openwrt/staging/blogic.git
12 years agoperf tools: Bump default sample freq to 4 kHz
Arnaldo Carvalho de Melo [Tue, 22 May 2012 16:14:18 +0000 (13:14 -0300)]
perf tools: Bump default sample freq to 4 kHz

Quoting Ingo:

"While at it I'd also suggest increasing the default sampling frequency,
from 1000 Hz per CPU to at least 4Khz auto-freq or so - this should work
well all across the board I think. CPUs are getting faster and command/app
run times are getting shorter, 1Khz is a bit low IMO."

Requested-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-2jafa6mkrufyekny9ei59lpu@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf buildid-list: Work better with pipe mode
Stephane Eranian [Tue, 15 May 2012 11:28:15 +0000 (13:28 +0200)]
perf buildid-list: Work better with pipe mode

In order for perf buildid-list to work with pipe-mode files, it needs to
process buildids and event attr structs.

$ perf record -o - noploop 2 | ./perf inject -b | perf buildid-list -i - -H
noploop for 2 seconds
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.084 MB - (~3678 samples) ]
0000000000000000000000000000000000000000 [kernel.kallsyms]
3a0d0629efe74a8da3eeba372cdbd74ad9b8f5d5 /usr/local/bin/noploop

The reason [kernel.kallsyms] shows a 0 build-id comes from the
way buildids are injected in the stream.

The buildid for the kernel is provided by a BUILD_ID record. The
[kernel.kallsyms] is provided by a MMAP record. There is no clean and
obvious way to link the two, unfortunately.

In regular mode, the kernel buildid is generated from reading the ELF
image or kallsyms and perf knows to associate [kernel.kallsyms] to it.
Later on, when perf processes the [kernel.kallsyms] MMAP record, it will
already have a dso for it.

So for now, make sure perf buildid-list shows the buildids for
everything but the kernel image.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1337081295-10303-6-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Fix piped mode read code
Stephane Eranian [Tue, 15 May 2012 11:28:12 +0000 (13:28 +0200)]
perf tools: Fix piped mode read code

In __perf_session__process_pipe_events(), there was a risk we would read
more than what a union perf_event struct can hold. this could happen in
case, perf is reading a file which contains new record types it does not
know about and which are larger than anything it knows about.

In general, perf is supposed to skip records it does not understand, but
in pipe mode, those have to be read and ignored.  The fixed size header
contains the size of the record, but that size may be larger than union
perf_event, yet it was used as the backing to the read in:

  union perf_event event;
  void *p;

  size = event->header.size;

  p = &event;
  p += sizeof(struct perf_event_header);
  if (size - sizeof(struct perf_event_header)) {
    err = readn(self->fd, p, size - sizeof(struct perf_event_header));

We fix this by allocating a buffer based on the size reported in the
header. We reuse the buffer as much as we can. We realloc in case it
becomes too small. In the  common case, the performance impact is
negligible.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1337081295-10303-3-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf inject: Fix broken perf inject -b
Stephane Eranian [Tue, 15 May 2012 11:28:11 +0000 (13:28 +0200)]
perf inject: Fix broken perf inject -b

perf inject -b was broken. It would not inject any build_id into the
stream. Furthermore, it would strip samples from the stream.

The reason was a missing initialization of the event attribute
structure. The perf_tool.tool.attr() callback was pointing to a simple
repipe. But there was no initialization of the internal data structures
to keep track of events and event ids. That later caused event id
lookups to fail, and sample would get removed.

The patch simply adds back the call to perf_event__process_attr() to
initialize the evlist structure and now build_ids are again injected.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1337081295-10303-2-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: rename HEADER_TRACE_INFO to HEADER_TRACING_DATA
Stephane Eranian [Tue, 15 May 2012 11:28:13 +0000 (13:28 +0200)]
perf tools: rename HEADER_TRACE_INFO to HEADER_TRACING_DATA

To match the PERF_RECORD_HEADER_TRACING_DATA record type.

This is the same info as the one used for pipe mode whereas the other
one is for regular file output. This will help in the later patch to add
meta-data infos in pipe mode.

Signed-off-by: Stephane Eranian <eranian@google.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1337081295-10303-4-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Add union u64_swap type for swapping u64 data
Jiri Olsa [Wed, 16 May 2012 06:59:04 +0000 (08:59 +0200)]
perf tools: Add union u64_swap type for swapping u64 data

The following union:
  union {
        u64 val64;
        u32 val32[2];
  } u;

is used on more than one place in perf code and will be used more in
upcomming patches.

Adding union u64_swap to have it defined globaly so we dont need to
redefine it all the time.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Tested-by: David Ahern <dsahern@gmail.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337151548-2396-4-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Carry perf_event_attr bitfield throught different endians
Jiri Olsa [Wed, 16 May 2012 06:59:03 +0000 (08:59 +0200)]
perf tools: Carry perf_event_attr bitfield throught different endians

When the perf data file is read cross architectures, the
perf_event__attr_swap function takes care about endianness of all the
struct fields except the bitfield flags.

The bitfield flags need to be transformed as well, since the bitfield
binary storage differs for both endians.

ABI says:
  Bit-fields are allocated from right to left (least to most significant)
  on little-endian implementations and from left to right (most to least
  significant) on big-endian implementations.

The above seems to be byte specific, so we need to reverse each byte of
the bitfield. 'Internet' also says this might be implementation specific
and we probably need proper fix and carry perf_event_attr bitfield flags
in separate data file FEAT_ section. Thought this seems to work for now.

Note, running following to test perf endianity handling:
test 1)
  - origin system:
    # perf record -a -- sleep 10 (any perf record will do)
    # perf report > report.origin
    # perf archive perf.data

  - copy the perf.data, report.origin and perf.data.tar.bz2
    to a target system and run:
    # tar xjvf perf.data.tar.bz2 -C ~/.debug
    # perf report > report.target
    # diff -u report.origin report.target

  - the diff should produce no output
    (besides some white space stuff and possibly different
     date/TZ output)

test 2)
  - origin system:
    # perf record -ag -fo /tmp/perf.data -- sleep 1
  - mount origin system root to the target system on /mnt/origin
  - target system:
    # perf script --symfs /mnt/origin -I -i /mnt/origin/tmp/perf.data \
     --kallsyms /mnt/origin/proc/kallsyms
  - complete perf.data header is displayed

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Tested-by: David Ahern <dsahern@gmail.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337151548-2396-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf record: Fix documentation for branch stack sampling
Anshuman Khandual [Fri, 18 May 2012 08:46:50 +0000 (14:16 +0530)]
perf record: Fix documentation for branch stack sampling

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Acked-by: Stephane Eranian <eranian@google.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/4FB60C7A.2080508@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Add cpu flag to sample_type if target has cpu
Namhyung Kim [Mon, 21 May 2012 01:42:07 +0000 (10:42 +0900)]
perf target: Add cpu flag to sample_type if target has cpu

Add PERF_SAMPLE_CPU flag into attr->sample_type if an user specified any
of cpu target (either system-wide or cpu list).

It will show correct values when cpu sort key is given for perf top and
perf report.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337564527-9367-1-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Always try to build libtraceevent
Namhyung Kim [Tue, 22 May 2012 09:03:54 +0000 (18:03 +0900)]
perf tools: Always try to build libtraceevent

Although perf depends on the libtraceevent, it cannot know when it needs
to be rebuilt. So just try to rebuild it always in order to make sure we
use the latest version.

While at it, silence annoying directory change messages.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1337677434-4881-2-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Rename libparsevent to libtraceevent in Makefile
Namhyung Kim [Tue, 22 May 2012 09:03:53 +0000 (18:03 +0900)]
perf tools: Rename libparsevent to libtraceevent in Makefile

Change some variable names according to new library name.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1337677434-4881-1-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf script: Rename struct event to struct event_format in perl engine
Frederic Weisbecker [Tue, 22 May 2012 14:30:49 +0000 (16:30 +0200)]
perf script: Rename struct event to struct event_format in perl engine

While migrating to the libtraceevent, the perl scripting engine
missed this structure rename.

This fixes:

     util/scripting-engines/trace-event-perl.c: In function "find_cache_event":
     util/scripting-engines/trace-event-perl.c:244: error: assignment from incompatible pointer type
     util/scripting-engines/trace-event-perl.c:248: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:248: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:250: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c: In function "perl_process_tracepoint":
     util/scripting-engines/trace-event-perl.c:286: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:286: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:307: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c: In function "perl_generate_script":
     util/scripting-engines/trace-event-perl.c:498: error: passing argument 1 of "trace_find_next_event" from incompatible pointer type
     util/scripting-engines/../trace-event.h:56: note: expected "struct event_format *" but argument is of type "struct event *"
     util/scripting-engines/trace-event-perl.c:498: error: assignment from incompatible pointer type
     util/scripting-engines/trace-event-perl.c:499: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:499: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:513: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:532: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:556: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:569: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:570: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:579: error: dereferencing pointer to incomplete type
     util/scripting-engines/trace-event-perl.c:580: error: dereferencing pointer to incomplete type

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/1337697049-30251-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf script: Explicitly handle known default print arg type
Frederic Weisbecker [Tue, 22 May 2012 14:30:48 +0000 (16:30 +0200)]
perf script: Explicitly handle known default print arg type

Handle the print argument types brought by the new libparsevent in perl
scripting engine.

PRINT_BSTRING and PRINT_DYNAMIC_ARRAY are treated just like strings
and thus don't require specific processing.

But PRINT_FUNC need specific plugins which are not yet handled, lets
warn if we meet this case.

This fixes:

     util/scripting-engines/trace-event-perl.c: In function define_event_symbol:
     util/scripting-engines/trace-event-perl.c:188: error: enumeration value PRINT_BSTRING not handled in switch
     util/scripting-engines/trace-event-perl.c:188: error: enumeration value PRINT_DYNAMIC_ARRAY not handled in switch
     util/scripting-engines/trace-event-perl.c:188: error: enumeration value PRINT_FUNC not handled in switch

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/1337697049-30251-1-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Add hardcoded name term for pmu events
Jiri Olsa [Mon, 21 May 2012 07:12:53 +0000 (09:12 +0200)]
perf tools: Add hardcoded name term for pmu events

Adding a new hardcoded term 'name' allowing to specify a name for the
pmu event. The term is defined along with standard pmu terms. If no
'name' term is given, the event name follows following template:

    "raw 0x<perf_event_attr::config>"

running:
    perf stat -e cpu/config=1,name=krava1/u ls

will produce following output:
    ...
    Performance counter stats for 'ls':
                 0 krava1
    ...

running:
    perf stat -e cpu/config=1/u ls

will produce following output:
    ...
    Performance counter stats for 'ls':
                 0 raw 0x1
    ...

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337584373-2741-6-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Separate 'mem:' event scanner bits
Jiri Olsa [Mon, 21 May 2012 07:12:52 +0000 (09:12 +0200)]
perf tools: Separate 'mem:' event scanner bits

Separating 'mem:' scanner processing, so we can parse out modifier
specifically and dont clash with other rules.

This is just precaution for the future, so we dont need to worry about
the rules clashing where we need to parse out any sub-rule of global
rules.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337584373-2741-5-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Use allocated list for each parsed event
Jiri Olsa [Mon, 21 May 2012 07:12:51 +0000 (09:12 +0200)]
perf tools: Use allocated list for each parsed event

Switch from using static temporary event list into dynamically allocated
one. This way we dont need to pass temp list to the parse_events_parse
which makes the interface more clear.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337584373-2741-4-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Add support for displaying event parser debug info
Jiri Olsa [Mon, 21 May 2012 07:12:50 +0000 (09:12 +0200)]
perf tools: Add support for displaying event parser debug info

Adding PARSER_DEBUG Makefile variable to enable building event scanner/
parser with debug enabled. This results in verbose output right out of
the scanner/parser.

It's useful for debuging the event parser. Keeping this only for event
parser so far.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337584373-2741-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf test: Move parse event automated tests to separated object
Jiri Olsa [Mon, 21 May 2012 07:12:49 +0000 (09:12 +0200)]
perf test: Move parse event automated tests to separated object

Moving event parsing specific tests into separated file:

  util/parse-events-test.c

Also changing the code a bit to ease running separate tests.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337584373-2741-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge branch 'perf/parse-events-4' of git://github.com/fweisbec/tracing into perf...
Ingo Molnar [Mon, 21 May 2012 08:42:09 +0000 (10:42 +0200)]
Merge branch 'perf/parse-events-4' of git://github.com/fweisbec/tracing into perf/core

Conflicts:
tools/perf/Makefile

This tree from Frederic unifies the perf and trace-cmd trace event format
parsing code into a single library.

Powertop and other tools will also be able to make use of it.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoMerge branch 'tip/perf/core-2' of git://git.kernel.org/pub/scm/linux/kernel/git/roste...
Ingo Molnar [Mon, 21 May 2012 07:44:36 +0000 (09:44 +0200)]
Merge branch 'tip/perf/core-2' of git://git./linux/kernel/git/rostedt/linux-trace into perf/core

12 years agoMerge branch 'perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux...
Ingo Molnar [Mon, 21 May 2012 07:17:31 +0000 (09:17 +0200)]
Merge branch 'perf/core' of git://git./linux/kernel/git/acme/linux into perf/core

Fixes for perf/core:

 - Rename some perf_target methods to avoid double negation, from Namhyung Kim.
 - Revert change to use per task events with inheritance, from Namhyung Kim.
 - Events should start disabled till children starts running, from David Ahern.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agotracing: Remove kernel_lock annotations
Richard Weinberger [Mon, 7 Nov 2011 08:23:22 +0000 (09:23 +0100)]
tracing: Remove kernel_lock annotations

The BKL is gone, these annotations are useless.

Link: http://lkml.kernel.org/r/1320654202-4433-1-git-send-email-richard@nod.at
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agotracing: Fix initial buffer_size_kb state
Vaibhav Nagarnaik [Thu, 3 May 2012 17:40:34 +0000 (10:40 -0700)]
tracing: Fix initial buffer_size_kb state

Make sure that the state of buffer_size_kb is initialized correctly and
returns actual size of the ring buffer.

Link: http://lkml.kernel.org/r/1336066834-1673-1-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoring-buffer: Merge separate resize loops
Vaibhav Nagarnaik [Fri, 18 May 2012 20:29:51 +0000 (13:29 -0700)]
ring-buffer: Merge separate resize loops

There are 2 separate loops to resize cpu buffers that are online and
offline. Merge them to make the code look better.

Also change the name from update_completion to update_done to allow
shorter lines.

Link: http://lkml.kernel.org/r/1337372991-14783-1-git-send-email-vnagarnaik@google.com
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoperf evsel: Create events initially disabled -- again
David Ahern [Mon, 14 May 2012 04:01:28 +0000 (22:01 -0600)]
perf evsel: Create events initially disabled -- again

764e16a changed perf-record to create events disabled by default and
enable them once perf initializations are done. This setting was dropped
by 0f82ebc. Now perf events are once again generated during perf's
initialization phase (e.g., generating maps).

As an example, perf opens a lot of files at startup. Unpatched:

perf record -e syscalls:sys_enter_open -ga -fo /tmp/perf.data -- sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.087 MB /tmp/perf.data (~3798 samples) ]

Using perf-script to look at the samples shows the perf command generating
563 of the 566 total events.

Patched:

perf record -e syscalls:sys_enter_open -ga -fo /tmp/perf.data -- sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.028 MB /tmp/perf.data (~1206 samples) ]

Using perf-script to look at the samples does not show perf command.

Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1336968088-11531-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge remote-tracking branch 'tip/perf/urgent' into perf/core
Arnaldo Carvalho de Melo [Fri, 18 May 2012 16:13:33 +0000 (13:13 -0300)]
Merge remote-tracking branch 'tip/perf/urgent' into perf/core

Merge reason: We are going to queue up a dependent patch:

"perf tools: Move parse event automated tests to separated object"

That depends on:

commit e7c72d8
perf tools: Add 'G' and 'H' modifiers to event parsing

Conflicts:
tools/perf/builtin-stat.c

Conflicted with the recent 'perf_target' patches when checking the
result of perf_evsel open routines to see if a retry is needed to cope
with older kernels where the exclude guest/host perf_event_attr bits
were not used.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Split term type into value type and term type
Jiri Olsa [Wed, 25 Apr 2012 16:24:57 +0000 (18:24 +0200)]
perf tools: Split term type into value type and term type

Introducing type_val and type_term for term instead of a single type
value. Currently the term type marked out the value type as well.

With this change we can have future string term values being specified
by user and translated into proper number along the processing.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1335371102-11358-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf hists: Fix callchain ip printf format
Jiri Olsa [Wed, 16 May 2012 06:59:08 +0000 (08:59 +0200)]
perf hists: Fix callchain ip printf format

The callchain address is stored as u64. Current code uses following
format string to display callchain address:

  "%p\n", (void *)(long)chain->ip

This way we lose upper 32 bits if we report 64 bit addresses in 32 bit
environment. Fixing this to always display whole 64 bits.

Note, running following to test perf endianity handling:
test 1)
  - origin system:
    # perf record -a -- sleep 10 (any perf record will do)
    # perf report > report.origin
    # perf archive perf.data

  - copy the perf.data, report.origin and perf.data.tar.bz2
    to a target system and run:
    # tar xjvf perf.data.tar.bz2 -C ~/.debug
    # perf report > report.target
    # diff -u report.origin report.target

  - the diff should produce no output
    (besides some white space stuff and possibly different
     date/TZ output)

test 2)
  - origin system:
    # perf record -ag -fo /tmp/perf.data -- sleep 1
  - mount origin system root to the target system on /mnt/origin
  - target system:
    # perf script --symfs /mnt/origin -I -i /mnt/origin/tmp/perf.data \
     --kallsyms /mnt/origin/proc/kallsyms
  - complete perf.data header is displayed

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337151548-2396-8-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Add uses_mmap field
Namhyung Kim [Wed, 16 May 2012 09:45:49 +0000 (18:45 +0900)]
perf target: Add uses_mmap field

If perf doesn't mmap on event (like perf stat), it should not create
per-task-per-cpu events. So just use a dummy cpu map to create a
per-task event for this case.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337161549-9870-3-git-send-email-namhyung.kim@lge.com
[ committer note: renamed .need_mmap to .uses_mmap ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoftrace: Remove selecting FRAME_POINTER with FUNCTION_TRACER
Steven Rostedt [Mon, 30 Apr 2012 13:17:03 +0000 (09:17 -0400)]
ftrace: Remove selecting FRAME_POINTER with FUNCTION_TRACER

The function tracer will enable the -pg option with gcc, which requires
that frame pointers. When FRAME_POINTER is defined in the kernel config
it adds the gcc option -fno-omit-frame-pointer which causes some problems
on some architectures. For those architectures, the FRAME_POINTER select
was not set.

When FUNCTION_TRACER was selected on these architectures that can not have
-fno-omit-frame-pointer, the -pg option is still set. But when
FRAME_POINTER is not selected, the kernel config would add the gcc option
-fomit-frame-pointer. Adding this option is incompatible with -pg
even on archs that do not need frame pointers with -pg.

The answer to this was to just not add either -fno-omit-frame-pointer
or -fomit-frame-pointer on these archs that want function tracing
but do not set FRAME_POINTER.

As it turns out, for archs that require frame pointers for function
tracing, the same can be used. If gcc requires frame pointers with
-pg, it will simply add it. The best thing to do is not select FRAME_POINTER
when function tracing is selected, and let gcc add it if needed.

Only add the -fno-omit-frame-pointer when something else selects
FRAME_POINTER, but do not add -fomit-frame-pointer if function tracing
is selected.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace/x86: Have x86 ftrace use the ftrace_modify_all_code()
Steven Rostedt [Fri, 27 Apr 2012 13:13:18 +0000 (09:13 -0400)]
ftrace/x86: Have x86 ftrace use the ftrace_modify_all_code()

To remove duplicate code, have the ftrace arch_ftrace_update_code()
use the generic ftrace_modify_all_code(). This requires that the
default ftrace_replace_code() becomes a weak function so that an
arch may override it.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Make ftrace_modify_all_code() global for archs to use
Steven Rostedt [Thu, 26 Apr 2012 18:59:43 +0000 (14:59 -0400)]
ftrace: Make ftrace_modify_all_code() global for archs to use

Rename __ftrace_modify_code() to ftrace_modify_all_code() and make
it global for all archs to use. This will remove the duplication
of code, as archs that can modify code without stop_machine()
can use it directly outside of the stop_machine() call.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Return record ip addr for ftrace_location()
Steven Rostedt [Wed, 25 Apr 2012 18:39:54 +0000 (14:39 -0400)]
ftrace: Return record ip addr for ftrace_location()

ftrace_location() is passed an addr, and returns 1 if the addr is
on a ftrace nop (or caller to ftrace_caller), and 0 otherwise.

To let kprobes know if it should move a breakpoint or not, it
must return the actual addr that is the start of the ftrace nop.
This way a kprobe placed on the location of a ftrace nop, can
instead be placed on the instruction after the nop. Even if the
probe addr is on the second or later byte of the nop, it can
simply be moved forward.

Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Consolidate ftrace_location() and ftrace_text_reserved()
Steven Rostedt [Wed, 25 Apr 2012 17:48:13 +0000 (13:48 -0400)]
ftrace: Consolidate ftrace_location() and ftrace_text_reserved()

Both ftrace_location() and ftrace_text_reserved() do basically the same thing.
They search to see if an address is in the ftace table (contains an address
that may change from nop to call ftrace_caller). The difference is
that ftrace_location() searches a single address, but ftrace_text_reserved()
searches a range.

This also makes the ftrace_text_reserved() faster as it now uses a bsearch()
instead of linearly searching all the addresses within a page.

Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Speed up search by skipping pages by address
Steven Rostedt [Wed, 25 Apr 2012 14:14:43 +0000 (10:14 -0400)]
ftrace: Speed up search by skipping pages by address

As all records in a page of the ftrace table are sorted, we can
speed up the search algorithm by checking if the address to look for
falls in between the first and last record ip on the page.

This speeds up both the ftrace_location() and ftrace_text_reserved()
algorithms, as it can skip full pages when the search address is
not in them.

Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Remove extra helper functions
Steven Rostedt [Wed, 25 Apr 2012 03:45:26 +0000 (23:45 -0400)]
ftrace: Remove extra helper functions

The ftrace_record_ip() and ftrace_alloc_dyn_node() were from the
time of the ftrace daemon. Although they were still used, they
still make things a bit more complex than necessary.

Move the code into the one function that uses it, and remove the
helper functions.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoftrace: Sort all function addresses, not just per page
Steven Rostedt [Wed, 25 Apr 2012 02:32:06 +0000 (22:32 -0400)]
ftrace: Sort all function addresses, not just per page

Instead of just sorting the ip's of the functions per ftrace page,
sort the entire list before adding them to the ftrace pages.

This will allow the bsearch algorithm to be sped up as it can
also sort by pages, not just records within a page.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agotracing: change CPU ring buffer state from tracing_cpumask
Vaibhav Nagarnaik [Fri, 4 May 2012 01:59:52 +0000 (18:59 -0700)]
tracing: change CPU ring buffer state from tracing_cpumask

According to Documentation/trace/ftrace.txt:

tracing_cpumask:

        This is a mask that lets the user only trace
        on specified CPUS. The format is a hex string
        representing the CPUS.

The tracing_cpumask currently doesn't affect the tracing state of
per-CPU ring buffers.

This patch enables/disables CPU recording as its corresponding bit in
tracing_cpumask is set/unset.

Link: http://lkml.kernel.org/r/1336096792-25373-3-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agotracing: Check return value of tracing_dentry_percpu()
Namhyung Kim [Mon, 23 Apr 2012 01:11:57 +0000 (10:11 +0900)]
tracing: Check return value of tracing_dentry_percpu()

If tracing_dentry_percpu() failed, tracing_init_debugfs_percpu()
will try to create each cpu directories on debugfs' root directory
as d_percpu is NULL.

Link: http://lkml.kernel.org/r/1335143517-2285-1-git-send-email-namhyung.kim@lge.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoring-buffer: Reset head page before running self test
Steven Rostedt [Wed, 16 May 2012 23:46:32 +0000 (19:46 -0400)]
ring-buffer: Reset head page before running self test

When the ring buffer does its consistency test on itself, it
removes the head page, runs the tests, and then adds it back
to what the "head_page" pointer was. But because the head_page
pointer may lack behind the real head page (held by the link
list pointer). The reset may be incorrect.

Instead, if the head_page exists (it does not on first allocation)
reset it back to the real head page before running the consistency
tests. Then it will be put back to its original location after
the tests are complete.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoring-buffer: Add integrity check at end of iter read
Steven Rostedt [Mon, 14 May 2012 21:02:33 +0000 (17:02 -0400)]
ring-buffer: Add integrity check at end of iter read

There use to be ring buffer integrity checks after updating the
size of the ring buffer. But now that the ring buffer can modify
the size while the system is running, the integrity checks were
removed, as they require the ring buffer to be disabed to perform
the check.

Move the integrity check to the reading of the ring buffer via the
iterator reads (the "trace" file). As reading via an iterator requires
disabling the ring buffer, it is a perfect place to have it.

If the ring buffer happens to be disabled when updating the size,
we still perform the integrity check.

Cc: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoring-buffer: Make addition of pages in ring buffer atomic
Vaibhav Nagarnaik [Fri, 4 May 2012 01:59:51 +0000 (18:59 -0700)]
ring-buffer: Make addition of pages in ring buffer atomic

This patch adds the capability to add new pages to a ring buffer
atomically while write operations are going on. This makes it possible
to expand the ring buffer size without reinitializing the ring buffer.

The new pages are attached between the head page and its previous page.

Link: http://lkml.kernel.org/r/1336096792-25373-2-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoring-buffer: Make removal of ring buffer pages atomic
Vaibhav Nagarnaik [Fri, 4 May 2012 01:59:50 +0000 (18:59 -0700)]
ring-buffer: Make removal of ring buffer pages atomic

This patch adds the capability to remove pages from a ring buffer
without destroying any existing data in it.

This is done by removing the pages after the tail page. This makes sure
that first all the empty pages in the ring buffer are removed. If the
head page is one in the list of pages to be removed, then the page after
the removed ones is made the head page. This removes the oldest data
from the ring buffer and keeps the latest data around to be read.

To do this in a non-racey manner, tracing is stopped for a very short
time while the pages to be removed are identified and unlinked from the
ring buffer. The pages are freed after the tracing is restarted to
minimize the time needed to stop tracing.

The context in which the pages from the per-cpu ring buffer are removed
runs on the respective CPU. This minimizes the events not traced to only
NMI trace contexts.

Link: http://lkml.kernel.org/r/1336096792-25373-1-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agotracing: Clean up tracing_mark_write()
Steven Rostedt [Sat, 12 May 2012 03:28:49 +0000 (23:28 -0400)]
tracing: Clean up tracing_mark_write()

On gcc 4.5 the function tracing_mark_write() would give a warning
of page2 being uninitialized. This is due to a bug in gcc because
the logic prevents page2 from being used uninitialized, and
gcc 4.6+ does not complain (correctly).

Instead of adding a "unitialized" around page2, which could show
a bug later on, I combined page1 and page2 into an array map_pages[].
This binds the two and the two are modified according to nr_pages
(what gcc 4.5 seems to ignore). This no longer gives a warning with
gcc 4.5 nor with gcc 4.6.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoRevert 'perf evlist: Fix creation of cpu map'
Namhyung Kim [Wed, 16 May 2012 09:45:48 +0000 (18:45 +0900)]
Revert 'perf evlist: Fix creation of cpu map'

The commit 55261f46702c ("perf evlist: Fix creation of cpu map") changed
to create a per-task event when no cpu target is specified. However it
caused a problem since perf-task do not allow event inheritance due to
scalability issues so that the result will contain samples only from
parent, not from its children.

So we should use perf-task-per-cpu events anyway to get the right
result. Revert it.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Analysed-by: Ingo Molnar <mingo@kernel.org>
Acked-and-tested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337161549-9870-2-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Rename functions to avoid double negation
Namhyung Kim [Wed, 16 May 2012 09:45:47 +0000 (18:45 +0900)]
perf target: Rename functions to avoid double negation

Rename perf_target__no_{cpu,task} to perf_target__has_{cpu,task} because
it's more intuitive and easy to parse (for human beings) when used with
negation.

The names are came out from David Ahern.  It is intended to be a
mechanical substitution without any functional change.

The perf_target__none remains unchanged since I couldn't find a right
name and it is hardly used with negation.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Suggested-by: David Ahern <dsahern@gmail.com>
Suggested-by: Ingo Molnar <mingo@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1337161549-9870-1-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf/x86/ibs: Fix undefined reference to `get_ibs_caps'
Robert Richter [Fri, 11 May 2012 09:44:59 +0000 (11:44 +0200)]
perf/x86/ibs: Fix undefined reference to `get_ibs_caps'

Fixing i386 allnoconfig built errors:

 arch/x86/built-in.o: In function `amd_pmu_hw_config':
 perf_event_amd.c:(.text+0xc3e1): undefined reference to `get_ibs_caps'

Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Mon, 14 May 2012 08:45:01 +0000 (10:45 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Arjan & Linus Annotation Edition

 - Fix indirect calls beautifier, reported by Linus.

 - Use the objdump comments to nuke specificities about how access to a well
   know variable is encoded, suggested by Linus.

 - Show the number of places that jump to a target, requested by Arjan.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf annotate browser: Add key bindings help window
Arnaldo Carvalho de Melo [Sat, 12 May 2012 19:36:55 +0000 (16:36 -0300)]
perf annotate browser: Add key bindings help window

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1txmtzf71eqie5xcukbfxors@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate browser: Show 'jumpy' functions
Arnaldo Carvalho de Melo [Sat, 12 May 2012 19:21:53 +0000 (16:21 -0300)]
perf annotate browser: Show 'jumpy' functions

Just press 'J' and see how many places jump to jump targets.

The hottest jump target appears in red, targets with more than one
source have a different color than single source jump targets.

Suggested-by: Arjan van de Ven <arjan@infradead.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-7452y0dmc02a20ooins7rn79@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate browser: Count the numbers of jump sources to a target
Arnaldo Carvalho de Melo [Sat, 12 May 2012 16:40:52 +0000 (13:40 -0300)]
perf annotate browser: Count the numbers of jump sources to a target

Instead of simply marking an offset as a jump target. So that we can
implement a new feature: showing "jumpy" targets, I.e. addresses that
lots of places jump to.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-vc7b0u5yxgrubig0q61ayhxf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: Introduce ->free() method in ins_ops
Arnaldo Carvalho de Melo [Sat, 12 May 2012 16:26:20 +0000 (13:26 -0300)]
perf annotate: Introduce ->free() method in ins_ops

So that we don't special case disasm_line__free, allowing each
instruction class to provide an specialized destructor, like is needed
for 'lock'.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-xxw4vs5n077tf35jsvjzylhb@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: Augment lock instruction output
Arnaldo Carvalho de Melo [Sat, 12 May 2012 16:15:34 +0000 (13:15 -0300)]
perf annotate: Augment lock instruction output

It just chops off the 'lock' and uses the ins__find, etc machinery to
call instruction specific parsers/beautifiers.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-4913ba2dzakz5rivgumosqbh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: Resolve symbols using objdump comment for single op ins
Arnaldo Carvalho de Melo [Fri, 11 May 2012 20:21:09 +0000 (17:21 -0300)]
perf annotate: Resolve symbols using objdump comment for single op ins

Starting with inc, incl, dec, decl.

Requested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-jvh0jspefr5jyn0l7qko12st@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: Resolve symbols using objdump comment
Arnaldo Carvalho de Melo [Fri, 11 May 2012 19:48:49 +0000 (16:48 -0300)]
perf annotate: Resolve symbols using objdump comment

This:

     mov    0x95bbb6(%rip),%ecx        # ffffffff81ae8d04 <d_hash_shift>

Becomes:

     mov    d_hash_shift,%ecx

Ditto for many more instructions that take two operands.

Requested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-i5opbyai2x6mn9e5yjmhx9k6@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: Use raw form for register indirect call instructions
Arnaldo Carvalho de Melo [Fri, 11 May 2012 15:28:55 +0000 (12:28 -0300)]
perf annotate: Use raw form for register indirect call instructions

   callq  *0x10(%rax)

was being rendered in simplified mode as:

   callq  *10

I.e. hexa, but without the 0x and omitting the register. In such cases
just use the raw form.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-m91tv004h2m1fkfgu6ovx3hb@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Fri, 11 May 2012 06:13:55 +0000 (08:13 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Fixes and improvements for perf/core:

- perf_target: abstraction for --uid, --pid, --tid, --cpu, --all-cpus handling,
  eliminating code duplicated in the tools, having constraints that apply to
  all of them, from Namhyung Kim

- Fixes for handling fallback to cpu-clock on PPC, from David Ahern

- Fix for processing events with unknown size, from Jiri Olsa

- Compilation fix on 32-bit, from Jiri Olsa

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agotracing: Do not enable function event with enable
Steven Rostedt [Thu, 10 May 2012 19:55:43 +0000 (15:55 -0400)]
tracing: Do not enable function event with enable

With the adding of function tracing event to perf, it caused a
side effect that produces the following warning when enabling all
events in ftrace:

 # echo 1 > /sys/kernel/debug/tracing/events/enable

[console]
event trace: Could not enable event function

This is because when enabling all events via the debugfs system
it ignores events that do not have a ->reg() function assigned.
This was to skip over the ftrace internal events (as they are
not TRACE_EVENTs). But as the ftrace function event now has
a ->reg() function attached to it for use with perf, it is no
longer ignored.

Worse yet, this ->reg() function is being called when it should
not be. It returns an error and causes the above warning to
be printed.

By adding a new event_call flag (TRACE_EVENT_FL_IGNORE_ENABLE)
and have all ftrace internel event structures have it set,
setting the events/enable will no longe try to incorrectly enable
the function event and does not warn.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoperf hists browser: Use '/' for search/filter instead of 's'
Arnaldo Carvalho de Melo [Wed, 9 May 2012 15:21:22 +0000 (12:21 -0300)]
perf hists browser: Use '/' for search/filter instead of 's'

That is what is used in vi and mutt, and as well on the 'annotate'
browser.

Eventually we can have keymappings to make people used to other key
associations more confortable.

Suggested-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-fyln9286b8gx5q4n277l0djs@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge branch 'tip/perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
Ingo Molnar [Wed, 9 May 2012 17:34:30 +0000 (19:34 +0200)]
Merge branch 'tip/perf/core' of git://git./linux/kernel/git/rostedt/linux-trace into perf/core

12 years agoperf stat: handle ENXIO error for perf_event_open
David Ahern [Tue, 8 May 2012 15:29:16 +0000 (09:29 -0600)]
perf stat: handle ENXIO error for perf_event_open

perf stat on PPC currently fails to run:

$ perf stat -- sleep 1
  Error: open_counter returned with 6 (No such device or address). /bin/dmesg may provide additional information.

  Fatal: Not all events could be opened.

The problem is that until 2.6.37 (behavior changed with commit b0a873e)
perf on PPC returns ENXIO when hw_perf_event_init() fails. With this
patch we get the expected behavior:

$ perf stat -v -- sleep 1
cycles event is not supported by the kernel.
stalled-cycles-frontend event is not supported by the kernel.
stalled-cycles-backend event is not supported by the kernel.
instructions event is not supported by the kernel.
branches event is not supported by the kernel.
branch-misses event is not supported by the kernel.

...

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336490956-57145-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate: shorten helpline so it fits in visible space
David Ahern [Tue, 8 May 2012 21:01:19 +0000 (15:01 -0600)]
perf annotate: shorten helpline so it fits in visible space

Additional toggles have pushed the help line out of view on a modestly
sized terminal (120 columns wide). Shorten it to just reminders.

Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1336510879-64610-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf record: Reset event name when falling back to cpu-clock
David Ahern [Tue, 8 May 2012 16:50:11 +0000 (10:50 -0600)]
perf record: Reset event name when falling back to cpu-clock

perf-record defaults to the H/W cycles event and if it is not supported
falls back to cpu-clock. Reset the event name as well.

Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1336495811-58461-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf top: Update event name when falling back to cpu-clock
David Ahern [Tue, 8 May 2012 16:49:49 +0000 (10:49 -0600)]
perf top: Update event name when falling back to cpu-clock

The 'perf top' command falls back to cpu-clock if the H/W cycles event
is not supported, but the event name is not updated leading to a
misleading header:

PerfTop: 8 irqs/sec  kernel:75.0%  exact:  0.0% [1000Hz cycles],  ...

Update the event name when the event type is changed so that the
header displays correctly:

PerfTop: 794 irqs/sec  kernel:100.0%  exact:  0.0% [1000Hz cpu-clock], ...

Signed-off-by: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/1336495789-58420-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf stat: handle ENXIO error for perf_event_open
David Ahern [Tue, 8 May 2012 15:29:16 +0000 (09:29 -0600)]
perf stat: handle ENXIO error for perf_event_open

perf stat on PPC currently fails to run:

$ perf stat -- sleep 1
  Error: open_counter returned with 6 (No such device or address). /bin/dmesg may provide additional information.

  Fatal: Not all events could be opened.

The problem is that until 2.6.37 (behavior changed with commit b0a873e)
perf on PPC returns ENXIO when hw_perf_event_init() fails. With this
patch we get the expected behavior:

$ perf stat -v -- sleep 1
cycles event is not supported by the kernel.
stalled-cycles-frontend event is not supported by the kernel.
stalled-cycles-backend event is not supported by the kernel.
instructions event is not supported by the kernel.
branches event is not supported by the kernel.
branch-misses event is not supported by the kernel.

...

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336490956-57145-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf record: Fix fallback to cpu-clock on ppc
David Ahern [Tue, 8 May 2012 15:28:57 +0000 (09:28 -0600)]
perf record: Fix fallback to cpu-clock on ppc

perf-record on PPC is not falling back to cpu-clock:

$ perf record -ag -fo /tmp/perf.data -- sleep 1

  Error: sys_perf_event_open() syscall returned with 6 (No such device or address).  /bin/dmesg may provide additional information.

  Fatal: No CONFIG_PERF_EVENTS=y kernel support configured?

The problem is that until 2.6.37 (behavior changed with commit b0a873e)
perf on PPC returns ENXIO when hw_perf_event_init() fails. With this
patch we get the expected behavior:

$ perf record -ag -fo /tmp/perf.data -v -- sleep 1
Old kernel, cannot exclude guest or host samples.
The cycles event is not supported, trying to fall back to cpu-clock-ticks
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.151 MB /tmp/perf.data (~6592 samples) ]

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336490937-57106-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf report: Fix format string for x86-32 compilation
Jiri Olsa [Wed, 2 May 2012 11:37:07 +0000 (13:37 +0200)]
perf report: Fix format string for x86-32 compilation

Using PRIu64 for printing out u64 nr_events to fix compilation
for x86 32 bits.

Cc: Arun Sharma <asharma@fb.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Frank C. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1335958638-5160-7-git-send-email-jolsa@redhat.com
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agosched, perf: Use a single callback into the scheduler
Peter Zijlstra [Tue, 8 May 2012 16:56:04 +0000 (18:56 +0200)]
sched, perf: Use a single callback into the scheduler

We can easily use a single callback for both sched-in and sched-out. This
reduces the code footprint in the scheduler path as well as removes
the PMU black spot otherwise present between the out and in callback.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-o56ajxp1edwqg6x9d31wb805@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Fix usage of IBS op current count
Robert Richter [Mon, 2 Apr 2012 18:19:18 +0000 (20:19 +0200)]
perf/x86-ibs: Fix usage of IBS op current count

The value of IbsOpCurCnt rolls over when it reaches IbsOpMaxCnt. Thus,
it is reset to zero by hardware. To get the correct count we need to
add the max count to it in case we received an ibs sample (valid bit
set).

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-13-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Catch spurious interrupts after stopping IBS
Robert Richter [Mon, 2 Apr 2012 18:19:17 +0000 (20:19 +0200)]
perf/x86-ibs: Catch spurious interrupts after stopping IBS

After disabling IBS there could be still incomming NMIs with samples
that even have the valid bit cleared. Mark all this NMIs as handled to
avoid spurious interrupt messages.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-12-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Implement workaround for IBS erratum #420
Robert Richter [Mon, 2 Apr 2012 18:19:16 +0000 (20:19 +0200)]
perf/x86-ibs: Implement workaround for IBS erratum #420

When disabling ibs there might be the case where hardware continuously
generates interrupts. This is described in erratum #420 (Instruction-
Based Sampling Engine May Generate Interrupt that Cannot Be Cleared).
To avoid this we must clear the counter mask first and then clear the
enable bit. This patch implements this.

See Revision Guide for AMD Family 10h Processors, Publication #41322.

Note: We now keep track of the last read ibs config value which is
then used to disable ibs. To update the config value we pass now a
pointer to the functions reading it.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-11-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Extend hw period that triggers overflow
Robert Richter [Mon, 2 Apr 2012 18:19:15 +0000 (20:19 +0200)]
perf/x86-ibs: Extend hw period that triggers overflow

If the last hw period is too short we might hit the irq handler which
biases the results. Thus try to have a max last period that triggers
the sw overflow.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-10-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Trigger overflow if remaining period is too small
Robert Richter [Mon, 2 Apr 2012 18:19:14 +0000 (20:19 +0200)]
perf/x86-ibs: Trigger overflow if remaining period is too small

There are cases where the remaining period is smaller than the minimal
possible value. In this case the counter is restarted with the minimal
period. This is of no use as the interrupt handler will trigger
immediately again and most likely hits itself. This biases the
results.

So, if the remaining period is within the min range, we better do not
restart the counter and instead trigger the overflow.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-9-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Rename some variables
Robert Richter [Mon, 2 Apr 2012 18:19:13 +0000 (20:19 +0200)]
perf/x86-ibs: Rename some variables

Simple patch that just renames some variables for better
understanding.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-8-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Precise event sampling with IBS for AMD CPUs
Robert Richter [Mon, 12 Mar 2012 11:54:32 +0000 (12:54 +0100)]
perf/x86-ibs: Precise event sampling with IBS for AMD CPUs

This patch adds support for precise event sampling with IBS. There are
two counting modes to count either cycles or micro-ops. If the
corresponding performance counter events (hw events) are setup with
the precise flag set, the request is redirected to the ibs pmu:

 perf record -a -e cpu-cycles:p ...    # use ibs op counting cycle count
 perf record -a -e r076:p ...          # same as -e cpu-cycles:p
 perf record -a -e r0C1:p ...          # use ibs op counting micro-ops

Each ibs sample contains a linear address that points to the
instruction that was causing the sample to trigger. With ibs we have
skid 0. Thus, ibs supports precise levels 1 and 2. Samples are marked
with the PERF_EFLAGS_EXACT flag set. In rare cases the rip is invalid
when IBS was not able to record the rip correctly. Then the
PERF_EFLAGS_EXACT flag is cleared and the rip is taken from pt_regs.

V2:
* don't drop samples in precise level 2 if rip is invalid, instead
  support the PERF_EFLAGS_EXACT flag

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20120502103309.GP18810@erda.amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Take instruction pointer from ibs sample
Robert Richter [Mon, 2 Apr 2012 18:19:11 +0000 (20:19 +0200)]
perf/x86-ibs: Take instruction pointer from ibs sample

Each IBS sample contains a linear address of the instruction that
caused the sample to trigger. This address is more precise than the
rip that was taken from the interrupt handler's stack. Update the rip
with that address. We use this in the next patch to implement
precise-event sampling on AMD systems using IBS.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-6-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Fix frequency profiling
Robert Richter [Mon, 2 Apr 2012 18:19:10 +0000 (20:19 +0200)]
perf/x86-ibs: Fix frequency profiling

Fixing profiling at a fixed frequency, in this case the freq value and
sample period was setup incorrectly. Since sampling periods are
adjusted we also allow periods that have lower 4 bits set.

Another fix is the setup of the hw counter: If we modify
hwc->sample_period, we also need to update hwc->last_period and
hwc->period_left.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-5-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Enable ibs op micro-ops counting mode
Robert Richter [Mon, 2 Apr 2012 18:19:09 +0000 (20:19 +0200)]
perf/x86-ibs: Enable ibs op micro-ops counting mode

Allow enabling ibs op micro-ops counting mode.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-4-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf: Pass last sampling period to perf_sample_data_init()
Robert Richter [Mon, 2 Apr 2012 18:19:08 +0000 (20:19 +0200)]
perf: Pass last sampling period to perf_sample_data_init()

We always need to pass the last sample period to
perf_sample_data_init(), otherwise the event distribution will be
wrong. Thus, modifiyng the function interface with the required period
as argument. So basically a pattern like this:

        perf_sample_data_init(&data, ~0ULL);
        data.period = event->hw.last_period;

will now be like that:

        perf_sample_data_init(&data, ~0ULL, event->hw.last_period);

Avoids unininitialized data.period and simplifies code.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-3-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86-ibs: Fix update of period
Robert Richter [Mon, 2 Apr 2012 18:19:07 +0000 (20:19 +0200)]
perf/x86-ibs: Fix update of period

The last sw period was not correctly updated on overflow and thus led
to wrong distribution of events. We always need to properly initialize
data.period in struct perf_sample_data.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1333390758-10893-2-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoMerge branch 'perf/x86-ibs' into perf/core
Ingo Molnar [Wed, 9 May 2012 13:22:23 +0000 (15:22 +0200)]
Merge branch 'perf/x86-ibs' into perf/core

12 years agotracing: Remove ftrace_disable/enable_cpu()
Steven Rostedt [Wed, 9 May 2012 00:57:53 +0000 (20:57 -0400)]
tracing: Remove ftrace_disable/enable_cpu()

The ftrace_disable_cpu() and ftrace_enable_cpu() functions were
needed back before the ring buffer was lockless. Now that the
ring buffer is lockless (and has been for some time), these functions
serve no purpose, and unnecessarily slow down operations of the tracer.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agotracing: Use seq_*_private interface for some seq files
Jiri Olsa [Wed, 25 Apr 2012 08:23:39 +0000 (10:23 +0200)]
tracing: Use seq_*_private interface for some seq files

It's appropriate to use __seq_open_private interface to open
some of trace seq files, because it covers all steps we are
duplicating in tracing code - zallocating the iterator and
setting it as seq_file's private.

Using this for following files:
  trace
  available_filter_functions
  enabled_functions

Link: http://lkml.kernel.org/r/1335342219-2782-5-git-send-email-jolsa@redhat.com
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
[
 Fixed warnings for:
   kernel/trace/trace.c: In function '__tracing_open':
   kernel/trace/trace.c:2418:11: warning: unused variable 'ret' [-Wunused-variable]
   kernel/trace/trace.c:2417:19: warning: unused variable 'm' [-Wunused-variable]
]

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoMerge branch 'perf/annotate' of git://git.kernel.org/pub/scm/linux/kernel/git/acme...
Ingo Molnar [Tue, 8 May 2012 14:55:14 +0000 (16:55 +0200)]
Merge branch 'perf/annotate' of git://git./linux/kernel/git/acme/linux into perf/core

Perf annotate browser improvements:

 - Get back the line separating the overheads from the disassembly, requested by
   Peter Zijlstra, Linus agreed now that it is a solid line and more column real
   state was harvested. Also it has the jump->arrow lines separated from it by
   the address/jump target column.

 - Don't change asm line color when toggling source code view. Requested by
   Peter Zijlstra.

Current snapshot:

 avtab_search_node
        │      push   %rbp
        │      mov    %rsp,%rbp
        │    → callq  mcount
        │      movzwl 0x6(%rsi),%edx
        │      and    $0x7fff,%dx
        │      test   %rdi,%rdi
        │    ↓ jne    20
   0.42 │17:┌─→xor    %eax,%eax
        │19:│  leaveq
   0.42 │   │← retq
        │   │  nopl   0x0(%rax,%rax,1)
        │20:│  mov    (%rdi),%rax
   0.08 │   │  test   %rax,%rax
        │   └──je     17
        │      movzwl (%rsi),%ecx
        │      movzwl 0x2(%rsi),%r9d
        │      movzwl 0x4(%rsi),%r8d
        │      movzwl %cx,%esi
        │      movzwl %r9w,%r10d
        │      shl    $0x9,%esi
        │      lea    (%rsi,%r10,4),%esi
        │      lea    (%r8,%rsi,1),%esi
        │      and    0x10(%rdi),%si
        │      movzwl %si,%esi
        │      mov    (%rax,%rsi,8),%rax
   1.01 │      test   %rax,%rax
        │    ↑ je     19
        │      nopw   0x0(%rax,%rax,1)
   3.19 │60:   cmp    %cx,(%rax)
        │    ↓ jne    7e
   0.08 │      cmp    %r9w,0x2(%rax)
        │    ↓ jne    7e
        │      cmp    %r8w,0x4(%rax)
        │    ↓ jne    79
        │      test   %dx,0x6(%rax)
        │    ↑ jne    19
        │79:   cmp    %r8w,0x4(%rax)
  83.45 │7e: ↑ ja     17
   3.36 │      mov    0x10(%rax),%rax
   7.98 │      test   %rax,%rax
        │    ↑ jne    60
        │      leaveq
        │    ← retq

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf top: Default to system wide using perf_target methods
Arnaldo Carvalho de Melo [Tue, 8 May 2012 13:47:09 +0000 (10:47 -0300)]
perf top: Default to system wide using perf_target methods

Additionally we were not checking if a cpu list had been provided by the
user. Fix that.

Reported-by: David Ahern <dsahern@gmail.com>
Reported-by: Namhyung Kim <namhyung@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ao3zrouylwmt7h9ikj0krubi@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agotracing: Prevent wasting time evaluating parameters in trace_preempt_on/off
Minho Ban [Mon, 7 May 2012 02:36:00 +0000 (11:36 +0900)]
tracing: Prevent wasting time evaluating parameters in trace_preempt_on/off

This fixes spending time for evaluating parameters in trace_preempt_on/off when
the tracer config is off.

The patch mainly inspired by Steven Rostedt, thanks Steven.

Link: http://lkml.kernel.org/r/4FA73510.7070705@samsung.com
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Turner <pjt@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Minho Ban <mhban@samsung.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoperf annotate browser: Compact 'nop' output
Arnaldo Carvalho de Melo [Mon, 7 May 2012 21:57:02 +0000 (18:57 -0300)]
perf annotate browser: Compact 'nop' output

Just suppress the nop operands, future infrastructure that will record
the instruction lenght (and its contents) in struct ins will allow
rendering them as nopN, i.e. nop5 for a 5-byte nop.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-qddbeglfzqdlal8vj2yaj67y@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf annotate browser: Do raw printing in 'o'ffset in a single place
Arnaldo Carvalho de Melo [Mon, 7 May 2012 21:54:16 +0000 (18:54 -0300)]
perf annotate browser: Do raw printing in 'o'ffset in a single place

Instead of doing the same in all ins scnprintf methods.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-8mfairi2n1nentoa852alazv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf stat: Use perf_evlist__create_maps
Namhyung Kim [Mon, 7 May 2012 05:09:04 +0000 (14:09 +0900)]
perf stat: Use perf_evlist__create_maps

Use same function with perf record and top to share the code checks
combinations of different switches.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-8-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Consolidate target task/cpu checking
Namhyung Kim [Mon, 7 May 2012 05:09:03 +0000 (14:09 +0900)]
perf target: Consolidate target task/cpu checking

There are places that check whether target task/cpu is given or not and
some of them didn't check newly introduced uid or cpu list. Add and use
three of helper functions to treat them properly.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-7-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Introduce perf_target__strerror()
Namhyung Kim [Mon, 7 May 2012 05:09:02 +0000 (14:09 +0900)]
perf tools: Introduce perf_target__strerror()

The perf_target__strerror() sets @buf to a string that describes the
(perf_target-specific) error condition that is passed via @errnum.

This is similar to strerror_r() and does same thing if @errnum has a
standard errno value.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-6-git-send-email-namhyung.kim@lge.com
[ committer note: No need to use PERF_ERRNO_TARGET__SUCCESS, use shorter idiom ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Introduce perf_target__parse_uid()
Namhyung Kim [Mon, 7 May 2012 05:09:01 +0000 (14:09 +0900)]
perf target: Introduce perf_target__parse_uid()

Add and use the modern perf_target__parse_uid() and get rid of the old
parse_target_uid().

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-5-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Introduce perf_target_errno
Namhyung Kim [Mon, 7 May 2012 05:09:00 +0000 (14:09 +0900)]
perf target: Introduce perf_target_errno

The perf_target_errno enumerations are used to indicate specific error
cases on perf target operations. It'd help libperf being a more generic
library.

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Suggested-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-4-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf evlist: Fix creation of cpu map
Namhyung Kim [Mon, 7 May 2012 05:08:59 +0000 (14:08 +0900)]
perf evlist: Fix creation of cpu map

Currently, 'perf record -- sleep 1' creates a cpu map for all online
cpus since it turns out calling cpu_map__new(NULL). Fix it.

Also it is guaranteed that cpu_list is NULL if PID/TID is given by
calling perf_target__validate(), so we can make the conditional bit
simpler.

This also fixes perf test 7 (Validate) failure on my 6 core machine:

  $ cat /sys/devices/system/cpu/online
  0-11
  $ ./perf test -v 7
   7: Validate PERF_RECORD_* events & perf_sample fields:
  --- start ---
  perf_evlist__mmap: Operation not permitted
  ---- end ----
  Validate PERF_RECORD_* events & perf_sample fields: FAILED!

Signed-off-by: Namhyung Kim <namhyung.kim@lge.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1336367344-28071-3-git-send-email-namhyung.kim@lge.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf top: Set target.system_wide
Arnaldo Carvalho de Melo [Mon, 7 May 2012 19:33:56 +0000 (16:33 -0300)]
perf top: Set target.system_wide

Check if neither of --pid, --tid or --uid was specified and if so, set
system_wide appropriately.

Namhyung's patch would make using any of the above target specifiers
emit a warning in perf_target__validate, since it would see
target.system_wide set and one of the others as well.

So set system_wide after validation.

Suggested-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-6e4zrji1uw0rinfyoitl0wi4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge branch 'tip/perf/core-4' of git://git.kernel.org/pub/scm/linux/kernel/git/roste...
Ingo Molnar [Mon, 7 May 2012 09:03:52 +0000 (11:03 +0200)]
Merge branch 'tip/perf/core-4' of git://git./linux/kernel/git/rostedt/linux-trace into perf/core

12 years agoperf: Turn off compiler warnings for flex and bison generated files
Greg Kroah-Hartman [Fri, 4 May 2012 18:32:54 +0000 (11:32 -0700)]
perf: Turn off compiler warnings for flex and bison generated files

We don't know what types of warnings different versions of flex
and bison combined with different versions of gcc is going to
generate, so just punt and don't warn about anything.

This fixes the build of perf for me on an openSUSE 12.1 system.

Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Link: http://lkml.kernel.org/r/20120504183254.GA11154@kroah.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf session: Fail on processing event with unknown size
Jiri Olsa [Mon, 16 Apr 2012 18:42:51 +0000 (20:42 +0200)]
perf session: Fail on processing event with unknown size

Currently if we cannot decide the size of the event, we guess next
event possition by:
  "... check alignment, and increment a single u64 in the hope
  to catch on again 'soon'"

This usually ends up with segfault or endless loop. It's better
to admit the failure right away, then pretend nothing happened.
It makes the life easier ;)

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20120416184251.GA11503@m.brq.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoftrace/x86: Use asm/kprobes.h instead of linux/kprobes.h
Steven Rostedt [Fri, 4 May 2012 13:26:16 +0000 (09:26 -0400)]
ftrace/x86: Use asm/kprobes.h instead of linux/kprobes.h

If CONFIG_KPROBES is not set, then linux/kprobes.h will not include
asm/kprobes.h needed by x86/ftrace.c for the BREAKPOINT macro.

The x86/ftrace.c file should just include asm/kprobes.h as it does not
need the rest of kprobes.

Reported-by: Ingo Molnar <mingo@elte.hu>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
12 years agoperf annotate browser: Don't change the asm line color when toggling source
Arnaldo Carvalho de Melo [Thu, 3 May 2012 16:23:00 +0000 (13:23 -0300)]
perf annotate browser: Don't change the asm line color when toggling source

Gets confusing. Remains to be chosen an appropriate different color for
source code.

This effectively reverts 58e817d997d1 ("perf annotate: Print asm code as
blue when source code is displayed")

Requested-by: Peter Zijlstra <peterz@infradead.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-qy9iq32nj3uqe5dbiuq9e3j9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>