uprobes/tracing: Fully initialize uprobe_trace_consumer before uprobe_register()

probe_event_enable() does uprobe_register() and only after that sets
utc->tu and tu->consumer/flags. This can race with uprobe_dispatcher()
which can miss these assignments or see them out of order. Nothing
really bad can happen, but this doesn't look clean/safe.

And this does not allow to use uprobe_consumer->filter() we are going
to add, it is called by uprobe_register() and it needs utc->tu.

Change this code to initialize everything before uprobe_register(), and
reset tu->consumer/flags if it fails. We can't race with event_disable(),
the caller holds event_mutex, and if we could the code would be wrong
anyway.

In fact I think uprobe_trace_consumer should die, it buys nothing but
complicates the code. We can simply add uprobe_consumer into trace_uprobe.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes/tracing: Fix dentry/mount leak in create_trace_uprobe()

create_trace_uprobe() does kern_path() to find ->d_inode, but forgets
to do path_put(). We can do this right after igrab().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Add exports for module use

The original pull message for uprobes (commit 654443e2) noted:

  This tree includes uprobes support in 'perf probe' - but SystemTap
  (and other tools) can take advantage of user probe points as well.

In order to actually be usable in module-based tools like SystemTap, the
interface needs to be exported.  This patch first adds the obvious
exports for uprobe_register and uprobe_unregister.  Then it also adds
one for task_user_regset_view, which is necessary to get the correct
state of userspace registers.

Signed-off-by: Josh Stone <jistone@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>

uprobes: Kill the bogus IS_ERR_VALUE(xol_vaddr) check

utask->xol_vaddr is either zero or valid, remove the bogus
IS_ERR_VALUE() check in xol_free_insn_slot().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Do not allocate current->utask unnecessary

handle_swbp() does get_utask() before can_skip_sstep() for no reason,
we do not need ->utask if can_skip_sstep() succeeds.

Move get_utask() to pre_ssout() who actually starts to use it. Move
the initialization of utask->active_uprobe/state as well. This way
the whole initialization is consolidated in pre_ssout().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>

uprobes: Fix utask->xol_vaddr leak in pre_ssout()

pre_ssout() should do xol_free_insn_slot() if arch_uprobe_pre_xol()
fails, otherwise nobody will free the allocated slot.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Do not play with utask in xol_get_insn_slot()

pre_ssout()->xol_get_insn_slot() path is confusing and buggy. This patch
cleanups the code, the next one fixes the bug.

Change xol_get_insn_slot() to only allocate the slot and do nothing more,
move the initialization of utask->xol_vaddr/vaddr into pre_ssout().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Turn add_utask() into get_utask()

Rename add_utask() into get_utask() and change it to allocate on
demand to simplify the caller. Like get_xol_area() it will have
more users.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Fold xol_alloc_area() into get_xol_area()

Currently only xol_get_insn_slot() does get_xol_area() + xol_alloc_area(),
but this will have more users and we do not want to copy-and-paste this
code. This patch simply moves xol_alloc_area() into get_xol_area() to
simplify the current and future code.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Move alloc_page() from xol_add_vma() to xol_alloc_area()

Move alloc_page() from xol_add_vma() to xol_alloc_area() to cleanup
the code. This separates the memory allocations and consolidates the
-EALREADY cleanups and the error handling.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Change handle_swbp() to expose bp_vaddr to handler_chain()

Change handle_swbp() to set regs->ip = bp_vaddr in advance, this is
what consumer->handler() needs but uprobe_get_swbp_addr() is not
exported.

This also simplifies the code and makes it more consistent across
the supported architectures. handle_swbp() becomes the only caller
of uprobe_get_swbp_addr().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>

uprobes/x86: Change __skip_sstep() to actually skip the whole insn

__skip_sstep() doesn't update regs->ip. Currently this is correct
but only "by accident" and it doesn't skip the whole insn. Change
it to advance ->ip by the length of the detected 0x66*0x90 sequence.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Teach handler_chain() to filter out the probed task

Currrently the are 2 problems with pre-filtering:

1. It is not possible to add/remove a task (mm) after uprobe_register()

2. A forked child inherits all breakpoints and uprobe_consumer can not
   control this.

This patch does the first step to improve the filtering. handler_chain()
removes the breakpoints installed by this uprobe from current->mm if all
handlers return UPROBE_HANDLER_REMOVE.

Note that handler_chain() relies on ->register_rwsem to avoid the race
with uprobe_register/unregister which can add/del a consumer, or even
remove and then insert the new uprobe at the same address.

Perhaps we will add uprobe_apply_mm(uprobe, mm, is_register) and teach
copy_mm() to do filter(UPROBE_FILTER_FORK), but I think this change makes
sense anyway.

Note: instead of checking the retcode from uc->handler, we could add
uc->filter(UPROBE_FILTER_BPHIT). But I think this is not optimal to
call 2 hooks in a row. This buys nothing, and if handler/filter do
something nontrivial they will probably do the same work twice.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Reintroduce uprobe_consumer->filter()

Finally add uprobe_consumer->filter() and change consumer_filter()
to actually call this method.

Note that ->filter() accepts mm_struct, not task_struct. Because:

1. We do not have for_each_mm_user(mm, task).

2. Even if we implement for_each_mm_user(), ->filter() can
   use it itself.

3. It is not clear who will actually need this interface to
   do the "nontrivial" filtering.

Another argument is "enum uprobe_filter_ctx", consumer->filter() can
use it to figure out why/where it was called. For example, perhaps
we can add UPROBE_FILTER_PRE_REGISTER used by build_map_info() to
quickly "nack" the unwanted mm's. In this case consumer should know
that it is called under ->i_mmap_mutex.

See the previous discussion at http://marc.info/?t=135214229700002
Perhaps we should pass more arguments, vma/vaddr?

Note: this patch obviously can't help to filter out the child created
by fork(), this will be addressed later.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Rationalize the usage of filter_chain()

filter_chain() was added into install_breakpoint/remove_breakpoint to
simplify the initial changes but this is sub-optimal.

This patch shifts the callsite to the callers, register_for_each_vma()
and uprobe_mmap(). This way:

- It will be easier to add the new arguments. This is the main reason,
  we can do more optimizations later.

- register_for_each_vma(is_register => true) can be optimized, we only
  need to consult the new consumer. The previous consumers were already
  asked when they called uprobe_register().

This patch also moves the MMF_HAS_UPROBES check from remove_breakpoint(),
this allows to avoid the potentionally costly filter_chain(). Note that
register_for_each_vma(is_register => false) doesn't really need to take
->consumer_rwsem, but I don't think it makes sense to optimize this and
introduce filter_chain_lockless().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill uprobes_mutex[], separate alloc_uprobe() and __uprobe_register()

uprobe_register() and uprobe_unregister() are the only users of
mutex_lock(uprobes_hash(inode)), and the only reason why we can't
simply remove it is that we need to ensure that delete_uprobe() is
not possible after alloc_uprobe() and before consumer_add().

IOW, we need to ensure that when we take uprobe->register_rwsem
this uprobe is still valid and we didn't race with _unregister()
which called delete_uprobe() in between.

With this patch uprobe_register() simply checks uprobe_is_active()
and retries if it hits this very unlikely race. uprobes_mutex[] is
no longer needed and can be removed.

There is another reason for this change, prepare_uprobe() should be
folded into alloc_uprobe() and we do not want to hold the extra locks
around read_mapping_page/etc.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Introduce uprobe_is_active()

The lifetime of uprobe->rb_node and uprobe->inode is not refcounted,
delete_uprobe() is called when we detect that uprobe has no consumers,
and it would be deadly wrong to do this twice.

Change delete_uprobe() to WARN() if it was already called. We use
RB_CLEAR_NODE() to mark uprobe "inactive", then RB_EMPTY_NODE() can
be used to detect this case.

RB_EMPTY_NODE() is not used directly, we add the trivial helper for
the next change.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill uprobe_events, use RB_EMPTY_ROOT() instead

uprobe_events counts the number of uprobes in uprobes_tree but
it is used as a boolean. We can use RB_EMPTY_ROOT() instead.

Probably no_uprobe_events() added by this patch can have more
callers, say, mmf_recalc_uprobes().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Anton Arapov <anton@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill uprobe->copy_mutex

Now that ->register_rwsem is safe under ->mmap_sem we can kill
->copy_mutex and abuse down_write(&uprobe->consumer_rwsem).

This makes prepare_uprobe() even more ugly, but we should kill
it anyway.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill UPROBE_RUN_HANDLER flag

Simply remove UPROBE_RUN_HANDLER and the corresponding code.

It can only help if uprobe has a single consumer, and in fact
it is no longer needed after handler_chain() was changed to use
->register_rwsem, we simply can not race with uprobe_register().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Change filter_chain() to iterate ->consumers list

Now that it safe to use ->consumer_rwsem under ->mmap_sem we can
almost finish the implementation of filter_chain(). It still lacks
the actual uc->filter(...) call but othewrwise it is ready, just
it pretends that ->filter() always returns true.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Introduce uprobe->register_rwsem

Introduce uprobe->register_rwsem. It is taken for writing around
__uprobe_register/unregister.

Change handler_chain() to use this sem rather than consumer_rwsem.

The main reason for this change is that we have the nasty problem
with mmap_sem/consumer_rwsem dependency. filter_chain() needs to
protect uprobe->consumers like handler_chain(), but they can not
use the same lock. filter_chain() can be called under ->mmap_sem
(currently this is always true), but we want to allow ->handler()
to play with the probed task's memory, and this needs ->mmap_sem.

Alternatively we could use srcu, but synchronize_srcu() is very
slow and ->register_rwsem allows us to do more. In particular, we
can teach handler_chain() to do remove_breakpoint() if this bp is
"nacked" by all consumers, we know that we can't race with the
new consumer which does uprobe_register().

See also the next patches. uprobes_mutex[] is almost ready to die.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: _register() should always do register_for_each_vma(true)

To support the filtering uprobe_register() should do
register_for_each_vma(true) every time the new consumer comes,
we need to install the previously nacked breakpoints.

Note:
- uprobes_mutex[] should die, what it actually protects is
  alloc_uprobe().

- UPROBE_RUN_HANDLER should die too, obviously it can't work
  unless uprobe has a single consumer. The consumer should
  serialize with _register/_unregister itself. Or this flag
  should live in uprobe_consumer->state.

- Perhaps we can do some optimizations later. For example, if
  filter_chain() never returns false uprobe can record this
  fact and avoid the unnecessary register_for_each_vma().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: _unregister() should always do register_for_each_vma(false)

uprobe_unregister() removes the breakpoints only if the last consumer
goes away. To support the filtering it should do this every time, we
want to remove the breakpoints which nobody else want to keep.

Note: given that filter_chain() is not actually implemented, this patch
itself doesn't change the behaviour yet, register_for_each_vma(false)
is a heavy "nop" unless there are no more consumers.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Introduce filter_chain()

Add the new helper filter_chain(). Currently it is only placeholder,
the comment explains what is should do. We will change it later to
consult every consumer to decide whether we need to install the swbp.
Until then it works as if any consumer returns true, this matches the
current behavior.

Change install_breakpoint() to call filter_chain() instead of checking
uprobe->consumers != NULL. We obviously need this, and this equally
closes the race with _unregister().

Change remove_breakpoint() to call this helper too. Currently this is
pointless because remove_breakpoint() is only called when the last
consumer goes away, but we will change this.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill uprobe_consumer->filter()

uprobe_consumer->filter() is pointless in its current form, kill it.

We will add it back, but with the different signature/semantics. Perhaps
we will even re-introduce the callsite in handler_chain(), but not to
just skip uc->handler().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Kill the pointless inode/uc checks in register/unregister

register/unregister verifies that inode/uc != NULL. For what?
This really looks like "hide the potential problem", the caller
should pass the valid data.

register() also checks uc->next == NULL, probably to prevent the
double-register but the caller can do other stupid/wrong things.
If we do this check, then we should document that uc->next should
be cleared before register() and add BUG_ON().

Also add the small comment about the i_size_read() check.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

uprobes: Move __set_bit(UPROBE_SKIP_SSTEP) into alloc_uprobe()

Cosmetic. __set_bit(UPROBE_SKIP_SSTEP) is the part of initialization,
it is not clear why it is set in insert_uprobe().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>

Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

. Check for flex and bison before continuing building, from Borislav Petkov.

. Make event_copy local to mmaps, fixing buffer wrap around problems, from
  David Ahern.

. Add option for runtime switching perf data file in perf report, just press
  's' and a menu with the valid files found in the current directory will be
  presented, from Feng Tang.

. Add support to display whole group data for raw columns, from Jiri Olsa.

. Fix SIGALRM and pipe read race for the rwtop perl script. from Jiri Olsa.

. Fix perf_evsel::exclude_GH handling and add a test to catch regressions, from
  Jiri Olsa.

. Error checking fixes, from Namhyung Kim.

. Fix calloc argument ordering, from Paul Gortmaker.

. Fix set event list leader, from Stephane Eranian.

. Add per processor socket count aggregation in perf stat, from Stephane Eranian.

. Fix perf python binding breakage.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf python: Link with sysfs.o

So that we fix this regression:

[root@sandy linux]# perf test -v 15
15: Try 'use perf' in python, checking link problems       :
--- start ---
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: /home/acme/git/build/perf/python/perf.so: undefined symbol: sysfs_find_mountpoint
---- end ----
Try 'use perf' in python, checking link problems: FAILED!
[root@sandy linux]#

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-8pf64bsdywg1gl9m55ul77hg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Pass the event_group info via perf_attr_details

So that we avoid dragging symbol.o into the python binding.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-izjubje7ltd1srji5wb0ygwi@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix calloc argument ordering

A sweep of the kernel for regex "kcalloc(sizeof" turned up 2 reversed
args, fixed in commit d3d09e18203dba16a9dbdb2b4cc673d90748cdd1 ("EDAC:
Fix kcalloc argument order") and also fixed in the networking commit
a1b1add07fa794974573d93483d68e373edfe7bd ("gro: Fix kcalloc argument
order").

I know that was the regex used, because on seeing the 1st of these
changes, I wondered "how many other instances of this are there" and I
happened to just use "calloc(sizeof" as a regex and it in turn found
these additional reversed args instances in the perf code.

In the kcalloc cases, the changes are cosmetic, since the numbers are
simply multiplied.  I had no desire to go data mining in userspace to
see if the same thing held true there, however.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Joe Perches <joe@perches.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359594349-25912-1-git-send-email-paul.gortmaker@windriver.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Adding automated parsing tests for group :GH modifiers

The ':GH' group modifier handling was just recently fixed, adding some
autommated tests to keep it that way. Adding tests for following events:

  "{cycles,cache-misses:G}:H"
  "{cycles,cache-misses:H}:G"
  "{cycles:G,cache-misses:H}:u"
  "{cycles:G,cache-misses:H}:uG"

Plus fixing test__group2 test.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359971803-2343-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix perf_evsel::exclude_GH handling

Let the perf_evsel::exclude_GH only prevent the reset of exclude_host
and exclude_guest attributes in case they were already set.

We cannot reset their values to 0, because they might have other
defaults set by event_attr_init.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359971803-2343-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf perl scripts: Fix SIGALRM and pipe read race for rwtop

Fixing rwtop script race. The issue is caused by rwtop script triggering
SIGALRM and underneath pipe reading layer reporting error when
interrupted.

Fixing this by setting SA_RESTART for rwtop SIGALRM handler, which
avoids interruption of the pipe reading layer.

The discussion for this issue & fix is here:
https://lkml.org/lkml/2012/9/18/123

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Original-patch-by: Andrew Jones <drjones@redhat.com>
Cc: Andrew Jones <drjones@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360080351-3246-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Add support to display whole group data for raw columns

Currently we don't display group members' values for raw columns like
'Samples' and 'Period' when in group report mode.

Uniting '__hpp__percent_fmt' and '__hpp__raw_fmt' function under new
function __hpp__fmt. It's basically '__hpp__percent_fmt' code with new
'fmt_percent' bool parameter added saying whether raw number or
percentage should be printed.

This way raw columns print out all the group members when
in group report mode, like:

  $ perf record -e '{cycles,cache-misses}' ls
  ...
  $ perf report --group --show-total-period --stdio
  ...
  #         Overhead                    Period  Command      Shared Object                             Symbol
  # ................  ........................  .......  .................  .................................
  #
      23.63%  11.24%       3331335         317       ls  [kernel.kallsyms]  [k] __lock_acquire
      12.72%   0.00%       1793100           0       ls  [kernel.kallsyms]  [k] native_sched_clock
       9.72%   0.00%       1369920           0       ls  libc-2.14.90.so    [.] _nl_find_locale
       0.03%   0.07%          4476           2       ls  [kernel.kallsyms]  [k] intel_pmu_enable_all
       0.00%  11.73%             0         331       ls  ld-2.14.90.so      [.] _dl_cache_libcmp
       0.00%  11.06%             0         312       ls  [kernel.kallsyms]  [k] vma_interval_tree_insert

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359981185-16819-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Add per processor socket count aggregation

This patch adds per-processor socket count aggregation for system-wide
mode measurements. This is a useful mode to detect imbalance between
sockets.

To enable this mode, use --aggr-socket in addition
to -a. (system-wide).

The output includes the socket number and the number of online
processors on that socket. This is useful to gauge the amount of
aggregation.

 # ./perf stat -I 1000 -a --aggr-socket -e cycles sleep 2
 #           time socket cpus             counts events
      1.000097680 S0        4          5,788,785 cycles
      2.000379943 S0        4         27,361,546 cycles
      2.001167808 S0        4            818,275 cycles

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1360161962-9675-3-git-send-email-eranian@google.com
[ committer note: Added missing man page entry based on above comments ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add cpu_map processor socket level functions

This patch adds:
- cpu_map__get_socket: get socked id from cpu
- cpu_map__build_socket_map: build socket map
- cpu_map__socket: gets acutal socket from logical socket

Those functions are used by uncore and processor socket-level
aggregation modes.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1360161962-9675-2-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Make event_copy local to mmaps

I am getting segfaults *after* the time sorting of perf samples where
the event type is off the charts:

(gdb) bt
\#0  0x0807b1b2 in hists__inc_nr_events (hists=0x80a99c4, type=1163281902) at util/hist.c:1225
\#1  0x08070795 in perf_session_deliver_event (session=0x80a9b90, event=0xf7a6aff8, sample=0xffffc318, tool=0xffffc520,
    file_offset=0) at util/session.c:884
\#2  0x0806f9b9 in flush_sample_queue (s=0x80a9b90, tool=0xffffc520) at util/session.c:555
\#3  0x0806fc53 in process_finished_round (tool=0xffffc520, event=0x0, session=0x80a9b90) at util/session.c:645

This is bizarre because the event has already been processed once --
before it was added to the samples queue -- and the event was found to
be sane at that time.

There seem to be 2 causes:

1. perf_evlist__mmap_read updates the read location even though there
are outstanding references to events sitting in the mmap buffers via the
ordered samples queue.

2. There is a single evlist->event_copy for all evlist entries.
event_copy is used to handle an event wrapping at the mmap buffer
boundary.

This patch addresses the second problem - making event_copy local to
each perf_mmap. With this change my highly repeatable use case no longer
fails.

The first problem is much more complicated and will be the subject of a
future patch.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1360098762-61827-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sort: Check return value of strdup()

When setup_sorting() is called, 'str' is passed to strtok_r() but it's
not checked to have a valid pointer.  As strtok_r() accepts NULL pointer
on a first argument and use the third argument in that case, it can
cause a trouble since our third argument, tmp, is not initialized.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1360130237-9963-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sort: Make setup_sorting returns an error code

Currently the setup_sorting() is called for parsing sort keys and exits
if it failed to add the sort key.  As it's included in libperf it'd be
better returning an error code rather than exiting application inside of
the library.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Suggested-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1360130237-9963-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sort: Drop ip_[lr] arguments from _sort__sym_cmp()

Current _sort__sym_cmp() function is used for comparing symbols between
two hist entries on symbol, symbol_from and symbol_to sort keys.  Those
functions pass addresses of symbols but it's meaningless since it gets
over-written inside of the _sort__sym_cmp function to a start address of
the symbol.  So just get rid of them.

This might cause a difference than prior output for branch stacks since
it seems not using start address of the symbol but branch address.
However AFAICS it'd be same as it gets overwritten anyway.

Also remove redundant part of code in sort__sym_cmp().

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1360130237-9963-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Check for flex and bison before continuing building

Check whether both executables are present on the system before
continuing with the build instead of failing halfway, if either are
missing.

Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/1359979554-9160-1-git-send-email-bp@alien8.de
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Fix set event list leader

The __perf_evlist__set_leader() was setting the leader for all events in
the list except the first. Which means it assumed the first event
already had event->leader = event.

Seems like this should be the role of the function to also do this. This
is a requirement for an upcoming patch set.

Signed-off-by: Stephane Eranian <eranian@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20130131125437.GA3656@quad
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Enable the runtime switching of perf data file

This is for tui browser only. This patch will check the returned key of
tui hists browser, if it's K_SWITH_INPUT_DATA, then recreate a session
for the new selected data file.

V2: Move the setup_brower() before the "repeat" jump point.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1359873501-24541-2-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Add option for runtime switching perf data file

Based on perf report/top/scripts browser integration idea from acme.

This will enable user to runtime switch the data file, when this option
is selected, it will popup all the legal data files in current working
directory, and the filename selected by user is saved in the global
variable "input_name", and a new key 'K_SWITCH_INPUT_DATA' will be
passed back to the built-in command which will perform the switch.

This initial version only enables it for 'perf report'.

v2: rebase to latest 'perf/core' branch (6e1d4dd) of acme's perf tree

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1359873501-24541-1-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/x86: Allow for architecture specific RDPMC indexes

Similar to config_base and event_base, allow architecture
specific RDPMC ECX values.

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Stephane Eranian <eranian@google.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360171589-6381-6-git-send-email-jacob.shin@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Move MSR address offset calculation to architecture specific files

Move counter index to MSR address offset calculation to
architecture specific files. This prepares the way for
perf_event_amd to enable counter addresses that are not
contiguous -- for example AMD Family 15h processors have 6 core
performance counters starting at 0xc0010200 and 4 northbridge
performance counters starting at 0xc0010240.

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Stephane Eranian <eranian@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360171589-6381-5-git-send-email-jacob.shin@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/amd: Use proper naming scheme for AMD bit field definitions

Update these AMD bit field names to be consistent with naming
convention followed by the rest of the file.

Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Stephane Eranian <eranian@google.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360171589-6381-4-git-send-email-jacob.shin@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/amd: Generalize northbridge constraints code for family 15h

Generalize northbridge constraints code for family 10h so that
later we can reuse the same code path with other AMD processor
families that have the same northbridge event constraints.

Signed-off-by: Robert Richter <rric@kernel.org>
Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Stephane Eranian <eranian@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360171589-6381-3-git-send-email-jacob.shin@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/amd: Rework northbridge event constraints handler

Code simplification. No functional changes.

Signed-off-by: Robert Richter <rric@kernel.org>
Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Stephane Eranian <eranian@google.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Robert Richter <rric@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1360171589-6381-2-git-send-email-jacob.shin@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/powerpc: Fix build error

Fix compile errors like those below:

  CC      arch/powerpc/perf/power7-pmu.o
   /home/git/linux/arch/powerpc/perf/power7-pmu.c:397:2: error: initialization from incompatible pointer type [-Werror]

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130205231938.GA24125@us.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'tip/perf/core' of git://git./linux/kernel/git/rostedt/linux-trace into perf/core

Pull tracing updated from Steve Rostedt.

Signed-off-by: Ingo Molnar <mingo@kernel.org>

tracing: Init current_trace to nop_trace and remove NULL checks

On early boot up, when the ftrace ring buffer is initialized, the
static variable current_trace is initialized to &nop_trace.
Before this initialization, current_trace is NULL and will never
become NULL again. It is always reassigned to a ftrace tracer.

Several places check if current_trace is NULL before it uses
it, and this check is frivolous, because at the point in time
when the checks are made the only way current_trace could be
NULL is if ftrace failed its allocations at boot up, and the
paths to these locations would probably not be possible.

By initializing current_trace to &nop_trace where it is declared,
current_trace will never be NULL, and we can remove all these
checks of current_trace being NULL which never needed to be
checked in the first place.

Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

. Make some POWER7 events available in sysfs, equivalent to
  what was done on x86, from Sukadev Bhattiprolu.

. Add event group view, from Namyung Kim:

  To use it, 'perf record' should group events when recording. And then perf
  report parses the saved group relation from file header and prints them
  together if --group option is provided.  You can use 'perf evlist' command to
  see event group information:

    $ perf record -e '{ref-cycles,cycles}' noploop 1
    [ perf record: Woken up 2 times to write data ]
    [ perf record: Captured and wrote 0.385 MB perf.data (~16807 samples) ]

    $ perf evlist --group
    {ref-cycles,cycles}

  With this example, default perf report will show you each event
  separately like this:

    $ perf report
    ...
    # group: {ref-cycles,cycles}
    # ========
    # Samples: 3K of event 'ref-cycles'
    # Event count (approx.): 3153797218
    #
    # Overhead  Command      Shared Object                      Symbol
    # ........  .......  .................  ..........................
        99.84%  noploop  noploop            [.] main
         0.07%  noploop  ld-2.15.so         [.] strcmp
         0.03%  noploop  [kernel.kallsyms]  [k] timerqueue_del
         0.03%  noploop  [kernel.kallsyms]  [k] sched_clock_cpu
         0.02%  noploop  [kernel.kallsyms]  [k] account_user_time
         0.01%  noploop  [kernel.kallsyms]  [k] __alloc_pages_nodemask
         0.00%  noploop  [kernel.kallsyms]  [k] native_write_msr_safe

    # Samples: 3K of event 'cycles'
    # Event count (approx.): 3722310525
    #
    # Overhead  Command      Shared Object                     Symbol
    # ........  .......  .................  .........................
        99.76%  noploop  noploop            [.] main
         0.11%  noploop  [kernel.kallsyms]  [k] _raw_spin_lock
         0.06%  noploop  [kernel.kallsyms]  [k] find_get_page
         0.03%  noploop  [kernel.kallsyms]  [k] sched_clock_cpu
         0.02%  noploop  [kernel.kallsyms]  [k] rcu_check_callbacks
         0.02%  noploop  [kernel.kallsyms]  [k] __current_kernel_time
         0.00%  noploop  [kernel.kallsyms]  [k] native_write_msr_safe

  In this case the event group information will be shown in the end of
  header area.  So you can use --group option to enable event group view.

    $ perf report --group
    ...
    # group: {ref-cycles,cycles}
    # ========
    # Samples: 7K of event 'anon group { ref-cycles, cycles }'
    # Event count (approx.): 6876107743
    #
    #         Overhead  Command      Shared Object                      Symbol
    # ................  .......  .................  ..........................
        99.84%  99.76%  noploop  noploop            [.] main
         0.07%   0.00%  noploop  ld-2.15.so         [.] strcmp
         0.03%   0.00%  noploop  [kernel.kallsyms]  [k] timerqueue_del
         0.03%   0.03%  noploop  [kernel.kallsyms]  [k] sched_clock_cpu
         0.02%   0.00%  noploop  [kernel.kallsyms]  [k] account_user_time
         0.01%   0.00%  noploop  [kernel.kallsyms]  [k] __alloc_pages_nodemask
         0.00%   0.00%  noploop  [kernel.kallsyms]  [k] native_write_msr_safe
         0.00%   0.11%  noploop  [kernel.kallsyms]  [k] _raw_spin_lock
         0.00%   0.06%  noploop  [kernel.kallsyms]  [k] find_get_page
         0.00%   0.02%  noploop  [kernel.kallsyms]  [k] rcu_check_callbacks
         0.00%   0.02%  noploop  [kernel.kallsyms]  [k] __current_kernel_time

  As you can see the Overhead column now contains both of ref-cycles and
  cycles and header line shows group information also - 'anon group {
  ref-cycles, cycles }'.  The output is sorted by period of group leader
  first.

  If perf.data file doesn't contain group information, this --group
  option does nothing.  So if you want enable event group view by
  default you can set it in ~/.perfconfig file:

    $ cat ~/.perfconfig
    [report]
    group = true

  It can be overridden with command line if you want:

    $ perf report --no-group

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf: Document the ABI of perf sysfs entries

This patchset addes two new sets of files to sysfs for POWER architecture.

- perf event config format in /sys/devices/cpu/format/event
- generic and POWER-specific perf events in /sys/devices/cpu/events/

The format of the first file is already documented in:

sysfs-bus-event_source-devices-format

Document the format of the second set of files '/sys/devices/cpu/events/*'
which would also become part of the ABI.

Changelog[v4]:
[Jiri Olsa]: Mention that multiple event= like terms can be specified
in the 'events' file.
[Jiri Olsa]: Remove the documentation for the 'config format' file
as it is already documented in 'Documentation/ABI/testing/'.
[Jiri Olsa]: Move ABI documentation from 'stable/' to 'testing/'

Changelog[v3]:
[Greg KH] Include ABI documentation.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130123062645.GG13720@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/POWER7: Make some POWER7 events available in sysfs

Make some POWER7-specific perf events available in sysfs.

$ /bin/ls -1 /sys/bus/event_source/devices/cpu/events/
branch-instructions
branch-misses
cache-misses
cache-references
cpu-cycles
instructions
PM_BRU_FIN
PM_BRU_MPRED
PM_CMPLU_STALL
PM_CYC
PM_GCT_NOSLOT_CYC
PM_INST_CMPL
PM_LD_MISS_L1
PM_LD_REF_L1
stalled-cycles-backend
stalled-cycles-frontend

where the 'PM_*' events are POWER specific and the others are the
generic events.

This will enable users to specify these events with their symbolic
names rather than with their raw code.

perf stat -e 'cpu/PM_CYC' ...

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130123062528.GE13720@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/POWER7: Make generic event translations available in sysfs

Make the generic perf events in POWER7 available via sysfs.

$ ls /sys/bus/event_source/devices/cpu/events
branch-instructions
branch-misses
cache-misses
cache-references
cpu-cycles
instructions
stalled-cycles-backend
stalled-cycles-frontend

$ cat /sys/bus/event_source/devices/cpu/events/cache-misses
event=0x400f0

This patch is based on commits that implement this functionality on x86.
Eg:
commit a47473939db20e3961b200eb00acf5fcf084d755
Author: Jiri Olsa <jolsa@redhat.com>
Date:   Wed Oct 10 14:53:11 2012 +0200

    perf/x86: Make hardware event translations available in sysfs

Changelog:[v2]
[Jiri Osla] Drop EVENT_ID() macro since it is only used once.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130123062454.GD13720@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf: Make EVENT_ATTR global

Rename EVENT_ATTR() to PMU_EVENT_ATTR() and make it global so it is
available to all architectures.

Further to allow architectures flexibility, have PMU_EVENT_ATTR() pass
in the variable name as a parameter.

Changelog[v2]
- [Jiri Olsa] No need to define PMU_EVENT_PTR()

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130123062422.GC13720@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/Power7: Use macros to identify perf events

Define and use macros to identify perf events codes This would make it
easier and more readable when these event codes need to be used in more
than one place.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130123062353.GB13720@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Add --group option

Add '-g/--group' option for showing event groups.  For simplicity it is
currently not compatible with other options.

  $ perf evlist --group
  {ref-cycles,cycles}

  $ perf evlist
  ref-cycles
  cycles

Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-20-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add report.group config option

Add report.group config option for setting default value of event
group view.  It affects the report output only if perf.data contains
event group info.

A user can write .perfconfig file like below to enable group view by
default:

  $ cat ~/.perfconfig
  [report]
  group = true

And it can be disabled through command line:

  $ perf report --no-group

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-19-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add --group option

Add --group option to enable event grouping.  When enabled, all the
group members information will be shown together with the leader.

  $ perf report --group
  ...
  # group: {ref-cycles,cycles}
  # ========
  #
  # Samples: 7K of event 'anon group { ref-cycles, cycles }'
  # Event count (approx.): 6876107743
  #
  #         Overhead  Command      Shared Object                      Symbol
  # ................  .......  .................  ..........................
  #
      99.84%  99.76%  noploop  noploop            [.] main
       0.07%   0.00%  noploop  ld-2.15.so         [.] strcmp
       0.03%   0.00%  noploop  [kernel.kallsyms]  [k] timerqueue_del
       0.03%   0.03%  noploop  [kernel.kallsyms]  [k] sched_clock_cpu
       0.02%   0.00%  noploop  [kernel.kallsyms]  [k] account_user_time
       0.01%   0.00%  noploop  [kernel.kallsyms]  [k] __alloc_pages_nodemask
       0.00%   0.00%  noploop  [kernel.kallsyms]  [k] native_write_msr_safe
       0.00%   0.11%  noploop  [kernel.kallsyms]  [k] _raw_spin_lock
       0.00%   0.06%  noploop  [kernel.kallsyms]  [k] find_get_page
       0.00%   0.02%  noploop  [kernel.kallsyms]  [k] rcu_check_callbacks
       0.00%   0.02%  noploop  [kernel.kallsyms]  [k] __current_kernel_time

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-18-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Show group description when event group is enabled

When using event group viewer, it's better to show the group description
rather than the leader information alone.

If a leader did not contain any member, it's a non-group event.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-17-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Bypass non-leader events when event group is enabled

Since we have all necessary information in the leader events and other
members don't, bypass members.  Member events will be shown along with
the leaders if event group is enabled.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-16-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf gtk/browser: Trim column header string when event group enabled

When event group feature is enabled, each column header is expanded to
match with the whole group column width.  But this is not needed for
GTK+ browser since ti usually use variable-width fonts.  So trim it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-15-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf gtk/browser: Add support for event group view

Show group members's overhead also when showing leader's if event
group is enabled.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranina@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-13-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Add suppport for event group view

Show group members' overhead also when showing the leader's if event
group is enabled.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-12-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Move coloring logic to hpp functions

Move coloring logic into the hpp functions so that each value can
be colored independently.  It'd required for event group view.

For overhead column, add a callback for printing 'folded_sign' of
callchains of a hist entry.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1358845787-1350-11-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ui/hist: Add support for event group view

Show group member's overhead also when showing the leader's if event
group is enabled.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-10-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf gtk/browser: Convert hpp helpers to a function

The hpp helpers do same job for each field so it was implemented as
macro in order to access those fields easily.  But it gets cumbersome to
maintain a large function in a macro as the function grows. Factor it
out to a function with a little helper macro to access field.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-9-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Convert hpp helpers to a function

The hpp helpers do same job for each field so it was implemented as
macro in order to access those fields easily.  But it gets cumbersome
to maintain a large function in a macro as the function grows. Factor
it out to a function with a little helper macro to access field.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-8-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ui/hist: Consolidate hpp helpers

Most of hpp helper functions do same jobs for different fields thus
consolidate them to appropriate functions/macros.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-7-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists: Resort hist entries using group members for output

When event group is enabled, sorting hist entries on periods for output
should consider groups members' period also.  To do that, build period
table using link/pair information and compare the table.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-6-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Make another loop for linking group hists

Now the event grouping viewing requires linking all member hists in a
group to the leader's.  Thus hists__output_resort should be called after
linking all events in evlist.

Introduce symbol_conf.event_group flag to determine whether the feature
is enabled or not.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-5-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf header: Add HEADER_GROUP_DESC feature

Save group relationship information so that it can be restored when perf
report is running.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-4-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Add group test conditions

As some new fields for handling groups added, check them to be sure to
have valid values in test__group* cases.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Keep group information

Add a few of group-related field in struct perf_{evlist,evsel} so that
the group information in a evlist can be known easily.  It only counts
groups which have more than 1 members since leader-only groups are
treated as non-group events.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1358845787-1350-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Delete maps on exit

Removing one more memory leak found with valgrind.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-gnb1gms0k8wictmtm2umpr8u@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Stop using exit()

Just return to the perf main() routine so that an unified exit path can
be followed and resources released, helping in finding memory leaks.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ro8oeodo96490nrhcph57atr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

. Fix some leaks in exit paths.

. Use memdup where applicable

. Remove some die() calls, allowing callers to handle exit paths
  gracefully.

. Correct typo in tools Makefile, fix from Borislav Petkov.

. Add 'perf bench numa mem' NUMA performance measurement suite, from Ingo Molnar.

. Handle dynamic array's element size properly, fix from Jiri Olsa.

. Fix memory leaks on evsel->counts, from Namhyung Kim.

. Make numa benchmark optional, allowing the build in machines where required
  numa libraries are not present, fix from Peter Hurley.

. Add interval printing in 'perf stat', from Stephane Eranian.

. Fix compile warnings in tests/attr.c, from Sukadev Bhattiprolu.

. Fix double free, pclose instead of fclose, leaks and double fclose errors
  found with the cppcheck tool, from Thomas Jarosch.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tracing: Add documentation of snapshot utility

This patch adds snapshot description in ftrace documentation.
This description includes what the snapshot is and how to use it.

Link: http://lkml.kernel.org/r/20121226025309.3252.150.stgit@liselsia
Cc: Rob Landley <rob@landley.net>
Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing: Make a snapshot feature available from userspace

Ftrace has a snapshot feature available from kernel space and
latency tracers (e.g. irqsoff) are using it. This patch enables
user applictions to take a snapshot via debugfs.

Add "snapshot" debugfs file in "tracing" directory.

  snapshot:
    This is used to take a snapshot and to read the output of the
    snapshot.

     # echo 1 > snapshot

    This will allocate the spare buffer for snapshot (if it is
    not allocated), and take a snapshot.

     # cat snapshot

    This will show contents of the snapshot.

     # echo 0 > snapshot

    This will free the snapshot if it is allocated.

    Any other positive values will clear the snapshot contents if
    the snapshot is allocated, or return EINVAL if it is not allocated.

Link: http://lkml.kernel.org/r/20121226025300.3252.86850.stgit@liselsia
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
[
   Fixed irqsoff selftest and also a conflict with a change
   that fixes the update_max_tr.
]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing: Replace static old_tracer check of tracer name

Currently the trace buffer read functions use a static variable
"old_tracer" for detecting if the current tracer changes. This
was suitable for a single trace file ("trace"), but to add a
snapshot feature that will use the same function for its file,
a check against a static variable is not sufficient.

To use the output functions for two different files, instead of
storing the current tracer in a static variable, as the trace
iterator descriptor contains a pointer to the original current
tracer's name, that pointer can now be used to check if the
current tracer has changed between different reads of the trace
file.

Link: http://lkml.kernel.org/r/20121226025252.3252.9276.stgit@liselsia
Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

tracing: Use sched_clock_cpu for trace_clock_global

For systems with an unstable sched_clock, all cpu_clock() does is enable/
disable local irq during the call to sched_clock_cpu().  And for stable
systems they are same.

trace_clock_global() already disables interrupts, so it can call
sched_clock_cpu() directly.

Link: http://lkml.kernel.org/r/1356576585-28782-2-git-send-email-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

ring-buffer: Add stats field for amount read from trace ring buffer

Add a stat about the number of events read from the ring buffer:

 #  cat /debug/tracing/per_cpu/cpu0/stats
entries: 39869
overrun: 870512
commit overrun: 0
bytes: 1449912
oldest event ts:  6561.368690
now ts:  6565.246426
dropped events: 0
read events: 112    <-- Added

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

perf header: Fix double fclose() on do_write(fd, xxx) failure

cppcheck reported:
[util/header.c:983]: (error) Used file that is not opened.

Thanks to Arnaldo Carvalho de Melo for pointing out that
fclose(NULL) is undefined behavior -> protect against it.

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Link: http://lkml.kernel.org/r/1751778.SZQB4fNdIh@storm
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf header: Fix memory leak for the "Not caching a kptr_restrict'ed /proc/kallsyms" case

cppcheck reported:
[util/header.c:316]: (error) Memory leak: filename
[util/header.c:316]: (error) Memory leak: linkname

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Link: http://lkml.kernel.org/r/9377388.0eFDp53iW6@storm
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix memory leak on error

cppcheck reported:
[util/event.c:480]: (error) Memory leak: event

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Link: http://lkml.kernel.org/r/2717013.8dV0naNhAV@storm
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf sort: Use pclose() instead of fclose() on pipe stream

cppcheck message:
[tools/perf/util/sort.c:277]: (error) Mismatching allocation and deallocation: fp

Also fix descriptor leak on error and always initialize the "fp" variable.

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Link: http://lkml.kernel.org/r/1359112354.yZcisNZ4k0@storm
Link: http://lkml.kernel.org/r/2266358.qvDXKLvJ67@storm
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix possible double free on error

Can only be triggered via CROSS_COMPILE env var.

Detected by cppcheck.

Signed-off-by: Thomas Jarosch <thomas.jarosch@intra2net.com>
Link: http://lkml.kernel.org/r/36736865.AIlztKhDqN@storm
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools, powerpc: Fix compile warnings in tests/attr.c

We print several '__u64' quantities using '%llu'. On powerpc, we by
default include '<asm-generic/int-l64.h> which results in __u64 being an
unsigned long. This causes compile warnings which are treated as errors
due to '-Werror'.

By defining __SANE_USERSPACE_TYPES__ we include <asm-generic/int-ll64.h>
and define __u64 as unsigned long long.

Changelog[v2]:
[Michael Ellerman] Use __SANE_USERSPACE_TYPES__ and avoid PRIu64
format specifier - which as Jiri Olsa pointed out, breaks on x86-64.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Anton Blanchard <anton@au1.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Michael Ellerman <ellerman@au1.ibm.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20130124054439.GA31588@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evsel: Fix memory leaks on evsel->counts

The ->counts field was never freed in the current code.  Add
perf_evsel__free_counts() function to free it properly.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1359078284-32080-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Add interval printing

This patch adds a new printing mode for perf stat.  It allows interval
printing. That means perf stat can now print event deltas at regular
time interval.  This is useful to detect phases in programs.

The -I option enables interval printing. It expects an interval duration
in milliseconds. Minimum is 100ms. Once, activated perf stat prints
events deltas since last printout. All modes are supported.

$ perf stat -I 1000 -e cycles noploop 10
noploop for 10 seconds
 #           time             counts events
      1.000109853      2,388,560,546 cycles
      2.000262846      2,393,332,358 cycles
      3.000354131      2,393,176,537 cycles
      4.000439503      2,393,203,790 cycles
      5.000527075      2,393,167,675 cycles
      6.000609052      2,393,203,670 cycles
      7.000691082      2,393,175,678 cycles

The output format makes it easy to feed into a plotting program such as
gnuplot when the -I option is used in combination with the -x option:

$ perf stat -x, -I 1000 -e cycles noploop 10
noploop for 10 seconds
1.000084113,2378775498,cycles
2.000245798,2391056897,cycles
3.000354445,2392089414,cycles
4.000459115,2390936603,cycles
5.000565341,2392108173,cycles

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1359460064-3060-3-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evsel: Add prev_raw_count field

This field will be used by commands which print counter deltas on
regular timer intervals, such as perf stat -I.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1359460064-3060-2-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Make numa benchmark optional

Commit "perf: Add 'perf bench numa mem'..." added a NUMA performance
benchmark to perf. Make this optional and test for required
dependencies.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359337882-21821-1-git-send-email-peter@hurleysoftware.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf: Add 'perf bench numa mem' NUMA performance measurement suite

Add a suite of NUMA performance benchmarks.

The goal was simulate the behavior and access patterns of real NUMA
workloads, via a wide range of parameters, so this tool goes well
beyond simple bzero() measurements that most NUMA micro-benchmarks use:

 - It processes the data and creates a chain of data dependencies,
   like a real workload would. Neither the compiler, nor the
   kernel (via KSM and other optimizations) nor the CPU can
   eliminate parts of the workload.

 - It randomizes the initial state and also randomizes the target
   addresses of the processing - it's not a simple forward scan
   of addresses.

 - It provides flexible options to set process, thread and memory
   relationship information: -G sets "global" memory shared between
   all test processes, -P sets "process" memory shared by all
   threads of a process and -T sets "thread" private memory.

 - There's a NUMA convergence monitoring and convergence latency
   measurement option via -c and -m.

 - Micro-sleeps and synchronization can be injected to provoke lock
   contention and scheduling, via the -u and -S options. This simulates
   IO and contention.

 - The -x option instructs the workload to 'perturb' itself artificially
   every N seconds, by moving to the first and last CPU of the system
   periodically. This way the stability of convergence equilibrium and
   the number of steps taken for the scheduler to reach equilibrium again
   can be measured.

 - The amount of work can be specified via the -l loop count, and/or
   via a -s seconds-timeout value.

 - CPU and node memory binding options, to test hard binding scenarios.
   THP can be turned on and off via madvise() calls.

 - Live reporting of convergence progress in an 'at glance' output format.
   Printing of convergence and deconvergence events.

The 'perf bench numa mem -a' option will start an array of about 30
individual tests that will each output such measurements:

 # Running  5x5-bw-thread, "perf bench numa mem -p 5 -t 5 -P 512 -s 20 -zZ0q --thp  1"
  5x5-bw-thread,                         20.276, secs,           runtime-max/thread
  5x5-bw-thread,                         20.004, secs,           runtime-min/thread
  5x5-bw-thread,                         20.155, secs,           runtime-avg/thread
  5x5-bw-thread,                          0.671, %,              spread-runtime/thread
  5x5-bw-thread,                         21.153, GB,             data/thread
  5x5-bw-thread,                        528.818, GB,             data-total
  5x5-bw-thread,                          0.959, nsecs,          runtime/byte/thread
  5x5-bw-thread,                          1.043, GB/sec,         thread-speed
  5x5-bw-thread,                         26.081, GB/sec,         total-speed

See the help text and the code for more details.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

tools: Correct typo in tools Makefile

It should be

make -C tools/ <tool>_install

Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Ingo Molnar <mingo@kernel.org>
Link: http://lkml.kernel.org/r/1359456492-22156-1-git-send-email-bp@alien8.de
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Fix leaks on PERF_RECORD_* test

This test:

 7: Validate PERF_RECORD_* events & perf_sample fields

needs to call perf_evlist__delete_maps().

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-t3181qy15avffdacqjcxfku2@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tracing/fgraph: Adjust fgraph depth before calling trace return callback

While debugging the virtual cputime with the function graph tracer
with a max_depth of 1 (most common use of the max_depth so far),
I found that I was missing kernel execution because of a race condition.

The code for the return side of the function has a slight race:

ftrace_pop_return_trace(&trace, &ret, frame_pointer);
trace.rettime = trace_clock_local();
ftrace_graph_return(&trace);
barrier();
current->curr_ret_stack--;

The ftrace_pop_return_trace() initializes the trace structure for
the callback. The ftrace_graph_return() uses the trace structure
for its own use as that structure is on the stack and is local
to this function. Then the curr_ret_stack is decremented which
is what the trace.depth is set to.

If an interrupt comes in after the ftrace_graph_return() but
before the curr_ret_stack, then the called function will get
a depth of 2. If max_depth is set to 1 this function will be
ignored.

The problem is that the trace has already been called, and the
timestamp for that trace will not reflect the time the function
was about to re-enter userspace. Calls to the interrupt will not
be traced because the max_depth has prevented this.

To solve this issue, the ftrace_graph_return() can safely be
moved after the current->curr_ret_stack has been updated.
This way the timestamp for the return callback will reflect
the actual time.

If an interrupt comes in after the curr_ret_stack update and
ftrace_graph_return(), it will be traced. It may look a little
confusing to see it within the other function, but at least
it will not be lost.

Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>