variable.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: do not init module on ftrace disabled
Steven Rostedt [Sat, 16 Aug 2008 01:40:04 +0000 (21:40 -0400)]
ftrace: do not init module on ftrace disabled

If one of the self tests of ftrace has disabled the function tracer,
do not run the code to convert the mcount calls in modules.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: fix some mistakes in error messages
Frédéric Weisbecker [Fri, 15 Aug 2008 19:08:22 +0000 (21:08 +0200)]
ftrace: fix some mistakes in error messages

This patch fixes some mistakes on the tracer in warning messages when
debugfs fails to create tracing files.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: srostedt@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: scripts/recordmcount.pl cross-build hack
Ingo Molnar [Mon, 9 Jun 2008 18:54:22 +0000 (20:54 +0200)]
ftrace: scripts/recordmcount.pl cross-build hack

hack around:

 ld: Relocatable linking with relocations from format elf32-i386 (init/.tmp_gl_calibrate.o) to format elf64-x86-64 (init/.tmp_mx_calibrate.o) i  CC      arch/x86/mm/extable.o
 objcopy: 'init/.tmp_mx_calibrate.o': No such file
 rm: cannot remove `init/.tmp_mx_calibrate.o': No such file or directory
 ld: Relocatable linking with relocations from format elf32-i386 (arch/x86/mm/extable.o) to format elf64-x86-64 (arch/x86/mm/.tmp_mx_extable.o) is not supported
 mv: cannot stat `arch/x86/mm/.tmp_mx_extable.o': No such file or directory
 ld: Relocatable linking with relocations from format elf32-i386 (arch/x86/mm/fault.o) to format elf64-x86-64 (arch/x86/mm/.tmp_mx_fault.o) is not supported

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: ftrace_kill_atomic() build fix
Ingo Molnar [Fri, 15 Aug 2008 16:22:09 +0000 (18:22 +0200)]
ftrace: ftrace_kill_atomic() build fix

fix:

 kernel/built-in.o: In function `ftrace_dump':
 (.text+0x2e2ea): undefined reference to `ftrace_kill_atomic'

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: build fix
Ingo Molnar [Fri, 15 Aug 2008 15:48:02 +0000 (17:48 +0200)]
ftrace: build fix

fix:

 In file included from init/main.c:65:
 include/linux/ftrace.h:166: error: expected ‘,' or ‘;' before ‘{' token
 make[1]: *** [init/main.o] Error 1
 make: *** [init/main.o] Error 2

Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: dump out ftrace buffers to console on panic
Steven Rostedt [Thu, 31 Jul 2008 02:36:46 +0000 (22:36 -0400)]
ftrace: dump out ftrace buffers to console on panic

At OLS I had a lot of interest to be able to have the ftrace buffers
dumped on panic.  Usually one would expect to uses kexec and examine
the buffers after a new kernel is loaded. But sometimes the resources
do not permit kdump and kexec, so having an option to still see the
sequence of events up to the crash is very advantageous.

This patch adds the option to have the ftrace buffers dumped to the
console in the latency_trace format on a panic. When the option is set,
the default entries per CPU buffer are lowered to 16384, since the writing
to the serial (if that is the console) may take an awful long time
otherwise.

[
 Changes since -v1:
  Got alpine to send correctly (as well as spell check working).
  Removed config option.
  Moved the static variables into ftrace_dump itself.
  Gave printk a log level.
]

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: ftrace_printk doc moved
Steven Rostedt [Fri, 1 Aug 2008 20:45:49 +0000 (16:45 -0400)]
ftrace: ftrace_printk doc moved

Based on Randy Dunlap's suggestion, the ftrace_printk kernel-doc belongs
with the ftrace_printk macro that should be used. Not with the
__ftrace_printk internal function.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: printk formatting infrastructure
Steven Rostedt [Fri, 1 Aug 2008 16:26:41 +0000 (12:26 -0400)]
ftrace: printk formatting infrastructure

This patch adds a feature that can help kernel developers debug their
code using ftrace.

  int ftrace_printk(const char *fmt, ...);

This records into the ftrace buffer using printf formatting. The entry
size in the buffers are still a fixed length. A new type has been added
that allows for more entries to be used for a single recording.

The start of the print is still the same as the other entries.

It returns the number of characters written to the ftrace buffer.

For example:

Having a module with the following code:

static int __init ftrace_print_test(void)
{
        ftrace_printk("jiffies are %ld\n", jiffies);
        return 0;
}

Gives me:

  insmod-5441  3...1 7569us : ftrace_print_test: jiffies are 4296626666

for the latency_trace file and:

          insmod-5441  [03]  1959.370498: ftrace_print_test jiffies are 4296626666

for the trace file.

Note: Only the infrastructure should go into the kernel. It is to help
facilitate debugging for other kernel developers. Calls to ftrace_printk
is not intended to be left in the kernel, and should be frowned upon just
like scattering printks around in the code.

But having this easily at your fingertips helps the debugging go faster
and bugs be solved quicker.

Maybe later on, we can hook this with markers and have their printf format
be sucked into ftrace output.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: new continue entry - separate out from trace_entry
Steven Rostedt [Fri, 1 Aug 2008 16:26:40 +0000 (12:26 -0400)]
ftrace: new continue entry - separate out from trace_entry

Some tracers will need to work with more than one entry. In order to do this
the trace_entry structure was split into two fields. One for the start of
all entries, and one to continue an existing entry.

The trace_entry structure now has a "field" entry that consists of the previous
content of the trace_entry, and a "cont" entry that is just a string buffer
the size of the "field" entry.

Thanks to Andrew Morton for suggesting this idea.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: remove old pointers to mcount
Steven Rostedt [Fri, 15 Aug 2008 02:47:19 +0000 (22:47 -0400)]
ftrace: remove old pointers to mcount

When a mcount pointer is recorded into a table, it is used to add or
remove calls to mcount (replacing them with nops). If the code is removed
via removing a module, the pointers still exist.  At modifying the code
a check is always made to make sure the code being replaced is the code
expected. In-other-words, the code being replaced is compared to what
it is expected to be before being replaced.

There is a very small chance that the code being replaced just happens
to look like code that calls mcount (very small since the call to mcount
is relative). To remove this chance, this patch adds ftrace_release to
allow module unloading to remove the pointers to mcount within the module.

Another change for init calls is made to not trace calls marked with
__init. The tracing can not be started until after init is done anyway.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: move notrace to compiler.h
Steven Rostedt [Fri, 15 Aug 2008 02:47:18 +0000 (22:47 -0400)]
ftrace: move notrace to compiler.h

The notrace define belongs in compiler.h so that it can be used in
init.h

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: do not show freed records in available_filter_functions
Steven Rostedt [Fri, 15 Aug 2008 02:47:17 +0000 (22:47 -0400)]
ftrace: do not show freed records in available_filter_functions

Seems that freed records can appear in the available_filter_functions list.
This patch fixes that.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: use only 5 byte nops for x86
Steven Rostedt [Thu, 14 Aug 2008 22:05:05 +0000 (18:05 -0400)]
ftrace: use only 5 byte nops for x86

Mathieu Desnoyers revealed a bug in the original code. The nop that is
used to relpace the mcount caller can be a two part nop. This runs the
risk where a process can be preempted after executing the first nop, but
before the second part of the nop.

The ftrace code calls kstop_machine to keep multiple CPUs from executing
code that is being modified, but it does not protect against a task preempting
in the middle of a two part nop.

If the above preemption happens and the tracer is enabled, after the
kstop_machine runs, all those nops will be calls to the trace function.
If the preempted process that was preempted between the two nops is executed
again, it will execute half of the call to the trace function, and this
might crash the system.

This patch instead uses what both the latest Intel and AMD spec suggests.
That is the P6_NOP5 sequence of "0x0f 0x1f 0x44 0x00 0x00".

Note, some older CPUs and QEMU might fault on this nop, so this nop
is executed with fault handling first. If it detects a fault, it will then
use the code "0x66 0x66 0x66 0x66 0x90". If that faults, it will then
default to a simple "jmp 1f; .byte 0x00 0x00 0x00; 1:". The jmp is
not optimal but will do if the first two can not be executed.

TODO: Examine the cpuid to determine the nop to use.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: x86 mcount stub
Steven Rostedt [Thu, 14 Aug 2008 19:45:12 +0000 (15:45 -0400)]
ftrace: x86 mcount stub

x86 now sets up the mcount locations through the build and no longer
needs to record the ip when the function is executed. This patch changes
the initial mcount to simply return. There's no need to do any other work.
If the ftrace start up test fails, the original mcount will be what everything
will use, so having this as fast as possible is a good thing.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: enable using mcount recording on x86
Steven Rostedt [Thu, 14 Aug 2008 19:45:11 +0000 (15:45 -0400)]
ftrace: enable using mcount recording on x86

Enable the use of the __mcount_loc infrastructure on x86_64 and i386.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: rebuild everything on change to FTRACE_MCOUNT_RECORD
Steven Rostedt [Thu, 14 Aug 2008 19:45:10 +0000 (15:45 -0400)]
ftrace: rebuild everything on change to FTRACE_MCOUNT_RECORD

When enabling or disabling CONFIG_FTRACE_MCOUNT_RECORD, we want a full
kernel compile to handle the adding of the __mcount_loc sections.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: enable mcount recording for modules
Steven Rostedt [Thu, 14 Aug 2008 19:45:09 +0000 (15:45 -0400)]
ftrace: enable mcount recording for modules

This patch enables the loading of the __mcount_section of modules and
changing all the callers of mcount into nops.

The modification is done before the init_module function is called, so
again, we do not need to use kstop_machine to make these changes.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: mcount call site on boot nops core
Steven Rostedt [Thu, 14 Aug 2008 19:45:08 +0000 (15:45 -0400)]
ftrace: mcount call site on boot nops core

This is the infrastructure to the converting the mcount call sites
recorded by the __mcount_loc section into nops on boot. It also allows
for using these sites to enable tracing as normal. When the __mcount_loc
section is used, the "ftraced" kernel thread is disabled.

This uses the current infrastructure to record the mcount call sites
as well as convert them to nops. The mcount function is kept as a stub
on boot up and not converted to the ftrace_record_ip function. We use the
ftrace_record_ip to only record from the table.

This patch does not handle modules. That comes with a later patch.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: create __mcount_loc section
Steven Rostedt [Thu, 14 Aug 2008 19:45:07 +0000 (15:45 -0400)]
ftrace: create __mcount_loc section

This patch creates a section in the kernel called "__mcount_loc".
This will hold a list of pointers to the mcount relocation for
each call site of mcount.

For example:

objdump -dr init/main.o
[...]
Disassembly of section .text:

0000000000000000 <do_one_initcall>:
   0:   55                      push   %rbp
[...]
000000000000017b <init_post>:
 17b:   55                      push   %rbp
 17c:   48 89 e5                mov    %rsp,%rbp
 17f:   53                      push   %rbx
 180:   48 83 ec 08             sub    $0x8,%rsp
 184:   e8 00 00 00 00          callq  189 <init_post+0xe>
                        185: R_X86_64_PC32      mcount+0xfffffffffffffffc
[...]

We will add a section to point to each function call.

   .section __mcount_loc,"a",@progbits
[...]
   .quad .text + 0x185
[...]

The offset to of the mcount call site in init_post is an offset from
the start of the section, and not the start of the function init_post.
The mcount relocation is at the call site 0x185 from the start of the
.text section.

  .text + 0x185  == init_post + 0xa

We need a way to add this __mcount_loc section in a way that we do not
lose the relocations after final link.  The .text section here will
be attached to all other .text sections after final link and the
offsets will be meaningless.  We need to keep track of where these
.text sections are.

To do this, we use the start of the first function in the section.
do_one_initcall.  We can make a tmp.s file with this function as a reference
to the start of the .text section.

   .section __mcount_loc,"a",@progbits
[...]
   .quad do_one_initcall + 0x185
[...]

Then we can compile the tmp.s into a tmp.o

  gcc -c tmp.s -o tmp.o

And link it into back into main.o.

  ld -r main.o tmp.o -o tmp_main.o
  mv tmp_main.o main.o

But we have a problem.  What happens if the first function in a section
is not exported, and is a static function. The linker will not let
the tmp.o use it.  This case exists in main.o as well.

Disassembly of section .init.text:

0000000000000000 <set_reset_devices>:
   0:   55                      push   %rbp
   1:   48 89 e5                mov    %rsp,%rbp
   4:   e8 00 00 00 00          callq  9 <set_reset_devices+0x9>
                        5: R_X86_64_PC32        mcount+0xfffffffffffffffc

The first function in .init.text is a static function.

00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices

The lowercase 't' means that set_reset_devices is local and is not exported.
If we simply try to link the tmp.o with the set_reset_devices we end
up with two symbols: one local and one global.

 .section __mcount_loc,"a",@progbits
 .quad set_reset_devices + 0x10

00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices
                 U set_reset_devices

We still have an undefined reference to set_reset_devices, and if we try
to compile the kernel, we will end up with an undefined reference to
set_reset_devices, or even worst, it could be exported someplace else,
and then we will have a reference to the wrong location.

To handle this case, we make an intermediate step using objcopy.
We convert set_reset_devices into a global exported symbol before linking
it with tmp.o and set it back afterwards.

00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 T set_reset_devices

00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 T set_reset_devices

00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices

Now we have a section in main.o called __mcount_loc that we can place
somewhere in the kernel using vmlinux.ld.S and access it to convert
all these locations that call mcount into nops before starting SMP
and thus, eliminating the need to do this with kstop_machine.

Note, A well documented perl script (scripts/recordmcount.pl) is used
to do all this in one location.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoftrace: mark lapic_wd_event() notrace
Ingo Molnar [Tue, 29 Jul 2008 10:36:02 +0000 (12:36 +0200)]
ftrace: mark lapic_wd_event() notrace

it can be called in the NMI path:

[    0.645999] calling  ftrace_dynamic_init+0x0/0xd6
[    0.647521] ------------[ cut here ]------------
[    0.647521] WARNING: at kernel/trace/ftrace.c:348 ftrace_record_ip+0x4e/0x252()
[    0.647521] Modules linked in:
[    0.647521] Pid: 15, comm: kstop1 Not tainted 2.6.27-rc1-tip #22686
[    0.647521]
[    0.647521] Call Trace:
[    0.647521]  <NMI>  [<ffffffff8024593f>] warn_on_slowpath+0x5d/0x84
[    0.647521]  [<ffffffff80220b99>] ? lapic_wd_event+0xb/0x5c
[    0.647521]  [<ffffffff80287b3b>] ftrace_record_ip+0x4e/0x252
[    0.647521]  [<ffffffff80211274>] mcount_call+0x5/0x31
[    0.647521]  [<ffffffff80220b9e>] ? lapic_wd_event+0x10/0x5c
[    0.647521]  [<ffffffff8083f3ec>] nmi_watchdog_tick+0x19d/0x1ad
[    0.647521]  [<ffffffff8083e875>] default_do_nmi+0x75/0x1e3
[    0.647521]  [<ffffffff8083f0b3>] do_nmi+0x5d/0x94
[    0.647521]  [<ffffffff8083e2d2>] nmi+0xa2/0xc2
[    0.647521]  [<ffffffff802b48c3>] ? check_bytes_and_report+0x11/0xcc
[    0.647521]  <<EOE>>  [<ffffffff80211274>] ? mcount_call+0x5/0x31
[    0.647521]  [<ffffffff802b49df>] check_object+0x61/0x1b0
[    0.647521]  [<ffffffff802b502a>] __slab_free+0x169/0x2ae
[    0.647521]  [<ffffffff80242dbf>] ? __cleanup_sighand+0x25/0x27
[    0.647521]  [<ffffffff80242dbf>] ? __cleanup_sighand+0x25/0x27
[    0.647521]  [<ffffffff802b60cd>] kmem_cache_free+0x85/0xb9
[    0.647521]  [<ffffffff80242dbf>] __cleanup_sighand+0x25/0x27
[    0.647521]  [<ffffffff80247b3d>] release_task+0x256/0x339
[    0.647521]  [<ffffffff802490b4>] do_exit+0x764/0x7ef
[    0.647521]  [<ffffffff8027624c>] __xchg+0x0/0x38
[    0.647521]  [<ffffffff8027619a>] ? stop_cpu+0x0/0xb2
[    0.647521]  [<ffffffff8027619a>] ? stop_cpu+0x0/0xb2
[    0.647521]  [<ffffffff8025922f>] kthread+0x4e/0x7b
[    0.647521]  [<ffffffff80212979>] child_rip+0xa/0x11
[    0.647521]  [<ffffffff80211c17>] ? restore_args+0x0/0x30
[    0.647521]  [<ffffffff802283a5>] ? native_load_tls+0x14/0x2e
[    0.647521]  [<ffffffff802591e1>] ? kthread+0x0/0x7b
[    0.647521]  [<ffffffff8021296f>] ? child_rip+0x0/0x11
[    0.647521]
[    0.647521] ---[ end trace 4eaa2a86a8e2da22 ]---
[    0.672032] initcall ftrace_dynamic_init+0x0/0xd6 returned 0 after 19 msecs

also mark it no-kprobes while at it.

Signed-off-by: Ingo Molnar <mingo@elte.hu>