linux-stable.git/tools/perf, branch v4.8.5

perf intel-pt: Fix MTC timestamp calculation for large MTC periods

2016-10-28T07:45:20+00:00

commit 3bccbe20f6d188ce7b00326e776b745cfd35b10a upstream.

The MTC packet provides a 8-bit slice of CTC which is related to TSC by
the TMA packet, however the TMA packet only provides the lower 16 bits
of CTC.  If mtc_shift > 8 then some of the MTC bits are not in the CTC
provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
packet by copying the missing bits from the current MTC assuming the
least difference between the two, and that the current MTC comes after
last_mtc.

Signed-off-by: Adrian Hunter 
Cc: Jiri Olsa 
Link: http://lkml.kernel.org/r/1475062896-22274-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf intel-pt: Fix estimated timestamps for cycle-accurate mode

2016-10-28T07:45:20+00:00

commit 51ee6481fa8e879cc942bcc1b0af713e158b7a98 upstream.

In cycle-accurate mode, timestamps can be calculated from CYC packets.
The decoder also estimates timestamps based on the number of
instructions since the last timestamp. For that to work in
cycle-accurate mode, the instruction count needs to be reset to zero
when a timestamp is calculated from a CYC packet, but that wasn't
happening, so fix it.

Signed-off-by: Adrian Hunter 
Cc: Jiri Olsa 
Link: http://lkml.kernel.org/r/1475062896-22274-1-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf intel-pt: Fix snapshot overlap detection decoder errors

2016-10-28T07:45:20+00:00

commit 810c398bc09b2f2dfde52a7d2483a710612c5fb8 upstream.

Fix occasional decoder errors decoding trace data collected in snapshot
mode.

Snapshot mode can take successive snapshots of trace which might overlap.
The decoder checks whether there is an overlap but only looks at the
current and previous buffer. However buffers that do not contain
synchronization (i.e. PSB) packets cannot be decoded or used for overlap
checking. That means the decoder actually needs to check overlaps between
the current buffer and the previous buffer that contained usable data.
Make that change.

Signed-off-by: Adrian Hunter 
Cc: Jiri Olsa 
Cc: Masami Hiramatsu 
Cc: Mathieu Poirier 
Link: http://lkml.kernel.org/r/1474641528-18776-10-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo 
Signed-off-by: Greg Kroah-Hartman

perf evsel: Do not access outside hw cache name arrays

2016-08-18T19:39:46+00:00

We have to check if the values are >= *_MAX, not just >, fix it.

From the bugzilla report:

''In file /tools/perf/util/evsel.c  function __perf_evsel__hw_cache_name
it appears that there is a bug that reads beyond the end of the buffer.
The statement "if (type > PERF_COUNT_HW_CACHE_MAX)" allows type to be
equal to the maximum value. Later, when statement "if
(!perf_evsel__is_cache_op_valid(type, op))" is executed, the function
can access array perf_evsel__hw_cache_stat[type] beyond the end of the
buffer.

It appears to me that the statement "if (type > PERF_COUNT_HW_CACHE_MAX)"
should be "if (type >= PERF_COUNT_HW_CACHE_MAX)"

Bug found with Coverity and manual code review. No attempts were made to
execute the code with a maximum type value.''

Committer note:

Testing it:

  $ perf record -e $(echo $(perf list cache | cut -d \[ -f1) | sed 's/ /,/g') usleep 1
  [ perf record: Woken up 16 times to write data ]
  [ perf record: Captured and wrote 0.023 MB perf.data (34 samples) ]
  $ perf evlist
  L1-dcache-load-misses
  L1-dcache-loads
  L1-dcache-stores
  L1-icache-load-misses
  LLC-load-misses
  LLC-loads
  LLC-store-misses
  LLC-stores
  branch-load-misses
  branch-loads
  dTLB-load-misses
  dTLB-loads
  dTLB-store-misses
  dTLB-stores
  iTLB-load-misses
  iTLB-loads
  node-load-misses
  node-loads
  node-store-misses
  node-stores
  $ perf list cache

  List of pre-defined events (to be used in -e):

    L1-dcache-load-misses        [Hardware cache event]
    L1-dcache-loads              [Hardware cache event]
    L1-dcache-stores             [Hardware cache event]
    L1-icache-load-misses        [Hardware cache event]
    LLC-load-misses              [Hardware cache event]
    LLC-loads                    [Hardware cache event]
    LLC-store-misses             [Hardware cache event]
    LLC-stores                   [Hardware cache event]
    branch-load-misses           [Hardware cache event]
    branch-loads                 [Hardware cache event]
    dTLB-load-misses             [Hardware cache event]
    dTLB-loads                   [Hardware cache event]
    dTLB-store-misses            [Hardware cache event]
    dTLB-stores                  [Hardware cache event]
    iTLB-load-misses             [Hardware cache event]
    iTLB-loads                   [Hardware cache event]
    node-load-misses             [Hardware cache event]
    node-loads                   [Hardware cache event]
    node-store-misses            [Hardware cache event]
    node-stores                  [Hardware cache event]
  $

Reported-by: Brian Sweeney 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Jiri Olsa 
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=153351
Signed-off-by: Arnaldo Carvalho de Melo

perf unwind: Use addr_location::addr instead of ip for entries

2016-08-16T18:23:29+00:00

This fixes the srcline translation for call chains of user space
applications.

Before we got:

    perf report --stdio --no-children -s sym,srcline -g address
     8.92%  [.] main                                 mandelbrot.h:41
            |
            |--3.70%--main +8390240
            |          __libc_start_main +139950056726769
            |          _start +8388650
            |
            |--2.74%--main +8390189
            |
             --2.08%--main +8390296
                       __libc_start_main +139950056726769
                       _start +8388650

     7.59%  [.] main                                 complex:1326
            |
            |--4.79%--main +8390203
            |          __libc_start_main +139950056726769
            |          _start +8388650
            |
             --2.80%--main +8390219

     7.12%  [.] __muldc3                             libgcc2.c:1945
            |
            |--3.76%--__muldc3 +139950060519490
            |          main +8390224
            |          __libc_start_main +139950056726769
            |          _start +8388650
            |
             --3.32%--__muldc3 +139950060519512
                       main +8390224

With this patch applied, we instead get:

    perf report --stdio --no-children -s sym,srcline -g address
     8.92%  [.] main                                 mandelbrot.h:41
            |
            |--3.70%--main mandelbrot.h:41
            |          __libc_start_main +241
            |          _start +4194346
            |
            |--2.74%--main mandelbrot.h:41
            |
             --2.08%--main mandelbrot.h:41
                       __libc_start_main +241
                       _start +4194346

     7.59%  [.] main                                 complex:1326
            |
            |--4.79%--main complex:1326
            |          __libc_start_main +241
            |          _start +4194346
            |
             --2.80%--main complex:1326

     7.12%  [.] __muldc3                             libgcc2.c:1945
            |
            |--3.76%--__muldc3 libgcc2.c:1945
            |          main mandelbrot.h:39
            |          __libc_start_main +241
            |          _start +4194346
            |
             --3.32%--__muldc3 libgcc2.c:1945
                       main mandelbrot.h:39

Suggested-and-Acked-by: Namhyung Kim 
Signed-off-by: Milian Wolff 
Tested-by: Arnaldo Carvalho de Melo 
LPU-Reference: 20160816153926.11288-1-milian.wolff@kdab.com
Signed-off-by: Arnaldo Carvalho de Melo

perf intel-pt: Fix occasional decoding errors when tracing system-wide

2016-08-15T21:11:18+00:00

In order to successfully decode Intel PT traces, context switch events
are needed from the moment the trace starts. Currently that is ensured
by using the 'immediate' flag which enables the switch event when it is
opened.

However, since commit 86c2786994bd ("perf intel-pt: Add support for
PERF_RECORD_SWITCH") that might not always happen. When tracing
system-wide the context switch event is added to the tracking event
which was not set as 'immediate'. Change that so it is.

Signed-off-by: Adrian Hunter 
Cc: Jiri Olsa 
Cc: stable@vger.kernel.org # v4.4+
Fixes: 86c2786994bd ("perf intel-pt: Add support for PERF_RECORD_SWITCH")
Link: http://lkml.kernel.org/r/1471245784-22580-1-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf probe: Release resources on error when handling exit paths

2016-08-15T21:10:59+00:00

Cc: Adrian Hunter 
Cc: Colin King 
Cc: David Ahern 
Cc: Jiri Olsa 
Cc: Masami Hiramatsu 
Cc: Namhyung Kim 
Cc: Wang Nan 
Link: http://lkml.kernel.org/n/tip-zh2j4iqimralugke5qq7dn6d@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf probe: Check for dup and fdopen failures

2016-08-15T20:06:19+00:00

dup and fdopen can potentially fail, so add some extra
error handling checks rather than assuming they always work.

Signed-off-by: Colin King 
Acked-by: Masami Hiramatsu 
Cc: Alexander Shishkin 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Wang Nan 
Link: http://lkml.kernel.org/r/1471038296-12956-1-git-send-email-colin.king@canonical.com
[ Free resources when those functions (now being verified) fail ]
Signed-off-by: Arnaldo Carvalho de Melo

perf symbols: Fix annotation of objects with debuginfo files

2016-08-15T19:49:57+00:00

Commit 73cdf0c6ea9c ("perf symbols: Record text offset in dso
to calculate objdump address") started storing the offset of
the text section for all DSOs:

       if (elf_section_by_name(elf, &ehdr, &tshdr, ".text", NULL))
               dso->text_offset = tshdr.sh_addr - tshdr.sh_offset;

Unfortunately this breaks debuginfo files, because we need to calculate
the offset of the text section in the associated executable file. As a
result perf annotate returns junk for all debuginfo files.

Fix this by using runtime_ss->elf which should point at the executable
when parsing a debuginfo file.

Signed-off-by: Anton Blanchard 
Reviewed-by: Naveen N. Rao 
Tested-by: Wang Nan 
Cc: Peter Zijlstra 
Cc: Ravi Bangoria 
Cc: stable@vger.kernel.org # v4.6+
Fixes: 73cdf0c6ea9c ("perf symbols: Record text offset in dso to calculate objdump address")
Link: http://lkml.kernel.org/r/20160813115533.6de17912@kryten
Signed-off-by: Arnaldo Carvalho de Melo

perf script: Don't disable use_callchain if input is pipe

2016-08-15T16:23:49+00:00

Because perf data from pipe do not have a header with evsel attr, we
should not check that and disable symbol_conf.use_callchain. Otherwise,
perf script won't show callchains even if the data stream contains
callchain.

Before:
  $ perf record -g -o - uname |perf script
  Linux
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.000 MB - ]
  uname  1828 182630.186578:  250000 cpu-clock:  ..b9499 setup_arg_pages
  uname  1828 182630.186850:  250000 cpu-clock:  ..83b20 ___might_sleep
  uname  1828 182630.187153:  250000 cpu-clock:  ..4b6be file_map_prot_ch
  ...

After:
  $ perf record -g -o - uname |perf script
  Linux
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.000 MB - ]
  uname  1833 182675.927099:     250000 cpu-clock:
                  ba5520 _raw_spin_lock+0xfe200040 ([kernel.kallsyms])
                  389dd4 expand_downwards+0xfe200154 ([kernel.kallsyms])
                  389f34 expand_stack+0xfe200024 ([kernel.kallsyms])
                  3b957e setup_arg_pages+0xfe20019e ([kernel.kallsyms])
                  40c80f load_elf_binary+0xfe20042f ([kernel.kallsyms])
                  ...

Signed-off-by: He Kuang 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Alexander Shishkin 
Cc: Peter Zijlstra 
Cc: Wang Nan 
Link: http://lkml.kernel.org/r/1470309943-153909-2-git-send-email-hekuang@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo