perf script: Remove lots of unused arguments

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf thread: Rename perf_event__preprocess_sample_addr to thread__resolve

Since none of the perf_event fields are used anymore, just the
perf_sample ones, and since this resolves to (map, symbol) from data
structures within struct thread, rename it to thread__resolve and make
the argument ordering similar to the one in machine__resolve().

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-2b33hs9bp550tezzlhl4kejh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf machine: Rename perf_event__preprocess_sample to machine__resolve

Since we only deal with fields in the passed struct perf_sample move
this method to struct machine, that is where the perf_sample fields
will be resolved to a struct addr_location, i.e. thread, map, symbol,
etc.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-a1ww2lbm2vbuqsv4p7ilubu9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add cpumode to struct perf_sample

To avoid parsing event->header.misc in many locations.

This will also allow setting perf.sample.{ip,cpumode} in a single place,
from tracepoint fields, as needed by 'perf kvm' with PPC guests, where
the guest hardware counters is not available at the host.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-qp3yradhyt6q3wl895b1aat0@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Forward the perf_sample in the dwarf unwind test

It _will_ be used, no sense in receiving it and nor fowarding it along.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-ht8v5et209wuoh5o6nh9pzyq@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Remove misplaced __maybe_unused

All over the tree.

Cc: David Ahern <dsahern@gmail.com>
cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/n/tip-8nzhnokxyp8y4v7gf0j00oyb@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Fix documentation of :ppp

Correctly document what is implemented for :ppp on Intel CPUs in recent
kernels.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1458575793-12091-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bench numa: Fix assertion for nodes bitfield

Comparing bits and bytes in numa benchmark assertion

I hit the issue on two socket Power8 machine presenting its numa nodes
as 0,1,16,17 (according to numactl). Therefore I got error (and hang of
parent process):

    perf: bench/numa.c:296: bind_to_memnode: Assertion `!(g->p.nr_nodes > (int)sizeof(nodemask))' failed.

This is obviously false positive. We can fit all the 18 nodes into
bitfield of 8 bytes (long on 64b architecture).

Signed-off-by: Jakub Jelen <jakuje@gmail.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jakub Jelen <jjelen@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: trivial@kernel.org
Link: http://lkml.kernel.org/r/1458388687-24421-1-git-send-email-jakuje@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf symbols: Record text offset in dso to calculate objdump address

Store DSO's .text offset into DSO, used for VDSOs and will also be used for
other needs, like handling kernel modules.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Cody P Schafer <dev@codyps.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kirill Smelkov <kirr@nexedi.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-2-git-send-email-wangnan0@huawei.com
[ Extracted from larger patch ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools: Move utilities.mak from perf to tools/scripts/

As it is used by several other tools, better move it outside tools/perf.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-34s9kue3xq9w5mijdmfrfx8s@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf test: Remove 'core_id' check in topo test

The topology test case of 'perf test' seems to be broken on my x86
system - due to the comparison of a "core-id" with # of CPUs online.

There are 8 online CPUs:

$ cat /sys/devices/system/cpu/online
0-7

but core-ids are not sequential and some core-ids exceed the number
of online CPUs.

$ cat /sys/devices/system/cpu/cpu?/topology/core_id
0
1
9
10
0
1
9
10

Looks like we can safely remove the check.  Output before:

$ perf --version
perf version 4.4.rc1.g34258a

$ perf test -v topo
36: Test topology in session                                 :
--- start ---
test child forked, pid 5906
templ file: /tmp/perf-test-vCwWG3
core_id number is too big.You may need to upgrade the perf tool.
test child interrupted
---- end ----
Test topology in session: FAILED!

and after:

$ perf test -v topo
36: Test topology in session                                 :
--- start ---
test child forked, pid 6532
templ file: /tmp/perf-test-y10wFJ
CPU 0, core 0, socket 0
CPU 1, core 1, socket 0
CPU 2, core 9, socket 0
CPU 3, core 10, socket 0
CPU 4, core 0, socket 1
CPU 5, core 1, socket 1
CPU 6, core 9, socket 1
CPU 7, core 10, socket 1
test child finished with 0
---- end ----
Test topology in session: Ok

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Jan Stancek <jstancek@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Link: http://lkml.kernel.org/r/20151203233219.GA27696@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-core-for-mingo-20160310' of git://git./linux/kernel/git/acme/linux into perf/core

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

User visible changes:

  - Implement 'perf stat --metric-only' (Andi Kleen)

  - Fix perf script python database export crash (Chris Phlipot)

Infrastructure changes:

  - perf top/report --hierarchy assorted fixes for problems introduced in this
    perf/core cycle (Namhyung Kim)

  - Support '~' operation in libtraceevent (Steven Rosted)

Build fixes:

  - Fix bulding of jitdump on opensuse on ubuntu systems when the DWARF
    devel files are not installed (Arnaldo Carvalho de Melo)

  - Do not try building jitdump on unsupported arches (Jiri Olsa)

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

perf stat: Add --metric-only support for -A

Add metric only support for -A too. This requires a new print function
that prints the metrics in the right order.

v2: Fix manpage
v3: Simplify nrcpus computation

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1457049458-28956-7-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Implement --metric-only mode

Add a new mode to only print metrics. Sometimes we don't care about the
raw values, just want the computed metrics. This allows more compact
printing, so with -I each sample is only a single line.  This also
allows easier plotting and processing with other tools.

The main target is with using --topdown, but it also works with -T and
standard perf stat. A few metrics are not supported.

To avoiding having to hardcode all the metrics in the code it uses a two
pass approach: first compute dummy metrics and only print the headers in
the print_metric callback. Then use the callback to print the actual
values.

There are some additional changes in the stat printout code to handle
all metrics being on a single line.

One issue is that the column code doesn't know in advance what events
are not supported by the CPU, and it would be hard to find out as this
could change based on dynamic conditions. That causes empty columns in
some cases.

The output can be fairly wide, often you may need more than 80 columns.

Example:

% perf stat -a -I 1000 --metric-only
     1.001452803 frontend cycles idle insn per cycle       stalled cycles per insn branch-misses of all branches
     1.001452803  158.91%               0.66                2.39                    2.92%
     2.002192321  180.63%               0.76                2.08                    2.96%
     3.003088282  150.59%               0.62                2.57                    2.84%
     4.004369835  196.20%               0.98                1.62                    3.79%
     5.005227314  231.98%               0.84                1.90                    4.71%

v2: Lots of updates.
v3: Use slightly narrower columns
v4: Add comment

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1457049458-28956-6-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Document CSV format in manpage

With all the recently added fields in the perf stat CSV output we should
finally document them in the man page. Do this here.

v2: Fix fields in documentation (Jiri)
v3: fix order of fields again (Jiri)
v4: Change order again.
v5: Document more fields (Jiri)
v6: Move time stamp first
v7: More fixes (Jiri)

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1457049458-28956-5-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Check sort keys before hot key actions

The context menu in TUI hists browser checks corresponding sort keys
when creating the menu item.  But hotkey actions lacks these checks so
it can filter using incorrect info.

For example, default sort key of 'perf top' doesn't contain 'comm' or
'pid' sort key so each hist entry's thread info is not reliable.  Thus
it should prohibit using thread filter on 't' key.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1457533253-21419-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Allow thread filtering for comm sort key

The commit 2eafd410e669 ("perf hists browser: Only 'Zoom into thread'
only when sort order has 'pid'") disabled thread filtering in hist
browser for the default sort key.  However the he->thread is still valid
even if 'pid' sort key is not given.  Only thing it should not use is
the pid (or tid) of the thread.  So allow to filter by thread when
'comm' sort key is given and show pid only if 'pid' sort key is given.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1457536490-24084-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add sort__has_comm variable

The sort__has_comm variable is to check whether the comm sort key is
given.  This is necessary to support thread filtering in the TUI hists
browser later.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1457533253-21419-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Recalc total periods using top-level entries in hierarchy

When hierarchy mode is enabled, each entry in a hierarchy level shares
the period.  IOW an upper level entry's period is the sum of lower level
entries.  Thus perf uses only one of them to calculate the total period
of hists.  It was lowest-level (leaf) entries but it has a problem when
it comes to filters.

If a filter is applied, entries in the same level will be filtered or
not.  But upper level entries still have period of their sum including
filtered one.  So total sum of upper level entries will not be same as
sum of lower level entries.

This resulted in entries having more than 100% of overhead and it can be
produced using perf top with filter(s).

Reported-and-Tested-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-8-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Remove nr_sort_keys field

The nr_sort_keys field is to carry the number of sort entries in a
hpp_list or hists to determine the depth of indentation of a hist entry.
As it's only used in hierarchy mode and now we have used nr_hpp_node for
this reason, there's no need to keep it anymore.  Let's get rid of it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-7-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Cleanup hist_browser__fprintf_hierarchy_entry()

The hist_browser__fprintf_hierarchy_entry() if to dump current output
into a file so it needs to be sync-ed with the corresponding function
hist_browser__show_hierarchy_entry().  So use hists->nr_hpp_node to
indent width and use first fmt_node to print overhead columns instead of
checking whether it's a sort entry (or dynamic entry).

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-6-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Remove hist_entry->fmt field

It's not used anymore and the output format is accessed by the hpp_list
pointer instead when hierarchy is enabled.  Let's get rid of it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-5-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix command line filters in hierarchy mode

When a command-line filter is applied in hierarchy mode, output is
broken especially when filtering on lower level.  The higher level
entries doesn't show up so it's hard to see the results.

Also it needs to handle multi sort keys in a single hierarchy level.

Before:

  $ perf report --hierarchy -s 'cpu,{dso,comm}' --comms swapper --stdio
  ...
  #    Overhead  CPU / Shared Object+Command
  # ...........  ...........................
  #
         13.79%     [kernel.vmlinux]  swapper
      31.71%     000
         13.80%     [kernel.vmlinux]  swapper
          0.43%     [e1000e]          swapper
         11.89%     [kernel.vmlinux]  swapper
          9.18%     [kernel.vmlinux]  swapper

After:

  #    Overhead  CPU / Shared Object+Command
  # ...........  ...............................
  #
      33.09%     003
         13.79%     [kernel.vmlinux]  swapper
      31.71%     000
         13.80%     [kernel.vmlinux]  swapper
          0.43%     [e1000e]          swapper
      21.90%     002
         11.89%     [kernel.vmlinux]  swapper
      13.30%     001
          9.18%     [kernel.vmlinux]  swapper

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-4-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add more sort entry check functions

Those functions are for checkinf if a given perf_hpp_fmt is a
filter-related sort entry.  With hierarchy mode, it needs to check
filters on the hist entries with its own hpp format list.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix hist_entry__filter() for hierarchy

When hierarchy mode is enabled each output format is in a separate hpp
list.  So when applying a filter it should check all formats in the
list.  Currently it only checks a single ->fmt field which was not set
properly.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457531222-18130-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf jitdump: Build only on supported archs

Build jitdump only on architectures defined in util/genelf.h file, to avoid
breaking the build on such arches.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Borislav Petkov <bp@suse.de>
Cc: Colin Ian King <colin.king@canonical.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Davidlohr Bueso <dbueso@suse.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Mel Gorman <mgorman@suse.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20160310164113.GA11357@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Add '~' operation within arg_num_eval()

When evaluating values for print flags, if the value included a '~'
operator, the parsing would fail. This broke kmalloc's parsing of:

__print_flags(REC->gfp_flags, "|", {(unsigned
long)((((((( gfp_t)(0x400000u|0x2000000u)) | (( gfp_t)0x40u) |
(( gfp_t)0x80u) | (( gfp_t)0x20000u)) | (( gfp_t)0x02u)) |
(( gfp_t)0x08u)) | (( gfp_t)0x4000u) | (( gfp_t)0x10000u) |
(( gfp_t)0x1000u) | (( gfp_t)0x200u)) & ~(( gfp_t)0x2000000u))
                                        ^
                                        |
                                      here

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: David Ahern <dsahern@gmail.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20160226181328.22f47129@gandalf.local.home
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Omit unnecessary cast in perf_pmu__parse_scale

There's no need to use a const char pointer, we can used char pointer
from the beginning and omit the unnecessary cast.

Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20160308184230.GB7897@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Pass perf_hpp_list all the way through setup_sort_list

Pass perf_hpp_list all the way through setup_sort_list so that the sort
entry can be added on the arbitrary list.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20160309100417.GA30910@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix perf script python database export crash

Remove the union in evsel so that the database id and priv pointer can
be used simultainously without conflicting and crashing.

Detailed Description for the fixed bug follows:

perf script crashes with a segmentation fault on user space tool version
4.5.rc7.ge2857b when using the python database export API. It works
properly in 4.4 and prior versions.

the crash fist appeared in:

cfc8874a4859 ("perf script: Process cpu/threads maps")

How to reproduce the bug:

Remove any temporary files left over from a previous crash (if you have
already attemped to reproduce the bug):

  $ rm -r test_db-perf-data
  $ dropdb test_db

  $ perf record timeout 1 yes >/dev/null
  $ perf script -s scripts/python/export-to-postgresql.py test_db

  Stack Trace:
  Program received signal SIGSEGV, Segmentation fault.
  __GI___libc_free (mem=0x1) at malloc.c:2929
  2929 malloc.c: No such file or directory.
  (gdb) bt
    at util/stat.c:122
    argv=<optimized out>, prefix=<optimized out>) at builtin-script.c:2231
    argc=argc@entry=4, argv=argv@entry=0x7fffffffdf70) at perf.c:390
    at perf.c:451

Signed-off-by: Chris Phlipot <cphlipot0@gmail.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: cfc8874a4859 ("perf script: Process cpu/threads maps")
Link: http://lkml.kernel.org/r/1457500314-8912-1-git-send-email-cphlipot0@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf jitdump: DWARF is also needed

While building on a Docker container for ubuntu and installing package
by package one ends up with:

    MKDIR    /tmp/build/util/
    CC       /tmp/build/util/genelf.o
  util/genelf.c:22:19: fatal error: dwarf.h: No such file or directory
   #include <dwarf.h>
                   ^
  compilation terminated.
  mv: cannot stat '/tmp/build/util/.genelf.o.tmp': No such file or directory

Because the jitdump code needs the DWARF related development packages to
be installed. So make it dependent on that so that the build can succeed
without jitdump support.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-le498robnmxd40237wej3w62@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bench mem: Prepare the x86-64 build for upstream memcpy_mcsafe() changes

The following upcoming upstream commit:

  92b0729c34ca ("x86/mm, x86/mce: Add memcpy_mcsafe()")

Adds _ASM_EXTABLE_FAULT(), which is not available in user-space
and breaks the build.

We don't really need _ASM_EXTABLE_FAULT() in user-space, so simply
wrap it to nothing.

Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Simplify quirk handling even more

Drop the quirk() function pointer in favor of a simple boolean which
says whether the quirk should be applied or not. Update comment while at
it.

Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-tip-commits@vger.kernel.org
Link: http://lkml.kernel.org/r/20160308164041.GF16568@pd.tnic
Signed-off-by: Ingo Molnar <mingo@kernel.org>

x86/nmi: Mark 'ignore_nmis' as __read_mostly

ignore_nmis is used in two distinct places:

 1. modified through {stop,restart}_nmi by alternative_instructions
 2. read by do_nmi to determine if default_do_nmi should be called or not

thus the access pattern conforms to __read_mostly and do_nmi() is a fastpath.

Signed-off-by: Kostenzer Felix <fkostenzer@live.at>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix PEBS data source interpretation on Nehalem/Westmere

Jiri reported some time ago that some entries in the PEBS data source table
in perf do not agree with the SDM. We investigated and the bits
changed for Sandy Bridge, but the SDM was not updated.

perf already implements the bits correctly for Sandy Bridge
and later. This patch patches it up for Nehalem and Westmere.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: jolsa@kernel.org
Link: http://lkml.kernel.org/r/1456871124-15985-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/pebs: Add proper PEBS constraints for Broadwell

This patch adds a Broadwell specific PEBS event constraint table.

Broadwell has a fix for the HT corruption bug erratum HSD29 on
Haswell. Therefore, there is no need to mark events 0xd0, 0xd1, 0xd2,
0xd3 has requiring the exclusive mode across both sibling HT threads.
This holds true for regular counting and sampling (see core.c) and
PEBS (ds.c) which we fix in this patch.

In doing so, we relax evnt scheduling for these events, they can now
be programmed on any 4 counters without impacting what is measured on
the sibling thread.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: acme@redhat.com
Cc: adrian.hunter@intel.com
Cc: jolsa@redhat.com
Cc: kan.liang@intel.com
Cc: namhyung@kernel.org
Link: http://lkml.kernel.org/r/1457034642-21837-4-git-send-email-eranian@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/pebs: Add workaround for broken OVFL status on HSW+

This patch fixes an issue with the GLOBAL_OVERFLOW_STATUS bits on
Haswell, Broadwell and Skylake processors when using PEBS.

The SDM stipulates that when the PEBS iterrupt threshold is crossed,
an interrupt is posted and the kernel is interrupted. The kernel will
find GLOBAL_OVF_SATUS bit 62 set indicating there are PEBS records to
drain. But the bits corresponding to the actual counters should NOT be
set. The kernel follows the SDM and assumes that all PEBS events are
processed in the drain_pebs() callback. The kernel then checks for
remaining overflows on any other (non-PEBS) events and processes these
in the for_each_bit_set(&status) loop.

As it turns out, under certain conditions on HSW and later processors,
on PEBS buffer interrupt, bit 62 is set but the counter bits may be
set as well. In that case, the kernel drains PEBS and generates
SAMPLES with the EXACT tag, then it processes the counter bits, and
generates normal (non-EXACT) SAMPLES.

I ran into this problem by trying to understand why on HSW sampling on
a PEBS event was sometimes returning SAMPLES without the EXACT tag.
This should not happen on user level code because HSW has the
eventing_ip which always point to the instruction that caused the
event.

The workaround in this patch simply ensures that the bits for the
counters used for PEBS events are cleared after the PEBS buffer has
been drained. With this fix 100% of the PEBS samples on my user code
report the EXACT tag.

Before:
  $ perf record -e cpu/event=0xd0,umask=0x81/upp ./multichase
  $ perf report -D | fgrep SAMPLES
  PERF_RECORD_SAMPLE(IP, 0x2): 11775/11775: 0x406de5 period: 73469 addr: 0 exact=Y
                           \--- EXACT tag is missing

After:
  $ perf record -e cpu/event=0xd0,umask=0x81/upp ./multichase
  $ perf report -D | fgrep SAMPLES
  PERF_RECORD_SAMPLE(IP, 0x4002): 11775/11775: 0x406de5 period: 73469 addr: 0 exact=Y
                           \--- EXACT tag is set

The problem tends to appear more often when multiple PEBS events are used.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: adrian.hunter@intel.com
Cc: kan.liang@intel.com
Cc: namhyung@kernel.org
Link: http://lkml.kernel.org/r/1457034642-21837-3-git-send-email-eranian@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Add definition for PT PMI bit

This patch adds a definition for GLOBAL_OVFL_STATUS bit 55
which is used with the Processor Trace (PT) feature.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: adrian.hunter@intel.com
Cc: kan.liang@intel.com
Cc: namhyung@kernel.org
Link: http://lkml.kernel.org/r/1457034642-21837-2-git-send-email-eranian@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix PEBS warning by only restoring active PMU in pmi

This patch tries to fix a PEBS warning found in my stress test. The
following perf command can easily trigger the pebs warning or spurious
NMI error on Skylake/Broadwell/Haswell platforms:

  sudo perf record -e 'cpu/umask=0x04,event=0xc4/pp,cycles,branches,ref-cycles,cache-misses,cache-references' --call-graph fp -b -c1000 -a

Also the NMI watchdog must be enabled.

For this case, the events number is larger than counter number. So
perf has to do multiplexing.

In perf_mux_hrtimer_handler, it does perf_pmu_disable(), schedule out
old events, rotate_ctx, schedule in new events and finally
perf_pmu_enable().

If the old events include precise event, the MSR_IA32_PEBS_ENABLE
should be cleared when perf_pmu_disable().  The MSR_IA32_PEBS_ENABLE
should keep 0 until the perf_pmu_enable() is called and the new event is
precise event.

However, there is a corner case which could restore PEBS_ENABLE to
stale value during the above period. In perf_pmu_disable(), GLOBAL_CTRL
will be set to 0 to stop overflow and followed PMI. But there may be
pending PMI from an earlier overflow, which cannot be stopped. So even
GLOBAL_CTRL is cleared, the kernel still be possible to get PMI. At
the end of the PMI handler, __intel_pmu_enable_all() will be called,
which will restore the stale values if old events haven't scheduled
out.

Once the stale pebs value is set, it's impossible to be corrected if
the new events are non-precise. Because the pebs_enabled will be set
to 0. x86_pmu.enable_all() will ignore the MSR_IA32_PEBS_ENABLE
setting. As a result, the following NMI with stale PEBS_ENABLE
trigger pebs warning.

The pending PMI after enabled=0 will become harmless if the NMI handler
does not change the state. This patch checks cpuc->enabled in pmi and
only restore the state when PMU is active.

Here is the dump:

  Call Trace:
   <NMI>  [<ffffffff813c3a2e>] dump_stack+0x63/0x85
   [<ffffffff810a46f2>] warn_slowpath_common+0x82/0xc0
   [<ffffffff810a483a>] warn_slowpath_null+0x1a/0x20
   [<ffffffff8100fe2e>] intel_pmu_drain_pebs_nhm+0x2be/0x320
   [<ffffffff8100caa9>] intel_pmu_handle_irq+0x279/0x460
   [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
   [<ffffffff811f290d>] ? vunmap_page_range+0x20d/0x330
   [<ffffffff811f2f11>] ?  unmap_kernel_range_noflush+0x11/0x20
   [<ffffffff8148379f>] ? ghes_copy_tofrom_phys+0x10f/0x2a0
   [<ffffffff814839c8>] ? ghes_read_estatus+0x98/0x170
   [<ffffffff81005a7d>] perf_event_nmi_handler+0x2d/0x50
   [<ffffffff810310b9>] nmi_handle+0x69/0x120
   [<ffffffff810316f6>] default_do_nmi+0xe6/0x100
   [<ffffffff810317f2>] do_nmi+0xe2/0x130
   [<ffffffff817aea71>] end_repeat_nmi+0x1a/0x1e
   [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
   [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
   [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
   <<EOE>>  <IRQ>  [<ffffffff81006df8>] ?  x86_perf_event_set_period+0xd8/0x180
   [<ffffffff81006eec>] x86_pmu_start+0x4c/0x100
   [<ffffffff8100722d>] x86_pmu_enable+0x28d/0x300
   [<ffffffff811994d7>] perf_pmu_enable.part.81+0x7/0x10
   [<ffffffff8119cb70>] perf_mux_hrtimer_handler+0x200/0x280
   [<ffffffff8119c970>] ?  __perf_install_in_context+0xc0/0xc0
   [<ffffffff8110f92d>] __hrtimer_run_queues+0xfd/0x280
   [<ffffffff811100d8>] hrtimer_interrupt+0xa8/0x190
   [<ffffffff81199080>] ?  __perf_read_group_add.part.61+0x1a0/0x1a0
   [<ffffffff81051bd8>] local_apic_timer_interrupt+0x38/0x60
   [<ffffffff817af01d>] smp_apic_timer_interrupt+0x3d/0x50
   [<ffffffff817ad15c>] apic_timer_interrupt+0x8c/0xa0
   <EOI>  [<ffffffff81199080>] ?  __perf_read_group_add.part.61+0x1a0/0x1a0
   [<ffffffff81123de5>] ?  smp_call_function_single+0xd5/0x130
   [<ffffffff81123ddb>] ?  smp_call_function_single+0xcb/0x130
   [<ffffffff81199080>] ?  __perf_read_group_add.part.61+0x1a0/0x1a0
   [<ffffffff8119765a>] event_function_call+0x10a/0x120
   [<ffffffff8119c660>] ? ctx_resched+0x90/0x90
   [<ffffffff811971e0>] ? cpu_clock_event_read+0x30/0x30
   [<ffffffff811976d0>] ? _perf_event_disable+0x60/0x60
   [<ffffffff8119772b>] _perf_event_enable+0x5b/0x70
   [<ffffffff81197388>] perf_event_for_each_child+0x38/0xa0
   [<ffffffff811976d0>] ? _perf_event_disable+0x60/0x60
   [<ffffffff811a0ffd>] perf_ioctl+0x12d/0x3c0
   [<ffffffff8134d855>] ? selinux_file_ioctl+0x95/0x1e0
   [<ffffffff8124a3a1>] do_vfs_ioctl+0xa1/0x5a0
   [<ffffffff81036d29>] ? sched_clock+0x9/0x10
   [<ffffffff8124a919>] SyS_ioctl+0x79/0x90
   [<ffffffff817ac4b2>] entry_SYSCALL_64_fastpath+0x1a/0xa4
  ---[ end trace aef202839fe9a71d ]---
  Uhhuh. NMI received for unknown reason 2d on CPU 2.
  Do you have a strange power saving mode enabled?

Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1457046448-6184-1-git-send-email-kan.liang@intel.com
[ Fixed various typos and other small details. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Use PAGE_SIZE for PEBS buffer size on Core2

Using PAGE_SIZE buffers makes the WRMSR to PERF_GLOBAL_CTRL in
intel_pmu_enable_all() mysteriously hang on Core2. As a workaround, we
don't do this.

The hard lockup is easily triggered by running 'perf test attr'
repeatedly. Most of the time it gets stuck on sample session with
small periods.

  # perf test attr -vv
  14: struct perf_event_attr setup                             :
  --- start ---
  ...
    'PERF_TEST_ATTR=/tmp/tmpuEKz3B /usr/bin/perf record -o /tmp/tmpuEKz3B/perf.data -c 123 kill >/dev/null 2>&1' ret 1

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: <stable@vger.kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20160301190352.GA8355@krava.redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/core: Fix perf_sched_count derailment

The error path in perf_event_open() is such that asking for a sampling
event on a PMU that doesn't generate interrupts will end up in dropping
the perf_sched_count even though it hasn't been incremented for this
event yet.

Given a sufficient amount of these calls, we'll end up disabling
scheduler's jump label even though we'd still have active events in the
system, thereby facilitating the arrival of the infernal regions upon us.

I'm fixing this by moving account_event() inside perf_event_alloc().

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: vince@deater.net
Link: http://lkml.kernel.org/r/1456917854-29427-1-git-send-email-alexander.shishkin@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'email/acme' into perf/core

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

    User visible changes:

 - Allow grouping multiple sort keys per 'perf report/top --hierarchy'
   level (Namhyung Kim)

 - Document 'perf stat --detailed' option (Borislav Petkov)

Infrastructure changes:

 - jitdump prep work for supporting it with Intel PT (Adrian Hunter)

 - Use 64-bit shifts with (TSC) time conversion (Adrian Hunter)

Fixes:

 - Explicitly declare inc_group_count as a void function (Colin Ian King)

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

perf report: Use hierarchy hpp list on gtk

Now hpp formats are linked using perf_hpp_list_node when hierarchy is
enabled.  Like in stdio, use this info to print entries with multiple
sort keys in a single hierarchy properly.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-8-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists browser: Use hierarchy hpp list

Now hpp formats are linked using perf_hpp_list_node when hierarchy is
enabled.  Like in stdio, use this info to print entries with multiple
sort keys in a single hierarchy properly.

Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-7-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf report: Use hierarchy hpp list on stdio

Now hpp formats are linked using perf_hpp_list_node when hierarchy is
enabled.  Use this info to print entries with multiple sort keys in a
single hierarchy properly.

For example, the below example shows using 4 sort keys with 2 levels.

  $ perf report --hierarchy -s '{prev_pid,prev_comm},{next_pid,next_comm}' \
   --percent-limit 1 -i perf.data.sched
  ...
  #    Overhead  prev_pid+prev_comm / next_pid+next_comm
  # ...........  .......................................
  #
      22.36%     0  swapper/0
          9.48%     17773  transmission-gt
          5.25%     109  kworker/0:1H
          1.53%     6524  Xephyr
      21.39%     17773  transmission-gt
          9.52%     0  swapper/0
          9.04%     0  swapper/2
          1.78%     0  swapper/3

Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-6-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists: Fix indent for multiple hierarchy sort key

When multiple sort keys are used in a single hierarchy, it should indent
using number of hierarchy levels instead of number of sort keys.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-5-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists: Support multiple sort keys in a hierarchy level

This implements having multiple sort keys in a single hierarchy level.
Originally only single sort key is supported for each level, but now
using the group syntax with '{ }', it can set more than one sort key in
one level.  Note that now it needs to quote in order to prevent shell
interpretation.

For example:

  $ perf report --hierarchy -s '{comm,dso},sym'
  ...
  #       Overhead  Command / Shared Object / Symbol
  # ..............  ..........................................
  #
      48.67%        swapper          [kernel.vmlinux]
         34.42%        [k] intel_idle
          1.30%        [k] __tick_nohz_idle_enter
          1.03%        [k] cpuidle_reflect
       8.87%        firefox          libpthread-2.22.so
          6.60%        [.] __GI___libc_recvmsg
          1.18%        [.] pthread_cond_signal@@GLIBC_2.3.2
          1.09%        [.] 0x000000000000ff4b
       6.11%        Xorg             libc-2.22.so
          5.27%        [.] __memcpy_sse2_unaligned

In the above example, the command name and the shared object name are
shown on the same line but the symbol name is on the different line.
Since the first two are grouped by '{}', they are in the same level.

Suggested-and-Tested=by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-4-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists: Use own hpp_list for hierarchy mode

Now each hists has its own hpp lists in hierarchy.  So instead of having
a pointer to a single perf_hpp_fmt in a hist entry, make it point the
hpp_list for its level.  This will be used to support multiple sort keys
in a single hierarchy level.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-3-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists: Introduce perf_hpp__setup_hists_formats()

The perf_hpp__setup_hists_formats() is to build hists-specific output
formats (and sort keys).  Currently it's only used in order to build the
output format in a hierarchy with same sort keys, but it could be used
with different sort keys in non-hierarchy mode later.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457361308-514-2-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf stat: Document --detailed option

I'm surprised this remained undocumented since at least 2011. And it is
actually a very useful switch, as Steve and I came to realize recently.

Add the text from

  2cba3ffb9a9d ("perf stat: Add -d -d and -d -d -d options to show more CPU events")

which added the incrementing aspect to -d.

Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Davidlohr Bueso <dbueso@suse.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mel Gorman <mgorman@suse.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 2cba3ffb9a9d ("perf stat: Add -d -d and -d -d -d options to show more CPU events")
Link: http://lkml.kernel.org/r/1457347294-32546-1-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf hists: Add level field to struct perf_hpp_fmt

The level field is to distinguish levels in the hierarchy mode.
Currently each column (perf_hpp_fmt) has a different level.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457103582-28396-2-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Use 64-bit shifts with (TSC) time conversion

Commit b9511cd761fa ("perf/x86: Fix time_shift in perf_event_mmap_page")
altered the time conversion algorithms documented in the perf_event.h
header file, to use 64-bit shifts.  That was done to make the code more
future-proof (i.e. some time in the future a 32-bit shift could be
allowed).  Reflect those changes in perf tools.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1457005856-6143-9-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf jit: Move clockid validation

Move clockid validation into jit_process() so it can later be made
conditional.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1457005856-6143-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf jit: Let jit_process() return errors

In preparation for moving clockid validation into jit_process().

Previously a return value of zero meant the processing had been done and
non-zero meant either the processing was not done (i.e. not the jitdump
file mmap event) or an error occurred.

Change it so that zero means the processing was not done, one means the
processing was done and successful, and negative values are an error.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1457005856-6143-5-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf session: Simplify tool stubs

Some of the stubs are identical so just have one function for them.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1457005856-6143-3-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf inject: Hit all DSOs for AUX data in JIT and other cases

Currently, when injecting build ids, if there is AUX data then 'perf
inject' hits all DSOs because it is not known which DSOs the trace data
would hit.

That needs to be done for JIT injection also, and in fact there is no
reason to distinguish what kind of injection is being done.  That is,
any time there is AUX data and the HEADER_BUID_ID feature flag is set,
and the AUX data is not being processed, then hit all DSOs.  This patch
does that.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1457005856-6143-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Explicitly declare inc_group_count as a void function

The return type is not defined, so it defaults to int, however, the
function is not returning anything, so this is clearly not correct. Make
it a void function.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1457008214-14393-1-git-send-email-colin.king@canonical.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'perf-core-for-mingo-20160303' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes:

User visible changes:

 - Check existence of frontend/backed stalled cycles in 'perf stat' (Andi Kleen)

 - Implement CSV metrics output in 'perf stat' (Andi Kleen)

 - Support metrics in 'perf stat' --per-core/socket mode (Andi Kleen)

 - Avoid installing .o files from tools/lib/ into the python extension (Jiri Olsa)

 - Rename the tracepoint '/format' field that carries the syscall ID from 'nr',
   that is also the name of some syscalls arguments, to "__syscall_nr", to
   avoid having multiple fields with the same name, that was breaking the
   python script skeleton generator from perf.data files (Taeung Song)

 - Support converting data from bpf events in 'perf data' (Wang Nan)

 - Fix segfault in 'perf test' hists related entries (Arnaldo Carvalho de Melo)

 - Fix output of %llu for 64 bit values read on 32 bit machines in libtraceevent (Steven Rostedt)

 - Fix time stamp rounding issue in libtraceevent (Chaos.Chen)

Infrastructure changes:

 - Fix setlocale() breakage in the pmu parsing code (Jiri Olsa)

 - Split libtraceevent's pevent_print_event() (Steven Rostedt)

 - Librarize some 'perf record' bits to allow handling multiple perf.data
   files per session (Wang Nan)

 - Ensure return non-zero rc when mmap fails in 'perf record' (Wang Nan)

 - Fix double free on 'command_line' in a error path in 'perf script' (Colin Ian King)

 - Initialize struct sigaction 'sa_flags' field in a 'perf test' entry (Colin Ian King)

 - Fix various build warnings in turbostat, detected with gcc6 (Colin Ian King)

 - Use .s extension for preprocessed assembler code (Masahiro Yamada)

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

perf stat: Check for frontend stalled for metrics

Add an extra check for frontend stalled in the metrics.  This avoids an
extra column for the --metric-only case when the CPU does not support
frontend stalled.

v2: Add separate init function

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1456858672-21594-8-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/power turbostat: fix various build warnings

When building with gcc 6 we're getting various build warnings that just
require some trivial function declaration and call fixes:

  turbostat.c: In function ‘dump_cstate_pstate_config_info’:
  turbostat.c:1973:1: warning: type of ‘family’ defaults to ‘int’
   dump_cstate_pstate_config_info(family, model)
  turbostat.c:1973:1: warning: type of ‘model’ defaults to ‘int’
  turbostat.c: In function ‘get_tdp’:
  turbostat.c:2145:8: warning: type of ‘model’ defaults to ‘int’
   double get_tdp(model)
  turbostat.c: In function ‘perf_limit_reasons_probe’:
  turbostat.c:2259:6: warning: type of ‘family’ defaults to ‘int’
   void perf_limit_reasons_probe(family, model)
  turbostat.c:2259:6: warning: type of ‘model’ defaults to ‘int’

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-wbicer8n0s9qe6ql8h9x478e@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Initialize sa.sa_flags

The sa_flags field is not being initialized, so a garbage value is being
passed to sigaction.  Initialize it to zero.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1456923322-29697-1-git-send-email-colin.king@canonical.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf test: Fix hists related entries

That got broken by d3a72fd8187b ("perf report: Fix indentation of
dynamic entries in hierarchy"), by using the evlist in setup_sorting()
without checking if it is NULL, as done in some 'perf test' entries:

  $ find tools/ -name "*.c" | xargs grep 'setup_sorting(NULL);'
  tools/perf/tests/hists_output.c:      setup_sorting(NULL);
  tools/perf/tests/hists_output.c:      setup_sorting(NULL);
  tools/perf/tests/hists_output.c:      setup_sorting(NULL);
  tools/perf/tests/hists_output.c:      setup_sorting(NULL);
  tools/perf/tests/hists_output.c:      setup_sorting(NULL);
  tools/perf/tests/hists_cumulate.c:    setup_sorting(NULL);
  tools/perf/tests/hists_cumulate.c:    setup_sorting(NULL);
  tools/perf/tests/hists_cumulate.c:    setup_sorting(NULL);
  tools/perf/tests/hists_cumulate.c:    setup_sorting(NULL);
  $

Fix it.

Before:

  [root@jouet ~]# perf test
  <SNIP>
  15: Test matching and linking multiple hists                 : FAILED!
  16: Try 'import perf' in python, checking link problems      : Ok
  17: Test breakpoint overflow signal handler                  : Ok
  18: Test breakpoint overflow sampling                        : Ok
  19: Test number of exit event of a simple workload           : Ok
  20: Test software clock events have valid period values      : Ok
  21: Test object code reading                                 : Ok
  22: Test sample parsing                                      : Ok
  23: Test using a dummy software event to keep tracking       : Ok
  24: Test parsing with no sample_id_all bit set               : Ok
  25: Test filtering hist entries                              : FAILED!
  26: Test mmap thread lookup                                  : Ok
  27: Test thread mg sharing                                   : Ok
  28: Test output sorting of hist entries                      : FAILED!
  29: Test cumulation of child hist entries                    : FAILED!
  <SNIP>

After the patch the above failed tests complete successfully.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: d3a72fd8187b ("perf report: Fix indentation of dynamic entries in hierarchy")
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Fix output of %llu for 64 bit values read on 32 bit machines

When a long value is read on 32 bit machines for 64 bit output, the
parsing needs to change "%lu" into "%llu", as the value is read
natively.

Unfortunately, if "%llu" is already there, the code will add another "l"
to it and fail to parse it properly.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20160209204237.337024613@goodmis.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Set int_array fields to NULL if freeing from error

Had a bug where on error of parsing __print_array() where the fields are
freed after they were allocated, but since they were not set to NULL,
the freeing of the arg also tried to free the already freed fields
causing a double free.

Fix process_hex() while at it.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20160209204237.188327674@goodmis.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Fix time stamp rounding issue

When rounding to microseconds, if the timestamp subsecond is between
.999999500 and .999999999, it is rounded to .1000000, when it should
instead increment the second counter due to the overflow.

For example, if the timestamp is 1234.999999501 instead of seeing:

  1235.000000

we see:

  1234.1000000

Signed-off-by: Chaos.Chen <rainboy1215@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20160209204236.824426460@goodmis.org
[ fixed incrementing "secs" instead of decrementing it ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Fix double free on command_line

The 'command_line' variable is free'd twice if db_export__branch_types()
fails. To avoid this, defer the free'ing of 'command_line' to after this
call so that the error return path will just free 'command_line' once.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Javi Merino <javi.merino@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1456875980-25606-1-git-send-email-colin.king@canonical.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Use .s extension for preprocessed assembler code

The "man gcc" says .i extension represents the file is C source code
that should not be preprocessed.  Here, .s should be used.

For clarification,
  .c  ---(preprocess)--->  .i
  .S  ---(preprocess)--->  .s

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Aaro Koskinen <aaro.koskinen@nokia.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Lukas Wunner <lukas@wunner.de>
Link: http://lkml.kernel.org/r/1454263140-19670-1-git-send-email-yamada.masahiro@socionext.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Support metrics in --per-core/socket mode

Enable metrics printing in --per-core / --per-socket mode. We need to
save the shadow metrics in a unique place. Always use the first CPU in
the aggregation. Then use the same CPU to retrieve the shadow value
later.

Example output:

  % perf stat --per-core -a ./BC1s

   Performance counter stats for 'system wide':

  S0-C0 2   2966.020381 task-clock (msec) #   2.004 CPUs utilized  (100.00%)
  S0-C0 2            49 context-switches  #   0.017 K/sec          (100.00%)
  S0-C0 2             4 cpu-migrations    #   0.001 K/sec          (100.00%)
  S0-C0 2           467 page-faults       #   0.157 K/sec
  S0-C0 2 4,599,061,773 cycles            #   1.551 GHz            (100.00%)
  S0-C0 2 9,755,886,883 instructions      #   2.12  insn per cycle (100.00%)
  S0-C0 2 1,906,272,125 branches          # 642.704 M/sec          (100.00%)
  S0-C0 2    81,180,867 branch-misses     #   4.26% of all branches
  S0-C1 2   2965.995373 task-clock (msec) #   2.003 CPUs utilized  (100.00%)
  S0-C1 2            62 context-switches  #   0.021 K/sec          (100.00%)
  S0-C1 2             8 cpu-migrations    #   0.003 K/sec          (100.00%)
  S0-C1 2           281 page-faults       #   0.095 K/sec
  S0-C1 2     6,347,290 cycles            #   0.002 GHz            (100.00%)
  S0-C1 2     4,654,156 instructions      #   0.73  insn per cycle (100.00%)
  S0-C1 2       947,121 branches          #   0.319 M/sec          (100.00%)
  S0-C1 2        37,322 branch-misses     #   3.94% of all branches

         1.480409747 seconds time elapsed

v2: Rebase to older patches
v3: Document shadow cpus. Fix aggr_get_id argument. Fix -A shadows (Jiri)

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1456785386-19481-4-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Implement CSV metrics output

Now support CSV output for metrics. With the new output callbacks this
is relatively straight forward by creating new callbacks.

This allows to easily plot metrics from CSV files.

The new line callback needs to know the number of fields to skip them
correctly

Example output before:

  % perf stat -x, true
  0.200687,,task-clock,200687,100.00
  0,,context-switches,200687,100.00
  0,,cpu-migrations,200687,100.00
  40,,page-faults,200687,100.00
  730871,,cycles,203601,100.00
  551056,,stalled-cycles-frontend,203601,100.00
  <not supported>,,stalled-cycles-backend,0,100.00
  385523,,instructions,203601,100.00
  78028,,branches,203601,100.00
  3946,,branch-misses,203601,100.00

After:

  % perf stat -x, true
  .502457,,task-clock,502457,100.00,0.485,CPUs utilized
  0,,context-switches,502457,100.00,0.000,K/sec
  0,,cpu-migrations,502457,100.00,0.000,K/sec
  45,,page-faults,502457,100.00,0.090,M/sec
  644692,,cycles,509102,100.00,1.283,GHz
  423470,,stalled-cycles-frontend,509102,100.00,65.69,frontend cycles idle
  <not supported>,,stalled-cycles-backend,0,100.00,,,,
  492701,,instructions,509102,100.00,0.76,insn per cycle
  ,,,,,0.86,stalled cycles per insn
  97767,,branches,509102,100.00,194.578,M/sec
  4788,,branch-misses,509102,100.00,4.90,of all branches

or easier readable

  $ perf stat  -x, -o x.csv true
  $ column -s, -t x.csv
  0.490635        task-clock              490635 100.00 0.489   CPUs utilized
  0               context-switches        490635 100.00 0.000   K/sec
  0               cpu-migrations          490635 100.00 0.000   K/sec
  45              page-faults             490635 100.00 0.092   M/sec
  629080          cycles                  497698 100.00 1.282   GHz
  409498          stalled-cycles-frontend 497698 100.00 65.09   frontend cycles idle
  <not supported> stalled-cycles-backend  0      100.00
  491424          instructions            497698 100.00 0.78    insn per cycle
                                                        0.83    stalled cycles per insn
  97278           branches                497698 100.00 198.270 M/sec
  4569            branch-misses           497698 100.00 4.70    of all branches

Two new fields are added: metric value and metric name.

v2: Split out function argument changes
v3: Reenable metrics for real.
v4: Fix wrong hunk from refactoring.
v5: Remove extra "noise" printing (Jiri), but add it to the not counted case.
Print empty metrics for not counted.
v6: Avoid outputting metric on empty format.
v7: Print metric at the end
v8: Remove extra run, ena fields
v9: Avoid extra new line for unsupported counters

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/1456785386-19481-3-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Ensure return non-zero rc when mmap fail

perf_evlist__mmap_ex() can fail without setting errno (for example, fail
in condition checking. In this case all syscall is success).

If this happen, record__open() incorrectly returns 0. Force setting rc
is a quick way to avoid this problem, or we have to follow all possible
code path in perf_evlist__mmap_ex() to make sure there's at least one
system call before returning an error.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-30-git-send-email-wangnan0@huawei.com
Signed-off-by: He Kuang <hekuang@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Introduce record__finish_output() to finish a perf.data

Move code for finalizing 'perf.data' to record__finish_output(). It will
be used by following commits to split output to multiple files.

Signed-off-by: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-23-git-send-email-wangnan0@huawei.com
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Extract synthesize code to record__synthesize()

Create record__synthesize(). It can be used to create tracking events
for each perf.data after perf supporting splitting into multiple
outputs.

Signed-off-by: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-20-git-send-email-wangnan0@huawei.com
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Use WARN_ONCE to replace 'if' condition

Commits in a BPF patchkit will extract kernel and module synthesizing
code into a separated function and call it multiple times. This patch
replace 'if (err < 0)' using WARN_ONCE, makes sure the error message
show one time.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-19-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf data: Explicitly set byte order for integer types

After babeltrace commit 5cec03e402aa ("ir: copy variants and sequences
when setting a field path"), 'perf data convert' gets incorrect result
if there's bpf output data. For example:

 # perf data convert --to-ctf ./out.ctf
 # babeltrace ./out.ctf
 [10:44:31.186045346] (+?.?????????) evt: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF810E7DD1, perf_tid = 23819, perf_pid = 23819, perf_id = 518, raw_len = 3, raw_data = [ [0] = 0xC028E32F, [1] = 0x815D0100, [2] = 0x1000000 ] }
 [10:44:31.286101003] (+0.100055657) evt: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF8105B609, perf_tid = 23819, perf_pid = 23819, perf_id = 518, raw_len = 3, raw_data = [ [0] = 0x35D9F1EB, [1] = 0x15D81, [2] = 0x2 ] }

The expected result of the first sample should be:

 raw_data = [ [0] = 0x2FE328C0, [1] = 0x15D81, [2] = 0x1 ] }

however, 'perf data convert' output big endian value to resuling CTF
file.

The reason is a internal change (or a bug?) of babeltrace.

Before this patch, at the first add_bpf_output_values(), byte order of
all integer type is uncertain (is 0, neither 1234 (le) nor 4321 (be)).
It would be fixed by:

perf_evlist__deliver_sample
 -> process_sample_event
   -> ctf_stream
      ...
      ->bt_ctf_trace_add_stream_class
        ->bt_ctf_field_type_structure_set_byte_order
          ->bt_ctf_field_type_integer_set_byte_order

during creating the stream.

However, the babeltrace commit mentioned above duplicates types in
sequence to prevent potential conflict in following call stack and link
the newly allocated type into the 'raw_data' sequence:

perf_evlist__deliver_sample
 -> process_sample_event
   -> ctf_stream
      ...
      -> bt_ctf_trace_add_stream_class
        -> bt_ctf_stream_class_resolve_types
           ...
           -> bt_ctf_field_type_sequence_copy
             ->bt_ctf_field_type_integer_copy

This happens before byte order setting, so only the newly allocated
type is initialized, the byte order of original type perf choose to
create the first raw_data is still uncertain.

Byte order in CTF output is not related to byte order in perf.data.
Setting it to anything other than BT_CTF_BYTE_ORDER_NATIVE solves this
problem (only BT_CTF_BYTE_ORDER_NATIVE needs to be fixed). To reduce
behavior changing, set byte order according to compiling options.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Jeremie Galarneau <jeremie.galarneau@efficios.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-10-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf data: Support converting data from bpf_perf_event_output()

bpf_perf_event_output() outputs data through sample->raw_data. This
patch adds support to convert those data into CTF. A python script then
can be used to process output data from BPF programs.

Test result:

  # cat ./test_bpf_output_2.c
  /************************ BEGIN **************************/
  #include <uapi/linux/bpf.h>
  struct bpf_map_def {
  unsigned int type;
  unsigned int key_size;
  unsigned int value_size;
  unsigned int max_entries;
  };
  #define SEC(NAME) __attribute__((section(NAME), used))
  static u64 (*ktime_get_ns)(void) =
  (void *)BPF_FUNC_ktime_get_ns;
  static int (*trace_printk)(const char *fmt, int fmt_size, ...) =
  (void *)BPF_FUNC_trace_printk;
  static int (*get_smp_processor_id)(void) =
  (void *)BPF_FUNC_get_smp_processor_id;
  static int (*perf_event_output)(void *, struct bpf_map_def *, int, void *, unsigned long) =
  (void *)BPF_FUNC_perf_event_output;

  struct bpf_map_def SEC("maps") channel = {
  .type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
  .key_size = sizeof(int),
  .value_size = sizeof(u32),
  .max_entries = __NR_CPUS__,
  };

  static inline int __attribute__((always_inline))
  func(void *ctx, int type)
  {
  struct {
  u64 ktime;
  int type;
  } __attribute__((packed)) output_data;
  char error_data[] = "Error: failed to output\n";
  int err;

  output_data.type = type;
  output_data.ktime = ktime_get_ns();
  err = perf_event_output(ctx, &channel, get_smp_processor_id(),
  &output_data, sizeof(output_data));
  if (err)
  trace_printk(error_data, sizeof(error_data));
  return 0;
  }
  SEC("func_begin=sys_nanosleep")
  int func_begin(void *ctx) {return func(ctx, 1);}
  SEC("func_end=sys_nanosleep%return")
  int func_end(void *ctx) { return func(ctx, 2);}
  char _license[] SEC("license") = "GPL";
  int _version SEC("version") = LINUX_VERSION_CODE;
  /************************* END ***************************/

  # ./perf record -e bpf-output/no-inherit,name=evt/ \
                 -e ./test_bpf_output_2.c/map:channel.event=evt/ \
                 usleep 100000
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.012 MB perf.data (2 samples) ]

  # ./perf script
          usleep 14942 92503.198504: evt:  ffffffff810e0ba1 sys_nanosleep (/lib/modules/4.3.0....
          usleep 14942 92503.298562: evt:  ffffffff810585e9 kretprobe_trampoline_holder (/lib....

  # ./perf data convert --to-ctf ./out.ctf
  [ perf data convert: Converted 'perf.data' into CTF data './out.ctf' ]
  [ perf data convert: Converted and wrote 0.000 MB (2 samples) ]

  # babeltrace ./out.ctf
  [01:41:43.198504134] (+?.?????????) evt: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF810E0BA1, perf_tid = 14942, perf_pid = 14942, perf_id = 1044, raw_len = 3, raw_data = [ [0] = 0x32C0C07B, [1] = 0x5421, [2] = 0x1 ] }
  [01:41:43.298562257] (+0.100058123) evt: { cpu_id = 0 }, { perf_ip = 0xFFFFFFFF810585E9, perf_tid = 14942, perf_pid = 14942, perf_id = 1044, raw_len = 3, raw_data = [ [0] = 0x38B77FAA, [1] = 0x5421, [2] = 0x2 ] }

  # cat ./test_bpf_output_2.py
  from babeltrace import TraceCollection
  tc = TraceCollection()
  tc.add_trace('./out.ctf', 'ctf')
  d = {1:[], 2:[]}
  for event in tc.events:
     if not event.name.startswith('evt'):
         continue
     raw_data = event['raw_data']
     (time, type) = ((raw_data[0] + (raw_data[1] << 32)), raw_data[2])
     d[type].append(time)
  print(list(map(lambda i: d[2][i] - d[1][i], range(len(d[1])))));

  # python3 ./test_bpf_output_2.py
  [100056879]

Committer note:

Make sure you have python3-devel installed, not python-devel, which may
be for python2, which will lead to some "PyInstance_Type" errors. Also
make sure that you use the right libbabeltrace, because it is shipped
in Fedora, for instance, but an older version.

To build libbabeltrace's python binding one also needs to use:

 ./configure --enable-python-bindings

And then set PYTHONPATH=/usr/local/lib64/python3.4/site-packages/.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1456479154-136027-9-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Check existence of frontend/backed stalled cycles

Only put the frontend/backend stalled cycles into the default perf stat
events when the CPU actually supports them.

This avoids empty columns with --metric-only on newer Intel CPUs.

Committer note:

Before:

  $ perf stat ls

    Performance counter stats for 'ls':

          1.080893     task-clock (msec)      #    0.619 CPUs utilized
                 0     context-switches       #    0.000 K/sec
                 0     cpu-migrations         #    0.000 K/sec
                97     page-faults            #    0.090 M/sec
         3,327,741     cycles                 #    3.079 GHz
   <not supported>     stalled-cycles-frontend
   <not supported>     stalled-cycles-backend
         1,609,544     instructions           #    0.48  insn per cycle
           319,117     branches               #  295.235 M/sec
            12,246     branch-misses          #    3.84% of all branches

       0.001746508 seconds time elapsed
  $

After:

  $ perf stat ls

    Performance counter stats for 'ls':

          0.693948     task-clock (msec)      #    0.662 CPUs utilized
                 0     context-switches       #    0.000 K/sec
                 0     cpu-migrations         #    0.000 K/sec
                95     page-faults            #    0.137 M/sec
         1,792,509     cycles                 #    2.583 GHz
         1,599,047     instructions           #    0.89  insn per cycle
           316,328     branches               #  455.838 M/sec
            12,453     branch-misses          #    3.94% of all branches

       0.001048987 seconds time elapsed
  $

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1456532881-26621-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix locale handling in pmu parsing

Ingo reported regression on display format of big numbers, which is
missing separators (in default perf stat output).

 triton:~/tip> perf stat -a sleep 1
         ...
         127008602      cycles                    #    0.011 GHz
         279538533      stalled-cycles-frontend   #  220.09% frontend cycles idle
         119213269      instructions              #    0.94  insn per cycle

This is caused by recent change:

  perf stat: Check existence of frontend/backed stalled cycles

that added call to pmu_have_event, that subsequently calls
perf_pmu__parse_scale, which has a bug in locale handling.

The lc string returned from setlocale, that we use to store old locale
value, may be allocated in static storage. Getting a dynamic copy to
make it survive another setlocale call.

  $ perf stat ls
         ...
         2,360,602      cycles                    #    3.080 GHz
         2,703,090      instructions              #    1.15  insn per cycle
           546,031      branches                  #  712.511 M/sec

Committer note:

Since the patch introducing the regression didn't made to perf/core,
move it to just before where the regression was introduced, so that we
don't break bisection for this feature.

Reported-by: Ingo Molnar <mingo@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20160303095348.GA24511@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/x86/uncore: Fix build on UP-IOAPIC configs

Commit:

  cf6d445f6897 ("perf/x86/uncore: Track packages, not per CPU data")

reorganized the uncore code to track packages, and introduced a dependency
on MAX_APIC_ID. This constant is not available on UP-IOAPIC builds:

  arch/x86/events/intel/uncore.c:1350:44: error: 'MAX_LOCAL_APIC' undeclared here (not in a function)

Include asm/apicdef.h explicitly to pick it up.

Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

tools lib traceevent: Split pevent_print_event() into specific functionality functions

Currently there's a single function that is used to display a record's
data in human readable format. That's pevent_print_event().
Unfortunately, this gives little room for adding other output within the
line without updating that function call.

I've decided to split that function into 3 parts.

 pevent_print_event_task() which prints the task comm, pid and the CPU
 pevent_print_event_time() which outputs the record's timestamp
 pevent_print_event_data() which outputs the rest of the event data.

pevent_print_event() now simply calls these three functions.

To save time from doing the search for event from the record's type, I
created a new helper function called pevent_find_event_by_record(),
which returns the record's event, and this event has to be passed to the
above functions.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20160229090128.43a56704@gandalf.local.home
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tracing/syscalls: Rename "/format" tracepoint field name "nr" to "__syscall_nr:

Some tracepoint have multiple fields with the same name, "nr", the first
one is a unique syscall ID, the other is a syscall argument:

  # cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_io_getevents/format
  name: sys_enter_io_getevents
  ID: 747
  format:
  field:unsigned short common_type; offset:0; size:2; signed:0;
  field:unsigned char common_flags; offset:2; size:1; signed:0;
  field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
  field:int common_pid; offset:4; size:4; signed:1;

  field:int nr; offset:8; size:4; signed:1;
  field:aio_context_t ctx_id; offset:16; size:8; signed:0;
  field:long min_nr; offset:24; size:8; signed:0;
  field:long nr; offset:32; size:8; signed:0;
  field:struct io_event * events; offset:40; size:8; signed:0;
  field:struct timespec * timeout; offset:48; size:8; signed:0;

  print fmt: "ctx_id: 0x%08lx, min_nr: 0x%08lx, nr: 0x%08lx, events: 0x%08lx, timeout: 0x%08lx", ((unsigned long)(REC->ctx_id)), ((unsigned long)(REC->min_nr)), ((unsigned long)(REC->nr)), ((unsigned long)(REC->events)), ((unsigned long)(REC->timeout))
  #

Fix it by renaming the "/format" common tracepoint field "nr" to "__syscall_nr".

Signed-off-by: Taeung Song <treeze.taeung@gmail.com>
[ Do not rename the struct member, just the '/format' field name ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20160226132301.3ae065a4@gandalf.local.home
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Check and discard not only 'nr' but also '__syscall_nr'

Format fields of a syscall have the first variable '__syscall_nr' or
'nr' that mean the syscall number.  But it isn't relevant here so drop
it.

'nr' among fields of syscall was renamed '__syscall_nr'.  So add
exception handling to drop '__syscall_nr' and modify the comment for
this excpetion handling.

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Taeung Song <treeze.taeung@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1456492465-5946-1-git-send-email-treeze.taeung@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix python extension build

The util/python-ext-sources file contains source files required to build
the python extension relative to $(srctree)/tools/perf,

Such a file path $(FILE).c is handed over to the python extension build
system, which builds the final object in the
$(PYTHON_EXTBUILD)/tmp/$(FILE).o path.

After the build is done all files from $(PYTHON_EXTBUILD)lib/ are
carried as the result binaries.

Above system fails when we add source file relative to ../lib, which we
do for:

  ../lib/bitmap.c
  ../lib/find_bit.c
  ../lib/hweight.c
  ../lib/rbtree.c

All above objects will be built like:

  $(PYTHON_EXTBUILD)/tmp/../lib/bitmap.c
  $(PYTHON_EXTBUILD)/tmp/../lib/find_bit.c
  $(PYTHON_EXTBUILD)/tmp/../lib/hweight.c
  $(PYTHON_EXTBUILD)/tmp/../lib/rbtree.c

which accidentally happens to be final library path:

  $(PYTHON_EXTBUILD)/lib/

Changing setup.py to pass full paths of source files to Extension build
class and thus keep all built objects under $(PYTHON_EXTBUILD)tmp
directory.

Reported-by: Jeff Bastian <jbastian@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Josh Boyer <jwboyer@fedoraproject.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: stable@vger.kernel.org # v4.2+
Link: http://lkml.kernel.org/r/20160227201350.GB28494@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf: Export perf_event_sysfs_show()

Required to use it in modular perf drivers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.930735780@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Convert it to a per package facility

RAPL is a per package facility and we already have a mechanism for a dedicated
per package reader. So there is no point to have multiple CPUs doing the
same. The current implementation actually starts two timers on two CPUs if one
does:

perf stat -C1,2 -e -e power/energy-pkg ....

which makes the whole concept of 1 reader per package moot.

What's worse is that the above returns the double of the actual energy
consumption, but that's a different problem to address and cannot be solved by
removing the pointless per cpuness of that mechanism.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.845369524@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Utilize event->pmu_private

Store the PMU pointer in event->pmu_private and use it instead of the per CPU
data. Preparatory step to get rid of the per CPU allocations. The usage sites
are the perf fast path, so we keep that even after the conversion to per
package storage as a CPU to package lookup involves 3 loads versus 1 with the
pmu_private pointer.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.748151799@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Make PMU lock raw

This lock is taken in hard interrupt context even on Preempt-RT. Make it raw
so RT does not have to patch it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.669411833@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Refactor the code some more

Split out code from init into seperate functions. Tidy up the code and get rid
of pointless comments. I wish there would be comments for code which is not
obvious....

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.588544679@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Clean up the printk output

The output is inconsistent. Use a proper pr_fmt prefix and split out the
advertisement into a seperate function.

Remove the WARN_ON() in the failure case. It's pointless as we already know
where it failed.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.504551295@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Calculate timing once

No point in doing the same calculation over and over. Do it once in
rapl_check_hw_unit().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.409238136@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Sanitize the quirk handling

There is no point in having a quirk machinery for a single possible
function. Get rid of it and move the quirk to a place where it actually
makes sense.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.311639465@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Add proper error handling

Like uncore the rapl driver lacks error handling. It leaks memory and leaves
the hotplug notifier registered.

Add the proper error checks, cleanup the memory and register the hotplug
notifier only on success.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.231222076@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Make Knights Landings support functional

The Knights Landings support added the events and the detection case, but then
returns 0 without actually initializing the driver.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dasaratharaman Chandramouli <dasaratharaman.chandramouli@intel.com>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Fixes: 3a2a7797326a4 "perf/x86/intel/rapl: Add support for Knights Landing (KNL)"
Link: http://lkml.kernel.org/r/20160222221012.149331888@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/cqm: Get rid of the silly for_each_cpu() lookups

CQM is a strict per package facility. Use the proper cpumasks to lookup the
readers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221012.054916179@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

cpumask: Export cpumask_any_but()

Almost every cpumask function is exported, just not the one I need to make the
Intel uncore driver modular.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.878299859@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Make PCI and MSR uncore independent

Andi wanted to do this before, but the patch fell down the cracks. Implement
it with the proper error handling.

Requested-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.799159968@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Clear all hardware state on exit

The only missing bit is to completely clear the hardware state on failure
exit. This is now a pretty simple exercise.

Undo the box->init_box() setup on all packages which have been initialized so
far.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.702452407@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/uncore: Track packages, not per CPU data

Uncore is a per package facility, but the code tries to mimick a per CPU
facility with completely convoluted constructs.

Simplify the whole machinery by tracking per package information. While at it,
avoid the kfree/alloc dance when a CPU goes offline and online again. There is
no point in freeing the box after it was allocated. We just keep proper
refcounting and the first CPU which comes online in a package does the
initialization/activation of the box.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.622258933@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

x86/topology: Create logical package id

For per package oriented services we must be able to rely on the number of CPU
packages to be within bounds. Create a tracking facility, which

- calculates the number of possible packages depending on nr_cpu_ids after boot

- makes sure that the package id is within the number of possible packages. If
  the apic id is outside we map it to a logical package id if there is enough
  space available.

Provide interfaces for drivers to query the mapping and do translations from
physcial to logical ids.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Luis R. Rodriguez <mcgrof@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.541071755@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Store box in event->pmu_private

Store the PMU pointer in event->pmu_private, so we can get rid of the
per CPU data storage.

We keep it after converting to per package data, because a CPU to
package lookup will be 3 loads versus one and these usage sites are
in the perf fast path.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.460851335@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf: Allow storage of PMU private data in event

For PMUs which are not per CPU, but e.g. per package/socket, we want to be
able to store a reference to the underlying per package/socket facility in the
event at init time so we can avoid magic storage constructs in the PMU driver.

This allows us to get rid of the per CPU dance in the intel uncore and RAPL
drivers and avoids a lookup of the per package data in the perf hotpath.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Harish Chegondi <harish.chegondi@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20160222221011.364140369@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>