perf tools xtensa: Add DWARF register names

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Marc Gauthier <marc@cadence.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: linux-xtensa@linux-xtensa.org
Link: http://lkml.kernel.org/r/1437208216-15729-9-git-send-email-jcmvbkbc@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Display cycles in branch sort mode

Display the cycles by default in branch sort mode.

To make enough room for the new column I removed dso_to. It is usually
redundant with dso_from.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-9-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Add branch annotation code to top

Now that we can process branch data in annotate it makes sense to
support enabling branch recording from top too. Most of the code needed
for this is already in shared code with report. But we need to add:

- The option parsing code (using shared code from the previous patch)
- Document the options
- Set up the IPC/cycles accounting state in the top session
- Call the accounting code in the hist iter callback

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-8-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Finally display IPC and cycle accounting

Add two new columns to the annotate display and display the average
cycles and the compute IPC if available.

When the LBR was not in any branch mode the IPC computation is
automatically disabled. We still display the cycle information.

Example output (with made up numbers):

The second column is the IPC and third average cycles.

                 │    __attribute__((noinline)) f2()
                 │    {
  5.15  0.07     │       push   %rbp
  0.01  0.07     │       mov    %rsp,%rbp
                 │            c = a / b;
  9.87  0.07     │       mov    a,%eax
        0.07     │       mov    b,%ecx
        0.07     │       cltd
  4.92  0.07  123│       idiv   %ecx
 70.79  0.07     │       mov    %eax,__TMC_END__
                 │    }
  9.25  0.07     │       pop    %rbp
  0.01  0.07  123│     ← retq

v2: Fix display problems.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-7-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Compute IPC and basic block cycles

Compute the IPC and the basic block cycles for the annotate display.

IPC is computed by counting the instructions, and then dividing the
accounted cycles by that count.

The actual IPC computation can only be done at annotate time, because we
need to parse the objdump output first to know the number of
instructions in the basic block.

The cycles/IPC are also put into the perf function annotation so that
the display code can show them.

Again basic block overlaps are not handled, with the longest winning,
but there are some heuristics to hide the IPC when the longest is not
the most common.

v2: Compute IPC correctly.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-6-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add processing for cycle histograms

Call the earlier added cycle histogram infrastructure from the perf
report hist iter callback. For this we walk the branch records.

This allows to use cycle histograms when browsing perf report annotate.

v2: Rename flag

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-5-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add infrastructure for a cycles histogram

This adds the basic infrastructure to keep track of cycle counts per
basic block for annotate. We allocate an array similar to the normal
accounting, and then account branch cycles there.

We handle two cases:

cycles per basic block with start and cycles per branch (these are later
used for either IPC or just cycles per BB)

In the start case we cannot handle overlaps, so always the longest basic
block wins.

For the cycles per branch case everything is accurately accounted.

v2: Remove unnecessary checks. Slight restructure. Move
symbol__get_annotation to another patch. Move histogram allocation.
v3: Merged with current tree

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-4-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Add flag for non ANY branch mode

Later patches need to cheaply check that the branch mode is in ANY.  Add
a new function to check all event attrs and add a flag to the report
state, which is then initialized.

v2: Rename flag

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-3-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add support for cycles, weight branch_info field

cycles is a new branch_info field available on some CPUs that indicates
the time deltas between branches in the LBR.

Add a sort key and output code for the cycles to allow to display the
basic block cycles individually in perf report.

We also pass in the cycles for weight when LBRs are processed, which
allows to get global and local weight, to get an estimate of the total
cost.

And also print the cycles information for perf report -D.  I also added
printing for the previously missing LBR flags (mispredict etc.)

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1437233094-12844-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add empty Build files for architectures lacking them

perf currently fails to build on MIPS as there is no
tools/perf/arch/mips/Build file.  Adding an empty file fixes this as
there are no MIPS-specific sources to build.

It looks like the same is needed for Alpha and PA-RISC, though I
haven't been able to test those.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Fixes: 5e8c0fb6a957 ("perf build: Add arch x86 objects building")
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1438704627.7315.2.camel@decadent.org.uk
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Move counter processing code into stat object

Moving counter processing code into stat object as
perf_stat__process_counter.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-8-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Pass 'struct perf_stat_config' into process_counter()

Passing 'struct perf_stat_config' into process_counter(), so that we can
make process_counter() non static and use it from other places.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-7-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Move 'interval' into struct perf_stat_config

Moving 'interval' into struct perf_stat_config. The point is to
centralize the base stat config so it could be used localy together with
other stat routines in other parts of perf code.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-6-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Move 'output' into struct perf_stat_config

Moving 'output' into struct perf_stat_config. The point is to centralize
the base stat config so it could be used localy together with other stat
routines in other parts of perf code.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-5-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Move 'scale' into struct perf_stat_config

Moving 'scale' into struct perf_stat_config. The point is to centralize
the base stat config so it could be used localy together with other stat
routines in other parts of perf code.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-4-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Introduce struct perf_stat_config

Moving 'aggr_mode' into new struct. The point is to centralize the base
stat config so it could be used localy together with other stat routines
in other parts of perf code.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
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/1437481927-29538-3-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add missing forward declaration of struct map to probe-event.h

Commit 7b6ff0bdbf4f7f429c2116cca92a6d171217449e ("perf probe ppc64le:
Fixup function entry if using kallsyms lookup") adds 'struct map' into
probe-event.h but not forward declares it. This patch fixes it.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Fixes: 7b6ff0bdbf4f ("perf probe ppc64le: Fixup function entry if using kallsyms lookup")
Link: http://lkml.kernel.org/n/1436445342-1402-30-git-send-email-wangnan0@huawei.com
[ No need to include map.h, just forward declare 'struct map' ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Introduce veprintf

va_args alternative to eprintf().

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/n/1436445342-1402-19-git-send-email-wangnan0@huawei.com
[ split from another patch ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tracing, perf: Implement BPF programs attached to uprobes

By copying BPF related operation to uprobe processing path, this patch
allow users attach BPF programs to uprobes like what they are already
doing on kprobes.

After this patch, users are allowed to use PERF_EVENT_IOC_SET_BPF on a
uprobe perf event. Which make it possible to profile user space programs
and kernel events together using BPF.

Because of this patch, CONFIG_BPF_EVENTS should be selected by
CONFIG_UPROBE_EVENT to ensure trace_call_bpf() is compiled even if
KPROBE_EVENT is not set.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

bpf: Use correct #ifdef controller for trace_call_bpf()

Commit e1abf2cc8d5d80b41c4419368ec743ccadbb131e ("bpf: Fix the build on
BPF_SYSCALL=y && !CONFIG_TRACING kernels, make it more configurable")
updated the building condition of bpf_trace.o from CONFIG_BPF_SYSCALL
to CONFIG_BPF_EVENTS, but the corresponding #ifdef controller in
trace_events.h for trace_call_bpf() was not changed. Which, in theory,
is incorrect.

With current Kconfigs, we can create a .config with CONFIG_BPF_SYSCALL=y
and CONFIG_BPF_EVENTS=n by unselecting CONFIG_KPROBE_EVENT and
selecting CONFIG_BPF_SYSCALL. With these options, trace_call_bpf() will
be defined as an extern function, but if anyone calls it a symbol missing
error will be triggered since bpf_trace.o was not built.

This patch changes the #ifdef controller for trace_call_bpf() from
CONFIG_BPF_SYSCALL to CONFIG_BPF_EVENTS. I'll show its correctness:

Before this patch:

   BPF_SYSCALL   BPF_EVENTS   trace_call_bpf   bpf_trace.o
   y             y           normal           compiled
   n             n           inline           not compiled
   y             n           normal           not compiled (incorrect)
   n             y          impossible (BPF_EVENTS depends on BPF_SYSCALL)

After this patch:

   BPF_SYSCALL   BPF_EVENTS   trace_call_bpf   bpf_trace.o
   y             y           normal           compiled
   n             n           inline           not compiled
   y             n           inline           not compiled (fixed)
   n             y          impossible (BPF_EVENTS depends on BPF_SYSCALL)

So this patch doesn't break anything. QED.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-2-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Add total time column to summary.

It is cumbersome to manually calculate the total time spent in a given
syscall by multiplying the average value with the number of calls.

Instead, we now do this directly inside perf trace.

Note that this is also done by 'strace', which even adds a column with
relative numbers - something we could do in the future.

Example:

  perf trace -s find /some/folder > /dev/null

   Summary of events:

   find (19976), 700123 events, 100.0%, 0.000 msec

     syscall            calls    total       min       avg       max      stddev
                                 (msec)    (msec)    (msec)    (msec)        (%)
     --------------- -------- --------- --------- --------- ---------     ------
     read                   4     0.006     0.001     0.002     0.003     27.42%
     write               8046     9.617     0.001     0.001     0.035      0.56%
     open               34196    40.384     0.001     0.001     0.071      0.30%
     close              68375    57.104     0.001     0.001     0.076      0.25%
     stat                   4     0.004     0.001     0.001     0.001      3.14%
     fstat              34189    27.518     0.001     0.001     0.060      0.34%
     mmap                  13     0.029     0.001     0.002     0.003     10.74%
     mprotect               6     0.018     0.002     0.003     0.005     17.04%
     munmap                 3     0.014     0.003     0.005     0.006     24.87%
     brk                   87     0.490     0.001     0.006     0.016      6.50%
     ioctl                  3     0.004     0.001     0.001     0.003     36.39%
     access                 1     0.004     0.004     0.004     0.004      0.00%
     uname                  1     0.001     0.001     0.001     0.001      0.00%
     getdents           68393   143.600     0.001     0.002     0.187      0.95%
     fchdir             68371    56.980     0.001     0.001     0.111      0.39%
     arch_prctl             1     0.001     0.001     0.001     0.001      0.00%
     openat             34184    41.737     0.001     0.001     0.102      0.41%
     newfstatat         34184    41.180     0.001     0.001     0.064      0.34%

Signed-off-by: Milian Wolff <milian.wolff@kdab.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
LPU-Reference: 1438853069-5902-1-git-send-email-milian.wolff@kdab.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

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

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

New features:

  - Deref sys_enter pointer args with contents from probe:vfs_getname, showing
    pathnames instead of pointers in many syscalls in 'perf trace'. (Arnaldo Carvalho de Melo)

  - Make 'perf trace' write to stderr by default, just like 'strace'. (Milian Woff)

Infrastructure changes:

  - color_vfprintf() fixes. (Andi Kleen, Jiri Olsa)

  - Allow enabling/disabling PERF_SAMPLE_TIME per event. (Kan Liang)

  - Fix build errors with mipsel-linux-uclibc compiler. (Petri Gynther)

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

perf tools: Fix build errors with mipsel-linux-uclibc compiler

linux/tools$ make ARCH=mips CROSS_COMPILE=mipsel-linux- perf
...
config/Makefile:256: *** No gnu/libc-version.h found, please install
glibc-dev[el].  Stop.
make[1]: *** [all] Error 2
make: *** [perf] Error 2

...
In file included from builtin-sched.c:13:0:
util/cloexec.h:8:12: error: redundant redeclaration of ‘sched_getcpu’
 [-Werror=redundant-decls]
 extern int sched_getcpu(void) __THROW;

mipsel-buildroot-linux-uclibc/sysroot/usr/include/bits/sched.h:88:12:
 note: previous declaration of ‘sched_getcpu’ was here
 extern int sched_getcpu (void) __THROW;

uclibc info:
sysroot/usr/include/bits/uClibc_config.h
__UCLIBC_MAJOR__ 0
__UCLIBC_MINOR__ 9
__UCLIBC_SUBLEVEL__ 33

sysroot/usr/include/features.h
__UCLIBC__ 1
__GLIBC__ 2
__GLIBC_MINOR__ 2

Signed-off-by: Petri Gynther <pgynther@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1438735081-24131-1-git-send-email-pgynther@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Write to stderr by default

Without this patch, it is cumbersome to read the trace output but
ignoring the normal, potentially verbose, output of the debuggee.  One
common example is doing something like the following:

 perf trace -s find /tmp > /dev/null

Without this patch, the trace summary will be lost. Now, it will still
be printed at the end. This behavior is also applied by strace.

Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/n/tip-tqnks6y2cnvm5f9g2dsfr7zl@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Do not include escape sequences in color_vfprintf return

color_vprintf was including the length of the invisible escape sequences
in its return argument. Don't include them to make the return value
usable for indentation calculations.

v2: Add comment, rebase

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1438649408-20807-3-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Remove trail argument to color vsprintf

Seems like it's always '\n' through color_fprintf_ln, which is not used
at all, removing.. ;-)

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1438649408-20807-2-git-send-email-andi@firstfloor.org
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Refine parse/config callchain functions

Pass global callchain_param into parse_callchain_record_opt and
perf_evsel__config_callgraph as parameter. So we can reuse these
functions to parse/config local param for callchain.

Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438677022-34296-3-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Per-event time support

This patchkit adds the ability to turn off time stamps per event.

One usaful case for partial time is to work with per-event callgraph to
enable "PEBS threshold > 1" (https://lkml.org/lkml/2015/5/10/196), which
can significantly reduce the sampling overhead.

The event samples with time stamps off will not be ordered.

Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438677022-34296-2-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Use vfs_getname syscall arg beautifier in more syscalls

Those were covered and tested in this cset:

 access, chdir, chmod, chown, chroot, creat, getxattr,
 inotify_add_watch, lchown, lgetxattr, listxattr,
 lsetxattr, mkdir, mkdirat, mknod, rmdir, faccessat,
 newfstatat, openat, readlink, readlinkat, removexattr,
 setxattr, statfs, swapon, swapoff, truncate, unlinkat,
 utime, utimes, utimensat.

E.g.:

  # trace -e statfs,access,mkdir mkdir /tmp/bla
   0.285 (0.020 ms): mkdir/2799 access(filename: /etc/ld.so.preload, mode: R         ) = -1 ENOENT No such file or directory
   1.070 (0.032 ms): mkdir/2799 statfs(pathname: /sys/fs/selinux, buf: 0x7ffeafbdc930) = 0
   1.087 (0.013 ms): mkdir/2799 statfs(pathname: /sys/fs/selinux, buf: 0x7ffeafbdc820) = 0
   1.189 (0.014 ms): mkdir/2799 access(filename: /etc/selinux/config                 ) = 0
   1.905 (0.610 ms): mkdir/2799 mkdir(pathname: /tmp/bla, mode: 511                  ) = 0
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Milian Wolff <mail@milianw.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-wbqtnlktquun3wtpjdz3okul@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
  and an empty message aborts the commit.

perf trace: Deref sys_enter pointer args with contents from probe:vfs_getname

To work like strace and dereference syscall pointer args we need to
insert probes (or tracepoints) right after we copy those bytes from
userspace.

Since we're formatting the syscall args at raw_syscalls:sys_enter time,
we need to have a formatter that just stores the position where, later,
when we get the probe:vfs_getname, we can insert the pointer contents.

Now, if a probe:vfs_getname with this format is in place:

 # perf probe -l
  probe:vfs_getname (on getname_flags:72@/home/git/linux/fs/namei.c with pathname)

That was, in this case, put in place with:

 # perf probe 'vfs_getname=getname_flags:72 pathname=filename:string'
 Added new event:
  probe:vfs_getname    (on getname_flags:72 with pathname=filename:string)

 You can now use it in all perf tools, such as:

perf record -e probe:vfs_getname -aR sleep 1
 #

Then 'perf trace' will notice that and do the pointer -> contents
expansion:

 # trace -e open touch /tmp/bla
  0.165 (0.010 ms): touch/17752 open(filename: /etc/ld.so.cache, flags: CLOEXEC) = 3
  0.195 (0.011 ms): touch/17752 open(filename: /lib64/libc.so.6, flags: CLOEXEC) = 3
  0.512 (0.012 ms): touch/17752 open(filename: /usr/lib/locale/locale-archive, flags: CLOEXEC) = 3
  0.582 (0.012 ms): touch/17752 open(filename: /tmp/bla, flags: CREAT|NOCTTY|NONBLOCK|WRONLY, mode: 438) = 3
 #

Roughly equivalent to strace's output:

 # strace -rT -e open touch /tmp/bla
  0.000000 open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3 <0.000039>
  0.000317 open("/lib64/libc.so.6", O_RDONLY|O_CLOEXEC) = 3 <0.000102>
  0.001461 open("/usr/lib/locale/locale-archive", O_RDONLY|O_CLOEXEC) = 3 <0.000072>
  0.000405 open("/tmp/bla", O_WRONLY|O_CREAT|O_NOCTTY|O_NONBLOCK, 0666) = 3 <0.000055>
  0.000641 +++ exited with 0 +++
 #

Now we need to either look for at all syscalls that are marked as
pointers and have some well known names ("filename", "pathname", etc)
and set the arg formatter to the one used for the "open" syscall in this
patch.

This implementation works for syscalls with just a string being copied
from userspace, for matching syscalls with more than one string being
copied via the same probe/trace point (vfs_getname) we need to extend
the vfs_getname probe spec to include the pointer too, but there are
some problems with that in 'perf probe' or the kernel kprobes code, need
to investigate before considering supporting multiple strings per
syscall.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Milian Wolff <mail@milianw.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-xvuwx6nuj8cf389kf9s2ue2s@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Use a constant for the syscall formatting buffer

We were using it as a magic number, 1024, fix that.

Eventually we need to stop doing it per line, and do it per
arg, traversing the args at output time, to avoid the memmove()
calls that will be used in the next cset to replace pointers
present at raw_syscalls:sys_enter time with its contents that
appear at probe:vfs_getname time, before raw_syscalls:sys_exit
time.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Milian Wolff <mail@milianw.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-4sz3wid39egay1pp8qmbur4u@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Remember if the vfs_getname tracepoint/kprobe is in place

So that we can later decide if we will store where to expand the
pathname once we are handling vfs_getname or if we should instead
just go on and straight away print the pointer.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Milian Wolff <mail@milianw.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ytxk5s5jpc50wahffmlxgxuw@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Do not show syscall tracepoint filter in the --no-syscalls case

We were accessing trace->syscalls.events members even when that struct
wasn't initialized, i.e. --no-syscalls was specified on the command
line, fix it to show that, still in debug mode, when we have an event
qualifier list, i.e. when we actually are doing subset syscall tracing.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Milian Wolff <mail@milianw.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Fixes: 19867b6186f3 ("perf trace: Use event filters for the event qualifier list")
Link: http://lkml.kernel.org/n/tip-7980ym6vujgh3yiai0cqzc88@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: No tracepoints? Don't call libtraceevent.

The libtraceevent handler (session->tevent) is only initialized when
there are tracepoints in a perf.data event list, so do not call
pevent_set_function_resolve() in those cases, fixing a segfault.

Reported-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-xyynkucl5p4bcs13zi4i4b1f@git.kernel.org
link: http://lkml.kernel.org/r/20150803174113.GA20282@krava.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/x86/intel/pebs: Robustify PEBS buffer drain

Vince Weaver and Stephane Eranian reported warnings in the PEBS
code when running the perf fuzzer. Stephane wrote:

  > I can reproduce the problem on my HSW running the fuzzer.
  >
  > I can see why this could be happening if you are mixing PEBS and non PEBS events
  > in the bottom 4 counters. I suspect:
  >         for (bit = 0; bit < x86_pmu.max_pebs_events; bit++) {
  >                 if ((counts[bit] == 0) && (error[bit] == 0))
  >                         continue;
  >
  > This test is not correct when you have non-PEBS events mixed with
  > PEBS events and they overflow at the same time. They will have
  > counts[i] != 0 but error[i] == 0, and thus you fall thru the loop
  > and hit the assert. Or it is something along those lines.

The only way I can make this work is if ->status only has !PEBS events
set, because if it has both set we'll take that slow path which masks
out the !PEBS bits.

After masking there are 3 options:

 - there is one bit set, and its @bit, we increment counts[bit].

 - there are multiple bits set, we increment error[] for each set bit,
   we do not increment counts[].

 - there are no bits set, we do nothing.

The intent was to never increment counts[] for !PEBS events.

Now if we start out with only a single !PEBS event set, we'll pass the
test and increment counts[] for a !PEBS and hit the warn.

Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/pebs: Fix event disable PEBS buffer drain

When disabling a PEBS event, we need to drain the buffer. Doing so
requires a correct cpuc->pebs_active mask.

The current code clears the pebs_active bit before draining the
buffer. Fix that.

Signed-off-by: "Liang, Kan" <kan.liang@intel.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: Vince Weaver<vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/37D7C6CF3E00A74B8858931C1DB2F07701885A65@SHSMSX103.ccr.corp.intel.com
[ Fixed the SOB. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Add an MSR PMU driver

This patch adds an MSR PMU to support free running MSR counters. Such
as time and freq related counters includes TSC, IA32_APERF, IA32_MPERF
and IA32_PPERF, but also SMI_COUNT.

The events are exposed in sysfs for use by perf stat and other tools.
The files are under /sys/devices/msr/events/

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Kan Liang <kan.liang@intel.com>
[ s/freq/msr/, added SMI_COUNT, fixed bugs. ]
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@kernel.org
Cc: adrian.hunter@intel.com
Cc: dsahern@gmail.com
Cc: eranian@google.com
Cc: jolsa@kernel.org
Cc: mark.rutland@arm.com
Cc: namhyung@kernel.org
Link: http://lkml.kernel.org/r/1437407346-31186-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Add Broadwell-DE uncore support

The uncore subsystem for Broadwell-DE is similar to Haswell-EP.  There
are some differences in pci device IDs, box number and constraints.

Please refer to the public document:

  http://www.intel.com/content/www/us/en/processors/xeon/xeon-d-1500-uncore-performance-monitoring.html

Signed-off-by: Kan Liang <kan.liang@intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1435839172-15114-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Use 0x11 as extra reg test value

The next patch adds a new perf extra register where 0x1ff is not a valid
value. Use 0x11 instead.

Signed-off-by: Andi Kleen <ak@linux.intel.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>
Link: http://lkml.kernel.org/r/1435707205-6676-3-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Make merge_attr() global to use from perf_event_intel

merge_attr() allows to merge two sysfs attribute tables.
Export it to be usable by other files too.

Next patch is going to use that to extend the sysfs format
attributes for a CPU.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1435612935-24425-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Limit LBR accesses to TOS in callstack mode

In callstack mode the LBR is not a ring buffer, but a stack that grows up
and down. This means in  this case we don't need to access all LBRs, only the
ones up to TOS. Do this optimization for the normal LBR read, and the context
switch save/restore code. For save/restore it can be done unconditionally, as
it only runs when call stack mode is active.

This recovers some of the cost of going to 32 LBRs on Skylake.

Signed-off-by: Andi Kleen <ak@linux.intel.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@kernel.org
Cc: eranian@google.com
Cc: jolsa@redhat.com
Link: http://lkml.kernel.org/r/1432786398-23861-6-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Use correct index to save/restore LBR_INFO with call stack

Use the correct index to save/restore the LBR_INFO_x MSR in
callstack mode. This is more a cleanup, as even with the wrong
index the register was correctly saved/restored, and also
LBR callgraph mode in perf tools do not really need anything in
LBR_INFO. But still better to use the right index.

Signed-off-by: Andi Kleen <ak@linux.intel.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@kernel.org
Cc: eranian@google.com
Cc: jolsa@redhat.com
Link: http://lkml.kernel.org/r/1432786398-23861-5-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Add Intel Skylake PMU support

Add perf core PMU support for future Intel Skylake CPU cores.

The code is based on Haswell/Broadwell.

There is a new cache event list, based on the updated Haswell
event list.

Skylake has removed most counter constraints on basic
events, so the basic constraints table now only has a single
entry (plus the fixed counters).

TSX support and various other setups are all shared with Haswell.

Skylake has 32 LBR entries. Add a new LBR init function
to set this up. The filters are all the same as Haswell.

It also has a new LBR format with a separate LBR_INFO_* MSR,
but that has been already added earlier.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-7-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Optimize v4 LBR unfreezing

In Arch perfmon v4 the GLOBAL_STATUS reset automatically unfreezes
LBRs. So no need to do it manually in the LBR code. Add a check
to skip it.

v2: Move test up to beginning of function.
Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-9-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Move PMU ACK to after LBR read

With Arch Perfmon v4 the PMU ack unfreezes the LBRs. So we need to do
the PMU ack after the LBR reading, otherwise the LBRs would be polluted
by the PMI handler.

This is a minimal change. In principle the ACK could be moved much later.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-10-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Handle new arch perfmon v4 status bits

ArchPerfmon v4 has some new status bits in GLOBAL_STATUS.

These need to be ignored when deciding whether a NMI
was an NMI, to avoid eating all NMIs when they
stay set, see:

    b292d7a10487 ("perf/x86/intel: ignore CondChgd bit to avoid false NMI handling")

This patch ignores the new ASIF bit, which indicates
that SGX interfered with the PMU, and also the new
LBR freezing bits, which are set when the LBRs get
frozen, plus the existing CondChange (set by JTAG
debuggers and some buggy BIOSes)

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-8-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Add support for LBRv5

Add support for the new LBRv5 format used on Intel Skylake CPUs.

The flags for mispredict, abort, in_tx etc. moved to range of separate
LBR_INFO_* MSRs. Teach the LBR code to read those. The original
LBR registers stay the same, except they have full sign
extension now.

LBR_INFO also reports a cycle count to the last branch.
Report the cycle information using the new "cycles" branch_info
output field.

In addition we have to context switch and clear the new INFO
MSRs to avoid any information leaks.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-6-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf: Add cycles to branch_info

Intel Skylake supports reporting the time in cycles a branch in the LBR
took, to give a rough indication of the basic block performance.

Export the cycle information in the branch_info structure.
This can be done by just reusing some currently zero padding.

This is just the generic header change. The architecture
still needs to fill it in.

There's no attempt to convert to real time, as we really
want cycles here.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-5-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

x86: Add new MSRs and MSR bits used for Intel Skylake PMU support

Add new MSRs (LBR_INFO) and some new MSR bits used by the Intel Skylake
PMU driver.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-4-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Allow time stamp for free running PEBSv3

With PEBSv3 the PEBS record contains a time stamp. That means we can allow
free-running PEBS without a PMI even if the user program requested a time stamp.
This avoids the need to use -T to get free running PEBS, and also avoids
any problems with mis-identifying MMAPs later.

Move the free_running_flags state into a variable in x86_pmu and use it.
This only works when no explicit clock_id is set.

Signed-off-by: Andi Kleen <ak@linux.intel.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@kernel.org
Cc: eranian@google.com
Cc: jolsa@redhat.com
Cc: kan.liang@intel.com
Link: http://lkml.kernel.org/r/1432786398-23861-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Add support for PEBSv3 profiling

PEBSv3 is the same as the existing PEBSv2 used on Haswell,
but it adds a new TSC field. Add support to the generic
PEBS handler to handle the new format, and overwrite
the perf time stamp using the new native_sched_clock_from_tsc().

Right now the time stamp is just slightly more accurate,
as it is nearer the actual event trigger point. With
the PEBS threshold > 1 patchkit it will be much more accurate,
avoid the problems with MMAP mismatches earlier.
The accurate time stamping is only implemented for
the default trace clock for now.

v2: Use _skl prefix. Check for default clock_id.
Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-3-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Add a native_perf_sched_clock_from_tsc()

PEBSv3 has a raw TSC time stamp in its memory buffer that
later needs to to be converted to perf_clock.

Add a native_sched_clock_from_tsc() that works the same
as native_sched_clock(), but starts with an already given
TSC value.

Paravirt is ignored, it will just get the native clock.
But there isn't a para virtualized PEBS anyway.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Link: http://lkml.kernel.org/r/1431285767-27027-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/pt: Add new timing packet enables

Intel PT chapter in the new Intel Architecture SDM adds several packets
corresponding enable bits and registers that control packet generation.
Also, additional bits in the Intel PT CPUID leaf were added to enumerate
presence and parameters of these new packets and features.

The packets and enables are:

  * CYC: cycle accurate mode, provides the number of cycles elapsed since
    previous CYC packet; its presence and available threshold values are
    enumerated via CPUID;

  * MTC: mini time counter packets, used for tracking TSC time between
    full TSC packets; its presence and available resolution options are
    enumerated via CPUID;

  * PSB packet period is now configurable, available period values are
    enumerated via CPUID.

This patch adds corresponding bit and register definitions, pmu driver
capabilities based on CPUID enumeration, new attribute format bits for
the new featurens and extends event configuration validation function
to take these into account.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.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@infradead.org
Cc: adrian.hunter@intel.com
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1438262131-12725-1-git-send-email-alexander.shishkin@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/pt: Do not force sync packets on every schedule-in

Currently, the PT driver zeroes out the status register every time before
starting the event. However, all the writable bits are already taken care
of in pt_handle_status() function, except the new PacketByteCnt field,
which in new versions of PT contains the number of packet bytes written
since the last sync (PSB) packet. Zeroing it out before enabling PT forces
a sync packet to be written. This means that, with the existing code, a
sync packet (PSB and PSBEND, 18 bytes in total) will be generated every
time a PT event is scheduled in.

To avoid these unnecessary syncs and save a WRMSR in the fast path, this
patch changes the default behavior to not clear PacketByteCnt field, so
that the sync packets will be generated with the period specified as
"psb_period" attribute config field. This has little impact on the trace
data as the other packets that are normally sent within PSB+ (between PSB
and PSBEND) have their own generation scenarios which do not depend on the
sync packets.

One exception where we do need to force PSB like this when tracing starts,
so that the decoder has a clear sync point in the trace. For this purpose
we aready have hw::itrace_started flag, which we are currently using to
output PERF_RECORD_ITRACE_START. This patch moves setting itrace_started
from perf core to the pmu::start, where it should still be 0 on the very
first run.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.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@infradead.org
Cc: adrian.hunter@intel.com
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1438264104-16189-1-git-send-email-alexander.shishkin@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/hw_breakpoints: Fix check for kernel-space breakpoints

The check looked wrong, although I think it was actually safe.  TASK_SIZE
is unnecessarily small for compat tasks, and it wasn't possible to make
a range breakpoint so large it started in user space and ended in kernel
space.

Nonetheless, let's fix up the check for the benefit of future
readers.  A breakpoint is in the kernel if either end is in the
kernel.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/136be387950e78f18cea60e9d1bef74465d0ee8f.1438312874.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/hw_breakpoints: Improve range breakpoint validation

Range breakpoints will do the wrong thing if the address isn't
aligned.  While we're there, add comments about why it's safe for
instruction breakpoints.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/ae25d14d61f2f43b78e0a247e469f3072df7e201.1438312874.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/hw_breakpoints: Disallow kernel breakpoints unless kprobe-safe

Code on the kprobe blacklist doesn't want unexpected int3
exceptions. It probably doesn't want unexpected debug exceptions
either. Be safe: disallow breakpoints in nokprobes code.

On non-CONFIG_KPROBES kernels, there is no kprobe blacklist.  In
that case, disallow kernel breakpoints entirely.

It will be particularly important to keep hw breakpoints out of the
entry and NMI code once we move debug exceptions off the IST stack.

Signed-off-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/e14b152af99640448d895e3c2a8c2d5ee19a1325.1438312874.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel: Fix SLM MSR_OFFCORE_RSP1 valid_mask

AVG_LATENCY(bit 38) is only available on MSR_OFFCORE_RSP0.
So the bit should be removed from RSP1 valid_mask.

Since RSP0 and RSP1 may have different valid_mask, intel_alt_er should
validate the config on the alternate offcore reg before replacing it.

Signed-off-by: Kan Liang <kan.liang@intel.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>
Link: http://lkml.kernel.org/r/1435170215-5017-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/lbr: Kill off intel_pmu_needs_lbr_smpl for good

The x86_lbr_exclusive commit (4807034248be "perf/x86: Mark Intel PT and
LBR/BTS as mutually exclusive") mistakenly moved intel_pmu_needs_lbr_smpl()
to perf_event.h, while another commit (a46a2300019 "perf: Simplify the
branch stack check") removed it in favor of needs_branch_stack().

This patch gets rid of intel_pmu_needs_lbr_smpl() for good.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.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@infradead.org
Cc: adrian.hunter@intel.com
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1435140349-32588-3-git-send-email-alexander.shishkin@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/bts: Drop redundant declarations

Both intel_pmu_enable_bts() and intel_pmu_disable_bts() are in perf_event.h
header file, no need to have them declared again in the driver.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.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@infradead.org
Cc: adrian.hunter@intel.com
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1435140349-32588-2-git-send-email-alexander.shishkin@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Use Sandy Bridge client PMU on Haswell/Broadwell

Haswell and Broadwell have the same uncore CBOX/ARB PMU as Sandy Bridge.
Add the respective model numbers to enable the SNB uncore PMU.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Cc: kan.liang@intel.com
Link: http://lkml.kernel.org/r/1434347862-28490-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Add support for ARB uncore PMU on Sandy/IvyBridge

Add a new "ARB" uncore PMU that is used to monitor the uncore queue
arbiter. This is useful to measure uncore queue occupancy and similar
statistics. The registers all have the same format as the
existing CBOX PMU.

Also move the event constraints from the CBOX to ARB. The 0x80+
events are ARB events and cannot be scheduled on a CBOX PMU.

Signed-off-by: Andi Kleen <ak@linux.intel.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: eranian@google.com
Cc: kan.liang@intel.com
Link: http://lkml.kernel.org/r/1434347862-28490-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/uncore: Remove use of macro DEFINE_PCI_DEVICE_TABLE()

The DEFINE_PCI_DEVICE_TABLE() macro is deprecated. Use
'struct pci_device_id' instead of DEFINE_PCI_DEVICE_TABLE(),
with the goal of getting rid of this macro completely.

This Coccinelle semantic patch performs this transformation:

@@
identifier a;
declarer name DEFINE_PCI_DEVICE_TABLE;
initializer i;
@@
- DEFINE_PCI_DEVICE_TABLE(a)
+ const struct pci_device_id a[] = i;

Signed-off-by: Vaishali Thakkar <vthakkar1994@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150717052759.GA6265@vaishali-Ideapad-Z570
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86/intel/rapl: Add support for Knights Landing (KNL)

Knights Landing DRAM RAPL supports PKG and DRAM RAPL domains.
DRAM RAPL has a different fixed energy unit (2^-16J) similar to
that of HSW.

Signed-off-by: Dasaratharaman Chandramouli <dasaratharaman.chandramouli@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Stephane Eranian <eranian@google.com>
Acked-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jacob Pan Jun <jacob.jun.pan@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Nikhil Rao <nikhil.rao@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/aa63b4a3af3160152fea1a10c807f4200527280c.1432665809.git.dasaratharaman.chandramouli@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Fix the waitqueue_active() check in xol_free_insn_slot()

The xol_free_insn_slot()->waitqueue_active() check is buggy. We
need mb() after we set the conditon for wait_event(), or
xol_take_insn_slot() can miss the wakeup.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Pratyush Anand <panand@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134036.GA4799@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Use vm_special_mapping to name the XOL vma

Change xol_add_vma() to use _install_special_mapping(), this way
we can name the vma installed by uprobes. Currently it looks
like private anonymous mapping, this is confusing and
complicates the debugging. With this change /proc/$pid/maps
reports "[uprobes]".

As a side effect this will cause core dumps to include the XOL vma
and I think this is good; this can help to debug the problem if
the app crashed because it was probed.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Pratyush Anand <panand@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134033.GA4796@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Fix the usage of install_special_mapping()

install_special_mapping(pages) expects that "pages" is the zero-
terminated array while xol_add_vma() passes &area->page, this
means that special_mapping_fault() can wrongly use the next
member in xol_area (vaddr) as "struct page *".

Fortunately, this area is not expandable so pgoff != 0 isn't
possible (modulo bugs in special_mapping_vmops), but still this
does not look good.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Pratyush Anand <panand@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134031.GA4789@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes/x86: Make arch_uretprobe_is_alive(RP_CHECK_CALL) more clever

The previous change documents that cleanup_return_instances()
can't always detect the dead frames, the stack can grow. But
there is one special case which imho worth fixing:
arch_uretprobe_is_alive() can return true when the stack didn't
actually grow, but the next "call" insn uses the already
invalidated frame.

Test-case:

#include <stdio.h>
#include <setjmp.h>

jmp_buf jmp;
int nr = 1024;

void func_2(void)
{
if (--nr == 0)
return;
longjmp(jmp, 1);
}

void func_1(void)
{
setjmp(jmp);
func_2();
}

int main(void)
{
func_1();
return 0;
}

If you ret-probe func_1() and func_2() prepare_uretprobe() hits
the MAX_URETPROBE_DEPTH limit and "return" from func_2() is not
reported.

When we know that the new call is not chained, we can do the
more strict check. In this case "sp" points to the new ret-addr,
so every frame which uses the same "sp" must be dead. The only
complication is that arch_uretprobe_is_alive() needs to know was
it chained or not, so we add the new RP_CHECK_CHAIN_CALL enum
and change prepare_uretprobe() to pass RP_CHECK_CALL only if
!chained.

Note: arch_uretprobe_is_alive() could also re-read *sp and check
if this word is still trampoline_vaddr. This could obviously
improve the logic, but I would like to avoid another
copy_from_user() especially in the case when we can't avoid the
false "alive == T" positives.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134028.GA4786@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Add the "enum rp_check ctx" arg to arch_uretprobe_is_alive()

arch/x86 doesn't care (so far), but as Pratyush Anand pointed
out other architectures might want why arch_uretprobe_is_alive()
was called and use different checks depending on the context.
Add the new argument to distinguish 2 callers.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134026.GA4779@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Change prepare_uretprobe() to (try to) flush the dead frames

Change prepare_uretprobe() to flush the !arch_uretprobe_is_alive()
return_instance's. This is not needed correctness-wise, but can help
to avoid the failure caused by MAX_URETPROBE_DEPTH.

Note: in this case arch_uretprobe_is_alive() can be false
positive, the stack can grow after longjmp(). Unfortunately, the
kernel can't 100% solve this problem, but see the next patch.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134023.GA4776@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Change handle_trampoline() to flush the frames invalidated by longjmp()

Test-case:

#include <stdio.h>
#include <setjmp.h>

jmp_buf jmp;

void func_2(void)
{
longjmp(jmp, 1);
}

void func_1(void)
{
if (setjmp(jmp))
return;
func_2();
printf("ERR!! I am running on the caller's stack\n");
}

int main(void)
{
func_1();
return 0;
}

fails if you probe func_1() and func_2() because
handle_trampoline() assumes that the probed function should must
return and hit the bp installed be prepare_uretprobe(). But in
this case func_2() does not return, so when func_1() returns the
kernel uses the no longer valid return_instance of func_2().

Change handle_trampoline() to unwind ->return_instances until we
know that the next chain is alive or NULL, this ensures that the
current chain is the last we need to report and free.

Alternatively, every return_instance could use unique
trampoline_vaddr, in this case we could use it as a key. And
this could solve the problem with sigaltstack() automatically.

But this approach needs more changes, and it puts the "hard"
limit on MAX_URETPROBE_DEPTH. Plus it can not solve another
problem partially fixed by the next patch.

Note: this change has no effect on !x86, the arch-agnostic
version of arch_uretprobe_is_alive() just returns "true".

TODO: as documented by the previous change, arch_uretprobe_is_alive()
      can be fooled by sigaltstack/etc.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134021.GA4773@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes/x86: Reimplement arch_uretprobe_is_alive()

Add the x86 specific version of arch_uretprobe_is_alive()
helper. It returns true if the stack frame mangled by
prepare_uretprobe() is still on stack. So if it returns false,
we know that the probed function has already returned.

We add the new return_instance->stack member and change the
generic code to initialize it in prepare_uretprobe, but it
should be equally useful for other architectures.

TODO: this assumes that the probed application can't use
      multiple stacks (say sigaltstack). We will try to improve
      this logic later.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134018.GA4766@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Export 'struct return_instance', introduce arch_uretprobe_is_alive()

Add the new "weak" helper, arch_uretprobe_is_alive(), used by
the next patches. It should return true if this return_instance
is still valid. The arch agnostic version just always returns
true.

The patch exports "struct return_instance" for the architectures
which want to override this hook. We can also cleanup
prepare_uretprobe() if we pass the new return_instance to
arch_uretprobe_hijack_return_addr().

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134016.GA4762@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Change handle_trampoline() to find the next chain beforehand

No functional changes, preparation.

Add the new helper, find_next_ret_chain(), which finds the first
!chained entry and returns its ->next. Yes, it is suboptimal. We
probably want to turn ->chained into ->start_of_this_chain
pointer and avoid another loop. But this needs the boring
changes in dup_utask(), so lets do this later.

Change the main loop in handle_trampoline() to unwind the stack
until ri is equal to the pointer returned by this new helper.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134013.GA4755@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Change prepare_uretprobe() to use uprobe_warn()

Turn the last pr_warn() in uprobes.c into uprobe_warn().

While at it:

   - s/kzalloc/kmalloc, we initialize every member of 'ri'

   - remove the pointless comment above the obvious code

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134010.GA4752@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Send SIGILL if handle_trampoline() fails

1. It doesn't make sense to continue if handle_trampoline()
   fails, change handle_swbp() to always return after this call.

2. Turn pr_warn() into uprobe_warn(), and change
   handle_trampoline() to send SIGILL on failure. It is pointless to
   return to user mode with the corrupted instruction_pointer() which
   we can't restore.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134008.GA4745@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Introduce free_ret_instance()

We can simplify uprobe_free_utask() and handle_uretprobe_chain()
if we add a simple helper which does put_uprobe/kfree and
returns the ->next return_instance.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134006.GA4740@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

uprobes: Introduce get_uprobe()

Cosmetic. Add the new trivial helper, get_uprobe(). It matches
put_uprobe() we already have and we can simplify a couple of its
users.

Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134003.GA4736@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

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

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

User visible changes:

  - Force period term to overload global settings, i.e. previously this
    command line:

     $ perf record -e 'cpu/instructions,period=20000/',cycles -c 1000 sleep 1

    would result in both events having a period equal to 1000, with the fix we
    get something saner:

     $ perf evlist -v | grep period
     cpu/instructions,period=20000/: ... { sample_period, sample_freq }: 20000, ...
     cycles: ... { sample_period, sample_freq }: 1000 ...
     $

   (Jiri Olsa)

Infrastructure changes:

  - Use the dummy software event with freq=0 in the twatch.py python
    binding example, to avoid disabling nohz. (Arnaldo Carvalho de Melo)

  - Add some missing constants to the python binding. (Arnaldo Carvalho de Melo)

  - Fix mismatched declarations for elf_getphdrnum, that happens
    only in the corner case where this function is not found on
    the system.  (Arnaldo Carvalho de Melo)

  - Add build test for having ending double slash. (Jiri Olsa)

  - Introduce callgraph_set for callgraph option. (Kan Liang)

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

Merge branch 'perf/urgent' into perf/core, to merge fixes before pulling more changes

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

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

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

  - Fix 'perf stat' transaction length metrics. (Andi Kleen)

  - Fix test build error when bindir contains double slash. (Pawel Moll)

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

perf tests: Adding build test for having ending double slash

Pawel Moll reported build issue for having extra slash (/) at the end of
the prefix variable.

  $ make prefix=/usr/local/

    CC       tests/attr.o
  tests/attr.c: In function ‘test__attr’:
  tests/attr.c:168:50: error: expected ‘)’ before ‘;’ token
    snprintf(path_perf, PATH_MAX, "%s/perf", BINDIR);
                                                ^
  tests/attr.c:176:1: error: expected ‘;’ before ‘}’ token
   }
   ^
  tests/attr.c:176:1: error: control reaches end of non-void function [-Werror=return-type]
   }
   ^
  cc1: all warnings being treated as errors

Adding automated test case for this.

Reported-by: Pawel Moll <pawel.moll@arm.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20150727182417.GD20509@krava.brq.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Introduce callgraph_set for callgraph option

Introduce callgraph_set to indicate whether the callgraph option was set
by user.

Signed-off-by: Kan Liang <kan.liang@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438162936-59698-4-git-send-email-kan.liang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Force period term to overload global settings

Currently the command line option settings beats the per event period
settings:

With no global settings, we get per-event configuration:

  $ perf record -e 'cpu/instructions,period=20000/' sleep 1
  $ perf evlist -v
  ... { sample_period, sample_freq }: 20000 ...

With 'c' option period setup, we get 'c' option value:
  $ perf record -e 'cpu/instructions,period=20000/' -c 1000 sleep 1
  $ perf evlist -v
  ... { sample_period, sample_freq }: 1000 ...

This patch makes the per-event settings overload the global 'c' option
setup:

  $ perf record -e 'cpu/instructions,period=20000/' -c 1000 sleep 1
  $ perf evlist -v
  ... { sample_period, sample_freq }: 20000 ...

I think the making the per-event settings to overload any other config
makes more sense than current state. However it breaks the current
'period' term handling, which might cause some noise.. so let's see ;-).

Also fixing parse event tests with the new behaviour.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438162936-59698-3-git-send-email-kan.liang@intel.com
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add support for event post configuration

Add support to overload any global settings for event and force user
specified term value. It will be useful for new time and backtrace
terms.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1438162936-59698-2-git-send-email-kan.liang@intel.com
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf session env: Rename exit method

The semantic associated in tools/perf/ with foo__delete(instance) is to
release all resources referenced by 'instance' members and then release
the memory for 'instance' itself.

The perf_session_env__delete() function isn't doing this, it just does
the first part, but the space used by 'instance' itself isn't freed, as
it is embedded in a larger structure, that will be freed at other stage.

For these cases we se foo__exit(), i.e. the usage is:

 void foo__delete(foo)
 {
         if (foo) {
                 foo__exit(foo);
                 free(foo);
         }
 }

But when we have something like:

 struct bar {
         struct foo foo;
         . . .
 }

Then we can't really call foo__delete(&bar.foo), we must have this
instead:

 void bar__exit(bar)
 {
         foo__exit(&bar.foo);
         /* free other bar-> resources */
 }

 void bar__delete(bar)
 {
         if (bar) {
bar__exit(bar);
                free(bar);
         }
 }

So just rename perf_session_env__delete() to perf_session_env__exit().

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-djbgpcfo5udqptx3q0flwtmk@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf symbols: Fix mismatched declarations for elf_getphdrnum

When HAVE_ELF_GETPHDRNUM_SUPPORT is false we trip on this problem:

    CC       /tmp/build/perf/util/symbol-elf.o
  util/symbol-elf.c:41:12: error: static declaration of ‘elf_getphdrnum’ follows non-static declaration
   static int elf_getphdrnum(Elf *elf, size_t *dst)
            ^
  In file included from util/symbol.h:19:0,
                   from util/symbol-elf.c:8:
  /usr/include/libelf.h:206:12: note: previous declaration of ‘elf_getphdrnum’ was here
   extern int elf_getphdrnum (Elf *__elf, size_t *__dst);
            ^
    MKDIR    /tmp/build/perf/bench/
  /home/git/linux/tools/build/Makefile.build:68: recipe for target '/tmp/build/perf/util/symbol-elf.o' failed
  make[3]: *** [/tmp/build/perf/util/symbol-elf.o] Error 1

Fix it.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-qcmekyfedmov4sxr0wahcikr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf python: Make twatch.py use soft dummy event, freq=0

To not sample, what we want are just the PERF_RECORD_ lifetime events
for threads, using the default, PERF_TYPE_HARDWARE +
PERF_COUNT_HW_CYCLES and freq=1 (the default), makes perf reenable
irq_vectors:local_timer_entry, disabling nohz, not good for some use
cases where all we want is to get notifications when threads comes and
goes...

Fix it by using PERF_TYPE_SOFTWARE (no counter rotation) and
PERF_COUNT_SW_DUMMY (created by Adrian so that we could have access to
those PERF_RECORD_ goodies).

Reported-by: Luiz Fernando Capitulino <lcapitulino@redhat.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jaroslav Skarvada <jskarvad@redhat.com>
Cc: Jeremy Eder <jeder@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-kfsijirfrs6xfhkcdxeoen06@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf python: Add missing PERF_RECORD_{MMAP2,AUX,etc}

Those were added to the kernel and tooling but we forgot to
expose them via the python binding, fix it.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-sg1m6t2c58gchidfce4hmitg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf python: Add macro to simplify maintainance of the constants array

Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/n/tip-ffuchgsbr5mqu91xl9oggfss@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf python: Remove dependency on 'machine' methods

The python binding still doesn't provide symbol resolving facilities,
but the recent addition of the trace_event__register_resolver() function
made it add as a dependency the machine__resolve_kernel_addr() method,
that in turn drags all the symbol resolving code.

The problem:

  [root@zoo ~]# perf test -v python
  17: Try 'import perf' in python, checking link problems      :
  --- start ---
  test child forked, pid 6853
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  ImportError: /tmp/build/perf/python/perf.so: undefined symbol: machine__resolve_kernel_addr
  test child finished with -1
  ---- end ----
  Try 'import perf' in python, checking link problems: FAILED!
  [root@zoo ~]#

Fix it by requiring this function to receive the resolver as a
parameter, just like pevent_register_function_resolver(), i.e. do
not explicitely refer to an object file not included in
tools/perf/util/python-ext-sources.

  [root@zoo ~]# perf test python
  17: Try 'import perf' in python, checking link problems      : Ok
  [root@zoo ~]#

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Fixes: c3168b0db93a ("perf symbols: Provide libtraceevent callback to resolve kernel symbols")
Link: http://lkml.kernel.org/n/tip-vxlhh95v2em9zdbgj3jm7xi5@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix test build error when bindir contains double slash

When building with a prefix ending with a slash, for example:

$ make prefix=/usr/local/

one of the perf tests fail to compile due to BUILD_STR macro mishandling
bindir_SQ string containing with two slashes:

-DBINDIR="BUILD_STR(/usr/local//bin)"

with the following error:

  CC       tests/attr.o
tests/attr.c: In function ‘test__attr’:
tests/attr.c:168:50: error: expected ‘)’ before ‘;’ token
  snprintf(path_perf, PATH_MAX, "%s/perf", BINDIR);
                                                  ^
tests/attr.c:176:1: error: expected ‘;’ before ‘}’ token
 }
 ^
tests/attr.c:176:1: error: control reaches end of non-void function [-Werror=return-type]
 }
 ^
cc1: all warnings being treated as errors

This patch works around the problem by "cleaning" the bindir string
using make's abspath function.

Signed-off-by: Pawel Moll <pawel.moll@arm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1438092613-21014-1-git-send-email-pawel.moll@arm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Fix transaction lenght metrics

The transaction length metrics in perf stat -T broke recently.

It would not match the metric correctly and always print K/sec.

This was caused by a incorrect update of the cycles_in_tx statistics.

Update the correct variable.

Also the check for zero division was reversed, which resulted in K/sec
being printed for no transactions. Fix this also up.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/1438039491-22091-1-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

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

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

New features:

  - Introduce PERF_RECORD_SWITCH(_CPU_WIDE) and use it in 'record' to
    ask for context switches, allowing non priviledged tasks to know
    when they are switched in and out, which wasn't possible with
    the other context switch tracepoint and software events, see the
    patch description for a comprehensive justification (Adrian Hunter)

  - Stop collecting /proc/kallsyms in perf.data files, saving about
    4.5MB on a typical x86-64 system, use the the symbol resolution
    routines used in all the other tools (report, top, etc) now that
    we can ask libtraceevent to use perf's symbol resolution code.
    (Arnaldo Carvalho de Melo)

User visible fixes:

  - Expose perf's symbol resolver to libtraceevent, so that its plugins can
    resolve tracepoint fields to kernel functions, like the 'function' field
    in the "timer:hrtimer_start tracepoint" (Arnaldo Carvalho de Melo)

Infrastructure changes:

  - Map propagation of thread and cpu maps improvements, prep work for
    'perf stat' new features (Jiri Olsa)

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

perf: Fix running time accounting

A recent fix to the shadow timestamp inadvertly broke the running time
accounting.

We must not update the running timestamp if we fail to schedule the
event, the event will not have ran. This can (and did) result in
negative total runtime because the stopped timestamp was before the
running timestamp (we 'started' but never stopped the event -- because
it never really started we didn't have to stop it either).

Reported-and-Tested-by: Vince Weaver <vincent.weaver@maine.edu>
Fixes: 72f669c0086f ("perf: Update shadow timestamp before add event")
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org # 4.1
Cc: Shaohua Li <shli@fb.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Linux 4.2-rc4

Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fix from Thomas Gleixner:
 "A single fix for the intel cqm perf facility to prevent IPIs from
  interrupt context"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/intel/cqm: Return cached counter value from IRQ context

Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "This update contains:

   - the manual revert of the SYSCALL32 changes which caused a
     regression

   - a fix for the MPX vma handling

   - three fixes for the ioremap 'is ram' checks.

   - PAT warning fixes

   - a trivial fix for the size calculation of TLB tracepoints

   - handle old EFI structures gracefully

  This also contains a PAT fix from Jan plus a revert thereof.  Toshi
  explained why the code is correct"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm/pat: Revert 'Adjust default caching mode translation tables'
  x86/asm/entry/32: Revert 'Do not use R9 in SYSCALL32' commit
  x86/mm: Fix newly introduced printk format warnings
  mm: Fix bugs in region_is_ram()
  x86/mm: Remove region_is_ram() call from ioremap
  x86/mm: Move warning from __ioremap_check_ram() to the call site
  x86/mm/pat, drivers/media/ivtv: Move the PAT warning and replace WARN() with pr_warn()
  x86/mm/pat, drivers/infiniband/ipath: Replace WARN() with pr_warn()
  x86/mm/pat: Adjust default caching mode translation tables
  x86/fpu: Disable dependent CPU features on "noxsave"
  x86/mpx: Do not set ->vm_ops on MPX VMAs
  x86/mm: Add parenthesis for TLB tracepoint size calculation
  efi: Handle memory error structures produced based on old versions of standard

x86/mm/pat: Revert 'Adjust default caching mode translation tables'

Toshi explains:

"No, the default values need to be set to the fallback types,
 i.e. minimal supported mode.  For WC and WT, UC is the fallback type.

 When PAT is disabled, pat_init() does update the tables below to
 enable WT per the default BIOS setup.  However, when PAT is enabled,
 but CPU has PAT -errata, WT falls back to UC per the default values."

Revert: ca1fec58bc6a 'x86/mm/pat: Adjust default caching mode translation tables'
Requested-by: Toshi Kani <toshi.kani@hp.com>
Cc: Jan Beulich <jbeulich@suse.de>
Link: http://lkml.kernel.org/r/1437577776.3214.252.camel@hp.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

perf/x86/intel/cqm: Return cached counter value from IRQ context

Peter reported the following potential crash which I was able to
reproduce with his test program,

[  148.765788] ------------[ cut here ]------------
[  148.765796] WARNING: CPU: 34 PID: 2840 at kernel/smp.c:417 smp_call_function_many+0xb6/0x260()
[  148.765797] Modules linked in:
[  148.765800] CPU: 34 PID: 2840 Comm: perf Not tainted 4.2.0-rc1+ #4
[  148.765803]  ffffffff81cdc398 ffff88085f105950 ffffffff818bdfd5 0000000000000007
[  148.765805]  0000000000000000 ffff88085f105990 ffffffff810e413a 0000000000000000
[  148.765807]  ffffffff82301080 0000000000000022 ffffffff8107f640 ffffffff8107f640
[  148.765809] Call Trace:
[  148.765810]  <NMI>  [<ffffffff818bdfd5>] dump_stack+0x45/0x57
[  148.765818]  [<ffffffff810e413a>] warn_slowpath_common+0x8a/0xc0
[  148.765822]  [<ffffffff8107f640>] ? intel_cqm_stable+0x60/0x60
[  148.765824]  [<ffffffff8107f640>] ? intel_cqm_stable+0x60/0x60
[  148.765825]  [<ffffffff810e422a>] warn_slowpath_null+0x1a/0x20
[  148.765827]  [<ffffffff811613f6>] smp_call_function_many+0xb6/0x260
[  148.765829]  [<ffffffff8107f640>] ? intel_cqm_stable+0x60/0x60
[  148.765831]  [<ffffffff81161748>] on_each_cpu_mask+0x28/0x60
[  148.765832]  [<ffffffff8107f6ef>] intel_cqm_event_count+0x7f/0xe0
[  148.765836]  [<ffffffff811cdd35>] perf_output_read+0x2a5/0x400
[  148.765839]  [<ffffffff811d2e5a>] perf_output_sample+0x31a/0x590
[  148.765840]  [<ffffffff811d333d>] ? perf_prepare_sample+0x26d/0x380
[  148.765841]  [<ffffffff811d3497>] perf_event_output+0x47/0x60
[  148.765843]  [<ffffffff811d36c5>] __perf_event_overflow+0x215/0x240
[  148.765844]  [<ffffffff811d4124>] perf_event_overflow+0x14/0x20
[  148.765847]  [<ffffffff8107e7f4>] intel_pmu_handle_irq+0x1d4/0x440
[  148.765849]  [<ffffffff811d07a6>] ? __perf_event_task_sched_in+0x36/0xa0
[  148.765853]  [<ffffffff81219bad>] ? vunmap_page_range+0x19d/0x2f0
[  148.765854]  [<ffffffff81219d11>] ? unmap_kernel_range_noflush+0x11/0x20
[  148.765859]  [<ffffffff814ce6fe>] ? ghes_copy_tofrom_phys+0x11e/0x2a0
[  148.765863]  [<ffffffff8109e5db>] ? native_apic_msr_write+0x2b/0x30
[  148.765865]  [<ffffffff8109e44d>] ? x2apic_send_IPI_self+0x1d/0x20
[  148.765869]  [<ffffffff81065135>] ? arch_irq_work_raise+0x35/0x40
[  148.765872]  [<ffffffff811c8d86>] ? irq_work_queue+0x66/0x80
[  148.765875]  [<ffffffff81075306>] perf_event_nmi_handler+0x26/0x40
[  148.765877]  [<ffffffff81063ed9>] nmi_handle+0x79/0x100
[  148.765879]  [<ffffffff81064422>] default_do_nmi+0x42/0x100
[  148.765880]  [<ffffffff81064563>] do_nmi+0x83/0xb0
[  148.765884]  [<ffffffff818c7c0f>] end_repeat_nmi+0x1e/0x2e
[  148.765886]  [<ffffffff811d07a6>] ? __perf_event_task_sched_in+0x36/0xa0
[  148.765888]  [<ffffffff811d07a6>] ? __perf_event_task_sched_in+0x36/0xa0
[  148.765890]  [<ffffffff811d07a6>] ? __perf_event_task_sched_in+0x36/0xa0
[  148.765891]  <<EOE>>  [<ffffffff8110ab66>] finish_task_switch+0x156/0x210
[  148.765898]  [<ffffffff818c1671>] __schedule+0x341/0x920
[  148.765899]  [<ffffffff818c1c87>] schedule+0x37/0x80
[  148.765903]  [<ffffffff810ae1af>] ? do_page_fault+0x2f/0x80
[  148.765905]  [<ffffffff818c1f4a>] schedule_user+0x1a/0x50
[  148.765907]  [<ffffffff818c666c>] retint_careful+0x14/0x32
[  148.765908] ---[ end trace e33ff2be78e14901 ]---

The CQM task events are not safe to be called from within interrupt
context because they require performing an IPI to read the counter value
on all sockets. And performing IPIs from within IRQ context is a
"no-no".

Make do with the last read counter value currently event in
event->count when we're invoked in this context.

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vikas Shivappa <vikas.shivappa@intel.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Will Auld <will.auld@intel.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/1437490509-15373-1-git-send-email-matt@codeblueprint.co.uk
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>